The R User Conference, useR!

2011
August 16-18 2011
University of Warwick, Coventry, UK

Book of Contributed Abstracts

Compiled 2011-07-28

1
Contents

Tuesday 16th August 5

Kaleidoscope Ia, 09:50-10:50 5
Spectroscopic Data in R and Validation of Soft Classifiers: Classifying Cells and Tissues by Raman
Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Revisiting Multi-Subject Random Effects in fMRI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Putting the R into Randomisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Kaleidoscope Ib, 09:50-10:50 9

Bringing the power of complex data analysis methods into R . . . . . . . . . . . . . . . . . . . . . . . . 9
Using the Google Visualisation API with R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Experimenting with a TTY connection for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Kaleidoscope Ic, 09:50-10:50 12

The R Ecosystem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Rc2: R collaboration in the cloud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
RStudio: Integrated Development Environment for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Portfolio Management, 11:15-12:35 15

R in the Practice of Risk Management Today . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Stress Testing with R-Adamant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Handling Multiple Time Scales in Hedge Fund Index Products . . . . . . . . . . . . . . . . . . . . . . . 17

Bioinformatics and High-throughput Data, 11:15-12:35 18

AFLP: generating objective and repeatable genetic data . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
The nz.seq package for handling genetic sequences in the Netezza Performance Server . . . . . . . . . . 19
Classification of Coverage Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Finite Mixture Model Clustering of SNP Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

High Performance Computing, 11:15-12:35 22

GPU computing and R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Deploying and Benchmarking R on a Large Shared Memory System . . . . . . . . . . . . . . . . . . . . 23
The CUtil package which enables GPU computation in R . . . . . . . . . . . . . . . . . . . . . . . . . . 24
OBANSoft: integrated software for Bayesian statistics and high performance computing with R . . . . . 25

Reporting Technologies and Workflows, 11:15-12:35 26

R2wd: writing Word Documents from R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
FRAD - Fast Results Analysis and Display . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
The Emacs Org-mode: Reproducible Research and Beyond . . . . . . . . . . . . . . . . . . . . . . . . . 28
Efficient data analysis workflow in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Teaching, 11:15-12:35 30
Teaching Statistics to Psychology Students using Reproducible Computing package RC and supporting
Peer Review Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Teaching applied statistics to students in natural sciences using the R Commander . . . . . . . . . . . . 31
Automatic generation of exams in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Modelling Systems and Networks, 16:00-17:00 33

An S4 Object structure for emulation - the approximation of complex functions . . . . . . . . . . . . . . 33
The structmcmc package: Structural inference of Bayesian networks using MCMC . . . . . . . . . . . . 34
Computation of generalized Nash equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Computational Physics and Chemometrics, 16:00-17:00 36

Segmented regression in thermo-physics modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Sparse Bayesian kernel projections for classification of near-infrared spectroscopy data . . . . . . . . . . 37
Recovering Signals and Information From Radio Frequencies Using R (A high school student’s experience) 38

Visualisation, 16:00-17:00 39
animatoR: dynamic graphics in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Graphical Syntax for Structables and their Mosaic Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
RMB: Visualising categorical data with Relative Multiple Barcharts . . . . . . . . . . . . . . . . . . . . 41

2
Official and Social Statistics, 16:00-17:00 42
ObsSensitivity: An R package for power analysis for sensitivity analyses of Observational Studies . . . . 42
Visualizing Multilevel Propensity Score Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Dimensionality Reduction and Variable Selection, 16:00-17:00 44

ClustOfVar: an R package for the clustering of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Variable Screening and Parameter Estimation for High-Dimensional Generalized Linear Mixed Models
Using L1-Penalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
gamboostLSS: boosting generalized additive models for location, scale and shape . . . . . . . . . . . . . 46

Business Management, 16:00-17:00 47

SCperf: An inventory management package for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Using R to test transaction cost measurement for supply chain relationship: A structural equation model 48
Integrating R and Excel for automatic business forecasting . . . . . . . . . . . . . . . . . . . . . . . . . 49

Wednesday 17th August 50

Kaleidoscope IIa, 09:50-10:50 50
Using R to quantify the buildup in extent of free exploration in mice . . . . . . . . . . . . . . . . . . . . 50
Changepoint analysis with the changepoint package in R . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Clustering patterns in streamflow to produce regionally or anthropogenically similar groups . . . . . . . 52

Panel Discussion I, 09:50-10:50 53

R User Group Panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

Kaleidoscope IIb, 09:50-10:50 54

RTextTools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
The Role of R in Lab Automation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Using R data functions with TIBCO Spotfire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

Spatio-Temporal Statistics, 11:15-12:35 57

Spatio-Temporal Bayesian Modelling using R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Applying geospatial techniques to temporal data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Structured Additive Regression Models: An R Interface to BayesX . . . . . . . . . . . . . . . . . . . . . 59

Molecular and Cell Biology, 11:15-12:35 60

The R package isocir for Isotonic Inference for Circular Data. Applications to Problems Encountered in
Cell Biology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
CircNNTSR: An R Package for the Statistical Analysis of Circular Data Based on Nonnegative Trigono-
metric Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Summary statistics selection for ABC inference in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Power and minimal sample size for multivariate analysis of microarrays . . . . . . . . . . . . . . . . . . 63

Mixed Effect Models, 11:15-12:35 64

Kenward-Roger modification of the F-statistic for some linear mixed models fitted with lmer . . . . . . 64
lqmm: Estimating Quantile Regression Models for Independent and Hierarchical Data with R . . . . . . 65
lcmm: an R package for estimation of latent class mixed models and joint latent class models . . . . . . 66
Mixed-effects Maximum Likelihood Difference Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Programming, 11:15-12:35 68
Tricks and Traps for Young Players . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Software design patterns in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Random input testing with R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
An Open Source Visual R Debugger in StatET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

Data Mining Applications, 11:15-12:35 72

Predicting the offender’s age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Leveraging Online Social Network Data and External Data Sources to Predict Personality . . . . . . . . 73
Using R to Model Click-Stream Data to Understand Users’ Path To Conversion . . . . . . . . . . . . . . 74

3
Development of R, 16:00-17:00 75
Packaging R for Ubuntu: Recent Changes and Future Opportunities . . . . . . . . . . . . . . . . . . . . 75
Interpreter Internals: Unearthing Buried Treasure with CXXR . . . . . . . . . . . . . . . . . . . . . . . 76
R’s Participation in the Google Summer of Code 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Geospatial Techniques, 16:00-17:00 78

Converting a spatial network to a graph in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Spatial modelling with the R–GRASS Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
sos4R - Accessing SensorWeb Data from R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Genomics and Bioinformatics, 16:00-17:00 81

MALDIquant: Quantitative Analysis of MALDI-TOF Proteomics Data . . . . . . . . . . . . . . . . . . 81
QuACN: Analysis of Complex Biological Networks using R . . . . . . . . . . . . . . . . . . . . . . . . . 82
Investigate clusters of co-expressed and co-located genes at a genomic scale using CoCoMap . . . . . . . 83

Regression Modelling, 16:00-17:00 84

Beta Regression: Shaken, Stirred, Mixed, and Partitioned . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Regression Models for Ordinal Data: Introducing R-package ordinal . . . . . . . . . . . . . . . . . . . . 85
Multiple choice models: why not the same answer? A comparison among LIMDEP, R, SAS and Stata. . 86

R in the Business World, 16:00-17:00 87

Odesseus vs. Ajax: How to build an R presence in a corporate SAS environment . . . . . . . . . . . . . 87
A Validation/Qualification Solution for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
R as a statistical tool for human factor engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

Hydrology and Soil Science, 17:05-18:05 90

GWSDAT (GroundWater Spatiotemporal Data Analysis Tool) . . . . . . . . . . . . . . . . . . . . . . . 90
IntR – Interactive GUI for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Visualisation and modelling of soil data using the aqp package . . . . . . . . . . . . . . . . . . . . . . . 92

Biostatistical Modelling, 17:05-18:05 93

survAUC: Estimators of Prediction Accuracy for Time-to-Event Data . . . . . . . . . . . . . . . . . . . 93
Higher-order likelihood inference in meta-analysis using R . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Gaussian copula regression using R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

Psychometrics, 17:05-18:05 96
Multinomial Processing Tree Models in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Detecting Invariance in Psychometric Models with the psychotree Package . . . . . . . . . . . . . . . . . 97
Investigating multidimensional unfolding models using R2WinBUGS . . . . . . . . . . . . . . . . . . . . 98

Multivariate Data, 17:05-18:05 99

Tests for Multivariate Linear Models with the car Package . . . . . . . . . . . . . . . . . . . . . . . . . . 99
missMDA: a package to handle missing values in and with multivariate exploratory data analysis methods100
MAINT.DATA: Modeling and Analysing Interval Data in R . . . . . . . . . . . . . . . . . . . . . . . . . 101

Interfaces, 17:05-18:05 102

Web 2.0 for R scripts & workflows: Tiki & PluginR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Browser Based Applications Supported by R in Pipeline Pilot . . . . . . . . . . . . . . . . . . . . . . . . 103
A new task-based GUI for R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

Thursday 18th August 105

Financial Models, 09:50-10:50 105
Computational aspects of continuous-time-arma (CARMA) models: The ctarma package . . . . . . . . . 105
robKalman - An R package for robust Kalman filtering revisited . . . . . . . . . . . . . . . . . . . . . . 106
(Robust) Online Filtering in Regime Switching Models and Application to Investment Strategies for
Asset Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Ecology and Ecological Modelling, 09:50-10:50 108

Using R for the Analysis of Bird Demography on a Europe-wide Scale . . . . . . . . . . . . . . . . . . . 108
Using OpenBUGS and lmer to study variation in plant demographic rates over several spatial and
temporal scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
An effort to improve nonlinear modeling practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

4
Generalized Linear Models, 09:50-10:50 111
brglm: Bias reduction in generalized linear models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Large Scale, Massively Parallel Logistic Regression in R with the Netezza Analytics Package . . . . . . . 112
The binomTools package: Performing model diagnostics on binomial regression models . . . . . . . . . . 113

Reporting Data, 09:50-10:50 114

uniPlot - A package to uniform and customize R graphics . . . . . . . . . . . . . . . . . . . . . . . . . . 114
sparkTable: Generating Graphical Tables for Websites and Documents with R . . . . . . . . . . . . . . 115
compareGroups package, updated and improved . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

Process Optimization, 09:50-10:50 117

Six Sigma Quality Using R: Tools and Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Process Performance and Capability Statistics for Non-Normal Distributions in R . . . . . . . . . . . . . 118
R-Package JOP: Optimization of Multiple Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

Inference, 11:15-12:35 120

Density Estimation Packages in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
The benchden Package: Benchmark Densities for Nonparametric Density Estimation . . . . . . . . . . . 121
Nonparametric Goodness-of-Fit Tests for Discrete Null Distributions . . . . . . . . . . . . . . . . . . . . 122
An algorithm for the computation of the power of Monte Carlo tests with guaranteed precision . . . . . 123

Population Genetics and Genetics Association Studies, 11:15-12:35 124

Simple haplotype analyses in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Mixed models of large pedigrees in genetic association studies . . . . . . . . . . . . . . . . . . . . . . . . 125
Graphical tools for assessing Hardy-Weinberg equilibrium for bi-allelic genetic markers . . . . . . . . . . 126

Neuroscience, 11:15-12:35 127

Detecting Drug Effects in the Brain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Statistical Parametric Maps for Functional MRI Experiments in R: The Package fmri . . . . . . . . . . 128
neuRosim an R package for simulation of fMRI magnitude data with realistic noise . . . . . . . . . . . . 129

Data Management, 11:15-12:35 130

It’s a Boy! An Analysis of Tens of Millions of Birth Records Using R . . . . . . . . . . . . . . . . . . . . 130
Challenges of working with a large database of routinely collected health data: Combining SQL and R . 131
Demographic: Classes and Methods for Data about Populations . . . . . . . . . . . . . . . . . . . . . . . 132
Correcting data violating linear restrictions using the deducorrect and editrules packages . . . . . . . . . 133

Interactive Graphics in R, 11:15-12:35 134

iWebPlots: Introducing a new R package for the creation of interactive web-based scatter plots . . . . . 134
Rocessing: Interactive Visualizations in R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Easy Interactive ggplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
RnavGraph and the tk canvas widget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

Kaleidoscope IIIa, 14:00-15:00 138

Using R for systems understanding - a dynamic approach . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Using multidimensional scaling with Duchon splines for reliable finite area smoothing . . . . . . . . . . 139
Studying galaxies in the nearby Universe, using R and ggplot2 . . . . . . . . . . . . . . . . . . . . . . . 140

Panel Discussion II, 14:00-15:00 141

Panel discussion: Challenges Bringing R into Commercial Environments . . . . . . . . . . . . . . . . . . 141

Kaleidoscope IIIb, 14:00-15:00 142

microbenchmark: A package to accurately benchmark R expressions . . . . . . . . . . . . . . . . . . . . 142
Vector Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Regular Posters 144

Late-breaking Posters 170

5
Kaleidoscope Ia, Tuesday 09:50-10:50 6

Spectroscopic Data in R and Validation of Soft Classifiers:

Classifying Cells and Tissues by Raman Spectroscopy
?
Claudia Beleites1,2, , Christoph Krafft2 , Jürgen Popp2,3 , and Valter Sergo1
1. CENMAT and Dept. of Industrial and Information Engineering, University of Trieste, Trieste/Italy
2. Institute of Photonic Technology, Jena/Germany
3. Institute of Physical Chemistry and Abbe Center of Photonics, University Jena/Germany
?
Contact author: [email protected]

Keywords: spectroscopy, soft classification, validation, brain tumour diagnosis

Medical diagnosis of cells and tissues is an important aim in biospectroscopy. The data analytical task
involved frequently is classification. Classification traditionally assumes both reference and prediction to
be hard, i. e. stating exactly one of the defined classes. In reality, the reference diagnoses may suffer from
substantial uncertainty, or the sample can comprise a mixture of the underlying classes, e.g. if sample
heterogeneity is not resolved or if the sample is actually undergoing a transition from one class to another
(e. g. rather continuous de-differentiation of tumour tissues). Such samples may be labelled with partial or
soft class memberships.
Many classification methods produce soft output, e. g. posterior probabilities. Methods like logistic regres-
sion can also use soft training data. Yet, for medical diagnostic applications it is even more important to
include soft samples into the model validation. Excluding ambiguous samples means retaining only clear
(i. e. easy) cases. Such a test set is not representative of the original unfiltered population, and creates a risk
of obtaining overly optimistic estimates of the model performance.
With softclassval (softclassval.r-forge.r-project.org), we introduce a framework to calculate commonly used
classifier performance measures like sensitivity and specificity also for samples with soft reference and
prediction. Briefly, if the soft class labels are interpreted as uncertainty, best and worst case as well as
expected performance are obtained via the weak, strong and product conjunction (and-operators, see e. g.
Gottwald, 2010). For the mixture interpretation, weighted versions of well-known regression performance
measures like mean absolute and root mean squared errors are derived.
As real world example, we classify 37 015 Raman (thereof 55 % soft) spectra of 80 brain tumour patients into
“normal”, “low grade”, and “high grade” tissue morphologies in order to delineate excision borders during
surgical treatment of the tumours. Thus, borderline cases are our actual target samples. We demonstrate
spectroscopy-related functionality supplied by hyperSpec (hyperspec.r-forge.r-project.org) and its conjoint
use with other packages.
Financial support by the Associazione per i Bambini Chirurgici del Burlo (IRCCS Burlo Garofolo Tri-
este) and of the European Union via the Europäischer Fonds für Regionale Entwicklung (EFRE) and the
“Thüringer Ministerium für Bildung, Wissenschaft und Kultur” (Project: B714-07037) is highly acknowl-
edged.

References
Gottwald, S. (2010). Many-valued logic. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy
(Spring 2010 ed.).
Kaleidoscope Ia, Tuesday 09:50-10:50 7

Revisiting Multi-Subject Random Effects in fMRI

Jonathan Rosenblatt1
1. Raymond and Beverly Sackler Faculty of Exact Sciences. Tel Aviv University, Tel-Aviv, Israel
?
Contact author: [email protected]

Keywords: fMRI, Finite Mixture, EM

Introduction Random Effects analysis has been introduced into fRMI research in order to generalize
findings from the study group to the whole population (see Friston et al.2002). Generalizing findings is
obviously harder then detecting activation in the study group since in order to be significant, an activation
has to be larger then the inter-subject variability which is assessed and controlled for within the study group.
Indeed, detected regions are smaller when using random effect analysis versus fixed effects. The statistical
assumptions behind the classic random effects model are that the effect is normally distributed over and
“activation” refers to a non-null mean effect. We argue this model is unrealistic and conservative compared
to the true population variability.

Method Inspired by the Two-Populations model widely used in genetics (Efron2008) and in fMRI (Hartvig and Jensen2000),
we develop a model for the inter-subject voxel-wise effect as a mixture of two populations: One in which
the voxel was activated by the paradigm with a normally distributed effect and a second, in which the voxel
is inactive, thus- a null effect. We suggest two justifications for this model: (a) Ill registration might map
active and non-active voxel to the same locations. (b) Brain plasticity permits the same anatomical location
to serve different functions.
The implementation of the method was done in R and included: (a) ML estimation of the parameters of the
proposed finite mixture using an EM algorithm over a non-convex constrained parameter space (to solve
identifiability issues). (b) Bootstrapping of the estimates’ distribution for inference on the number of mixed
populations, known to be an analytically unsolved problem (Garel2007) . (c) Optional parallelization of the
Bootstrapping using Condor and/or the snow package.

Results We demonstrate our method on real fMRI dataset of 67 subjects at 60,000 brain locations. We
construct estimate maps and p-value maps of the voxel-wise proportion of population responding to an
experimental paradigm. Once these have been constructed, we define “activation” as locations where more
then a given percentage of the population has been found active. We revisit the activation maps created
under the classical definition of activation and compare them to activation found under this new definition
to show the power gained using the finite Gaussian mixture.

References
[Efron2008] Efron, B. 2008. Microarrays, empirical Bayes and the two-groups model. Statistical science
23 (1): 1–22.
[Friston et al.2002] Friston, K. J., D. E. Glaser, R. N. A. Henson, S. Kiebel, C. Phillips, and J. Ashburner.
2002. Classical and Bayesian Inference in Neuroimaging: Applications. NeuroImage 16 (2): 484–512
(June).
[Garel2007] Garel, Bernard. 2007. Recent asymptotic results in testing for mixtures. Computational
Statistics & Data Analysis 51 (11): 5295–5304 (July).
[Hartvig and Jensen2000] Hartvig, N.V., and J. L. Jensen. 2000. Spatial mixture modeling of fMRI data.
Human Brain Mapping 11 (4): 233–248.
Kaleidoscope Ia, Tuesday 09:50-10:50 8

Putting the R into Randomisation

⋆
Zoe Hoare 1,
1. NWORTH, Bangor trials unit, Bangor University, Y Wern, LL57 2PZ http://www.bangor.ac.uk/imscar/nworth/
⋆
Contact author: [email protected]

Keywords: Randomisation, Clinical trials, Simulation, Validation

When introducing a new dynamic adaptive randomisation method to the clinical trials field the functionality
and flexibility of R allowed the programming of the algorithm without any restrictions. Whilst using R for
implementing randomisation procedures is not a new idea, expanding its use to include a modelling stage
for optimisation of individual trials was a particularly useful addition.
The setting up of the novel randomisation procedure in R, linking it to Excel using RExcel for simulation
purposes, validation of the processes and the possibilities for linking to the web for web based centralized
randomisation processes will all be addressed.
The randomisation algorithm used is fully tuneable by setting model parameters, reproducing anything
between deterministic allocation (minimisation) and simple randomisation. Simulation with R allowed in-
vestigation the robustness of the algorithm, the statistical properties of bias and loss in terms of a clinical
trial and the plausible outcomes of a wide variety of possible situations that may arise. The algorithm can
now be tuned before the trial starts to give confidence intervals around the split of the final allocation.
Inputs to the simulations allow trials to be customised for differing number of participants, number of
treatment groups, ratios of allocation to treatment groups, number of stratification variables and their defined
levels and sets of parameters to be tested. The algorithm is currently used in two ways: for sequential and
for complete list randomisation. While the base code is the same the wrapper for the implementation varies
slightly.
The use of R in clinical trials is slowly becoming more accepted and has certainly been discussed at length
on the R help mailing lists. Validation of any system is integral to ensuring the system is functioning
correctly. There have been instances where systems that have not been properly validated have resulted in
significant costs within the clinical trial world. However, mistakes like this are not a reason to avoid using
more complex methods altogether. All the code written within R was easily tested and validated within an
expected operating range ensuring that the system is doing exactly what is expected.
Future work entails linking into web based system, developing reporting functionality of the simulation tool
and improve the robustness of the tool for multiple users.

References
Baier, T. and E. Neuwirth (2007, April). Excel :: Com :: R. Computational Statistics 22(1), 91–108.

Group, C. S. (2001). Effect of low dose mobile versus traditional epidural techniques on mode of delivery :
a randomised controlled trial. The Lancet 358, 19–23.

Henry, S., D. Wood, and B. Narasimhan (2009). Subject Randomization System. In useR! 2009, The R User
Conference (Rennes France), pp. 85.

Hewitt, C. and D. Torgerson (2006). Is restricted randomisation necessary? BMJ 332, 1506–1508.

Russell, D., Z. Hoare, R. Whitaker, C. W. C, and I. Russell (2011). Generalized method for adaptive ran-
domisation in clinical trials. Statistics in Medicine, Early view published online DOI 10.1002/sim.4175.
Kaleidoscope Ib, Tuesday 09:50-10:50 9

Bringing the power of complex data analysis methods into R

?
Teh Amouh1, , Benot Macq2 , Monique Noirhomme-Fraiture1
1. School of Computer Science, University of Namur, Belgium
2. School of Engineering, University of Louvain-la-neuve, Belgium
?
Contact author: [email protected]

Keywords: Set-valued data, Modal data, Histogram data, Multidimensional data

In classical statistical data analysis, individuals are described by single-valued scalar type variables. A
rectangular data matrix defines the relation between a set of individuals and a series of variables such that
each cell of the data matrix contains one single scalar value (quantitative or categorical).
Sometimes, however, individuals require to be described by set-valued type variables. For example, in a
time budget study, a variable that records the daily time spent watching television would not allow single-
valued answer (such as 3 hours), because this value normally varies from day to day for each individual. An
interval-valued answer like ”between 2.5 and 3.5 hours”, reflecting an internal variability, would be more
appropriate. When considering a stock market, information on an unpredictable share price would include
the degree of uncertainty. A possible statement would be: ”the price of share S varies between 130 and 140,
with probability 30%”, leading to data expressed as an histogram for variable price and individual S. The
classical single-valued cell data table can hardly cope with such complex descriptions. There is a need on a
data table model in which each cell could contain a (weighted) listing of values taken from a predefined set,
and not just a single quantitative or categorical value. The data.frame model provided by R does not
apply.
A research field named symbolic data analysis (Bock and Diday (2000)) and defined as the extension of
standard data analysis to complex data, proposes a great deal of methods for set-valued data analysis (Diday
and Noirhomme-Fraiture (2008)). These data are called symbolic data and encompass interval type data
(which means subsets of the set of real values), multi-valued categorical type data (which means listings of
ordinal or nominal categories) and multi-valued quantitative type data (which means listings of numerical
values). A listing of categories can be summarized as a weighted set of distinct categories or as a discrete
probability distribution. In either case we talk about modal data. A listing of numerical values can be
summarized as an interval with the lower and upper bounds being respectively the minimum and maximum
values in the listing. A listing of numerical values can also be summarized or as a histogram (if the range
of values in the listing is segmented into intervals) or as a cumulative density function. A symbolic data
table is a rectangular data matrix which allows such complex data values in each of its cells. Powerful data
analysis methods are available for these types of data.
In order to bring the power of set-valued data analysis methods into R, we develop appropriate data structures
using both S3 and S4 object approaches available in R (Chambers (2008)). Our data structures include table
objects that extend the data.frame object and allow complex data values in each cell. This talk is about
the design and implementation of these data structures. An R package containing these data structures will
be available as a basic building bloc for the R implementation of complex data analysis methods.

References
Bock, H.-H. and E. Diday (Eds.) (2000). Analysis of Symbolic Data: Exploratory Methods for Extracting
Statistical Information from Complex Data. Berlin: Springer-Verlag.
Chambers, J. (2008). Software for Data Analysis: Programming with R. Springer.
Diday, E. and M. Noirhomme-Fraiture (Eds.) (2008). Symbolic Data Analysis and the SODAS Software.
Wiley.
Kaleidoscope Ib, Tuesday 09:50-10:50 10

Using the Google Visualisation API with R

Markus Gesman, Diego de Castillo
Contact authors: [email protected]
Project site: http://code.google.com/p/google-motion-charts-with-r/

Keywords: Google Visualisation API, R.rsp, RApache, JSON

In 2006 Hans Rosling gave an inspiring talk at TED (Rosling, 2006) about social and economic develop-
ments in the world over the last 50 years, which challenged the views and perceptions of many listeners.
Rosling had used extensive data analysis to reach his conclusions. To visualise his talk, he and his team at
Gapminder (Gapminder Foundation, 2010) had developed animated bubble charts, aka motion charts.
Rosling’s presentation popularised the idea and use of interactive charts, and as a result the software behind
Gapminder was bought by Google and integrated as motion charts into their Visualisation API (Google Inc.,
2010h) one year later.
In 2010 Sebastián Pérez Saaibi (Saaibi, 2010) presented at the R/Rmetrics Workshop on Computational
Finance and Financial Engineering the idea to link Google motion charts with R using the R.rsp pack-
age (Bengtsson, 2009).
Inspired by those talks and the desire to use interactive data visualisation tools to foster the dialogue between
data analysts and others the authors of this article started the development of the googleVis package (Ges-
mann and de Castillo, 2011).
The googleVis package provides an interface between R and the Google Visualisation API. The Google
Visualisation API offers interactive charts which can be embedded into web pages. With the googleVis
package users can create easily web pages with interactive charts based on R data frames and display them
either via the R HTTP help server locally or within their own sites. The current version 0.2.4 of the package
provides interfaces to Motion Charts, Annotated Time Lines, Maps, Geo Maps, Tables and Tree Maps.
This session will provide an overview of the package functionality and the authors will share examples and
experiences with the audience.

References
H. Bengtsson. R.rsp: R server pages. http://CRAN.R-project.org/package=R.rsp, 2009. R
package version 0.4.0.

Gapminder Foundation. Gapminder. http://www.gapminder.org, 2010.

M. Gesmann and D. de Castillo. googleVis: Using the Google Visualisation API with R. http://code.
google.com/p/google-motions-chart-with-r/, 2011. R package version 0.2.4.

Google Inc. Google Visualization API. http://code.google.com/apis/visualization/

documentation/gallery.html, 2010h.

H. Rosling. TED Talk: Hans Rosling shows the best stats you’ve ever seen. http://www.ted.com/
talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen.html, 2006.

S. P. Saaibi. R/RMETRICS Generator Tool for Google Motion Charts. https://www.rmetrics.org/, 2010.
Meielisalp, Lake Thune Switzerland, June 27 - July 1, 2010.
Kaleidoscope Ib, Tuesday 09:50-10:50 11

Experimenting with a tty connection for R

Matthew S Shotwell
Department of Biostatistics, Vanderbilt University
Contact author: [email protected]

Keywords: terminal, devices, hardware, user interface

The Portable Operating System Interface (POSIX; IEEE, 2004) offers a standard concept of the computer
terminal and an associated Application Programming Interface (API). Operating systems that natively sup-
port the POSIX concept include GNU Linux, BSD, and Mac OS X. Support for Microsoft Windows is pro-
vided by third-party software, such as Cygwin (Red Hat, 2011). The POSIX terminal serves as a common
mechanism for asynchronous data transfer, including keyboard input, USB and other hardware communica-
tions, and software-to-software connections. We present a parsimonious extension to R that implements the
POSIX terminal API as an R connection. The new feature is styled as a ‘tty connection’.
Applications of the tty connection are broad in scope. This presentation highlights applications to the useR
interface, and in hardware communications. We demonstrate how the tty connection facilitates common
user interactions, such as ’press any key to continue’ functionality, and password entry without displaying
the password text. A live demonstration is prepared using R to control an external USB temperature sensor,
as well as a GPS navigation device. We conclude with a discussion of integrated data collection and signal
processing in medical devices.

References

Institute of Electrical & Electronics Engineers, Inc. (2004). IEEE Std 1003.1 (POSIX), 2004 Edition,
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1309816.
Red Hat, Inc. (2011). Cygwin (version 1.7) [Software], http://www.cygwin.com
Kaleidoscope Ic, Tuesday 09:50-10:50 12

The R Ecosystem
David M Smith1,*

1. VP Marketing and Community, Revolution Analytics

*Contact author: david@revolutionanalytics

Keywords: community, users, ecosystem

In the last couple of years, and especially since the New York Times feature article about R1, I have
observed a dramatic shift in the perceptions of R. Long gone are the days when it was regarded as a
“niche” statistical computing tool mainly used for research and teaching. Today, R is viewed as a
critical component of any data scientist’s toolbox, and has been adopted as a layer of the analytics
infrastructure at many organizations. As a result, a broad ecosystem has sprung up around open-source
R. In this talk, I’ll share my observations on this evolving ecosystem, including:

• Estimates of the number of R users

• The R Community: people sharing R software and experiences

• How R is being used in academia and in industry

• R’s role in the Data Science movement

• How R is bringing modern Statistics to business applications

The talk will conclude with some thoughts on what the future holds for R’s role in the worlds of
statistics, data science, and business.

References

Vance, Ashlee (2009). “Data Analysts Captivated by R’s Power”. The New York Times, January 6
2009.
Kaleidoscope Ic, Tuesday 09:50-10:50 13

Rc2: R collaboration in the cloud

?
E. James Harner1 and Mark Lilback1
1. Department of Statistics, West Virginia University
?
Contact author: [email protected]

Keywords: R, cloud computing, collaboration, web 2.0

Rc2 is a cloud-based, collaborative, web 2.0 interface to R that works with WebKit-based browsers (e.g.,
Safari and Chrome). In addition to desktop browsers, client-specific style sheets and scripting provide a
touch-optimized interface for mobile devices such as the iPad. With Rc2 , R sessions are no longer tied to a
specific computer or user.
Rc2 is designed for simplicity and ease-of-use from the start, allowing students to learn R without solely
imposing a command-line interface. At the same time, power users will find the system flexible enough to
meet most of their needs. Users can have per-project workspaces, and instructors can predefine workspaces
containing the required data and R packages for each assignment.
The groundbreaking features of Rc2 arise from the collaborative and instructional benefits that cloud-based
computing brings. Instructors can schedule classroom sessions where students can watch the instructor
interact with R in real-time. The instructor can turn control of R over to individual students as a virtual
blackboard to pose questions and work through problems while still communicating via voice chat. Integra-
tion with the social web allows instructors to provide notifications to students over the messaging platform
of their choice, be it email, SMS, Twitter, or Facebook.
The same features allows researchers to collaborate over the Internet without concern for data becoming
out of sync. Users can start long-running computations and Rc2 will notify the user(s) when the process is
complete. Full support for Sweave allows users to easily include, update, and format R output within LATEX
documents for both classroom assignments and publishable papers.
Rc2 is designed to be a fast, scalable, distributed system with features like load-balancing, multiple authen-
tication mechanisms (password, LDAP, Kerberos), dynamic auto-configuration using mDNS and DNS-SD,
fast and persistent client-server communications using WebSockets, and custom R packages for database
interaction and graphics generation. Rc2 can start on a single server and expand to a cluster of computers
optimized for each tier of the system (web server, application server, R instances, and database).
Kaleidoscope Ic, Tuesday 09:50-10:50 14

RStudio: Integrated Development Environment for R

?
J.J. Allaire1
1. CEO of RStudio
?
Contact author: [email protected]

Keywords: GUI, Tools, Sweave, Web

RStudio is a new Integrated Development Environment (IDE) for the R programming language. RStudio is
an open-source project intended to combine the various components of R (console, source editing, graphics,
history, help, etc.) into one seamless and productive workbench. It is designed to both ease the learning
curve for new R users as well as provide high productivity tools for more advanced users. RStudio runs
on all major platforms including Windows, Mac OS X, and Linux. In addition to the desktop application,
RStudio can be deployed as a server to enable web access to R sessions running on remote systems.

RStudio running on Windows

For the useR! 2011 conference, we would like to introduce RStudio as a new open-source IDE for the R
community. We will cover its various features in depth as well as tips on how to optimize workflow. We
will cover the basics (console, editing, plots, etc) and also discuss more advanced features including code
completion, searchable history and authoring LATEX and Sweave documents.

The presentation will also cover the details of setting up an RStudio server. We will show a hands on demon-
stration of installing RStudio on an Ubuntu Server as well as describe the various configuration options and
administrative utilities.

NOTE: Currently, RStudio is not publicly available (it is in beta) however will be released well in advance
of the conference. Our website http://www.rstudio.org/ will also be live at this time. For the purposes of
considering this presentation, please contact us to receive access to the website.
Portfolio Management, Tuesday 11:15-12:35 15
R in the Practice of Risk Management Today
Jagrata Minardi1,*

1. TIBCO Software, Inc.

*Contact author: [email protected]

Keywords: financial risk management, portfolio management, simulation

This talk explores several concrete computational tasks in risk management and portfolio
management, in the true scale at which they occur in large financial institutions. Using R and
well-known packages, there is strong coverage of analytical methods, but challenges remain as
size of problems grow and answers are required more quickly.

How does today's software improve our ability to manage risk in the time needed by business?
When models change, how quickly can they be brought to market? How has recent developments
in both regulatory bodies and the competitive landscape changed the business problems
themselves? We'll identify the state of the art against several key challenges in financial risk
management, and indicate how software---and the industry itself---are changing the playing field.
Portfolio Management, Tuesday 11:15-12:35 16

Stress Testing with R-Adamant

?
Rocco Claudio Cannizzaro1, , Fausto Molinari2
1. R-Adamant
2. R-Adamant
?
Contact author: [email protected]

Keywords: R-Adamant, Stress Testing, VaR, Expected-Shortfall

Value-at-Risk (VaR) and Expected-Shortfall are standard and widely used risk-management tools.
The calculation of such metrics is generally based either on some a-priori assumption on the statistical dis-
tribution of the time series or on historical data analysis and simulations.

Even when assumptions are made on the form of the distribution, the estimation of the related parameters is
typically based on historical data and the choice of how much history should be used is a non trivial prob-
lem: short data history produces noisy estimates while long historical analysis results in biased estimations
which may not be reflective of the changing nature of the markets and their volatilities.

An alternative approach is to exploit the correlation between the performance of the managed portfolios and
the macro economic conditions in which the underlying sectors of the investment are operating.
Once the link is established, it is possible to evaluate the impact of macro economical changes on the perfor-
mance of the given investments and generate a multitude of possible forecast scenario outcomes from which
the risk metrics (VaR, ES) can be computed.

Stress scenarios/events can also be analysed and the associated risks quantified, leading to a more dynamic
and consistent assessment of the financial position of the investments.
During the presentation we will demonstrate how to exploit R-Adamant to simulate several possible sce-
narios for the economy and the market, and attempt to forecast financial trend and returns of a portfolio in
each of those scenarios.
We will use the statistical and graphical tools contained in R-Adamant, like: Efficient Portfolio estimation,
VaR and ES estimation, Vector Autoregressive models, and Monte-Carlo simulations.
The tutorial scope is to show how a valuable tool as R-Adamant can be a powerful ally for researchers and
university students for their thesis and for companies.

References

Johnston J. (1984), Econometric Methods - third edition, McGraw-Hill

Harvey A.C. (1993), Time Series Models 2nd Edition, The MIT Press.

Markowitz, Harry M. (1991), Portfolio Selection, second edition, Blackwell.

Steven M. Kay (1993), Fundamentals of Statistical Signal Processing, Volume 2: Detection Theory.
Portfolio Management, Tuesday 11:15-12:35 17

Handling Multiple Time Scales in Hedge Fund Index Products

Jovan Njegic1,*, Drago Indjic2,3

1. Business School of Novi Sad, Serbia

2. Sunningdale Capital LLP, UK
3. London Business School, UK
*Contact author: [email protected]

Keywords: hedge fund data sources, time series, risk management, R functions

The analysis of portfolio of hedge funds suffers from numerous data quality problems. In particular case
of hybrid fund structure investing across the full universe of hedge fund products in marketplace with
contractual liquidity ranging from intra-daily to annual, we have to handle time series with varying
periodicity and possibly multiple price status (estimated, executed, audited etc).

We have been using virtually all commonly known R time series packages libraries (such as xts, zoo) to
clean and „synchronize‟ the time series in order to provide standard risk reporting and risk limit
management functionality, exposing several design flaws and fixing minor bugs (e.g. to.weekly,
Return.portfolio) and attempting to develop a new wrapper function specifically addressing
hedge fund data sources and features.

This work is based on “R in Modern Portfolio of Hedge Fund Analysis” presentation given at London R
group meeting in December 2010.

References

Indjic, D., (2010), R in Modern Portfolio of Hedge Fund Analysis, http://www.londonr.org/LondonR-

20090331/R%20in%20modern%20FoHF.pdf
Bioinformatics and High-throughput Data, Tuesday 11:15-12:35 18

AFLP: generating objective and repeatable genetic data.

?
Thierry Onkelinx 1, , Pieter Verschelde1 , Sabrina Neyrinck1 , Gerrit Genouw1 , Paul Quataert1
1. Research Institute for Nature and Forest (INBO), Scientific Services, Gaverstraat 4, B-9500 Geraardsbergen, Belgium
?
Contact author: [email protected]

Keywords: Genetics, normalisation, classification, repeatability, randomisation

Amplified fragment length polymorphisms (AFLP) is a technique to get a DNA fingerprint from an or-
ganism. The total genomic DNA is split into fragments at fixed combinations of nucleotides, resulting in
fragments of different lengths. After electropherese we get a DNA profile (“bands” or “peaks”) of intensity
versus fragment length. High intensity indicates the presence of fragment(s), low intensity the absence.
Our package AFLP, available on R-Forge1 , supports the analysis of AFLP data on several points: i) ran-
domise the samples and add replicates, ii) normalise the intensity for differences due to laboratory handlings
and equipment, iii) classify the normalised intensity into presence/absence data, iv) calculate the repeatabil-
ity of the analysis and v) apply multivariate analysis on the classified data.
Replicates are added during the randomisation of the samples. This allows to measure the repeatability and
helps the normalisation between batches. Therefore both within batch as between batch replicates are added.
Bonin et al. (2004) suggest a “technical difference rate” (TDR) to estimate the repeatablity of two replicates
from the same sample. We adapted this TDR formula to work with more than two replicates per sample.
Furthermore we use this TDR not only to measure the repeatability of the samples, but also the fragments
and the overall repeatability.
The normalisation is based on linear mixed models (lme4). The idea is to model the average intensity in
function of the different sources of variability (DNA extraction, lab batch, capillary, . . . ). The raw residuals
of the model are used as the normalised intensity. Several diagnostic plots are available to highlight potential
problems. The normalised intensity is classified based on their distribution per fragment. The distribution
is bimodal for a polymorph fragment (both presences and absences) and unimodal for a monomorph frag-
ment (only presences or only absences). The lowest density between the modi is used as threshold for the
classification. The user can get graphs displaying the distribution and the selected threshold. The build-in
analysis capabilities are currently limited to methods for hclust and princomp. Both the normalised
and classified data can be exported to other packages when needed.
Advantages of this package: i) The package starts from the design in the lab, ensuring the necessary repli-
cates and randomisation. ii) The normalisation considers a much wider variety of sources than other soft-
ware. iii) The intensity, typically lognormal distributed, can be transformed during the normalisation. iv)
The classification threshold depends not an arbitrary threshold, but on the distribution of the normalised
intensity. v) TDR is used as a measure of overall quality, quality per sample and quality per fragment.

References
Bonin, A., E. Bellemain, P. Bronken Eidessen, F. Pompanon, C. Borchmann, and P. Taberlet (2004). How
to track and assess genotyping errors in population genetic studies. Molecular Ecology 13, 3261–3273.

1
http://r-forge.r-project.org/projects/aflp/
Bioinformatics and High-throughput Data, Tuesday 11:15-12:35 19

The nz.seq package for handling genetic sequences

in the Netezza Performance Server

?
Przemysław Biecek1,2, , Paweł Chudzian1,3 , Peter Gee1 , Jeff Foran1 , Justin Lindsey1

1. NzLabs, Netezza, an IBM company,

2. Faculty of Mathematics, Informatics, and Mechanics, University of Warsaw,
3. Faculty of Electronics and Information Technology, Warsaw University of Technology
?
Contact author: [email protected]

Keywords: genetic sequences, parallel processing, relational database, next generation sequencing.

In the first part of this talk we will show why it is advantageous to store genetic sequences in a relational
database instead of flat files. In the second part we will present the R interface supporting massively parallel
operations on genetic sequences which are stored in the relational database. The presented implementation
of nz.seq package is specific to the Netezza Performance Server (NPS) database but the presented approach
is general and might be easily applied to any other database with a similar architecture, i.e. the parallel share
nothing environment.
DNA, RNA and amino acid sequences are usually stored in flat files instead of relational databases. The
three main reasons for that are: it is easier to copy or download flat files, it is easier to load flat files into
a statistical packages like R and it is easier to use command line tools like SAMtools, bowtie or bwa. On the
other hand storing data in a database allows one to incorporate the data structure, and allows for fast access
to required sequences based on their properties or metadata information. Note that the volume of genetic
data may be huge, dozens or thousands of terabytes, thus the access time to a required sequence may be
significant. Moreover, databases support efficient merging of different kinds of data, e.g. genetic sequences
might be combined with other features like sex, age, medical history, geographical coordinates etc.
In the NPS one can take advantages of both approaches. Sequences are stored in the relational database, but
some operations might be performed in all computing nodes transparently to the internal database storage
model.
The package nz.seq leverages the communication with the database in two ways:
• From the R client one can easily download any sequence or set of sequences from the remote database
using their names, IDs or some matching criteria. Afterwards in the local R client one can operate on
such sequences and store results from local operations in the remote database.
• Instead of downloading the data from the remote database to local R client, one can upload the R code
from the local R client to the remote database. The submitted R script is then applied to all selected
sequences in a fully parallel way. That decreases the overall computation time by reducing the data
download time, which can be a significant improvement due to the large volume of data. Most of the
R functions from the R packages available on CRAN or Bioconductor might be used in such remote
operations. One can easily apply a variety of R operations to a large number of sequences stored in
the remote database.
At the end of this talk we will present a real life application using data from The 1000 Genomes Project.
Bioinformatics and High-throughput Data, Tuesday 11:15-12:35 20

Classification of Coverage Patterns

Stefanie Tauber 1,? , Fritz Sedlazeck1 , Lanay Tierney2 , Karl Kuchler2 , Arndt von Haeseler1
1. Center for Integrative Bioinformatics Vienna, Max F Perutz Laboratories, University of Vienna, Medical University of Vienna,
University of Veterinary Medicine, Dr.-Bohr-Gasse 9, A-1030 Vienna, Austria
2. Christian Doppler Laboratory Infection Biology, Max F Perutz Laboratories, Medical University of Vienna, Campus Vienna
Biocenter, Dr.-Bohr-Gasse 9, A-1030 Vienna, Austria
?
Contact author: [email protected]

Keywords: Next Generation Sequencing Data, Coverage, Fractals

The advent of DNA sequencing technologies [Metzker (2009)] has brought along an enormous amount of
data that still poses a fundamental data-analysis challenge for bioinformaticians and biostatisticians.
When speaking of sequencing data the term ’coverage’ is widely used but, at the same time, not well defined.
It has to be distinguished between theoretical (’sequencing depth’) and observed (’local’) coverage. The
local coverage can be defined as an integer vector counting per nucleotide the number of reads mapping
to the respective nucleotide. In the following the term ’coverage’ always refers to the observed local per
nucleotide coverage.
In genome resequencing we expect and aim for uniform coverage whereas technologies like RNA-Seq [Oz-
solak and Milos (2010)] or ChIP-Seq [Park (2009)] are especially interested in coverage jumps. However,
any kind of differential expression analysis relies on a count table containing the number of mapped reads
per gene model. This summarization step is not well investigated and its implications on the downstream
analysis are not fully understood yet. It is obvious that a summarization value like the sum of reads per gene
model is not able to exhaustively capture the underlying coverage information.
Therefore we introduce the fractal dimension (FD) [Kaplan and Glass (1995)] and the Hurst exponent (H)
[Peitgen et al. (1992)] in order to distinguish between more or less ’reliable’ coverage patterns. FD, as
well as H do not make use of any user-defined parameters and are hence free of any ad-hoc heuristics. We
propose a re-weighting of the read counts with both FD and H yielding a more reliable count table.
Additionally we show the influence of different mapping strategies on the observed coverage patterns and
read counts. This is of course of special interest as any mapping peculiarity propagate to all downstream
analysis. We discuss our results on a large Illumina RNA-Seq data set. The host-pathogen interaction of
Candida albicans and dendritic mouse cells are investigated by a time course design with three replicates
per time point.
We illustrate the entire analysis as well as all up-mentioned methods by means of a R package we are
developing.

References
Kaplan, D. and L. Glass (1995). Understanding Nonlinear Dynamics. New York: Springer.
Metzker, M. L. (2009, December). Sequencing technologies the next generation. Nature Reviews Genet-
ics 11(1), 31–46.
Ozsolak, F. and P. M. Milos (2010, December). RNA sequencing: advances, challenges and opportunities.
Nature reviews. Genetics 12(February).
Park, P. J. (2009, October). ChIP-seq: advantages and challenges of a maturing technology. Nature reviews.
Genetics 10(10), 669–80.
Peitgen, H.-O., H. Jürgens, and D. Saupe (1992). Chaos and Fractals. New York: Springer.
Bioinformatics and High-throughput Data, Tuesday 11:15-12:35 21

Finite Mixture Model Clustering of SNP Data

?
Norma Coffey1, , John Hinde1 , Augusto Franco Garcia2
1.National University of Ireland, Galway
2. Department of Genetics, ESALQ/USP, Piracicaba, Brazil
?
Contact author: [email protected]

Keywords: Clustering, finite mixture models, orthogonal regression, SNPs.

Sugarcane is polypoid, i.e. has 8 to 14 copies of every chromosome, with individual alleles in varying
numbers. It is therefore important to develop methods that identify the many different alleles and associated
genotypes. One way of doing this is through the analysis of single nucleotide polymorphisms (SNPs). The
frequency of a SNP base at a gene locus will vary depending on the number of alleles of the gene containing
the SNP locus. Capturing this information accurately across several SNPs can give an indication as to the
number of allele haplotypes present for a gene. Such information could have implications for sugarcane
breeding since high yield potential may be due to the presence of and/or different number of copies of, a
specific allele(s) present at a gene locus. The figures below display the data collected for two SNPs of the
sugarcane plant. Each point represents the intensity of two SNP bases; h.L is the intensity of the A base,
h.H is the intensity of the T base. The data in Figure A can clearly be clustered into two groups - the group
along the y-axis and the group along the line with a particular (unknown) angle. These groups correspond to
two genotypes and thus clustering is essential for genotyping. In Figure B it is not clear how many clusters
(genotypes) are present and therefore it is necessary to develop a technique that can determine the number
of clusters present, determine the angles between the clusters to identify different genotypes, and provide a
probabilistic clustering to identify points that have high probability of belonging to a particular cluster (have
a particular genotype) and those that are regarded as an unclear genotype.
A B
80

● ● ●
50

●
●
● ●
●
● ● ●
●
● ●
●
● ● ●
●
60

40

●
● ●● ● ●● ● ●
●
●
● ●
●
●
● ● ● ● ●
●
●
● ● ●●● ●
●
● ● ● ● ●
●
● ● ● ●
●
● ● ●● ● ● ●
●
● ●● ● ●
●
●● ● ● ●
● ● ● ● ●
● ● ●
30

●
● ● ●● ● ● ● ● ●● ●● ●
● ● ● ● ●
●
h.H

h.H

● ● ●
●● ●● ●● ●
40

● ●
●
● ● ● ●● ● ● ●
●● ● ●
● ●● ●
● ● ● ● ● ● ● ●●● ●
● ● ● ● ● ●● ●
● ● ●
● ● ●
● ●● ●● ● ●
20

● ●● ● ● ● ●
● ● ●● ● ● ●● ●
●
●
● ● ● ●●
●
●●● ● ● ● ●● ●
● ● ●● ● ● ● ●
● ● ● ● ●● ●● ● ●
● ●● ● ● ● ●
20

● ● ●
●● ● ● ● ●
●● ● ●
● ●● ●● ●● ● ●
● ● ● ● ● ●
●● ● ● ● ●● ● ●●●● ●● ● ● ●
10

● ● ●
● ● ● ●● ● ●● ● ●
● ●
●
●●● ●
●● ●● ● ●
● ● ●
● ● ●● ●●●
● ●
● ●
● ●
● ●
●●●● ●●● ● ●
●
● ●●
0

0 5 10 15 0 5 10 15 20 25

h.L h.L

The above criteria indicate that model-based cluster analysis techniques could be useful for analysing these
data. However standard model-based cluster analysis techniques such as those implemented in the R package
mclust (Fraley and Raftery, 2002) attempt to fit sperical/ellipsoildal components thus failing to cluster these
data in an appropriate way and do not provide estimates of the angles between the clusters. To determine
these angles it is necessary to fit a regression line to the data in each cluster. Using finite mixtures of linear
regression lines is also inappropriate since it is not clear for these data which is the response variable and
which is the explanatory variable. Problems are also encountered when attempting to fit a regression line to
the group parallel to the y-axis in Figure A since this line has infinite slope in the usual least squares setting.
As a result we propose to use finite mixtures of orthogonal regression lines to cluster the data, which ensures
that using either variable as the response variable yields the same clustering results and that a regression line
can be fitted to the group parallel to the y-axis. We implement this technique in R and show its usefulness
in clustering these data.

