R cheat sheet: C.

Hurst

R cheat sheet

Contents
1 Data I/O and manipulation 2
1.1 Reading in data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 (very) Basic manipulation of a data frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Creating factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Data summary 3
2.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Summarizing categorical variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.2 Desriptive statistics for a continuous variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Basic graphics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Classical bivariate tests 5

3.1 χ2 test of independance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Independant and paired t-tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.3 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4 Modelling 6
4.1 Linear regression and the general linear model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.2 Logistic regression and other GLMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3 Longitudinal and other correlated outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.1 Continuous longitudinal or otherwise correlated outcomes . . . . . . . . . . . . . . . . . . . . . 8
4.3.2 Categorical longitudinal or otherwise correlated outcomes . . . . . . . . . . . . . . . . . . . . . 8

Preamble
This document is to provide generic syntax for many of the more common R functions and analyses you are likely to
use. Any idenitfier (name) prefixed with ’my” is generic and is not to be taken literately (i.e. Your code shouldn’t
have these names in it), instead you should adapt the code for your own purposes. Specifically:
• my.y, my.x1 and my.x2 are three continuous variables with my.y assumed to be the outome (endpoint) variable
and my.x1 and my.x2 assumed to be predictors.

• my.a and my.b are two categorical variables that are initially just number coded (i.e. R doesn’t yet know they
are categorical)
• my.a.fac and my.b.fac are the ”Factors” correponding to my.a and my.b (see code below)
• my.c.fac.within is a within-subject effect (e.g. for longitudinal data)

• my.pat.id is a variable that idenitfies the patient (for longitudinal data)

• myfoo.csv is some comma delimited texte file (can be created by excel) that stores our data
• mydata.df is an R data frame that holds all of the above variables

1
R cheat sheet: C.Hurst

1 Data I/O and manipulation

1.1 Reading in data
There are MANY ways to input data in R. Personally, I like to keep it simple. I tend to just use comma delimited
text file, or ”csv” files that I have created in R.

READING IN DATA

##Read in data from a comma delimited file##

#1. Set working directory (note forwardslash)

setwd("C:/myrdata")

#2. Read data into mydata.df (dataframe)

mydata.df<-read.csv("somedatafile.csv")

Also, see the foreign library help to see how to read in files from other statistics packages like Stata or SPSS.

1.2 (very) Basic manipulation of a data frame

Now let’s do some very simple manipulation of the data. Note that R is VERY powerful when it comes to data
manipulation (subsetting etc). I will only touch on it VERY briefly here. Look up some of the free texts available for
R for MUCH more in this area.

BASIC SUBSETTING

#1a. New data frame with subset of variables (cols 1 to 3 only)

mynew.df<-mydata.df[,c(1:3)]

#1b. New df with subset of variables (excluding cols 1 to 3)

mynew.df<-mydata.df[,-c(1:3)]

#1c. Alternatively, use the variable names

mynew.df<-mydata.df[,c("age", "sex", "bmi")]

#2a. Now a subset of row (e.g. males only)

mynew.df<-mydata.df[mydata.df$sex==0,]

#2b. Or only patients less than 60 years old

mynew.df<-mydata.df[mydata.df$age<60,]

2
R cheat sheet: C.Hurst

1.3 Creating factors

Factors are useful because they are our way of telling R that we want a variable to be considered as categorical. If
a variable is a text variable (e.g. ”Male” and ”Female”), R automatically makes it a factor. BUT if a variable is
coded numerically (e.g. Sex = 0 and 1), we have to let R know that this is not a continuous variable We do this using
factors. This is also a useful way of providing labels for our variables. Now how do we create these factors.

