07-12 FR PDF

ESTIMATING COST PER
LANE MILE FOR ROUTINE

HIGHWAY OPERATIONS
AND MAINTENANCE
Project 07-12
January 2011
Midwest Regional University Transportation Center

College of Engineering
Department of Civil and Environmental Engineering
University of Wisconsin, Madison
Authors: Teresa M. Adams, Ph.D.

University of Wisconsin-Madison, Madison, Wisconsin
Principal Investigator: Teresa M. Adams, Ph.D.

Professor, Department of Civil and Environmental Engineering, University of Wisconsin-Madison
DISCLAIMER
This research was funded by the Midwest Regional University Transportation Center and the
Wisconsin Department of Transportation. The contents of this report reflect the views of
the authors, who are responsible for the facts and the accuracy of the information presented
herein. This document is disseminated under the sponsorship of the Department of
Transportation, University Transportation Centers Program, in the interest of information
exchange. The U.S. Government assumes no liability for the contents or use thereof. The
contents do not necessarily reflect the official views of the Midwest Regional University
Transportation Center, the University of Wisconsin, the Wisconsin Department of
Transportation, or the USDOT’s RITA at the time of publication.
The United States Government assumes no liability for its contents or use thereof. This
report does not constitute a standard, specification, or regulation.
The United States Government does not endorse products or manufacturers. Trade and
manufacturers names appear in this report only because they are considered essential to the
object of the document.
ii
TECHNICAL REPORT DOCUMENTATION PAGE
1. Report No. 2. Government Accession No. 3. Recipient’s Catalog No.
MRUTC 07-12 CFDA 20.701
4. Title and Subtitle 5. Report Date January 2011

Estimating Cost Per Lane Mile for Routine Highway Operations and Maintenance
6. Performing Organization Code
7. Author/s 8. Performing Organization Report No.

Teresa M. Adams, Ph.D. MRUTC 07-12
9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)
Midwest Regional University Transportation Center
University of Wisconsin-Madison
1415 Engineering Drive, Madison, WI 53706 11. Contract or Grant No.
0092-06-12
12. Sponsoring Organization Name and Address 13. Type of Report and Period Covered
Wisconsin Department of Transportation Final Report [12/14/06 –
Hill Farms State Transportation Building 3/31/2010]
4802 Sheboygan Avenue
Madison, WI 53707 14. Sponsoring Agency Code
15. Supplementary Notes

Project completed for the Midwest Regional University Transportation Center with support from the Wisconsin Department of Transportation.
16. Abstract
The disparity between maintenance budgets and maintenance requirements causes agencies to make difficult choices
about maintenance priorities. There is a growing need to effectively link maintenance costs and condition to provide clear
evidence of impacts due to budget tradeoffs decisions. The focus of this research was to develop mathematical relationships
between expenditures for highway maintenance and the resulting maintenance condition by analyzing historic cost and
condition data. The research involved analyzing maintenance condition and cost data over three years, 2004, 2005, and 2006
from the state transportation agencies in Michigan, Ohio, and Wisconsin for highway components including pavement,
shoulders, roadside vegetation, drainage, signs, and pavement markings. The maintenance management systems at the states
use different categorization schemes for their maintenance activity costs and different rating systems for maintenance
condition, thus direct comparison between states was not possible.
The primary result of this research is a set of probabilistic distribution functions for annual maintenance costs for a
wide range of maintenance activities. Confidence intervals can be constructed around the average using the chosen level of
confidence (i.e., 95%). The functions are useful for sensitivity and simulation analyses.
The researchers hypothesized that data would reveal relationships between cost and condition. A regression tree
analysis approached was used to search for relevant model equations. However, the statistical analysis of the data revealed
weak evidence of these relationships. This finding is common for all three of the states that were investigated. There are at
least two clear limitations of the data. First, to see trends overtime, three years of data may not be enough. Furthermore,
even with budget cuts, noticeable deterioration in condition, deficiencies, or maintenance backlog may take longer than three
years. Second, the available cost and condition data are aggregated over many highway miles. Maintenance management and
cost records generally do not include precise highway locations where maintenance was performance, the specific activities
that were performed, nor the precise cost and timing of those activities. Consequently the trends and relationships between
cost and condition are “washed out” by the lack of precision.
17. Key Words 18. Distribution Statement
No restrictions. This report is available through the Transportation Research
Maintenance, budgets, pavement condition, Information Services of the National Transportation Library.
Michigan, Ohio, Wisconsin.
19. Security Classification (of this report) 20. Security Classification (of this page) 21. No. Of Pages 22. Price
Unclassified Unclassified 67 -0-
Form DOT F 1700.7 (8-72) Reproduction of form and completed page is authorized.
iii
Table of Contents
DISCLAIMER ..................................................................................................................................... ii
TECHNICAL REPORT DOCUMENTATION PAGE ............................................................ iii
LIST OF TABLES ............................................................................................................................. vi
LIST OF FIGURES .......................................................................................................................... vii
EXECUTIVE SUMMARY ................................................................................................................ 1
Chapter 1. INTRODUCTION .................................................................................................... 3
1.1 Problem Statement ............................................................................................................ 3
1.2 Objectives............................................................................................................................ 4
1.3 Scope.................................................................................................................................... 4
1.4 Methodology ....................................................................................................................... 5
1.5 Organization of Report ..................................................................................................... 7
Chapter 2. MAINTENANCE ACTIVITY COSTS ................................................................. 9
2.1 Adjustment for Inflation ................................................................................................... 9
2.2 Adjustment for Location .................................................................................................. 9
2.3 Cost Transformation Functions .................................................................................... 10
2.3.1 Box-Cox Transformation ........................................................................................... 10
2.3.2 Johnson Transformation ............................................................................................ 12
2.4 Maintenance Activity Costs per Lane Mile (CPLM) ................................................... 13
2.4.1 Highway Maintenance in Michigan........................................................................... 14
2.4.2 Highway Maintenance in Ohio.................................................................................. 14
2.4.3 Highway Maintenance in Wisconsin......................................................................... 15
2.5 Distribution Functions for Maintenance Expenditures ............................................. 15
Chapter 3. CATEGORICAL MAINTENANCE COSTS .................................................... 25
3.1 Michigan’s Maintenance Costs per Lane Mile ............................................................. 25
3.1.1 Maintenance Activity Categories and Lane Mile (LM) Groups ............................ 25
3.1.2 Cost Trends for LM Groups ..................................................................................... 28
3.1.3 Annual Cost Trends during the Study Period ......................................................... 30
3.1.4 Cost Distribution Functions for Maintenance Categories .................................... 32
3.2 Ohio’s Maintenance Costs per Lane Mile .................................................................... 34
3.3 Wisconsin’s Maintenance Costs per Lane Mile ........................................................... 37
Chapter 4. MAINTENANCE CONDITION ASSESSMENT ........................................... 43
4.1 Michigan’s Maintenance Quality Assurance Program ................................................ 43
4.2 Ohio’s Maintenance Quality Survey (MQS) Program ................................................ 44
iv
4.3 Wisconsin Maintenance Quality Assurance Program ................................................. 45
Chapter 5. MODELS RELATING COST TO CONDITION ........................................... 51
5.1 Analysis Methodology ..................................................................................................... 51
5.2 Modeling Cost to Condition For Michigan ................................................................. 51
5.3 Modeling Cost to Condition For Ohio......................................................................... 52
5.4 Modeling Cost to Condition For Wisconsin ............................................................... 54
5.4.1 Aggregating Element Condition and Maintenance Costs ..................................... 55
5.4.2 Regression Tree Modeling Approach ....................................................................... 58
5.4.3 Regression Tree Models ............................................................................................. 61
5.4.4 Discussion .................................................................................................................... 63
Chapter 6. Conclusions ................................................................................................................ 65
REFFERENCES ............................................................................................................................... 67
v
LIST OF TABLES
Table 1. MQA Questions and Statistical Approaches (Schmitt et al. 2006) .................................. 5
Table 2. Urban Consumer Price Index ............................................................................................... 9
Table 3. Estimated hourly labor rates for highway maintenance workers in Michigan ............... 10
Table 4. Activities and costs per lane mile for contracts with Michigan’s cities ........................... 16
Table 5. Activities and costs per lane mile for contracts with Michigan’s counties ..................... 17
Table 6. Activities and costs per lane mile for Michigan’s direct maintenance forces................. 18
Table 7. Activities and costs per lane mile for counties in Ohio .................................................... 20
Table 8. Activities and costs per lane mile for contracts with Wisconsin’s counties ................... 23
Table 9. Maintenance categories for analysis of Michigan’s maintenance costs ........................... 26
Table 10. Lane mile (LM) groups for Michigan’s highway maintenance providers ..................... 26
Table 11. Analysis of the mean CPLM for LM groups – Michigan (95% CI) .............................. 29
Table 12. Analysis of the mean annual CPLM – Michigan (95% CI) ............................................ 31
Table 13. Categorical CPLM – contracts with Michigan’s cities ..................................................... 32
Table 14. Categorical CPLM – Michigan’s direct forces at state maintenance garages ............... 32
Table 15. Categorical CPLM – contracts with Michigan’s counties ............................................... 33
Table 16. Maintenance categories for analysis of Ohio’s maintenance costs ................................ 35
Table 17. Analysis of the mean CPLM for LM groups and fiscal years – Ohio (95% CI) ......... 36
Table 18. Categorical CPLM and distribution functions – Ohio .................................................... 37
Table 19. Categories for analysis of Wisconsin’s maintenance costs ............................................. 38
Table 20. Analysis of the mean CPLM for LM groups and years – Wisconsin (95% CI). ......... 40
Table 21. Categorical CPLM and distribution functions – contracts with Wisconsin’s
counties .................................................................................................................................................... 41
Table 22. Michigan’s sufficiency rating system for pavement surface and joint condition ......... 43
Table 23. Michigan’s sufficiency rating system for drainage condition ......................................... 43
Table 24. Michigan’ sufficiency rating system for shoulder condition ........................................... 44
Table 25. Mapping cost to condition categories for Michigan. ....................................................... 44
Table 26. Ohio’s maintenance quality survey (MQS) criteria. ......................................................... 47
Table 27. WisDOT compass elements, features, and thresholds (Compass 2005). ..................... 48
Table 28. HMS cost activity codes mapped to Compass elements/features. ............................... 50
Table 29. Model parameters for guide analysis of cost and condition in Ohio. ........................... 52
Table 30. Models that relate maintenance cost to condition in Ohio. ........................................... 52
Table 31. Deficiencies per lane mile ranges for Ohio’s Maintenance OPI index ......................... 54
Table 32. Correlation among features shoulder elements ................................................................ 55
Table 33. Correlation among features of mowing, litter pickup and woody vegetation ............. 55
Table 34. Correlation among features of asphalt travel ways .......................................................... 56
Table 35. Correlation among features of concrete travel ways ....................................................... 56
Table 36. Model parameters for guide analysis of cost and condition in Wisconsin. .................. 58
Table 37. Best Subsets Regression for Litter Pickup Cost ............................................................... 62
vi
LIST OF FIGURES
Figure 1. Trend for Miles Traveled and Maintenance Expenditures in Wisconsin ...................... 3
Figure 2. Methodology .......................................................................................................................... 6
Figure 3. Histogram and probability plot for non-normal cost data. ............................................. 11
Figure 4. Transformed costs using Box Cox transformation function. ......................................... 12
Figure 5. Histogram of non-normalized cost per lane mile. ............................................................ 13
Figure 6. Histogram of transformed and normalized cost per lane mile ....................................... 13
Figure 7. Regions and counties of the Michigan DOT including maintenance responsibilities
and garage locations ............................................................................................................................... 14
Figure 8. Transportation districts and counties in Ohio .................................................................. 14
Figure 9. Transportation regions and counties in Wisconsin .......................................................... 15
Figure 10. Lane mile (LM) groups for Michigan’s highway maintenance providers.................... 27
Figure 11. CPLM trends for LM groups – contracts with Michigan’s cities. ................................ 28
Figure 12. CPLM trends for LM groups – contracts with Michigan’s counties ........................... 29
Figure 13. CPLM trends for LM groups – direct forces at Michigan’s maintenance garages. ... 30
Figure 14. Annual CPLM trends - contracts with Michigan’s counties ......................................... 31
Figure 15. Annual CPLM trends - contracts with Michigan’s cities ............................................... 31
Figure 16. Annual CPLM trends - direct forces at Michigan’s maintenance garages .................. 31
Figure 17. County lane mile (LM) groups - Ohio.............................................................................. 34
Figure 18. Cost per lane mile trends for LM groups – Ohio’s Counties ....................................... 34
Figure 19. Annual trends for cost per lane mile – Ohio’s Counties ............................................... 36
Figure 20. Lane mile (LM) groups – Wisconsin’s counties .............................................................. 38
Figure 21. Wisconsin’s cost per lane mile for maintenance categories and lane mile groups ..... 39
Figure 22. Annual trends for cost per lane mile for contracts with Wisconsin’s counties. ......... 40
Figure 23. Observed deficiencies versus fitted models for Ohio .................................................... 53
Figure 24. %Backlog for asphalt pavements in Wisconsin .............................................................. 57
Figure 25. Latitude groups .................................................................................................................... 59
Figure 26. Regression tree for maintenance backlog of concrete pavements ............................... 61
Figure 27. Observed versus fitted backlog for concrete traveled ways .......................................... 61
Figure 28. Regression tree for mowing maintenance expenditures ................................................ 62
Figure 29. Regression tree for managing woody vegetation ............................................................ 63
vii
EXECUTIVE SUMMARY
When transportation agencies prepare a design for new highway construction or major
improvements to existing highways, they consider the agency’s life-cycle and user costs in the
project design decisions. However, after highway projects are completed, maintenance budgets
rarely get adjusted to accommodate the maintenance of new lane miles. This problem is
worsened by the fact that the number of vehicle miles traveled (VMT) per highway mile tends to
increase over time. Maintenance budgets remain relatively constant, regardless of an increase in
the number of vehicle miles traveled.
The disparity (real or perceived) between maintenance budgets and maintenance requirements
causes agencies to make difficult choices about maintenance priorities. When this happens,
maintenance concerns related to safety and operational efficiency tend to override preservation
of capital investment. There is a growing need to effectively link both maintenance costs and
condition to provide clear evidence of impacts due to budget tradeoffs decisions. In addition,
estimates for maintenance costs are useful for establishing and evaluating bid prices in
maintenance contract agreements.
The focus of this research was to develop mathematical relationships between expenditures for
highway maintenance and the resulting maintenance condition by analyzing historic cost and
condition data. The research involved analyzing data from the state transportation agencies in
Michigan, Ohio, and Wisconsin. The scope includes maintenance of all highway components
within an agency’s right-of-way including pavement, shoulders, roadside vegetation, drainage,
signs, and pavement markings. Winter operations, bridges, and roadside facilities such as rest
areas and weigh stations were not included in the project scope. The analysis looks at
maintenance cost and condition in 83 Michigan counties, 88 Ohio counties, and 72 Wisconsin
counties over three years, 2004, 2005, and 2006.
The project did not involve direct comparisons between the states. The maintenance
management systems at the states use different categorization schemes for their maintenance
activity costs and different rating systems for maintenance condition. Similarly, the maintenance
costs available from the states include different cost items. Maintenance costs in Michigan
include labor and material but not overhead. The available maintenance costs for Ohio include
labor, materials, and overhead. Similarly, maintenance costs for Wisconsin are total expenditures
including labor, materials, equipment and overhead. Consequently, direct comparisons between
the states were not possible. For Michigan, the maintenance components are divided into 5
categories: surface maintenance, shoulder maintenance, roadside maintenance, general
maintenance, and traffic services. For Ohio, the maintenance components are organized into 8
categories including: guardrail, pavement, pavement drop-off, vegetation obstruction, litter,
drainage ditch obstruction, sign deterioration, and pavement markings. For Wisconsin, the
maintenance components are organized into nine groups: asphalt traveled way, concrete traveled
way, unpaved shoulders, paved shoulders, mowing, litter pickup, woody vegetation, noxious
weeds, and ditches. Maintenance condition in Michigan is expressed as an averaged rating per
county on a scale of 0-5. In Ohio, maintenance condition is expresses as absolute total deficiency
for each maintenance item, In Wisconsin, maintenance conditions is expressed as percentage of
features that are deficient.
The researchers hypothesized that data would reveal various relationships between cost and
condition. For example, that maintenance condition deteriorates as maintenance expenditures
1
decrease or that various county-level characteristics such as size, responsible lane miles, and
weather would have some impact on the relationship between maintenance cost and
maintenance condition. However, the statistical analysis of the data revealed weak evidence of
these intuitive relationships.
The researchers used regression analysis to find relevant model equations. Various tools were
used to analyze the data. The primary analysis tool is the regression tree modeling algorithm,
GUIDE (Generalized, Unbiased, Interaction, Detection and Estimation). A regression tree is a
piecewise constant or piecewise linear estimate of a regression function, constructed by
recursively partitioning the data and sample space. The regression tree approach provides a way
to find county groups and the corresponding model for each group. Within each tree, models
capture differences among the counties depending upon county size or lane miles, vehicle miles
traveled, population density, and other relevant variables such as soils and precipitation. The
algorithm automatically searches over numerous multiple linear models to fit different subsets of
the data and uses cross-validation to pick groupings and corresponding models that minimize
the least squares of error on the prediction. The result is sets of models that may be applicable to
counties in other states with similar characteristics. Other analysis tools include scatter plots and
MINITAB regression analysis.
The overall finding from this research is that the available data is not useful for optimizing
maintenance budget management and allocation. This finding is common for all three of the
states that were investigated. A wise person once said, “If we torture data long enough, it will
tell us whatever we want.” This project may be the exception.
There are at least two clear limitations of the data. First, to see trends overtime, three years of
data may not be enough. With biennial state budgets, three years is only one and one-half
budget cycles. The analysis showed little change over the three year period. Furthermore, even
with budget cuts, noticeable deterioration in condition, deficiencies, or maintenance backlog may
take longer than three years. Second, trends and relationships between cost and condition are
washed out by the lack of precision and accuracy of the maintenance activities and associated
costs. The available cost and condition data are aggregated and generalized over large areas for
many highway miles. Maintenance management and cost records generally do not include the
precise highway location where maintenance was performance, the specific activities that were
performed, the precise cost and timing of those activities. To overcome the problem,
maintenance activities and costs must be more clearly tied to the maintenance elements that are
assessed for maintenance condition.
Among the salient results of this research are a set of probabilistic distribution functions for
annual maintenance costs for a wide range of maintenance activities. Confidence intervals can be
constructed around the average using the chosen level of confidence (i.e., 95%). The functions
are useful for sensitivity and simulation analyses.
2
Chapter 1. INTRODUCTION
1.1 Problem Statement