References
Fraley, C. and A. Raftery (2002). Model-based clustering, discriminant analysis, and density estimation.
Journal of the American Statistical Association 97, 611–631.
High Performance Computing, Tuesday 11:15-12:35 22

GPU computing and R

?
Willem Ligtenberg1,
1. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: R, GPU, OpenCL, High-performance computing

Modern CPUs currently have 4 to 8 cores. In contrast an affordable GPU ( 200 euro) has 384 cores. This
makes them ideally suited for parallel processing. Examples of problems that can benefit from processing
on the GPU are Fast Fourier Transforms, the k-nearest neighbour algorithm or sequence alignment. If the
problem can be cut into small pieces, it might be interesting to port it to the GPU.
In order to make this compute power more accessible there a few options. One is to make some of the basic
R functionality use the GPU directly, a second option is to wrap C(++) applications that make use of the
GPU with R. Finally, we can build an interface between R and OpenCL, which leaves all possibilities open
for the user and therefore provides maximum flexibility.
In the first part of our presentation we will provide an overview of current R packages that use the GPU. In
the second part, we will present our progress on the ROpenCL package. ROpenCL provides functions to
transfer data frames to and from the GPU and to push kernel functions to the GPU, thereby unlocking the
full potential of the GPU. These kernel functions still need to be written in OpenCL, however the interfacing
with the GPU can all be done with R. The ROpenCL package is similar in spirit and based on the PyOpenCL
Python library Klöckner (Klöckner).

References
Klöckner, A. PyOpenCL. http://mathema.tician.de/software/pyopencl.
High Performance Computing, Tuesday 11:15-12:35 23

Deploying and Benchmarking R on a Large Shared Memory System

Pragneshkumar Patel1 , George Ostrouchov1,2, Luke Tierney3

1. University of Tennessee
2. Oak Ridge National Laboratory
3. University of Iowa
*Contact author: [email protected]

Keywords: high performance computing, parallel computing, multicore architecture, multithreading

We describe our experience in deploying R on a large shared memory system, Nautilus (SGI UltraViolet), which
has 1024 cores (Intel Nehalem EX processors), 4 terabytes of global shared memory, and 8 NVIDIA Tesla GPUs
in a single system image. This system is part of RDAV, the University of Tennessee's Center for Remote Data
Analysis and Visualization sponsored by the National Science Foundation as part of TeraGrid XD.

One of our goals on Nautilus is to provide implicit parallel computing for serial R codes. Implicit parallelism
automatically exploits multiple cores without changes to user code. Another goal is to provide an environment
where users can effectively explore explicit parallelism through the use of many R packages that have been
developed recently for parallel computing.

For implicit parallelism, we report on Intel’s Math Kernel Library (MKL) and on the pnmath package. MKL
includes a high performance multithreaded BLAS implementation. The pnmath package provides multithreading
to many R math functions for operating on large vectors. We report on benchmark runs for many core count and
vector size combinations, which we use for optimal speedup calibration. We observe speedups in excess of 600
for some compute intensive pnmath functions when operating on large vectors.

In addition to reporting on specific BLAS and pnmath functions, we include R Benchmark 2.5 and some of its
modifications. If time permits, we also mention experiences with RDAV center customers and some packages for
exploiting explicit parallelism.

References

Luke Tierney (2008). Implicit and Explicit Parallel Computing in R, COMPSTAT 2008, Part II, 43-51.

Markus Schmidberger, Martin Morgan, Dirk Eddelbuettel, Hao Yu, Luke Tierney, Ulrich Mansmann (2009).
State of the Art in Parallel Computing with R. Journal of Statistical Software, August 2009, Volume 31, Issue 1.

Luke Tierney (2010). pnmath, http://www.stat.uiowa.edu/~luke/R/experimental/.

Intel Math Kernel Library,

http://software.intel.com/sites/products/documentation/hpc/composerxe
/en-us/mklxe/mkl_userguide_lnx/index.htm.
High Performance Computing, Tuesday 11:15-12:35 24

The “CUtil” package which enables GPU computation in R

Kazutaka Doi1,*† , Kei Sakabe†

1. Regulatory Science Research Program, National Institute of Radiological Sciences, Japan

†
Both authors contributed equally to this work
*Contact author: [email protected]

Keywords: CUDA, GPU computation, operator overriding, Windows

There are some R packages which uses GPU for computation, and the better performance is provided by
GPU-equipped computers. However, the current number of implemented functions in their packages is
limited because special procedures are required for utilization of GPU, such as memory allocation in
video memory, and data transfer between main memory and video memory. Furthermore, these
packages are not easy to use for Windows users because binary packages are not provided, and it is not
easy for Windows users to build a developing environment. Since there seems to be a large number of
potential Windows users, it is beneficial to develop a package which enables GPU computing easily on
Windows. Therefore, we began to develop a new package which utilize GPU in computation with
minimum modification of R source code, and also can be used easily for Windows users. The package
development is ongoing, and the detail specifications are not fixed, except that the package is developed
with NVIDIA’s CUDA toolkit. Our package CUtil (CUDA Utility package) will be equipped with the
following features.

As described above, the first feature is that Windows users will be able to use this package easily. This
package will be available from CRAN, and we will also try to make Windows binary packages
available. The second feature is that after loading the package, the frequently used operators and
functions can be overridden. With this overriding, we can enjoy GPU computation performance with
minimum modification of R source code. The third feature is that the package minimize the data
transfer cost between main memory and video memory. The data on video memory is treated as
external pointer in R objects, and once the R object on main memory is transferred to video memory for
GPU computation, the data is kept on video memory as R objects. The next time of the GPU
computation, the data can be used directly by GPU without the data transfer. Because it is known that
the proportion of time needed for the data transfer is relatively high, this advantage will be beneficial
especially for a long series of computations, such as Markov chain Monte Carlo methods in Bayesian
statistics. Other than these features, we are going to implement double precision floating-point and
complex number computation in addition to single floating-point, and garbage-collection, which
enables us to use a small amount of video memory effectively. This package will require computers
with NVIDIA’s GPU (compute capability is equal to or greater than 2.0).

References

Buckner J, Wilson J, Seligman M, Athey B, Watson S, Meng F (2010). The gputools package enables
GPU computing in R. Bioinformatics 26, 134-135.

Adelino Ferreira da Silva (2010). cudaBayesreg: Bayesian computation in CUDA. The R Journal 2, 48-
55.
High Performance Computing, Tuesday 11:15-12:35 25

OBANSoft: integrated software for Bayesian statistics and high

performance computing with R
?
Manuel Quesada1, , Domingo Giménez1 , Asunción Martı́nez2
1. Universidad de Murcia
2. Universidad Miguel Hernández
?
Contact author: [email protected]

Keywords: Bayesian statistics, parallelism, OpenMP, MPI, CUDA

Among all the applications for Bayesian analysis, there is none that integrates the whole process of Objective
Bayesian Analysis. The aim of the “Software for Objective Bayesian Analysis” (OBANSoft) is to cover this
gap. The first version includes the easiest models (Yang and Berger (1996)) but its design will allow for
more complex algorithms in time.
The statistics engine has been implemented using R. Implementing the algorithms with this language lets
us take advantage of the increasing number of new routines that the R community is developing, so we can
focus the effort on other purposes such as high performance computing. For this reason, OBANSoft has been
designed taking into account that it is going to incorporate a variety of parallelism levels, and it will be the
first R parallel in a shared memory architecture. Moreover, the heterogeneity of parallel architectures will
obstruct non advance users in parallelism optimizing the performance of their programs without the help
of an expert. Concerning parallel performance, OBANSoft is ready to include an auto tuning module that
configures the best parameters to execute the parallel algorithms (Katagiri et al. (2004)).
Although the statistics engine is developed in R, this is envolved in a top layer defined like a Java interface
(using JRI to link them). This main layer coordinates the lower layers where the different engines based on
several architectures are implemented. Indeed, this top layer will implement the auto-tuning algorithms. The
final user will only use a desktop application that will solve a problem in the most efficient way for each
platform.
The first version is a complete and integrated Java Desktop Application (OBANSoft) implementing the
first models used to teach Bayesian analysis (Quesada (2010)). All the models are implemented using R
libraries that have been reorganized to compose our R engine. The SnowFall library (SnowFall (2011))
allows exploitation of parallelism in multicore systems. However, the performance of this library is not
satisfactory enough and other parallelism strategies need to be considered.
In conclusion, we have developed the base application where we are going to include more complex models
with higher computational needs. Our concern is to implement and link the R engine with other parallel
languages and high performance libraries (BLAS, OpenMP, MPI, CUDA, etc.). OBANSoft will assume all
those new routines, including auto-tuning decisions. Our goal is to distance the final user from the problems
of parallel computation.

Referencias
Katagiri, T., K. Kise, H. Honda, and T. Yuba (2004). Effect of auto-tuning with user’s knowledge for
numerical software. In Proceedings of the 1st conference on Computing frontiers, pp. 12–25. ACM.
Quesada, M. (2010, Julio). Obansoft: aplicación para el análisis bayesiano objetivo y subjetivo. estudio de
su optimización y paralelización. Master’s thesis, Universidad de Murcia.
SnowFall (2011). Url. http://cran.r-project.org/web/packages/snowfall/.
Yang, R. and J. O. Berger (1996). A catalog on noninformative priors. Discussion Paper, 97-42, ISDS, Duke
University, Durham, NC.
Reporting Technologies and Workflows, Tuesday 11:15-12:35 26

R2wd: Writing MS-Word Documents from R

Christian Ritter1,2 , Nathan Uyttendaele1

1. Université Catholique de Louvain, Louvain-la-Neuve, Belgium.

2. Ritter and Danielson Consulting, Brussels, Belgium.
*Contact author: [email protected]

Keywords: COM services, automatic reporting.

R2wd is a package for creating new and modifying existing MS-Word documents. It allows to outline
the document structure (title, headings, sections), insert text content, R graphs in metafile and bitmap
formats, R dataframes as Word tables, mathematical formulas written in a LaTeX like style, and to apply
Word themes to the document. Bookmarks of all inserted elements are created automatically which can
be useful in building cross references and in creating tagged PDF documents from the original Word
documents. R2wd helps creating documents automatically from R scripts and thereby supports
automatic and traceable reporting. In addition, the code behind the R2wd functions serves as a tutorial
on how to use COM objects under Windows. In this talk, we shall show how R2wd works and point out
which tasks already work well and where there are still challenges. Although R2wd has already been
used in real automatic reporting situation, it is still work-in-progress. We encourage the community of
R users to help us in making R2wd more reliable and powerful.

By default R2wd is based on the rcom package which, in turn, makes use of the statconnDCOM
connector. It can also be adapted to work with the legacy RDCOM package.

References

Thomas Baier and Erich Neuwirth (2007) Excel :: COM :: R, Computational Statistics 22/1, pp. 91-108.
Reporting Technologies and Workflows, Tuesday 11:15-12:35 27

FRAD - Fast Results Analysis and Display

Dmitry Bliznyk1,*

1. Researcher and PhD student at Riga Technical University

*Contact author: [email protected]

Keywords: R, fast analysis, universal usage, template

Paper presents approach for statistical data visualization and analysis. For this purposes the template
based on R package were created. The FRAD - Fast Results Analysis and Display template provides
possibility to make fast data analysis even for inexperienced user without the need of programming. It is
ready for processing ASCII input files (e.g. CSV) and producing timeline, histogram and summary plots
as well as summary table in .csv format. Possible application areas: data quick overview, visual
comparison of data (timeline, histogram), big datasets analysis and results publishing.

FRAD template is not limited to specific task. It’s designed to provide the power of R language for
broad range of users, which are dealing with data analysis. The template contains many features, which
are aimed to reduce time that is needed to get the first data analysis results. Some examples of features:
automatic range selection accordingly to minimal and maximal values in all analyzed files, skipping
corrupted data, automatic axis labeling. One more valuable feature is that proposed approach prevents
“human made mistakes” in analysis, specifically in result files naming. In FRAD result files are named
automatically using input file name and analyzed data type.

User of FRAD template is able to choose from three levels of control according to his needs and
available time. After analyzing big group of user needs, the default parameters where chosen for the
first level of control, which is fully automatic. If user needs more freedom, there is the parameters file
that gives ability to select analysis features and tune visual look of the output plots. And finally user is
able to change template’s R code, to adjust anything.

FRAD template proves proved itself as valuable tool in research work of the “ifak”[1] institute,
specifically in ZESAN[2] project. Previously used Origin[3] software could not satisfy demands in fast
test case data analysis. Since Origin does not support automated analysis of multiple files, it could be
effectively used for test case data analysis, where number of files per test case is more than ten. For one
test case analysis (consisting of four files) user previously should spend around an hour, with FRAD
template it takes just two minutes (at 1 million records per file). Advantage is huge!

References

[1] Ifak - Institut für Automation und Kommunikation e.V. Magdeburg, www.ifak.eu

[2] ZESAN project - Reliable, energy-efficient wireless sensor/actor networks for building automation,
supervision and process control, http://www.ifak.eu/index.php?id=651&L=3

[3] Origin software, http://www.originlab.com/

Reporting Technologies and Workflows, Tuesday 11:15-12:35 28

The Emacs Org-mode: Reproducible Research and Beyond

?
Andreas Leha1, , Tim Beißbarth1
1. Department of Medical Statistics, University Medical Center Göttingen
?
Contact author: [email protected]

Keywords: Literate Programming, Reproducible Research, emacs, Org-mode

Reproducible research and literate programming are widespread techniques in the R community using
the power of the well established Sweave package [Leisch, 2002] and other approaches (e.g. R2HTML
[Lecoutre, 2003], odfWeave [Kuhn et al., 2010], via docutils [Dasgupta, 2010]). The main advantage of all
literate programming efforts is, that (R-)code and results, mainly tables and plots, are kept together. This en-
sures that the reported results originate from the current version of the code, making the report reproducible
and adding other benefits, like simplifying version control for results and code in conjunction.
Org-mode (http://orgmode.org/) can be used to achieve similar results as Sweave. But while in contrast to
Sweave Org-mode is bound to the editor emacs (although there are different ports to vi(m)), it is a far more
flexible approach in other aspects:
The most prominent feature is the variability of output formats: the output can be chosen from LATEX (PDF),
HTML, and soon also ODF.
The language that is woven into the text is not limited to R, but includes Perl, Python, Octave, shell, Matlab,
Lisp, SQL, and many more. This is a key feature, as quite often data analysis requires the use of different
tools on the same data: For example, Perl is frequently used during the data pre-processing step. Likewise,
external programs are often integrated for specialized tasks (e.g. short read alignment) and might best be
called from command line. Source code blocks in different languages can easily interact within Org-mode
by passing the output from one block as input to another block. Results from a code block can be cached to
be re-calculated only when the code changed, avoiding repeated execution of lengthy calculations.
It is even possible to do metaprogramming in Org-mode, meaning that one block of source code generates
source code again, which will be handled correctly by Org-mode, too.
And finally, Org-mode is much more than a tool for literate programming. Being initially developed for
outlining, (TODO) lists, and project and task management, all this functionality is also available to the R
programmer.
We present workflows of how to perform data analyses, create reports, and publish results on your website
from within one tool.

References
Abhijit Dasgupta. Flexible report generation and literate programming using r and python’s docutils module. In useR!,
2010. URL http://user2010.org/slides/Dasgupta.pdf.

Max Kuhn, Steve Weston, Nathan Coulter, Patrick Lenon, and Zekai Otles. odfWeave: Sweave processing of Open
Document Format (ODF) files, 2010. URL http://CRAN.R-project.org/package=odfWeave. R pack-
age version 0.7.17.

Eric Lecoutre. The R2HTML package. R News, 3(3):33–36, December 2003. URL http://cran.r-project.
org/doc/Rnews/Rnews_2003-3.pdf.

Friedrich Leisch. Sweave: Dynamic generation of statistical reports using literate data analysis. In Wolfgang Härdle
and Bernd Rönz, editors, Compstat 2002 — Proceedings in Computational Statistics, pages 575–580. Physica
Verlag, Heidelberg, 2002. URL http://www.stat.uni-muenchen.de/~leisch/Sweave. ISBN 3-
7908-1517-9.
Reporting Technologies and Workflows, Tuesday 11:15-12:35 29

Efficient data analysis workflow in R

Peter Baker1
1. Statistical Consultant & Senior Lecturer, School of Population Health, University of Queensland, Herston, QLD. Australia
? Contact author: [email protected]

Keywords: workflow, unix tools, reproducible research, project templates, data handling

After the fifth large data set came through my door in as many months, I thought it would be more efficient
to automate workflow rather than start afresh each time. For instance, I knew my collaborators and clients
were likely to tweak their data a number of times because even though they had provided data with hundreds
or even thousands of variables, they’d undoubtedly left out a few important variables or discovered that some
were coded wrongly or recorded inconsistently. And of course, I also thought that my collaborators were
likely to contact me a couple of days before a grant application or final report was due with a revised data
set and new questions in the hope that I could just press the button to instantly extract some final answers.
About the same time, I noticed a few interesting posts on R–bloggers and stackoverflow. I particularly liked
the software engineering term DRY (don’t repeat yourself) with suggestions about automating processing
with R functions or packages. Another post referred to Long (2009) which provides a useful guide to man-
aging workflow for data analysis in large projects. Long’s ideas revolve around using stata in a Windows
environment in order to efficiently facilitate replication of work by following a cycle of Planning, Organis-
ing, Computing and Documenting. stata has some useful features like using variable “labels” in plots and
tables (unlike standard R), datasignatures and Long provides good strategies for using codebooks for data
handling and checking. However, the approach concentrates on manual methods rather than programming
tools like make, automatic initial data processing, regular expressions for text processing or version control.
Many tools are available for efficiently managing projects and carrying out routine programming tasks. One
such tool is GNU make. It is standard on linux and MacOSX and available via Rtools or cygwin for
Windows. Originally developed for programming languages like C it is well suited to statistical analysis.
Since the late 80’s I’ve used make to project manage data analysis using GENSTAT, SAS, R and other
statistical packages. It is very efficient in only re–running analyses or producing reports when dependencies
such as program files (R, Rnw, : : :) or data files change. R is used for all data analysis steps described below.
Using the R ProjectTemplate as a starting point, the following will be outlined:
 the set up of project directories, Makefiles, R program files to read and check data, initial documenta-
tion and log files for use with emacs Org-mode or other editor;
 using codebook(s) to label variables, set up factors with suitable labels;
 producing initial data summaries and plots for data checking and exploration;
 setting up an initial git repository for version control; and
 producing initial summaries using literate programming via Sweave.
An R package which automates the steps above is under development.

References
Kuhn, M. (2011). ReproducibleResearch cran task view. http://www.r-project.org/.
Long, J. S. (2009, February). The Workflow of Data Analysis Using Stata (1 ed.). Stata Press.
White, J. M. (2010). ProjectTemplate: Automates the creation of new statistical analysis projects. R package
version 0.1-3.
Teaching, Tuesday 11:15-12:35 30

Teaching Statistics to Psychology Students using Reproducible

Computing package RC and supporting Peer Review Framework
Ian E. Holliday1* and Patrick Wessa2
1. Aston University, School of Life and Health Sciences, Aston Brain Centre, Birmingham B4 7ET UK
2. Leuven Institute for Research on Information Systems, University of Leuven, Belgium
*Contact author: [email protected]
Keywords: statistics education, peer review, reproducible computing, package RC
There is increasing impetus towards a reform of statistics teaching in the face of a
widespread recognition that traditional approaches to statistics education focussed on
computational and analytical skills does not provide students with the ability to apply statistical
thinking in real world situations (Garfield et al 2002). Moore (1997) argues for a revised
approach requiring “the abandonment of an ‘information transfer’ model in favour of a
‘constructivist’ view of learning:..” (Moore, 1997). Furthermore, recent research suggests that
there are clear benefits of collaborative working involving opportunities for peer review and
feedback, particularly for female students who tend to use such opportunities most effectively
(Wessa, 2008).
We have developed an undergraduate second year statistics course for psychology
students based on principles of constructivist education, whose design is modelled on an
established course in a business education setting, aiming to encourage “statistical thinking and
literacy”. The key software components applied were developed and introduced by Patrick
Wessa: 1) the reproducible computing framework available at http://www.freesatistics.org via
package RC; and, 2) a supporting online peer review (PR) management system (Wessa, 2009).
Our presentation illustrates these components from the point of view of the student and the
instructor.
Students receive instruction via traditional lectures and supporting workshops, however
the workshop material is presented as a ‘compendium’, an enhanced document form containing
all the data and computations necessary to fully reproduce and communicate the results of a
statistical analysis. The students’ task is to complete workshop assignments and create a new
compendium (report) based on their own analysis and interpretation that contain links to their
‘blogged’ (i.e. archived in the reproducible computations repository) computations; their
documents are then uploaded to the PR system and circulated anonymously to students’ peers
for review. The students’ assessed assignment is to provide peer review feedback on up to 5
workshop compendiums they receive each week. The course design provides a social
constructivist framework within which independent learning can flourish.
The instructor is able to monitor, evaluate, and control the learning process through a
series of statistical reports and accompanying query tools which are available through the RC
package and the PR system. The data that is retrieved through these systems can be used for the
purpose of educational research based on objectively measured observations. Our analysis of
students’ performance over two years provides evidence that the combination of RC and PR
leads to deep learning of statistical concepts. We briefly present the latest findings to support
this conclusion, based on a structural equation model (PLS-PM) and a randomized experiment.

Garfield J; Hogg B, Schau C & Whittinghill D. (2002) “First Courses in Statistical Science:
The Status of Educational Reform Efforts.” Journal of Statistics Education Vol. 10( 2)
Moore, DS (1997) “New Pedagogy and New Content: The Case of Statistics” International
Statistical Review vol. 65 (2), pp123-165
Wessa, P. (2008). “How Reproducible Research Leads to non-Rote Learning Within a Socially
Constructivist e-Learning Environment.” 7th. Europ. Conf. on E-Learning, Vol 2, 651-658.
Wessa, P. (2009) “A framework for statistical software development, maintenance, and
publishing within an open-access business model.” Computational Statistics, 24(2), 183-193
Teaching, Tuesday 11:15-12:35 31

Teaching applied statistics to students in natural sciences

using the R Commander
Kristian Hovde Liland 1,* , Solve Sæbø1, Thore Egeland1,
Ellen Sandberg1, Lars Snipen1, Trygve Almøy1

1. The Norwegian University of Life Sciences, Dept. of Chemistry, Biotechnology and Food Science,
P.O. Box 5003, N-1432 Ås, Norway
*Contact author: [email protected]

Keywords: R Commander, natural sciences, applied statistics

Statistical software has become an important part of teaching applied statistics on all levels of higher
education. The assortment of available applications and packages is formidable, and choosing correctly
can be a key to successful learning. At many universities different software has been used for the
students that focus on statistics than those that have statistics as an obligatory subject in a study of
biology, chemistry or other experimental sciences. Typically software with a graphical user interface
and a limited set of statistical methods, e.g. Minitab or SPSS, is chosen for those who need easy access
to basic analysis, while more advanced software with a programming based interface, e.g. R, is chosen
for those in need of more flexible and adaptable analyses.

With the advent and continual expansion and refinement of the R Commander universities do not have
to make this choice anymore as everything is available through R. At the Norwegian University of Life
Sciences we are phasing out the other rigid and expensive programs. Through a local repository,
http://repository.umb.no/R, and automatic updates of our own plug-in to the R Commander,
RcmdrPlugin.UMB, we achieve instant distribution of the improvements and additions we make to our
statistical toolbox. We continually adapt the plug-in to the courses we teach, tweaking the built in
functions of the R Commander and adding what we need.

Utilizing the above set-up, students of natural sciences get a soft introduction to computer aided
statistical analysis, they get used to seeing and interpreting code, and they use the same base as the
students aiming for a statistical education. Teachers and researchers only have to relate to one software
package, and they can use their expertise when adapting it to their needs and research. Finally, the
institutes save money on expensive licenses.

In addition to this we have a dedicated course for teaching basic use of R for master and PhD students
in need of customized analysis in their research. This has become very popular.
Teaching, Tuesday 11:15-12:35 32

Automatic generation of exams in R

?
Bettina Grün1 , Achim Zeileis2,
1. Department of Applied Statistics, Johannes Kepler Universität Linz
2. Department of Statistics, Universität Innsbruck
?
Contact author: [email protected]

Keywords: Exams, Multiple Choice, Arithmetic Problems

Package exams provides a framework for automatic generation of standardized statistical exams which is
especially useful for large-scale exams. To employ the tools, users just need to supply a pool of exercises
and a master file controlling the layout of the final PDF document. The exercises are specified in separate
Sweave files (containing R code for data generation and LATEX code for problem and solution description)
and the master file is a LATEX document with some additional control commands. An overview is given of the
main design aims and principles as well as strategies for adaptation and extension. Hands-on illustrations –
based on example exercises and control files provided in the package – are presented to get new users started
easily.

References
Grün, B. and A. Zeileis (2009). Automatic generation of exams in R. Journal of Statistical Software 29(10),
1–14. http://www.jstatsoft.org/v29/i10/
Modelling Systems and Networks, Tuesday 16:00-17:00 33

An S4 Object structure for emulation - the approximation of

complex functions
Rachel Oxlade1 , Peter Craig1
1. Durham University, UK
?
Contact author: [email protected]

Keywords: Bayesian, computer model, emulation, nested models, S4 objects

We aim to facilitate the R user who has data from a complex function which is costly to evaluate directly
(for example a deterministic computer model, see for example Craig et al. (1997), Kennedy and O’Hagan
(2001)). We approximate the function by a gaussian process, and thereby produce a posterior distribution
for the function’s behaviour throughout its domain. Where the function’s value is already known, that is
at the input points belonging to the data mentioned above, the posterior will predict the correct value with
certainty. Elsewhere, the distribution will give an approximation which represents our uncertainty. Typically
the domain of the function may cover twenty or more dimensions, and there may be several thousand data
points. An object oriented structure is presented for dealing with this problem, enabling the user to interact
with the posterior distribution of the function across the domain. This is similar in its goals to the package
BACCO, but with a different approach.
A particular scenario of interest to us is where we are given a second model (or function), whose inputs
extend those of the first. It may contain an additional process, and it is useful for us to be able to compare
the two models, to assess the value of including this new process, and to see how the two functions differ
across the input domain. We deal with this problem by creating a hierarchical structure, which we again use
S4 objects to capture. Joint emulation of models with differing input spaces has, to our knowledge, not yet
been addressed, and so this approach is a novel one. The object oriented structure will be presented, along
with some methods for dealing with such objects.
In constructing these emulators there are many choices to make, for example whether or not to include a
regression surface to capture large scale variation, and if so, how it is to be formed. Incorporated into the R
code described above are functions which automate such choices given the user’s stipulations. This means
that, given some initial data showing the function’s behaviour and some directions as to how R is to proceed
at certain points, the functions’ behaviour at any set of previously unexplored points can be predicted, and
a measure of uncertainty given. The S4 object structure allows the user to interact with the objects, and in
particular provides methods to adjust them by making alterations to the original choices made.

References
Craig, P. S., M. Goldstein, A. H. Seheult, and J. A. Smith (1997). Pressure matching for hydrocarbon
reservoirs: a case study in the use of Bayes linear strategies for large computer experiments. Case Studies
in Bayesian Statistics 3, 37–93.

Kennedy, M. C. and A. O’Hagan (2001). Bayesian Calibration of computer models. Journal Of The Royal
Statistical Society Series B 63, 425–464.
Modelling Systems and Networks, Tuesday 16:00-17:00 34

The structmcmc package: Structural inference of Bayesian

networks using MCMC
Robert J. B. Goudie1
1. University of Warwick, UK
?
Contact author: [email protected]

Keywords: Bayesian networks, Graphical models, MCMC, MC3

I will describe the structmcmc package, which implements the widely-used MC3 algorithm (Madigan et al.,
1994), as well as a number of variants of the algorithm. The MC3 algorithm is a Metropolis-Hastings
sampler for which the target distribution is the posterior distribution of Bayesian networks.
The implementation allows the local conditional distributions to be multinomial or Gaussian, using standard
priors. Arbitrary structural priors for the Bayesian network can be specified. The main difficulty in sampling
Bayesian networks efficiently is ensuring the acyclicity constraint is not violated. The package implements
the cycle-checking methods introduced by King and Sagert (2002), which is an alternative to the method
introduced by Giudici and Castelo (2003). To enable convergence to be assessed, a number of tools for
creating diagnostic plots are included.
Interfaces to a number of other R packages for Bayesian networks are available, including deal (hill-climbing
and heuristic search), bnlearn (a number of constraint-based and score-based algorithms) and pcalg (PC-
algorithm). An interface to gRain is also included to allow its probability propagation routines to be used
easily.

References
Giudici, P. and R. Castelo (2003). Improving Markov Chain Monte Carlo Model Search for Data Mining .
Machine Learning 50, 127–158.

King, V. and G. Sagert (2002). A Fully Dynamic Algorithm for Maintaining the Transitive Closure. Journal
of Computer and System Sciences 65(1), 150–167.

Madigan, D., A. E. Raftery, J. C. York, J. M. Bradshaw, and R. G. Almond (1994). Strategies for Graphical
Model Selection. In P. Cheeseman and R. W. Oldford (Eds.), Selecting Models from Data: AI and
Statistics IV, pp. 91–100. New York: Springer-Verlag.
Modelling Systems and Networks, Tuesday 16:00-17:00 35

Computation of (generalized) Nash equilibria

?
Christophe Dutang1,
1. Université de Lyon, Université Lyon 1, Institut de Sciences Financière et d’Assurance, 50 avenue Tony Garnier, 69007 Lyon,
France
?
Contact author: [email protected]

Keywords: Generalized Nash equilibrium, Fixed-point method, Regularized gap function, Optimization

When John Nash introduces an equilibrium concept in his seminar paper Nash (1950) for n-player games.
Soon after, Debreu (1952) generalized the Nash equilibrium concept for his abstract economy by allowing
the strategy space of players to depend on others player actions. From that time, the concept was known as
generalized Nash equilibrium (GNE).
Applications of GNE can be found in many fields, e.g. economics, engineering, mathematics, computer
science, operational research. However, the computational side of the GNE has not been very studied
since his introduction in the 50’s, because of its mathematical-economic origin according to Facchinei and
Kanzow (2009). Only recently, papers such as Facchinei and Kanzow (2009), von Heusinger and Kanzow
(2009) and Nabetani et al. (2009) focus on this topic.
The presentation will focus on computational methods when working with continuous strategy space, i.e.
excluding matrix games. In the current litterature, there are three main approaches to solve GNE problem:
(i) fixed-point algorithms, (ii) gap function minimization and (iii) extended KKT system solving methods.
These approaches are implemented in the R package GNE1 by three distinct functions: fixedpoint,
minGap and NewtonKKT.

References
Debreu, G. (1952). A social equilibrium existence theorem. Proc. Nat. Acad. Sci. U.S.A..

Facchinei, F. and C. Kanzow (2009). Generalized nash equilibrium problems. Updated version of the
’quaterly journal of operations research’ version.

Nabetani, K., P. Tseng, and M. Fukushima (2009). Parametrized variational inequality approaches to gener-
alized nash equilibrium problems with shared constraints. Computational Optimization and Applications.

Nash, J. F. (1950). Equilibrium points in n-person games. Proc. Nat. Acad. Sci. U.S.A. 36, 48–49.

von Heusinger, A. and C. Kanzow (2009). Optimization reformulations of the generalized nash equilibrium
problem using the nikaido-isoda type functions. Computational Optimization and Applications 43(3).

1
available on R-forge.
Computational Physics and Chemometrics, Tuesday 16:00-17:00 36

Segmented regression in thermo-physics modeling

Irina Roslyakova1*, Holger Dette2, Mauro Palumbo1

1 - Department of Scale Bridging Thermodynamic and Kinetic Simulation, ICAMS, Ruhr-Universität Bochum, UHW
10/1022, Stiepeler Str. 129, 44801 Bochum

2 - Ruhr-Universität Bochum, Fakultät für Mathematik, Mathematik III, Universitätsstraße 150, 44780 Bochum
*
Contact author: [email protected]

Keywords: segmented regression, bootstrap, confidence intervals, thermophysical properties of

materials, heat capacity

Traditionally in applied thermo-physics the temperature dependence of the heat capacity is described
by polynomials in temperature (from 298K to the melting point) [1], with adjustable parameters fitted
to experimental data. The effort to extend that description to low temperature demands more physical
modeling which takes into account the recently available theoretical data.

The more physical approach requires the modeling of several contributions (e.g. electronic,
vibrational, etc.) that appear in different temperature ranges [2].

In this work we propose the development of statistical segmented models, that can be used as a
support tool for CALPHADa modeling [3] and we implement them in R [4, 5]. Several segmented
regression functions were considered and analyzed, with the corresponding confidence intervals being
calculated using the bootstrap method. Moreover we validate the consistency of underlying fitting
results, by calculating other physical properties of interest, such as the enthalpy, that can be derived
directly from the heat capacity of the studied material.

An overview of available functionalities in the R software package for segmented regression [6, 7] will
be discussed in solving the nonlinear equations arising from complex model that has to be applied.

References

[1] O. Kubaschewski, C.B. Alcock, P.J. Spencer (1993). Materials thermochemistry.

[2] G. Grimvall (1986). Thermophysical properties of materials.
[3] B. Sundman, H. Lukas, S. Fries (2007). Computational Thermodynamics: The Calphad Method.
[4] G. A. F. Seber, C. J. Wild (1989). Nonlinear Regression.
[5] J. D. Toms, M. L. Lesperance (2003). Piecewise regression: A tool for identifying ecological
thresholds. Ecology 84(8), pp.2034-2041
[6] Vito M.R. Muggeo (2010). Segmented relationships in regression models. R-Package: segmented
(Version 0.2-7.3)
[7] Derek Sonderegger (2010). SiZer: Significant Zero Crossings. R-Package: SiZer (Version 0.1-3)

a
CALPHAD = CALculation of Phase Diagramm
Computational Physics and Chemometrics, Tuesday 16:00-17:00 37

Sparse Bayesian kernel projections for classification of

near-infrared spectroscopy data
?
Katarina Domijan1
1. Department of Mathematics and Statistics, NUI Maynooth, Maynooth, Ireland

?
Contact author: [email protected]

Keywords: Bayesian inference, reproducing kernel Hilbert spaces, Bayesian decision theory, NIR spec-
troscopy

A Bayesian classification method is applied to near infrared spectroscopic data obtained from from several
food authenticity studies. The datasets record spectra over the visible and near infra-red wavelength range
for relatively small numbers of training samples. Typically the dimension of the variables exceeds ten to
twenty - fold the number of samples, however, the reflectance measurements are sequentially highly cor-
related. The classifier is based on the reproducing kernel Hilbert spaces (RKHS) theory which allows for
nonlinear generalization of linear classifiers by implicitly mapping the classification problem into a high
dimensional feature space where the data are thought to be linearly separable. The proposed algorithm
performs the classification of the projections of the data to the principal axes of the feature space. The
projections are uncorrelated and sparse, leading to large computational savings and improved classification
performance. The likelihood is modeled through the multinomial logistic regression model and the rela-
tively standard hierarchical prior structure for Bayesian generalized linear models is assumed. The degree
of sparsity is regulated in a novel framework based on Bayesian decision theory. The Gibbs sampler is
implemented to find the posterior distributions of the parameters, thus probability distributions of prediction
can be obtained for new data points, which gives a more complete picture of classification. The classifier
obtains good classification rates for the datasets considered and does not require pre-processing steps to
reduce their dimensionality. The results from several classification algorithms available in R programming
language are included for comparison. A collection of R functions for classification and data visualization
is available.
Computational Physics and Chemometrics, Tuesday 16:00-17:00 38

Recovering Signals and Information From Radio Frequencies

Using R (A high school student’s experience)
?
Jinhie Skarda1
1. Montgomery Blair High School, Silver Spring. MD 20901
?
Contact author: [email protected]

Keywords: RF signal analysis and demodulation, Time-series analysis, Software oscilloscope and radio

Radio frequency (RF) is part of the electromagnetic spectrum (10 kHz-300 GHz) and is essential to com-
munications and information transfer. For this study, RF measurements were carried out with antennas
connected to a PC software oscilloscope, and a digital receiver combined with a software-defined radio tool
kit (GNU radio). The R language and environment for statistical computing was used to analyze RF signal
strengths as well as the information carried on amplitude modulated (AM) radio signals. The oscilloscope
was programmed to take ten RF time histories (with 40,000 points over 2 milliseconds) approximately 2
seconds apart. The RF power spectra (up to 10 MHz in frequency) corresponding to the time histories were
computed using the fft function, and averaged. Variations in RF signal strengths were then studied by
taking the ten time history sets every 6 hours for several weeks, and every half hour over a couple of days.
The signal strengths of the four AM stations chosen for detailed analysis were found to vary by as much
as two orders of magnitude, with strongest signals typically occurring at noon. The information content of
multiple AM signals was successfully analyzed using mixing, filtering, decimation, and interpolation rou-
tines written in R to demodulate a digital signal. The multiple carrier waves were contained in a single RF
signal, which was received and stored using the GNU radio and a digital receiver. Where feasible, the results
obtained using these routines were compared with the results from standard or specialized R packages such
as signal and tuneR. The reduction of undesirable RF signals inside RF-free environments of homemade
Faraday cages was also investigated, with data analysis performed in R. Faraday cages were effective in
reducing RF signal strengths to 1/1000 of the unshielded signals outside the cages. The plotting capabilities
of R were used for graphical analysis of the RF data in the form of conventional frequency and time history
plots, box plots, and contour plots (sometimes referred to as waterfall plots in RF analysis).

References

Adler, Joseph (2010). R in a Nutshell, A Desktop Quick Reference, California, O’Reilly.

Greensted, Andrew (2010). FIR Filters by Windowing The Lab Book Pages, http://www.labbookpages.
co.uk/audio/firWindowing.html#code/.
Hamming, R. W. (1998). Digital Filters, New York, Dover Publications Inc.
Lyons, Richard G. (2004). Understanding Digital Signal Processing, New Jersey, Prentice Hall.
Rouphael, Tony J. (2009). RF and Digital Signal Processing for Software-Defined Radio, A Multi-Standard
Multi-Mode Approach, New York, Elsevier.
Schmitt, Ron (2002). Electromagnetics Explained, A Handbook for Wireless/RF, EMC, and High-Speed
Electronics, New York, Elsevier.
Spector, Phil (2008). Data Manipulation with R, New York, Springer.
Straw, Dean R., Ed. (2007). The ARRL Antenna Book, Connecticut, The AARL Inc.
Visualisation, Tuesday 16:00-17:00 39

animatoR: dynamic graphics in R

?
Andrej Blejec1,2,
1. National Institute of Biology, Ljubljana, Slovenia
2. University of Ljubljana, Department of Biology, Ljubljana, Slovenia
?
Contact author: [email protected]

Keywords: Graphics, Dynamic graphics, Teaching

Graphics, especially dynamic graphics, is an impressive tool in various demonstrations. In statistics teach-
ing, there are many situations where graphics with animation of certain elements of the picture communicate
the concepts in obvious way.
Since R graphic devices are in a sense static, several approaches towards dynamic graphics are used. On
many occasions, one would like to move certain graphical element, for example one point, on otherwise
static picture. One way is to hide the point by re-plotting it in exclusive OR (XOR) mode and plotting the
point in a new position. This method can be fast since one is plotting only the elements that are changing
on the otherwise static background which can be very complex. R graphic devices are not suitable for such
technique. Another way, which is close to this technique, is hiding the dynamic elements by re-plotting
them in the background color. This works only for pictures with solid single color background and proves to
be unsatisfactory. Another technique is to simply plot a series of complete pictures, each one with relocated
picture elements. If the pictures are not very complex, R is fast enough (if not too fast) for producing a
flicker free dynamic impression. This is the most popular technique, which can provide satisfactory results.
To get an impression of smooth movement, the changes in successive pictures should be small and one
needs to get many intermediate point or line positions. Here we provide technique and a set of functions
that complement base graphics function for production of dynamic graphics. The basic idea is to define
the starting and finishing coordinates of moving picture elements (points, lines, segments, etc.). Then we
plot a series of pictures for successive intermediate positions, which are calculated by homotopy between
starting and finishing values. If start position is x0 and end position is x1 than positions between them can
be determined as
xt = x0 (1 − t) + x1 t, t ∈ [0, 1]
for different values of homotopy parameter t. Selection of suitable sequence for homotopy parameter t
provides an impression of smooth movement along trajectories from starting to finishing positions.
Package animatoR is a developing package utilizing homotopy for production of smooth dynamic graphics,
with the motive of presentations and use in teaching. In the presentation, several examples that demonstrate
the use of dynamic graphics in statistics teaching will be shown.
Visualisation, Tuesday 16:00-17:00 40

Graphical Syntax for Structables and their Mosaic Plots

?
Erich Neuwirth1 , Richard M. Heiberger2,
1. University of Vienna
2. Temple University
?
Contact author: [email protected]

Keywords: RExcel, Rcmdr, aperm, graphical syntax, mosaic

A "structable" object is a representation in the vcd package in R of a k-dimensional contingency

table. The structable object has an attribute "split vertical" that carries two pieces of information:
assignment of the factors to row or columns, and sequencing of the factors. The default plot of a structable is
as a mosaic plot with recursive splits of the factors in the specified sequence. Each split is along the vertical
or horizontal direction associated with the column or row assignment of its factor.
The printed display of a structable as a flat table in two dimensions shows the row and column assignment
but is unable to illustrate the sequencing of the splits. As a consequence, multiple structables—and their
associated mosaic plots—can yield the same printed flat table.
We have developed a graphical notation (with RExcel and Rcmdr implementations), and corresponding
R functions with a command-line notation, that simplify the specification of the alternate sequencing of
splits—hence alternate mosaic plots—associated with a printed flat table. The notation also permits re-
assignment of factors from rows to columns. The primary R function is an aperm (array permutation)
method designed for structables. It extends the permutation argument of the default aperm method for ar-
rays and tables to include the "split vertical" information that distinguishes the multiple structables
associated with the same flat printed table.
Visualisation, Tuesday 16:00-17:00 41

RMB: Visualizing categorical data with Relative Multiple

Barcharts
Alexander Pilhöfer
1. Alexander Pilhöfer
Department of Computer Oriented Statistics and Data Analysis
Institute of Mathematics
University of Augsburg
86135 Augsburg, Germany
E-mail: [email protected]
URL: http://www.rosuda.org

Keywords: categorical data, visualization, multiple barcharts, logistic regression

This talk will present a new graphic for displaying categorical data called “Relative Multiple Barcharts”
(rmb-plot) and its implementation in the R package extracat. It is a new attempt to enrich the family
of mosaicplots by combining the most important advantages of multiple barcharts (see Hofmann, 2000)
and classical mosaicplots (see Friendly, 1994) in one display. The intention of rmb-plots is to precisely
display relative frequencies of a target variable for each combination of explanatory variables divided over
a grid-like graphical display and, simultaneously, their corresponding weights. The breakup of absolute
frequencies into conditional distributions and weights is a common procedure in many methodologies for
categorical data analysis, such as generalized linear models or correspondence analysis, but even so there
seems to be a lack of graphical solutions for exploratory as well as illustrative purposes. After a brief
introduction to the concepts of the plot the talk will focus on the implementation in R. Using the well-
known Copenhagen housing dataset for the examples the talk will first present the main variants of the plot
such as a generalized version of spineplots as well as the most important options like color schemes, ceiling
censored zooming and residual-based shadings according to Poisson models or logistic regression models.
Moreover an interactive version of the graphic based on the iWidgets package will be presented which
provides interactive controls for all important options as well as a connection to classical mosaicplots in the
vcd (Hornik et al., 2006) package.

References
Friendly, M. (1994). Mosaic displays for multi-way contingency tables. Journal of the American Statistical
Association 89, 190–200.

Hofmann, H. (2000). Exploring categorical data: Interactive mosaic plots. Metrika 51(1), 11–26.

Hornik, K., A. Zeileis, and D. Meyer (2006). The strucplot framework: Visualizing multi-way contingency
tables with vcd. Journal of Statistical Software 17.

Pilhoefer, A. and A. Unwin (2010). Multiple barcharts for relative frequencies and parallel coordinates plots
for categorical data - package extracat. Journal of Statistical Software. submitted.
Official and Social Statistics, Tuesday 16:00-17:00 42

ObsSensitivity: An R package for power analysis for sensitivity

analyses of Observational Studies
?
Neil Diamond1, , Ewa Sztendur2
1. Monash University, Melbourne, Australia
2. Victoria University, Melbourne, Australia
?
Contact author: [email protected]

Keywords: Observational Studies, Power, Sensitivity Analysis

Researchers in the social sciences have been encouraged to use randomised controlled experiments (see,
for example, Mosteller and Boruch(2002)). This has been aided by the development of easily used software
such as that of Spybook et. al. (2009) which allows researchers to design both person randomised and cluster
randomised trials. Despite this, observational studies are still often used. The design of observational
studies is just as important as in randomised controlled experiments and “a well designed observational
study resembles, as closely as possible, a simple randomized experiment . . . [except that in an observational
study] . . . randomization is not used to assign treatments”. (Rosenbaum, 2010, p.4). Many observational
studies now use propensity scores to make the treatment and control groups comparable.
Although propensity scores remove the effect of measured covariates, they do not remove bias due to un-
measured variables. An essential aspect of the analysis and reporting of an observational study is to carry
out a sensitivity analysis which determines the magnitude of the bias that would be needed to alter the con-
clusions of the study. The bias is measured in terms of a parameter Γ (Gamma): the odds of receiving the
treatment rather than the control given the observed covariates. A value of Γ = 1 indicates that there is no
bias due to unmeasured covariates while a value of Γ = 2 indicates that an unobserved covariate has the
effect of making one of two apparently equal subjects twice as likely to be in the treatment group than the
control group.
This talk summarises an R package, ObsSensitivity, that assists researchers to determine the appropriate
sample size for an observational study. The software tool is provided in R/Excel using the R Commander in-
terface to R, giving the advantages of R but with a convenient and easily learnt environment. Demonstration
of the use of the software with examples of actual observational studies will be given.
Mosteller, F. and and Boruch, R., (Eds.), (2002). Evidence Matters: Randomized Trials in Education
Research, Brookings Institution Press: Washington, DC.
Rosenbaum, P.R (2010). Design of Observational Studies Springer: New York.
Spybook, J., Raudenbush, S.W., Congdon, R., and Martı́nez, A. (2009). Optimal Design for Longitudinal
and Multilevel Research: Documentation of the “Optimal Design” Software.
http://www.wtgrantfoundation.org/resources/overview/research_tools
Downloaded 31 March, 2011.
Official and Social Statistics, Tuesday 16:00-17:00 43

Visualizing Multilevel Propensity Score Analysis

Jason M. Bryer1,*

1. University at Albany
*Contact author: [email protected]

Keywords: propensity score analysis, multilevel analysis, causal inference, graphics

The use of propensity score analysis (Rosenbaum & Rubin, 1983) has gained increasing popularity for
the estimation of causal effects within observational studies. However, its use in situations where data is
multilevel, or clustered, is limited (Arpino & Mealli, 2008). This talk will introduce the
multilevelPSA (Bryer, 2011) package for R that provides functions for estimating propensity
scores for large datasets using logistic regression and conditional inference trees. Furthermore, a set of
graphical functions that extends the framework of visualizing propensity score analysis introduced by
Helmreich and Pruzek (2009) to multilevel analysis will be discussed. An application for estimating the
effects of private schools on reading, mathematics, and science outcomes from the Programme for
International Student Assessment (PISA; Organization for Economic Co-operation and Development,
2009) is provided.

References

Arpino, B. & Mealli, F. (2008). The specification of the propensity score in multilevel observational
studies. Dondena Working Paper 6.

Bryer, J. (2011). multilevelPSA: Package for estimating and visualizing multilevel propensity
score analysis. http://multilevelpsa.r-forge.r-project.org

Helmreich, J & Pruzek, R.M. (2009). PSAgraphics: An R package to Support Propensity Score
Analysis. Journal of Statistical Software, 29, 06.