CREATING FACTORS

#1. Create a factor for gender

# (currently coded as 0[males] and 1[females])
sex.fac<-factor(sex, labels=c("Males", "Females"))

#2. Now see if it works

table(sex.fac)

2 Data summary
2.1 Summary statistics
2.1.1 Summarizing categorical variables
To tabulate a single categorical variable (frequency table):

FREQUENCY TABLES

#1. Generate feq table for A factor

my.tab<-table(mydata.df$mya.fac)
my.tab

To cross tabulate two categorical variable

CROSS TABULATIONS

#1. Generate X-tab of A by B factor variables

my.tab<-table(mydata.df$mya.fac, mydata.df$myb.fac)
my.tab

3
R cheat sheet: C.Hurst

2.1.2 Desriptive statistics for a continuous variable

Many ways to do this, but I think it is easiest to use the R library psych (You may need to download this from CRAN
first).

#Load psych library into memory

library(psych)

#Generate descriptive stats for the continuous variable ’myx’

describe(mydata.df$myx)

#Generate descriptive stats for ’myx’ by groups (e.g.by gender)

describeBy(mydata.df$myx, group=mydata.df$mya.fac)

Note the describe function gives you most summary stats (mean, median, sd, IQR, range, min, max, n, nmiss etc)

2.2 Basic graphics

Now for continuous variable (univariate). To get a histogram and a boxplot for a single variable

UNIVARIATE GRAPHICS FOR A CONTINUOUS VARIABLE

#1. Histogram of continuous variable

hist(mydata.df$my.y)

#2. Boxplot of continuous variable

boxplot(mydata.df$my.y)

and now the bivariate relationship between a continuous outcome in terms of a categorical explanatory (factor)

BIVARIATE CONTINUOUS Y AND CATEGORICAL X

#1. Side-by-side boxplot

boxplot(my.y~my.a, data=mydata.df)

Now for two continuous variables.

BIVARIATE CONTINUOUS Y AND CONTINUOUS X

4
R cheat sheet: C.Hurst

#1. Scatter plot

plot(x=mydata.df$my.x, y=mydata.df$my.y)

3 Classical bivariate tests

Now for the classical bivariate tests. I will consider three:
1. χ2 test of independance

2. t-tests (both independant and paired)

3. Correlation (Pearson’s and Spearman’s)

3.1 χ2 test of independance

χ2 TEST OF INDEPENDANCE

#1. Generate X-tab of A by B factor variables

my.tab<-table(mydata.df$mya.fac, mydata.df$myb.fac)

#2. Input table into the chi.sq test

chisq.test(my.tab)

3.2 Independant and paired t-tests

T-TESTS

#1. Indpendant t-test

t.test(my.y ~ my.a, data = mydata.df)

#2. Paired t-test

t.test(y1, y2, paired=TRUE, data = mynewdata.df)
#Note data needs to be in wide format for this type of test

3.3 Correlation
CORRELATION

5
R cheat sheet: C.Hurst

#1. Pearson correlation

cor(mydata.df$my.x1, mydata.df$my.x2 , method ="pearson")
#Note Pearson’s is default so you could just write
cor(mydata.df$my.x1,mydata.df$my.x2)

#1b. Above only generates coeffeicent, need to tests using

cor.test(mydata.df$my.x1,mydata.df$my.x2)

2. Spearmans
cor(mydata.df$my.x1, mydata.df$my.x2 , method ="spearman")

4 Modelling
One of the things you will notice about R is that modelling is very simple. Once you understand the basics of mod-
eling in one situation (e.g. a continous outcome), extending to other types and outcomes and situations is very easy.
Modelling in R is based on the concept of a formula:

y ∼ x1 + x2
in a linear regression (assumeing y, x1 and x2 are continuous) implies:

y = β0 + β1 x 1 + β2 x 2

If this same formula is used (for example) in a Poisson regression (for count data) using a log link then

y ∼ x1 + x2
implies

y = eβ0 +β1 x1 +β2 x2

Regardless, the same basic form of the formula is used throughout all R modelling

4.1 Linear regression and the general linear model

Linear regression and the general linear model use exactly the same model (lm(), they only differ in that factors
(categorical predictors) will occur in the general linear model

LINEAR REGRESSION AND THE GENERAL LINEAR MODEL

#1. A multivariable linear regression

my.model<-lm(my.y~my.x1+my.x2, data=mydata.df)