When transportation agencies prepare designs for new highway construction or major
improvements to existing highways, the agency and user life-cycle costs are considered in project
design decisions. However, after highway projects are completed, maintenance budgets are rarely
adjusted to accommodate the routine maintenance of new lane miles.
This problem is worsened by the fact that each year the number of vehicle miles traveled (VMT)
per highway mile has a tendency to increase over time, while on the other hand, maintenance
budgets do not seem to be adjusted to accommodate these increases. Figure 1 shows the trends
of million VMT and expenditures/million VMT in Wisconsin from 2000 to 2006. It shows that
between the years 2001 and 2004 while VMT was increasing from 57 million to 60 million, the
money spent for highway maintenance per million VMT was decreasing from 3.5 to 2.5 million
dollars.
61 $4.00
$3.50
60
$3.00
59
$2.50
Million VMT
58 $2.00
Million $
$1.50
57
$1.00
56 Million VMT
$0.50
Expenditures/million VMT
55 $0.00
2000 2001 2002 2003 2004 2005 2006
Year
Figure 1. Trend for Miles Traveled and Maintenance Expenditures in Wisconsin

The disparity between maintenance budgets and maintenance requirements causes agencies to
make difficult choices about maintenance priorities. When this happens, maintenance activities
will usually not get the priority because in addition to concerns about preserving capital
investment, highway operations and maintenance bureaus have concerns about safety and loss of
operational efficiency due to deteriorating condition of roadways. There is a growing need to
efficiently link both maintenance costs and condition to allow decision makers to better address
concerns about preserving capital investment and maintaining operational efficiency (1).
3
1.2 Objectives
The focus of this research is to develop a way to estimate cost of ongoing routine operations
and the maintenance cost components of total highway life-cycle cost using historical data. The
research involves analysis of data from three state transportation agencies: MDOT, ODOT and
WisDOT. There are two main tasks:
 Perform a correlation study to characterize the relationship between maintenance
cost and maintenance condition.
 Develop model equations for estimating the recurring annual cost for routine,
county-level highway maintenance.
It is hypothesized that maintenance condition will deteriorate as maintenance expenditures
decrease. It is also expected that different sized counties based on parameters such as lane miles
will perform differently with regard to the relationship between maintenance cost and
maintenance condition. The objective of this project is to determine if this is indeed the case for
both Ohio and Michigan over the 3 year period of 2004-2006. Addition objectives include
developing equations that will give transportation decision makers the tools required to make
informed decisions regarding transportation maintenance budget. Two types of equations will be
modeled, one predicts condition given cost and the other predicts cost given condition.
Equations that predict condition can be used to determine the effects of increased or decreased
funding on condition. Equations that predict cost can be used to determine the cost to achieve
certain levels of condition.
1.3 Scope
The scope of the project includes maintenance of all highway components within an agency’s
right-of-way including pavement, shoulders, roadside vegetation, drainage, signs, and pavement
markings. Winter operations, bridges, and roadside facilities such as rest areas and weigh stations
are not included in the project scope.
The analysis uses data from the state transportation agencies in Michigan, Ohio, and Wisconsin.
The maintenance and cost data cover fiscal years 2004, 2005, and 2006. The geographic extent
of is the county level in each state. The analysis looks at maintenance cost and condition in 83
Michigan counties, 88 Ohio counties, and 72 Wisconsin counties over a three year period. The
maintenance management system at each state influences the level of detail for the available data.
For Michigan, the maintenance components are divided into 5 categories: surface maintenance,
shoulder maintenance, roadside maintenance, general maintenance, and traffic services. For
Ohio, the maintenance components are organized into 8 categories including: guardrail,
pavement, pavement drop-off, vegetation obstruction, litter, drainage ditch obstruction, sign
deterioration, and pavement markings. For Wisconsin, the maintenance components are
organized into nine groups: asphalt traveled way, concrete traveled way, unpaved shoulders,
paved shoulders, mowing, litter pickup, woody vegetation, noxious weeds, and ditches.
It is difficult to directly compare cost and condition among the states. Maintenance conditions
are expressed using different measurement scales and some specific maintenance items are not
directly comparable. Maintenance condition in Michigan is expressed as an averaged rating per
county on a scale of 0-5. In Ohio, maintenance condition is expresses as absolute total deficiency
for each maintenance item, In Wisconsin, maintenance conditions is expressed as percentage of
4
features that are deficient. Similarly, the maintenance costs available from the states include
different cost items. Maintenance costs in Michigan include labor and material but not overhead.
The available maintenance costs for Ohio include labor, materials, and overhead. Similarly,
maintenance costs for Wisconsin are total expenditures including labor, materials, equipment
and overhead.
A set of independent variables were evaluated to model the relationships between maintenance
cost and condition among the counties in each state. The variables are limited to what is
available in each agency’s inventory and condition inspection data sets. In addition, the research
team added environmental, operational, and socio-economic variables.
1.4 Methodology
The research involved statistical analysis of cost and condition data with the goal of finding
relationships and trends that reveal insights for improving the effectiveness of maintenance
expenditures. Much work has been done on developing statistical guideline for designing a
Maintenance Quality Assurance (MQA) data collection and analysis plan. One example is shown
in Table 1. Much less work has been done on relating MQA cost to condition and there are no
guidelines upon which the methodology could be modeled.
Table 1. MQA Questions and Statistical Approaches (Schmitt et al. 2006)
Question Statistical Approach
Determining Statistical distributions and statistical parameters can assist in determining sample
number of size. LOS confidence interval estimation equations can provide an interrelationship
samples to yield of sample size, variability, confidence level, precision, and data distribution
valid information parameters.
Developing Confidence intervals can be constructed around the average using the chosen level
confidence in an of confidence (i.e., 95%), underlying variability, and sample size.
estimate
Stratifying data in Analysis of variance can detect whether there is a difference between features of
terms of data (region, crews, etc.), while incorporating the variability into the determination.
geographical or Some states may have significant differences between regions that can be detected
highway features using Analysis of Variance.
Comparing results T-tests can detect the mean difference between data sets generated by two different
of MQA data MQA data collectors. Then, with the assistance of power curves, the “true” mean
collectors difference between them can be measured and controlled. A whole host of standard
statistical tests are available with which to compare data sets obtained by different
MQA technicians or teams.
Are years F-tests and t-tests can provide a statistical comparison of means. Paired-sample t-
different or not tests are used when data are collected from the same roadway segment from year-to-
year, while a two-sample t-test is used when the roadway segments are independent
of each other.
Looking for Outliers, or data points that are abnormal from a distribution, can detect trouble
trouble signs signs. Several standard tests for outliers exist, and the chi-square or other goodness-
of-fit tests can be used to check normality.
Reporting data Beginning with simple fundamental statistical measures is always the best start (plot
the data, calculate the average and standard deviation, etc.). The sampling design
largely drives if/how a statistically valid analysis can proceed, so effort must be
placed on sampling design at the beginning.
5
The research process for data analysis is shown in Figure 2. The first step was to collect cost and
condition data from the state Departments of Transportation. The research team collected
various county-level environmental and socio-economic parameters as well.
Data Collection Michigan, Ohio, and

Wisconsin DOTs
Prepare and Format Cost and Environmental,

Condition Data operational, and socio-
economic parameters
Synthesize and Characterize Data
Map Cost to Condition Figure 2. Methodology
Analyze Correlation of all Available Data
Tools: Scatter Plots,

Regression Analysis MINITAB and GUIDE
Next, the team prepared and formatted the available data for characterization and analysis. The
synthesis and characterization step included determining sample sizes, averages, standard
deviations and ranges. In addition, we characterize the cost and condition data independently for
the 3-year period of 2004-06. This includes scatter plots of cost versus total lane mile and
condition versus total lane mile for each maintenance category. Scatter plots also serve to
visualize outliers or any abnormal data within the 3-year period.
The forth step is to map cost to condition using available parameters followed by analysis of
correlation among the parameters to identify the parameters that most influence cost and
condition.
The final step in the analysis methodology is the regression analysis to find relevant model
equations. Various tools were used to analyze the data. The primary analysis tool is the
regression tree modeling algorithm, GUIDE (Generalized, Unbiased, Interaction, Detection and
Estimation). Other analysis tools include scatter plots and MINITAB regression analysis.
A regression tree is a piecewise constant or piecewise linear estimate of a regression function,
constructed by recursively partitioning the data and sample space (6). The regression tree
approach provides a way to find county groups and the corresponding model for each group.
Within each tree, models capture differences among the counties depending upon county size or
lane miles, vehicle miles traveled, population density, and other relevant variables. The algorithm
automatically searches over numerous multiple linear models to fit different subsets of the data
and uses cross-validation to pick groupings and corresponding models that minimize the least
squares of error on the prediction. The result is models with improved prediction accuracy and
the effect of the outliers alleviated. Additionally, the models are much simpler and easier to use
than transformed (non-linear) models based on the scatter plots (7). The result is sets of models
6
that may be applicable to counties in other states with similar characteristics.
1.5 Organization of Report