Organization for Economic Co-operation and Development (2009). Programme for International
Student Assessment. http://www.pisa.oecd.org/

Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational
studies for causal effects. Biometrica, 70, 1, pp. 41-55.
Dimensionality Reduction and Variable Selection, Tuesday 16:00-17:00 44

ClustOfVar: an R package for the clustering of variables

?
Marie Chavent1,2, , Vanessa Kuentz3 , Benoı̂t Liquet4 , Jérôme Saracco1,2
1. IMB, University of Bordeaux, France
2. CQFD team, INRIA Bordeaux Sud-Ouest, France
3. CEMAGREF, UR ADBX, France
4. ISPED, University of Bordeaux, France
?
Contact author: [email protected]

Keywords: Mixture of quantitative and qualitative variables, hierarchical clustering of variables, k-means
clustering of variables, dimension reduction.

Clustering of variables is as a way to arrange variables into homogeneous clusters i.e. groups of variables
which are strongly related to each other and thus bring the same information. Clustering of variables can then
be useful for dimension reduction and variable selection. Several specific methods have been developed for
the clustering of numerical variables. However concerning qualitative variables or mixtures of quantitative
and qualitative variables, much less methods have been proposed. The ClustOfVar package has then been
developped specifically for that purpose. The homogeneity criterion of a cluster is the sum of correlation
ratios (for qualitative variables) and squared correlations (for quantitative variables) to a synthetic variable,
summarizing “as good as possible” the variables in the cluster. This synthetic variable is the first principal
component obtained with the PCAMIX method. Two algorithms for the clustering of variables are proposed:
iterative relocation algorithm, ascendant hierarchical clustering. We also propose a bootstrap approach in
order to determine suitable numbers of clusters. The proposed methodologies are illustrated on real datasets.

References

Chavent M, Kuentz V., Saracco J. (2009). A Partitioning Method for the clustering of Categorical vari-
ables. In Classification as a Tool for Research, Hermann Locarek-Junge, Claus Weihs (Eds), Springer,
Proceedings of the IFCS’2009, Dresden.
Dhillon, I.S., Marcotte, E.M. and Roshan, U. (3003). Diametrical clustering for identifying anti-correlated
gene clusters, Bioinformatics, 19(13), 1612-1619.
Kiers, H.A.L., (1991). Simple structure in Component Analysis Techniques for mixtures of qualitative and
quantitative variables, Psychometrika, 56, 197–212.
Pagès, J. (2004). Analyse factorielle de données mixtes, Revue de Statistique Appliquée, 52(4), 93-111.
Vigneau, E. and Qannari, E.M., (2003). Clustering of Variables Around Latent Components, Communica-
tions in Statistics - Simulation and Computation, 32(4), 1131–1150.
Dimensionality Reduction and Variable Selection, Tuesday 16:00-17:00 45

Variable Screening and Parameter Estimation for

High-Dimensional Generalized Linear Mixed Models Using
`1-Penalization
?
Jürg Schelldorfer1, , Peter Bühlmann1

1
Seminar für Statistik, Department of Mathematics, ETH Zurich, CH-8092 Zurich, Switzerland
?
Contact author: [email protected]

Keywords: generalized linear mixed models, Lasso, high-dimensional data, coordinatewise optimization,
variable selection

We propose a two step procedure for dealing with high-dimensional generalized linear mixed models. Gen-
eralized linear mixed models (Breslow and Clayton, 1993; Bates, 2009a,b) are straighforward extensions of
generalized linear models for clustered observations. In a first step, we perform a Lasso-type (Tibshirani,
1996) variable screening procedure in order to select a relatively small set of covariates. In the second step,
we perform ordinary maximum likelihood using only the variables selected in the first step. The latter step
is necessary for overcoming bias problems steming from the variable screening step.

In this talk, we present the key ingredients for fitting high-dimensional generalized linear mixed models and
demonstrate the performance of the prodecure by presenting the new R package glmmlasso.

This work is an extension of Schelldorfer et al. (2011) for gaussian linear mixed models and the R package
lmmlasso, which is available from R-Forge (http://lmmlasso.R-forge.R-project.org) and the first author’s
website (http://stat.ethz.ch/people/schell).

References
Bates, D. M. (2009a). Computational methods for mixed models. Vignette for lme4 on http://www.r-
project.org/.

Bates, D. M. (2009b). Linear mixed model implementation in lme4. Vignette for lme4 on http://www.r-
project.org/.

Breslow, N. and D. G. Clayton (1993). Approximate inference in generealized linear mixed models. Journal
of the American Statistical Association 88, 9–25.

Schelldorfer, J., P. Bühlmann, and S. van de Geer (2011). Estimation for high-dimensional linear mixed-
effects models using `1 -penalization. To appear in the Scandinavian Journal of Statistics, available at
http://arxiv.org/abs/1002.3784.

Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical
Society Series B 58, 267–288.
Dimensionality Reduction and Variable Selection, Tuesday 16:00-17:00 46

gamboostLSS: boosting generalized additive models for

location, scale and shape
Benjamin Hofner1,∗
1. Institut für Medizininformatik, Biometrie und Epidemiologie; Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany
?
Contact author: [email protected]

Keywords: GAMLSS, high-dimensional data, prediction inference, spatial models, variable selection

Generalized additive models for location, scale and shape (GAMLSS) model not only the mean but every
parameter of the conditional distribution of the outcome (e.g. location, scale and shape) using distinct sets of
covariates (Rigby and Stasinopoulos, 2005). We present a boosting algorithm for GAMLSS (gamboostLSS;
Mayr et al., 2010), which was developed to allow model fitting for potentially high-dimensional data and
that overcomes limitations of the original fitting algorithm related to variable selection based on (potentially
problematic) information criteria. Furthermore, our approach allows to include a wide variety of possible
covariate effects such as linear, smooth, random, or even spatial effects. As the algorithm relies on boosting,
estimation of the effects with intrinsic variable selection is possible.
We apply the gamboostLSS approach to data of the Munich Rental Guide, which is used by landlords and
tenants as a reference for the rent of a flat depending on its characteristics and spatial features. The net-rent
predictions resulting from the high-dimensional GAMLSS are highly competitive to classical GAMs while
covariate-specific prediction intervals show a major improvement.
A software implementation of the algorithm is available in the R package gamboostLSS (Hofner et al.,
2010).

References
Hofner, B., A. Mayr, N. Fenske, and M. Schmid (2010). gamboostLSS: Boosting Methods for GAMLSS
Models. R package version 0.5-0.

Mayr, A., N. Fenske, B. Hofner, T. Kneib, and M. Schmid (2010). GAMLSS for high-dimensional data - a
flexible approach based on boosting. Technical Report 98, Department of Statistics, Ludwig-Maximilans-
Universität München.

Rigby, R. A. and D. M. Stasinopoulos (2005). Generalized additive models for location, scale and shape
(with discussion). Applied Statistics 54, 507–554.
Business Management, Tuesday 16:00-17:00 47

SCperf: An inventory management package for R

?
Marlene S. Marchena1
1. Department of Electrical Engineering, Pontifical Catholic University of Rio de Janeiro
?
Contact author: [email protected]

Keywords: Supply chain management, inventory control, EOQ model, safety stock, bullwhip effect.

Supply Chain Management, i.e., the control of the material flow from suppliers of raw material to final
customers, is a crucial problem for companies. If appropriately designed and executed, it may offer efficient
business solutions, thereby minimizing costs and improving readiness or competitiveness. In this context,
the use of mathematical inventory models can give a significant competitive advantage.
We have developed the SCperf package as the first package which implements different inventory models
that can be used when developing inventory control systems. There are several basic considerations that
must be reflected in the inventory model. For instance, models can be divided into deterministic models and
stochastic models according to the predictability of demand involved. Our package presents functions to
estimate the order quantity and the reorder point regarding different models. Also, other important variables
like the safety stock level and the bullwhip effect are calculated. During the presentation, examples will be
used to illustrate different inventory situations.

References

[1] Sven Axsäter (2006) Inventory control. Springer, New York. 2nd
[2] Frederick Hillier and Gerald Lieberman (2001). Introduction to operational research. McGraw-Hill.
New York, 7th.
[3] Marlene Marchena (2010). The bullwhip effect under a generalized demand process: an R implemen-
tation. In Book of Contributed Abstract useR! 2010, The R User Conference, (Gaithersburg, USA), pp.
101.
[4] Marlene Marchena (2011). Measuring and implementing the bullwhip effect under a generalized demand
process. PhD Thesis Pontifical Catholic University of Rio de Janeiro, 2011.
[5] R Development Core Team (2010). R: A language and environment for statistical computing. R Foun-
dation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0,
http://www.R-project.org.
Business Management, Tuesday 16:00-17:00 48

Using R to test transaction cost measurement for supply chain

relationship: A structural equation model
?
Pairach Piboonrungroj1,2, , Stephen M. Disney1
1. Logistics Systems Dynamics Group, Cardiff Business School, Cardiff University, United Kingdom
2. Chiang Mai School of Economics, Chiang Mai University, Thailand
?
Contact author: [email protected]

Keywords: Transaction cost economics, Supply chain relationship, Structural equation model

Transaction cost economics (TCE) has been widely used to explain the existence and boundary of the firm
(Williamson, 2005). Recently, TCE has been extended to inter-firm relationship in supply chains (Hobbs,
1996). However, a measurement of the transaction cost has rarely been tested empirically. Grover and Mal-
hotra (2005) attempted to do so but that measurement has limited to an industrial context and did not cover
transaction costs related to the governance problem and the opportunity cost. Thus, we revisited the mea-
surement of transaction cost using both industrial and service perspectives. Moreover, we also considered
the transaction cost metric that associated with the governance and the opportunity cost. Then a revised
transaction cost metric was tested with empirical data from the tourism supply chains in Thailand using a
structural equation model (SEM).

In this study, we used packages in R, specifically sem (Fox, 2006) and OpenMx (Boker et al., 2011) to test
the measurement model. Results from R were then compared to those from other popular SEM software i.e.,
AMOS (Arbuckle, 1995) and LISREL (Jöreskog and Sörbom, 1997). An evaluation of SEM using packages
in R and other softwares is also discussed.

Acknowledgement
The authors are grateful to the Royal Thai Government through the Commission on Higher Education for fi-
nancial support of Mr. Piboonrungrojs study in Cardiff University (under the program Strategic Scholarships
for Frontier Research).

References
Arbuckle, J. L. (1995). Amos 16.0 User’s Guide. Chicago: SPSS.
Boker, S., M. Neale, H. Maes, M. Wilde, M. Spiegel, T. Brick, J. Spies, R. Estabrook, S. Kenny, T. Bates,
P. Mehta, and J. Fox (2011). Openmx: An open source extended structural equation modeling framework.
Psychometrika (In Press), 1–12.
Fox, J. (2006). Structural equation modeling with the sem package in r. Structural Equation Modeling 13(3),
465–486.
Grover, V. and M. K. Malhotra (2005). Transaction cost framework in operations and supply chain manage-
ment research: theory and measurement. Journal of Operations Management 21(4), 457–473.
Hobbs, J. E. (1996). A transaction cost approach to supply chain management. Supply Chain Manage-
ment 1(2), 15–27.
Jöreskog, K. G. and D. Sörbom (1997). LISREL 8 user’s reference guide. Lincolnwood, IL: Scientific
Software International.
Williamson, O. E. (2005). Transaction cost economics and business administration. Scandivavian Journal
of Management 21(1), 19–40.
Business Management, Tuesday 16:00-17:00 49

Integrating R and Excel for automatic business forecasting

?
Giovanni Millo1, , Fabrizio Ortolani1
1. Assicurazioni Generali S.p.A., R&D Department

?
Contact author: Giovanni [email protected]

Keywords: RExcel, Forecasting.

We present a simple exercise in bridging the gap between statistics and everyday business practice, based on
two powerful tools already available in the R system: the forecast package Hyndman (2011) for automatic
time series forecasting and the RExcel add-in for MS Excel Baier and Neuwirth (2007), allowing to embed
R functionality into spreadsheets and to interact with their built-in macro language. The application we
developed makes forecasting practice accessible to those who are not familiar with statistical programs and,
possibly, do not even have a sound statistical background.
Many processes inside the firm involve forecasting. Some build on models and relationships between bal-
ance sheet items, but sometimes an a-theoretical extrapolation of past tendencies is needed. As few firms
can afford to have trained statisticians dedicated to supply-chain forecasting and the like, budgeting and
other activities are often based on simple, heuristic extrapolation of past data. It is commonplace, especially
in small enterprises, to ”pick last year/month’s value”, either in terms of stocks or of increments, as the best
estimate for the coming period.
Fully automatic forecasting of time series, based on model fitting and model comparing algorhythms select-
ing the ’best’ model for the data at hand, provides a statistically well founded solution to the forecasting
problem and can be of great use to the firm in obtaining accurate predictions for variables like sales, com-
modities’ input needs and the like, where forecast errors cost money.
Such fully automatic procedures are implemented in a variety of commercial software. We show how an
open-source solution is also very easy to set up.
The ideal way is thus to have the forecaster dealing only with Excel for data input, commmand issuing and
results’ retrieval, while a ’real’ statistical engine transparently does the computing in the background.
Now the forecaster just has to select the data vector, press the trigger keys for showing up the userform,
select the data frequency and press OK. He will get the forecasts at the end of the original series.

References
Baier, T. and E. Neuwirth (2007). Excel :: Com :: R. Computational Statistics 22(1), 91–101.

Hyndman, R. J. (2011). forecast: Forecasting functions for time series. R package version 2.13.
Kaleidoscope IIa, Wednesday 09:50-10:50 50

Using R to quantify the buildup in extent

of free exploration in mice
Tal Galili1*, Yoav Benjamini1

1. Department of Statistics and Operations Research, Tel Aviv University, Tel Aviv 69978, Israel
*Contact author: [email protected]

Keywords: Quantile loess, sequences of repeated motion, dynamics of behavior, open field test,
phenotyping mouse behavior

Background: (ref 1) To obtain a perspective on an animal's own functional world, we study its behavior
in situations that allow the animal to regulate the growth rate of its behavior and provide us with the
opportunity to quantify its moment-by-moment developmental dynamics. Thus, we are able to show
that a mouse's exploratory behavior consists of sequences of repeated motion: iterative processes that
increase in extent and complexity, whose presumed function is a systematic active management of input
acquired during the exploration of a novel environment. We use this study to demonstrate our approach
to quantifying behavior: targeting aspects of behavior that are shown to be actively managed by the
animal, and using measures that are discriminative across strains and treatments and replicable across
laboratories.

The R perspective: In our research, R was our central tool of choice (ref 1, citation 39). We employed
various existing R facilities for preparing, analyzing, and visualizing the data. We implemented the
Quantile.loess algorithm in R (the code was published online ref 2) in order to quantify various
measurements of the buildup in the mouse’s exploration of the ring.

In this talk I will provide the context of this study and present how R was used for devising and
implementing known and new methods in order to support our investigation.

References

1. Yoav Benjamini, Ehud Fonio, Tal Galili, Gregor Z. Havkin, and Ilan Golani (2011).
Quantifying the buildup in extent and complexity of free exploration in mice. PNAS (Published
online before print March 7, 2011)
2. Tal Galili (2010). R-statistcs blog, http://www.r-statistics.com/2010/04/quantile-
loess-combining-a-moving-quantile-window-with-loess-r-function/
Kaleidoscope IIa, Wednesday 09:50-10:50 51

Changepoint analysis with the changepoint package in R

?
R. Killick1, , I.A. Eckley1
1. Department of Mathematics & Statistics, Lancaster University, Lancaster, U.K.
?
Contact author: [email protected]

There exists a rich literature exploring changepoint problems which dates back to the 1950s. To date there
have been few implementations of changepoint methods within existing R packages. At the time of writing,
those packages which are currently available implement novel methodology only. As a consequence it
is more difficult (i) to compare the performance of new methods with those proposed in the established
literature and (ii) for practitioners to test changepoint methods on their data.
The changepoint package attempts to resolve the above by providing both well-established and new meth-
ods within a single package. In doing so, we hope to make it easier for practitioners to implement existing
methods and for researchers to compare the performance of new approaches against the established liter-
ature.. The package has therefore been designed to give users access to many techniques for changepoint
analysis within a few easy to use functions. It includes functions that detect changes in mean, variance and
regression under various distributional and distribution-free assumptions.
Given the capacity to sample time series at ever higher frequencies the need for computationally efficient
search methods is increasing. This is currently an active area of research and so we include four search
options in this initial version of the package. The search options which are available are At Most One
Change (AMOC), Binary Segmentation (Scott and Knott, 1974), Segment Neighbourhoods (Auger and
Lawrence, 1989) and PELT (Killick et al., 2011).
The presentation will be structured as an introduction to changepoint analysis followed by a demonstration
of the methods within the changepoint package. The methods will be illustrated using datasets from a
variety of application areas including genetics, oceanography, finance and linguistics.

References
Auger, I. E. and Lawrence, C. E. (1989). Algorithms for the optimal identification of segment neighbor-
hoods. Bulletin of Mathematical Biology, 51(1):39–54.

Killick, R., Fearnhead, P., and Eckley, I. A. (2011). Optimal detection of changepoints with a linear compu-
tational cost. In Submission.

Scott, A. J. and Knott, M. (1974). A cluster analysis method for grouping means in the analysis of variance.
Biometrics, 30(3):507–512.
Kaleidoscope IIa, Wednesday 09:50-10:50 52

Clustering patterns in streamflow to produce regionally or

anthropogenically similar groups
?
Karen R. Ryberg1,2,
1. U.S. Geological Survey
2. North Dakota State University
?
Contact author: [email protected]

Keywords: streamflow, environmetrics, clustering, trends, hydrology

Trends in streamflow characteristics (flood peaks, critical low flows, annual-mean flows) are often discussed
in various climate change scenarios, whether that climate change is human induced or the result of natural
climatic cycles. Increasing or decreasing trends are important for emergency management, agricultural, and
municipal supply, as well as other industrial uses. Trends are often summarized graphically with a few sites
highlighted in a paper, for example, or summarized in tables of trend values. However, when looking at a
large number of sites with varying drainage basins sizes over a large geographic region, the number of sites
involved makes these graphical and tabular methods difficult to comprehend and summarize.
R is used here to calculate a number of streamflow-related variables and extract features from the daily
streamflow time series of approximately 500 long-term, mostly unregulated, streamgage sites operated by
the U.S. Geological Survey. Cluster analysis is then performed to identify regionally similar areas and to
identify sites that may be geographically distant but experiencing similar trends, such as a downward trend
in streamflow because of large-scale irrigation which depletes groundwater and baseflow to the streams. A
number of cluster variables and techniques (including techniques in cluster and pvclust) are explored and
the results compared to existing knowledge of hydrologic trend and regional differences. The clustering is
then visualized geospatially (using maptools and rgdal) with other hydrologic and geographic data. The
clustering and visualization then allows researchers to focus on key sites in the regions (sites with the longest
periods of record, largest basins, critical sites where flooding has economic impacts, and so on), as well as
identify outliers. Outliers are of interest because they could be sites with unique conditions such as extensive
land use change, urbanization, or regulation not indicated in the source database.
Panel Discussion I, Wednesday 09:50-10:50 53

Panel on Starting & Building a Local R User Group

Derek McCrae Norton, Atlanta RUG, groups.google.com/group/atlanta-r-users

Szilard Pafka, Los Angeles RUG, www.meetup.com/LAarea-R-usergroup/
Richard Pugh, Mango Solutions, www.mango-solutions.com
David Smith, Revolution Analytics, www.revolutionanalytics.com

In the past two years the a number of local R User Groups (RUG’s) have started or ramped up their mem-
bership. This panel discussion will focus on actual experiences:
• Building interest in your community
• The kick off meeting
• Finding speakers
• Finding sponsors
• Promotion tricks
• Organizational tricks & traps
• Joys of RUG’ing
Audience participation & questions will be encouraged!

Call for panelists: If you would like to participate on the panel, please contact Derek Norton.
Kaleidoscope IIb, Wednesday 09:50-10:50 54

RTextTools
Loren Collingwood,1,2,* Tim Jurka3, Amber Boydstun3, Emiliano Grossman4,
and Wouter Van Atteveldt5
1. Political Science Department, University of Washington
2. Center for Statistics and the Social Sciences, University of Washington
3. Political Science Department, University of California, Davis
4. Centre of European Studies, Sciences Po
5. Communication Science Department, Vrije Universiteit
*Contact author: [email protected]

Keywords: Supervised Learning, Text Analysis, Machine Learning, Social Science

Machine learning has only recently entered the world of social and political science. For years,
scholars have used undergraduate research assistants for various classification tasks—such as
labeling congressional bill titles, classifying parliamentary speeches, and coding party
platforms. Many social scientists are in search of new tools to automate tedious coding tasks,
and social scientists have now begun using supervised learning to automate the labeling of
documents. RTextTools is a recently developed R package that provides a uniform interface to
several existing R algorithms to label text.

From the tm package, we include functions to generate document term matrices, stopword
removal, and stemming. Currently, the package includes standardized training and
classification access to svm, NaiveBayes, glmnet, randomForest, tree, AdaBoost, Bagging, and
nnet algorithms. In addition, we include a C++ maximum entropy train and classification
function. We also provide a function for cross validating each algorithm. Finally, several
accuracy and ensemble agreement functions are provided to examine how well each algorithm
does in terms of predictive accuracy and ensemble agreement. Researchers can quickly identify
which text documents are coded with high degrees of accuracy and which documents need to
be coded by humans for active learning.
Kaleidoscope IIb, Wednesday 09:50-10:50 55

The Role of R in Lab Automation

?
Jason Waddell1 , Tobias Verbeke1,
1. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: lab automation, LIMS, R, software integration, R Service Bus

Wet labs of life science companies increasingly make use of advanced measurement technologies that are
organized in automated workflows. Many of the steps in these workflows require statistical analyses prior to
feeding the results of one step into the next step. Also, the final experiment data require automated reporting
to support scientists in the experiment assessment and decision making.
The advantages of using R in this context are well known as it provides companies with cutting edge sta-
tistical algorithms that can be directly transferred from Academia to the workbench in industry. In this
presentation we will share our experiences in bridging R and lab equipment computers, LIMS systems and
end user applications from a variety of settings. We will distill best practices from examples in the analysis
of RTqPCR data, EEG signal processing, behavioral experiment data or compound activity screens. The
secret weapon to support the diverse workflows and computational requirements is the R Service Bus, an
open source integration tool specifically designed to cope with R integration exercises.
OpenAnalytics (2010–2011). The R Service Bus. http://www.openanalytics.eu/r-service-bus.
Kaleidoscope IIb, Wednesday 09:50-10:50 56

Using R data functions with TIBCO Spotfire

Peter Shaw1*, Louis Bajuk-Yorgan1

1. TIBCO Spotfire Software

*Contact author: [email protected]

Keywords: Spotfire, R

R scripts and functions can be called from TIBCO Spotfire, and the results returned to the Spotfire
environment. Spotfire is a highly dynamic, interactive graphical environment for visualizing data, with
filtering and drill-down capabilities. A user can thus manipulate data, drill down and highlight data of
interest, and seamlessly send this subset to R for analysis, returning the results to Spotfire.

The combination of Spotfire with R scripts and functions enables powerful, dynamic analyses. Spotfire
provides a convenient means of manipulating data and selecting the data for analysis; R extends the
capabilities of Spotfire. The results returned by R to Spotfire can be any combination of numeric (e.g.,
model coefficients, forecasts etc), text (e.g., summary diagnostics) or R graphical objects.

We will show how to install an R script and a CRAN package into Spotfire and demonstrate the
resulting R-Spotfire interactive environment, including an example of an R graphic returned into
Spotfire.
Spatio-Temporal Statistics, Wednesday 11:15-12:35 57

spTimer: Spatio-Temporal Bayesian Modelling

using R

Khandoker Shuvo Bakar & Sujit K. Sahu

April 1, 2011

Abstract
Hierarchical Bayesian modelling of large point referenced space-time data
are increasingly becoming feasible in many environmental applications due
to the recent advances in both statistical methodology and computation
power. Bayesian model based analysis methods using the Markov chain
Monte Carlo (MCMC) techniques are, however, a very formidable task for
large data sets rich in both space and time. Most such analyses are best per-
formed using user written computer code in a low level language that takes
hours and sometimes days to run in most personal computers. Currently
there does not exist an R package which can fit and predict space-time data
effectively, although the package spBayes (Finley, et al., 2007)can analyse
some moderately sized data sets.
This paper develops the R package, spTimer specifically for modelling spatio-
temporal data. This allows us to fit and forecast temporal data over large
spatial domains even in mid-range personal computers available today. The
package implements a recently developed Bayesian hierarchical autoregres-
sive model (Sahu, et al., 2007) suitable for moderately sized problems. For
modelling even larger space-time data, a predictive process approximation
method proposed by Banerjee, et al., (2008) is extended, implemented and
illustrated with a large data set on ozone monitoring data observed in the
eastern United States. mean square error results over other competitive
methods currently available in the literature.

1
Spatio-Temporal Statistics, Wednesday 11:15-12:35 58

Applying geospatial techniques to temporal data

Jason Lessels? , Thomas Bishop, Michael Nelson
The Faculty of Agriculture, Food and Natural Resources. The University of Sydney. Australia.
?
Contact author: [email protected]

Keywords: Water quality, Geospatial, Generalised linear mixed models

Studies involving water quality often focus on load estimation to quantify the characteristics of a catchment,
whilst government guidelines often provide concentration threshold targets. Water quality studies often
involve the use of unequally spaced temporal data, which prohibits the use of traditional time series methods.
Currently simple linear regression models are the most commonly used to estimate concentration levels
through time. The use of such models fail to account for the temporal auto-correlation in the observations
or provide meaningful results in regards to threshold guidelines. This paper will demonstrate the use of
spatial generalised liner (mixed) models to account for the temporal auto-correlation. There are currently
two packages within R to provide statistically sound methods to improve water quality estimation through
time. The package geoR provides the ability to use a general linear model with a temporal auto-correlation
structure to estimate concentration through time. The second package geoRglm provides the ability to use
a generalised linear mixed model (GLMM) to estimate the probability of a threshold being exceeded. Both
models account for the temporal auto-correlation, where simple linear regression models have failed. In
addition the GLMM provides a new method for the estimation of the probability of threshold exceedence
through time.
Spatio-Temporal Statistics, Wednesday 11:15-12:35 59

Structured Additive Regression Models:

An R Interface to BayesX
?
Nikolaus Umlauf1, , Thomas Kneib2 , Stefan Lang1 , Achim Zeileis1
1. Department of Statistics, Universität Innsbruck, Austria
2. Department of Mathematics, Carl von Ossietzky Universität Oldenburg, Germany
?
Contact author: [email protected]

Keywords: MCMC, geoadditive models, mixed models, space-time regression, structured additive regres-
sion.

Structured additive regression (STAR) models provide a flexible framework for modeling possible nonlinear
effects of covariates: They contain the well established frameworks of generalized linear models (GLM)
and generalized additive models (GAM) as special cases but also allow a wider class of effects, e.g., for
geographical or spatio-temporal data. This allows for the specification of complex and realistic models that
can typically be conveniently estimated using Bayesian inference based on modern Markov chain Monte
Carlo (MCMC) simulation techniques or a mixed model representation.
Although there is already a quite extensive existing toolset in R supporting GLMs and GAMs, many of the
more complex models from the STAR class, especially those utilizing Bayesian inference, are currently not
easily available. They are, however, provided in the standalone software package BayesX: a very compre-
hensive Bayesian semiparametric regression toolbox based on open-source C++ code. BayesX not only
covers models for responses from univariate exponential families, but also models from non-standard re-
gression situations such as models for categorical responses with either ordered or unordered categories,
continuous time survival data, or continuous time multi-state models.
Since there has been increasing interest in an R interface to BayesX, the already existing CRAN package
BayesX, which previously provided only functions for exploring estimation results, is now extended to a
full interactive interface. With the new version of the package, STAR models can be conveniently specified
using R’s formula language (with some extended terms), fitted using the BayesX binary, represented in R
with objects of suitable classes, and finally printed/summarized/plotted.
The talk outlines the usage of the R interface to BayesX and its application in complex regression problems,
emphasizing its strength in estimating and visualizing models with geographical effects.

References
Brezger, A., T. Kneib, and S. Lang (2005). BayesX: Analyzing Bayesian structural additive regression
models. Journal of Statistical Software 14(11), 1–22. http://www.jstatsoft.org/v14/i11/

Brezger, A. and S. Lang (2006). Generalized structured additive regression based on Bayesian P-splines.
Computational Statistics & Data Analysis 50, 967–991.

Fahrmeir, L., T. Kneib, and S. Lang (2004). Penalized structured additive regression for space-time data: A
Bayesian perspective. Statistica Sinica 14, 731–761.

Kneib, T. (2005). Mixed model based inference in structured additive regression. Dr. Hut-Verlag, München,
available at http://edoc.ub.uni-muenchen.de/5011/.
Molecular and Cell Biology, Wednesday 11:15-12:35 60

The R package isocir for Isotonic Inference for Circular Data.

Applications to Problems Encountered in Cell Biology.
?
Sandra Barragán1, , Cristina Rueda1 , Miguel A. Fernández1 , Shyamal D. Peddada2
1. Universidad de Valladolid, Spain.
2. National Institute of Environmental Health Sciences, USA.
?
Contact author: [email protected]

Keywords: Circular Data, Isotropic Order, CIRE, Conditional Test, R package isocir.

Estimation of angular parameters when they are intrinsically ordered around a unit circle is a question of
great interest for several researchers. Standard statistical methods, developed for Euclidean parameter space,
are not directly applicable for circular data. Particularly, in the presence of restrictions among the param-
eters, estimators and hypotheses tests of hypotheses have to be properly defined in order to cope with the
peculiarities of circular data. Motivated by applications to problems encountered in cell biology, Rueda
et al. (2009) introduced the notion of isotropic order and developed a methodology for estimating param-
eters under this constraint. Given the recent interest among cell biologists in identifying cell cycle genes
that are conserved among multiple species, Fernandez et al. (2011) developed a methodology for dealing
with isotropic testing problems. The R package isocir provides a user fiendly software for running all these
methods in any context where circular data may appear.

References
Fernandez, M., C. Rueda, and S. Peddada (2011). Isotropic order among core set of orthologs conserved
between budding and fission yeast. Preprint.

Rueda, C., M. Fernandez, and S. Peddada (2009). Estimation of parameters subject to order restrictions
on a circle with application to estimation of phase angles of cell-cycle genes. Journal of the American
Statistical Association 104(485), 338–347.
Molecular and Cell Biology, Wednesday 11:15-12:35 61

CircNNTSR: An R Package for the Statistical Analysis of

Circular Data Based on Nonnegative Trigonometric Sums
★
Juan José Fernández-Durán1,2, , Marı́a Mercedes Gregorio-Domı́nguez1,3
1. Instituto Tecnológico Autónomo de México, ITAM
2. Department of Statistics and School of Business
3. Department of Actuarial Science
★
Contact author: [email protected]

Keywords: Circular Data, Fourier Series, R Software.

In Fernández-Durán (2004), a new family of circular distributions based on nonnegative trigonometric sums
(NNTS models) is developed. Contrary to the great majority of families of circular distributions, this family
allows to model datasets that present multimodality and/or skewness. Initially, the maximum likelihood
estimates of the parameters of the NNTS family were obtained by algorithms based on Sequential Quadratic
Programming (SQP) that were difficult to implement in R and take a long time to converge. Because the
parameter space of the NNTS family is the surface of the hypersphere, an efficient Newton-like algorithm
on manifolds is generated in order to obtain the maximum likelihood estimates of the parameters. This
algorithm is implemented in the R package CircNNTSR. Examples of the application of CircNNTSR
for testing seasonality and homogeneity in problems in biology, actuarial science and environmetrics are
presented.

References
[1] Fernández-Durán, J.J. (2004) Circular Distributions Based on Nonnegative Trigonometric Sums. Bio-
metrics, 60, pp. 499-503.

[2] Fernández-Durán, J.J. and Gregorio-Domı́nguez, M.M. (2010) Maximum Likelihood Estimation of
Nonnegative Trigonometric Sum Models Using a Newton-like Algorithm on Manifolds. Electronic
Journal of Statistics, 4: 1402-1410.

[3] Fernández-Durán, J.J. and Gregorio-Domı́nguez, M.M. (2010b) CircNNTSR: An R package for the
statistical analysis of circular data using nonnegative trigonometric sums (NNTS) models. R package
version 1.0-1. http://CRAN.R-project.org/package=CircNNTSR

[4] R Development Core Team. R: A language and environment for statistical computing. R Foundation for
Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org/, 2010.
Molecular and Cell Biology, Wednesday 11:15-12:35 62

Summary statistics selection for ABC inference in R

?
Matthew A. Nunes1, , David J. Balding2
1. Department of Mathematics & Statistics, Lancaster University, Lancaster, U.K.
2. UCL Genetics Institute, University College London, London, U.K.
?
Contact author: [email protected]

Keywords: Approximate Bayesian Computation, entropy minimization, summary statistics

For the purposes of statistical inference, high-dimensional datasets are often summarized using a number
of statistics. Due to the intractable likelihoods involved in models for these complex datasets, this is often
performed using Approximate Bayesian Computation (ABC), in which parameter inference is achieved by
comparing the summaries of an observed dataset to those from simulated datasets under a chosen model
(Beaumont et al., 2002).
The question of how best to summarize high-dimensional datasets to maximize potential for inference has
been recently addressed in Nunes and Balding (2010). The authors propose two techniques to select a good
set of summaries for ABC inference from a collection of possible statistics.
Since high-dimensional datasets arise in many areas of science, these methods could provide practical guid-
ance to scientists on how to represent datasets for optimal inference. The minimum entropy (ME) and two-
stage error prediction methods introduced in Nunes and Balding (2010) are implemented for the “rejection-
ABC” algorithm (Tavaré et al., 1997) in the ABCME R package. The package also includes code to perform
optional “regression adjustments” for the mean and variance of parameter values which can improve overall
quality of inference of ABC algorithms (Fan and Yao, 1998; Yu and Jones, 2004; Beaumont et al., 2002).
Some of the computational cost of the ABC algorithm and summary selection procedures is reduced through
the use of C routines.
We provide an example of the techniques in the ABCME package, demonstrating posterior inference of the
parameters from a coalescent model of population DNA sequences (Nordborg, 2007).

References
Beaumont, M. A., W. Zhang, and D. J. Balding (2002). Approximate Bayesian Computation in population
genetics. Genetics 162(4), 2025–2035.

Fan, J. and Q. Yao (1998). Efficient estimation of conditional variance functions in stochastic regression.
Biometrika 85(3), 645–660.

Nordborg, M. (2007). Coalescent theory. In D. J. Balding, M. J. Bishop, and C. C (Eds.), Handbook of

Statistical Genetics (3rd ed.)., pp. 179–208. Wiley: Chichester.

Nunes, M. A. and D. J. Balding (2010). On optimal selection of summary statistics for Approximate
Bayesian Computation. Stat. Appl. Genet. Mol. Biol. 9(1).

Tavaré, S., D. J. Balding, R. C. Griffiths, and P. Donnelly (1997). Inferring coalescence times from DNA
sequence data. Genetics 145(2), 505–518.

Yu, K. and M. C. Jones (2004). Likelihood-based local linear estimation of the conditional variance function.
J. Am. Stat. Assoc. 99(465), 139–144.
Molecular and Cell Biology, Wednesday 11:15-12:35 63

Power and minimal sample size for multivariate analysis of

microarrays
?
Maarten van Iterson1, , José A. Ferreira2 , Judith M. Boer1,3,4 and Renée X. Menezes5
1. Center for Human and Clinical Genetics, LUMC, Leiden
2. RIVM, Bilthoven
3. Department of Pediatrics Oncology and Hematology, Erasmus MC Sophia Childrens Hospital, Rotterdam
4. Netherlands Bioinformatics Center
5. Department of Epidemiology and Biostatistics, VUmc, Amsterdam.
?
Contact author: M.van [email protected]

Keywords: average power, sample size determination, high-throughput data, kernel deconvolution esti-
mator, moderated t test statistics.

Choosing the appropriate sample size for high-throughput experiments, such as those involving microar-
rays and next-generation sequencing is complicated. Traditional univariate sample size determinations
relate power and significance level to sample size, effect size and sample variability. However, for high-
dimensional data these quantities need to be redefined: average power instead of power, significance level
needs to take multiple testing into account, and both effect sizes and variances have many values.
Some authors (see Ferreira and Zwinderman, 2006 and Dobbin and Simon, 2005) have proposed such
methods for two-group comparisons of high-dimensional data. The most general of those, by Ferreira and
Zwinderman, uses the entire set of test statistics from pilot data to estimate the effect size distribution, power
and minimal sample size, as opposed to most other published methods that either fix an effect size of in-
terest, or assume (partial) homoscedasticity. Ferreira and Zwinderman assume that the test statistics follow
a normal distribution, which is unlikely to hold in practice as many comparisons involve a Student-t test
statistic, and the performance of the method in such cases was not evaluated.
We aimed at a generalization of power and sample size estimation more applicable to high-throughput ge-
nomics data. First, we extended Ferreira and Zwindermans method to the case of a Student-t test. Second,
we considered t-test statistics generated by testing if a coefficient of a general linear model is equal to zero.
Furthermore, we considered Student-t tests that use a shrunken variance estimator, such as those produced
by empirical Bayes linear models as implemented in the BioConductor package limma (Smyth, 2005).
These extensions represent a considerable improvement on the power and sample size estimation compared
to when the normal assumption is used, which we illustrate via a simulation study. Finally, we will ex-
tend the method to generalized linear models aimed at power and sample size estimation for RNA-seq data.
The extensions will be implemented as part of our BioConductor package SSPA (van Iterson et al., 2009),
forming a valuable tool for experimental design of microarray experiments.

References

Dobbin K, Simon R. (2005). Sample size determination in microarray experiments for class comparison
and prognostic classification. Biostatistics. 6(1):27-38.
Ferreira JA, Zwinderman A. (2006). Approximate sample size calculations with microarray data: an illus-
tration. Statistical Applications in Genetics and Molecular Biology. 5(25).
Smyth GK (2005). Limma: linear models for microarray data, Bioinformatics and Computational Biology
Solutions using R and Bioconductor, R. Gentleman and V. Carey and S. Dudoit and R. Irizarry and W.
Huber, Springer, New York, 397–420.
van Iterson et al. (2009). van Iterson M, ’t Hoen PA, Pedotti P, Hooiveld GJ, den Dunnen JT, van Ommen
GJ, Boer JM, Menezes RX. Relative power and sample size analysis on gene expression profiling data.
BMC Genomics. 10:439.
Mixed Effect Models, Wednesday 11:15-12:35 64

Kenward-Roger modification of the F-statistic for some linear

mixed models fitted with lmer
?
Ulrich Halekoh1 , Søren Højsgaard1
1. Department of Molecular Biology and Genetics, Aarhus University, Denmark
?
Contact author: Ulrich Halekoh

Keywords: Kenward-Roger, degrees of freedom, linear mixed models, bootstrap tests

For linear mixed models a frequent task is to test for the reduction of a model by the removal of some fixed
effect. For models fitted with the lmer function of the R package lme4 (Bates et al., 2011) such tests can be
performed by via a maximum likelihood ratio test and the approximate χ2 distribution of the test statistic.
This approximation may yield rather anti-conservative tests (Pinheiro and Bates, 2000, p. 88) with er-
roneously small p-values. An alternative is to use F-tests which are exact in some balanced situations.
Kenward and Roger (1997) proposed a modification of the F-test statistic possibly shrinking the statistic and
adjusting the residual degrees of freedom in order to achieve a better approximation to a F-distribution. Our
function implements such an approximation for linear mixed models fitted with lmer. The implementation
is restricted to covariance structures which can be expressed as a linear combination of known matrices.
Such models comprise variance component and random coefficient models.
It has been shown in simulation studies (Kenward and Roger, 1997; Spilke et al., 2005) that sometimes the
Kenward-Roger modification represents a satisfactory approximation to the F-distribution. But others have
argued (e.g. in a discussion on the R help list ) that it is generally not clear that such an approximation holds
under general circumstances.
We offer therefore additional functions to calculate a p-value via parametric bootstrap and possible improve-
ments by the Bartlett correction of the test statistic.
In the talk we will shortly depict the theory and show some applications also pointing to some numerical
problems we encounter.
The functions will be made available via the CRAN package doBy (Højsgaard and Halekoh, 2011).

References
Bates, D., M. Maechler, and B. Bolker (2011). lme4: Linear mixed-effects models using S4 classes. R
package version 0.999375-39.

Højsgaard, S. and U. Halekoh (2011). doBy: Groupwise summary statistics, general linear contrasts,
LSMEANS (least-squares-means), and other utilities. R package version 4.3.0.

Kenward, M. G. and J. H. Roger (1997). Small sample inference for fixed effects from restricted maximum
likelihood. Biometrics 53(3), 983–997.

Pinheiro, J. C. and D. M. Bates (2000). Mixed Effects Models in S and S-plus. Springer.

Spilke, J., H.-P. Piepho, and X. Hu (2005). A simulation study on tests of hypotheses and confidence inter-
vals for fixed effects in mixed models for blocked experiments with missing data. Journal of Agricultural,
Biological, and Environmental Statistics 10(3), 374–389.
Mixed Effect Models, Wednesday 11:15-12:35 65

lqmm: Estimating Quantile Regression Models for Independent

and Hierarchical Data with R
?
Marco Geraci1,
1. MRC Centre of Epidemiology for Child Health, UCL Institute of Child Health, London, UK
?
Contact author: [email protected]

Keywords: Quantile regression, random effects, longitudinal data, hierarchical models

Conditional quantile regression (QR) pertains to the estimation of unknown quantiles of an outcome as a
function of a set of covariates and a vector of fixed regression coefficients. In the last few years, the need for
extending the capabilities of QR for independent data to deal with clustered sampling designs (e.g., repeated
measures) has led to several and quite distinct approaches. Here, I consider the likelihood–based approach
that hinges on the strict relationship between the weighted L1 norm problem associated with a conditional
QR model and the asymmetric Laplace distribution (Geraci and Bottai, 2007).
In this presentation, I will illustrate the use of the R package lqmm to perform QR with mixed (fixed and
random) effects for a two–level nested model. The estimation of the fixed regression coefficients and of
the random effects’ covariance matrix is based on a combination of Gaussian quadrature approximations
and optimization algorithms. The former include Gauss-Hermite and Gauss-Laguerre quadratures for, re-
spectively, normal and double–exponential (i.e., symmetric Laplace) random effects; the latter include a
modified compass search algorithm and general purpose optimizers (optim and optimize). Modelling
and inferential issues are detailed in Geraci and Bottai (2011) (a preliminary draft is available upon request).
The package also provides commands for the case of independent data.

References

Geraci, M. and Bottai, M. (2007). Quantile regression for longitudinal data using the asymmetric Laplace
distribution. Biostatistics 8(1), 140–154.
Geraci, M. and Bottai, M. (2011). Linear quantile mixed models, in preparation.
Mixed Effect Models, Wednesday 11:15-12:35 66

lcmm: an R package for estimation of latent class mixed models

and joint latent class models
?
Cécile Proust-Lima1,2, , Benoit Liquet1,2
1. INSERM U897, Bordeaux, France
2. ISPED, Bordeaux Segalen University, France
?
Contact author: [email protected]

Keywords: classification, joint models, mixed models, growth mixture model, longitudinal data, non-
Gaussian data, time-to-event data.

The linear mixed model is routinely used to describe change over time of a quantitative outcome in longitu-
dinal studies. However, it is limited to the analysis of a quantitative Gaussian outcome, in a homogeneous
population, and does not handle association with a time-to-event although it is frequent in practice. The
R package lcmm extends the linear mixed model to (1) the study of heterogeneous populations through
the estimation of latent class mixed model, and (2) the joint analysis of longitudinal and time-to-event data
through the estimation of joint latent class models. In each case, both Gaussian or non Gaussian quantitative
and ordinal outcomes can be analysed.
Latent class mixed models consist in exploring the latent profiles of trajectories in heterogeneous popula-
tion. They combine the mixed models theory to account for the individual correlation in repeated measures,
and the latent class models theory to discriminate homogeneous latent groups when modelling trajectories.
Despite a large interest in this approach also known as growth mixture models, implementation in free soft-
wares is very limited. Within the lcmm package, the function hlme estimates latent class mixed models
assuming a quantitative Gaussian outcome and lcmm extends this approach to handle non Gaussian quanti-
tative and ordinal outcomes that are very frequent especially in psychological and quality of life studies.
Joint models to analyse jointly longitudinal and time-to-event data have also become increasingly popular
in statistics. There exist two kinds of joint models, shared random-effect models in which functions of the
random-effects from mixed model are included in the survival model, and joint latent class models which
assume the population is constituted of latent classes with a specific longitudinal outcome trajectory and
a specific risk of event. While an R package JM was recently developed to estimate shared random-effect
models, no free software exists for joint latent class models. We propose in lcmm the function jointlcmm
to estimate such joint models, both for Gaussian or non Gaussian quantitative and ordinal outcomes.
Whatever the models, estimation in lcmm is performed using a maximum likelihood method using a mod-
ified Marquardt algorithm with strict convergence criteria. Only for ordinal outcomes, a numerical integra-
tion is required. hlme, lcmm and jointlcmm allow any shape of trajectory, and covariates as predictors of
the latent class structure as well as of class-specific trajectories. The joint model includes different baseline
functions and common or class-specific covariate effects. Finally, posterior classification and goodness-of-
fit measures are provided within post estimation functions.

References

Proust, Jacqmin-Gadda (2005). Estimation of linear mixed models with a mixture of distribution for the
random-effects. Computer Methods and Programs in Biomedicine,78,165-73
Proust-Lima, Joly et al. (2009). Joint modelling of multivariate longitudinal outcomes & a time-to-event: a
nonlinear latent class approach, CSDA, 53, 1142-54
Proust-Lima & Taylor (2009). Development and validation of a dynamic prognostic tool for prostate cancer
recurrence using repeated measures of posttreatment PSA: a joint modeling approach, Biostatistics, 10,
535-49
Mixed Effect Models, Wednesday 11:15-12:35 67

Mixed-effects Maximum Likelihood Difference Scaling

?
Kenneth Knoblauch1 , Laurence T. Maloney2
1. INSERM, U846, Stem Cell and Brain Research Institute, Dept. Integrative Neurosciences, Bron, France
2. Psychology Department & Center for Neurosciences, New York University, New York, NY, USA
?
Contact author: [email protected]

Keywords: psychophysics, difference-scaling, glm, mixed-effects

Difference scaling is a psychophysical procedure used to estimate interval, perceptual scales along a stimulus
continuum. On each of a set of trials, an observer is presented with a quadruple, (Ia , Ib , Ic , Id ), drawn
from an ordered set of stimuli, {I1 < I2 < . . . < Ip }. The observer judges which pair shows the greater
perceptual difference. Alternatively, a trial can consist of a triple, (Ia , Ib , Ic ) and the observer judges whether
the difference between (Ia , Ib ) or (Ib , Ic ) appears greatest. The fitting procedure estimates perceptual scale
values, {ψ1 , ψ2 , . . . , ψp }, that best capture the observer’s judgments of the perceptual differences between
the stimuli. The perceived difference between a pair of stimuli, (Ia , Ib ), is modeled as the difference of
perceptual scale values, Lab = ψb − ψa . In comparing, pairs of pairs, we assume that the observer forms the
noise-contaminated decision variable, ∆ = Lab − Lcd + ,  ∼ N (0, σ 2 ) and judges the difference between
the pair (Ia , Ib ) to be greater than that between (Ic , Id ) precisely when ∆ > 0. The parameter σ represents
the observer’s precision in judgment. We estimate σ and the remaining free parameters, ψ2 , . . . , ψp , either
by directly maximizing the likelihood, using the optim function or, because the decision rule is linear, via
a generalized linear model, using glm. The package MLDS on CRAN implements these two approaches.
We extend these analyses to mixed-effects models, using glmer and lmer in the lme4 package. This
allows us to incorporate variability of observers and other sources as random effects, The design matrix
constructed for glm, however, distributes the stimulus levels over different columns, somewhat like a factor
variable, making it difficult to treat them as a single entity. We consider two strategies to deal with this
situation using glmer and one using lmer.
1. redefining the decision variable in terms of a parametric function that describes well the perceptual
scale and using this decision variable as a regressor in the model formula,
2. redefining the decision variable in terms of an observer’s own estimated perceptual scale and using
this decision variable as a regressor in the model formula,
3. fitting the estimated perceptual scale values as a linear mixed-effects model, using lmer.
We demonstrate the three approaches with example data sets and discuss their pros and cons.

References
Bates, D., M. Maechler, and B. Bolker (2011). lme4: Linear mixed-effects models using S4 classes. R
package version 0.999375-39/r1298.

Knoblauch, K. and L. T. Maloney (2008). MLDS: Maximum likelihood difference scaling in R. Journal of
Statistical Software 25, 1–26.