#1b. Get summary (R-sq and coeffcients) and ANOVA table

summary(my.model)

6
R cheat sheet: C.Hurst

anova(my.model)

2. A general linear model (using a temporary factors)

my.model<-lm(my.y~my.x1+as.factor(my.a), data=mydata.df)

2b. ...or defining factor (permanantly) first

mya.fac<-factor(my.a, labels=c("low", "med", "high")
my.model<-lm(my.y~my.x1+mya.fac, data=mydata.df)

2c. Now get summary and ANOVA table

summary(my.model)
anova(my.model)

4.2 Logistic regression and other GLMs

The Generalized Linear Model (GLM) in R is a simple extension of the command used for the standard Linear Model
(Linear Regression, ANOVA and the General Linear Model). Note that the only REAL difference (as far as coding
goes), is the addition of the family argument. Here we will use Binary Logistic regression as an example, but we can
also fit any of the other GLMs (e.g. Possion regression) using the same command (just change the ’family’ argument).
Assume that the outcome variable, my.a is a binary outcome variable:

#1. A binary logistic regression

my.model<-lm(my.a~my.x1+my.x2, data=mydata.df, family=binomial)

#1b. Get summary (R-sq and coeffcients) and ANOVA table

summary(my.model)
anova(my.model)

#Note I wrote the below function myself-get the R code from me

print.ORCIs.glmm.wald(my.model)

4.3 Longitudinal and other correlated outcomes

Longitudinal data (and correlated data) is where we have several observations associated with each observational unit.
For example:
• Longitudianl data: We have several observations (over time) for each patient
• For clustered data (e.g. a multicentre study), we might have many patients associated with each clinic or hospital
(and to not account for this in our analysis would render our results invalid).

Here, I will just give an example of a longitudunal analysis (using the patient identifier my.pat.id), but the
approach used for clustered data (e.g. Hospital ID) is very similar.

7
R cheat sheet: C.Hurst

4.3.1 Continuous longitudinal or otherwise correlated outcomes

For a continuous outcome variable, we would use a Linear Mixed Model (LMM) to analysis our data. Typically we
will have a ’within-subject effect’ (e.g. Time), and one or more ’between level effect(s)’ (e.g. Treatment, Gender).
Between-subject effects are the ones we are used to.

Note that there are several R libraries for LMMs. My prefered library is called lme4. Again, the first time you
use this, you may have to download from CRAN.

library(lme4)

#1. A linear mixed model (Random intercept model)

# Note: As long as you specify the patient ID, R can workout
# the between- and within-subject effects itself
my.model<-lmer(my.y~my.x1+my.c.fac.within + (1|my.pat.id), data=mydata.df)

#1b. Summaries model and coeffcients

summary(my.model)
anova(my.model)
#Note I wrote the below function myself-get the R code from me
print.BetaCI.lmm(my.model)

#2. A linear mixed model (Random coeffcients model)

my.model<-lmer(my.y~my.x1+my.c.fac.within + (my.c.fac.within|my.pat.id), data=mydata.df)

#2b. Summaries model and coeffcients

summary(my.model)
anova(my.model)
print.BetaCI.lmm(my.model)

4.3.2 Categorical longitudinal or otherwise correlated outcomes

Like a standard linear regession (Genral Linear Model) can be extended to a Generalized Linear model, a LMM can
also be extended to a Generalized Linear Mixed Model. In terms of R, the only real difference is the addition of the
family argument.

library(lme4)

#1. A Generalized linear mixed model (Random intercept model)

# Note: This is a binary logistic mixed effect regression
# Assume that my.a is a binary avriable
my.model<-glmer(my.a~my.x1+my.c.fac.within+(1|my.pat.id), data=mydata.df, family=binomial)

#1b. Summaries model and coeffcients

8
R cheat sheet: C.Hurst

summary(my.model)
anova(my.model)
#Note I wrote the below function myself-get the R code from me
print.ORCIs.glmm.wald(my.model)

#Note: As with the LMM we can also fit a Random coeffcients

#version of the GLMM