This report is organized into five chapters. Chapter 2 summarizes our findings on maintenance
activity costs per lane mile in each of the three states. The chapter includes distribution functions
and statistical parameters for cost at the activity level.
In Chapter 3, maintenance activities are rolled up to the category level. Typical categories are
pavements, roadsides, shoulders, drainage, signs, etc. The state use similar categories but not
there are slight differences that make comparisons unlikely. Trends for maintenance costs at the
category level were explored by partitioning the data set by year and by responsible lane miles
for of the maintenance jurisdiction. Finally the chapter includes distribution functions and
statistical parameters for cost at the category level.
Chapter 4 contains a brief introduction to the maintenance condition assessment program in
each state. The agencies organized the condition evaluation data into maintenance categories.
The categories are the same as those used in Chapter 3.
Chapter 5 presents the analysis for relating maintenance cost to condition at the category level. .
The chapter presents the procedure and results of the GUIDE regression tree and MINITAB
regression analyses for determining analytical models that relate condition and cost.
Finally, Chapter 6 presents conclusions.
7
Chapter 2. MAINTENANCE ACTIVITY COSTS
This chapter looks at distributions of individual maintenance activities cost using three years of
data from the state departments of transportation at Michigan, Ohio and Wisconsin. This study
examined the data from 2004, 2005 and 2006. The costs were adjusted for inflation and location
to make the results useful for predicting the maintenance costs in future years and other
locations.
The detail and scope of the cost data varies from state to state. For Ohio and Wisconsin, the
analysis of costs focuses at the county-level. For Michigan, cost data for maintenance performed
by contracts with counties and cities was available by route segments. Thus segment-level data
was used to prepare distribution functions. Costs were at the county-level for work performed
by direct forces at Michigan’s state garages.
Costs for each state included labor, equipment and materials and the labor costs included
fringes. The analyses for Michigan and Wisconsin used direct costs for maintenance activities –
administrative costs were ignored. For Ohio, the analysis used total costs including direct and
indirect.
2.1 Adjustment for Inflation

Cost data was adjusted for inflation using the urban
consumer price index (CPI) to base year 2006 Table 2. Urban Consumer Price Index
costs. The urban CPI allows data from the years Year Urban CPI
2004 and 2005 to be compared with year 2006 data. 2004 186.1
The urban CPI for the years 2004-2006 are listed in 2005 191.6
Table 2 and the formula for adjusting cost is show 2006 195.9
below.
Urban CPI i 
Costi = Costj 
 Urban CPI j 
 
2.2 Adjustment for Location

Labor costs within each state were adjusted to accommodate the variations in labor rates across
the state. In this section the researches present the development of the location indices for
Michigan. Hourly labor rates across the state of Michigan were adjusted to the Detroit-Warren-
Livonia area as the base location. For Ohio, labor costs were adjusted to be equivalent with the
Cleveland area. The researchers adjusted labor rates for Michigan and Ohio based on the hourly
labor rates published by the Bureau of Labor Statistics – Occupational Employment Statistics
website (http://www.bls.gov/oes/2007/ may/oes_2600004.htm). Individual counties were identified
for each area using the Bureau of Labor Statistics metropolitan and nonmetropolitan area
definitions (http://www.bls.gov/oes/2007/may/msa_def.htm#12980).
The Bureau of Labor Statistics website provides the highway maintenance labor rate for many
areas but not all. The website did not defined hourly rates for ten counties in Michigan (Bay,
Berrien, Calhoun, Clinton, Eaton, Ingham, Jackson, Monroe, Muskegon, and Ottawa). For these
counties, the rate was obtained by comparing common rates (e.g. construction laborer hourly
rates) between the county and its neighbors. The highway maintenance hourly rate was then
9
derived according to the relationship (increase or decrease) between the common rates from the
neighboring counties. The results are shown in Table 3. The values in the index column were
used to convert labor costs to equivalent Detroit labor costs.
Table 3. Estimated hourly labor rates for highway maintenance workers in Michigan
Area Hourly Rate Index
Ann Arbor, MI $18.81 96.41
Detroit-Warren-Livonia, MI $19.51 100.00
Flint, MI $17.65 90.47
Grand Rapids-Wyoming, MI $19.30 98.92
Kalamazoo-Portage, MI $17.33 88.83
Saginaw-Saginaw Township North, MI $20.32 104.15
South Bend-Mishawaka, IN-MI $15.38 78.83
Warren-Troy-Farmington Hills, MI Metro Division $19.76 101.28
Upper Peninsula of Michigan non-Metro Area $17.79 91.18
Northeast Lower Peninsula of Michigan non-Metro Area $17.37 89.03
Northwest Lower Peninsula of Michigan non-Metro Area $17.46 89.49
Balance of Lower Peninsula of Michigan non-Metro Area $17.18 88.06
Source: http://www.bls.gov/oes/2007/may/oes_2600004.htm
For Wisconsin, labor rates were obtained from the state’s Highway Maintenance System (HMS).
Labor rates across the state were adjusted to be equivalent to highway maintenance labor rates in
Milwaukee County (county with the highest labor rate in 2006).
2.3 Cost Transformation Functions

The histogram in Figure 3 shows the distribution of maintenance costs per lane mile (litter
pickup). A normal fit is super imposed on the histogram. Clearly the data does not follow a
normal distribution. The probability plot in Figure 3 confirms that the values do not follow a
normal distribution because they do not resemble a straight line.
The maintenance costs do not follow normal distributions but many can be transformed to fit a
normal distribution. Box-Cox or Johnson transformations correct the non-normality so that the
data can be used for analysis methods that require a normal distribution. Various simulation and
risk analysis procedures require normalized data. This section presents a brief review of the Box-
Cox and Johnson transformations and how the transformation functions would be used in a
cost stimulation analysis.
2.3.1 Box-Cox Transformation

The formula for the Box-Cox transformation function is,
Y X
Where, Y is the transformed value of X. Each value of X is one of the original cost values in
dollars, and  is the transformation parameter. λ is in the range of -0.5 to 0.5. For 0, the
Box-Cox transformation function is:
If a set of cost data can fit a normal distribution using the Box-Cox transformation, then we
would be given the transformation parameter, along with the mean, and standard deviation,
10
, of the fitted normal distribution. For example, if 0.214, then the cost data could be
transformed using the Box-Cox transformation function:
Y  X 0.214
The formula was applied to each cost value in Figure 3. A histogram of the transformed cost
data is shown in Figure 4. The histogram of the transformed values now resembles a normal
distribution. As shown, the transformed data fits a normal distribution with  3.46, and  =
0.4717. The probability plot shows that the values now follow a normal distribution after the
Box-Cox transformation since they resemble a straight line.
60 Mean 384.0
StDev 238.8
N 264
50
40
Frequency
30
20
10
0
0 300 600 900 1200
Cost Figure 3.
Histogram and
probability plot for
99.9
Mean
StDev
384.0
238.8
non-normal cost
99 N 264 data.
AD 5.890
95 P-Value <0.005
90
80
Percent
70
60
50
40
30
20
10
5
0.1
-500 0 500 1000 1500
Cost
The transformation function can be used to model the cost data for a simulation analysis. Cost
per lane mile can be randomly sampled from the normal distribution  = 3.46,  = 0.4717). The
sampled value can be input to the inverse transformation function to obtain an estimated cost
from the original cost non-normal distribution. For this example, the inverse of the
transformation function is:
11
35 Mean 3.460
StDev 0.4717
N 264
30
25
Frequency
20
15
10
0
2.4 2.8 3.2 3.6 4.0 4.4
Figure 4.
Transformed Costs Transformed costs
using Box Cox
transformation
99.9
Mean
StDev
3.460
0.4717
function.
99 N 264
AD 0.260
95 P-Value 0.707
90
80
Percent
70
60
50
40
30
20
10
5
0.1
2.0 2.5 3.0 3.5 4.0 4.5 5.0
Transformed Costs
2.3.2 Johnson Transformation

The Johnson transformation is another method to correct non-normality. Figure 5 shows
another histogram of the cost samples. The data are not normal but can be normalized using a
Johnson transformation function:
Y = -3.897 + 3.563 Asinh((X - 1303.29) / 2246.62)
Where, Y is the transformed value of X. Each X is one of the original sampled costs in Figure 5.
The transformed costs follow a normal distribution with mean μ = -0.026, and standard
deviation, σ = 1.035. Each data set has unique a unique Johnson transformation function.
In order to predict the cost for an activity we need to sample from the normal distribution and
then compute the associated cost for the original distribution. For example, in order to calculate
the cost associated with a value, Y, from the normalized curve we use the following inverse of
the transformation function: The value X would give us the dollar amount.
X=2246.62 sinh((Y+3089719)/3056296)+1303.29
12
50
M ean 4381
40 S tD ev 1151
N 264
Frequency
30
20
10
0
2000 3000 4000 5000 6000 7000 8000 9000
Cost per Lane Mile ($2006)
Figure 5. Histogram of non-normalized cost per lane mile.
50
M ean -0.02563
S tDev 1.035
40 N 264
Frequency
30
20
10
0
-3 -2 -1 0 1 2 3
Transformed Cost per Lane Mile
Figure 6. Histogram of transformed and normalized cost per lane mile
2.4 Maintenance Activity Costs per Lane Mile (CPLM)

Distribution functions were found for the maintenance activity costs per lane mile for three
states: Michigan, Ohio and Wisconsin. The researchers used the quality tools built into
Minitab®. These tools provide statistical assessment for various candidate distributions. The
distribution with the best fit (lowest AD value and a high enough P-value (>0.05 for alpha =
0.95) was selected.
Each state handles their maintenance contracting and cost accounting differently. The important
differences for this project are the level of detail for maintenance work breakdown, resolution of
the roadway sections to which costs are assigned, and finally details of the cost items such as
whether administrative overhead is included or not. Before presenting summary tables of the
cost distribution functions, this report briefly reviews the maintenance management and cost
accounting in each state.
13
2.4.1 Highway Maintenance in Michigan
In Michigan, maintenance is conducted through contracts with the counties and local agencies,
or by direct forces at state maintenance garages. Figure 7 shows the geographic location and
maintenance mechanism
for each of the state’s 83
counties. The
approximate distribution
of lane miles is 5% for
cities and 70% for
counties through
contract agreements,
and 25% for the state
using direct forces.
Maintenance in certain
counties are managed by
one or more state
maintenance garages, so
county level estimates Figure 7. Regions and
are not possible. counties of the
The MDOT Michigan DOT
maintenance cost data including maintenance
was obtained from the responsibilities and
department’s Michigan garage locations
Source: Michigan DOT Maintenance
Architectural Project Support (created in 2004)
(MAP) database. Table 4
and Table 5 summarize
the costs per lane mile for the contracted maintenance and Table 6 summaries costs per lane
mile for maintenance activities performed by direct forces. The costs are listed in equivalent
2006 dollars for the Detroit area with no
administrative overhead. For direct forces,
fringe benefits are included. The original
data set did not contain costs for direct
forces for the year 2006. Thus, Table 6 is
based on years 2004 and 2005, while
Table 4 and Table 5 are based on data for
years 2004, 2005 and 2006.
2.4.2 Highway Maintenance in

Ohio
The Ohio Department of Transportation
(ODOT) manages 12 districts that oversee
88 counties (Figure 8). The district offices
are responsible for maintenance including
winter services such as snow and ice
removal. Figure 8. Transportation districts and counties in Ohio
Ohio DOT, http://www.dot.state.oh.us/dist.asp
14
Ohio’s maintenance costs are managed by ODOT’s web-based project management system,
Ellis (4). This contains the contracted portion of ODOT’s activities in maintaining assets. Table
7 lists the eight cost groups (guardrail, pavement deficiency, vegetation obstruction, litter,
drainage/ditch obstruction, sign deterioration, and pavement markings) from the Ellis system
that were made available for analysis.
2.4.3 Highway Maintenance in

Wisconsin
In Wisconsin, county crews perform
maintenance on state roads within its
jurisdiction. The state monitors
maintenance costs with the Highway
Maintenance System (HMS).
Maintenance costs are documented in the
Highway Maintenance System (HMS).
Maintenance expenditures are recorded
according to designated activity codes (3).
Maintenance activity groups that are used
for this study are show in the Table 8.
Figure 9. Transportation regions and
counties in Wisconsin
http://www.dot.state.wi.us/about/locate/dtd.htm#nc
2.5 Distribution Functions for Maintenance Expenditures