Maloney, L. T. and J. N. Yang (2003). Maximum likelihood difference scaling. Journal of Vision 3(8),
573–585.

Schneider, B., S. Parker, and D. Stein (1974). The measurement of loudness using direct comparisons of
sensory intervals. Journal of Mathematical Psychology 11, 259–273.
Programming, Wednesday 11:15-12:35 68

Tricks and Traps for Young Players

?
Ray Brownrigg1
1. School of Engineering and Computer Science, Victoria University of Wellington, NZ
?
Contact author: [email protected]

Keywords: programming, performance, vectorisation

This presentation will illustrate for new users to R some of its very useful features that are frequently over-
looked, and some frequently misunderstood features. Emphasis will be on achieving results efficiently,
so there may be some value for (moderately) seasoned users as well as beginners. Many of the features
discussed will be illustrated by following the development of an actual simulation project.
Issues to be discussed include:
• Vectorisation
– user-defined functions (using curve, optimise)
– pseudo vectorisation
– multi-dimensional vectorisation
• sort, order and rank
• When to use Fortran
• Local versions of standard functions
• Using a matrix to index an array
• Resolution of pdf graphs
• get
• file.choose
Note: This paper was presented at UseR’08 in Dortmund. It seemed to be very popular, so I feel it is worth
presenting the same material again.
Programming, Wednesday 11:15-12:35 69

Software design patterns in R

Friedrich Schuster1,*

1. HMS Analytical Software GmbH, Heidelberg, Germany

*Contact author: [email protected]

Keywords: Software engineering, Design patterns, Software quality, Development process

The long-term development of R-programs and packages not only leads to a growing code base, but
also to increased code complexity and stronger interdependencies between packages and functions. As a
result the required training times as well as the development and maintenance costs tend to rise
substantially, especially for large applications. In a corporate environment, software quality and
maintenance are a major cost factor. One answer to this challenge of software engineering are
"Software Design Patterns". Design patterns are reusable general solutions to commonly occurring
problems in software design. Design patterns assist the software development process in different ways:
(1) by providing tested and proven development paradigms, patterns speed up development and reduce
errors. (2) Patterns facilitate the familiarization with unknown code. (3) Patterns improve
communication between developers. This presentation will focus on well-known development patterns
and their applicability to R and on the identification of specific patterns for R.

References
Erich Gamma, Richard Helm, Ralph Johnson, John Vlissides (1994). Design Patterns. Elements of
Reusable Object-Oriented Software. Addison-Wesley.
Programming, Wednesday 11:15-12:35 70

Random input testing with R

Patrick Burns1,*

1. Burns Statistics
*Contact author: [email protected]

Keywords: Software testing, Random generation

Traditional software testing uses specific inputs and then sees if the results are correct. This is
necessary but not always sufficient to have faith that the software operates properly. When the number
of inputs is large, the combinatorial explosion means that full coverage is impossible.

An alternative form of testing is to create random inputs and then infer the suitability of the result. This
means many more combinations can be tested, and in particular avoids bias that may be in the
traditional test suite. Another advantage of this type of testing is that it exercises the code throwing
errors and warnings, which is seldom the case for traditional tests.

R is an excellent environment both for generating random inputs, and for examining the resulting
output. We'll highlight a specific example of portfolio optimization.
Programming, Wednesday 11:15-12:35 71

An Open Source Visual R Debugger in StatET

Stephan Wahlbrink1 , Tobias Verbeke2,?
1. WalWare / wahlbrink.eu
2. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: IDE, debugger, developer tools, R

The StatET plug-ins turn Eclipse into a first class integrated development environment for R programmers
and data analysts. It offers a set of mature tools for R coding and package building and provides a fully
integrated R Console, Object Browser and R Help System. In this paper we will present a visual debugger
for R that has been seamlessly implemented using the Eclipse debugger framework and allows developers
to visually step through the code and inspect variables and expressions in dedicated Eclipse views. We will
provide a detailed overview of all debugging functionality that has been implemented (over the last year)
and (convincingly) demonstrate the added value a visual debugger can bring to the R development process.
WalWare et al (2004 – 2011). Eclipse Plug-In for R: StatET. http://www.walware.de/goto/
statet/.
Data Mining Applications, Wednesday 11:15-12:35 72

Predicting the offender’s age

?
Stephan Stahlschmidt1,2,
1. Humboldt-Universität zu Berlin, School of Business and Economics, Chair of Statistics, Spandauer Straße 1, 10178 Berlin,
Germany
2. Ruprecht-Karls-Universität Heidelberg, Centre for Social Investment, Adenauerplatz 1, 69115 Heidelberg, Germany
?
Contact author: [email protected]

Keywords: Offender Profiling, Prediction Power, Statistical Learning

Predicting the age of an unknown offender is notoriously difficult for police profilers. These experts rely
solely on traces found on the crime scene and combine this information to deduce the exact events which
happened on the crime scene. Only afterwards they try to characterise the offender based upon the unassured
knowledge drawn from the crime scene. These statements on the offender are often based on asumptions
and lack certitude. However, an assured prediction on, for example, the offender’s age would prove very
useful to narrow the group of potential suspects.

We therefore compare the performance of several prediction techniques on a data set of sex-related homi-
cides. The applied techniques include linear regression, k-Nearest-Neighbour, regression trees, Random
Forest and Support Vector Machine. We evaluate each approach by its prediction power concerning the
offender’s age and account for the specific requirements of forensic data, namely the restricted access to
information via the crime scene and uncertainty of available information. Our results show, if and how the
police’s investigation could be enhanced by implementing one of these well-known prediction techniques.

References

Friedman, J., Hastie, T. and Tibshirani, R. (2010). Regularization Paths for Generalized Linear Models via
Coordinate Descent. Journal of Statistical Software, 33, 1-22.
Hornik, K. , Buchta, C. and Zeileis, A. (2009). Open-Source Machine Learning: R Meets Weka. Computa-
tional Statistics, 24, 225-232.
Karatzoglou, A., Smola, A., Hornik, K. and Zeileis, A. (2004). kernlab - An S4 Package for Kernel Methods.
Journal of Statistical Software, 11, 1-20.
Liaw, A. and Wiener, M. (2002). Classification and Regression by randomForest. R News, 2, 18–22.
Data Mining Applications, Wednesday 11:15-12:35 73

Leveraging Online Social Network Data and External Data

Sources to Predict Personality
Daniel Chapsky*

*Contact author: [email protected]

Keywords: Bayesian Networks, Personality, Facebook

People express their personalities through online social networks in a variety of ways, such as their
relationships with their friends and their listed interests. In this work I present a method for
automatically predicting an individual’s personality by combining his Facebook profile information
with external data sources using a machine learning method known as a Bayesian Network. The
developed models use representations of people’s connections to other people, places, cultures, and
ideas, as expressed through Facebook. Due to the nature of Bayesian Networks, the semantics
underlying the models are clear enough to not only predict personality, but also use knowledge of one’s
personality to predict his behavioral attributes and actions. I will present some of the more interesting
models of personality that my systems have produced thus far. These models demonstrate the potential
of my methodology in two ways: First, they are able to explain up to 70% of all variation in a
personality trait from a sample of 615 individuals. Second, they are able to clearly describe underlying
relationships in the model through findings such as how to predict a man’s agreeableness based on his
age, hometown, number of Facebook wall posts, and his willingness to disclose his preference for
music made by Lady Gaga.

I will also present the necessary background in Bayesian Networks and Personality theory to understand
the above results, and present how all data collection and modeling was automated using an academic
edition of REvolution R Enterprise and various R packages.

Figure 1: An Example Bayesian

Network modeling personality.
Blue nodes are personality traits.
Dotted lines denote a positive
relationship while straight lines
denote a negative one.
Data Mining Applications, Wednesday 11:15-12:35 74

Using R to Model Click-Stream Data to Understand Users' Path

To Conversion
Douglas Galagate1*,2 , Professor Wolfgang Jank1,3

1. University of Maryland, College Park

2. Department of Mathematics
3. Robert H. Smith School of Business
*Contact author: [email protected]

Keywords: Click-stream data, data mining, advertising, internet

Advertisers spend a lot of time and effort directing consumers' attention to their client's
website, enticing them to purchase a product or a service. While most models attribute a
consumer's conversion only to the last site they visited ("last click model"), it is generally
accepted that all of the information in a consumer's search path (i.e. in the sequence of all
websites visited) play some role in the ultimate purchase decision. In this project, we use data
mining techniques to characterize and model a consumer's "path to conversion." That is, we
characterize the network of websites and sequence of clicks in order to gauge the impact of
different types of online content, its order and its relationship to one another on the probability
of a consumer's conversion. We use many R packages in the analysis.

References

Agresti, Alan (2002). Categorical Data Analysis

Wickham, Hadley (2009). ggplot2: Elegant Graphics for Data Analysis (Use R)
Liu, Bing (2010). Web Data Mining: Exploring Hyperlinks, Contents, and Usage Data (Data-Centric
Systems and Applications)
Development of R, Wednesday 16:00-17:00 75

Packaging R for Ubuntu: Recent Changes and Future

Opportunities
?
Michael A. Rutter1,
1. Department of Mathematics, Penn State Erie, The Behrend College
?
Contact author: [email protected]

Keywords: R, CRAN, Packages, Ubuntu

Ubuntu is a GNU/Linux distribution based on Debian Linux that has become popular on desktops, servers,
and in cloud-based environments. In addition to being free and open source, one of the important features
of Ubuntu is the packaging system. The packaging system allows users and administrators to easily install
and update software packages while automatically handling the software dependencies that may be needed
in order for the software to run.
CRAN currently has binary packages of the latest versions of R as well as the recommended R packages
in both 32-bit (i386) and 64-bit (amd64) versions for a number of Ubuntu releases, including the latest two
long term service (LTS) releases. Due to changes in hardware availability, the building process for these
packages has changed. In this presentation, I will detail how the build process has moved from personal
servers to Launchpad, a suite of tools provided by Ubuntu. Launchpad includes the ability to build binary
packages for a variety of different architectures and Ubuntu releases. These builds are then synced to CRAN,
allowing them to be mirrored on various servers across the world.
However, even with the strong packaging system of Ubuntu, only a small number of the 2,000 plus R
packages are available via CRAN or the main Ubuntu repositories. Installing additional packages can be
done within R, but this sometimes requires tracking down the required libraries in Ubuntu. In addition, once
these packages are installed they are not automatically updated during Ubuntu system updates, which may
cause users to miss bug fixes or feature updates. In order to make more packages available to R users via
Ubuntu, I will also discuss the possibility of using Launchpad as the back-end of an automated package
building system for Ubuntu. Using components of cran2deb, developed by Charles Blundell and Dirk
Eddelbuettel, it should be possible to create the source packages on a single server and allow Launchpad
do to the heavy lifting of building the packages for multiple architectures. Initially, the number of packages
available will be limited to some of the more popular packages (e.g. ggplot2) and those packages used
in popular R references. While this type of resource will be helpful for users new to R and Ubuntu, an
additional goal is to provide the packages needed to easily deploy R in cloud-based Ubuntu environments.

References

cran2deb website: http://debian.cran.r-project.org/

Launchpad website: https://launchpad.net/
Development of R, Wednesday 16:00-17:00 76

Interpreter Internals: Unearthing Buried Treasure with CXXR

Andrew Runnalls1
1. School of Computing, University of Kent, UK, [email protected]

Keywords: R, CXXR, C++, packages, bignum

CXXR (www.cs.kent.ac.uk/projects/cxxr) is a project to reengineer the interpreter of the R language, cur-

rently written for the most part in C, into C++, whilst as far as possible retaining the full functionality of
the standard R environment, including compatibility with existing R packages (cf. Runnalls, 2010). It is in-
tended that by reorganising the code along object-oriented lines, by deploying the tighter code encapsulation
that is possible in C++, and by improving the internal documentation, the project will open up new avenues
of development for the R project. Development of CXXR started in May 2007, then shadowing R 2.5.1; at
the time of this abstract it reflects the functionality of R 2.12.1.
From the beginning, different kinds of R object have been represented within the CXXR interpreter using a
C++ class hierarchy, with the intention that this hierarchy could be extended by R developers. This reflects
two strategic objectives:
• Given a C++ class (provided for example by a third-party library) it should be straightforward to
introduce an R (CXXR) class wrapper around it, so that the functionality of that class can be exploited
within the R language framework.
• Conversely, given an existing R class (provided typically by an R package), it should be straightfor-
ward to migrate its functionality—to whatever extent desired—into a new underlying C++ class.
Unfortunately, much C code within the standard R interpreter that could greatly support these objectives is
not offered to packages as an API (Application Program Interface), and is not documented to API standards.
Moreover, in large measure this code is hard-wired around the built-in R data types (REALSXP etc.), and so
cannot easily be applied to other C++ data types.
This paper will explain how CXXR is endeavouring to change this situation by rewriting key pieces of
interpreter code at a higher level of abstraction (often using C++ templates), and making it available to
package writers via carefully documented interfaces.
Current progress will be illustrated by considering how to implement a package allowing the use of ‘BigInts’
(integers of arbitrarily large magnitude) within R. This is a task that has already been ably accomplished,
building on the standard R interpreter, by Lucas et al. as part of their gmp package, which utilises GMP, the
GNU Multiple Precision Arithmetic Library. But the paper will show how the CXXR framework makes it
very much easier to establish key elements of the necessary functionality.

References

Andrew Runnalls (2010). CXXR and Add-on Packages. http://user2010.org/slides/Runnalls.

pdf. Presented at useR! 2010, Gaithersburg, MD.
Antoine Lucas, Immanuel Scholz, Rainer Boehme and Sylvain Jasson (2010). Package ‘gmp’. http:
//cran.r-project.org/web/packages/gmp/index.html.
Torbjörn Granlund et al. (2010). The GNU Multiple Precision Arithmetic Library. http://gmplib.
org/
Development of R, Wednesday 16:00-17:00 77

R’s Participation in the Google Summer of Code 2011

?
Claudia Beleites1,2, , . . . , and John C. Nash3
1. CENMAT and Dept. of Industrial and Information Engineering, University of Trieste, Trieste/Italy
2. Institute of Photonic Technology, Jena/Germany
3.Telfer School of Mangement, University of Ottawa, Ottawa/Canada
?
Contact author: [email protected]

Keywords: R packages, Google Summer of Code

We plan to present this year’s Google Summer of Code projects that take place under the umbrella of The R
Project for Statistical Computing. The work will start end of May and finish in August (suggested “pencils
down” at August 16th ). Thus the UseR! 2011 conference will be a good occasion to present the brand new
tools and packages and also to make the google summer of code more widely known to the user community
of R.
Clearly the deadline for abstracts for the UseR! 2011 conference is prior to the final assignment of the
students (April 25th ), so this “dummy” abstract will be expanded with a more detailed one on the GSoC-R
projects page once we have active projects.
Geospatial Techniques, Wednesday 16:00-17:00 78

Converting a spatial network to a graph in R

Binbin Lu, Martin Charlton
National Centre for Geocomputation, National University of Ireland, Maynooth
*Contact author: [email protected]

Keywords: spatial network, graph, igraph, shp2graph

Spatial networks have been an important subject of many studies in quantitative geography and
sociology (Barthelemy, 2010). In practical applications, they are traditionally managed in a geospatial
vector format, in which spatial entities are recorded as polylines that are spaghetti collections of
2D/3D geospatial coordinates (George & Shekhar, 2008). On the other hand, a network can be
regarded as a set of nodes, occupying particular positions and joined in pairs by physical or ephemeral
constructs (Gastner & Newman, 2006). In this sense, networks are visualized in a link-node mode,
termed as graph data model. Due to its great advantages in simplifying the representation of a network
and facilitating related computations powered by graph theory, this model has been widely used as
one of the central tools in spatial analysis (Hostis, 2007). Based on the R platform, a considerable
number of packages have emerged for processing both kinds of objects. On the R-spatial website
(Bivand, 2007), numerous packages are available for processing spatial data, especially like sp and
maptools. As for graph, there are two popular packages to create and manipulate undirected and
directed graphs, i.e. graph and igraph. However, there is no such thing to convert a spatial object to a
graph-class object. This work aims to fill this gap.

Overall, all the functions are encapsulated in an R package named shp2graph. Its main utility is to
convert a SpatialLines or SpatialLinesDataFrame object to a graphNEL, graphAM or igraph object.
The principle is to abstract geospatial details, select two endpoints of each spatialline as spatially
enabled nodes and then link them with an edge. Furthermore, the following issues have been provided
for this conversion:

 Attribute heritage. Properties are guaranteed to inherit properly.

 Structure optimization. As for the potential redundant info, like self-loops, multiple-edges and
pseudo-nodes, three optimization functions have been designed to remove them.
 Topology. Identified from the converting principle, topology errors could cause serious
connectivity problems, and correct topology is the key to a successful conversion.
Consequently, a self-test function of topology has been included in this package.
 Integration of data points and network. When we are dealing with a set of data points together
with a spatial network, they may not be closely related. In this case, how to integrate these
points into the network is concerned according to different accuracy requirements.

References

Barthelemy M (2010). Spatial Networks, http://arxiv.org/abs/1010.0302v2

George B, Shekhar S (2008). Road Maps, Digital. In: Shashi S and Hui X (Ed) Encyclopedia of GIS,
Springer-Verlag. New York.
Gastner MT, Newman MEJ (2006). The spatial structure of networks, The European Physical Journal
B 29, 247-252
Hostis AL (2007). Graph theory and representation of distances: chronomaps, and other
representations. In: Mathis P(Ed.), Graphs and Networks: Multilevel Modeling, Wiley-ISTE,
London.
Bivand R(2007), Spatial data in R, http://r-spatial.sourceforge.net/
Geospatial Techniques, Wednesday 16:00-17:00 79

Spatial modelling with the R–GRASS Interface

?
Rainer M Krug1,
1. Centre of Excellence for Invasion Biology, Stellenbosch University, South Africa
?
Contact author: [email protected]

Keywords: Ecological modelling, spatial, monte carlo simulation, GRASS, GIS

R is a widely used tool for analysis and simulation of non-spatial data, even considered by many as the
“lingua franca” for statistical modelling. But for working with spatial data, the toolbox provided by R does
not consist of all the functionality to do complex analysis in R directly, unless one is prepared to spend a
substantial amount of time with programming routines. This problem extends to the storage of spatial data,
where, especially when dealing with large areas, the data sets can become extremely large and one is quickly
confronted with memory problems. As a solution to these problems, interfaces with different GIS programs
are provided for in R. These interfaces (for linking to GIS spgrass6, rsaga, importing and exporting rgdal,
sqlitemap and others, but also spgrass6, and rsaga) provide functions to import the data into R and export it
again into a format readable by the GIS, but also to utilise the functions of the GIS applications from within
R, i. e. to use R as a scripting language for those GIS applications. This scripting becomes an important tool
in the analysis and simulation of spatial data, as large maps can be manipulated without reading them into
R, instead using functions of GIS programs which were written to accommodate large data sets (there is a
package in R (raster) which uses the same approach, i. e. only loading smaller sections of the whole map
into memory and processing those, but it obviously does not have all the functions included in dedicated
GIS applications).
As elegant the solution is of using R as a scripting language to combine the power of R with the power
of GIS applications like GRASS and SAGA, it has still problems. These, however, show only when these
interfaces are used intensively.
To illustrate my points and to highlight a few areas which the R–GRASS interface could be improved, I will
present a study conducted to investigate the impact of different budgets and management scenarios on the
effectiveness of alien plant clearing. The study was conducted using a spatio-temporal explicit simulation
model, covering areas of up to 215.000 hectares with a cell size of one hectare. The simulation used several
vector and raster input layers, and generated hundreds of raster output layers. The interface between R and
GRASS (spgrass6) was used intensively, as certain processes were not available in R or the implementation
in GRASS was considerably faster due to implementation in C.
Main aspects identified during the project as “could be improved” range from direct reading of GRASS
data without the need of additional software and simple implementation of processing with a MASK to the
general question of backends for spatial data to parallisation especially of spatial routines.
Geospatial Techniques, Wednesday 16:00-17:00 80

sos4R - Accessing Sensor Web Data from R

?
Daniel Nüst1,2,
1. 52◦ North Initiative for Geospatial Open Source Software GmbH
2. Institute for Geoinformatics, University of Muenster, Germany
?
Contact author: [email protected]

Keywords: Sensor Observation Service, Open Geospatial Consortium, Web-service Client, XML, Repro-
ducible Research

The Sensor Web is a collection of standards and software solutions that allows the management of and data
retrieval from sensors through the internet. The Open Geospatial Consortium’s (OGC) Sensor Observation
Service (SOS) is the core web service to provide sensor data in an interoperable, standardized way. Because
its goals are so generic, this service is relatively complex, e.g. it needs other standards for request and query
markup, and standards for data markup and encoding. This complexity makes it hard for non-Sensor Web
specialists to benefit from publishing their data using SOSs, or analyzing data retrieved from other’s or even
someone’s own SOS instances. The recently published sos4R CRAN package tries to overcome this hurdle
by providing a relatively simple R interface to the SOS. The main contribution of the package is a set of
R functions for the core SOS operations for data retrieval. These suffice for many common use cases by
encapsulating the complexity of the SOS interface.
This work presents the technical background of accessing a XML-based web service from R using the
packages RCurl and XML. It shows how the generic work flow of request building, encoding, data transfer
and data decoding can be modelled in R classes and methods. Flexible mechanisms allow users to easily
add features and adapt entire processing steps, or just the required parts, to their needs.
We also present example analyses based on publicly available SOS that illustrate the potential and advan-
tages of building analyses and visualizations in R directly on SOS. These programmes can be based on near
real-time data, but can also be the base for reproducibility of analyses. Reproducible research is already
supported well by a variety of R packages, and these are now further complemented with open, online data
sources such as those found in the Sensor Web.
Finally, we describe future work, especially on the integration of the SOS respectively OGC data models
with existing and upcoming endeavours for spatio-temporal data in R, like the spacetime CRAN package.

References
Botts, M., G. Percivall, C. Reed, and J. Davidson (2008). OGC Sensor Web Enablement: Overview and
High Level Architecture. In S. Nittel, A. Labrinidis, and A. Stefanidis (Eds.), GeoSensor Networks,
Volume 4540 of Lecture Notes in Computer Science, pp. 175–190. Springer Berlin / Heidelberg.

Lang, D. T. (2007). R as a Web Client the RCurl package. Journal of Statistical Software.

Lang, D. T. (2010). XML: Tools for parsing and generating XML within R and S-Plus. online. version
3.2-0.

Nüst, D. (2010). sos4R - The OGC Sensor Observation Service Client for the R Project.

Open Geospatial Consortium, Inc. (2010). Sensor Observation Service. http://www.

opengeospatial.org/standards/sos.

Pebesma, E. (2010). spacetime: classes and methods for spatio-temporal data. R package version 0.1-6.
Genomics and Bioinformatics, Wednesday 16:00-17:00 81

MALDIquant: Quantitative Analysis of MALDI-TOF

Proteomics Data
?
Sebastian Gibb1, and Korbinian Strimmer1
1. Institute for Medical Informatics, Statistics and Epidemiology (IMISE), University of Leipzig, Härtelstr. 16–18, D-04107
Leipzig, Germany
?
Corresponding author: [email protected]

Keywords: Proteomics, mass spectrometry, MALDI-TOF, Bruker *flex.

MALDI-TOF is a well established technology for mass spectrometric profiling of proteomics data. There are
several ongoing efforts to provide open-source analysis software for proteomics studies, such as OpenMS
(Kohlbacher et al., 2007) or PROcess (Li, 2005). However, a complete analysis pipeline for MALDI-TOF
data for the R platform is lacking.
Common tasks and challenges in mass spectrometric analysis are data input, normalization, calibration,
peak peaking and other preprocessing task. Due to technological limitations the intensity values of identified
peaks cannot be directly compared across multiple spectra. This renders MALDI-TOF data difficult to use
for quantitative analysis.
Here we introduce the MALDIquant R package for analysis of MALDI-TOF proteomics data. Our package
provide routines for importing native Bruker *flex binary format, baseline removal, peak picking and, most
importantly, procedures for coherent assignment of intensity values across multiple spectra. This approach
generalizes the widely used standard procedures that rely on total ion count calibration or 0/1 truncation of
peak intensities. The analysis pipeline is illustrated as well as compared with its competitors by classification
of cancer proteomics data from the University Hospital Leipzig.
MALDIquant and associated R packages are available from CRAN.

References
Kohlbacher, O., K. Reinert, C. Gröpl, E. Lange, N. Pfeifer, O. Schulz-Trieglaff, and M. Sturm (2007).
TOPP–the OpenMS proteomics pipeline. Bioinformatics 23(2), 191–197.

Li, X. (2005). PROcess: Ciphergen SELDI-TOF Processing. Bioconductor R package archive. R package
version 1.26.0.
Genomics and Bioinformatics, Wednesday 16:00-17:00 82

QuACN: Analysis of Complex Biological Networks using R

Laurin AJ Mueller1,∗,† , Karl G Kugler1,† , Matthias Dehmer1
1. Institute for Bioinformatics and Translational Research, UMIT, Hall in Tirol, Austria
?
Contact author: [email protected] † contributed equally

Keywords: Network analysis, bioinformatics, network topology, information-theory

Classical analysis of biological data is mainly based on investigating single, isolated features. As it is now
understood that most diseases are driven by sets of interacting genes or proteins this has shifted towards a
more complex and holistic perception of this problem Strohman (2002). The first step in network-based anal-
ysis of complex biological data is inferring valid and robust network representations of the data. A plethora
of packages for this task are available in R, e.g. Langfelder and Horvath (2008); Meyer et al. (2008); Altay
and Emmert-Streib (2010). A topological analysis of the network provides new insights, as the structure
of a network represents the biological function. A quantitative approach is to use so called topological net-
work descriptors, which represent a network structure by a numeric value. Note, that different descriptors
capture different structural patterns of the network topology. A small selection of topological network de-
scriptors is available in the igraph package Csardi and Nepusz (2006). Recently we published the package
QuACN on CRAN Mueller et al. (2011). QuACN provides a multitude of different topological network
descriptors. We want to draw ones attention to the group of parametric graph entropy measures Dehmer and
Emmert-streib (2008); Dehmer and Mowshowitz (2011), which are exclusively available in QuACN. These
measures assign a probability value to each vertex of a graph by using information functionals to calculate
the graph’s information-content. Topological network descriptors can be used for a multitude of applications
in the structural analysis of (biological) networks, e.g. supervised and unsupervised machine learning, or
the integrative analysis of networks. Here, we present selected examples of descriptor-based approaches.

References
Altay, G. and F. Emmert-Streib (2010). Inferring the conservative causal core of gene regulatory networks.
BMC Systems Biology 4(1), 132.

Csardi, G. and T. Nepusz (2006). The igraph software package for complex network research. InterJour-
nal Complex Systems, 1695.

Dehmer, M. and F. Emmert-streib (2008). Structural information content of networks : Graph entropy based
on local vertex functionals. Computational Biology and Chemistry 32, 131–138.

Dehmer, M. and A. Mowshowitz (2011, January). A history of graph entropy measures. Information
Sciences 181(1), 57–78.

Langfelder, P. and S. Horvath (2008, January). WGCNA: an R package for weighted correlation network
analysis. BMC Bioinformatics 9, 559.

Meyer, P. E., F. Lafitte, and G. Bontempi (2008). minet: A R/Bioconductor package for inferring large
transcriptional networks using mutual information. BMC Bioinformatics 9, 461.

Mueller, L. A., K. G. Kugler, A. Dander, A. Graber, and M. Dehmer (2011, November). QuACN - An
R Package for Analyzing Complex Biological Networks Quantitatively. Bioinformatics (Oxford, Eng-
land) 27(1), 140–141.

Strohman, R. (2002). Maneuvering in the Complex Path from Genotype to Phenotype. Science 296, 701–
703.
Genomics and Bioinformatics, Wednesday 16:00-17:00 83

Investigate clusters of co-expressed and co-located genes at a

genomic scale using CoCoMap
Marion Ouedraogo 1 , Frédéric Lecerf 2 , Sébastien Lê 3
1. INRA, Agrocampus Ouest UMR 0598 Génétique Animale, Rennes, France
2. Agrocampus Ouest, INRA UMR 0598 Génétique Animale, Rennes, France
3. Agrocampus Ouest, CNRS UMR 6625 Mathématiques Appliquées, Rennes, France
?
Contact author: [email protected]

Keywords: Multivariate analysis, Biostatistics, Genomics, Genes co-expression, Genes co-location

The study of the genome structure and its role in the gene function regulation had revealed new insights in
the regulation of genes expressions by their chromosomal locations. The regulation of the genes expression
is in some part due to the genome architecture. The organization of the chromosomes within the nucleus al-
lows interactions between distant regions. Those interacting regions might present some co-expressed genes.
The biological interest is to identify the co-located genes within a region which present a co-expression. In
our data, genes are considered as statistical variables and individuals as statistical units; we will say that
genes are co-expressed if they induce the same structure on the individuals.
The aim of the CoCoMap package is to identify co-expression between genomic regions. It uses transcrip-
tomic data and the corresponding chromosomal locations of the genes and is mainly based on multivariate
exploratory approaches:
i) A sequence of Principal Component Analyses among the genome is used to identify the co-located genes
which present a co-expression.
We introduce the autovariogram representation to identify and discriminate regions of co-expressed genes.
ii) The Multiple Factor Analyses and its features are used to describes the co-expression between the chro-
mosomal regions.
The final output is the groups of co-located genes which present a co-expression. The method has proven
its efficiency on simulated data. Preliminary analyses on two different experimental sets identified some
common clusters in interaction.
The CoCoMap package proposes an interface to the method and its features to investigate transcriptomic
data.

References

Baker M. (2011). Genome in three dimensions. Nature 470: 289–294

De S. and Babu M. (2010). Genomic neighbourhood and the regulation of gene expression. Current Opin-
ions in Cell Biology, 22:326–333.

Sexton T. et al. (2009). Genomic interactions: Chromatin loops and gene meeting points in transcriptional
regulation. Seminars in Cell & Developmental Biology 20: 849–855
Regression Modelling, Wednesday 16:00-17:00 84

Beta Regression: Shaken, Stirred, Mixed, and Partitioned

?
Achim Zeileis1 , Bettina Grün2,3 , Francisco Cribari-Neto4
1. Department of Statistics, Universität Innsbruck
2. Institut für Statistik, Ludwig-Maximilians-Universität München
3. Institut für Angewandte Statistik, Johannes Kepler Universität Linz
4. Departamento de Estatı́stica, Universidade Federal de Pernambuco
?
Contact author: [email protected]

Keywords: beta regression, finite mixture, proportion, rate, model-based recursive partitioning

The class of beta regression models is commonly used by practitioners to model variables that assume val-
ues in the open standard unit interval (0, 1). It is based on the assumption that the dependent variable is
beta-distributed with its mean and precision parameters depending on some regressors. We explore various
flavors of parametrizations for the dependence of the response on the regressors. To shake and stirr, either a
single or double index model is employed for mean and/or precision, i.e., the parameters are related to po-
tentially different sets of regressors through a linear predictor plus link function(s). This approach naturally
incorporates features such as heteroskedasticity or skewness which are commonly observed in data taking
values in the standard unit interval, such as rates or proportions. Furthermore, additional heterogeneity in
the data can be captured by mixing or partitioning: If covariates are available that explain the heterogeneity,
a model-based recursive partitioning approach can be employed. If not, latent class regression is an alter-
native. All flavors of models are implemented in the R package betareg, leveraging the flexmix and party
packages.

References
Cribari-Neto, F. and A. Zeileis (2010). Beta regression in R. Journal of Statistical Software 34(2), 1–24.

Grün, B. and F. Leisch (2008). Flexmix version 2: Finite mixtures with concomitant variables and varying
and constant parameters. Journal of Statistical Software 28(4), 1–35.

Zeileis, A., T. Hothorn, and K. Hornik (2008). Model-based recursive partitioning. Journal of Computa-
tional and Graphical Statistics 17(2), 492–514.
Regression Modelling, Wednesday 16:00-17:00 85

Regression Models for Ordinal Data:

Introducing R-package ordinal
?
Rune Haubo B Christensen1,
1. Technical University of Denmark, Department of Informatics and Mathematical Modelling, Section for Mathematical Statistics,
Richard Petersens Plads, Building 305, Room 122, DK-2800 Kgs. Lyngby, Denmark
?
Contact author: [email protected]

Keywords: Cumulative Link Models, Mixed Effects, Location-Scale, Partial Proportional Odds

Ordered categorical data, or simply ordinal data, are commonplace in scientific disciplines where humans
are used as measurement instruments. Examples include school gradings, ratings of preference in consumer
studies, degree of tumor involvement in MR images and animal fitness in field ecology. Cumulative link
models (Agresti, 2002) are a powerful model class for such data since observations are treated rightfully
as categorical, the ordered nature is exploited and the flexible regression framework allows in-depth anal-
yses. A pertinent latent variable interpretation of cumulative link models is an important aspect in many
applications in sensometrics, psychometrics and other social sciences. Cumulative link (mixed) models are
implemented in functions clm and clmm in package ordinal (Christensen, 2011).
The MASS function polr is a popular implementation of basic cumulative link models taking its name
from the proportional odds model—a cumulative link model with a logit link. clm and clmm extends polr
in a number of ways by allowing for random effects, scale effects, nominal effects, flexible link functions
and structured thresholds, further, several estimation methods are available including a fast Newton scheme.
Mixed effects models are estimated with standard Gauss-Hermite quadrature, the highly accurate adaptive
Gauss-Hermite quadrature or the flexible Laplace approximation accommodating nested as well as crossed
random effect structures.
Collectively these options facilitate a fuller analysis of ordinal data. The model framework embrace location-
scale models (McCullagh, 1980; Cox, 1995), allows for so-called partial proportional odds (Peterson and
Harrell Jr., 1990), facilitates inference for the link function (Genter and Farewell, 1985) and allows assess-
ment of linearity of the response scale. profile likelihood methods help visualize the likelihood function
and provide accurate confidence intervals via a confint method. dropterm, addterm and anova
methods facilitate model comparison. The implementation is primarily in R-code with computer intensive
parts in C.

References
Agresti, A. (2002). Categorical Data Analysis (Second ed.). Wiley.

Christensen, R. H. B. (2011). ordinal—regression models for ordinal data. R package version 2010.12-15
http://www.cran.r-project.org/package=ordinal/.

Cox, C. (1995). Location-scale cumulative odds models for ordinal data: A generalized non-linear model
approach. Statistics in medicine 14, pp. 1191–1203.

Genter, F. C. and V. T. Farewell (1985). Goodness-of-link testing in ordinal regression models. The Cana-
dian Journal of Statistics 13(1), pp. 37–44.

McCullagh, P. (1980). Regression models for ordinal data. Journal of the Royal Statistical Society, Series B
42, pp. 109–142.

Peterson, B. and F. E. Harrell Jr. (1990). Partial proportional odds models for ordinal response variables.
Applied Statistics 39, pp. 205–217.
Regression Modelling, Wednesday 16:00-17:00 86

Multiple choice models: why not the same

answer? A comparison among LIMDEP, R,
SAS and Stata
1
Giuseppe Bruno

1. Bank of Italy Economic Research and International relations. [email protected]

Keywords: Multinomial models, Probit, Gibbs sampler, MCMC.

The recent past has seen a great deal of research into categorical response models espe-
cially in the econometrics and behavioral literature. The most popular categorical response
models are the Multinomial Logit (MNL) and the Multinomial Probit (MNP). The choice
space for these models can be dichotomous (yes or no) or polytomous (e.g. points on a Likert
scale). When dealing with more than two choices, a trade-off comes out between numerical
tractability and model flexibility. In particular the more simple mathematical features of the
MNL model entail the validity of the Independence of Irrelevant Alternatives (IIA) assump-
tion. On the other hand, the adoption of the MNP model provides the maximum modeling
flexibility at a much higher computational price. In this paper we investigate the availability
and the accuracy of canned estimation algorithms for the multinomial models in the most
spread packages employed by the researchers at the Bank of Italy. Here we compare the clas-
sical Maximum Likelihood method along with the Markov Chain Monte Carlo (MCMC)
method available in the R package named MNP. A thorough comparison of the algorithms
available in the packages LIMDEP, R, SAS and Stata is provided along with accuracy
and performances indications. Preliminary results are quite insightful. As it is foreseeable,
the availability of a closed-form expression for the choice probabilities in the multinomial
exp(Xi ·βj +Cij λ)
logit model, shown in the following equation: P r(yi = j|Xi ) = ΣJ exp(X i ·βk +Cik λ)
, allows
k=1
the employment of stable and replicable algorithms. On the other hand the estimation of a
multinomial probit with Maximum Likelihood involves the integration of a multivariate nor-
mal distribution. This task is not numerically straightforward. Different packages provides
different answers and sometimes they don’t even provide an answer. R provides the MNP
package for fitting a Bayesian multinomial probit model with Markov chain Monte Carlo
methods (see Imai and van Dyk 2005) . SAS, STATA and LIMDEP fit the multinomial
probit model via Maximum Likelihood. Some empirical applications taken from the health
insurance and the travel mode choices are presented. It is shown how the employment of the
estimation algorithms for these models would benefit from the comparison with the imple-
mentation of the same algorithm in different packages and how helpful the comparison with
already available numerical benchmarks would be (see Econometric Benchmarks 2010).

References

Econometric Benchmarks (2010).

http://www.stanford.edu/~clint/bench/#logit.

G. Glasgow and R.M. Alvarez (2008). Discrete Choice Methods,

http://www.polsci.ucsb.edu/faculty/glasgow/glasgow_alvarez_final.pdf.
R in the Business World, Wednesday 16:00-17:00 87

Odysseus vs. Ajax: How to build an R presence in a corporate

SAS environment
?
Derek McCrae Norton1,
1. Norton Analytics
?
Contact author: [email protected]

Keywords: Business Analyitcs, Open Source, SAS

A lot has changed in the year since I gave my talk at the UseR!2010 [1]. There have been multiple prominent
articles about R as detailed in [2]. There has been a noticable decline in SAS and a steady increase in R as
detailed in [3]. As well as many other factors, but there is still resistance in the corporate environment.
There is certainly less than there was a year ago, but it is still there.

How does one break through the resistance?

By following some simple steps (simple doesn’t necessarily mean easy), you can build a strong R following
at your corporation. This work will provide those steps as well as some methods to implement them.
• Start Small.
• Spread the word.
• Show the ROI.
• Focus on what R does better than SAS.
• . . . Tune in to find more.

Last year, I likened R to David and SAS to Goliath, however I revise that now as shown in the title. Both R
and SAS are powerful forces in a war against poor analytics (Troy), but R is the crafty intelligent fighter like
Odysseus while R is the formitable, but brutish fighter like Ajax [4].

References

1. Revolution Analytics Blog - RMedia, http://blog.revolutionanalytics.com/rmedia/.

2. Norton, Derek M. (2010). David v. Goliath: How to build an R presence in a corporate SAS environment.
In useR! 2010, The R User Conference, (Gaithersburg, Maryland, USA).
3. Muenchen, Robert A. (2011). The Popularity of Data Analysis Software. http://sites.google.
com/site/r4statistics/popularity.
4. Homer. The Odyssey.
R in the Business World, Wednesday 16:00-17:00 88

A Validation/Qualification Solution for R

Michael O’Connell1, Ian Cook1,*

1. Spotfire, TIBCO Software Inc.

*Contact author: [email protected]

Keywords: validation, qualification, regulatory compliance, FDA, clinical trial

To date, R is not widely used in regulated environments, e.g., clinical trials for pharmaceuticals and
medical devices. A common misperception exists that R cannot support the various regulatory
requirements for validation/qualification. We present a straightforward framework for successfully
complying with regulatory software validation requirements including FDA 21 CFR Part 11 and other
GxP documents. Recognizing that validation/qualification applies to the software and to its installation
and operation, we present a solution that facilitates qualification of an R installation to meet IQ/OQ/PQ
standards. We cite previous useR! proceedings on the topic, and discuss the combination of factors
enabling growth in the use of R in regulated environments, including guidance from the R Foundation
and the availability of tools supporting validation/qualification.

References

FDA (2010). Code of Federal Regulations Title 21, Part 11,

http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?cfrpart=11.

Harrell, Frank E Jr. (2007). R for Clinical Trial Reporting: Reproducible Research, Quality and
Validation. useR! 2007 (Iowa State University),
http://biostat.mc.vanderbilt.edu/twiki/pub/Main/FHHandouts/dmcreport.pdf.

R Foundation for Statistical Computing (2008). R: Regulatory Compliance and Validation Issues, A
Guidance Document for the Use of R in Regulated Clinical Trial Environments, http://www.r-
project.org/doc/R-FDA.pdf.

Schwartz, Marc (2007). Use of R in Clinical Trials and Industry-Sponsored Medical Research. useR!
2007 (Iowa State University), http://user2007.org/program/presentations/schwartz.pdf.
R in the Business World, Wednesday 16:00-17:00 89

R as a statistical tool for human factor engineering

?
Enrico Branca1,
1. R-Adamant
?
Contact author: [email protected]

Keywords: R-Adamant, human resources, jobs, applied statistics

In the last decades global market has evolved toward a new dimension by creating an environment where
different countries , attitudes , cultures and work organization coexist together. Companies are so exposed
to a new management challenge to keep high business performances.

In this scenario Human Resources Management has become more than ever a key factor to build company’s
success. Tasks as: find and leverage talents, promote a culture of excellence by incouraging outstanding
achievements, assemble a collective knowledge by sharing experience for common good, reduce employee
turnovers are, among others, crucial to ensure company’s survival.
Considering that in some cases Universities have not been providing with enough graduates, offset the gap
in those sectors with skills shortage is a competitive investment in corporate branding.
As reported by a recent research at Yale School of Management in 2009: ”Those who graduate in bad
economies may suffer from underemployment and are more likely to experience job mismatching since they
have fewer jobs from which to choose. The disadvantage might be eliminated if workers can easily shift into
jobs and career paths.”
An analytical approach can be used to create a functional relation between the HR function and enterprise
risk management process. This allows to quantify and mitigate the possible risks related with human capital
that can affect company’s business performances.
Time series models (ARIMA, moving average) and multivariate analysis techniques (ANOVA, Generalised
linear models) are indeed used as a support for both HR managers and risk managers.
Statistical methods and tools widely used for Business Intelligence and financial risk management may
promote a new corporate culture more focused on Sense of Accomplishment and networks building than
compliance to static and outdated behavioural models.
The tutorial scope is to show how a valuable tool as R-Adamant can be a powerful ally for researchers and
university students as well as for companies.

References

Lisa B. Kahn (First Draft: March, 2003 Current Draft: August 13, 2009). The Long-Term Labor Market
Consequences of Graduating from College in a Bad Economy,Yale School of Management.
U.S. Department of Labor - U.S. Bureau of Labor Statistics (Summary 10-05 / June 2010), Issues in Labor
Statistics.
Council for Adult and Experiential Learning (CAEL) and Southern Regional Education Board (SREB)
(2007). Increasing Degree Completion Among Adult Learners: Policies and Practices to Build Greater
State Capacity. Chicago, IL: CAEL.
Hydrology and Soil Science, Wednesday 17:05-18:05 90

GWSDAT (GroundWater Spatiotemporal Data Analysis Tool)

Wayne Jones1, * , Michael Spence2

1. Shell Projects and Technology, Statistics and Chemometrics, Shell Technology Centre Thornton, Chester, United Kingdom
2. Shell Projects and Technology, Soil and Groundwater Group, Shell Technology Centre Thornton, Chester, United Kingdom
*Contact author: [email protected]

Keywords: Environmental monitoring, spatiotemporal modelling, application development, interface

with other languages.

The GroundWater Spatiotemporal Data Analysis Tool (GWSDAT) is a user-friendly software

application developed by Shell for the analysis and visualisation of visualise trends in environmental
(groundwater) monitoring data. In this presentation we will discuss the underlying software architecture
and demonstrate GWSDAT functionality which includes:

• Linking GWSDAT to Microsoft products for a user friendly application entry point (e.g. Excel,
PowerPoint, Word) and automatic report generation (R packages: RExcel and rcom).

• Graphical User Interface (R packages: tcltk and rpanel).

• Time series trend detection (R packages: sm, Kendall and zoo).

• Methods for visualising and handling Spatial data (R packages deldir, sp, splancs and maptools).

• Smoothing and spatial plot animations for spatiotemporal trend detection.

GWSDAT Screenshot.
Hydrology and Soil Science, Wednesday 17:05-18:05 91

IntR – Interactive GUI for R

Fabio Veronesi1,*, Ron Corstanje1

1. Cranfield University, Bld.37 - School of Applied Sciences, MK43 0AL, Cranfield, Bedfordshire
*Contact author: [email protected]

Keywords: GUI, interactive interface, R, geostatistics, python,

In recent years, R is increasingly being used by soil scientists. Its popularity is due to the fact that it
is free, open-source, but also due to the large number of packages dedicated to spatial statistics and
geostatistical interpolation. On the other hand, the necessity to learn a brand new programming
language, with a relatively steep learning curve, has kept and still keeps most soil scientists away from
R.

This software is designed to solve this problem and speed up the learning process. IntR is a graphical
user interface, wrote in Python, that eases the procedure of the creation and execution of an R script for
2D and 3D geostatistical interpolation. This interface works by asking “questions” to the user: for
instance to select the data file, the model of the variogram ect., and compiling the R script based on the
commands inserted. The user is therefore in full control of every part of the process. The script is run in
batch mode and the results are shown at the end of the process and saved. Normally the output of a
geostatistical analysis is a table of the results, an image of the variogram, and images of the prediction
and uncertainty maps and, where available, a plot of the observed versus the predicted values. The
script is also saved for the user to control it and thus learn the language. The algorithms that can be used
in IntR are: for the 2D version, inverse distance interpolation, ordinary kriging, universal kriging and
regression kriging, random forest and CART, plus a module for obtaining a point grid from a polygon
shape file and a module for the variogram and the anisotropy analysis. For the 3D version, the user can
choose between inverse distance, ordinary kriging and universal kriging, plus a module to create a 3D
prediction grid, a module for variogram and anisotropy analysis, a module for creating a series of slice
images of the 3D map and a module for sticking the slices in an animation video. The packages used to
perform all the analysis are: gstat, sp, maptools, randomForest, tree, rgdal, lattice, akima.

The software was developed with two different datasets, one in 2D and the other in 3D. The
bidimensional dataset is composed by 30 textural data samples (sand, clay and silt) collected at three
depth interval: 0-10 cm, 10-30 cm and 30-70 cm. The second dataset is composed by 57 cone-index
samples taken on a regular grid, with a vertical resolution of 4 cm. In both cases we used geophysical
covariates for the prediction, namely EM38, EM31 and Gamma-ray data. Regarding the 2D case study,
the results show that universal kriging is the best performer among the major prediction interpolators
incorporated into IntR. In the 3D case study, a comparison between a 3D ordinary kriging and 3D
universal kriging was undertaken. As expected, the best predictor in 3D universal kriging. For this
reason, universal kriging was used to create a 3D map of the Lany field.

References

Bivand R.S., Pebesma E.J., Gomez-Rubio V. (2008). Applied spatial data analysis with R. Springer,
NY.

ESRI (1998). ESRI Shapefile Technical Description.

Liaw A. and Wiener M. (2002). Classification and Regression by randomForest. In R News 2(3).

Pebesma, E.J. (2004). Multivariable geostatistics in S: the gstat package. In Computers & Geosciences,
30: 683-691.
Hydrology and Soil Science, Wednesday 17:05-18:05 92

Visualisation and modelling of soil data using the aqp package

?
Pierre Roudier1, , Dylan Beaudette2
1. Landcare Research, New Zealand
2. Natural Resources Conservation Service, USDA, USA
?
Contact author: [email protected]

Keywords: Soil science, visualisation, aggregation, classification

Soils support nearly all terrestrial food webs, are composed of a dynamic mixture of organic and mineral
constituents, and occur within the relatively fragile intersection between atmosphere, biosphere, and litho-
sphere. With the growth of the human population and a limited soil resource to feed those populations,
the study of such a complex environment arises as a prime challenge (Sanchez et al., 2009). However, the
importance of soils has been acknowledged for a long time. An impressive quantity of soil information has
been collected to support soil survey operations, natural resource inventories, and research over at least the
last 100 years. Soils are routinely sampled and studied according to characteristic layers (called horizons),
resulting in a complex data structure, with the dimensions of location, depth, and property space.
The high dimensionality and grouped nature of large soil profile collections can complicate standard anal-
ysis, summarization, and visualization, especially within a research community that is a traditional spread-
sheet user. Soil data manipulation and analysis is also complicated by difficulties associated with processing
horizon data that vary widely in depth and thickness. Scalability and reproducibility are also becoming sig-
nificant issues as the size of the databases grow. Finally, while the investigation of soil profile characteristics
and horizon-level morphology is strongly based on visual and tactile cues, the challenge of communicating
these data is traditionally addressed using written narrative or tabular form.
R is a suitable platform to address those challenges, and to develop tools that would provide soil scientists
with aggregation, modelling and visualisations for soil data. An R package, aqp (Algorithms for quantitative
pedology), has been developed to extend R’s methods to the specificities of soil data. Specialized S4 classes
have recently been added to support the multivariate hierarchy of linked spatial data (e.g. coordinates), site
data (e.g. landscape position), and horizon data (e.g. clay content at 10 cm). Various new aggregation and
classification methods are also available, and make use of the parallelised environment provided by the plyr
package (Wickham, 2011). Examples of the aqp functionalities are proposed on different soil databases,
for data visualisation, analysis and classification. The future developments of the package, especially its
interactions with other packages used for soil data analysis (e.g. sp Pebesma and Bivand, 2005), is also
discussed.

References
Pebesma, E. J. and R. S. Bivand (2005, November). Classes and methods for spatial data in R. R News 5(2),
9–13.

Sanchez, P. A., S. Ahamed, F. Carre, A. E. Hartemink, J. Hempel, J. Huising, P. Lagacherie, A. B. McBrat-

ney, N. J. McKenzie, M. d. L. Mendonca-Santos, B. Minasny, L. Montanarella, P. Okoth, C. A. Palm,
J. D. Sachs, K. D. Shepherd, T.-G. Vagen, B. Vanlauwe, M. G. Walsh, L. A. Winowiecki, and G.-L.
Zhang (2009). Digital Soil Map of the World. Science 325(5941), 680–681.

Wickham, H. (2011). plyr: Tools for splitting, applying and combining data. R package version 1.4.
Biostatistical Modelling, Wednesday 17:05-18:05 93

survAUC: Estimators of Prediction Accuracy for

Time-to-Event Data
?
Matthias Schmid1, , Sergej Potapov1 and Werner Adler1
1. Department of Medical Informatics, Biometry and Epidemiology, Friedrich-Alexander University Erlangen-Nuremberg
?
Contact author: [email protected]

Keywords: Survival analysis; prediction accuracy; risk prediction

The evaluation of predictions for continuous time-to-event outcomes has become a key interest in biostatis-
tical research. In contrast to the situation where predictions for uncensored outcomes have to be evaluated,
deriving measures of prediction accuracy is not straightforward in the presence of censored observations.
This is because traditional performance measures for continuous outcomes (such as the mean squared error
or the R2 fraction of explained variation) are biased if applied to censored data.
The survAUC package (Potapov et al. 2011) implements a variety of estimators to evaluate survival predic-
tive accuracy. The focus is on estimators of discrimination indices that measure how well a prediction model
separates observations having an event from those having no event (Pepe et al. 2008). In addition, survAUC
provides R functions to estimate likelihood-based coefficients (O’Quigley et al. 2005) and measures based
on scoring rules (Gerds and Schumacher 2006).

References

Potapov S, Adler W, Schmid M (2011). survAUC: Estimators of Prediction Accuracy for Time-to-Event
Data. R package version 1.0-0.
http://cran.at.r-project.org/web/packages/survAUC/index.html.
Pepe MS, Zheng Y, Jin Y, Huang Y, Parikh CR, Levy WC (2008). Evaluating the ROC performance of
markers for future events. Lifetime Data Analysis 14, 86–113.
Gerds TA, Schumacher M (2006). Consistent estimation of the expected Brier score in general survival
models with right-censored event times. Biometrical Journal 48, 1029–1040.
O’Quigley J, Xu R, Stare J (2005). Explained randomness in proportional hazards models. Statistics in
Medicine 24, 479–489.
Biostatistical Modelling, Wednesday 17:05-18:05 94

Higher-order likelihood inference in meta-analysis using R

?
Annamaria Guolo1 , Cristiano Varin2
1. Università degli Studi di Verona, Italy
2. Università Ca’ Foscari, Venezia, Italy
?
Contact author: [email protected]

Keywords: higher-order asymptotics, linear mixed-effects model, meta-analysis, Skovgaard’s statistic,

small sample inference

Standard likelihood inference is known to improve common non-likelihood techniques in meta-analysis

(Hardy and Thompson, 1996). Nevertheless, relying on typical first-order approximations, as for example
the χ2 distribution for Wald and likelihood ratio statistics, can give rise to inaccurate results. This draw-
back is a consequence of small sample sizes, which are typical in meta-analysis. Resorting to the theory
of higher-order asymptotics provides remarkably more precise results than first-order counterpart. See, for
example, Severini (2000) and Brazzale et al. (2007) for a general overview of higher-order asymptotics. We
present an R package which implements first-order likelihood inference and the second-order adjustment to
the log-likelihood ratio statistic of Skovgaard (2006), either for meta-analysis and meta-regression problems,
following the results in Guolo (2011). The package allows inference on fixed- and random-effect compo-
nents of linear mixed models used in meta-analysis. The functionality of the package will be illustrated on
a real example from the medical literature.

References

Brazzale, A R, Davison, A C, and Reid N (2007). Applied Asympotics: Case Studies in Small Sample
Statistics. Cambridge University Press: Cambridge.
Guolo A (2011). Higher-order likelihood inference in meta-analysis and meta-regression. Submitted.
Hardy R J, Thompson S G (1996). A likelihood approach to meta-analysis with random effects. Statistics
in Medicine 15, 619–629.
Severini, T A (2000). Likelihood Methods in Statistics. Oxford University Press: Oxford.
Skovgaard I M (1996). An explicit large-deviation approximation to one-parameter tests. Bernoulli 2, 145–
165.
Biostatistical Modelling, Wednesday 17:05-18:05 95

Gaussian copula regression using R

?
Guido Masarotto1 , Cristiano Varin2
1. Università degli Studi di Padova, Italy
2. Università Ca’ Foscari, Venezia, Italy
?
Contact author: [email protected]

Keywords: discrete time series; Gaussian copula; likelihood inference; longitudinal data; spatial data.

Marginal regression models for non-normal correlated responses are typically fitted by the popular gener-
alized estimating equations approach of Liang and Zeger (1986). Despite several theoretical and practical
advantages, likelihood analysis of non-normal marginal regression models is much less diffuse, see e.g.
Diggle et al. (2002). The main reason is the difficult identification of general classes of multivariate distri-
butions for categorical and discrete responses. Gaussian copulas provide a possible solution with a general
framework for modelling dependent responses of any type (Song, 2000). Gaussian copulas combine the
simplicity of interpretation in marginal modelling with the flexibility in the specification of the dependence
structure. Despite this, Gaussian copula regression had still a limited use since for noncontinuous depen-
dent responses the likelihood function requires the approximation of high-dimensional integrals. Masarotto
and Varin (2010) propose an adaptation of the Geweke-Hajivassiliou-Keane importance sampling algorithm
(Keane, 1994; Train 2003) to overcome the numerical difficulties of the likelihood inference.
The R package mr implements the methodology discusses in Masarotto and Varin (2010). The package
allows a flexible specification of the marginals and the dependence structure. At the time of writing, the
package contains methods for inference in regression models for longitudinal and clustered responses, time
series, spatially correlated observations and cross-correlated studies. The functionality of the package will
be illustrated on several real data examples arising in Biostatistics.

References

Diggle, P J , Heagerty, P, Liang, K-Y, Zeger, S L (2002). Analysis of Longitudinal Data. Second edition.
Oxford University Press: Oxford.
Keane, M P (1994). A computationally practical simulation estimator for panel data. Econometrica 62,
95–116.
Liang, K L, Zeger, S L (1986). Longitudinal data analysis using generalized linear models. Biometrika 73,
13–22.
Masarotto, G, Varin, C (2010). Gaussian dependence models for non-Gaussian marginal regression. Sub-
mitted.
Song, P X-K (2000). Multivariate dispersion models generated from Gaussian copula. Scandinavian Jour-
nal of Statistics 27, 305–320.
Train, K E (2003). Discrete Choice Methods with Simulation. Cambridge University Press: Cambridge.
Psychometrics, Wednesday 17:05-18:05 96

Multinomial Processing Tree Models in R

?
Florian Wickelmaier1,
1. Department of Psychology, University of Tübingen
?
Contact author: [email protected]

Keywords: Cognitive Psychometrics, Multinomial Models, Latent Variables, Categorical Data

Multinomial processing tree models are a class of statistical models for categorical data with latent parame-
ters. These parameters are the link probabilities of a tree-like graph and represent the cognitive processing
steps executed to arrive at observable response categories (Batchelder & Riefer, 1999; Erdfelder et al., 2009;
Riefer & Batchelder, 1988).
In this presentation, the mpt package (Wickelmaier, 2011) in R is introduced which provides functions for
fitting and testing such models. The model structure is represented symbolically using a simple formula
interface. Parameter estimation is carried out by the expectation-maximization algorithm described in Hu
and Batchelder (1994). The statistical procedures are illustrated using examples from cognitive psychology
and memory research.

Category pairs Singletons

r E1
c
1−r E4
u F1
u E2
u 1 − u F2
1 − u E3
1−c
u E3
1−u
1 − u E4

References

Batchelder, W. H. & Riefer, D. M. (1999). Theoretical and empirical review of multinomial process tree
modeling. Psychonomic Bulletin & Review, 6, 57–86.
Erdfelder, E., Auer, T., Hilbig, B. E., Aßfalg, A., Moshagen, M., & Nadarevic, L. (2009). Multinomial
processing tree models: A review of the literature. Zeitschrift für Psychologie, 217, 108–124.
Hu, X. & Batchelder, W. H. (1994). The statistical analysis of general processing tree models with the EM
algorithm. Psychometrika, 59, 21–47.
Riefer, D. & Batchelder, W. H. (1988). Multinomial modeling and the measurement of cognitive processes.
Psychological Review, 95, 318–339.
Wickelmaier, F. (2011). mpt: Multinomial processing tree (MPT) models. R package version 0.3-0.
http://CRAN.R-project.org/package=mpt
Psychometrics, Wednesday 17:05-18:05 97

Detecting Invariance in Psychometric Models with the

psychotree Package
?
Carolin Strobl1, , Florian Wickelmaier2 , Julia Kopf1 and Achim Zeileis3
1. Deparment of Statistics, LMU Munich
2. Psychological Institute, Universität Tübingen
3. Department of Statistics, Universität Innsbruck
?
Contact author: [email protected]

Keywords: Bradley-Terry Model, Rasch Model, Differential Item Functioning (DIF), Parameter Instabil-
ity, Model-Based Recursive Partitioning.

The psychotree package offers a statistical toolbox for detecting parameter differences in psychometric
models, including different worth parameters in Bradley-Terry models (Strobl, Wickelmaier and Zeileis,
2011) and differential item functioning (DIF) in the Rasch model (Strobl, Kopf and Zeileis, 2010a,b). The
method for detecting different worth parameters in Bradley-Terry models is implemented in the bttree
function, the DIF detection method for the the Rasch model is implemented in the raschtree function.
Both methods are based on a general model-based recursive partitioning framework employing generalized
M-fuctuation tests for detecting differences in the model parameters between different groups of subjects
(Zeileis and Hornik, 2007; Zeileis, Hothorn and Hornik, 2008). The main advantage of this approach is that
it allows to detect groups of subjects exhibiting different model parameters, that are not pre-specifed, but
are detected automatically from combinations of covariates. The talk outlines the statistical methodology
behind psychotree as well as its practical application by means of illustrative examples.

References
Strobl, C., J. Kopf, and A. Zeileis. A New Method for Detecting Differential Item Functioning in the
Rasch Model. Technical Report 92, Department of Statistics, Ludwig-Maximilians-Universität München,
Germany, 2010. URL: http://epub.ub.uni-muenchen.de/11915/.

Strobl, C., J. Kopf, and A. Zeileis. “Wissen Frauen weniger oder nur das Falsche? – Ein statistisches Modell
für unterschiedliche Aufgaben-Schwierigkeiten in Teilstichproben.” Allgemeinbildung in Deutschland –
Erkenntnisse aus dem SPIEGEL Studentenpisa-Test. Ed. S. Trepte and M. Verbeet Wiesbaden: VS Verlag,
2010, 255–272.

Strobl, C., F. Wickelmaier, and A. Zeileis. “Accounting for Individual Differences in Bradley-Terry Models
by Means of Recursive Partitioning.” (To appear.). Journal of Educational and Behavioral Statistics
(2011).

Zeileis, A., T. Hothorn, and K. Hornik. “Model-Based Recursive Partitioning.” Journal of Computational
and Graphical Statistics 17 (2008): 492–514.

Zeileis, Achim and Kurt Hornik. “Generalized M-Fluctuation Tests for Parameter Instability.” Statistica
Neerlandica 61 (2007): 488–508.
Psychometrics, Wednesday 17:05-18:05 98

Investigating multidimensional unfolding models using

R2WinBUGS
Shiu-Lien Wu1 , Wen-Chung Wang2

1. National Chung-Cheng University, Taiwan

2. The Hong Kong Institute of Education
*Contact author: [email protected]

Abstract

It has been argued that unfolding models could be more appropriate for describing responses to attitude
items or Likert items than dominance models. However, most existing unfolding models are either
unidimensional or fitting to binary data under the framework of item response theory (IRT). When there
are multiple tests of Likert items, or when a Likert item measures more than one latent trait
simultaneously, unidimensional unfolding models become inefficient or inappropriate. Meanwhile, if a
model can be fit only to binary data, the applications are limited. To resolve these problems, we
developed the confirmatory multidimensional generalized graded unfolding model, which is a
multidimensional extension of the generalized graded unfolding model (Roberts, Donoghue, &
Laughlin, 2000), and conducted a series of simulations to evaluate its parameter recovery by using R
and the R package R2WinBUGS. The simulation study demonstrated that the parameters of the new
model can be recovered fairly well. In addition, we analyzed a real data set about tattoo attitude to
depict the implication and applications of the new model and to demonstrate its advantages over the
unidimensional model. The results showed that the multidimensional model had a better fit than the
unidimensional one (log PsBF = 27.2); the multidimensional model yielded higher reliability estimates
(.92, .89, .83) for the 3 latent traits than the unidimensional one (.84, .85, .83); and the
multidimensional model yielded higher correlation estimates among the 3 latent traits (.20 ~ .84) than
the unidimensional model (.04 ~ .30).

Keywords: multidimensional item response theory, unfolding models, Markov chain Monte Carlo,
R2WinBUGS

References

Sturtz, S., Ligges, U., & Gelman, A. (2005). R2WinBUGS: A package for running WinBUGS from R.
Journal of Statistical Software, 12, 1-16.
Multivariate Data, Wednesday 17:05-18:05 99

Tests for Multivariate Linear Models with the car Package

?
John Fox1,
1. McMaster University, Canada
?
Contact author: [email protected]

Keywords: Multivariate Analysis of Variance, Repeated Measures, Linear Hypothesis Tests

It is straightforward to fit multivariate linear models in R with the lm function: Simply specify the left-
hand side of the linear-model formula as a matrix of responses. R also has facilities for testing multivariate
linear models via the anova function (as described in Dalgaard, 2007). Although the anova function is
very flexible, applied to a multivariate linear model it calculates sequential (often termed “type I”) tests,
which rarely are of interest, and performing other common tests, especially for repeated-measures designs,
is relatively inconvenient. In contrast, the Anova function in the car package (associated with Fox and
Weisberg, 2011) can perform partial tests for the terms in a multivariate linear model, either obeying the
principle of marginality (“type II” tests) or violating it (“type III” tests), including simply specified multi-
variate and univariate tests for repeated-measures models. In addition, the linearHypothesis function
in the car package can test arbitrary linear hypothesis for multivariate linear models, including models for
repeated measures. Both the Anova and linearHypothesis functions and their associated summary
methods return a variety of information useful in further computation on multivariate linear models, such as
the graphical display of hypothesis tests (see, e.g., Fox et al., 2009).

References
Dalgaard, P. (2007). New functions for multivariate analysis. R News 7(2), 2–7.

Fox, J., M. Friendly, and G. Monette (2009). Visualizing hypothesis tests in multivariate linear models: The
heplots package for r. Computational Statistics 24, 233–246.

Fox, J. and S. Weisberg (2011). An R Companion to Applied Regression. Sage Publication.

Multivariate Data, Wednesday 17:05-18:05 100

missMDA: a package to handle missing values in and with

multivariate exploratory data analysis methods
Julie Josse1 , Francois Husson1
1. Applied Mathematics Department, Agrocampus, Rennes, France
?
Contact author: [email protected]

Keywords: Missing values, Principal component methods, Multiple imputation, Confidence ellipses

In this presentation, we describe the missMDA package which is dedicated to handle missing values in
exploratory data analysis methods such as principal component analysis (PCA) and multiple correspondence
analysis. This package provides the classical outputs (scores, loadings, graphical representations, etc.) of
principal component methods despite the missing values. It also gives confidence areas around the position
of the points (individuals and variables) representing the uncertainty due to missing values. The package can
also be used to perform single or multiple imputation for continuous and categorical variables in a general
framework. In this presentation, we describe the underlying method through PCA.
A common approach to handle missing values in PCA consists in minimizing the loss function (the recon-
struction error) over all nonmissing elements. This can be achieved by the iterative PCA algorithm (also
named expectation maximization PCA, EM-PCA) described in Kiers (1997). It consists in setting the miss-
ing elements at initial values, performing the analysis (the PCA) on the completed data set, filling-in the
missing values with the reconstruction formula (the PCA model, Caussinus (1986)) and iterate these two
steps until convergence. The parameters (axes and components) and the missing values are then simul-
taneously estimated. Consequently, this algorithm can be seen as a single imputation method. To avoid
overfitting problems, regularized version of the EM-PCA algorithm have been proposed (Josse and Husson,
2011; Ilin and Raiko, 2010).
After the point estimate, it is natural to focus on the variability of the parameters. However, the variance
of the axes and components estimated from the completed data set (obtained with the EM-PCA algorithm)
is underestimated. Indeed, the imputed values are considered as observed values and consequently the
uncertainty of the prediction is not taken into account in the subsequent analysis. It is possible to resort to
multiple imputation (Rubin, 1987) to avoid this problem. Multiple imputation consists first in generating
different plausible values for each missing values. Then it consists in performing the statistical analysis on
each imputed data set and combining the results. Josse and Husson (2011) have proposed a new method to
generate multiple imputed data sets from the PCA model. They have also proposed two ways to visualize
the influence of the different predictions of the missing values onto the PCA results. It leads to confidence
areas around the position of the individuals and of the variables on the PCA maps.

References
Caussinus, H. (1986). Models and uses of principal component analysis. In J. de Leeuw, W. Heiser, J. Meul-
man, and F. Critchley (Eds.), Multidimensional Data Analysis, pp. 149–178. DSWO Press.
Ilin, A. and T. Raiko (2010). Practical approaches to principal component analysis in the presence of missing
values. Journal of Machine Learning Research 11, 1957–2000.
Josse, J. and F. Husson (2011). Multiple imputation in principal component analysis. Advances in Data
Analysis and Classification.
Kiers, H. A. L. (1997). Weighted least squares fitting using ordinary least squares algorithms. Psychomet-
rica 62, 251–266.
Rubin, D. B. (1987). Multiple imputation for non-response in survey. Wiley.
Multivariate Data, Wednesday 17:05-18:05 101

MAINT.DATA: Modeling and Analysing Interval Data in R

?
A. Pedro Duarte Silva1, , Paula Brito2
1. Faculdade de Economia e Gestão & CEGE, Universidade Católica Portuguesa at Porto, Porto, Portugal
2. Faculdade de Economia & LIAAD-INESC Porto LA, Universidade do Porto, Porto, Portugal
?
Contact author: [email protected]

Keywords: Symbolic data, Interval data, Parametric modelling of interval data, Statistical tests for interval
data, Skew-Normal distribution

In the classical model of multivariate data analysis, data is represented in a n × p data-array where n
“individuals” (usually in rows) take exactly one value for each variable (usually in columns). Symbolic
Data Analysis (Diday and Noirhomme-Fraiture (2008), Noirhomme-Fraiture and Brito (2011)) provided
a framework where new variable types allow to take directly into account variability and/or uncertainty
associated to each single “individual”, by allowing multiple, possibly weighted, values for each variable.
New variable types - interval, categorical multi-valued and modal variables - have been introduced. We
focus on the analysis of interval data, i.e., where elements are described by variables whose values are
intervals of IR. Parametric inference methodologies based on probabilistic models for interval variables
are developed in Brito and Duarte Silva (2011) where each interval is represented by its midpoint and log-
range, for which Normal and Skew-Normal (Azzalini and Dalla Valle (1996)) distributions are assumed.
The intrinsic nature of the interval variables leads to special structures of the variance-covariance matrix,
which are represented by five different possible configurations.
In this work, we introduce the package MAINT.DATA, which implements the proposed methodologies in R.
It introduces a data class for representing interval data. MAINT.DATA includes functions for modeling and
analysing interval data, in particular maximum likelihood estimation and statistical tests for the different
considered configurations. Methods for (M)ANOVA and Linear and Quadratic Discriminant Analysis of
this data class are also provided.

References

Azzalini, A. and Dalla Valle, A. (1996). The multivariate Skew-Normal distribution, Biometrika 83 (4), pp.
715–726.
Brito, P. and Duarte Silva, A.P. (2011). Modelling interval data with Normal and Skew-Normal distributions.
Journal of Applied Statistics, (in press).
Noirhomme-Fraiture, M. and Brito, P. (2011). Far Beyond the Classical Data Models: Symbolic Data
Analysis. Statistical Analysis and Data Mining (in press).
Diday, E. and Noirhomme-Fraiture, M. (2008). Symbolic Data Analysis and the SODAS Software, Wiley,
Chichester.
Interfaces, Wednesday 17:05-18:05 102

Web 2.0 for R scripts & workflows: Tiki & PluginR

?
Xavier de Pedro Puente1,2, , Alex Sánchez Pla1,2
1. Department of Statistics. Faculty of biology. University of Barcelona.
2. Statistics and Bioinformatics Unit. Vall d’Hebron Research Institute. Catalonia. Spain
?
Contact author: [email protected]

Keywords: GUI, Web 2.0, Free/libre Software, Tiki Wiki CMS Groupware, PluginR.

Abstract
The need to work with colleagues from other institutions is very common around Academia and
Science. Teams often find tools to communicate and coordinate with other web platforms to improve
collaboration across space and time. Although analysis and visualization of data with R is becoming
very popular, development teams frequently look also for web-based graphical user interfaces for the
end users of those R scripts. The list of prototypes and publicly announced free tools (R (2011)) includes
programs of all kinds. However, a quick review of these tools led us to similar conclusion reached by
other researchers such as Saunders (2009): most of these programs seem to present problems in the
short to medium term. Those problems arise from the fact that either such programs no longer work with
current stable versions of standard and free web technology, because its development seems to have been
discontinued for years. Or because they are too difficult to install or use for most scientists or people who
are not professionals in web technology. Therefore, we decided in our research groups to contribute to
the development of a relatively new approach, different from the latest approaches presented in the latest
years (Ooms (2009), Nakano and Nakama (2009) and others): a plugin for Tiki Wiki CMS Groupware
(also known as ”Tiki”), a mature collaborative web 2.0 framework released as free/libre open source
software, somewhat similar to the R extension for Mediawiki, but with all the extra features from this
“Tightly Integrated Knowledge Infrastructure“ that Tiki represents), along with its decentralized but truly
successful development model (Tiki (2011)). This new PluginR (De Pedro (2011)), has so far allowed
the development of a Web application of use in research on the Teaching and Learning field (De Pedro
et al. (2010)), as well as to develop web interfaces for Basic Pipelines in Bioinformatics for medical
research (De Pedro and Sánchez (2010b)). The communication will highlight the main advantages (and
disadvantages) found up to date with the use of Tiki + PluginR to solve many of the needs of our research
groups, including the new progresses achieved after the presentation at the last Spanish R Users meeting
(De Pedro and Sánchez (2010a))

References
De Pedro, X. (2011). Tiki documentation: Plugin r. https://doc.tiki.org/PluginR.

De Pedro, X., M. Calvo, A. Carnicer, J. Cuadros, and A. Miñarro (2010). Assessing student activity
through log analysis from computer supported learning assignments. In Proceedings of the Interna-
tional Congress of University Teaching and Innovation. http://cochise.bib.ub.es.

De Pedro, X. and A. Sánchez (2010a). Usando de forma segura r vı́a web con tiki. In II Jornadas de
Usuarios de R en Castellano. http://r-es.pangea.org/II+Jornadas.

De Pedro, X. and A. Sánchez (2010b). Using r in tiki for bioinformatics (*poster*). In Xth Spanish
Symposium on Bioinformatics (JBI2010). http://estbioinfo.stat.ub.es/?p=219.

Nakano, J. and E.-j. Nakama (2009). Web interface to r for high-performance computing. In The R UseR
Conference 2009, Rennes.

Ooms, J. (2009). Building web applications with r. In The R UseR Conference 2009, Rennes.

R (2011). R faq. cran.r-project.org/doc/FAQ/R-FAQ.html#R-Web-Interfaces.

Saunders, N. (2009). A brief survey of r web interfaces. http://nsaunders.wordpress.com/

2009/11/30/a-brief-survey-of-r-web-interfaces/.

Tiki (2011). Development model. https://tiki.org/Model.

Interfaces, Wednesday 17:05-18:05 103

Browser Based Applications Supported by R in Pipeline Pilot

DaveNicolaides1,*, Noj Malcolm1, Stephane Vellay1, Dana Honeycutt2, Tim Moran2

1. Accelrys Ltd., 334 Cambridge Science Park, Milton Road, Cambridge UK, CB4 0WN
2. Accelrys Inc., 10188 Telesis Court, Suite 100, San Diego, CA 92121, USA
*Contact author: [email protected]

Examples of the use of R as the “statistical engine” behind a simplified interface designed to be used by
scientists abound (Bajuk-Yorgan, Kaluzny, 2010; Weiss 2008; Neuwirth 2008; there are perhaps
another dozen examples in the useR! meetings over the last two years alone).

We wish to share the success we’ve had in deploying R more widely to the research community in
standard browsers, via our Pipeline Pilot informatics platform (Hassan et al. 2006). While there are
similarities between our approach and those referred to previously, we think that a reliance on the
scientific domain knowledge increasingly to be found in R packages distinguishes our approach. We
will give specific examples of the use of R behind the scenes, in such diverse areas of research as
pharmaceutical discovery, automated image classification, gene expression, heterogeneous catalysis,
and process analytical technology.

The success of these activities have largely stemmed from their approach to integration not as an
activity based on software technologies, but instead based on people and the ways they work and learn.
Allowing scientists to leverage a sophisticated statistics engine in a familiar interface gives them greater
insight into and confidence in their decisions. In seeking to extend this success we will have to address
many complex issues, including:

 distinguishing transfer of statistical expertise (where the goal is learning) from mere automation
(where the goal is research efficiency), and
 understanding how we might “enable ideas as software” in R, where those ideas already have a
strong presence in our own and other commercial software products.

References

Lou Bajuk-Yorgan, Stephen Kaluzny (2010). Making R accessible to Business Analysts with
TIBCO Spotfire. In useR! 2010, The R User Conference (Gaithersburg, Maryland), Book of Abstracts
p. 14.

ChristianWeiss (2008). Commercial meets Open Source - Tuning STATISTICA with R.In useR! 2008,
The R User Conference (Dortmund, Germany), Book of Abstracts p. 188.

Erich Neuwirth (2008). R meets the Workplace - Embedding R into Excel and making it more
accessible.In useR! 2008, The R User Conference (Dortmund, Germany), Book of Abstracts p. 135.

Hassan M, Brown RD, Varma-O'brien S and Rogers D. (2006). Cheminformatics analysis and learning
in a data pipelining environment. Moleculardiversity 10(3), 283-99.
Interfaces, Wednesday 17:05-18:05 104

A new task-based GUI for R

Sheri Gilley1,*

1. Principal UI Designer, Revolution Analytics

*Contact author: [email protected]

Keywords: GUI, Revolution R, user interface design

One of Revolution Analytics' main goals is to make data analysis done in R accessible to a wider
community of people. To that end, we have designed a graphical user interface for R.

Last year at useR!, I presented1 the design methodology used to develop a GUI for R. This year, I will
follow up on that theme by discussing some of the issues discovered during usability testing at various
stages in the development. A demo of the soon-to-be released product will show how these issues were
addressed, including:

• A task-based interface with an easy way to browse through the dialogs to find the task you want
to perform.

• Task dialogs to allow users to generate and run R code. This can be done without ever seeing
the code, although a user may click a button to reveal the code at any time.

• Script editor to write code that does not have a corresponding dialog.

• Workspace explorer to help you to learn more about your data as well as make it easy to drop
variables into a task or into code written in the script editor.

• Output and history also organized by the tasks.

References

1. Gilley (2010). Designing a flexible GUI for R.

http://user2010.org/abstracts/Gilley.pdf.  
 
Financial Models, Thursday 09:50-10:50 105

Computational aspects of continuous-time-arma (CARMA)

models: The ctarma package
Helgi Tomasson1
1. University of Iceland, Faculty of Ecnomics

?
Contact author: [email protected]

Keywords: Estimation, simulation, continuous-time, ARMA models.

Some computational aspects of the continuous-time ARMA, CARMA are reviewed. Methods of simu-
lation and estimation have been implemented in an R-package, ctarma. The simulations can be either
frequency-domain based or time-domain based. Several approaches of simulating CARMA processes with
time- and frequency-domain methods are implemented in the package. The estimation is based on numer-
ically maximizing the normal likelihood. The likelihood is computed via the Kalman-filter algorithm. The
process is constrained to be stationary by transforming the parameter space. Two alternative transformations
enforcing the stationary conditions on the parameter space are given. The package can model irregularly
spaced time series. Starting values of the parameters can be generated by using the Whittle-estimator when
a usable empirical spectrum is available. A scheme of increasing the size of the model, i.e., generating a
CARMA(p+1,q+1) model from a CARMA(p,q) is built into the package. The functionality of the package
is illustrated with simulations and some real-data series.
Financial Models, Thursday 09:50-10:50 106

robKalman—An R package for robust Kalman filtering

revisited
?
Bernhard Spangl1, , Peter Ruckdeschel2
1. University of Natural Resources and Life Sciences, Vienna
2. Fraunhofer ITWM, Department of Financial Mathematics, Kaiserslautern
?
Contact author: [email protected]

Keywords: EM algorithm, Kalman filter, recursive nonlinear filter, robustness

Building up on talks on this issue given at previous UseR! conferences, we report on progress made in the
development of the package robKalman. The focus of this talk will be on
• (robust) Kalman filtering and smoothing
– enhancing the functionality allowing for time-invariant and time-variant hyper parameters, even
functions
– but keeping the easy extensibility and modular approach of the general recursive filtering and
smoothing infrastructure
• (robust) estimation of hyper parameters
– via the EM Algorithm and its robustification
• (robust) recursive nonlinear filtering implementations, namely,
– the extended Kalman filter and
– the unscented Kalman filter

References

Durbin, J. and S. Koopman (2001). Time Series Analysis by State Space Methods. New York: Oxford
University Press.

R Development Core Team (2011). R: A Language and Environment for Statistical Computing. Vienna,
Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0. http://www.R-project.
org/.

Ruckdeschel, P. (2010, May). Optimally Robust Kalman Filtering. Technical Report 185, Fraunhofer ITWM
Kaiserslautern, Fraunhofer Platz 1, 67663 Kaiserslautern, Germany. http://www.itwm.fraunhofer.
de/fileadmin/ITWM-Media/Zentral/Pdf/Berichte_ITWM/2010/bericht_185.pdf.

Ruckdeschel, P. and B. Spangl (2010). robKalman: Robust Kalman Filtering. R package version 0.3,
http://robkalman.r-forge.r-project.org.

Shumway, R. and D. Stoffer (2000). Time Series Analysis and Its Applications. New York: Springer.

Spangl, B. (2008). On Robust Spectral Density Estimation. Ph. D. thesis, Dept. of Statistics and Probability
Theory, Vienna University of Technology, Vienna.

Wan, E. and R. van der Merwe (2001). The unscented kalman filter. In Haykin (Ed.), Kalman Filtering and
Neural Networks, Chapter 7, pp. 221–280. New York: Wiley.
Financial Models, Thursday 09:50-10:50 107

(Robust) Online Filtering in Regime Switching Models and

Application to Investment Strategies for Asset Allocation
?
Christina Erlwein1 , Peter Ruckdeschel1,
Fraunhofer ITWM, Dept. of Fin. Mathematics, Fraunhofer-Platz 1, 67663 Kaiserslautern, Germany
?
Contact author: [email protected]

Keywords: robustness, Hidden Markov Models, filtering, substitutive outliers, asset allocation

We implement to R, an online filtering algorithm for Hidden Markov Models with conditionally Gaussian
observations by Elliott (1994), and are currently about to package this functionality to a new R package
robHMM.
This algorithm consists of several steps: It involves a change of measure to an equivalent measure under
which we have independence as well as a filtering and and a (ML-) parameter estimation step where the last
two steps form an EM-algorithm.
The algorithm is modularized correspondingly such that in each step the respective function realizing it may
easily be replaced by a suitable alternative (robust) function.
We study the vulnerability of each of these steps against substitutive outliers and propose corresponding
robust alternatives extending Ruckdeschel (2010).
In a similar setting as in Erlwein et al. (2009), we apply this robustified algorithm to investment strategies
for asset allocation with the rationale to better handle possible peaks or missings in asset returns, limiting
their impact on optimal parameter estimates. The parameter estimates obtained are in turn used to make
investment decisions within a regime-switching framework.

References
Elliott, R. (1994). Exact adaptive filters for markov chains observed in gaussian noise. Automatica 30,
1399–1408.

Erlwein, C., R. Mamon, and M. Davison (2009). An examination of hmm-based investment strategies for
asset allocation. Applied stochastic models in business and industry. DOI: 10.1002/asmb.820.

R Development Core Team (2011). R: A language and environment for statistical computing. Vienna,
Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0. http://www.R-project.
org.
Ruckdeschel, P. (2010, May). Optimally Robust Kalman Filtering. Technical Report 185, Fraunhofer ITWM
Kaiserslautern, Fraunhofer Platz 1, 67663 Kaiserslautern, Germany. http://www.itwm.fraunhofer.
de/fileadmin/ITWM-Media/Zentral/Pdf/Berichte_ITWM/2010/bericht_185.pdf.
Ecology and Ecological Modelling, Thursday 09:50-10:50 108

Using R for the Analysis of Bird Demography on a Europe-wide Scale

Christian Kampichler1,2, Henk van der Jeugd1, Alison Johnston3, Rob Robinson3, Stephen Baillie3
1
Vogeltrekstation – Dutch Centre for Avian Migration and Demography, Netherlands Institute of Ecology, Wageningen, Netherlands;
2
División de Ciencias Biológicas, Universidad Juárez Autónoma de Tabasco, Villahermosa, Mexico;
3
British Trust for Ornithology, Thetford, U.K.
Contact: [email protected]

Bird populations are not constant in time but experience variation due to internal dynamics or to the action of
external forces such as climatic change or habitat destruction. They are often used as indicators for the trends of
wider biodiversity because our knowledge on them is generally good [1]. Beginning in the early 1980s, a Europe-
wide network of more than 400 constant effort sites operating in 12 countries has been established, where birds
are captured and ringed with a unique mark, according to a standardised protocol [2]. This scheme allows the
monitoring of demographic parameters (abundance, reproduction rate, survival rate) of a large number of songbird
species. The European database currently holds records of more than five million captures and recaptures.
Annual and site-specific reproduction rates are calculated by GLM based on the ratio of occurrences of adult
and juvenile birds per site and year. The estimation of survival rates is somewhat more sophisticated and requires
GLM with multinomial response variables for the application of specific capture-recapture models [3]. Basically,
input data consist of a capture history for each ringed specimen of a given species where years are coded as 1
(specimen captured) and 0 (specimen not captured). Each possible capture history—for example, 111, 110, 101 or
100 for a three-years mark-recapture study—has a corresponding probability which is composed of the probability
that a bird has survived from the preceding year (Φ) and the probability it is captured conditional on having
survived (p). For the four capture histories above, the probabilities are Φ1p2Φ2p3, Φ1p2(1 – Φ2p3), Φ1(1–p2)Φ2p3 and
1 – Φ1p2 – Φ1(1 – p2)Φ2p3. The capture history probabilities are weighed by their frequency in the dataset and the
parameters are estimated using maximum likelihood or Markov chain Monte Carlo methods.
The most widely used application currently available for survival analysis using data from marked
individuals is Program MARK [4] (URL http://www.phidot.org/software/mark/). On the one hand, it is extremely
comprehensive, offering a wide range of different models. On the other hand, its data and model manipulation
facilities are rather restricted—they are based on a conventional GUI with numerous pull-down menus and check-
box options—making the repeated analyses of many species (or for many locations) a cumbersome and extremely
error-prone task. We thus use RMark [5] (URL http://www.phidot.org/software/mark/rmark/) as an interface,
allowing for the use of the powerful formula and design matrix functions in R. A key advantage of R for
monitoring is that one does not need to recreate the analyses each year with extra data. Having a set of automated
scripts makes producing trends easy when resources are often tight. Furthermore, the possibility of using and
exchanging scripts facilitates the communication among the various institutions running national constant effort
site programmes and assures a unified analytical approach.

References
[1] Gregory RD, van Strien A, Vorisek P, Gmelig Meyling AW, Noble DG, Foppen RPB, Gibbons DW, 2005.
Developing indicators for European birds. Philosophical Transactions of the Royal Society B 360, 269–288
[2] Robinson RA, Julliard R, Saracco JF, 2009. Constant effort: studying avian population processes using
standardised ringing. Ringing & Migration 24, 199-204
[3] Amstrup S, MacDonald L, Manly B, 2006. Handbook of Capture-Recapture Analysis. Princeton University
Press
[4] White GC, Burnham KP 1999. Program MARK: Survival estimation from populations of marked animals.
Bird Study 46 Supplement, 120-138
[5] Laake J, 2010. RMark: R Code for MARK Analysis. R package version 1.9.6.
Ecology and Ecological Modelling, Thursday 09:50-10:50 109

Using OpenBUGS and lmer to study variation in plant

demographic rates over several spatial and temporal scales
Johan P. Dahlgren

Department of Botany, Stockholm University, Stockholm, Sweden.

Contact author: [email protected]

Keywords: Bayesian lasso, Ecology, Hierarchical OpenBUGS model, lmer, Population dynamics

I am using the lme4 and R2OpenBUGS R packages to study patterns in temporal and spatial variation
in vital rates of the forest herb Lathyrus vernus. The combined effect of vital rate variation on
population dynamics is examined in an Integral Projection Model (IPM), also run in R. Hierarchical
plant demographic studies only recently became practically possible and have rarely been performed
previously (Buckley et al. 2003), but available modeling techniques now enable such approaches
(McMahon & Diez 2007, Evans et al. 2010). Understanding where variation occurs can shed much light
on what factors affect plant demography. In contrast to previous studies I also use Openbugs and lmer
to examine at what spatial scales temporal variation occurs. In addition to this study of variance
components, I attempt to determine which out of a large set of environmental factors that were
measured at the plot level have the largest effect on population dynamics. This is done by including
Bayesian lasso components into the OpenBUGS vital rate models (Tibshirani 1996, Yi & Xu 2008).

Preliminary analyses suggest that most of the variation of all vital rates occur at the scale o f individuals,
as opposed to plot or sub-population scales. Temporal variation seems to be equally distributed over
spatial scales. Environmental factors have differing effects different years and are generally weak. This
is in accordance with the fact that little variation occurred at the plot scale, where environmental factors
were measured. I discuss the potential of density dependence to cause observed patterns.

On the technical side, I conclude that apart from the obvious “limitations” of a specialized function of
not easily allowing utilization of techniques such as the Bayesian lasso used here, at least two aspects of
lmer should make it an often more desirable choice than OpenBUGS for ecologists planning similar
analyses (see also Gelman & Hill 2007). First, as seen from comparisons with our BUGS models, it is
very accurate - even when estimating variance components for factors with few levels. Second, it is
much faster than OpenBUGS, not only in computation time of correctly specified models, but because
of the time it can take to specify a working BUGS model.

References
Buckley, Y.M., D.T. Briese & M. Rees (2003). Demography and management of the invasive plant
species Hypericum perforatum. I. Using multi-level mixed-effects models for characterizing
growth, survival and fecundity in a long-term data set. Journal of Applied Ecology 40, 481–93.
Evans, M.E.K., K.E. Holsinger & E.S. Menges (2010). Fire, vital rates, and population viability: a
hierarchical Bayesian analysis of the endangered Florida scrub mint. Ecological Monographs 80,
627–49.
Gelman, A. & J. Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models.
University Press, Cambridge.
McMahon, S.M. & J.M. Diez (2007). Scales of association: hierarchical linear models and the
measurement of ecological systems. Ecology Letters 10, 437–52.
Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. J. R. Stat. Soc. B, 58, 267–88.
Yi, N. & S. Xu (2008). Bayesian LASSO for Quantitative Trait Loci Mapping. Genetics 179, 1045-55.
Ecology and Ecological Modelling, Thursday 09:50-10:50 110

An effort to improve nonlinear modeling practice

John C. Nash1 with the NCEAS Nonlinear Modeling Working Group

1. Telfer School of Management (retired), University of Ottawa, Ottawa, ON K1N 6N5 Canada
Contact author: [email protected]

Keywords: nonlinear modeling, optimization, BUGS

Funded by the National Center for Ecological Analysis and Synthesis, the Nonlinear Modeling Working
Group is an assembly of approximately 20 researchers from a number of countries who are meeting for
two 1-week working seminars, one in January 2011 and the other scheduled for late July 2011. At the
first meeting, we chose a suite of test problems of an ecological nature requiring nonlinear models to
be estimated. Each problem was then approached using open source modeling tools, in particular, R,
AD Model Builder, and an open version of BUGS (either OpenBUGS or JAGS). This work will be
continued at the second meeting and individually outside the formal events.

The goal of the Working Group (http://www.nceas.ucsb.edu/projects/12602) is to document the effort

to build solutions in each package, to provide a set of ecological nonlinear modelling problems that are
well-described, to provide advice on how to carry out the process, and to assess where each package has
advantages or disadvantages with a view to improving the tools.

This paper will aim to share our preliminary findings and look to enlarge the activity and understanding
of benchmarking not only software but the human process of problem-solving.
Generalized Linear Models, Thursday 09:50-10:50 111

brglm: Bias reduction in generalized linear models

?
Ioannis Kosmidis1,
1. Department of Statistical Science, University College, Gower Street London, WC1E 6BT, London, United Kingdom
?
Contact author: [email protected]

Keywords: glm fitting methods, asymptotic bias reduction, adjusted score functions, Fisher scoring.

The brglm R package provides an alternative fitting method for the glm function for reducing the bias of
the maximum likelihood estimator in generalized linear models (GLMs). The fitting method is based on
the generic iteration developed in Kosmidis and Firth (2010a) for solving the bias-reducing adjusted score
equations (Firth, 1993). It relies on the implementation of the first-order term in the asymptotic expansion of
the bias of the maximum likelihood estimator for GLMs which has been derived in Cordeiro and McCullagh
(1991). The bias-corrected estimates derived in the latter study are by-products of the general fitting method.
The benefits of reducing the bias in estimation are discussed, especially in models for discrete responses.
Specifically, in such models there is a positive probability that the maximum likelihood estimate has infinite
components, which can potentially cause problems in the use of standard inferential procedures. In contrast,
for many well-used GLMs, the reduced-bias estimates have been found to always have finite components,
motivating their study and use in practice (see, for example, Firth 1992; Mehrabi and Matthews 1995; Heinze
and Schemper 2002; Bull et al. 2002; Zorn 2005; Kosmidis 2009; Kosmidis and Firth 2010b).
The brglm package also provides methods for the construction of confidence intervals through the profiles
of appropriately constructed inference functions (Kosmidis, 2008; Kosmidis and Firth, 2009).

References
Bull, S. B., C. Mak, and C. Greenwood (2002). A modified score function estimator for multinomial logistic regression
in small samples. Computational Statistics and Data Analysis 39, 57–74.
Cordeiro, G. M. and P. McCullagh (1991). Bias correction in generalized linear models. Journal of the Royal Statistical
Society, Series B: Methodological 53(3), 629–643.
Firth, D. (1992). Bias reduction, the Jeffreys prior and GLIM. In L. Fahrmeir, B. Francis, R. Gilchrist, and G. Tutz
(Eds.), Advances in GLIM and Statistical Modelling: Proceedings of the GLIM 92 Conference, Munich, New York,
pp. 91–100. Springer.
Firth, D. (1993). Bias reduction of maximum likelihood estimates. Biometrika 80(1), 27–38.
Heinze, G. and M. Schemper (2002). A solution to the problem of separation in logistic regression. Statistics in
Medicine 21, 2409–2419.
Kosmidis, I. (2008). The profilemodel R package: Profiling objectives for models with linear predictors. R News 8/2,
12–18.
Kosmidis, I. (2009). On iterative adjustment of responses for the reduction of bias in binary regression models.
Technical Report 09-36, CRiSM working paper series.
Kosmidis, I. and D. Firth (2009). Bias reduction in exponential family nonlinear models. Biometrika 96(4), 793–804.
Kosmidis, I. and D. Firth (2010a). A generic algorithm for reducing bias in parametric estimation. Electronic Journal
of Statistics 4, 1097–1112.
Kosmidis, I. and D. Firth (2010b). Multinomial logit bias reduction via poisson log-linear model. Technical Report
10-18, CRiSM working paper series. Accepted for publication in Biometrika.
Mehrabi, Y. and J. N. S. Matthews (1995). Likelihood-based methods for bias reduction in limiting dilution assays.
Biometrics 51, 1543–1549.
Zorn, C. (2005). A solution to separation in binary response models. Political Analysis 13, 157–170.
Generalized Linear Models, Thursday 09:50-10:50 112

Large Scale, Massively Parallel Logistic

Regression in R with the Netezza Analytics

Package

C.Dendek, M. Klopotek, P. Biecek

Netezza Corporation, an IBM Company

April 4, 2011



Abstract
One of the important limitations of the standard glm procedure for estimation
of the parametres of logistic regression model in R is the size of the data
that is kept in the memory, i.e. the original sample and algorithmspecic
temporary data, eectively restricting both cardinality and dimensionality of
the sample.
The biglm package makes possible to overcome the restriction on the
number of observations present in the sample. But it still leaves the dimen-
sionality limitation, due to the secondorder algorithm being used to t the
models.
The possiblity of use of the rstorder, stochastic gradientdescent op-
timization method creates a tradeo between the rate of convergence and
the maximal dimensionality of the sample. Interestingly, in case of smoothly
regularized logistic regression (e.g. L2based, rigde estimate), it is possible
to parallelize the rstorder method w.r.t. data sample, greatly improving
the computation time of a single iteration and  in practice  reducing the
adventage of secondorder method.
The gradient-based approach outlined above has been implemented in the
Netezza Analytics package using Netezza Performance Server as a database

1
Generalized Linear Models, Thursday 09:50-10:50 113

The binomTools package: Performing model

diagnostics on binomial regression models
Merete K Hansen1,*, Rune Haubo B Christensen1

1. Technical University of Denmark, Department of Informatics and Mathematical Modelling, Section for Mathematical
Statistics, Richard Petersens Plads, Building 305, Room 122, DK-2800 Kgs. Lyngby, Denmark
*Contact author: [email protected]

Keywords: Binomial regression models, diagnostics

Binomial regression models are widely used for modelling observations with dichotomous outcome.
Since diagnostics are important for validation of model adequacy, the accessibility of suitable
diagnostic methods for binomial regression models is crucial. The binomTools package (Hansen and
Christensen, 2011) provides a range of diagnostic methods for this class of models extending the basic
tool set in base R and enabling a thorough examination of the fitted model.

Appropriate use of the deviance or Pearson’s X2 reported by glm for goodness-of-fit assessment
requires that observations are suitably grouped, e.g., binary observations should be grouped to binomial
form. Our group method for glm objects performs this otherwise cumbersome task. Additional
goodness-of-fit tests for binary data available in binomTools include the Hosmer-Lemeshow test.

Residual analysis in binary and binomial models is complicated by the non-unique definition of
residuals and the sparseness of appropriate graphical methods. rstudent from stats provide
approximate deletion residuals, the so-called likelihood residuals (Williams, 1987); binomTools
enhance this by providing the exact counterparts. The half-normal plot with simulated envelopes
implemented in our halfnorm function is described by Collett (2003) and one of the most effective and
sensitive residual plots for the notoriously difficult binary models. Further, the coefficient of
discrimination (Tjur, 2008) and related figures are implemented for summarizing and describing
predictive performance.

Our aim with binomTools is to provide a comprehensive collection of methods for residual analysis,
model diagnostics and presentation of results from binomial regression models.

References:

Collett, D. (2003). Modelling binary data (Second edition). Chapman & Hall/CRC.

Hansen, M. K. and Christensen, R. H. B. (2011). binomTools – diagnostic tools for binomial regression
models. 1.0.1. https://r-forge.r-project.org/projects/binomtools/.

Tjur, T. (2008). Coefficients of determination in logistic regression models – a new proposal: The
coefficient of discrimination. The American statistician 63(4), 366-372.