Distribution functions for the cost per lane mile (CPLM) of individual maintenance activities are
listed in Table 4 to Table 8. The column for “statewide average” is the total cost for the activity
divided by the total number of responsible lane miles for which costs were incurred. Using this
method, only the lane miles on which work was performed get included in computing the
average cost. The “average” cost for the county, contract agency, or garage were calculated by
averaging of the cost per lane mile for each county, agency or garage. For the latter method,
large maintenance projects were weighted the same as small maintenance projects.
The columns for average and standard deviation do not include route segments for which no
maintenance was performed. Instead, the table shows a probability of expenditures based on the
number of nonzero cost samples for each of the activity. The total number of samples is the
same for each activity. The probability column is the percentage of statewide lane miles for
which costs were incurred.
The distribution functions are useful for conducting simulation and sensitivity analysis of
maintenance cost estimates. The Mean (μ) and standard deviation (σ) are for normal
distributions using the Box Cox and Johnson Transformation functions
Some activities have very few cost observations. For this reason, distributions could not be
computed for some activities.
15
Table 4. Activities and costs per lane mile for contracts with Michigan’s cities
Maintenance Activity Statewide Contracts with Michigan’s Cities
Activity Average Average Std. dev Probability
Group Code Description ($2006) ($2006) ($2006) % Distribution
Pavements 109 Surface Maintenance 371.02 422.75 862.74 75 Box-Cox: μ=2.416, σ=0.568, λ = 0.165
Johnson: μ=0.063, σ=0.92
Shoulders 119 Shoulders 77.39 108.30 128.75 17 1.682+0.635 Ln((X-0.083)/(842.968-X))
Johnson: μ=0.05, σ=1.03
121 Trees & Shrubs 142.86 204.94 371.73 35 -3.392+0.774 Ln(X+4.724)
Vegetation Johnson: μ=-0.05, σ=1.042
Control 126 Grass & Weed Control 96.69 327.29 516.16 40 2.665+0.682 Ln((X+2.383)/(7139.90-X))
127 Brush Cutting NA NA NA NA No data
122 Drainage & Back Slopes 243.07 305.29 628.15 59 Box-Cox: μ=1.452, σ=0.175, λ =0.078
123 Ditch Cleaning NA NA NA NA Only 1 data point
Drainage
Culvert/Under Drain Johnson: μ=-0.044, σ=0.942
128 218.55 406.61 497.86 4 1.033+0.476 Ln((X-5.630)/(2078.87-X))
Maintenance
Johnson: μ=0, σ=0.974
124 Full Width Litter Pickup 97.32 122.18 260.87 24 2.434+0.56 Ln((X-0.747)/( 3195.44-X))
Litter
132 Sweeping & Flushing 444.77 654.01 2518.57 90 No Fit
Johnson: μ=-0.008, σ=0.986
Signs 160 Traffic Signs 126.92 99.47 238.15 48 4.957+0.681 Ln((X+0.528)/(51435.0-X))
Johnson: μ= 0.09, σ= 0.99
Pavement 162 Pavement Marking 59.49 87.69 131.97 13
2.497+0.68 Ln((X+1.066)/(1446.91-X))
Markings
164 Delineators NA NA NA NA No data
165 Attenuators NA NA NA NA No data
Guardrails Guardrail & Guard Johnson: μ= -0.064, σ= 1.015
130 117.45 153.08 187.20 4
Posts 1.261+0.559 Ln((X+1.286)/(903.645-X))
137 ROW Fence Repair NA NA NA NA Only 2 data points
Johnson: μ=-0.054, σ= 1.02
139 Structures 335.21 761.71 975.92 3
Other 1.039+0.466 Ln((X+9.259)/(3909.56-X))
Johnson: μ=0.03, σ= 1.025
161 Traffic Signals 310.61 535.79 2008.47 37
-3.725+0.591 Asinh((X-0.437)/0.374808)
16
Table 5. Activities and costs per lane mile for contracts with Michigan’s counties
Maintenance Activity Statewide Contracts with Michigan’s Counties
Johnson: μ=-0.023, σ=1.029
Pavements 109 Surface Maintenance 417.35 433.34 1606.64 97
-2.84+0.61 Ln(X+0.812)
Shoulders 119 Shoulders 164.24 231.10 501.70 92 No Fit
Johnson: μ=-0.513, σ=1.025
121 Trees & Shrubs 55.01 61.55 86.72 86
Vegetation 2.316+0.697 Ln((X+0.268)/(886.107-X))
Control 126 Grass & Weed Control 220.54 173.94 233.53 95 No Fit
127 Brush Cutting 27.04 47.73 79.49 3 No Fit
122 Drainage & Back Slopes 111.47 137.07 288.37 88 Box-Cox: μ=1.213, σ=0.094, λ =0.05
123 Ditch Cleaning 41.33 47.50 55.84 12 No Fit
Drainage
Culvert/Under Drain Johnson: μ=0.056, σ=1.03
128 56.83 68.41 120.53 27
Maintenance 2.162+0.554 Ln((X+0.02)/(1004.89-X))
124 Full Width Litter Pickup 193.65 130.54 311.76 96 No Fit
Litter Johnson: μ=0.033, σ=0.997
132 Sweeping & Flushing 173.65 132.78 294.13 77
-2.392+0.652 Ln(X+0.39)
Signs 160 Traffic Signs 86.37 78.36 120.19 92 No Fit
Johnson: μ=-0.009, σ= 0.899
Pavement 162 Pavement Marking 18.61 104.04 264.54 2
-0.86+0.33 Asinh((X-1.918)/0.739)
Markings
164 Delineators 22.38 19.38 29.12 33 Box-Cox: μ=1.299, σ= 0.190, λ= 0.113
Johnson: μ= -0.030, σ= 1.030
165 Attenuators 267.39 329.13 741.48 9
2.001+ 0.443 Ln((X - 0.173) / (6348.31 - X))
Guardrails
Guardrail & Guard
130 123.00 99.17 177.33 68 Box-Cox: μ=1.289, σ=0.15, λ = 0.071
Posts
Johnson: μ= -0.021, σ= 0.997
137 ROW Fence Repair 66.06 63.74 127.01 18
2.633+0.595 Ln((X+0.295)/(1796.81-X))
Other 139 Structures 115.17 97.88 261.52 26 Box-Cox: μ=1.285, σ= 0.21, λ= 0.08
Johnson: μ= -0.02, σ= 0.925
161 Traffic Signals 124.54 157.07 277.87 20
1.693+0.434 Ln((X-0.006)/(2207.00-X))
17
Table 6. Activities and costs per lane mile for Michigan’s direct maintenance forces
Maintenance Activity Statewide Maintenance Garage
10100 Joint & Crack Filling 126.82 143.09 192.17 77 No Fit
Remove/Replace Pave-
10200 127.95 140.33 254.11 69 Box Cox: μ=1.461, σ=0.313; λ = 0.1071
ment (Fast Set Concrete)
Surface 10300 Patrol Patching 125.53 130.24 134.31 91 Box Cox: μ=3.5378, σ=1.233; λ = 0.2845
Maintenance Bituminous Johnson: μ=-0.035, σ=1.015
10500 58.01 60.42 74.02 62
Maintenance & Repair 1.182 + 0.509 Ln((X + 0.114) / (329.933 - X))
Johnson: μ=-0.011, σ=1.113
10800 Bump Removal 19.99 23.86 34.19 53
-2.361 + 0.885 Asinh((X - 1.275) / 1.485)
Johnson: μ=0.028, σ=0.9334
11000 Routine Blading 26.76 30.63 36.54 75
1.406 + 0.549 * Ln((X - 0.0585) / (207.514 - X))
Shoulder Gravel Shoulder Johnson: μ=0.029, σ=0.998
11200 128.62 137.11 159.38 86
Maintenance Maintenance 2.051 + 0.744 Ln((X - 0.431) / (1268.98 - X))
Paved Shoulder Johnson: μ=0.049, σ=0.991
11400 67.28 76.80 150.13 46
Maintenance -2.005 + 0.553 Asinh((X - 0.857) / 0.964)
12000 Tree Removal 39.93 43.78 44.81 78 No Fit
Johnson: μ=-0.011, σ=0.986
12100 Stump Removal 6.33 8.40 11.38 51
1.588 + 0.562 Ln((X + 0.038) / (66.411 - X))
Johnson: μ=0.023, σ=1.011
12600 Area Moving 27.55 31.40 42.22 74
Vegetation 1.407 + 0.526 Ln((X + 0.061) / (202.380 - X))
Control Johnson: μ=-0.061, σ=1.047
12700 Brush Control 34.67 37.14 39.74 82
1.087+0.605 Ln ((X+0.5996)/(171.709-X))
Johnson: μ=0.093, σ=0.883
17100 Tree Trimming 11.37 12.92 31.76 43
-1.075+0.691 Ln(X+0.705)
17200 Vegetation Control 11.65 11.75 13.45 29 Box Cox: μ=1.480, σ=0.478; λ = 0.215
Johnson: μ=-0.063, σ=0.885
12200 Catch Basin Cleanout 27.26 31.75 53.42 51
1.108 + 0.312 Ln((X - 0.093) / (236.214 - X))
Johnson: μ=-0.011, σ=0.941
Drainage 12300 Ditch Cleanout 56.52 57.52 64.45 74
-5.0698 + 1.266 * Ln(X + 14.720)
Culvert & Underdrain
12800 19.26 21.12 27.84 69 Box Cox: μ=1.614, σ=0.405; λ = 0.192
Cleaning
18
Maintenance Activity Statewide Maintenance Garage
Culvert & Underdrain Johnson: μ=-0.027, σ=0.972
12810 97.22 104.02 127.69 83
Cleaning 2.866 + 0.903 Ln((X + 8.357) / (1822.56 - X))
12400 Litter Pickup 65.05 70.41 75.35 90 No Fit
13200 Approach Sweeping 10.38 12.45 14.83 77 Box Cox: μ=1.572, σ=0.4758; λ = 0.228
Litter
Johnson: μ=0.00006, σ=0.9993
13600 Curb Sweeping 17.23 17.96 19.77 34
0.864 + 0.434 Ln((X - 0.107) / (73.414 - X))
Johnson: μ=-0.029, σ=0.9901
13000 Guardrail Repair 52.09 52.53 55.08 83
2.040 + 0.888 * Ln((X + 4.984) / (444.600 - X))
Guardrails
Johnson: μ=0.0007, σ=0.944
13010 Guardrail Ending Repair 15.72 18.93 30.33 62
-1.608 + 0.738 Ln(X + 0.532)
Right of Way Fence Johnson: μ=0.033, σ=0.948
Other 13700 21.42 25.23 50.01 62
Repair 1.785+0.426 Ln((X-0.051)/(368.019-X))
19
Table 7. Activities and costs per lane mile for counties in Ohio
Maintenance Activity Statewide County
End Assembly Installation and Johnson: μ=0.003, σ=1.00,
6230 34.45 30.08 47.63 59
Maintenance 2.0690 + 0.5909 Ln((X + 0.0169) / (399.70 - X))
6233 Repair, Replacement, or Removal 161.21 164.93 139.58 97 No Fit
Guardrail
Crash Attenuator Repair or Johnson: μ=0.072, σ=0.921 1.8658 + 0.4306

6235 23.77 18.24 39.15 19
Replacement Ln((X - 0.0393) / (265.16 - X))
Concrete Median Barrier Repair, Johnson: μ=0.005, σ=0.945,
6237 Replacement, Restoration, Removal 8.34 7.46 8.26 14
3.2354 + 0.9300 Ln((X + 0.2239) / (167.43 - X))
Johnson: μ=0.032, σ=1.031
6333 Guardrail Betterment 28.62 24.97 63.57 27
-0.8990 + 0.5100 Asinh((X - 2.6384) / 1.1621)
6120 Pavement Underseal/Fill Voids 2.23 2.298 1.81 4 Box Cox: μ=0.533, SC=0.834, λ = 0
Johnson: μ=-0.033, σ=1.004,
6121 Pothole Patching 337.21 318.3 262.60 100
2.6524 + 1.1335 Ln((X + 18.72) / (3029.60 - X))
6122 Surface Repairs 212.56 223 280.20 87 No Fit.
Johnson: μ=-0.006, σ= 1.00
Pavement Deficiency
6123 Full Depth Repair 102.15 118.62 249.34 78

-1.6388 + 0.5871 Asinh((X - 4.4339) / 3.6576
6124 Fill and Seal Joints and Cracks 167.33 176.03 160.78 72 Box-Cox: μ=5.264, σ=1.847, λ = 0..341
6125 Surface Treatment 218.03 237.4 244.7 35 No Fit
6126 Pavement Jacking 4.04 4.15 3.44 2 Box Cox: μ= 0.596. σ=0.214, λ = -0.5
Planning Bituminous Johnson: μ=-0.0124, σ=1.000,
6127 29.50 29.23 36.15 76
Pavement 2.2019 + 0.7175 Ln((X - 0.3079) / (341.43 - X))
6129 Surface Paving 215.08 216.4 265.80 43 Gamma: S=0.535, SC=404.06
Johnson: μ=-0.0004, σ=0.961,
6136 Partial Depth Repair 117.17 120.02 157.63 40
1.3218 + 0.4945 Ln((X - 0.2150) / (788.36 - X))
Johnson: μ=-0.0177, σ=0.968,
6321 Roadway Betterment 31.72 34.33 60.03 30
-1.4509 + 0.4380 Asinh((X - 1.0177) / 0.4418)
Johnson: μ=-0.020, σ=1.026,
6130 Spot Berming 186.75 191.47 171.12 100
4.4614 + 1.3696 Ln((X + 25.73) / (4648.43 - X))
Drop Off
Pavement
Blading-Restore Unpaved
6131 308.46 331 257.73 100 Box Cox: μ=16.648, σ=7.36, λ= 0.5
Berm and/or Shoulders
Johnson: μ=-0.054, σ=1.022,
6132 Repair Curbs and or Gutters 8.79 9.26 15.33 19
2.2233 + 0.605 Ln((X + 0.0084) / (148.74 - X))
20
Berm and/or Shoulder Johnson: μ=-0.0152, σ=1.022,
6331 52.41 56.38 110.79 73
Betterments 3.76 + 0.6848 Ln((X + 0.2848) / (5024.23 - X))
6220 Weed Eating 34.59 36.08 34.29 35 Box Cox: μ= 2.516 σ=0.784. λ =0.285
Johnson: μ=-0.013, σ=0.988,
6221 Mowing 397.74 415.83 185.14 100
Vegetation
0.4544 + 1.9991 Ln((X + 294.54) / (1302.4 - X))

Control
Chemical Spray Vegetation

6222 Management
47.44 47.66 39.17 78 No Fit
Care of Shrubs, Plants and Johnson: μ=0.0037, σ=0.981,
6223 355.19 366.02 327.30 100
Trees -2.3852 + 1.4605 Asinh((X - 26.01) / 108.45)
Cleaning Curbs, Gutters & Along
6231 Median Barriers
65.48 57.94 97.94 83 Box Cox: μ=1.28, σ=0.145. λ = 0.077
Litter
6232 Litter Pickup 399.54 384.01 238.78 100 Box Cox: μ=3.471, σ=0.474, λ = 0.214
6236 Litter Patrol 107.22 106.19 138.09 82 No Fit
Johnson: μ=0.073, σ=0.998
6133 Repairing Slopes 34.32 34.66 37.43 75
1.664 + 0.6570 Ln((X - 0.5027) / (239.35 - X))
Johnson: μ=-0.0197, σ=0.969,
6134 Repairing Slips and Slides 131.72 129.7 308.20 67
2.377 + 0.4634 Ln((X - 0.2774) / (4136.29 - X))
Johnson: μ=-0.0421, σ=1.025,
6135 Ditch & Shoulder Relocation 38.29 43.95 104.91 35
-1.7415 + 0.6568 Ln(X + 0.1358)
Johnson: μ=-0.0178, σ=0.983,
6140 Culvert Inspection 21.73 22.96 32.80 76
1.6963 + 0.556 Ln((X - 0.1422) / (210.70 - X))
Drainage
Johnson: μ=-0.0069, σ=0.995,