Williams, D. A. (1987). Generalized linear model diagnostics using the deviance and single case
deletions. Applied Statistics 36, 181-191.
Reporting Data, Thursday 09:50-10:50 114

”uniPlot” – A package to uniform and customize R graphics

?
Sina Rüeger1,
1. Institute of Data Analysis and Process Design (Zurich University of Applied Sciences)
?
Contact author: [email protected]

Keywords: uniform layout, customizing graphics, visual perception

Three packages widely used for producing graphics in R are graphics, ggplot2 and lattice. Each has its
own strengths and limitations. However, they all differ in design. This is undesirable if graphics produced
by different packages are used in the same report. Therefore the package uniPlot, which uniforms graphics
produced by graphics, ggplot2 or lattice, is presented.
Daily R users who use the program to graph data within a particular data analysis are familiar with differ-
ences in appearance; they can extract information easily and mask the layout differences. Readers unfamiliar
with R will be distracted when interpreting three different graphics made by graphics, ggplot2 and lattice
graphics because variation in layout affects the decoding of information. Therefore, a way should be found
to produce consistent and calm graphics. In R this can be solved by adjusting several parameters, options or
themes. Everyone who has ever prepared plots by adjusting parameters, options and themes knows that this
is cumbersome and time consuming.
The package uniPlot solves this problem. It (a) designs an uniform look of graphics and (b) adapts graphics
from packages graphics, ggplot2 and lattice to achieve this uniform look. The idea is that the user would
load different layout-styles by a simple function. Changes of settings and parameters could be made within
this function. Restoring of the settings would be possible.
A first version of uniPlot is currently under development and will be available at CRAN by the end of July
2011.
Reporting Data, Thursday 09:50-10:50 115

sparkTable: Generating Graphical Tables for Websites and

Documents with R
?
Alexander Kowarik1, , Bernhard Meindl1 and Matthias Templ1,2
1. Statistics Austria, Guglgasse 13, 1110 Vienna, Austria
2. Vienna University of Technology, Wiedner Hauptstraße 8-10, 1040 Vienna, Austria
?
Contact author: [email protected]

Keywords: Visualisation, Graphical Tables, Sparklines

With the R package sparkTable (Kowarik et al. (2010)) tables presenting quantitative information can
be enhanced by including sparklines and sparkbars (initially proposed by Tufte (2001)). Sparklines and
sparkbars are simple, intense and illustrative graphs, small enough to fit in a single line. Therefore they can
easily enrich tables and continuous texts with additional information in a comprehensive visual way.
Another feature of sparkTable is the optimal allocation of geographical units to cells in a 2-dimensional ta-
ble and the presentation of graphical and quantitative information within these cells. For example, graphical
tables including information about countries with each country representing a cell in a 2-dimensional grid
can be generated. Hereby, the position of countries in a thematic map are projected to the given grid in an
optimal manner by solving a linear program.
The usage of S4-classes provide an easy and fast way to create fancy output for websites, presentation and
documents. The output is presented and explained with real-world applications.

References
Kowarik, A., B. Meindl, and S. Zechner (2010). sparktable: Sparklines and graphical tables for tex and
html. http://CRAN.R-project.org/package=sparkTable.

Tufte, E. R. (2001). Visual Display of Quantitative Information. Graphics Press.

Reporting Data, Thursday 09:50-10:50 116

compareGroups package, updated and improved

?
Héctor Sanz2,1 , Isaac Subirana3,1,4 , Joan Vila1,3
1. Cardiovascular Epidemiology & Genetics group, Inflammatory and Cardiovascular Disease Programme, IMIM, Hospital del
Mar Research Institute, Spain
2. UCICEC CAIBER. IMIM-Hospital del Mar
3. CIBER Epidemiology and Public Health (CIBERESP), Spain
4. Statistics Department, University of Barcelona, Spain
?
Contact author: [email protected]

Keywords: Software Design, Bivariate Table, LATEX, Descriptive Analysis

In many studies, such as epidemiological ones, it is needed to compare characteristics between groups of
individuals or disease status. Usually these comparisons are presented in the form of tables (also called
Bivariate Tables) of descriptive statistics where rows are characteristics, and each column is a group / status.
Usually the number of characteristics is large, and thus construction of these tables is laborious.
To build them in an easy, quick and efficient way, we created the compareGroups package [1]. Here we
present package improvements and extensions in the following issues:
• descriptives by status in a cohort study allowing to include right-censored time-to-response
• descriptives for groups or the entire sample.
• exporting tables to HTML format
• new GUI aspect with a single frame and a main menu
• importing data from workspace using the GUI
• subset specifically for each variable using the GUI
• incidence for right-censored time-to-event row-variables.
• extended and improved vignette
• exporting tables to LATEX under the longtable environment
• proper plots for survival analysis

References

1. Hector Sanz, Isaac Subirana, Joan Vila (2010) ”Bivariate Analyses” UseR! 2010, The R User Conference
2010, (National Institute of Standards and Technology, Gaithersburg, Maryland, US), July 2010.
Process Optimization, Thursday 09:50-10:50 117

Six Sigma Quality Using R: Tools and Training

?
Emilio López1, , Andrés Redchuk1 , Javier M.Moguerza1
1. Department of Statistics and Operations Research. Rey Juan Carlos University (Madrid)
?
Contact author: [email protected]

Keywords: Six Sigma, Process Improvement, Engineering Statistics, Quality Control, Lean Six Sigma

Six Sigma is a known methodology for Process and Quality Improvement. It is also a philosophy, and a
set of tools. It is based on the methodology DMAIC (Define, Measure, Analyze, Improve, Control). There
are other business-process management methodologies related to Six Sigma, such as DFSS (Design For Six
Sigma) or Lean Manufacturing (Lean Six Sigma).
Six Sigma is notable for using the Scientific Method, and Statistical Techniques. Some of the statistical
tools that are used in Six Sigma projects are:
• Graphic Analysis: Pareto Charts, Histograms, Scatterplots, Box-Whisker Charts, Group Charts, Lo-
cation Charts, Control Charts, Multivari Charts
• Design of Experiments
• Regression and Analysis of Variance (ANOVA)
• Confidence Intervals, Hypothesis Testing
• Gage R&R Studies
• Acceptance Sampling, Capability Analysis, Reliability Analysis
Commercial statistical software usually includes specific options for Quality Management, for example
control charts. There are also a couple of contribution packages in R regarding control charts (qcc, IQCC)
but there is no a complete set of tools for Six Sigma.
We are currently working in several initiatives aimed at explain and facilitate Six Sigma practitioners to
carry on their Six Sigma projects with R, such as a SixSigma package (already at CRAN repositories)
with some functions deployed and many others in mind, as well as an on-line training course within the EC
Lifelong Learning Programme VRTUOSI project, and several publications in preparation.

References
Allen, T. T. (2010). Introduction to Engineering Statistics and Lean Six Sigma - Statistical Quality Control
and Design of Experiments and Systems. Springer.

Box, G. (1991). Teaching engineers experimental design with a paper helicopter. Report 76, Center for
Quality and Productivity Improvement. University of Wisconsin.

Chambers, J. M. (2008). Software for data analysis. Programming with R. Statistics and Computing.
Springer.

ISO (2009). ISO/TS 16949: Quality management systems – Particular requirements for the application of
ISO 9001:2008 for automotive production and relevant service part organizations. International Organi-
zation for Standardization.

Montgomery, D. (2005). Introduction to Statistical Quality Control (5th ed.). New York: Wiley.

Murrell, P. (2005). R Graphics. Chapman & HallCRC.

Process Optimization, Thursday 09:50-10:50 118

Process Performance and Capability Statistics for Non-Normal

Distributions in R
?
Thomas Roth1
The Department of Quality Science - Technical University of Berlin
?
Contact author: Thomas Roth

Keywords: Distribution Identification, Process Capability, Process Performance

Estimating performance and capability statistics is a widespread practice in industry to assess and communi-
cate the state of (manufacturing) processes. The calculation of these statistics is among others defined in the
international standard ISO 21747:2006. Although the provided definitions are generic (i.e. distribution inde-
pendent), calculation of these statistics is mainly based on the assumption of a normally distributed process.
This practice causes difficulties with regularly occurring non-normally distributed process characteristics.
Methods are presented for conveniently fitting distributions and subsequently assessing process performance
and capability regardless of the underlying distribution of the measured characteristic. In addition an insight
into the terminology of the ISO 21747:2006 document is provided.

References
ISO (2005). Quality management systems - Fundamentals and vocabulary (ISO 9000:2005).

ISO (2007). Statistical methods - Process performance and capability statistics for measured quality char-
acteristics (ISO 21747:2006).

Mittag, H.-J. and H. Rinne (1999). Prozessfähigkeitsmessung für die industrielle Praxis. München: Hanser.

Montgomery, D. C. (2005). Introduction to statistical quality control (5 ed.). Hoboken, N.J: John Wiley.

Roth, T. (2010). qualityTools: Statistical Methods for Quality Science.

Process Optimization, Thursday 09:50-10:50 119

R-Package JOP: Optimization of Multiple Responses

Nikolaus Rudak1 , Sonja Kuhnt1
1. Faculty of Statistics, TU Dortmund University, Germany
?
Contact author: [email protected]

Keywords: Multiple responses, Robust parameter design, Simultaneous optimization, Pareto Optimality.

In technical applications, often a set of response variables with corresponding target values depends on a
number of control variables. In these cases an off-line quality control prior to the actual manufacturing
process frequently implies optimizing the mean as well as minimizing the variance of the responses.
Many existing methods for the optimization of multiple responses require some kind of weighting of these
responses, for instance in terms of costs or desirabilities. Kuhnt and Erdbruegge (2004) present an alternative
strategy using loss functions and a penalty matrix which can be decomposed into a standardizing and a
weight matrix. The Joint Optimization Plot displays the effect of different weight matrices in terms of
predicted response means and variances. Furthermore Erdbruegge et al. (2011) show that every point that
minimizes the conditional mean of the loss function is Pareto optimal.
The new R package JOP (Kuhnt and Rudak, 2011) is an implementation of the Joint Optimization Plot and
is available on CRAN in the version 2.0.1. JOP includes an automated procedure to fit double generalized
linear models by means of the R package Joint Modeling (see (Ribatet and Iooss, 2010)). For the optimiza-
tion actual three different optimization routines can be chosen, depending on the complexity of the problem,
more precisely by nlminb, gosolnp (see (Ghalanos and Theussl, 2011)) or genoud (see (Walter et al.,
2009)). The resulting optimal responses together with corresponding settings of the control variables are
displayed by the Joint Optimization Plot. JOP returns an object containing the optimal response and control
variable values as well as the fitted double generalized linear models. Furthermore the user can choose a
compromise with the mouse directly on the plot and JOP returns the corresponding optimal control variable
settings.
We demonstrate the use of the Joint Optimization Plot in various applications from mechanical engineering.

References
Erdbruegge, M., S. Kuhnt, and N. Rudak (2011). Joint optimization of several responses based on loss
functions. submitted.

Ghalanos, A. and S. Theussl (2011). Rsolnp: General Non-linear Optimization Using Augmented Lagrange
Multiplier Method. R package version 1.0-8.

Kuhnt, S. and M. Erdbruegge (2004). A strategy of robust paramater design for multiple responses. Statis-
tical Modelling 4, 249–264.

Kuhnt, S. and N. Rudak (2011). JOP: Joint optimization plot. R Package version 2.0.1.

Ribatet, M. and B. Iooss (2010). JointModeling: Joint Modelling of Mean and Dispersion. R package
version 1.0-2.

Walter, R. M., Jr., and J. S. Sekhon (2009). Genetic optimization using derivatives: The rgenoud package
for R. Journal of Statistical Software. Forthcoming.
Inference, Thursday 11:15-12:35 120

Density Estimation Packages in R

Henry Deng (Rice University)
Mentor: Hadley Wickham (Rice University)

Contact author: [email protected]

Density estimation is an important statistical tool, and within R, there are over 10 packages for density
estimation, including np, locfit, and KernSmooth. At different times, it is often difficult to know which to
use. In this project, we will summarize the results of our study comparing these packages. We will
present a brief outline of the theory behind each package, as well as a description of the functionality
and comparison of performance. Some of the factors we touch on are dimensionality, flexibility, and
control over bounds.
Inference, Thursday 11:15-12:35 121

The benchden Package: Benchmark Densities for

Nonparametric Density Estimation
?
Henrike Weinert1, , Thoralf Mildenberger2
1. Faculty of Statistics, TU Dortmund University, 44221 Dortmund, Germany
2. Department of Mathematics, University of Bayreuth, 95440 Bayreuth, Germany

?
Contact author: [email protected]

Keywords: nonparametic statistics, density estimation, benchmark densities

This talk gives an introduction to the R-package benchden which implements a set of 28 example densities
for nonparametric density estimation (Mildenberger and Weinert, 2011). In the last decades, nonparametric
curve estimation has become an important field of research. To assess the performance of nonparametric
methods aside from theoretical analysis often simulation studies are used. Indeed, most published articles
suggesting a new method contain a simulation study in which the proposed method is compared to at least
a few competitors. The comparison of methods on real-life data sets has the disadvantage that the correct
solution is unknown. By using artificial data sets generated under a completely known mechanism one
can overcome this disadvantage. Examples of such data sets are the Donoho-Johnstone functions Blocks,
Bumps, Doppler and HeaviSine originally introduced in Donoho and Johnstone (1994). These four functions
have well-known features like discontinuities or certain textures that resemble the difficulties encountered
in specific applications.
In the case of density estimation there seems to be no generally used set of test densities. Our package
benchden (Mildenberger et al., 2011) aims at closing this gap. The set of 28 test bed densities first in-
troduced by Berlinet and Devroye (1994) is sufficiently large to cover a wide variety of situations that are
of interest for the comparison of different methods. These densities also differ widely in their mathemati-
cal properties such as smoothness or tail behaviour and include some densities with infinite peaks that are
not square-integrable. They include both densities from standard families of distributions as well as some
examples specifically constructed to pose special challenges to estimation procedures.
The package contains the usual R-functions for densities that evaluate the density, distribution and quantile
functions or generate random variates. In addition, a function which gives some information on special
properties of the densities is included, which should be useful in large simulation studies.

References
Berlinet, A. and L. Devroye (1994). A comparison of kernel density estimates. Publications de l’Institute
de Statistique de L’Universite de Paris 38, 3–59.

Donoho, D. L. and I. Johnstone (1994). Ideal spatial adaption by wavelet shrinkage. Biometrika 81, 425–
455.

Mildenberger, T. and H. Weinert (2011). The benchden package: Benchmark densities for nonparametric
density estimation. Discussion Paper 14/2011, SFB 823, Technische Universitt Dortmund.

Mildenberger, T., H. Weinert, and S. Tiemeyer (2011). benchden: 28 benchmark densities from
Berlinet/Devroye (1994). R package version 1.0.4.
Inference, Thursday 11:15-12:35 122

Nonparametric Goodness-of-Fit Tests for Discrete Null

Distributions
?
Taylor B. Arnold1,
1. Yale University, Department of Statistics
?
Contact author: [email protected]

Keywords: Kolmogorov-Smirnov, Numerical Instability, Power Analysis

Goodness-of-fit tests are used to assess whether data are consistent with a hypothesized null distribution.
The χ2 test is the best-known parametric goodness-of-fit test, while the most popular nonparametric tests are
the classic test proposed by Kolmogorov and Smirnov followed closely by several variants on Cramér-von
Mises tests.
In their most basic forms, these nonparametric goodness-of-fit tests are intended for continuous hypothe-
sized distributions, but they have also been adapted for discrete distributions by Conover (1972), Choulakian
et al. (1994), and Gleser (1985). Unfortunately, most modern statistical software packages and programming
environments have failed to incorporate these discrete versions. As a result, researchers would typically rely
upon the χ2 test or a nonparametric test designed for a continuous null distribution. For smaller sample
sizes, in particular, both of these choices can produce misleading inferences.
We will present a revision of R’s ks.test() function and a new cvm.test() function to fill this void
for researchers and practitioners in the R environment. This work was motivated by the need for such
goodness-of-fit testing in a study of Olympic figure skating scoring (Emerson and Arnold, 2010). We will
first present overviews of the theory and general implementation of the discrete Kolmogorov-Smirnov and
Cramér-von Mises tests. We then will discuss the particular implementation of the tests in R and provide
examples paying close attention to the many numerical issues that arise in the implementation.

References
Arnold, T. B. and J. W. Emerson (2010). ks.test: A Kolmogorov Goodness-of-Fit Test for Discontinuous
Distributions. R package version 1.0.

Choulakian, V., R. A. Lockhart, and M. A. Stephens (1994). Cramér-von Mises statistics for discrete distri-
butions. The Canadian Journal of Statistics 22(1), 125–137.

Conover, W. J. (1972). A Kolmogorov goodness-of-fit test for discontinuous distributions. Journal of the
American Statistical Association 67(339), 591–596.

Cramér, H. (1946). Mathematical Methods of Statistics. Princeton, NJ: Princeton University Press.

Emerson, J. W. and T. B. Arnold (2010, August). The power of human nature in a study of olympic figure
skating. Unpublished manuscript.

Gleser, L. (1985). Exact power of goodness-of-fit tests of Kolmogorov type for discontinuous distributions.
Journal of the American Statistical Association 80(392), 954–958.

Imhof, J. P. (1961). Computing the distribution of quadratic forms in normal variables. Biometrika 48,
419–426.

Kolmogorov, A. (1933). Sulla determinazione empirica di una legge di distribuzione. Giornale dell’ Instuto
Italiano degli Attuari 4, 83–91.
Inference, Thursday 11:15-12:35 123

An algorithm for the computation of the power of Monte Carlo

tests with guaranteed precision
?
Patrick Rubin-Delanchy1 , Axel Gandy1,
1. Deparment of Mathematics, Imperial College London, SW7 2AZ, London, UK
?
Contact author: [email protected]

Keywords: Monte Carlo testing, power of a test, sequential testing.

The power of a hypothesis test is the probability of rejecting the null hypothesis if the data follows a given
distribution. For Monte Carlo tests, for example bootstrap or permutation tests, this quantity must usually
be evaluated by simulation. N datasets are generated from the specified data distribution, a Monte Carlo
test is performed on each, and the frequency with which the null hypothesis is rejected serves as an estimate
of the power of the test.
The existing research into computing the power of Monte Carlo tests focuses on obtaining an estimate that
is as accurate as possible for a fixed computational effort. However, the methods that are proposed have no
guarantee of precision, in the sense that they cannot report a (non-trivial) confidence interval for the power
with a certain coverage probability.
In this presentation, we describe an algorithm that runs N Monte Carlo tests simultaneously and indefi-
nitely until a user-specified confidence interval length and coverage probability is met. We demonstrate that
under some minor regularity conditions the algorithm terminates in finite expected time, and discuss some
optimisation issues.
We are currently augmenting the R simctest package to incorporate this work. The user must input a
confidence interval length and coverage probability, and a function that generates N binary streams, where
each value is one if the simulated statistic under the null is at least as extreme as the base test-statistic for this
stream. The method returns an estimate of the power with a confidence interval that meets the user-specified
requirements.
Population Genetics and Genetics Association Studies, Thursday 11:15-12:35 124

Simple haplotype analyses in R

Benjamin French1,*, Nandita Mitra1, Thomas P Cappola2, Thomas Lumley3

1. Department of Biostatistics and Epidemiology, University of Pennsylvania, Philadelphia, PA USA

2. Penn Cardiovascular Institute, Philadelphia, PA USA
3. Department of Statistics, University of Auckland, Auckland, New Zealand
*Contact author: [email protected]

Keywords: Statistical genetics, regression models

Statistical methods of varying complexity have been proposed to efficiently estimate haplotype effects
and haplotype-environment interactions in case-control and prospective studies. We have proposed an
alternate approach that is based on a non-iterative, two-step estimation process: first, an expectation-
maximization algorithm is used to compute posterior estimates of the probability of all potential
haplotypes consistent with the observed genotype for each subject; second, the estimated probabilities
are used as weights in a regression model for the disease outcome, possibly including environmental
factors. Standard error estimates are based on a robust variance estimator. We have shown that the two-
step process provides valid tests for genetic associations and reliable estimates of modest genetic effects
of common haplotypes for case-control studies (French et al, 2006). The two-step process has also been
applied to prospective studies with a survival outcome subject to censoring (Neuhausen et al, 2009). An
advantage of the two-step process is its straightforward implementation in software, so that analyses
combining genetic and environmental information can be conducted by researchers expert in that
subject matter using standard software, rather than by statisticians using specialized software. We
illustrate the use of the two-step process for case-control studies using our R package haplo.ccs, which
implements weighted logistic regression, and for prospective studies with a survival outcome using our
working R package haplo.cph, which implements weighted Cox regression. We illustrate our method
and software using data from a study of chronic heart failure patients (Cappola et al, 2011) to estimate
the effect of CLCNKA haplotypes on time to death or cardiac transplantation.

References

Cappola TP, Matkovich SJ, Wang W, et al. (2011). Loss-of-function DNA sequence variant in the
CLCNKA chloride channel implicates the cardio-renal axis in interindividual heart failure risk variation.
Proceedings of the National Academy of Sciences, doi:10.1073/pnas.1017494108

French B, Lumley T, Monks SA, et al. (2006). Simple estimate of haplotype relative risks for case-
control data. Genetic Epidemiology 30, 485–494.

Neuhausen SL, Brummel S, Ding YC, et al. (2009). Genetic variation in insulin-like growth factor
signaling genes and breast cancer risk among BRAC1 and BRAC2 carriers. Breast Cancer Research 11,
R76.
Population Genetics and Genetics Association Studies, Thursday 11:15-12:35 125

Mixed models of large pedigrees in genetic association studies

?
Jing Hua Zhao1
1. MRC Epidemiology Unit, Institute of Metabolic Science, Addenbrooke’s Hospital, Hills Road, Cambridge CB2 0QQ, UK
?
Contact author: [email protected]

Keywords: mixed models, pedigree data, genetic linkage and association analysis

Standard approaches in genetic linkage and association analysis of data from large pedigrees have limited
ability to handle covariates, which are important for studying their main effects and interactions with genetic
markers. We desribed mixed models appropriate for analyzing such data. In particular, our motivating exam-
ple was a simulated data distributed through the Genetic Analysis Workshop (GAW17) for which three types
of mixed models have been fitted: a linear model of quantitative trait, a logistic model of binary trait, and a
Cox model of binary trait and age at onset. The relevant functions lmekin for linear model and coxme for
Cox model are from the R package kinship and able to accommodate kinship and identity-by-descent (IBD)
information, while function pedigreemm from the R package pedigreemm allows for Gaussian, binary
and Poisson responses with kinship but not IBD information. We compared these with procedures in SAS.
We found that avaibility of a good IBD information can be useful for positional cloning and fine mapping
in genome-wide association studies involving single nucleotide polymorphisms. We believe that our work
will be of practical use to researchers in their analysis of pedigree data.

References
Epstein, M. P., J. E. Hunter, E. G. Allen, S. L. Sherman, X. Lin, and M. Boehnke (2009). A variance-
component framework for pedigree analysis of continuous and categorical outcomes. Statistics in Bio-
science 1, 181–198.

Therneau, T. (2007). On mixed-effect Cox models, sparse matrices, and modeling data from large pedigrees.
http://mayoresearch.mayo.edu/mayo/research/biostat/upload/kinship.pdf.

Vazquez, A. L., D. M. Bates, G. J. M. Rose, D. Gianola, and K. A. Weigel (2010). Technical note: An R
package for fitting generalized linear mixed models in animal breeding. Journal of Animal Science 88,
497–504.

Zhao, J. H. (2005). Mixed-effects Cox models of alcohol dependence in extended families. BMC Genetics 6
(Suppl I), S127.
Population Genetics and Genetics Association Studies, Thursday 11:15-12:35 126

Graphical tools for assessing Hardy-Weinberg equilibrium for

bi-allelic genetic markers
?
Jan Graffelman1,
1. Department of Statistics and Operations Research, Universitat Politècnica de Catalunya, Barcelona, Spain
?
Contact author: [email protected]

Keywords: Exact test, chi-square test, ternary plot, acceptance region, log-ratio transformation.

If there are no disturbing forces (migration, mutation, selection, etc.), then the genotype frequencies AA,
AB , and BB of a bi-allelic genetic marker are expected to occur with frequencies p2 , 2pq and q 2 respectively,
where p is the allele frequency of A, and q = 1 − p is the allele frequency of B. This basic principle
in genetics is known as Hardy-Weinberg equilibrium (HWE). In genetics, markers are statistically tested
prior to their subsequent use in for instance, association studies. A significant deviation from HWE can
be ascribed to many factors. Gross genotyping error can cause deviation from HWE and forms one of the
reasons to test markers. The HapMap project (2007) excludes genetic markers that have a p-value below
0.001 in an exact test for HWE.
Several statistical tests are available to check markers for HWE: the χ2 test, the exact test, the likelihood
ratio test and Bayesian tests. Weir (1996, Chapter 4) gives an overview of several tests for HWE. Several of
these tests are implemented in The R package HardyWeinberg. The package is currently being extended
with routines for power calculation and Bayesian tests. Most testing for HWE is done in an entirely numer-
ical manner. Graffelman and Morales (2008) showed that testing can be done graphically inside a ternary
plot representation. The HWE law defines a parabola inside a ternary plot of the three genotype frequencies.
An acceptance region for the different tests can be drawn around the HWE parabola. The graphical testing
facilities are implemented in the function HWTernaryPlot of the R package HardyWeinberg. The ob-
tained graphics are very informative because they display genotype frequencies, allele frequencies and the
(statistical) equilibrium condition in a single graph.
The genotype counts can also be treated as three-way compositions that sum up to 1. Tools from the field
of compositional data analysis yield alternative graphical representations of the HWE law. Several different
log-ratio transformations (additive, centred or isometric) of the genotype counts can be used. Graphically,
the HWE parabola in the ternary plot is converted into a straight line in log-ratio coordinates. Functions
HWalrPlot, HWclrPlot and HWilrPlot of the HardyWeinberg package can be used to create the
log-ratio plots for HWE. Zero genotype counts are a problem for the log-ratio approach, and some adjust-
ment for zero counts is necessary.
The proposed graphics can be used to assess HWE for multiple samples that are all typed for one marker
(e.g. cases and controls), but can also be used to screen many markers simultaneously. This way, a whole
genomic region can be screened for an excess or lack of heterozygotes. Several genetic data sets will be
used in the talk to illustrate the proposed graphs.

References
Graffelman, J. and J. Morales-Camarena (2008). Graphical tests for Hardy-Weinberg equilibrium based on
the ternary plot. Human Heredity 65(2), 77–84. DOI: 10.1159/000108939.

The International HapMap Consortium (2007). A second generation human haplotype map of over 3.1
million SNPs. Nature 449, 851–861.

Weir, B. S. (1996). Genetic Data Analysis II. Sunderland, Massachusetts: Sinauer Associates.
Neuroscience, Thursday 11:15-12:35 127

Detecting Drug Effects in the Brain

Heather Turner1*, Foteini Strimenopoulou1, Phil Brain1

1. Pfizer Global Research & Development, UK

*Contact author: [email protected]

Keywords: Multivariate analysis, Statistical modelling, Pharmaceutical research

In the early phases of drug development, the effect of a candidate drug on the brain can be of key
importance, either because the drug is specifically targeted at the brain or because of safety concerns
regarding an effect on the brain. Quantitative electroencephalography (qEEG) uses multiple electrodes
placed on the scalp of a subject (typically rat or human) to record the electrical activity produced by the
firing of neurons within the brain. By monitoring these “brainwaves” under different treatment
conditions, the effect of a candidate drug can be inferred.

qEEG produces a virtually continuous signal over time that is typically processed using a Fast Fourier
Transform, producing a set of power spectra at successive time slices for each subject. The challenge is
to relate these power spectra (a multivariate response representing the brain activity), to the
concentration of the drug in the brain at the corresponding time. Of course, the actual concentration of
the drug in the brain at a given time cannot be measured, but can be modelled using pharmacokinetic
models typically with unknown parameters.

In this talk we will introduce an R package for Extended Semi-linear Canonical Correlation Analysis
(ESLCCA, described in Brain et al, 2011). ESLCCA estimates the parameters of a given nonlinear
pharmacometric model to maximize the correlation with a linear combination of multiple response
variables (in this case, the power spectra). We shall illustrate how this method has been used to
characterize brain activity under different treatment regimens in research and development projects at
Pfizer.

References

Brain, P, Strimenopoulou, F & Ivarsson, M (2011). Analysing electroencephalogram (EEG) data using
Extended Semi-Linear Canonical Correlation Analysis. Submitted.
Neuroscience, Thursday 11:15-12:35 128

Statistical Parametric Maps for Functional MRI Experiments

in R: The Package fmri
?
Karsten Tabelow1,2, , Jörg Polzehl1
1. Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS) Leibniz-Institut im Forschungsverbund Berlin e.V.
2. DFG Research Center M ATHEON “Mathematics for key technologies”
?
Contact author: [email protected]

Keywords: functional magnetic resonance imaging, structural adaptive smoothing, structural adaptive
segmentation, random field theory, multiscale testing

The purpose of the package fmri is the analysis of single subject functional magnetic resonance imaging
(fMRI) data. It provides fMRI analysis from time series modeling by a linear model to signal detection and
publication quality images. Specifically, it implements structural adaptive smoothing methods with signal
detection for adaptive noise reduction which avoids blurring of activation areas.
We describe the complete pipeline for fMRI analysis using fmri, i.e., reading from various medical imaging
formats, the linear modeling used to create the statistical parametric maps, signal detection and visualization
of results. We review the rationale behind the structural adaptive smoothing algorithms and explain their
usage from the package fmri. We demonstrate how such an analysis is performed using experimental data.

References

K. Tabelow, J. Polzehl, H.U. Voss, V. Spokoiny (2006). Analyzing fMRI experiments with structural adap-
tive smoothing procedures. NeuroImage 33, 55–62.
K. Tabelow, V. Piëch, J. Polzehl, H.U. Voss, High-resolution fMRI: Overcoming the signal-to-noise prob-
lem, J. Neurosci. Meth., 178 (2009) pp. 357–365.
J. Polzehl, H.U. Voss, K. Tabelow (2010). Structural adaptive segmentation for statistical parametric map-
ping. NeuroImage 52, 515–523.
K. Tabelow, J.D. Clayden, P. Lafaye de Micheaux, J. Polzehl, V.J. Schmid and B. Whitcher (2011). Image
analysis and statistical inference in neuroimaging with R, NeuroImage 55, pp. 1686–1693.
K. Tabelow (2010). Statistical parametric maps for functional MRI experiments in R: The package fmri.
Preprint no. 1562, WIAS, Berlin, http://www.wias-berlin.de/preprint/1562/wias_
preprints_1562.pdf
Neuroscience, Thursday 11:15-12:35 129

neuRosim: an R package for simulation of fMRI magnitude

data with realistic noise
?
Marijke Welvaert1, , Yves Rosseel1
1. Department of Data Analysis, Ghent University, Belgium
?
Contact author: [email protected]

Keywords: fMRI, simulation, physiological noise

Statistical analysis techniques for highly complex structured data such as fMRI data should be thoroughly
validated. In this process, knowing the ground truth is essential. Unfortunately, establishing the ground truth
of fMRI data is only possible with highly invasive procedures (i.e. intracranial EEG). Therefore, generating
the data artificially is often the only viable solution. However, there is currently no consensus among
researchers on how to simulate fMRI data. Research groups develop their own methods and use only in-
house software routines. A general validition of these methods is lacking, probably due to the nonexistance
of well-documented and freely available software.
In a response to this gap, neuRosim is developed to offer a software package for the simulation of fMRI
data. The ultimate goal of the package is to create a general standardized platform that contains fMRI
simulation methods that are validated. To this end, the package will contain several functions to simulate
BOLD activation and fMRI noise.
In the current version, the functionalities of the package are separated in two layers. First, low-level func-
tions are intended for advanced useRs who want in-depth control over their simulated data. With these
functions it is possible to build fMRI data consisting of activation and noise while keeping full knowledge
of the structure of the data and having the possibility to manipulate several parameters. Second, high-level
functions are created to be used in more standard simulation studies. The power of these functions lies in
the fact that they allow the useR to generate a full 4D fMRI dataset using only 3 command lines.
During the presentation, we will discuss the current features of neuRosim and our plans for coming updates.
Using a variety of examples, we will demonstrate the useR-friendliness for first time useRs and how these
examples can be extended to more advanced simulations. Finally, we will show the difference between the
implemented simulation methods and discuss briefly how the simulated data can be used in other related R
packages.
Data Management, Thursday 11:15-12:35 130

It’s a Boy!
An Analysis of Tens of Millions of Birth Records Using R
Susan I. Ranney, Ph.D.1*

1. VP Product Development, Revolution Analytics, Inc.

*Contact author: [email protected]

Keywords: Visualization, Data, Birth

The fact that more boys than girls are born each year is well established – across time and
across cultures. But there are variations in the degree to which this is true. For example, there is
evidence that the sex ratio at birth declines as the age of the mother increases, and babies of a higher
birth order are more likely to be girls. Different sex ratios at birth are seen for different racial groups,
and a downward trend in the sex ratio in the United States since the early 1970s has been observed.
Although these effects are very small at the individual level, the impact can be large on the number of
“excess” males born each year. To analyze the role of these factors in the sex ratio at birth, it is
appropriate to use data on many individual births over multiple years.
Such data are in fact readily available. Public-use data sets containing information on all births
in the United States are available on an annual basis from 1985 to 2008. But, as Joseph Adler points out
in R in a Nutshell, “The natality files are gigantic; they’re approximately 3.1 GM uncompressed. That’s
a little larger than R can easily process.” An additional challenge to using these data files is that the
format and contents of the data sets often change from year to year.
Using the RevoScaleR package, these hurdles are overcome and the power and flexibility of the
R language can be applied to the analysis birth records. Relevant variables from each year are read from
the fixed format data files into RevoScaleR’s .xdf file format using R functions. Variables are then
recoded in R where necessary in order to create a set of variables common across years. The data are
combined into a single, multi-year .xdf file containing tens of millions of birth records with information
such as the sex of the baby, the birth order, the age and race of the mother, and the year of birth.
Detailed tabular data can be quickly extracted from the .xdf file and easily visualized using
lattice graphics, as shown in the plot below. Trends in births, and more specifically the sex ratio at
birth, are examined across time and demographic characteristics. Finally, logistic regressions are
computed on the full .xdf file examining the conditioned effects of factors such age of mother, birth
order, and race on the probability of having a boy baby.

References

Adler, Joseph (2010). R in a Nutshell.

CDC (1985-2008). Birth Data Files, http://www.cdc.gov/nchs/data_access/Vitalstatsonline.htm.
Matthews TJ, Hamilton BE (2005). Trend analysis of the sex ratio at birth in the United States,
http://www.cdc.gov/nchs/data/nvsr/nvsr53/nvsr53_20.pdf.
Data Management, Thursday 11:15-12:35 131

Challenges of working with a large database of

routinely collected health data: Combining SQL and R
Joanne Demmler1, Caroline Brooks1, Sarah Rodgers1, Frank Dunstan2, Ronan Lyons1

1. Swansea University, College of Medicine, Grove Building, Singleton Park, Swansea SA2 8PP
2. Cardiff University, Department of Primary Care & Public Health, Neuadd Meirionnydd, Heath Park, Cardiff CF14 4YS
*Contact author: [email protected]

Keywords: SAIL, anonymised data, RODBC, child health, data linkage

Introduction: Vast amounts of data are collected about patients and service users in the course of
health and social care service delivery. Electronic data systems for patient records have the potential
to revolutionise service delivery and research. But in order to achieve this, it is essential that the
ability to link the data at the individual record level be retained whilst adhering to the principles of
information governance. One such example is the Secure Anonymised Information Linkage (SAIL)
databank, which contains health, social and education data for three million residents for a contiguous
area in Wales, UK. There are currently 21 major datasets containing about 1.6 billion records, which
can be linked anonymously at the individual level and household level.

Background: All work on SAIL data is executed through a secure remote desktop environment via a
virtual private network (VPN), which has no internet connection and all output requires approval
before it can be released for external viewing or publication. The processor speed equals 1 core of a
Xeon X5550 @ 2.67 GHz processor, with an allocated memory of 2GB RAM per user.

Methods: We present here an example from the National Community Child Health Database, which
contains height and weight measurements from school entry examinations for 849,238 children. Data
are preselected using SQL and saved as a temporary table in SAIL to remove children with negative
age at examination and to restrict examinations to the years 1990 to 2008. After removal of
biologically unfeasible records with SQL 1,764,728 records for 594,720 children are imported into R
using the sqlfetch command of the RODBC package. A simple algorithm is explored to remove
remaining outliers in the data.

Results: Although R is very effective in some basic analysis and in the exploration of the data, it is
not very efficient in dealing with such a vast dataset in certain situations. Memory limitations within
the secure gateway platform mean that quite simple R scripts might run for a considerable time (days)
and data might get lost in transfer operations (saving back to SAIL might fail depending on the table
dimensions). At the present, this prevents the usage of more advanced statistical methods as well as
modelling of the data.

Outlook: As a result of this analysis we have decided to migrate R onto a more powerful server. We
are also investigating the possibility of multithreading R code, using the College of Medicine’s BlueC
replacement supercomputer (2 nodes, each with 30 Power7 cores @ 3.3GHz and 100GB of Ram).
Data Management, Thursday 11:15-12:35 132

Demographic: Classes and Methods for Data about Populations

?
John Bryant1
1. Statistics New Zealand
?
Contact author: [email protected]

Keywords: S4 Classes, Demography, Official Statistics, Data Manipulation

Population data cross-classified by variables such as age, sex, and time is ubiquitous in official statistics and
elsewhere. The presentation introduces Demographic, an R package under development that aims to ease
some of the burden of manipulating this sort of data.
It is natural to represent cross-classified data as a multidimensional array, and to describe operations on
cross-classified data as manipuations of arrays, such as splitting, collapsing, or expanding. Package De-
mographic facilitates this approach by providing an S4 class, Demographic, that builds on standard R
arrays. Among other things, an object of class Demographic
• holds metadata such as the precise definitions of age intervals, or key words describing of the nature
of each dimension (eg ‘time’, ‘origin’, or ‘category’),
• has methods for most standard functions, many of which take advantage of the metadata by, for
instance, checking that age intervals align correctly,
• encourages the use of names rather than numeric indices, in the interest of more transparent, less
error-prone code,
• permits data.frame-like expressions such as subset(x, age >= 65).
There are two main subclasses of Demographic: Counts and Rates. Objects from these two classes
behave differently, reflecting differences in the way that analysts conventionally treat counts and rates data.
For instance, when one Counts object is divided by another Counts object, a Rates object is produced,
and if one of the Counts objects has a dimension that the other does not, the extra dimension is collapsed
before the division is carried out.
The presentation will show how Demographic can be used for common tasks such as tidying messy data or
doing simple projections. It will also provide an example of how a specialised package can take advanage
of the general-purpose services provided by Demographic.
Data Management, Thursday 11:15-12:35 133

Correcting data violating linear restrictions using the

deducorrect and editrules packages
Mark van der Loo1,? , Edwin de Jonge1 and Sander Scholtus1
1. Statistics Netherlands, Henri Faasdreef 312, 2492 JP The Hague, The Netherlands
?
Contact author: [email protected]

Keywords: Data editing, official statistics, linear restrictions, errror correction

editrules In many computational and statistical problems one needs to represent a set of m linear
restrictions on a data record x of the form Ax − b = 0 or Ax − b ≥ 0,where A is a matrix with real
coefficients and b a vector in Rm . Constructing and maintaining A manually is tedious and prone
to errors. Moreover, in many cases the restrictions are stated verbosely, for example as “profit +
cost must equal turnover”. The editrules package can parse restrictions written in R language to
matrix form. For example:
> editmatrix(c("x + 3*y == -z", "x>0"))
Edit matrix:
x y z CONSTANT
e1 1 3 1 0
e2 1 0 0 0

Edit rules:
e1 : x + 3*y == -z
e2 : x > 0
The result is an S3 object of class editmatrix which extends the standard matrix object. Here, the
editmatrix function accepts linear restrictions in character or data.frame format. The latter
offers the opportunity to name and comment the restrictions. The package also offers functionality
to check data against the imposed restrictions and summarize errors in a useful way. In fact, the
error checking functionality is independent of restrictions being of linear form, and can be used for
any restriction including numerical and/or categorical data.

deducorrect Raw survey data is often plagued with errors which need to be solved before one can
proceed with statistical analysis. The deducorrect package offers functionality to detect and correct
typing errors (based on the Damerau-Levenshtein distance) and rounding errors in numerical data
under linear restrictions. It also solves sign errors and value swaps, possibly masked by rounding
errors. The methods used are (slight) generalizations of the methods described by Scholtus (2008)
and Scholtus (2009). All datacorrection functions return corrected data where possible, a log of the
applied corrections and the correction status. The package also offers functionality to determine if
a matrix is totally unimodular, which is useful for solving errors in data involving balance accounts.

References
Scholtus, S. (2008). Algorithms for correcting obvious inconsistencies and rounding errors in busi-
ness data. Technical Report 08015, Statistics Netherlands, Den Haag. Accepted by J.Official
Stat.

Scholtus, S. (2009). Automatic correction of simple typing error in numerical data with balance
edits. Technical Report 09046, Statistics Netherlands, Den Haag. Accepted by J. Offical Stat.
Interactive Graphics in R, Thursday 11:15-12:35 134

iWebPlots: Introducing a new R package for the creation

of interactive web-based scatter plots
Eleni-Anthippi Chatzimichali1,* , Conrad Bessant1

1. Cranfield Health, Cranfield University, Bedfordshire, MK43 0AL. UK.

*Contact author: [email protected]

Keywords: Visualization, Image Maps, HTML, Web Graphics

Scatter plots constitute a most widely-used tool employed to investigate the correlations
underlying a dataset. Nowadays, the demand for online data visualization in addition to the increasing
need for availability of dynamic features and interactions, necessitate the construction of interactive
web graphics.
The iWebPlots package simplifies the implementation of interactive web-based scatter plots,
generated directly via the R statistical environment. The package’s functions take as input a matrix of
coordinates with associated metadata and generate a bitmap image with an HTML wrapper.
Interactivity is implemented using the fundamental HTML image map technology, so no additional
software such as applets or plug-ins is required. Developers can easily modify and expand the generated
HTML pages, or incorporate them into web applications; thus, great extensibility is ensured.
Additional features include dynamic tooltips and text annotations as well as asynchronous
alternation between two- and three-dimensional scatter plots. Furthermore, each plot can be interlinked
with a fully interactive data table, which displays more information about the data in the plot.
The iWebPlots package has been used to develop the web front-end of a multivariate analysis
pipeline, featuring techniques such as Principal Component Analysis (PCA) and k-means clustering,
among many others. This work was carried out as part of the SYMBIOSIS-EU project, funded by
European Commission Framework 7.

References

SYMBIOSIS-EU (2011), http://www.symbiosis-eu.net/

Interactive Graphics in R, Thursday 11:15-12:35 135

Rocessing: Interactive Visualizations in R

Ian Hansel1

1. Ernst & Young, Fraud Investigation and Dispute Services

*Contact author: [email protected]

Keywords: Visualization, Processing, Rjava, Interactive Graphs

Rocessing is a new package under development which combines R with Processing to produce highly
interactive graphs and plots for Exploratory Data Analysis and presentation of data. Processing is an
open source Java based language and environment which allows users to create images, animations and
interact with data. There are two main objectives of this project:

1. Provide an in-depth visualization tool for analysing data in R.

2. Establish a platform that allows interactive applications written in Processing to be called from
R.

Visualizing is often a key component of any analytical activity whether this is in the data cleansing
stage, statistical modelling stage or the presentation of results. Being able to perform these tasks in an
interactive manner can make these tasks simpler and quicker. In my presentation I will be going
through a case study to show how Rocessing can be used at each step from data preperation and
profiling right up to displaying the results of the analysis.

Rocessing utilises the rJava package to link R to Processing. The initial release scheduled for 20 April
2011 will provide interactive graphs such as scatterplots, histograms and mapping. In addition it will
allow for user created Processing scripts to be called from R.

References

Fry, Ben (2008), Visualizing Data, O'Reilly Media, ISBN 0596514557.

Fry, Ben; Reas, Casey (2011). Processing, http://processing.org.

R Development Core Team (2010). R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL
http://www.Rproject.org .

Tufte, Edward R. (2001), The Visual Display of Quantitative Information, Graphics Press LLC.

Urbanek, Simon(2010). rJava: Low-level R to Java interface. R package version 0.8-6.

http://CRAN.Rproject.org/package=rJava
Interactive Graphics in R, Thursday 11:15-12:35 136

Easy Interactive ggplots

Richard Cotton,1*

1. Health and Safety Laboratory, Buxton, UK

*Contact author: [email protected]

Keywords: ggplot2, gWidgets, plotting, data-viz, interactive

The ggplot2 package allows the creation of high quality static graphics, but it does not currently support
interaction. In general, making ggplot2 graphics interactive is a difficult problem to solve, and is
beyond the scope of this talk. Instead, a simpler problem is solved here: the ability to quickly and
easily make small numbers of interactive features for specific usage, via control panels built using the
gWidgets* packages. This talk demonstrates the creation of several of pieces of functionality that can
be easily created and mixed together to provide a custom interactivity for your ggplots. Features
demonstrated will include changing scale transformations, faceting, the title and axis labels, the dataset
and geom layers.
Interactive Graphics in R, Thursday 11:15-12:35 137

RnavGraph and the tk canvas widget

?
Adrian Waddell1 , Wayne Oldford1
1. University of Waterloo
?
Contact author: [email protected]

Keywords: Visualization, Clustering, Multivariate statistics, Graphs, Images

We will demonstrate RnavGraph, an R package to explore and/or cluster interactively high dimensional
data such as images, microarray or text data. The navigational infrastructure is provided by graphs (see
Hurley and Oldford, 2011, Comp. Stat.). That is, the user moves on a graph a ”You are here” circle,
or ”bullet”, from one node to another along defined edges, causing some data visualization to be smoothly
morphed from one plot (first node) into another plot (second node). Nodes of such graphs could for example
represent 2d scatterplots and the edges rigid 3d rotations- or 4d transitions- from one scatterplot into another.
We implemented our own 2d scatterplot display, called tk2d for this working example. tk2d can display
either dots, images, star glyphs or text, and the data can be linked between displays.

interactive graph linked 2d image scatterplot linked 2d glyph scatterplot

A:B

A:D C:D

B:C

The RnavGraph package heavily uses tcl and tk through the tcltk R package. Both the graph display and
the tk2d display build upon the tk canvas widget. The tk2d display also makes use of the C API to achieve
smooth morphing and image resizing results. Users can also easily extend the package and its predefined
visualization instructions; this will be demonstrated during the talk.

For the second part of our presentation, we will present some of our insights gained while working with
tcl, tk and R, especially with regard to the performance of the tk canvas widget. This will include some
experimentation results, demonstration of some self-containing simple examples and some remarks about
the C API to the tk canvas widget.
Kaleidoscope IIIa, Thursday 14:00-15:00 138

Using R for systems understanding – a dynamic approach

Thomas Petzoldt1, Karline Soetaert2

1. Technische Universiotät Dresden, Institut für Hydrobiologie, 01062 Dresden

2. Centre for Estuarine and Marine Ecologoy (CEME), Netherlands Institute of Ecology (NIOO)
*Contact author: [email protected]

Keywords: Simulation Models, Differential Equations, Stoichiometric Matrix, Ecology

The R system is not only a statistics and graphics system. It is a general-purpose high-level
programming language that can be used for scientific computing in general. It is increasingly
superseding conventional spreadsheet computing and became one of the standard environments for data
analysis and modeling in ecology. An increasing collection of packages explicitly developed for
dynamic modeling (deSolve, simecol, FME, ReacTran, cf. Soetaert et al 2010) and a growing number
of textbooks teaching systems understanding by using R examples (Bolker, 2008; Soetaert and Herman,
2009; Stevens, 2009) are just an indicator for this trend.

The contribution will focus on practical experience with implementing and using dynamic models in R
from an ecological modeler’s perspective who works together with field and lab ecologists. Here,
dynamic models play an essential role for improving qualitative and quantitative systems
understanding. Aspects of two different case studies with increasing level of complexity are presented
to demonstrate different application scenarios and modeling techniques:

 A model of semi-continuous laboratory cultures using package deSolve and the event
mechanism,
 A water quality model for a polluted river using package ReacTran for transport and a compact
representation using stoichiometric matrices (Peterson matrices) for matter turnover (cf.
Reichert and Schuwirth, 2010).

The examples are organized in package simecolModels, a collection of simulation models that cover
the range from teaching demos up to the ecosystem level. It is discussed how model implementations
can be organized in a readable and computation-efficient way and how the model outcome can be
visualized.

References

Bolker, B. (2008). Ecological Models and Data in R. Princeton University Press.

Reichert, P. and Schuwirth, N. (2010) A generic framework for deriving process stoichiometry in
environmental models. Environmental modelling and Software 25, 1241-1251.

Soetaert, K. and Herman, P. M. J. (2009). A Practical Guide to Ecological Modelling Using R as a

Simulation Platform. Springer.

Soetaert, K., Petzoldt, T. and Setzer, R. W. (2010) Solving differential equations in R. The R Journal
2/2, 5-15.

Stevens, M. H. H. (2009). A Primer of Theoretical Population Ecology with R. Springer.

Kaleidoscope IIIa, Thursday 14:00-15:00 139

Using multidimensional scaling with Duchon splines for reliable

finite area smoothing
?
David L. Miller1 , Simon N. Wood1
1. Mathematical Sciences, University of Bath, UK
?
Contact author: [email protected]

Keywords: spatial smoothing, generalised additive models, within-area distances, splines

Splines are often used to perform smoothing over geographical regions. However when boundary features
intrude into the study region, splines may smooth across the features in a way that makes little sense and may
be misleading. Other smoothers also suffer form this problem of inappropriately linking parts of the domain
(for example either side of a river or peninsula). One view is that the problem is caused by the smoother
using of an unsuitable metric to measure the distance between points. A simple substitution of within-region
distances in place of Euclidean distances has been proposed previously to combat this problem, with some
success, but at the price of some loss of clarity about what the resulting smoothing objective means, and
quite high computational cost.
Our new approach takes within-area distances and uses them to project the location data into a new (often
high-dimensional) space using multidimensional scaling. Conventional smoothing can then take place in
this new space in which the euclidean distances between points now approximates the original within-region
distances. We show that Duchon’s 1977 generalisation of thin plate splines has particular advantages for
smoothing in the MDS space.
As well as finite area smoothing, the new method provides a means for smoothing with respect to general
distance measures, and examples of both spatial smoothing and more general distance-based smoothing are
given.
Kaleidoscope IIIa, Thursday 14:00-15:00 140

Studying galaxies in the nearby Universe, using R and ggplot2

?
Alastair Sanderson
School of Physics & Astronomy, University of Birmingham, UK
?
Contact author: [email protected] (http://www.sr.bham.ac.uk/˜ajrs)

Keywords: Astronomy, galaxies, multivariate data visualization, ggplot2

Galaxies are cosmic building blocks whose properties tell us about the origins and fate of the Universe.
These objects are drawn together under the influence of gravity to form a complex pattern of filaments,
and at the intersection of these filaments lie groups and clusters of galaxies– the largest gravitationally
bound structures in the Universe. Here, large concentrations of dark matter tightly hold the fast-moving
galaxies together, enabling close interactions and mergers to take place, which transform their properties.
The location, luminosity and morphology of ∼105 galaxies in the local Universe represents a 5 dimensional
multivariate dataset which is ideally suited to exploration with R, and ggplot2 in particular. I will show how
R is being used to develop new insights into our understanding of galaxy groups and the process of cosmic
feedback in the Universe, based on results from the CLoGS (Complete Local-Volume [galaxy] Groups
Sample1 ) project, using the HyperLeda2 galaxy database.

1
http://www.sr.bham.ac.uk/˜ejos/CLoGS.html
2
http://leda.univ-lyon1.fr
Panel Discussion II, Thursday 14:00-15:00 141

Panel discussion: Challenges Bringing R into Commercial

Environments
Louis Bajuk-Yorgan1,*

1. TIBCO Software Inc.

*Contact author: [email protected]

Keywords: Commercial, Applications, Panel

Champions of wider R usage within commercial environments often face multiple challenges, such as a
lack of IT knowledge or acceptance, concerns about technical support, questions around validation or
other regulatory compliance, difficulties in maintaining multiple versions of R, etc.

In this panel discussion, R champions and commercial vendors will share their views and experiences.
Questions from the audience are highly encouraged.

NOTE:

I organized a panel discussion on this topic for useR 2010. The discussion was well-attended, with lots
of questions from the audience, and several attendees approached me afterwards to suggest this be a
regular feature of useR conferences. The members of the panel from last year are listed below. If my
proposal is accepted, I would invite participation from Revolution and Mango again, and seek out new
R champions to broaden the discussion.

Panel composition last year:

- Revolution Analytics (Norman Nie, CEO)
- TIBCO (Lou Bajuk-Yorgan, Sr. Director Product Management)
- Bret Musser, Director, Clinical Biostatistics at Merck
- Mango Solutions (Richard Pugh, CTO)
- Jim Porzak, Senior Director of Marketing Analytics at Ancestry.com
- Thomas G Filloon, Principal Statistician, Procter & Gamble

Suggested logistics:
- Each speaker sends one slide to moderator (TBD), moderator controls the slides
- Each speaker has 3 minutes to talk
- Rest of the time is devoted to Q&A and discussion
Kaleidoscope IIIb, Thursday 14:00-15:00 142

microbenchmark: A package to accurately benchmark R

expressions
?
Olaf Mersmann1 , Sebastian Krey1
1. TU Dortmund, Department of Statistics
?
Contact author: [email protected]

Keywords: Benchmarking, Timing, High Performance Computing

We present the R package microbenchmark. It provides functions to accurately measure the execution time
of R expressions. This enables the user to benchmark and compare different implementation strategies for
performance critical functions, whose execution time may be small, but which will be executed many times
in his program. In contrast to the often used system.time and replicate combination, our package
offers the following advantages: firstly it attempts to use the most accurate method of measuring time made
available by the underlying operating system, secondly it estimates to overhead of the timing routines and
subtracts it from all measurements automatically and lastly it provides utility functions to quickly compare
the timing results for different expressions.
In our presentation we will first describe the implementation of the timing routines for each platform (Win-
dows, MacOS X, Linux and Solaris) and highlight some of the difficulties and limitations when measuring
sub-millisecond execution times. Topics covered include strategies for dealing with CPU frequency scaling
and multi-core system, why elapsed time is measured and not CPU time, why time is measured and not
clock cycles as well as ways to estimate the granularity of the underlying timing routine. We conclude with
a practical guideline for benchmark experiments on each of the operating systems mentioned.
After the viability of timing very short running R expressions has been established, some examples are
provided to show why it is useful to study the performance characteristics of such expressions at all. First
off, a baseline for the “speed” of the R interpreter is established by measuring the time it takes to execute the
simplest possible function1 . We consider this to be the equivalent of the cycle time of a microprocessor. Next
we extend this idea to S3 and S4 methods to see how much overhead one incurs for the method dispatch
in comparison to a simple function call. Afterwards we investigate the time it takes to generate a vector
containing the numbers from 1 to n (for different n) using the different methods available in base R. Time
permitting we conclude with a real world example where using high-level vectorized R functions actually
hurt performance when compared to a naive implementation using low level operations.

1
The simplest possible function being function() NULL.
Kaleidoscope IIIb, Thursday 14:00-15:00 143

Vector Image Processing

?
Paul Murrell1
1. Department of Statistics, The University of Auckland
?
Contact author: [email protected]

Keywords: image processing, vector graphics, importing graphics, SVG, javascript

This talk will describe a project to convert a static PDF map of a university campus into an interactive SVG
map, using R.
The conversion involves three steps: importing the original map into R, using the grImport package; pro-
cessing the map contents to identify important features; and using the gridSVG package to add interactivity
to the map and export it in an SVG format.
Both the import and export steps depend on some recent improvements in the handling of lines and text in
the grImport and gridSVG packages, plus the introduction of a new “path” primitive in the grid package.
These new graphics features and enhancements will be described and demonstrated.
There will also be a discussion of the image processing step and the addition of interactivity to the map. The
former will demonstrate the usefulness of treating an image as a source of data, which can be manipulated
using the standard data processing tools in R, and the latter will demonstrate a way to add simple interactivity
to the SVG output that is produced by the gridSVG package.

References
Murrell, P. gridSVG: Export grid graphics as SVG. R package version 0.6-0.

Murrell, P. (2009). Importing vector graphics: The grImport package for R. Journal of Statistical Soft-
ware 30(4), 1–37.

R Development Core Team (2011). R: A Language and Environment for Statistical Computing. Vienna,
Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0.
Regular Posters 144

Adding direct labels to plots

?
Toby Dylan Hocking1,2,
1. INRIA Sierra team for machine learning research, 23 avenue d’Italie, Paris, France
2. Institut Curie/INSERM U900/Mines ParisTech cancer bioinformatics group, 26 rue d’Ulm, Paris, France
?
Contact author: [email protected]

Keywords: Visualization, graphics, legends, direct labels

High-level plotting systems such as lattice and ggplot2 provide automatic legends for decoding
color labels in R plots (Sarkar, 2008; Wickham, 2009). However, with many colors, legends become
difficult to read, and direct labels are a more suitable decoding method:
> library(lattice)
> data(mpg, package = "ggplot2")
> p <- xyplot(jitter(cty) ~ jitter(hwy), mpg, groups = class,
+ main = "City and highway fuel efficiency depends on car class")
> library(directlabels)
> print(direct.label(p))
City and highway fuel efficiency depends on car class

35 ●
●

30 ●
● ●

subcompact ● ●
jitter(cty)

25 ● ●
● ● ● ●
● ● ●
● ● ●
●
● ● ●●●
20 ● ●
● ●
●
● ●
●●
●
●
●
● ●
●
● ●
●
compact
●●● ●●
● ● ●
●● ●
minivan ●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
● ●● ● ●●●
● ● ● ● ●
●
●
●● ●● ●
●●
● ●● ●
● ●●●
15 ●● ● ●●● ●●● ● ●● ● ●● ●●● midsize
●
●●
●● ● ●
●
● ●●
● ●
●
●
●
● ●● ●●
pickup ●
● ●
●
●
●
●● ●
● ● ● ●●
●
2seater
● ●
●
●●
● ●● ●●
● ● ●
10 ●●●
suv

20 30 40

jitter(hwy)

Direct labels are inherently more intuitive to decode than legends, since they are placed near the
related data. However, direct labels are not widely used because they are often much more difficult
to implement than legends, and their implementation varies between plotting systems.
The directlabels package solves these problems by providing a simple, unified interface for direct
labeling in R. Given a lattice or ggplot2 plot saved in the variable p, direct labels can be added by
calling direct.label(p,f) where f is a Positioning Method that describes where labels should be
placed as a function of the data. The power of this system lies in the fact that you can write your
own Positioning Methods, and that any Positioning Method can be used with any plot. So once
you have a library of Positioning Methods, direct labeling becomes trivial and so can more easily
used as a visualization technique in everyday statistical practice.

References
Sarkar, D. (2008). Lattice: Multivariate Data Visualization with R. New York: Springer. ISBN
978-0-387-75968-5.
Wickham, H. (2009). ggplot2: elegant graphics for data analysis. Springer New York.
Regular Posters 145

iRegression: a regression library to symbolic interval-valued

variables
Lima Neto Eufrásio1*, Souza Filho Cláudio1, Anjos Ulisses1

1. Departamento de Estatística, Universidade Federal da Paraíba, Cidade Universitária s/n, João Pessoa, PB, Brazil
*Contact author: [email protected]

Keywords: Regression, interval variable, symbolic data analysis.

Symbolic data analysis (SDA) has been introduced as a domain related to multivariate analysis, pattern
recognition and artificial intelligence in order to introduce new methods and to extend classical data
analysis techniques and statistical methods to symbolic data (Billard and Diday 2006). In SDA, a
variable can assume as a value an interval from, a set of real numbers, a set of categories, an ordered list
of categories or even a histogram. These new variables take into account the variability and/or
uncertainty presented in the data. Interval variables have been studied in the area of SDA, where very
often an object represents a group of individuals and the variables used to describe it need to assume a
value which express the variability inherent to the description of a group. Moreover, interval-valued
data arise in practical situations such as recording monthly interval temperatures in meteorological
stations, daily interval stock prices, among others. Nowadays, different approaches have been
introduced to analyze interval-valued variables. In the field of SDA, approaches to fit a regression
model to interval-valued data have been discussed in the literature. However, the access to such
methods still is restricted, being necessary to request to the authors. Billard and Diday (2000) were first
to propose an approach to fit a linear regression model to symbolic interval-valued data sets. Lima Neto
and De Carvalho (2008) improved the previous approach presenting a new method based on two linear
regression models, the first regression model over the midpoints of the intervals and the second one
over the ranges, which reconstruct the boundaries of the interval-values of the dependent variable in a
more efficient way when compared with the Billard and Diday's method. The aim of this work is to
development a R library, called iRegression, that includes some regression methods for interval-valued
variables. This new library will be the first one developed to treat symbolic data in the regression
context. Thus, some regression methods for symbolic interval-valued variables will be accessible to
students, teachers and professionals.

Acknowledgments : The authors would like to thank CNPq (Brazilian Agency) for their financial
support.

References

Billard, L. and Diday, E. (2006). Symbolic Data Analysis: Conceptual Statistics and Data Mining, John
Wiley, New York.

Lima Neto, E.A. and De Carvalho, F.A.T.. (2008), Centre and range method to fitting a linear
regression model on symbolic interval data. Computational Statistics and Data Analysis 52, pp. 1500-
1515
Regular Posters 146

Mixtures of Unimodal Distributions

?
Carlos E. Rodrı́guez1,2, , Stephen G. Walker1
We present a new Bayesian mixture model. The main idea of our proposal is to change the components
distribution of the mixture. Whereas the normal distribution is typically used as the kernel distribution,
it does have some serious issues for the modeling of clusters, see for example Escobar & West (1995) or
Richardson & Green (1997). If a cluster is skewed or heavy tailed, then the normal will be inefficient and
many may be needed to model a single cluster. Our intention is to use as kernel a family of distribution
functions for which the only constraint is that they are unimodal. Hence, we define a cluster as a set of data
which can be adequately modeled via a unimodal distribution. To construct unimodal distributions we use
a mixture of unform distributions with the Dirichlet Process, (Ferguson (1973) and Sethuraman (1994)), as
the mixing distribution. To sample from the correct posterior distribution we use the ideas of Kalli, Griffin
& Walker (2009), and Damien & Walker (2001), Tierney (1994), Green (1995) and Godsill (2001).

Keywords: Unimodal Distribution, Dirichlet Process, Markov chain Monte Carlo and Slice Sampler
1. International Society for Bayesian Analysis (ISBA)
2. Mexican Statistical Association (AME)
?
Contact author: [email protected]

References
[1] Ferguson, T. S. (1973). A Bayesian analysis of some nonparametric problems. Annals of Statistics, 1,
209-230.

[2] Sethuraman, J. (1994). A constructive definition of Dirichlet priors. Statistica Sinica, 4, 639-650.

[3] Godsill, S. J. (2001). On the relationship between Markov chain Monte Carlo methods for model
uncertainty. J. Computational Graphical Statistics, 10, 230-248.

[4] Green, P. J. (1995). Reversible jump Markov chain Monte Carlo computation and Bayesian model
determination. Biometrika, 82, 711-732.

[5] Richardson, S. and Green, P. J. (1997). On Bayesian analysis of mixtures with an unknown number of
components (with discussion). Journal of the Royal Statistical Society, Series B 59, 731-792.

[6] Escobar, M.D. and West, M. (1995) Bayesian density estimation and inference using mixtures. Journal
of the American Statistical Association. 90, 577-588.

[7] Tierney, L. (1994). Markov chains for exploring posterior distributions (with discussion). Annals of
Statistics 22, 1701-1762.

[8] Damien, P. and Walker, S.G. (2001). Sampling truncated normal, beta and gamma densities. Journal
of Computational and Graphical Statis- tics 10, 206-215.

[9] Kalli, M., Griffin, J.E. and Walker, S.G. (2009). Slice sampling mixture models. Statistics and Com-
puting. 21, 93-105.
Regular Posters 147

Accelerating Simulations in R using Automatically Generated

GPGPU-Code
Frank Kramer1*, Andreas Leha1, Tim Beissbarth1

1. Department of Medical Statistics, University Medical Center Göttingen, Göttingen, Germany

*Contact author: [email protected]

Keywords: GPGPU, CUDA, Simulation, simR2CUDA

Newly developed classifiers, estimators, and other algorithms are often tested in simulations to assess
power, control of alpha-level, accuracy and related quality criteria.

Depending on the required number of simulation runs

and the complexity of algorithms, especially the
estimation of parameters and the drawing of random
numbers under certain distributions, these simulations
can run for several hours or even days. Such simulations
are embarrassingly parallel and, therefore, benefit
massively from parallelization. Not everyone has access
to clusters or grids, though, but highly parallel graphics
cards suitable for general purpose computing are
installed in many computers. While there is a distinct
number of maintained R-packages available
(Eddelbuettel2011) that are able to interface with GPUs
and allow the user to speed up computations, integrating
these methods in simulation runs is often complicated
and mechanisms are not easily understood.

We present a new package that acts as a wrapper for

CUDA-implementations and offers a systematic way for
a statistician to accelerate simulations written in R. After
including the package simR2CUDA the user defines the
algorithm for a single simulation run in an R function and Figure 1: Illustration of generic process
passes the function, other required parameters and the flow in generated code
number of simulation runs to a compiler function. The
allowed syntax concerning flow-control for the function defining a single simulation run is restricted to
comply with GPU and implementation limitations. Given that a CUDA environment is correctly
installed and configured, a wrapper function, representing the whole simulation, and a CUDA function,
representing an individual simulation step, are generated and integrated into R using dyn.load.
Upon calling the generated code the simulation wrapper, a simple C function, splits up the independent
simulation runs for parallel execution on GPU threads and collects and returns the results, after
execution of all n runs on the GPU has finished (see Figure 1).

References

Eddelbuettel (2011). CRAN Task View: High-Performance and Parallel Computing with R,
http://cran.r-project.org/web/views/HighPerformanceComputing.html.
Regular Posters 148

The State of StatET

?
Stephan Wahlbrink1 , Tobias Verbeke2,
1. WalWare / wahlbrink.eu
2. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: R, IDE, Eclipse, StatET, integrated development environment

The StatET plug-ins for Eclipse offer the professional R developer and statistician an integrated develop-
ment environment since 2004. This presentation will look back on an interesting year of development and
demonstrate some of the hidden and less hidden new functionality that was introduced between UseR!2010
and UseR!2011.
WalWare et al. (2004–2011). Eclipse Plug-In for R: StatET. http://www.walware.de/goto/statet.
Regular Posters 149

The R Service Bus

?
Tobias Verbeke1,
1. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: R, SOA, ESB, enterprise service bus, software integration

R is a unique computational and visualization engine that can be useful to power data-driven decision making
in about any analytics problem. The complexity of many integration tasks, though, requires dedicated tools.
The R Service Bus is a swiss army knife that allows you to plug R into your processes independently of the
technology used by other software applications involved in the workflow. It has been in development for
over two years and released under open source license to the R community in April 2011.
In this presentation we will provide an overview of the architecture of the R Service Bus and show how the
architectural decisions guarantee reliability, flexibility, high availability and scalability of R-based analytics
applications. Next we will demonstrate how applications can connect to RSB using RESTful and SOAP-
based web services, e-mail protocols, messaging protocols (JMS, STOMP) etc. and which different payloads
can be used to unleash to power of R.
OpenAnalytics (2010–2011). The R Service Bus. http://www.openanalytics.eu/r-service-bus.
Regular Posters 150

Teaching Measurement Systems Analysis to Engineers Using R

?
Thomas Roth1
The Department of Quality Science - Technical University of Berlin
?
Contact author: Thomas Roth

Keywords: MSA, Capability, Gage R&R, ANOVA, DoE

Measurement Systems Analysis is an important aspect within the statistical education of engineers. While
classical design of experiments (DoE) is dealt with in statistical courses for engineers, more specific designs
and their analysis to identify the components of variation of measurement systems are rarely discussed.
These Gage Repeatability & Reproducibility studies as well as procedures relating to terms such as Bias,
Linearity and Gage Capability are subject of national and international standards and obligatory within but
not restricted to automotive industry.
Methods utilizing the comprehensive qualityTools package, summaries and graphs as well as an example for
teaching Measurement Systems Analysis to engineers as part of an obligatory statistics course for engineers
are illustrated. The conceptual design of the methods and the relation to national and international standards
are presented.

References
A.I.A.G. (2010). Measurement systems analysis: Reference manual (4 ed.). Detroit, Mich: DaimlerChrysler
and Ford Motor and General Motors.

Burdick, R. K., C. M. Borror, and D. C. Montgomery (2005). Design and analysis of gauge R&R studies:
Making decisions with confidence intervals in random and mixed ANOVA models. Philadelphia, Pa,
Alexandria, Va: Society for Industrial Applied Mathematics and American Statistical Association.

Herrmann, J. and T. Roth (2010). Qualitätsmanagement als Pflichtfach für Bachelor an der TU-Berlin.
In R. Schmitt (Ed.), GQW 2010 , (Aachen, Germany), // Unternehmerisches Qualitätsmanagement, pp.
205–217. Aachen: Apprimus-Verl.

ISO (2008). Quality management systems – Requirements (ISO 9001:2008).

ISO (2010). Statistical methods in process management – Capability and performance – part 7: Capability
of measurement processes (ISO 22514-7).

Roth, T. (2010). qualityTools: Statistical Methods for Quality Science.

Verband der Automobilindustrie e.V. (2010). Prüfprozesseignung: Eignung von Messsystemen, Mess- und
Prüfprozessen, Erweiterte Messunsicherheit, Konformitätsbewertung.
Regular Posters 151

The cards Package

?
Jason Waddell1,
1. OpenAnalytics BVBA
?
Contact author: [email protected]

Keywords: R, cards, card games, teaching, probability

Many statistics courses use cards as examples for presenting probability theory. The cards package pro-
vides tools for plotting of cards and card games, along with adaptable interactive examples for card-based
probability exercises and games.
In this presentation we will demonstrate functions for generating randomized decks and graphically dis-
playing cards. In addition we will demonstrate how the package can be used to create fully interactive card
games (such as casino-solitaire) that record and report play statistics.
Regular Posters 152

Using R in Insurance: Examples from Lloyd’s

Markus Gesmann1,*, Viren Patel1, Gao Yu1

1. Lloyd’s, One Lime Street, London EC3M 7HA

*Contact author: [email protected]

Keywords: Statistical Modelling, Data Management, Financial Analysis, Graphics, R packages

We propose to present about how the Analysis team of Lloyd’s is using R in its operations. The poster
focuses on 4 areas: statistical modelling, visualisation, usage of R and LaTeX to create management
information, and collaboration via R packages.

Statistical modelling: Understanding the volatility of the risks insured at Lloyd’s is crucial for pricing,
business planning and capital setting. The Analysis team uses R to fit loss distribution to the historical
data and Monte Carlo simulation to review the performance of syndicates and the market.

Visualisation: Data visualisation can help to provide early insight into favourable or unfavourable
development of the business environment. The Analysis team uses lattice and googleVis to visualize
the results of our work in a form that is easy to disseminate. As an example, we introduce the Statistics
Relating to Lloyd’s document, and how to use the googleVis package to display some of its
information.

Using R and LaTeX to create management information: In a market with over 80 syndicates it is
necessary to automatise the creation of performance reports for management information. Our team
plays back Benchmarking reports which allows management and underwriters to view their own
performance in the context of competitors. These reports are being produced using R and LaTex, among
other software tools.

Creating internal R packages: For easier availability and dissemination teams in Lloyd’s create
packages of their R functions. This includes both packages tailored to a specific task and general
packages to be used to comply with internal branding guidelines.

References

Lloyd’s (2010). Statistics Relating to Lloyd’s, http://www.lloyds.com/stats/.

Regular Posters 153

Solving Norm Constrained Portfolio Optimizations via

Coordinate-Wise Descent Algorithms
Yu-Min Yen1 and Tso-Jung Yen1
1. Department of Finance, LSE and Institute of Statistical Science, Academia Sinica
?
Contact author: [email protected]

Keywords: Portfolio optimization, norm constraint, coordinate-wise descent algorithm

In this paper we demonstrate that coordinate-wise descent algorithms can be used to solve portfolio selection
problems in which asset weights are constrained by lq norms for 1 ≤ q ≤ 2. A special case of the such
problems is when q = 1. The l1 norm constraint promotes zero values for the weight vector, leading
to an automatic asset selection for the portfolio. We first consider the case of minimum (global) variance
portfolio (mvp) in which the asset weights are constrained by weighted l1 and squared l2 norms. We use two
benchmark data sets to examine performances of the norm constrained portfolio. When the sample size is
not large in comparison with the number of assets, the norm constrained portfolio tends to have a lower out-
of-sample portfolio variance, lower turnover rate, less numbers of active assets and short-sale positions, but
higher Sharpe ratio than the one without such norm constraints. We then show some extensions; particularly
we derive an efficient algorithm for solving an mvp problem in which assets are allowed to be chosen
grouply. All of the program codes for the algorithms are written by R.
Regular Posters 154

Using R for air quality data analysis: A tool for designing

improved large-scale air pollution prevention programs
?
A. Alija1 , M. N. Gonzalez1 , A. Junquera-Perez1 , A. Ayesta1 , E. Setien1 , L. Garcia1 and J. Blanco1
1. Ingenieros Asesores S.A. Parque Tecnologico de Asturias. Parcela 39, 33428. Principado de Asturias

?
Contact author: [email protected]

Keywords: Air quality, Air pollution, openair, data mining.

Emissions of air pollutants derive from almost all economic and societal activities. The majority of green-
house gas emissions (GHG), acidifying substances, tropospheric ozone precursor emissions and material
input caused by the life-cycles of activities related to consumption can be allocated to the main consumption
areas of eating and drinking, housing and infrastructures, and mobility [Age11].
In this work, we present an intensive use of R and more concretely package openair [CR11]. Package
openair lead out to analyse long historical time series of air pollution data coming from a large urban region
somewhere [pri]. Long historical series of data have the inconvenience of they are very difficult to analyze
for several reasons. Perhaps, some of these reasons are: first, air pollution data is a data extremely correlated
in time and second, air pollution data is a complex data which require a very optimized visualization tools
to carry out even the simplest analysis. Therefore, much of the data available is only briefly analysed;
perhaps with the aim of comparing pollutant concentrations with national and international air quality limits.
Package openair, thanks to the specific functions, helps us to overcome some of these barriers.
We would like to thanks PhD. Lucas Fernandez Seivane for our helpfull discussions.

References
[Age11] EEA (European Environment Agency). Soer2010. http://www.eea.europa.eu/soer,
2011.

[CR11] David Carslaw and Karl Ropkins. openair: Open-source tools for the analysis of air pollution
data, 2011. R package version 0.4-0.

[pri] private. To preserve the confindenciality of the data source we will keep the location anonymous.
Regular Posters 155

Invasions by polymorphic species

Elizabeth Elliott1,* and Stephen Cornell1

1. University of Leeds
*Contact author: [email protected]

Keywords: Range-expansion; Evolution; Reaction-diffusion model; Numerical simulations

Two issues of current importance in ecology are the range expansion of species and invasion of exotic
organisms. Climate change is affecting the environments that species are currently adapted to, and to
survive species will either need to adapt to the new conditions or shift with their current environment.
Empirical observations and simulation models have shown that higher dispersal ability may lead to
increased rates of spread but little is known about the effect of having a community of dispersal
phenotypes on the rate of range expansion. We use a spatially explicit analytical model based on
partial-differential equations to investigate the invasion of a species with two dispersal phenotypes into
a previously unoccupied landscape. These phenotypes differ in both their dispersal and population
growth rate. Using analytical techniques and carrying out numerical simulations in R (R Development
Core Team, 2010) using the deSolve package (Soetaert et al., 2010), we find that the presence of both
phenotypes can result in faster range expansions than if only a single phenotype were present in the
landscape, and that typically the invasion can occur up to twice as fast as a result of this polymorphism.
This has implications for predicting species invasion speeds, suggesting that speeds cannot just be
predicted from looking at a single phenotype and that the presence of a community of phenotypes needs
to be taken into consideration.

References

R Development Core Team (2010). R: A language and environment for statistical computing. Vienna,
Austria, R Foundation for Statistical Computing.

Soetaert, K., Petzdolt, T. and Setzer, R. W (2010). Solving differential equations in R: Package
deSolve. Journal of Statistical Software 33, 1-25.
Regular Posters 156

Using R to Empower a New Plant Biology  

Naim Matasci1,*, Matthew Vaughn2, Nirav Merchant1,3

1. The iPlant Collaborative, BIO5 Institute, University of Arizona

2. The iPlant Collaborative, Texas Advanced Computing Center, University of Texas at Austin.
3. Biotechnology Computing Facility, BIO5 Institute, University of Arizona
*Contact author: [email protected]

Keywords: cyberinfrastructure, plant sciences, bioinformatics, high performance computing

The iPlant Collaborative is a U.S. National Science Foundation program to build a cyberinfrastructure
for the plant sciences that will enable new conceptual advances through integrative, computational
thinking. Teams composed of plant biologists and computer scientists work on grand challenges that
address fundamental questions in plant biology, in particular focusing on elucidating the link between
genotype and phenotype and on building a tree of life to represent the evolutionary history of all green
plants.
Countless computer programs address individual aspects of these grand challenges but the complexity
and sheer magnitude of the problem makes it almost impossible for individual researchers or research
groups to tackle such problems alone.
iPlant provides a cyberinfrastructure that allows researchers in the plant science community to
collaborate and share and integrate data and algorithms. Furthermore, by providing access to High
Performance Computing resources it permits the generation and analysis of very large datasets as well
as the usage of computationally intensive algorithms.
Given the growing importance of R in virtually every field of biology and bioinformatics, iPlant’s
cyberinfrastructure has been enriched by an interface to facilitate the development, execution and
distribution of R scripts.
Developers can work on constructing new tools in Atmosphere, iPlant’s virtual environment, through an
IDE and access data and workspaces located in their iPlant home directory on in shared locations. All
I/O operations are handled through iRODS, the Integrated Rule-Oriented Data System which permits
the sharing of data and provides direct read access to plant genomes and other large files. The use of
iRODS also ensures consistency of data across the different access point to iPlant’s cyberinfrastructure
(Atmosphere, API, Discovery Environment). Through iPlant’s API users and developers can execute
computationally intensive scripts on Texas Advanced Computing Center’s HPC resources. Finally, by
providing a simple metadata description, developers can create GUIs for their applications and make
them available to colleagues and collaborators only or to the community at large through iPlant’s
Discovery Environment.

References

Atmosphere: http://atmo.iplantcollaborative.org/
iRODS: https://www.irods.org/
iPlant’s Discovery Environment: http://preview.iplantcollaborative.org/de/
Regular Posters 157

Rknots, an R package for the topological analysis of knotted

biological polymers
?
Federico Comoglio1 , Maurizio Rinaldi2
1. Department of Biosystems Science and Engineering, Swiss Federal Institute of Technology (ETH) Zurich, Basel, Switzerland
2. Department of Chemical, Food, Pharmaceutical and Pharmacological Sciences (DiSCAFF), University of Piemonte Orientale
”A. Avogadro”, Novara, Italy
?
Contact author: [email protected]

Keywords: Rknots, knotted polymers, computational knot theory, topology, proteins

The topological study of biological polymers has led to important insights into their structural properties
and evolution Virnau et al. (2006). From a topological point of view polymers can be naturally modeled
as sequences of 3D points, i.e. open polygonal paths. Their closure generates classical objects in topology
called knots. Boosted by Taylor’s work Taylor (2000), the characterization of knotted proteins is a subject
of increasing interest in both experimental and computational biology due to the close structure-function
relationship and reproducible entangled folding. The exponential growth of the total number of structures
deposited into the Protein Data Bank (PDB) requires dedicated computational high-throughput methods able
to deal with a large amount of data. These methods combine a structure reduction scheme of a protein back-
bone model with the computation of a knot invariant. Being easy to compute, the Alexander polynomial
represents the current default choice Kolesov et al. (2007). Although this choice is primarily supported by
the fact that the knots so far identified in proteins are of the simplest types, the Alexander polynomial is
not able to discern knots chirality and its limited power is not optimal to develop a generalized framework
aiming to describe topological properties of 3D structures.
Recently, we developed a topological framework for the computation of the HOMFLY polynomial Freyd
et al. (1985), a more powerful invariant able to define knots chirality. By screening the entire PDB, we
obtained an up-to-date table of knotted proteins that also includes two newly detected right-handed trefoil
knots in recently deposited protein structures Comoglio and Rinaldi (2011).
However, the application of our framework is not limited to proteins and in order to provide an open-source
package to the scientific community working in the field, we developed and here we present Rknots, an
R package that includes functions and utilities to process and topologically explore polymers along with
dedicated utilities for working with protein structures. Among the implemented functions, this first release
includes PDB entry import or fetching (exploiting the flexibility of the R package bio3d), structure reduction
via the Alexander-Briggs algorithm Alexander and Briggs (1926) and the MSR (Minimal Structure Reduc-
tion algorithm) Comoglio and Rinaldi (2011), topological invariants and linking number computation, 3D
and knot diagram graphics function along with a knots and links dataset repository and utilities.
We welcome external contributions to Rknots in order to further extend the package functionalities and
provide a general purpose tool to identify knots in three-dimensional structures.

References
Alexander, J.-W. and G.-B. Briggs (1926). On types of knotted curves. Ann of Math 28, 562–586.

Comoglio, F. and M. Rinaldi (2011). A topological framework for the computation of the homfly polynomial
and its application to proteins. PLoS ONE in the press.

Freyd, P., D. Yetter, J. Hoste, W.-B.-R. Lickorish, K. Millett, and A. Ocneanu (1985). A new polynomial
invariant of knots and links. Bull Amer Math Soc (NS) 12, 239–246.

Kolesov, G., P. Virnau, M. Kardar, and L.-A. Mirny (2007). Protein knot server: detection of knots in protein
structures. Nucleic Acid Res 35, W425–W428.
Regular Posters 158

A specialised software for statistical applications in

macromolecular crystallography
?
James Foadi1, , Gwyndaf Evans2 , David Waterman3
1. MPL, Imperial College, London SW7 2AZ
2. Diamond Light Source Ltd, Harwell Science and Innovation Campus, Didcot, Oxfordshire OX11 0DE
3. CCP4 - Research Complex at Harwell (RCaH) - Harwell Science and Innovation Campus, Didcot, Oxfordshire OX11 0FA
?
Contact author: [email protected]

Keywords: Macromolecular crystallography

A new package for applications in macromolecular structural crystallography, cRy, has been developed in
the renowned R statistical platform. This is the first software of its kind and it is supposed to provide a bridge
between the large communities of crystallographers and professional statisticians. At present macromolec-
ular crystallographers make heavy use of large systems of programs, like CCP4 [1] or PHENIX [2], mainly
written in Fortran, C/C++ or Python. These programs handle the several statistical operations, normally
carried out in crystallography, with ad hoc routines, usually developed by different authors, often not shar-
ing a common statistical platform. Data and results are exchanged through files with well-defined formats.
The cRy package reads and writes files in the most commonly used crystallographic formats, carries out
all major crystallographic calculations and provides an interface between the crystallographic data structure
and the statistical objects and tools offered by R. cRy provides, thus, that common statistical platform that,
at present, is still lacking in structural crystallography. The code has been developed using S4 classes.

References

[1] Collaborative Computational Project 4 (1994). The CCP4 Suite of Programs for Protein Crystallography.
Acta Cryst. D, 50, 760–763.
[2] P. D. Adams, P. V. Afonine, G. Bunkóczi, V. B. Chen, I. W. Davis, N. Echoo Is, J. J. Headd, L. -W.
Hung, G. J. Kapral, R. W. Grosse-Kuntsleve, A. J. Terwilliger and P. H. Zwart (2010). PHENIX: a
comprehensive Python-based system for macromolecular structure solution. Acta Cryst. D, 66, 213–
221.
Regular Posters 159

EMA - A R package for Easy Microarray data Analysis

Nicolas Servant1,2,3,? , Eleonore Gravier1,2,3,4, , Pierre Gestraud1,2,3 , Cecile Laurent1,2,3,6,7,8 , Caroline
Paccard1,2,3 , Anne Biton1,2,3,5 , Isabel Brito1,2,3 , Jonas Mandel1,2,3 , Bernard Asselain1,2,3 , Emmanuel
Barillot1,2,3 , Philippe Hupé1,2,3,5
1. Institut Curie, Paris F-75248, France
2. INSERM, U900, Paris F-75248, France
3. Ecole des Mines ParisTech, Fontainebleau, F-77300 France
4. Institut Curie, Departement de Transfert, Paris F-75248, France
5. CNRS, UMR144, Paris F-75248, France
6. CNRS, UMR3347, Orsay F-91405, France
7. INSERM, U1021, Orsay F-91405, France
8. Université Paris-Sud 11, Orsay F-91405, France
?
Contact author: [email protected]

Keywords: Microarray analysis

The increasing number of methodologies and tools currently available to analyse gene expression microar-
ray data can be confusing for non specialist users. Based on the experience of biostatisticians of Institut
Curie, we propose both a clear analysis strategy and a selection of tools to investigate microarray gene ex-
pression data. The most usual and relevant existing R functions were discussed, validated and gathered in
an easy-to-use R package (EMA) devoted to gene expression microarray analysis.
Removing noise and systematic biases is performed using the most famous techniques for Affymetrix
GeneChip normalisation. The data are then filtered to both reduce the noise and increase the statistical
power of the subsequent analysis. Exploratory approaches based on R packages such as FactoMineR, or
mostclust and classically used to find clusters of genes (or samples) with similar profiles are also offered.
Supervised approaches, as Significance Analysis of Microarrays (siggenes package) approach or ANOVA
functions, are proposed to identify differentially expressed genes (DEG) and functional enrichment of the
DEG list is assessed based on GOstat package.
The package includes a vignette which describes the detailed biological/clinical analysis strategy used at
Institut Curie. Most of the functions were improved for ease of use (fewer command lines, default parame-
ters tested and chosen to be optimal). Relevant, enhanced and easy-to-interpret text and graphic outputs are
offered. The package is available on The Comprehensive R Archive Network repository.

References
Bertoni, A. and G. Valentini (2007). Model order selection for bio-molecular data clustering. BMC Bioin-
formatics 8 Suppl 2, S7.

Falcon, S. and R. Gentleman (2007). Using GOstats to test gene lists for GO term association. Bioinformat-
ics 23, 257–258.

Le, S., J. Josse, and F. Husson (2008). Factominer: an R package for multivariate analysis. Journal of
statistical software 25, 1–18.

Servant, N., G. Eleonore, P. Gestraud, C. Laurent, C. Paccard, A. Biton, I. Brito, J. Mandel, B. Asselain,
E. Barillot, and P. Hupé (2010). Ema - a R package for easy microarray data analysis. BMC Research
Notes 3, 277.

Tusher, V. G., R. Tibshirani, and G. Chu (2001). Significance analysis of microarrays applied to the ionizing
radiation response. Proc Natl Acad Sci U S A 98, 5116–21.
Regular Posters 160

Classification of Enzymes via Machine Learning Approaches

Neetika Nath*, John B. O. Mitchell

Biomedical Sciences Research Complex and School of Chemistry, University of St Andrews, KY16 9ST
*[email protected]

Keywords: Enzyme Classification, Machine Learning, R, Protein Function Prediction.

Abstract: We compare enzyme mechanistic descriptors derived from the MACiE (Mechanism,
Annotation and Classification in Enzymes) database [Holliday et al., 2006] and use multivariate
statistical analysis for assessment of enzyme classification. Each enzyme has an Enzyme Commission
(EC) number, a numerical code designed to classify enzymes by describing the overall chemistry of
the enzymatic reaction. The EC number system was devised five decades ago, in a pre-bioinformatics
age. As the volume of available information is increasing, a large number of informatics groups have
tried to use protein sequence and structural information to understand and reproduce the classification,
some of which have been successful [Cai et al., 2004]. Other groups have effected automatic EC
classification using chemoinformatics descriptions of the underlying reactions. Our objective is to
develop a computational protocol using the R package CARET [Kuhun et al., 2007] to predict EC
number from MACiE-derived descriptors. We evaluate 260 well annotated chemical reaction
mechanisms of enzymes using machine learning methods, placing them into the six top level EC
classes. Moreover, we compare the classification performances of three supervised learning
techniques, Support Vector Machine (SVM) [Vapnik, 1998], Random Forest (RF) [Breiman, 2001]
and K Nearest Neighbour (kNN), for the reaction mechanism classification task using five different
descriptor sets from MACiE data. Results: We found that all classifiers performed similarly in terms
of overall accuracy with the exception of K Nearest Neighbour analysis, which has the lowest
performance. The best performance was achieved by the Random Forest classifier.

References

Holliday, G.L., et al.. (2006) MACiE (Mechanism, Annotation and Classification in Enzymes): novel
tools for searching catalytic mechanisms http://www.ebi.ac.uk/thornton-srv/databases/MACiE/

Cai, C., L. Han, et al. (2004). "Enzyme family classification by support vector machines." Proteins
55: 66 - 76.

Kuhn, M., Wing, J., Weston, S, Williams A., Keefer, C. & Engelhardt, A. (2007) caret: Classification
and Regression Training, http://cran.r-project.org/web/packages/caret/

Vapnik, V. N. (1998) Statistical Learning Theory. New York: John Wiley and Sons.

Breiman, L. (2001) “Random forests,” Machine Learning, vol. 45, no. 1, pp. 5– 32.
Regular Posters 161

BeadDataPackR: A Compression Tool For Raw Illumina

Microarray Data
?
Mike L. Smith1, , Andy G. Lynch1
1. Department of Oncology, University of Cambridge, Cancer Research UK Cambridge Research Institute, Robinson Way, Cam-
bridge CB2 0RE, UK
?
Contact author: [email protected]

Keywords: Bioinformatics, Microarray, Illumina, Compression

Microarray technologies have been an increasingly important tool in biological research in the last decade,
and a number of initiatives have sought to stress the importance of the provision and sharing of raw microar-
ray data. Illumina BeadArrays provide a particular problem in this regard, as their random construction and
high number of replicate probes simultaneously adds value to analysis of the raw data and obstructs the shar-
ing of those data, with many of the standard repositories having no facilities for storing the raw data. This
leaves the burden of distributing such data with the individual researcher. The BeadDataPackR package
provides an interface to a novel compression scheme for raw Illumina BeadArray data, designed to ease the
storage and bandwith concerns that providing access to such data brings. It offers two key advantages over
off-the-peg compression tools. First it uses knowledge of the data formats to achieve greater compression
than other approaches, and second it does not need to be decompressed for analysis, but rather the values
held within can be directly accessed by existing tools, such as the popular beadarray package.

References
M.L.Smith and A.G.Lynch (2010). BeadDataPackR: A Tool to Facilitate the Sharing of Raw Data from
Illumina BeadArray Studies. Cancer Informatics 9, 217–227.
Regular Posters 162

GWAtoolbox: An R Package for Time Efficient Quality Control

of Multiple GWAS Data Files
?
Daniel Taliun1,2, , Christian Fuchsberger3 , Peter P. Pramstaller1 , Cristian Pattaro1 on behalf of the
CKDGen consortium
1. Institute of Genetic Medicine, European Academy Bozen-Bolzano (EURAC), Bolzano, Italy
2. Free University of Bozen-Bolzano, Bolzano, Italy
3. Department of Biostatistics, University of Michigan, Ann Arbor, MI
?
Contact author: [email protected]

Keywords: genome-wide association study, quality control, visualization

In the recent years, Genome-Wide Association Studies (GWASs) have been proven to be a very powerful
approach to uncover common genetic variants affecting human disease risk and quantitative outcome levels.
To date, 1212 genetic loci were reported to be significantly associated with at least one of 210 traits (Hin-
dorff et al. (2011)). To allow sufficient power to identify variants with small effects, GWAS sample size
has been augmented by pooling results from dozens of individual GWASs into large meta-analyses efforts.
However, combining results from a large number of GWASs, which differ in terms of study design, popula-
tion structure, data management, and statistical analysis, poses several challenges regarding the consistency
and quality of data which are usually difficult to be addressed systematically. This is mainly due to the
GWAS file size, which typically includes 2.5 to 7 million rows (corresponding to genetic variants) and > 9
columns (attributes). Consequently, the data harmonization across studies usually takes several weeks or
months.
While working in the CKDGen Consortium, aim to detect renal function genes (Köttgen et al. (2010)), we
have been performing meta-analyses of several GWASs. To standardize and speed up the quality control
(QC) process we developed the GWAtoolbox, an R package which lightens and accelerates the handling of
huge amounts of data from GWASs. GWAtoolbox provides time efficient QC of data stored in dozens of
files. Based on a simple configuration script, GWAtoolbox can process any number of files and produce
QC reports in a matter of minutes. QC reports consist of an extensive list of quality statistics and graphical
output presented using DHTML, which allow fast and easy inspection of individual data files. Additional
statistics and graphs allow quick identification of studies that are systematically different from the other
(outliers). The high time efficiency was achieved through the data reduction technique integrated into the
visualization pipeline. In particular, instead of passing all the million data points to the R plotting routines,
only a small part of data points is selected in a such way that preserves the quality of the final graphical
output. Additionally, the implementation of all computationally intensive steps was transferred to C++.
Through its extensive use in several current GWAS meta-analyses, GWAtoolbox has been proven to sig-
nificantly speed up the data management and to improve the overall meta-analysis quality. GWAtoolbox is
open source and available for MacOS, Linux, and Windows OS, at http://www.eurac.edu/gwatoolbox.

References
Hindorff, L. et al. (2011). A catalog of published genome-wide association studies. Available at:
www.genome.gov/gwastudies. Accessed March 23, 2011.

Köttgen, A. et al. (2010). New loci associated with kidney function and chronic kidney disease. Nat
Genet 42(5), 376–84.
Regular Posters 163

Is the seasonal variation in hospitalisations rates of atrial

fibrillation induced strokes in Denmark and New Zealand
dynamic?
?
Anette Luther Christensen1, , Simon Hales2 , Sren Lundbye-Christensen1 , Lars HVilsted
Rasmussen1 , Kim Overvad1 , Claus Dethlefsen1
1. Department of Cardiology, Center of Cardiovascular Research, Aalborg Hospital, Aarhus University Hospital, Aalborg, Denmark
2. Department of Public Health, University of Otago, Wellington, New Zealand
?
Contact author: [email protected]

Keywords: State space model, Model selection, Seasonal variation, Atrial Fibrillation, Stroke

Atrial Fibrilation (AF) is the most common cardiac arrythmia Friberg et al. (2003) and the hospitalisations
rates of AF have increased during the last two decades; in fact AF is considered an epidemic. The frequency
of AF increases with age and considering the population demographic it is expected that the number of
people with AF will be increasing remarkably. Having AF often leads to palpitations, respiratory distress,
and fatigue. Furthermore, AF is considered an independent risk factor for stroke Wolf et al. (1991). The
consequences of stroke on patients are crucial and may be a considerable burden on society regarding reha-
bilitation of stroke patients. It has been reported that hospitalisations with stroke exhibit seasonal variation
during the calendar year, however seasonal variations in AF induced strokes have not been investigated.
Furthermore, it has not been investigated whether there has been changes in the seasonal variation of stroke
hospitalisations. Knowledge of the seasonal variation of AF induced stroke, and possible changes over time,
may contribute to knowledge of the etiology of AF and may improve prophylaxis treatment in AF patients,
which may lead to improved prognoses for AF patients.
Using a state space model to fit daily incidence rates of AF induced strokes, accounting for a secular trend
and modelling the seasonal variation as a sum of sinusoids with different frequencies, it is possible to inves-
tigate whether hospitalisations of AF induced strokes exhibit seasonal variation. Furthermore, performing
model selection we may be able to investigate the dynamic nature of the seasonal variation over time. How-
ever, modelling the daily incidence rates of AF induced strokes as being Poisson distributed using a state
space model, as proposed by Lundbye-Christensen et al. Lundbye-Christensen et al. (2009), is not trivial in
regards to estimation algorithms and especially performing model selection.
We identified daily hospitalisations in Denmark using the Danish National Patients Registry, and in New
Zealand using the National Minimum Data Set. Both registries hold records of all hospitalisations, each
record includes information on date of hospitalisation and primary diagnosis as well as secondary diagnoses.
Analyses will be performed on data sets from each country separately, to make comparisons of seasonal
variation in incidence rates of AF induced strokes between the two countries. All analyses will be performed
in R using the package sspir.

References
Friberg, J., P. Buch, H. Scharling, N. Gadsbll, and G. B. Jensen (2003). Rates of Hospital Admissions for
Atrial Fibrillation. Epidemiology 13(6), 666–672.

Lundbye-Christensen, S., C. Dethlefsen, A. Gorst-Rasmussen, T. Fischer, H. C. Schønheyder, K. J. Roth-

man, and H. T. Sørensen (2009). Examing Secular Trends and Seasonality in Count Data Using Dynamic
Generalized Linear Models: a New Methodological Approach Illustrated with Hospital Discharge Data
on Myocardial Infarction. European Journal of Epidemiology 24(5), 225–230.