6141 Cleaning & Reshaping Ditches 310.46 314.81 299.62 99
-7.8403 + 1.383 Ln(X + 62.383)
Johnson: μ=-0.021, σ=1.099,
6142 Cleaning Channels 31.72 33.86 45.52 73
1.915 + 0.665 Ln((X - 0.1084) / (300.85 - X))
Johnson: μ=0.010. σ=0.9501,
6143 Clean Drainage Structures 120.64 120.25 87.88 100
2.192 + 1.132 Ln((X + 6.295) / (814.97 - X))
6144 Repair Drainage Structures 169.20 174.57 147.65 100 Box Cox: μ=5.368, σ=1.727, λ = 0.3423
Clean Ditches with Tiger Johnson: μ=-0.0215, σ=0.964
6145 47.67 51.23 47.24 35
Ditcher 0.8839 + 0.543 Ln((X + 0.2176) / (206.23 - X))
Inspect and Clean Underdrain
6146 13.44 12.64 17.51 59 Box Cox: μ=1.288, σ=0.221, λ = 0.1322
Outlets
21
Johnson: μ=-0.0470, σ=1.017,
6343 Culvert Betterment 128.01 134.63 168.12 84
1.792 + 0.627 Ln((X + 0.6676) / (1299.1 - X))
Catch Basin Repair, Replace, Johnson: μ=0.0116, σ=1.0335,
6344 26.48 27.41 34.90 77
or Install 1.836 + 0.644 Ln((X - 0.0558) / (242.81 - X))
Ground-Mounted Flat Sheet Johnson: μ=0, σ=0.9924
6521 191.10 193.13 93.66 100
Sign Maintenance 3.074 + 1.948 Ln((X + 34.238) / (1211.29 - X))
6522 Delineator Maintenance 36.19 37.56 42.37 74 No Fit
Ground Mounted Extrusheet
6523 30.33 29.33 37.42 81 Box-Cox: μ=1.5507, σ=0.3071, λ = 0.1561
Signs
Signs
Overhead Mounted Johnson: μ=-0.0057, σ=0.947,

6524 10.00 9.90 15.91 32
Extrusheet Signs 2.3639 + 0.608 Ln((X - 0.0322) / (197.38 - X))
Overhead Sign Support Johnson: μ=0.0374, σ= 0.993,
6525 3.81 3.69 4.68 19
Maintenance 1.8876+ 0.639 Ln((X - 0.1276) / (33.08 - X))
Overhead Mounted Flat Sheet Johnson: μ=-0.0696, σ=0.929,
6554 60.93 72.60 124.20 14
Sign Maintenance 1.0513 + 0.286 Ln((X - 0.2942) / (474.20 - X))
Johnson: μ=0.0027, σ=1.015,
6531 Auxiliary Pavement Markings 20.79 22.64 29.33 89
-2.849 + 1.059 Ln(X + 0.9864)
Pavement Marking
6532 Center Line and Tee 11.52 11.77 11.93 66 Box Cox: μ=1.461, σ=0.280, λ = 0.1827
Johnson: μ=-0.0093, σ=1.039,
6533 Edge Line Marking 50.95 51.71 63.65 77
-2.924 + 0.874 Ln(X + 1.464)
Johnson: μ=0.0433, σ=0.988,
6534 Lane Line Marking 3.37 3.51 3.30 18
-1.501 + 0.920 Asinh((X - 0.636) / 0.663)
Raised Pavement Markers Johnson: μ=0.0177, σ=1.000
6535 41.38 43.20 53.25 45
Maintenance 1.615 + 0.6230 Ln((X + 0.8518) / (320.33 - X))
6536 Remove Markings 14.40 15.77 68.26 37 Box-Cox: μ=0.8557, σ=0.1898, λ = -0.1638
22
Table 8. Activities and costs per lane mile for contracts with Wisconsin’s counties
Maintenance Activity Statewide Wisconsin’s Counties
Spot Repair / Pothole Johnson: μ=0.0767, σ =1.0402
1 149.67 121.81 101.95 100
Repair -8.35 + 1.74*Ln(X + 29.11)
2 Crack Sealing / Filling 122.19 126.56 109.99 94 Box Cox: μ= 5.72943; σ= 2.28057; λ = 0.382913
Roadways - 3 Sealcoating 36.99 57.36 133.07 27 No Fit
Asphalt Johnson: μ= -0.045, σ = 0.9507
4 Wedging / Rut Filling 38.74 44.62 72.23 61
1.35 + 0.38*Ln((X - 0.02)/(380.99 - X))
Johnson: μ= 0.028, σ =1.022
5 Milling 13.70 13.59 20.84 56
2.096+0.602*Ln((X-0.05)/(173.78-X))
8 Thin Resurfacing 70.58 91.96 149.37 65 Box Cox: μ= 1.57; σ= 0.49; λ = 0.141
Johnson: μ= -0.048, σ = 1.013
11 Emergency Repair 9.27 9.17 19.91 48
2.71+0.57*Ln((X-0.01)/(325.84-X))
Roadways -
12 Non-Emergency Repair 45.88 47.24 70.01 68 Box Cox: μ= 1.68; σ= 0.53; λ = 0.177
Concrete
Repair of Distressed Johnson: μ= 0.009, σ = 1.065
13 33.74 26.32 46.50 50
Pavement 1.72+0.43*Ln((X-0.0056)/(291.92-X))
Johnson: μ= -0.011, σ = 1.007
21 Grading Gravel Shoulders 157.33 166.93 91.93 100
Roadways - -12.4203+2.36754*Ln(X+39.8533)
Shoulders Johnson: μ= 0.015, σ = 1.107
22 Repairing Paved Shoulders 9.32 9.83 26.03 64
-2.48+0.62*Asinh((X-0.05)/0.08)
Johnson: μ= 0.021, σ =1.011
31 Sweeping Pavement 41.12 30.84 38.35 98
-2.863+1.051*Asinh((X+0.588)/2.554)
Litter
Johnson: μ= -0.0001, σ=0.958
42 Litter Pickup 178.90 128.00 192.66 100
-2.22+0.81*Asinh((X-3.27256)/7.90)
Johnson: μ= 0.010, σ =0.992
41 Mowing 140.40 129.00 60.08 100
1.75+1.63 Ln((X+12.38)/(508.32-X))
Vegetation
Johnson: μ= -0.018, σ = 1.023
Control 43 Woody Vegetation 99.23 104.19 70.33 100
3.32+1.65*Ln((X+24.88)/(959.57-X))
44 Noxious Weed Control 22.85 22.01 28.52 86 Box Cox: μ= 1.65; σ= 0.47; λ = 0.203
Clean / Repair Drainage Johnson: μ= 0.03, σ = 1.015
51 59.44 51.27 71.99 99
Drainage Structure -2.90+0.88*Ln(X+1.19)
52 Roadside Drainage 67.51 66.42 61.65 100 Box Cox: μ= 2.28; σ= 0.44; λ = 0.216
Guiderails 55 Safety Appurtenances 45.56 38.77 35.53 96 Box Cox: μ= 2.731; σ= 0.849; λ = 0.301
23
Maintenance Activity Statewide Wisconsin’s Counties
81 Routine Repair 27.85 28.06 19.68 100 Box Cox: μ= 4.90; σ= 2.02; λ = 0.5
Sign
Johnson: μ= -0.027, σ = 0.987
Repair 85 Emergency Repair 4.51 4.77 6.40 88
2.30+0.69 Ln((X+0.00082)/(71.23-X))
Pavement Johnson: μ= -0.009, σ = 1.039
90 Pavement Marking 166.85 256.91 384.77 28
Marking -1.70+0.69 Asinh((X-27.31)/14.51)
24
Chapter 3. CATEGORICAL MAINTENANCE COSTS
This chapter presents statistical analysis and distribution functions for cost per lane mile (CPLM)
of common maintenance activities grouped into typical maintenance categories. The scope of
maintenance categories includes pavements, shoulders, drainage, litter, vegetation control,
pavement markings, signs, and guardrails. The purpose for defining maintenance categories is to
explore economies scope that might exist depending on the range of activities. The specific
activities in each category vary among the three states, thus direct comparisons of the magnitude
of costs are not possible.
The analysis does provide insight on relative trends in CPLM. The statistical analysis of the mean
CPLM for each maintenance category focuses on significant annual trends over the three year
study period and trends in CPLM among the service providers in each state. The CPLM
observations were assigned to lane mile (LM) groups based on responsible lane miles. The
purpose for defining LM groups is to be able to explore economies of scale and scope that
might exist depending on the volume of maintenance work and range of activities. The
significance differences were based on a 95% confidence interval.
Individual distribution functions were identified for characterizing the maintenance CPLM for
each category and state. For most categories, maintenance CPLM follows normal distributions
after transformation using Box-Cox or Johnson transformation functions. If analysis of the
mean indicated significant difference among costs for LM groups, then specialized distribution
functions were identified. Normal distribution functions are desirable for statistical simulation
and sensitivity analysis and could be use for the forecasting.
3.1 Michigan’s Maintenance Costs per Lane Mile
3.1.1 Maintenance Activity Categories and Lane Mile (LM) Groups

Maintenance responsibility for Michigan’s is divided among the cities, counties and state forces.
Data sets for Michigan include activity costs for contracts with cities and counties, and direct
forces at state maintenance garages in 83 counties over the three year period. Table 9 lists the
activity breakdown for each of the eight each maintenance categories.
The CPLM observations were assigned to lane mile (LM) groups based on responsible lane miles
of the service provider and to a lesser degree by the length in the route sections maintained each
of the three maintenance provider groups. Maintenance activities were performed on sections
of highways of various lengths which allowed the researchers to separate the data in groups by
the corresponding lane miles. The data set did not include route segments for the maintenance
costs incurred for direct forces at the state’s maintenance garages. As such, route segment lane
miles were not used to group the maintenance garages. The LM groups were found using the K-
means clustering of observations. This procedure uses non-hierarchical clustering of
observations according to MacQueen's algorithm (11). Each observation was assigned to the
cluster whose centroid it is closest to.
Table 10 and Figure 10 show the ranges of responsible lane miles and route segment lane miles
for each group. The table lists the ranges for each group and the boxplots illustrate those
ranges. The samples column in Table 10 indicates the number of maintenance activity costs for
25
the full scope of activities in the maintenance categories. The asterisks in Figure 10 indicate
outliers. (Section 3.1.2 contains a full description of how to interpret the boxplots.) Multiple
asterisks may be for a single county or route segment because there are multiple activities
observations. The outliers were not removed.
Table 9. Maintenance categories for analysis of Michigan’s maintenance costs
Maintenance Activities for Michigan Maintenance Activities for Michigan State
Maintenance Local Agency and Counties Maintenance Garages
Category Code Description Code Description
10100 Joint & Crack Filling
Remove/Replace Pavement (Fast Set
10200
Concrete)
Surface 1090 Surface Maintenance
10300 Patrol Patching
10500 Bituminous Maintenance & Repair
10800 Bump Removal
11000 Routine Blading
Shoulders 1190 Shoulders 11200 Gravel Shoulder Maintenance
11400 Paved Shoulder Maintenance
1220 Drainage & Back Slopes 12200 Catch Basin Cleanout
1230 Ditch Cleaning 12300 Ditch Cleanout
Drainage
Culvert/Under Drain 12800 Culvert & Underdrain Cleaning
1280
Maintenance 12810 Culvert & Underdrain Cleaning
12000 Tree Removal
1210 Trees & Shrubs 12100 Stump Removal
Vegetation 17100 Tree Trimming
Control 12600 Area Mowing
1260 Grass & Weed Control
17200 Vegetation Control
1270 Brush Cutting 12700 Brush Control
1240 Full Width Litter Pickup 12400 Litter Pickup
Litter 13200 Approach Sweeping
1320 Sweeping & Flushing
13600 Curb Sweeping
Signs 1600 Traffic Signs
1620 Pavement Marking No Activities
Markings
1640 Delineators
1300 Guardrail & Guard Posts 13000 Guardrail Repair
Guardrails
1650 Attenuators 13010 Guardrail Ending Repair
Table 10. Lane mile (LM) groups for Michigan’s highway maintenance providers
LM Responsible Lane Miles Route Segment Lane Miles
Provider Group Samples Min Mean Max Min Mean Max
1 318 0.77 7.41 15.59 0.75 4.99 11.70
Cities 2 180 13.31 21.69 35.16 1.60 9.74 23.35
3 69 34.6 53.75 77.04 3.81 16.74 37.51
1 435 10.7 195.3 265.5 1.34 46.38 136.86
2 273 265.5 334.9 421.4 1.70 59.86 194.95
Counties
3 234 453.9 544.3 688.6 4.25 62.64 195.80
4 120 1125.0 1499.5 1774.8 7.60 110.42 541.53
1 20 120.5 178.5 200.0
State
2 35 230.3 275.1 351.0
Garages
3 34 393.6 497.9 659.4
26
The decision to use
1 2 3 three lane mile groups
RLM Route Lane Miles for cities and garages and
80 40
four groups for counties
70
is not arbitrary. If more
60 30 groups were used, the
50 samples are fewer and
40
thus less representative
20
sizes become smaller. If
30
fewer groups are used
20 10 then data trends
10 attributed to responsible
0
lane miles may not be
0
visible. As shown, the
1 2 3
LM Group - Michigan's cities proving state highway maintenance ranges of responsible
lane miles for the cities
and garages are narrower
1 2 3 4 than for the counties. On
2000
RLM
600
Route Lane Miles average the cities are
responsible for
500 maintaining fewer lane
1500 miles and shorter
400
sections than the
1000 300
counties and state
maintenance garages.
200
500
100
0 0
1 2 3 4
LM Group - Michigan's counties providing highway maintenance
Figure 10. Lane mile
(LM) groups for
700
Michigan’s highway
600
maintenance providers
Responsibile Lane Miles
500
400
300
200
100
1 2 3
LM Group - Michigan maintenance garages
27
3.1.2 Cost Trends for LM Groups
This analysis looks for
significant differences 1 2 3
and trends in cost per Pavements $/LM(no outliers) Shoulders $/LM(no outliers)
lane mile (CPLM) 3000 300
among the service
2000 200
providers grouped by
responsible miles. 1000 100
The plots in Figure 11
0 0
to Figure 13 show the Vegetation $/LM(no outliers) Drainage $/LM(no outliers)
3000
expenditures in 2400
CPLM for each 1800 2000
maintenance category 1200
for the three 1000
600
procurement 0 0
methods: contracts 1 2 3
with cities and LM Group - Michigan's Cities
counties, and direct
forces at state
maintenance garages. 1 2 3
Litter $/LM(no outliers) Signs $/LM(no outliers)
The boxplots (also 4800 1000
called box-and- 3600 750
whisker plots) are 2400 500
useful to assess and
1200 250
compare the cost
distributions. Each 0 0
Markings $/LM(no outliers) Guardrails $/LM
box encloses the 300 800
range for half of the 600

200
cost data for the LM 400
group and 100 200
maintenance category.
0 0
The top of the box is
1 2 3
the third quartile (Q3)
LM Group - Michigan's Cities
- 75% of the CPLM
values are less than or Figure 11. CPLM trends for LM groups – contracts with Michigan’s
equal to this value. cities.
The bottom of the
box is the first quartile (Q1) - 25% of the CPLM values are less than or equal to this value. The
line across each box is the median - the middle of the data. Half of the observations are less than
or equal to the median. The lower whisker extends to the lowest CPLM value within the lower
limit. (Lower limit = Q1- 1.5 (Q3 - Q1).) The upper whisker extends to the highest CPLM value
within the upper limit. (Upper limit = Q3 + 1.5 (Q3 - Q1).)
28
The asterisks beyond the whiskers are outliers - unusually large observations. A few extremely
high CPLM values were removed from the data to prevent skewing of results. Those costs
appear to be for major rehabilitation maintenance projects. Most of the remaining outliers occur
in the plots for contracts with Michigan’s cities (Figure 11). This is most likely due having fewer
samples for costs and the narrow range of lane miles within each LM group for the cities.
Among the service
1 2 3 4 providers, CPLM for
Pavements $/LM(no outliers) Shoulders $/LM(no outliers)
3000
600
contracts with Michigan’s
450
counties show clear trends
2000
300
(Figure 12). CPLM for
1000
150
pavements, vegetation
control, drainage, litter,
0 0
1600
Vegetation $/LM(no outliers) Drainage $/LM(No outliers) signs, guardrails, and to a
800
lesser extent markings, tend
1200 600
to increase as the number of
800 400
responsible lane miles
400 200 increases. From the plots, it
0 0 might be concluded that the
1 2 3 4
CPLM increases as the scale
LM Group - Michigan's Counties
of the maintenance
operation increases. The
1 2 3 4 CPLM for shoulder
Litter $/LM(no outliers) Signs $/LM(no outliers)
400 maintenance shows an
2000
300 inverse relationship. As the
1500
200 number of responsible lane
1000
100
miles increases, the
500
maintenance CPLM
0 0
Markings $/LM(no outliers)
1600
Guardrails $/LM(no outliers) decreases. Areas with many
60 lane miles may be more likely
1200
45 to have paved shoulders.
30
800
Paved shoulders require less
15 400
maintenance than the
0 0 unpaved shoulders which
1 2 3 4
likely occur in areas with
LM Group - Michigan's Counties
fewer lane miles.
Figure 12. CPLM trends for LM groups – contracts with Table 11 summarizes the
Michigan’s counties statistical
Table 11. Analysis of the mean CPLM for LM groups – Michigan (95% CI) analysis of the
Maintenance Mean CPLM differs from population mean for LM groups: mean. The
Category Cities Counties State Garages mean CPLM
Pavements No difference 1,2,3,4 No difference of each LM
Shoulders 1 1,2,4 No difference group was
Drainage 2 1,2,4 2,3
compared to
Vegetation Control 3 1,2,3,4 No difference
Litter 2 1,2,3,4 No difference the population
Signs No difference 1,2,3,4 No data mean for all
Markings 1 1,2,3 No data LM groups.
Guardrail No difference 1,2,4 No difference The analysis
29
used a confidence interval
1 2 3
of 95%. The numeric
Surface $/LM(no outliers) Shoulders $/LM(no outliers) values in Table 11 are the
800
1000
LM groups with mean
600 CPLM that differs from the
750
500
400 population mean for the
250 200 maintenance category. For
0 0 drainage maintenance, the
Vegetation $/LM Drainage $/LM(no outliers)
800 mean CPLM for LM group
400
600 2 is different from the
300
400 population mean for all
200
200
cities; the means of LM
100
groups 1, 2 and 4 are
0 0
1 2 3
different from the
LM Group - Michigan's Maintenance Garages population mean for all
counties; and the mean
CPLM for LM groups 2 and
1 2 3
Litter $/LM(no outliers) Guardrails $/LM(no oultiers)
3 are different from the
300
250 population mean for the
state garages. The analysis
250 200 indicated little difference in
200
CPLM among state garages
150 except for the drainage
150 category. Overall, CPLM
100
100
for drainage maintenance
experienced the most
50
50 difference among LM
0 0
groups for all maintenance
providers.
1 2 3
LM Group - Michigan's Maintenance Garages
Figure 13. CPLM trends for LM groups – direct forces at

Michigan’s maintenance garages.
3.1.3 Annual Cost Trends during the Study Period

The boxplots in Figure 14 show the range of annual CPLM for each maintenance categories for
contracts with the counties and Figure 15 shows the same plots for contracts with the cities. The
plots indicate no significant changes in expenditures during the study period.
Statistical analysis of the mean (Table 12) indicated that average CPLM for vegetation control in
2006 is significantly different than the average cost over the three year period. This may be due
to the effects of the numerous outliers.
Annual CPLM for maintenance performed by direct forces at state maintenance garages are
shown in Figure 16. Only two years of data was available. After adjusting to 2006$, the plots
clearly show significant differences in 2004 and 2005 for all categories.
30
2004 2005 2006 2004 2005 2006
Pavements $/LM(no outliers) Shoulders $/LM(no outliers) Pavements $/LM(no outliers) Shoulders $/LM(no outliers)
3000
600 3000 300
2000 450
2000 200
300
1000 1000 100
150
0 0 0 0
Vegetation $/LM(no outliers) Drainage $/LM(No outliers) Vegetation $/LM(no outliers) Drainage $/LM(no outliers)
1600 800 3000
2400
1200 600 1800 2000
800 400 1200
1000
400 200 600
0 0 0 0
2004 2005 2006 2004 2005 2006
Fiscal Year - Michigan's Counties Fiscal Year - Michigan's Cities
2004 2005 2006 2004 2005 2006

Litter $/LM(no outliers) Signs $/LM(no outliers) Litter $/LM(no outliers) Signs $/LM(no outliers)
400 4800 1000
2000
300 3600 750
1500
1000 200 2400 500
500 100 1200 250

0 0 0 0
Markings $/LM(no outliers) Guardrails $/LM(no outliers) Markings $/LM(no outliers) Guardrails $/LM
1600 300 800
60
1200 600
45 200
800 400
30
400 100 200
15
0 0 0 0
2004 2005 2006 2004 2005 2006
Fiscal Year - Michigan's Counties Fiscal Year - Michigan's Cities
Figure 14. Annual CPLM trends - contracts Figure 15. Annual CPLM trends - contracts
with Michigan’s counties with Michigan’s cities
Table 12. Analysis of the mean annual CPLM – Michigan (95% CI)
Maintenance Mean CPLM differs from population mean for fiscal years:
Category Cities Counties State Garages
Pavements No difference No difference 2004, 2005
Shoulders No difference No difference 2004, 2005
Drainage No difference No difference 2004, 2005
Vegetation Control No difference 2006 2004, 2005
Litter No difference No difference 2004, 2005
Signs No difference No difference No data
Markings No difference No difference No data
Guardrail No difference No difference 2004, 2005
2004 2005 2004 2005

Surface $/LM(no outliers) Shoulders $/LM(no outliers) Litter $/LM(no outliers) Guardrails $/LM(no oultiers)
800 250
1000 300
600
750
400 250 200
500
250 200 200
150
0 0
Vegetation $/LM Drainage $/LM(no outliers) 150
800
400 100
600 100
300
200 400 50
50
100 200
0 0
0 0
2004 2005 2004 2005
Fiscal Year - Michigan's Maintenance Garages Fiscal Year - Michigan's Maintenance Garages
Figure 16. Annual CPLM trends - direct forces at Michigan’s maintenance garages
31
3.1.4 Cost Distribution Functions for Maintenance Categories
Michigan’s maintenance costs include labor plus benefits, equipment, and materials cost. The
administrative costs were removed for this analysis. Table 13 to Table 15 list the cost
distribution functions for the eight maintenance categories. The specific activities included in
these categories are listed in Table 9. The ranges for the LM groups are listed in Table 10.
The set of LM groups for each distribution function was determined based on a pair-wise
comparison of mean CPLM. The pair-wise comparisons used a 95% confidence level.
Organizing the LM groups by this criterion generalized the distribution models for the CPLM.
The maintenance costs can be represented by normal distributions with only a few categories
having no fit. The Box-Cox and Johnson transformation functions are explained in Section 2.3.
The mean and standard deviation of CPLM are listed in the tables. For many categories the
standard deviation is greater than the mean indicating very wide ranges.
Table 13. Categorical CPLM – contracts with Michigan’s cities
Maintenance LM Probability Mean Std.dev. Transformation Function and
Category Group % ($2006) ($2006) Distribution
Pavements 1,2,3 75 388.01 479.02 Box-Cox: λ=0.2, μ = 2.93, σ=0.808
Johnson: μ=0.076, σ=0.936
Shoulders 1,2,3 17 82.11 76.26
1.074+0.609 Ln((X-0.366)/(344.77-X))
Johnson: μ=-0.002, σ=0.976
Drainage 1,3 56 336.11 490.33
1.966+0.577 Ln((X+0.48)/(4136.07-X))
Johnson: μ=-0.043, σ=0.999
1,2 57 276.55 401.03
Vegetation 2.980+0.749 Ln((X+3.462)/(7709.74-X))
Control Johnson: μ=-0.028, σ=1.015
1,3 54 343.77 491.25
2.971+0.744 Ln((X+6.605)/( 9293.41-X))
1,3 93 537.61 706.09 No Fit
Litter Johnson: μ=0.036, σ=1.033
2,3 90 434.28 522.39
-2.435+1.344 Asinh((X+11.582)/104.74)
Johnson: μ=-0.02, σ=0.988
Signs 1,2,3 48 83.39 140.46
2.874+0.671 Ln((X+0.316)/(2579.69-X))
Johnson: μ=0.055, σ=1.127
1,3 13 86.59 92.93
1.416+0.656 Ln((X-1.060)/(428.32-X))
Markings
Johnson: μ=-0.046, σ=0.991
2,3 9 38.91 39.60
0.779+0.503 Ln((X+0.386)/(149.01-X))
Johnson: μ=-0.064, σ=1.015
Guardrails 1,2,3 4 153.08 187.20
1.261+0.559 Ln((X+1.286)/(903.65 - X))
Table 14. Categorical CPLM – Michigan’s direct forces at state maintenance garages
Mean Std.dev. Transformation Function and
Maintenance Category LM Group ($2006) ($2006) Distribution
Surface 1,2,3 346.78 299.90 Box-Cox: λ=0.5, μ = 16.445, σ=8.79
Johnson: μ=-0.001, σ=0.92
Shoulders 1,2,3 172.46 176.42
1.512+0.683 Ln((X+2.659)/(1126.26-X))
Johnson: μ=0.013, σ=1.022
Vegetation Control 1,2,3 108.48 104.45
0.948+0.586 Ln((X+4.402)/(441.55-X))
1,2 203.38 175.03 Box-Cox: λ=0.5, μ = 12.61, σ=6.73
Drainage
1,3 126.31 122.90 Box-Cox: λ=0.5, μ = 9.746, σ=5.657
Johnson: μ=0.028, σ=0.984
Litter 1,2,3 83.48 74.59
0.825+0.521 Ln((X-0.282)/(302.09-X))
Johnson: μ=-0.004, σ=1.083
Guardrails 1,2,3 62.10 57.02
1.764+0.963 Ln((X+6.624)/(366.53-X))
32
The tables list the probability of expenditure for each category and LM grouping. The
probabilities were computed using joint probability theory. The probability of expenditure for
the category is equal to the probability of expenditure for at least one of the activities in the
category. More simply, the probability of expenditure is equal one minus the probability of no
expenditures. The probability of no expenditures can be computed by counting the number of
samples with no expenditures for any of the activities and dividing by the total number of
samples for the category.
Table 15. Categorical CPLM – contracts with Michigan’s counties