Wolf, P. A., R. D. Abbott, and W. B. Kannel (1991). Atrial Fibrillation as an Independent Risk Factor for
Stroke: the Framingham Study. Stroke 22(8), 983–988.
Regular Posters 164

Predictive Tool Development

A Time-to-event study of Diabetes Complications in Ontario
1* 1,2
Kelvin Lam , Douglas Manuel

1. Institute for Clinical Evaluative Sciences, Toronto, Ontario, Canada

2. The Ottawa Hospital Research Institute, Ottawa, Ontario, Canada
* Contact author : [email protected]

Keywords: Diabetes Complications, Competing Risk, Data Visualization, Administrative

Database, Diagnostic Tools

For physicians and people with diabetes, a simple and accurate predictive tool for
complications such as acute myocardial infarction (AMI) and cardiovascular disease (CVD)
will be invaluable for clinical decision making. Our study looked at how age and previous
medical history affects duration to these complications as well as death using the competing
risk methodology, where the latter served as the main competing event. R package cmprsk
was used to derive model parameters and baseline cumulative hazards. To determine the
model validity, diagnostics including c-index, deciles plots and quantile risk ratio plots will be
presented using ggplot2.

References
[1] Fine JP and Gray RJ (1999) A proportional hazards model for the subdistribution of a
competing risk. JASA 94:496-509.

[2] Putter H, Fiocco M, Geskus RB (2007). Tutorial in biostatistics: Competing risks and
multi-state models. Statistics in Medicine 26, 2389–2430.

[3] Bob Gray <[email protected]> (2010). cmprsk: Subdistribution Analysis of

Competing Risks. R package version 2.2-1. http://CRAN.R-project.org/package=cmprsk
Regular Posters 165

Analyzing the American Community Survey with R

Anthony Damico1,* , Rachel Licata1

1. Kaiser Family Foundation

*Contact author: [email protected]

Topic Area: Statistics in the Social and Political Sciences 
 
Keywords: American Community Survey, Health Policy, Big Data, SQL, County‐Level Data 
 
Research Objective: Designed to replace the information gathered by the decennial Census, the American 
Community Survey (ACS) is likely to play an increasingly important role in the future of health policy research, 
especially for regional and state‐level studies.  Despite powerful statistical analysis, data manipulation, and 
presentation capabilities, the open‐source R Statistical Software Package has not become widely adopted in the 
fields of health economics, health policy, or healthcare management and evaluation. User guidelines and other 
technical documents for government‐funded and publicly‐available data sets rarely provide appropriate syntax 
examples to R users. The objective of this presentation will be to describe the steps required to import ACS data 
into R, to prepare that data for analysis using the replicate weight variance and generalized variance formula 
calculation techniques, and to produce the principal set of statistical estimates sought by health policy 
researchers. 

Study Design: This presentation reviews the step‐by‐step method explanations and syntax needed to analyze the 
ACS with the R Statistical Software package.  This includes importation instructions, estimate calculations, 
variance and error term calculations, as well as linkages to other data sets.  In order to equip healthcare 
researchers with the tools needed to analyze this large dataset on their personal computers, each of these steps 
include a brief discussion of absolute minimum computing requirements, as well as detailed workarounds and 
shortcuts for the more memory‐intensive processes.  

Population Studied: The ACS represents all civilian, noninstitutionalized Americans.  The examples used in this 
presentation include state and regional estimates; however, all instructions and syntax are presented with the 
intention of allowing the researcher to re‐define a population of interest with minimal effort. 

Principal Findings: Depending on research budget, computing resources, and level of programming skill, 
conducting analyses of the ACS with R often presents a viable alternative to other statistical analysis packages. 

Conclusions: Given the large file size of the ACS, interested health policy researchers may be limited in their 
ability to analyze this survey with off‐the‐shelf statistical software packages due to memory overload issues.  
Although R users often experience similar memory limits and problems, the flexibility of the core R programming 
language and its integration with both parallel processing engines and Structured Query Language (SQL) allows 
for the relatively straightforward analysis of large, complex‐sample survey data such as the ACS. 

Implications for Policy, Practice or Delivery: Providing health policy researchers and statisticians with a strategy 
to work with the American Community Survey using freely available software will increase their ability to 
examine a variety of health and demographic indicators at the sub‐national level.  By outlining the technical 
steps to analyze this data, researchers could study topics such as regional characteristics of Health Professional 
Shortage Areas, state demographic factors associated with Medicare Advantage plan premiums, or county‐level 
demographics of uninsured populations.  An understanding of the methods needed to work with the ACS will 
open up the field of geographic region‐based analyses to health policy researchers. 
Regular Posters 166

Nonparametric estimation of a heaping mechanism for precise

and heaped self-report data
?
Sandra D. Griffith1, , Saul Shiffman2 , Daniel F. Heitjan1
1. Department of Biostatistics & Epidemiology, University of Pennsylvania
2. Department of Psychology, University of Pittsburgh
?
Contact author: [email protected]

Keywords: Digit preference, Coarse data, Rounded data, Smoking cessation, Measurement error

Open-ended numerical measures, often used in self-report to assess quantities or frequencies, exhibit a form
of measurement error termed heaping. Heaping occurs when quantities are reported with varying levels
of precision. Digit preference is a special case of heaping where the preferred values are round numbers.
Daily cigarette counts, for example, commonly exhibit heaps at multiples of 20, and to a lesser extent, 2, 5,
and 10, when measured by retrospective recall methods. Because heaping can introduce substantial bias to
estimates, conclusions drawn from data subject to heaping are suspect. Several methods have been proposed
to estimate the true underlying distribution from heaped data, but all depend on unverifiable assumptions
about the heaping mechanism. A data set in which subjects reported cigarette consumption by both a precise
method (ecological momentary assessment as implemented with a hand-held electronic device) and a more
traditional, imprecise method (timeline followback, or periodic retrospective recall) motivates our method.
We propose a nonparametric method to estimate the conditional distribution of the heaping mechanism given
the precise measurement. We measure uncertainty in the heaping mechanism with a bootstrap approach. We
describe our implementation of the method using R and graphically illustrate our results with the ggplot2
package. Application to our data suggests that recall errors are a more important source of bias than actual
heaping.

References
Griffith, S. D., S. Shiffman, and D. F. Heitjan (2009, November). A method comparison study of timeline
followback and ecological momentary assessment of daily cigarette consumption. Nicotine & tobacco
research 11(11), 1368–73.

Shiffman, S. (2009, September). How many cigarettes did you smoke? Assessing cigarette consumption by
global report, Time-Line Follow-Back, and ecological momentary assessment. Health psychology 28(5),
519–26.

Wang, H. and D. F. Heitjan (2008, August). Modeling heaping in self-reported cigarette counts. Statistics
in medicine 27(19), 3789–804.
Regular Posters 167

Forecasting multivariate time series using the DLM package:

An application to road traffic networks
Osvaldo Anacleto-Junior
Dept. of Mathematics and Statistics, The Open University, Milton Keynes, UK
Contact author: [email protected]

Keywords: DLM package, traffic networks, time series, graphical models

Traffic data have some characteristics that can be quite challenging to deal with from a statistical modelling
perspective. To have a broad view of the traffic network under analysis, information have to be collected
from a series of sites, which can generate a high-dimensional multivariate time series. Also, as management
decisions have to be taken in real time during periods like rush hours, traffic systems require forecasts in a
on-line environment.
Traffic flows in a road network can be modelled by a class of models known as graphical dynamic models.
These models use a directed acyclic graph (DAG) in which the nodes represent the time series of traffic
flows at the various data collection sites in a network, and the links between nodes represent the conditional
independence and causal structure between flows at different sites. The DAG breaks the multivariate model
into simpler univariate components, each one being a dynamic linear model (DLM). This makes the model
computationally simple, no matter how complex the traffic network is, and allows the forecasting model to
work in real-time. Also, graphical dynamic models can accommodate changes that may happen in a network
due to external events like traffic accidents and roadworks, which can heavily affect traffic flows and also
the structure of the network, leading to sudden changes in the data being analysed.
This poster will report an application of this class of model using the DLM package in a busy motorway
junction in the UK. It will be shown the advantage of some package features over standard DLM model
fitting approaches, focusing in particular on its numerically stable singular value decomposition-based al-
gorithms implemented for filtering.
Regular Posters 168

Who’s in the Waiting Room? Modelling Multivariate Time

Series of Counts of Patients to Hospital Emergency
Departments
?
Sarah Bolt1, , Ross Sparks1 , James Lind2
1. CSIRO Mathematics, Informatics and Statistics, North Ryde, NSW, Australia
2. Gold Coast Hospital, Southport, QLD, Australia
?
Contact author: [email protected]

Keywords: Multivariate, Times Series, Health Informatics.

While predictive tools are already being implemented to assist in forecasting the total volume of patients to
Emergency Departments [Jessup et al. (2010)], early detection of any changes in the types of patients presenting
would help authorities to manage limited health resources and communicate effectively about risk, both in
a timely fashion. But before we are able to detect changes we must understand and model the expected
counts of presentations for all possible subgroups. For example, we need to forecast the expected number
of patients on a given day at a particular hospital, with a particular disease, of a particular age, sex, etc.
So our objective was to model this large collection of interdependent time series. This problem presented
a number of issues including the sheer number of them and the fact that many of these time series had
very low counts and displayed overdispersion relative to a Poisson model. Furthermore, since the goal of
these models was to model away known behaviours, we had to incorporate effects including significant
seasonality, day of the week effects and some strongly interacting variables.
The method we present here divided the problem into two different components. The first used a regression
approach to model time series of aggregated counts. The second allocated proportions of these counts to
subgroups using a binary regression tree analysis. This method thus drew together R functionality from two
different areas:
• Analysis of times series of counts using glm {stats} [R Development Core Team (2010)], glm.nb {MASS}
[Venables and Ripley (2002)] and gamlss {gamlss} [Rigby and Stasinopoulos (2005)]. The latter function al-
lowed us to model counts with a negative binomial distribution and a modelled dispersion parameter.
• Allocation of counts to subgroups using the poisson regression tree approach of rpart {rpart}
[Therneau et al. (2010)].

References
Jessup, M., M. Wallis, J. Boyle, J. Crilly, J. Lind, D. Green, P. Miller, and G. Fitzgerald (2010). Imple-
menting an emergency department patient admission predictive tool. Journal of Health Organization and
Management 24, 306–318.

R Development Core Team (2010). R: A Language and Environment for Statistical Computing. Vienna,
Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0.

Rigby, R. A. and D. M. Stasinopoulos (2005). Generalized additive models for location, scale and
shape,(with discussion). Applied Statistics 54, 507–554.

Therneau, T. M., B. Atkinson, and B. D. Ripley. (2010). rpart: Recursive Partitioning. R package version
3.1-48.

Venables, W. N. and B. D. Ripley (2002). Modern Applied Statistics with S (Fourth ed.). New York:
Springer. ISBN 0-387-95457-0.
Regular Posters 169

Algorithm for defining hospital stays

Katrine Damgaard1*, Jon Helgeland1,

1. Norwegian Knowledge Centre for the Health Services

*Contact author: [email protected]

Keywords: Healthcare, hospitals, Multicore

Objective: The objective was to link hospital stays which constituted a complete chain of admissions
for patients transferred between wards/department within/between hospitals, in order to compare
mortality between hospitals

Material and methods: The Norwegian Knowledge Centre for the Health Services has developed a
system for retrieving data from the patient administrative systems (PAS) at all Norwegian hospitals.
By use of this system, we retrieved PAS data from each hospital for patients discharged during 2005-
2009.

Each record corresponds to one admission at one ward within a department in a hospital. To define a
complete patient stay, we have to aggregate the admissions; stays at one department, hospital stays and
stays involving more than one hospital. The latter is important for patients transferred between
hospitals to obtain their complete medical treatment history. All permanent residents in Norway have a
personal identification number (PIN) which enables linking between hospitals.

We have developed an algorithm that concatenates the ward admissions to a chain of admissions. The
function gives you all the levels of stays described above. The input to the function is a serial number,
PIN, date of admission, date of discharge, hospital, department and further optional parameters. The
function uses the package multicore in R, for the ability to use more than one processor for large
datasets. One of the properties of the algorithm is that you can choose different time windows, or
tolerances, for concatenating the ward admissions. We chose 24 hours as tolerance for our purpose.

Results: With the choice of 24 hours time window for difference between time of discharge and next
admission, we found 10 485 022 hospital stays involving one or more hospitals out of 16 370 163
ward admissions from 3 304 546 patients.

Conclusion: The algorithm can be used to aggregate large patient administrative data sets, with
acceptable running times. The tolerance limit can be adjusted to suit the purpose and the data at hand.
Late-breaking Posters 170

Using merror 2.0 to Determine Measurement Bias and Imprecision

Richard A. Bilonick1,2,*

1. University of Pittsburgh School of Medicine, Dept. of Ophthalmology

2. University of Pittsburgh Graduate School of Public Health, Dept. of Biostatistics
*Contact author: [email protected]

Keywords: Measurement Error, Calibration, Bias, Imprecision, Grubbs' Estimator

Researchers often need to evaluate the agreement between n=2 or more devices (instruments,
methods) based on measurements made jointly on N items (subjects). In the simplest case, each
measurement X ij for device i and item j is described as:
X ij =i i  jij
where  i and i describe how device i distorts (biases) the unknown true values  j , and  ij is a
Normally distributed random error with mean 0 and standard deviation (SD)  i for device i.  could
be distributed Normally with mean   and SD  in some applications. For uniqueness, the geometric

average of all of the i is constrained to equal one. Jaech (1985) describes many approaches to
estimating the parameters including method of moments (MOM) and maximum likelihood (ML).
Estimates of these parameters will provide calibration curves relating any two devices, for example.
For simple cases of 3 or more devices and only one measurement from each device, the merror
package provides several useful functions. ncb.od (non-constant bias - original data) computes ML
estimates of the bias and imprecision parameters based on the N ×n rectangular array of
measurements. You must have at least n=3 devices (or independently repeat at least 1 of two devices)
because otherwise there is not enough information to compute all the parameter estimates uniquely
(without making very strong assumptions about the parameters). Essentially, ncb.od fits a
measurement error structural equation model with just one latent variable (for  ). ncb.od will
compute estimates of the  i , i ,  i for each device, and  (with confidence intervals for the  i
and  ). You should be aware, however, that only certain functions have meaning: 1) when the i do
not all equal 1, then you need the ratios of the i to check for scale bias (devices have different sized
measurement units, i.e., a non-constant bias or bias that changes with the level of the true value), 2)
when the i all equal 1, then differences among the  i estimate constant relative biases, and 3) the
device imprecision  i must be divided by the corresponding i in order to be compared (otherwise,
you are dealing with incomparable results from different measuring systems). Also included is the
function lrt to do likelihood ratio tests of the i . To visualize the data, use pairs.me to make
modified pairs plots where all the axis limits are the same for all plots with the line of equality (no
bias) indicated. Finally, precision.grubbs.cb.pd and precision.grubbs.ncb.od are
provided to compute estimates of the imprecision SD's based on, respectively, using paired data under
the constant bias model, and using the original data under the non-constant bias model. To implement
these functions, the Fortran code from Jaech (1985) was translated to R. Some minor computational
mistakes in the Fortran code were fixed when re-writing in the R programming language.

References

Jaech, J. L. (1985). Statistical Analysis of Measurement Errors, John Wiley & Sons, New York.
Late-breaking Posters 171

Using Metabolomics to monitoring kidney transplanted patients:

transplant from a living donor.
M. Calderisi1 , A. Vivi1 , M. Tassini1 , M. Carmellini2

1. NMR Centre University of Siena, Italy

2. Department of Surgery and Bioengineering, University of Siena, Italy
*Contact author: [email protected]
The work was partially supported by Consorzio Interuniversitario Trapianti d'Organo (www.consorziotrapianti.it)

Keywords: metabolomics, chemometrics, transplant, living donor, NMR

NMR spectroscopy is a analytical platform for biological matrices, it enables the quick generation of
spectral profiles of low molecular weights metabolites. Metabonomics is a system approach used also to
investigate the metabolic profile by multivariate data analysis tools. The aim of the work is to control
the kidney graft recovery process in an non-invasive way, estimating the process by 1H-NMR
spectroscopy and chemometrics on urine samples as they contains metabolites concerning the
pathological and clinical state of the patients. The goal of this statistical analysis is the identification of
the dynamic biopatterns (individual patient recovery trajectories) related to the transplant. Samples have
been collected every day, for both patients, along all the recovery period. The donor is a 60 years old
healthy woman (6 samples), the recipient is his son, a 31 years old man (10 samples).
Data have been pre-processed (baseline correction, spectra alignment, water signal exclusion and
normalization to unit length) and pre-treated (mean centering and log10 transformation). The log10
transformation has been especially useful to reduce the natural difference between high-intensity and
low-intensity peaks, without boosting noise. Data have been analyzed by means of PCA and significant
signal regions have been identified by selection of the highest loadings (in absolute value).
The scores plot in figure clearly shows the
difference between the metabolic profile
before (right side) and after (left side) the
surgery (that happen during day 3) for
recipient (ric) and his recovery trajectory
occurring from day 5 to day 10. The
metabolic profile of the donor (don) goes
from a "normal" state (bottom right of the
scores plot) to an altered state (top right) after
the surgery day, then goes back to its original
condition.
In conclusion, it is possible to state that NMR
spectra plus chemometrics elaboration could
be used to monitor a kidney graft recovery
progress. Data elaboration has been carried
out using FactoMiner and ptw and packages.
References
Francois Husson, Julie Josse, Sebastien Le and Jeremy Mazet (2010). FactoMineR: Multivariate
Exploratory Data Analysis and Data Mining with R. R package version 1.14. http://CRAN.R-
project.org/package=FactoMineR

Bloemberg, T. G. et al. (2010) "Improved Parametric Time Warping for Proteomics", Chemometrics
and Intelligent Laboratory Systems, 104 (1), 65-74
Late-breaking Posters 172

hyperSpecGUI:
Graphical Interaction in Spectroscopic Data Analysis
?
Sebastian Mellor1 , Claudia Beleites2, , Colin Gillespie1 , Christoph Krafft2 , and Jürgen Popp2,3
1. School of Mathematics & Statistics, Newcastle University, Newcastle/UK.
2. Institute of Photonic Technology, Jena/Germany.
3. Institute of Physical Chemistry and Abbe Center of Photonics, University Jena/Germany.
?
Corresponding author: [email protected]

Keywords: spectroscopy; Google Summer of Code; GUI.

Spectroscopic data analysis includes highly visual tasks that need graphical user interaction. We present
hyperSpecGUI, a companion package enhancing hyperSpec. Both packages are hosted at hyperspec.r-
forge.r-project.org.
The hyperSpecGUI packages follows the “specialise on a single thing” paradigm, as opposed to implement-
ing an integrated development environment (IDE) or all-(spectroscopic)-purpose GUI. Visual interaction is
provided by small, specialised GUIs (applets, dialogs) that are called as functions. This allows full GUI
interaction with scripts and easy integration with IDEs. In particular, this allows:
Fast work-flow creation: rather than a single GUI, we use a combination of small, bespoke GUIs.
Flexibility: the package is extensible and easily adapted to new and unforeseen tasks.
Compatibility with batch processing: spectroscopic data sets are large with data analysis often including
high performance computations on remote servers that do not provide graphical interaction. Swapping
between GUI-based and command line based work/batch processing are smooth and easy.
Reproducibility: interactive tasks can be recorded, allowing literate programming techniques to be used,
e.g. Sweave.
Interactive spike correction of Raman spectra: when cosmic rays hit the detector, spikes are observed in
the Raman spectra. Several strategies to identify these artifacts exist and work well with high, sharp spikes.
However, manual control and adjustment of the results is necessary as broader artifacts may be confused
with sharp Raman bands and vice versa. Also the borders of broader artifacts are difficult to detect.
As an example, we present a GUI for spike filtering of Raman data, demonstrating the wrapping of this GUI
with existing IDEs/editors.
Acknowledgements. SM’s work is funded by the Google Summer of Code 2011. CB, CK, and JP acknowl-
edge funding by the European Union via the Europäischer Fonds für Regionale Entwicklung (EFRE) and
the “Thüringer Ministerium für Bildung, Wissenschaft und Kultur” (Project: B714-07037).
Late-breaking Posters 173

Visualisation for three-dimensional shape data

?
Stanislav Katina1, and Adrian Bowman1
1. School of Mathematics and Statistics, The University of Glasgow, University Gardens, Glasgow G12 8QQ, Scotland, UK
?
Contact author: [email protected] [Late Breaking Poster Abstract, topic: Visualization & Graphics]

Keywords: Procrustes shape coordinates, thin-plate splines, 3D facial surfaces, shape index

In Late Breaking Poster we present visualisation of shape-space Principal Component Analysis (PCA)
of (semi)landmarks from 3D facial surfaces. This multivariate model decomposes shape signal into low-
dimensional linear combinations of high-dimensional measurements. Using Generalized Procrustes Analy-
sis (Bookstein 1991), Procrustes Shape Coordinates (PSC) are calculated after suitable re-scaling, transla-
tion, and rotation of the data and then decomposed to affine and non-affine components. PCA is performed
by singular value decomposition of the covariance matrix of centered PSC.
The data consists of 3D coordinates of 23 landmarks and 1664 semilandmarks on curves and surface patches
taken from facial surfaces. Then, suitably scaled eigenvectors of the covariance matrix mentioned above
were added to the symmetrised mean (reference shape) calculated by relabeling and reflecting the mean
shape (semi)landmark coordinates. Using a carefully chosen mesh of 59242 points triangulated by 117386
faces and derived 1694 (semi)landmarks, we calculated standardized mesh points of the reference shape
(called Thin-Plate Spline morph) by TPS interpolation model. This process is performed in an anatomically
and geometrically meaningful manner, similar to the dense correspondence model of Mao et al. (2006)
where the sum of principal curvatures is equivalent to bending energies. To explore the nature of particular
PCs, we visualised 4D maps of signed pointwise Euclidean distances (in the sense of inwards/outwards
facial direction) between shapes equivalent to extremes of the PC scores. Furthermore, these distances
were decomposed to x, y, and z components of a suitably oriented face. Additionally, a shape index (SI)
was calculated from principal curvatures derived from the Weingarten curvature matrix as a function of
coefficients of a locally fitted parabolic regression model to facial coordinates (Goldfeather and Interrante
2004). The differences in SI between shapes equivalent to both extremes of PC scores were visualised as
well.
The programs are written in R language (R Development Core Team 2011). The proposed visualisation
is useful for interpretation of the effects of particular shape changes with direct practical implications in
applied craniometrics and in anthropometrics more generally.

Acknowledgement

The research was supported by Wellcome Trust grant WT086901MA.

References

Bookstein, F.L., 1991: Morphometric Tools for Landmark Data. New York, Cambridge University Press
Goldfeather, J., Interrante, V., 2004: A novel Cubic-Order Algorithm for Approximating Principal Direction
Vectors. ACM Transactions on Graphics 23, 1: 45-63
Mao, Z., Ju, X., Siebert, J.P., Cockshott, W.P., Ayoub, A., 2006: Constructing dense correspondences for
the analysis of 3D facial morphology. Pattern Recognition Letters 27: 597–608
R Development Core Team, 2011: R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria
Late-breaking Posters 174

Bayesian Hierarchical Clustering for Microarray Time Series Data

with Replicates and Outlier Measurements
Emma J Cooke1, Paul D W Kirk2, Richard S Savage2*, David L Wild2*

1. Department of Chemistry, University of Warwick, Coventry UK

2. Systems Biology Centre, University of Warwick, Coventry UK
*Contact authors: [email protected], [email protected]

Keywords: Systems biology, Bayesian hierarchical clustering

A key aim in systems biology is to link modelling of the interactions of system components with high
throughput data. Time series experiments have become increasingly common, necessitating the
development of novel analysis tools that capture the resulting data structure. R/BHC is a widely used
package in R/Bioconductor for fast Bayesian hierarchical clustering of gene expression data (Savage et
al, 2009). We present an extension to this package that enables time series data to be modelled and
clustered, using Gaussian process regression to capture the structure of the data. Using a wide variety
of experimental data sets, we show that our algorithm consistently yields higher quality and more
biologically meaningful clusters than current state-of-the-art methodologies. Our approach permits a
small proportion of the data to be modelled as outlier measurements, which allows noisy genes to be
grouped with other genes of similar biological function. Our method exploits replicate observations to
inform a prior distribution of the noise variance, which enables the discrimination of additional distinct
expression profiles. These extensions will be included in the next release of R/BHC.

References
 
Savage, R. S., Heller, K., Xu, Y., Ghahramani, Z., Truman, W. M., Grant, M., Denby, K. J., and Wild,
D. L. (2009). R/BHC: fast Bayesian hierarchical clustering for microarray data, BMC
Bioinformatics 10, 242
 
 
Late-breaking Posters 175

Arbitrary Accurate Computation with R: Package

’Rmpfr’
?
Martin Mächler1,2,

1. ETH Zurich (Seminar for Statistics), Switzerland

2. R Core Development Team
?
Contact author: [email protected]

Keywords: Arbitrary Precision, High Accuracy, Multiple Precision Floating-Point, Rmpfr

The R package Rmpfr allows to use arbitrary high precision numbers instead of R’s double precision
numbers in many R computations and functions.
This is achieved by defining S4 classes of such numbers and vectors, matrices, and arrays thereof,
where all arithmetic and mathematical functions work via the (GNU) MPFR C library, where
MPFR is acronym for “Multiple Precision Floating-Point Reliably”. MPFR is Free Software,
available under the LGPL license, and itself is built on the free GNU Multiple Precision arithmetic
library (GMP).
Consequently, by using Rmpfr, you can often call your R function or numerical code with mpfr–
numbers instead of simple numbers, and all results will automatically be much more accurate.
> options(digits = 17)# to print to full "standard R" precision
> .N <- function(.) mpfr(., precBits = 200)
> exp( 1 )
[1] 2.7182818284590451
> exp(.N(1))
1 'mpfr' number of precision 200 bits
[1] 2.7182818284590452353602874713526624977572470936999595749669679

Applications by the package author include testing of Bessel or polylog functions and distribution
computations, e.g. for stable distributions. In addition, the Rmpfr has been used on the R-
help or R-devel mailing list for high-accuracy computations, e.g., in comparison with results from
commercial software such as Maple, and in private communications with Petr Savicky about fixing
R bug PR#14491.
We expect the package to be used in more situations for easy comparison studies about the accuracy
of algorithms implemented in R, both for “standard R” and extension packages.

References
Fousse L, Hanrot G, Lefèvre V, Pélissier P, Zimmermann P (2007). “MPFR: A multiple-precision
binary floating-point library with correct rounding.” ACM Trans. Math. Softw., 33, 1–13. ISSN
0098-3500. URL http://doi.acm.org/10.1145/1236463.1236468.
Fousse L, Hanrot G, Lefèvre V, Pélissier P, Zimmermann P (2011). MPFR: A multiple-precision
binary floating-point library with correct rounding. URL http://mpfr.org/.
Granlund T, the GMP development team (2011). GNU MP - The GNU Multiple Precision Arith-
metic Library. URL http://gmplib.org/.
Maechler M (2011). Rmpfr: R MPFR - Multiple Precision Floating-Point Reliable. R package
version 0.3-0, URL http://rmpfr.r-forge.r-project.org/.
Late-breaking Posters 176

A dissimilarity based on relevant statistics

⋆
Antonio Miñarro1, , Marta Cubedo1 , Josep M. Oller1
1. Department of Statistics. University of Barcelona
⋆
Contact author: [email protected]

Keywords: Information metric, Rao distance, Multivariate distances.

A dissimilarity index between statistical populations in the absence of model assumptions is constructed.
The index is based on some properties of the information metric as can be seen in Cubedo et al. (2011).
We are interested to define a distance between p different statistical populations Ω1 , . . . , Ωp , obtaining, for
every statistical population, a data matrix Xi of order ni × m.
We shall assume that the statistical populations may differ in some vectorial parameter of interest ξ =
(ξ1 , . . . , ξk )t , although we shall not assume a particular parametric statistical model for the distribution of
X.
We define certain convenient statistics T = (T1 , . . . , Tk )t , that we assume are unbiased and consistent
estimators of the parameters of interest ξ for each population. The dissimilarity index δij , is defined by:
√
1 b
δij = (Tj − Tbi )t Ψ
b −1 (Tbj − Tbi )
i
ni
where Ψb i is an unbiased estimation of Ψi = cov(T |Ωi ) i = 1, . . . , p, constructed by Bootstrap methods
by re-sampling the rows of the data matrix Xi .
∆ = (δij ) is a p × p non symmetric dissimilarity matrix between the populations Ω1 , . . . , Ωp . It is possible
to approach ∆ by a symmetric matrix D using the trace norm D = (∆ + ∆t )/2
We have used simulations to assess the performance of the dissimilarity defined above, which we called
Relevant Statistics Dissimilarity (RSD), in comparison to the usual Mahalanobis distance and to the Siegel
distance between multivariate normal populations, Calvo et al. (2002).
An R script has been developed to perform simulations, bootstrap and compute dissimilarity matrices. In
addition, a Multidimensional Scaling has been realized to visualize and compare the differences between
our dissimilarity and Mahalanobis and Siegel distances. Implementing the methodology in a R package will
be discussed.

References

Cubedo, M., Miñarro, A., Oller, J.M. (2011). Some Geometrical Remarks in Statistics. In XIIIth Spanish
Biometry Conference and 3rd Ibero-American Biometry Meeting, (Barcelona, Spain).
Calvo, M., Villarroya, A., Oller, J.M. (2002). A biplot method for multivariate normal populations with
unequal covariance matrices. Test 11, 143–165.
Late-breaking Posters 177

BMEA: An R package for the analysis of Affymetrix® Exon

Array Data

Stephen Pederson1*, Gary Glonek2, Simon Barry1

1. School of Paediatrics & Reproductive Health, University of Adelaide, SA, Australia 5005
2. School of Mathematical Sciences, University of Adelaide, SA, Australia 5005
*Contact author: [email protected]

Keywords: Bayesian Statistics, Bioinformatics, Exon Array, Gene Expression, Alternate Splicing

Microarrays have been shown to provide consistent & accurate estimates of comparative gene
expression levels1. Early generations of Affymetrix® arrays use probes targeting the 3’ end of an
mRNA transcript and were restricted to detection of changes in gene expression levels. The newer
generation of arrays, such as Exon arrays, use probes which target the entire length of a transcript and
are additionally capable of detecting changes in gene structure via alternate splicing.

Building on the earlier probe-level modelling2 approach for 3’ Arrays, Bayesian Modelling for Exon
Arrays as implemented in the BMEA package, has been developed using a hierarchical Bayesian
model for detection of differentially expressed genes and alternate splicing events when applied to
Exon Array data. Comparison of the BMEA approach with existing methods such as FIRMA3 shows
a strong improvement in performance for detection of both differentially expressed genes & alternate
splicing events, and through a lack of reliance on annotated transcripts, will serve to complement
existing Bayesian approaches such as MMBGX4.

References

1. Robinson, M. and Speed, T (2007) A comparison of Affymetrix gene expression arrays BMC
Bioinformatics 8 (449)
2. Bolstad, BM (2004) Low Level Analysis of High-density Oligonucleotide Array Data:
Background, Normalization and Summarization. PhD Dissertation. University of California,
Berkeley.
3. Purdom, E., Simpson, K., Robinson, M., Conboy, J., Lapuk, A. and Speed, T (2008) FIRMA:
a method for detection of alternative splicing from exon array data. BMC Bioinformatics
24(15):1707–1714
4. Turro, E., Lewin, A., Rose, A., Dallman, M. and Richardson, S (2009) MMBGX: a method
for estimating expression at the isoform level and detecting differential splicing using whole-
transcript Affymetrix arrays Nucleic Acids Research 38(1):e4
Late-breaking Posters 178

Interactive Applets for Teaching with R

Andrew Rich, Danny Kaplan, Randall Pruim, Nicholas J. Horton,
Joseph J. Allaire

The perceived lack of “user friendliness” in R is sometimes cited as an ob-

stacle to the adoption of R in teaching. Many teachers believe a mouse- and
menu-driven graphical interface is essential to success in the classroom, while
others think that learning a notation of computer commands is an important
component of a student’s education in the contemporary world shaped by pow-
erful computation and ubiquitous data. There are elements of truth in both
views, suggesting that a synthesis is appropriate: providing interactive graphi-
cal “applets” within the context of a command-driven system.
In this poster, we describe an easy-to-program facility for making easy-to-
use interactive “applets” in R using the manipulate package in the RStudio
interface for R. We have been developing a suite of such applets for teaching
introductory statistics, calculus, and epidemiology. Some examples: a graphical
system for selecting model terms in statistical models; a display of a mathe-
matical function together with its derivatives and integrals to show the rela-
tionship among them; an interface for exploring parameters in SIR models to
demonstrate the different possible outcomes of in the spread of infectious dis-
ease. Together with a set of command-line utilities for teaching statistics and
calculus provided through the mosaic package, the applets should increase the
attractiveness of teaching with R, and the manipulate package can empower
instructors, even those with modest programming abilities, to create their own
applets.

1
Late-breaking Posters 179

How engineers will learn R from motorcycle tires !

Koo Rijpkema*

Eindhoven University of Technology, Department of Mathematics and Computer Science, the Netherlands
*Contact author: [email protected]

Keywords: Statistics Education, Design of Experiments, packages BHH2, DoE.base, DoE.wrapper,

rsm

For 1st and 2nd years’ students from the Mechanical Engineering
department at the Eindhoven University of Technology in the
Netherlands the basic ideas of engineering statistics and
statistical/experimental reasoning are introduced through a Design-
Based Learning project. In this project about 15 groups of 7
students each are (simultaneously) introduced to the problem of
instability of motorcycle road handling, such as wobble. v FL
Characteristics of the motor cycle tire play an important role in this
and students have to find out through real physical experiments

how pressure of the tire and vertical load influence these
characteristics. For conducting the experiment they can use facilities we have available in the
Automotive Lab at the Mechanical Engineering department, more specifically the flat-plank
experimental setup that is available for measuring both static and dynamic characteristics of tires.

measuring hub
obstacle road

wheel carrier air spring system

The experimental set-up used in the project.

In the project a ‘distributed experimental set-up’ is performed, where different groups measure
characteristics at different factor-levels combinations. Data obtained by individual groups have to be
combined, explored and analysed, aiming at an adequate fit of a Response Surface Model, to be used
for prediction. As this project is the students’ first introduction to modern computer aided statistical
analysis they will primarily use the R Commander interface. However, the project serves as a start-up
for more advanced courses on Engineering Statistics and Design of Experiments [1], where they will
use R for more customized analyses, using and adapting contributions from relevant packages such as
BHH2, DoE.base, DoE.wrapper and rsm [2,3,4].
 

References
[1] Box, G.E.P., J.S. Hunter and W.G. Hunter, (2005). Statistics for Experimenters: Design,
Innovation, and Discovery, 2nd Edition. New York: Wiley.
[2] Groemping, U. (2009). Design of Experiments in R . Presentation at UseR! 2009 in Rennes, France
[3] Lenth, R.V. (2009). Response-Surface Methods in R, Using rsm . Journal of Statistical Software
32 (7), 1-17
[4] Vikneswaran (2005). An R companion to "Experimental Design". URL http://CRAN.R-
project.org/doc/contrib/Vikneswaran-ED_companion.pdf
Late-breaking Posters 180

plotKML: a framework for visualization of space-time data in

virtual globes
?
Pierre Roudier1, , Tomislav Hengl2 , Dylan Beaudette3
1. Landcare Research, New Zealand
2. ISRIC – World Soil Information, The Netherlands
3. Natural Resources Conservation Service, USDA, USA
?
Contact author: [email protected]

Keywords: Spatial data, data visualisation

KML (formerly Keyhole Markup Language), an XML-based format, is an OGC (Open Geospatial Consor-
tium) standard to represent spatial data. Use of KML has been popularized with the development of virtual
globe Google Earth (McGavra et al. 2009). Since it approval as an OGC standard, support for KML has
been introduced in other virtual globes and geographic information systems (Craglia et al. 2008).
KML support is already available in R through the GDAL external library, whose bindings are provided the
rgdal package, or directly implemented in R (packages maptools and raster). However, the support of the
KML specifications is only partial. On the other hand, there is an increasing demand for tools to visually
explore spatio-temporal patterns in a variety of environmental data ( Andrienko & Andrienko, 2006).
The plotKML package provides the user with tools to plot spatial and spatio-temporal objects as described
by sp classes into a KML file, and proposes an easy syntax to make complete use of the KML specifications.
It aims at bringing virtual globes as a full-featured plotting canvas for spatial data, and thus provide not
only methods to write various geometries in KML, but also plotting essentials, such as legend frames and
metadata subsets.
Three levels of functions are provided: (i) a general, user-friendly kml() method, (ii) a set of intermediary
functions to generate a KML file layer by layer, and (iii) and advanced templates for creating more advanced
KML files.

(a) (b)

Figure 1: Examples of plotKML outputs. (a) Multi-layer visualisation of predictions and sampling loca-
tions. (b) Soil profile plot showing changes of measured soil organic carbon with depth.

Andrienko, N., and Andrienko, G. (2006) Exploratory analysis of spatial and temporal data: a systematic approach. Springer, 703 p.
Craglia, M., Goodchild, M., Annoni, A., Camara, G., Gould, M., Kuhn, W., Mark, D., Masser, I., Maguire, D., Liang, S., and Par-
sons, E. (2008). Next-Generation Digital Earth: A position paper from the Vespucci Initiative for the Advancement of Geographic
Information Science. International Journal of Spatial Data Infrastructures Research 3, 146–167.

McGavra, G., Morris, S., and Janée, G. (2009). Technology Watch Report: Preserving Geospatial Data. DPC Technology Watch
Series Report 09-01. Digital Preservation Coalition, York, UK.
 Late-breaking Posters 181
1+1=3: Combining cross-sectional multiple test data to find the historic
shape of an epidemic
Gustaf Rydevik1, 2, 3*, Mike Hutchings2, Giles Innocent1, Glenn Marion1, Piran White3
1. Biomathematics and Statistics Scotland
2. Scottish Agricultural College
3. University of York
*Contact author: [email protected]

Keywords: Disease surveillance, Empirical Likelihood, EM, Bayesian Statistics, Biostatistics

We have written a set of R functions to simulate, estimate and visualize the historic incidence of pathogen infections,
given a cross-sectional sample where two or more tests with different characteristics have been taken of each subject.
We use the example of antibody and nucleic acid tests to illustrate this scenario.

In the surveillance of infectious diseases, it is often the case that resources and practical issues limit data collection to
a prevalence study at a single point in time. Previous studies have used antibody titers to calculate incidence rates for
various diseases, including Salmonella [Simonsen et al.], HIV [Heisterkamp SH et al.] and Pertussis[Hallander et al.].
However, to our knowledge, no published work has looked at combining antibody data with, for example, nucleic acid
or culture data when calculating incidence rates.

The effect of the time interval between infection and test is different for different diagnostic tests. For example,
antibodies tend to remain for a long time after infection, while tests designed to demonstrate the presence of a
pathogen such as cultures or nucleic acid tests are much more sensitive to time elapsed since infection. By exploiting
this difference, we can get an idea of whether the incidence is stable, on the increase, or decreasing, using only a
single prevalence study.

We have developed a model for this situation which combines five different components: the incidence distribution
over time, the distribution of antibodies over time after infection, the distribution of nucleic acid (or some other
indicator) over time after infection, and the observed distributions of antibodies/nucleic acid in the cross-sectional
sample.

Using a combined Bayesian/empirical likelihood approach, building on functions from the mclust package, we then
use the observed distribution of antibodies and nucleic acid together with a prior for the infection incidence, to
generate a shifted, posterior estimate of how the historical incidence behaved.

Because the results of these analyses differ depending on the specific relationship between antibodies and nucleic acid
in the studied pathogen, simulation tools built on the simecol package are used for generating data to test how our
approach works across various different antibody, nucleic acid and incidence distributions.

Acknowledgement: This work is being conducted within the FP7 WildTech Project (http://www.wildtechproject.com).

References
Simonsen, J., Mølbak, K., Falkenhorst, G., Krogfelt, K. A., & Linneberg, A. (2009). Estimation of incidences of infectious
diseases based on antibody measurements. Statistics in medicine, Volume 28, Issue 14, pages 1882–1895,

Heisterkamp SH, de Vries R, Sprenger HG, Hubben GAA, Postma MJ(2008). Estimation and prediction of the hiv- aids-
epidemic under conditions of Haart using mixtures of incubation time distributions. Statistics in Medicine 2008;
27:781–794.

Hallander, H. O., Andersson, M., Gustafsson, L., Ljungman, M. and Netterlid, E. (2009), Seroprevalence of pertussis
antitoxin (anti-PT) in Sweden before and 10 years after the introduction of a universal childhood pertussis
vaccination program. APMIS, 117: 912–922
Late-breaking Posters 182

A Package for Temporal Disaggregation of Time Series

?
Christoph Sax1,2, , Peter Steiner1,3
1. State Secretariat for Economic Affairs Seco, Switzerland
2. University of Basel, Switzerland
3. University of Bern, Switzerland
?
Contact author: [email protected]

Keywords: Econometrics, Time Series, Temporal Disaggregation, Chow-Lin

Temporal disaggregation methods are used to disaggregate and interpolate a low frequency time series into
a higher frequency series, while either the sum or the average of the resulting high-frequency series is
consistent with the low frequency series. Disaggregation can be performed with or without the help of one
or more indicator series; most indicator-based methods are variants of the method proposed by Chow and
Lin (1971). Currently, there is no package in R to perform these tasks. tempdisagg offers a solution for the
general task of temporally disaggregating a time series in R.

Use cases

Researchers in need of high frequency data can generate these data by themselves. For example, estimating a
VAR (package vars) requires all variables to have the same frequency. tempdisagg allows the construction
of a missing series by the researcher. In many European countries, quarterly figures of Gross Domestic
Product (GDP) are estimated by statistical agencies using temporal disaggregation methods. Thanks to
tempdisagg, at the State Secretariat for Economic Affairs, we are able to estimate Switzerland’s official
quarterly GDP figures almost fully automated in R.

Implementation

tempdisagg implements the standard methods for temporal disaggregation: Denton, Chow-Lin, Fernandez
and Litterman. Denton, in its most basic form, does not require additional information besides the low-
frequency series, Chow-Lin, Fernandez and Litterman, on the other hand, are using one or more indicators.
tempdisagg is not restricted to annual or quarterly data, it works as long as the ratio of high to low frequency
is an integer number (e.g. biannual to monthly data).
The selection of a temporal disaggregation model is similar to the selection of a linear regression model.
Thus, td, the main function of tempdisagg, closely mirrors the working of the lm function (package stats),
including taking advantage of the formula interface. Unlike lm, td can handle ts and mts time-series
objects, as a typical use-case involves these objects. Nevertheless, it also can handle vectors and matrices in
the same way as lm. td is a general function for temporally disaggregating time series and invokes particular
methods (Denton, Chow-Lin, . . . ) based on the method argument. Internally, td uses the optimize
function (package stats) to solve the one-dimensional optimization problem, which is at the core of some
variants of the Chow-Lin procedure.

References
Chow, G. C. and A.-l. Lin (1971). Best linear unbiased interpolation, distribution, and extrapolation of time
series by related series. The Review of Economics and Statistics 53(4), 372–375.
Late-breaking Posters 183

cloudnumbers.com - Your Calculation Cloud for R

Markus Schmidberger1,*, Moritz von Petersdorff-Campen1, Markus Fensterer1, Erik

Muttersbach1

1. Cloudnumbers.com
*Contact author: [email protected]

Keywords: R, cloud computing, computer cluster, high performance computing

Cloudnumbers.com provides scientists, as well as companies, with the resources to perform high
performance calculations in the cloud. We help our customers to fight epidemics, develop highly
advanced drugs and manage financial risk. Our aim is to change the way research collaboration is done
today by bringing together scientists and businesses from all over the world on a single platform.
Cloudnumbers.com is a Berlin, Germany based international high-tech startup striving for enabling
everyone to benefit from the High Performance Computing (HPC) related advantages of the cloud. We
regard reducing the difficulty of accessing the pooled computational power and abolishing the need for
initial infrastructure investments as the key challenges in achieving our mission.

We pursue our goals by using the innovative potential of cloud computing. Leveraging the power of
several pooled computer capacities and providing these on-demand via the Internet allows for a more
efficient capacity utilization and highly elevated computational power. The cloud exhibits the
possibility to speed up CPU and memory intensive calculations performed with applications like the
well-known open-source statistics program R (www.r-project.org). Additionally, the cloud’s
characteristics predestine it for being the optimal mean for collaborative data sharing. Cooperation of
project members, academics or even business networks thus is facilitated in an unprecedented way with
regard to HPC calculations. User created knowledge databases as well as access to the most prominent
public data resources take the collaboration and cooperation potential to yet another level and scale. All
these features are integrated into the safe and efficient platform solution available on a pay-as-you-go
basis at cloudnumbers.com.

Currently we focus on R (CRAN and Bioconductor packages are available). Computer clusters running
with the packages multicore, snow, Rmpi or snowfall are available out of the box. At this point our
main goal is to get feedback and consumer insights. That is why we will provide you with some free
credit in order to allow you to test our service :

http://www.cloudnumbers.com

References

M. Schmidberger, M. Morgan, D. Eddelbuettel, H. Yu, L. Tierney, U. Mansmann; State of the Art in

Parallel Computing with R; Journal of Statistical Software 2009: Vol. 31, Issue 1, Page 1-27;
Late-breaking Posters 184

A tentative implementation of R for finding the non-dominated

fronts in multi-objective optimization

Ching-Shih Tsou1,* , Bo-Han Wu1, Ying-Hao Lee1, Ya-Chu Chiang1

1. Institute of Information and Decision Sciences, National Taipei College of Business, Taipei 10051, Taiwan, R.O.C.
*Contact author: [email protected]

Keywords: multi-objective optimization, non-dominated solutions

Most decisions involve more than one conflicting objectives. This leads to the concept of non-
dominance in multi-objective optimization that is different from the optimality concept of single
objective one. Non-dominance means that the improvement of some objective could only be achieved at
the expense of other objectives. Practically speaking, the effort on multi-objective optimization should
be made in finding the non-dominated solutions archived as the non-dominated set by considering all
objectives to be equally important. However, generating the non-dominated set could be
computationally expensive and is often inefficient. This paper intends to implement three algorithms,
the naive and slow, continuously updated, and an efficient method proposed by Kung et al., in R for
finding the non-dominated fronts. Comparisons are made in order to highlight the applicability of R for
building feasible multi-objective optimizers.

References

Deb, K. (2001). Multi-objective Optimization Using Evolutionary Algorithms. John Wiley &
Sons, pp.33-43.

Deb, K., Agrawal, S., Pratap, A., and Meyarivan, T. (2000). A fast elitist non-dominated
sorting genetic algorithm for multi-objective optimization: NSGA-II. In Proceeding of the 6th
International Conference on Parallel Problem Solving from Nature, Pages 849-858.

Kung, H., Luccio, F., and Preparata, F. (1975). On finding the maxima of a set of vectors.
Journal of the Association Computing Machinery 22(4), 469- 476.
Late-breaking Posters 185

A hybrid machine learning algorithm for unlabeled

gene expression data classification
Gokmen Zararsiz1,*, Ahmet Ozturk1, Erdem Karabulut2, Ferhan Elmalı1

1. Department of Biostatistics and Medical Informatics, Erciyes University, Kayseri, Turkey

2. Department of Biostatistics, Hacettepe University, Ankara, Turkey
*Contact author: [email protected]

Keywords: unlabeled data, clustering, classification, gene expression datasets

In gene expression data analysis, classification algorithms are widely used to classify biological
samples and to predict clinical or other outcomes. But, these algorithms can not be used directly, if the
data is unlabeled. Ignoring unlabeled data leads information loss and labeling them manually is a very
difficult and expensive process. There are semi-supervised algorithms which were produced to label
the unlabeled data. However, these algorithms are impractical in wholly unlabeled datasets. Thus, it is
very significant to generate the class label for this kind of wholly unlabeled datasets and clustering
algorithms can be used to obtain such labels. We proposed a hybrid machine learning algorithm for
gene expression datasets to determine the best clustering with an optimum cluster number, then
classify the dataset using new generated class labels. In this project, we will present the applicability
and effectiveness of this algorithm by using several R packages: clValid [Brock et al., 2011],
RankAggreg [Pihur et al., 2009], e1071 [Dimitriadou et al., 2011].

Acknowledgement
The project described is financially supported by Research Fund of the Erciyes University
(Project Number: TSY-11-3587).

References

D. Meyer (2001). Support Vector Machines The Interface to libsvm in package e1071. R News Volume
1/3, 23-26.

G. Brock, V. Pihur, S. Datta, and S. Datta (2008). clValid, an R package for cluster validation. Journal
of Statistical Software, 25(4).

J. Xie, C. Wang, Y. Zhang, S. Jiang (2009). Clustering Support Vector Machines for Unlabeled Data
Classification. 2009 International Conference on Test and Measurement (Hong Kong, China),
pp. 34-38.

V. Pihur, S. Datta, S. Datta (2009). RankAggreg, an R package for weighted rank aggregation. BMC
Bioinformatics, 10:62