Maintenance LM Probability Mean Std.dev. Transformation Function and
1,2 97 80.42 84.44 Box-Cox: λ=0.234, μ = 2.514, SC=0.696
Johnson: μ=0.021, σ=0.985
3 96 378.55 423.30
Pavements 1.116+0.533 Ln((X+1.25)/(1821.7-X))
Johnson: μ=-0.019, σ=0.958
4 97 682.75 641.01
1.007+0.63 Ln((X+17.06)/(2920.6-X))
1,2,3 94 163.31 130.29 No Fit
Shoulders Johnson: μ=0.003, σ=0.955
4 73 38.45 36.65
1.336+0.693 Ln((X-0.387)/(201.25-X))
Johnson: μ=-0.026, σ=1.009
1,2 89 72.94 73.32
1.031+0.595 Ln((X+0.613)/( 318.5-X))
Johnson: μ=0.003, σ=1.013
Drainage 3 93 120.44 108.05
1.104+0.709 Ln((X+3.18)/(508.7-X))
Johnson: μ=0.014, σ=0.998
4 94 241.70 198.89
0.933+0.702 Ln((X+1.777)/( 876.8-X))
1,2 98 137.64 79.13 No Fit
Johnson: μ=0.028, σ=0.978
Vegetation 3 95 264.27 176.75
0.602+0.756 Ln((X+3.699)/(770.9-X))
Control
Johnson: μ=-0.11, σ=0.995
4 90 466.11 378.82
0.979+0.84 Ln((X+45.42)/(1884.7-X))
Johnson: μ=0.028, σ=0.978
1,2 98 82.34 61.39
0.943+0.783 Ln((X+2.36)/(291.97-X))
Litter 3 94 187.00 148.13 Box-Cox: λ=0.5, μ = 12.38, SC=5.82
Johnson: μ=-0.047, σ=0.984
4 98 501.20 440.67
1.581+0.802 Ln((X-25.86)/(2878.7-X))
Johnson: μ=-0.014, σ=1.0
1,2 90 37.53 30.47
0.855+0.701 Ln((X+0.861)/(132.6-X))
Johnson: μ=0.023, σ=0994
Signs 3 98 73.94 51.12
1.297+1.053 Ln((X+8.23)/( 302.91-X))
Johnson: μ=0.001, σ=0.952
4 98 109.70 75.26
0.811+0.757 Ln((X+0.958)/( 363.5-X))
Johnson: μ=0.011, σ=0.987
1,2 26 9.68 9.24
1.211+0.661 Ln((X-0.059)/(45.442-X))
Markings
Johnson: μ=0.013, σ=0.946
3,4 48 19.73 16.60
0.850+0.643 Ln((X+0.421)/(71.21-X))
Johnson: μ=0.029, σ=1.001
1,2 65 29.52 29.39
1.155+0.628 Ln((X+0.376)/(134.9-X))
Johnson: μ=-0.061, σ=1.002
Guardrail 3 72 116.98 127.55
1.02+0.552 Ln((X+0.765)/(533.82-X))
Johnson: μ=0.024, σ=0.903
4 87 343.06 370.78
0.938+0.462 Ln((X-0.559)/(1525.6-X))
33
3.2 Ohio’s Maintenance Costs per Lane Mile
3.2.1 Maintenance
Activity Categories 1000
and Lane Mile (LM) 900
Responsible Lane Miles

800
Groups
700
The Ohio county cost data were 600
assigned to lane mile (LM) groups
500
based on responsible lane miles.
400
Figure 17 shows the range of lane
miles in each LM group. 300
Ohio data included 88 counties 200

over the three year period. Table 1 2 3 4 5
LM Group - Ohio's Counties
16 lists the maintenance activities
in eight categories.
Figure 17. County lane mile (LM) groups - Ohio
3.2.2 Cost Trends for LM

1 2 3 4 5
Guardrail $/LM Pavement Def $/LM Groups
3000
800
2000
The box plots in Figure 18 show the
600
400
ranges of CPLM for the eight
200
1000 maintenance categories and five LM
0 0 groups. The plots show some
Pavement DOff $/LM Mowing&Veg $/LM
2000 4800 variations in expenditures across the
1500 3600 LM groups. The cost per lane mile
1000 2400 for litter pickup shows an increasing
500 1200 trend as the responsible lane miles
0 0 increase. The plots for pavement
1 2 3 4 5
drop off, and mowing and vegetation
LM Group - Ohio Counties
maintenance categories show an
inverse relationship. As the number
Litter $/LM
1 2
Drainage $/LM
3 4 5
of responsible lane miles increases
2000
4000
the maintenance CPLM decreases.
3000
1500
1000
2000
Figure 18. Cost per lane
500 1000
mile trends for LM
0
Sign $/LM
0
Pavement Marking $/LM
groups – Ohio’s Counties
1000
600
750
450
500 300
250 150
0 0
1 2 3 4 5
LM Group - Ohio's Counties
34
Table 16. Maintenance categories for analysis of Ohio’s maintenance costs
Category Code Description
6230 End Assembly Installation and Maintenance
6233 Repair, Replacement, or Removal
Guardrail 6235 Crash Attenuator Repair or Replacement
6237 Concrete Median Barrier Repair, Replacement, Restoration, Removal
6333 Guardrail Betterment
6120 Pavement Underseal/Fill Voids
6121 Pothole Patching
6122 Surface Repairs
6123 Full Depth Repair
6124 Filling and Sealing Joints and Cracks
Pavement
6125 Surface Treatment
Deficiency
6126 Pavement Jacking
6127 Planning Bituminous Pavement
6129 Surface Paving
6136 Partial Depth Repair
6321 Roadway Betterment
6130 Spot Berming
Pavement Drop 6131 Blading-Restoring Unpaved Berm and/or Shoulders
Off 6132 Repairing Curbs and or Gutters
6331 Berm and/or Shoulder Betterments
6220 Weed Eating
Mowing and
6221 Mowing
Vegetation
6222 Chemical Spraying Vegetation Management
Control
6223 Care of Shrubs, Plants and Trees
6231 Cleaning Curbs, Gutters & Along Median Barriers
Litter 6232 Litter Pickup
6236 Litter Patrol
6133 Repairing Slopes
6134 Repairing Slips and Slides
6135 Ditch & Shoulder Relocation
6140 Culvert Inspection
6141 Cleaning & Reshaping Ditches
Drainage 6142 Cleaning Channels
Obstruction 6143 Cleaning Drainage Structures
6144 Repairing Drainage Structures
6145 Cleaning Ditches with Tiger Ditcher
6146 Inspection and Cleaning Underdrain Outlets
6343 Culvert Betterment
6344 Catch Basin Repair, Replace, or Install
6521 Ground-Mounted Flat Sheet Sign Maintenance
6522 Delineator Maintenance
Sign 6523 Ground Mounted Extrusheet Sign Maintenance
Deterioration 6524 Overhead Mounted Extrusheet Sign Maintenance
6525 Overhead Sign Support Maintenance
6554 Overhead Mounted Flat Sheet Sign Maintenance
6531 Auxiliary Pavement Markings
6532 Center Line Marking and Tee Marking
Pavement 6533 Edge Line Marking
Marking 6534 Lane Line Marking
6535 Raised Pavement Markers Maintenance
6536 Pavement Marking Removal
35
show the annual 2004 2005 2006
CPLM for the Litter $/LM
4000
Drainage $/LM
maintenance 2000
3000
categories over the 3- 1500
year study period. 1000
2000
Expenditures for 500 1000
most are constant 0 0
with the exception of 1000
Sign $/LM Pavement Marking $/LM
signs and pavement 600

750
markings. For these 450
the statistical analyses 500 300
indicate the mean 250 150
CPLM differs across 0 0
the years. 2004 2005 2006
Figure 19. Annual Fiscal Year - Ohio's Counties

trends for cost per
lane mile – Ohio’s 2004 2005 2006
Counties Guardrail $/LM Pavement Def $/LM
3000
Table 17 summarizes 800
the results of the 600 2000
analyses of mean 400
CPLM across LM 200
1000
groups and fiscal 0 0
years. Groups and Pavement DOff $/LM Mowing&Veg $/LM
2000 4800
years listed in this
table have significant 1500 3600
differences in mean 1000 2400
CPLM when 500 1200
compared to the
0 0
overall mean of the 2004 2005 2006
category. Fiscal Year - Ohio's Counties
Table 17. Analysis of the mean CPLM for LM groups and fiscal years – Ohio (95% CI)
Mean CPLM differs from population mean for LM groups and year:
Maintenance Category LM Group Year
Guardrail No difference No difference
Pavement Deficiency No difference No difference
Pavement Drop Off 2, 4, 5 No difference
Mowing and Vegetation Control 3, 4, 5 No difference
Litter 4, 5 No difference
Drainage Obstruction No difference No difference
Sign Deterioration 1 2005, 2006
Pavement Markings No difference 2004, 2006
36
Table 18 lists the observed means and standard deviations for each of the 8 maintenance
categories and provides the best fit distribution function for each category. The sample size for
each category was the 264 cost records. The transformation functions fit the observed data to a
normal distribution with mean of 0 and standard deviation of 1. In this table μ and σ are the
mean and standard deviation of the transformation functions. Normal distribution functions are
desirable for statistical simulation and sensitivity analysis.
Table 18. Categorical CPLM and distribution functions – Ohio
Maintenance LM Probability Mean Std. dev. Transformation Function and
Guardrail 1,2,3,4,5 98 192.50 173.28 No Fit
Pavement Johnson: μ=-0.038, σ=0.994
1,2,3,4,5 100 988.52 478.26
Deficiency 4.126 + 2.24 Ln((X + 271.8)/8562.1 - X))
Johnson: μ=-0.01, σ=1.029
1,2,3 100 611.70 307.58
2.217+1.435 Ln((X - 48.69)/(2912.8 - X))
Pavement
4 100 464.39 185.15 Box Cox: λ=0.5, μ=21.119, σ=4.336
Drop Off
Johnson: μ=0.006, σ=1.014
5 100 227.42 82.46
-7.961 + 1.674*Ln(X - 89.5265)
Johnson: μ=0.024, σ=0.933
1,2,3 100 896.54 450.78
Vegetation -18.392 + 2.653 Ln(X + 206.879)
Control Johnson: μ=0.011, σ=1.01
4,5 100 615.57 315.27
0.332 + 0.602 Ln((X - 166.5)/(1272.8-X))
Johnson: μ=0.049, σ=0.835
1 100 664.14 597.39
-0.622 + 0.367 Asinh((X - 306.72)/22.06)
Litter Johnson: μ=-0.007, σ=1.036
2,3,4 100 463.77 257.31
-16.777 + 2.633 Ln(X + 166.441)
5 100 1232.28 652.69 Box Cox: λ=0.5, μ=33.87, σ=9.55
Drainage Johnson: μ=-0.024, σ=1.049
1,2,3,4,5 100 936.28 585.98
Obstruction -2.874 + 1.684 Asinh((X - 131.7)/252.1)
Johnson: μ=-0.055, σ=0.948
1 100 363.34 241.35
Sign 1.034+0.79 Ln((X - 57.42)/(1277.2 - X))
Deterioration Johnson: μ=0.043, σ=1.047
2,3,4,5 100 251.94 130.12
-19.832+3.352*Ln(X + 142.95)
Pavement Johnson: μ=-0.030, σ=1.081
1,2,3,4,5 98 95.92 89.86
Markings 2.375 + 1.072 Ln((X + 6.8) / (786.1 - X))
3.3 Wisconsin’s Maintenance Costs per Lane Mile
3.3.1 Maintenance Activity Categories and Lane Mile (LM) Groups

The Wisconsin data set contains annual maintenance costs for various activities for the state’s 72
counties over three a year period. Table 19 lists the activities that were grouped into nine
categories. Cost data for Wisconsin is divided into labor, benefits, equipment and materials cost.
The overhead cost was removed for this analysis. All highway maintenance is performed by the
counties through contracts with the state. For this analysis, cost is analyzed on a county level.
For Wisconsin, pavement maintenance costs are organized by surface type. The researchers kept
them in separate categories.
37
Table 19. Categories for analysis of Wisconsin’s maintenance costs
Category Code Description
1 Spot Repair / Pothole Repair
2 Crack Sealing / Filling
3 Sealcoating
Roadways - Asphalt
4 Wedging / Rut Filling
5 Milling
8 Thin Resurfacing
11 Emergency Concrete Repair
Roadways - Concrete 12 Non-Emergency Concrete Repair
13 Repair of Distressed Pavement
21 Grading Gravel Shoulders
Roadways - Shoulders
22 Repairing Paved Shoulders
31 Sweeping Pavement
Litter
42 Litter Pickup
41 Mowing
Roadsides - Vegetation 43 Woody Vegetation
44 Noxious Weed Control
51 Clean / Repair Drainage Structure
Drainage
52 Maintain Roadside Drainage
Guiderails 55 Maintain Safety Appurtenances
81 Permanent Sign Repair
Sign Repair
85 Emergency Sign Repair
Pavement Marking 90 Pavement Marking
The counties were assigned to lane mile (LM) groups according to the number of responsible
lane miles. Figure 20 shows the ranges of lane miles in each of the LM groups.
2000
LM Responsible Lane Miles
Group Samples Min Mean Max
1 20 84.62 140.49 216.07
Responsible Lane Miles
1500 2 72 221.68 198.71 377.95

3 76 383.70 461.43 558.75
4 42 563.9 660.4 1027.1
5 6 1329.7 1480.2 1777.0
1000
500
0
1 2 3 4 5
LM Group - Wisconsin's Counties
Figure 20. Lane mile (LM) groups – Wisconsin’s counties
38
3.3.2 Cost Trends for LM Groups
1 2 3 4 5
Asphalt $/LM Concrete $/LM
1200
400
900 300
600 200
300 100
0 0
Shoulders $/LM Litter $/LM
600 1600
450 1200
300 800
150 400
0 0
1 2 3 4 5
1 2 3 4 5
Vegetation $/LM Drainage $/LM Guiderails $/LM
600 200
600
450 450 150
300 300 100
150 150 50
0 0
0
Sign $/LM Pav ement Mark ings $/LM 1 2 3 4 5
2000
100
1500
75
1000
50
25 500
0 0
1 2 3 4 5
Figure 21. Wisconsin’s cost per lane mile for maintenance categories
and lane mile groups
The box plots in Figure 21 show the ranges of CPLM for the nine maintenance categories and
five LM groups. For some maintenance categories the CPLM varies across the LM groups
indicating possible economies of scale. For the litter, drainage and guiderails categories, the
CPLM increases with increasing lane miles. However, the CPLM for the pavement markings
category shows the opposite pattern. For pavement markings, the CPLM decreases with
increasing lane miles.
Overall the mean CPLM does vary among the LM groups. The variance for LM group 5 is
higher than for the other LM groups. This could be explained by the small number of samples;
only six data samples. Two counties, Dane and Milwaukee, are in LM group 5.
39
show Wisconsin’s
2004 2005 2006 categorical CPLM for
Asphalt $/LM Concrete $/LM
each year of the study
1200 400 period. Expenditures
900 300 for all maintenance
600 200 categories do not
300 100 appear to differ from
0 0 year to year. Although,
Shoulders $/LM Litter $/LM
600 1600 results of the analysis of
450 1200 the mean CPLM for
300 800
asphalt pavements,
drainage, and sign
150 400
repair indicates some
0 0 differences.
2004 2005 2006
Fiscal Year - Wisconsin's Counties Figure 22. Annual
trends for cost per
2004 2005 2006
lane mile for
Vegetation $/LM Drainage $/LM Guiderails $/LM contracts with
600 200
600 Wisconsin’s
450 450 150 counties.
300 300 100
The statistical analysis
150 50
150 of the mean across LM
0
Sign $/LM
0
Pav ement Mark ings $/LM
0 groups and the fiscal
100
2000 2004 2005 2006 years in Table 20 does
1500 not consider LM group
75
1000
5 since the number of
50
samples is small. The
500
25 groups and years listed
0 0 in the table have
2004 2005 2006
significant difference in
Fiscal Year - Wisconsin's Counties
means when compared
to the overall mean CPLM for
Mean CPLM differs from population
mean for LM groups and years the maintenance category.
Maintenance Category LM Group Year
Roadways-Asphalt 3 2004, 2005 Table 20. Analysis of the
Roadways-Concrete X X mean CPLM for LM
Roadways-Shoulders 2 X groups and years –
Litter 2, 3 X Wisconsin (95% CI).
Roadside-Vegetation X X
Drainage X 2004
Guiderails 1, 2, 4 X
Sign Repair 2, 4 2006
Pavement Markings 1, 4 X
40
Table 21 lists the observed means and standard deviations for each of the nine maintenance
categories and the best fit distribution function for each category. The transformation functions
fit the observed data to a normal distribution with mean of 0 and standard deviation of 1. In
this table μ represents the new normalized mean and σ represents the new normalized standard
deviation.
The pair-wise comparison was done only for LM groups 1, 2, 3, and 4. Group 5 was excluded
since it has only 6 samples from 2 counties. And in the report we did ANOVA only for 1,2,3,4
Table 21. Categorical CPLM and distribution functions – contracts with Wisconsin’s counties
LM Probability Mean Std. dev. Transformation Function and
Roadways- Johnson: μ =-0.009; σ=0.934
1,2,3,4 100 345.39 217.41
Asphalt -14.704+2.38*Ln(X + 174.13)
Roadways-
1,2,3,4 71 66.58 90.33 Box-Cox:: λ=0.181,μ =1.829, σ=0.567
Concrete
Johnson: μ =-0.025; σ=1.014
1,3,4 100 156.72 82.18
Roadways- 6.75+2.34 Ln((X+38.01)/(3389.96-X))
Shoulders Johnson: μ =0.022; σ=0.983
2 100 207.71 104.68
-1.72+1.461 Asinh((X-76.142)/71.029)
Johnson: μ =0.007; σ=1.043
1,2 100 81.34 58.89
1.864+0.989 Ln((X-6.925)/(437.6-X))
Litter
Johnson: μ =-0.024; σ=1.04
3,4 100 186.38 219.87
-4.948+1.044 Ln(X-4.067)
Roadside- Johnson: μ =-0.04; σ=1.01
1,2,3,4 100 248.42 104.9
Vegetation 4.432+2.308*Ln((X+35.85)/(2092-X))
Johnson: μ =0.294; σ=0.962
Drainage 1,2,3,4 100 113.04 101.56
-1.607+1.169 Asinh((X-16.717)/35.85)
1,2 92 25.66 30.82 Box-Cox:: λ=0.174,μ = 1.593, σ=0.364
Guiderails Johnson: μ =0.294; σ=0.962
3,4 99 46.11 35.91
-4.792+1.321 Ln(X+2.248)
Johnson: μ =-0.031; σ=1.03
1,4 100 22.75 17.75
0.812+0.717 Ln((X+0.616)/( 78.09-X))
Sign Repair
Johnson: μ =0.027; σ=1.035
2,3 100 35.41 21.92
-13.778+3.305*Ln(X+33.044)
Johnson: μ =-0.035; σ=1.147
1,2 16 632.57 612.25
Pavement 0.742+0.461*Ln((X-54.1)/( 2051.6-X))
Markings Johnson: μ =-0.065; σ=0.972
3,4 33 141.16 127.71
-2.298+0.886*Asinh((X-18.01)/11.53)
41
Chapter 4. MAINTENANCE CONDITION ASSESSMENT
This chapter contains brief overviews of the maintenance condition assessment methods at each
state. Michigan uses a sufficiency ratings assigned to each segment of roadway. Ohio uses a
system that measures countywide deficiencies for a range of features. Wisconsin’s system,
somewhat similar to Ohio’s, estimates the percentage of assets backlogged for maintenance.
4.1 Michigan’s Maintenance Quality Assurance Program

Michigan’s maintenance quality assurance program (MQA) is overseen by Michigan’s
Transportation Asset Management Council. The Council is responsible for tracking the overall
condition of roads and bridges, and the spending of allocated public dollars.
The Michigan DOT provided sufficiency ratings for pavements, shoulders, and drainage
conditions for roadway sections for the years 2004, 2005, and 2006. The data is reported in lane
miles. The sufficiency rating is determined from an annual “windshield survey” of the state’s
43,000 miles of federal-aid eligible roads. Sufficiency ratings are based on observed road
conditions, including cracking, potholes, sinking, drainage, and rutting. Table 22 to Table 24 list
the sufficiency ratings for the predominant pavement surface, drainage, and shoulder condition
for the segments. Bridge decks are not included. Intermediate codes (1.5, 2.5, 3.5 and 4.5) could
be used to describe surface conditions that lie between those defined in the table. The terms
occasional, frequent, or continuous refer to the time and resources spent to maintain the surface
so that its condition is as good as the deterioration will permit.
Table 22. Michigan’s sufficiency rating system for pavement surface and joint condition
Rating Condition Description
1 Excellent: No deterioration. Distresses are non-existent.
Good: Some indication of initial deterioration, but not yet requiring appreciable amounts of
2 maintenance. Distress items may include the start of transverse and/or longitudinal cracks. Slight
rutting may be apparent in the wheel path.
Fair: Average deterioration requiring occasional routine maintenance. Distresses may include minor
3 cracks becoming continuous throughout the segment. Severe cracking has been patched or sealed but
may start to show through patching.
Poor: Excessive deterioration requiring frequent maintenance warrants repair or resurfacing soon.
4 Distress may be evident in widening cracks and faulting joints. Cracks could be breaking through
patches continuously.
Very poor: Extreme deterioration, requiring continuous maintenance, warrants repair, resurfacing, or
5 total replacement. Distress items may include severe transverse and/or longitudinal cracking. Joints are
failing. Rutting in wheel path may be severe and patches are no longer beneficial to pavement condition.
Table 23. Michigan’s sufficiency rating system for drainage condition
Excellent: Excellent cross section design, ditch slopes 1:4 or flatter, ditches deep enough to carry very
1
heavy rainfall. Curb and gutter sections have catch basins at all curb returns and usually at mid-block.
Good: Good cross section design, ditch slopes 1:4 or better, ditches can carry heavy rainfall. Curb and
2
gutter sections have catch basins at the curb returns.
Fair: Cross section usually narrow, steep ditch slopes, ditches intermittent and not very deep. Curb and
3
gutter sections do not have sufficient catch basins. Rutting may cause hydroplaning.
Poor: Cross section narrow, grade is at natural ground level or slightly above, ditches intermittent and
4
shallow. Curb and gutter sections have few catch basins.
Very poor: Ground level higher than roadway and no ditches. Curb and gutter sections have no catch
5
basins. Rutting could cause severe hydroplaning.
43
Table 24. Michigan’ sufficiency rating system for shoulder condition
Excellent: Curbs are structurally in very good condition and curb heights are adequate for more than one
1
resurfacing. Shoulders show no visible or apparent deterioration of surface.
Good: Curbs and shoulders show some deterioration. Curb heights may be adequate for more than one
2
resurfacing. Cracking on paved shoulders is minor.
Fair: Curbs in average condition and height may be adequate for one resurfacing. Shoulders show
3 average deterioration that may require occasional maintenance. Cracking on paved shoulders is
continuous and edge breakaway may be evident. Lane/Shoulder drop may be apparent.
Poor: Structural condition of curbs is poor and curb heights are inadequate for resurfacing. Paved
4 shoulders show frequent deterioration and may require continuous maintenance. Cracks are continuous
and edges are breaking away. Lane/Shoulder drop may affect driver reaction.
Very Poor: Structural condition of curbs is very poor and curb heights are inadequate for resurfacing.
5 Shoulders show extreme deterioration and are probably beyond normal maintenance repair. Pavement
pop-outs are continuous and shoulder edge is broken away. Lane/Shoulder drop off may be severe.
While cost data are available for a wide range of maintenance activities, condition data are
limited to the pavement, shoulder and drainage maintenance categories. Thus, only a subset of
the cost activities codes is applicable. Table 25 shows the mapping and scope of cost and
condition used in the analysis.
Table 25. Mapping cost to condition categories for Michigan.
Condition Contract Maintenance Direct Forces
Maintenance Activity Activity
Features Description Description
Category Code Code
10100 Joint & Crack Filling
Remove/Replace Pavement (Fast
Surface & 10200
Surface Set Concrete)
Pavement Joint 109
Maintenance 10300 Patrol Patching
Condition
10500 Bituminous Maintenance & Repair
10800 Bump Removal
11000 Routine Blading
Shoulder
Shoulder Shoulder 119 11200 Gravel Shoulder Maintenance
Maintenance
11400 Paved Shoulder Maintenance
Drainage &
122 Back Slopes 12200 Catch Basin Cleanout
Drainage
Maintenance
&
Drainage 123 Ditch Cleaning 12300 Ditch Cleanout
Subbase
Culvert/Under 12800 Culvert & Underdrain Cleaning
Condition
128 Drain
12810 Culvert & Underdrain Cleaning
Maintenance
4.2 Ohio’s Maintenance Quality Survey (MQS) Program

This study uses Ohio’s Maintenance Quality Survey (MQS) data for each county. The MQS
program routinely assesses the condition of all highways under the jurisdiction of each district
and county. The MQS is performed by two 2-person crews working out of the DOT’s Office of
Maintenance Administration. The crews survey one quarter of each county’s state-maintained
highways every three months. MQS deficiencies are collected via touch-screen laptop computers
utilizing GPS technology (3).
The MQS data are used by the District Highway Management Administrators and County
Managers to allocate their available resources on their County Work Plans. The MQS records
maintenance deficiencies for eight categories: guardrail, pavement deficiency, pavement drop-
44
off, vegetation obstruction, litter, drainage ditch obstruction, sign deterioration, and pavement
marking as shown in Table 26. The deficiency standards describe the characteristics for
determining whether features are deficient. The deficiency count indicates the unit for counting
the number of deficiencies. Total deficiencies are compiled for each County and District, as well
as statewide from the most recent rolling four quarters by the Office of Maintenance
Administration. MQS furnishes a basis against which the success of maintenance efforts may be
measured.
In addition, the department uses the Organizational Performance Index (OPI) to evaluate
employee performance. The OPI system grades maintenance efforts according to the ranges of
total deficiencies per lane mile as determined from the MQS data. The goal of OPI is to ensure
the highest level of service possible with the resources available. OPI grades for the counties are
fairly constant throughout the 3-year analysis period while the MQS deficiency counts are not.
For Ohio, maintenance cost and condition are related at the category level because they are
consistently defined in both the Ellis and MQS systems. This means the sum of the deficiency
counts for all features in a maintenance category is related to the sum of the expenditures for all
activities in that category. As an example, for guardrails, the sum of the deficiencies for rails,
posts, spacer blocks, concrete medians, bridge parapets, anchor assemblies, end terminals, and
crash attenuators is related to the sum of the expenditures for activities codes 6230, 6233, 6235,
6237, and 6333.
4.3 Wisconsin Maintenance Quality Assurance Program

In the state of Wisconsin maintenance condition is documented in Compass (1), the quality
assurance and asset management program for highway operations. Maintenance condition is
cataloged according to elements and features. A maintenance element is defined as a logical
grouping of features based on location or function along a highway. Examples of elements
include asphalt and concrete pavements, shoulders, and traffic management. (An element is similar
to a category used by other states.) A maintenance feature is defined as a physical asset or activity
and its condition is measured in the field. There are one or more maintenance features for each
element.
The state conducts an annual Compass survey of randomly selected roadway segments. The
sample size is approximately 1% of the state-maintained highways. If a feature is in a condition
at which it requires maintenance in the next 12 months, then it is designated as “backlogged”.
Table 27 lists the elements, their corresponding features, and the condition thresholds for being
backlogged.
For continuous features measured by the mile, the threshold value includes both sides of the
roadway for a mile segment. Condition information for asphalt and concrete pavements are
taken from the agency’s PMMS (pavement maintenance management system). State-maintained
highway pavements in Wisconsin are inspected on a two-year cycle, with half of the state’s
pavements inspected in one year, and the other half in the next year. Compass and the PMMS
are the data sources for maintenance condition in this study.
For this research, we strived to relate cost and condition at the lowest level possible. Table 28
shows the relationship between HMS cost codes and Compass elements/features as per the
HMS cost code descriptions found in Exhibit 6.07 of the Wisconsin State Highway Maintenance
Manual (3) and Compass feature descriptions.
45
The problems encountered when relating cost to condition result from the granularity of the
information available. The relationships are either one-to-one, many-to-one, or many-to-many.
The ideal situation is a one-to-one direct relationship between one HMS cost code and one
Compass feature leaving little room for ambiguity. The two examples of one-to-one
relationships in Table 28 are noxious weeds and ditches.
One-to-many relationships have one HMS cost code mapped to several Compass features.
Relating one cost to maintenance of many features requires that we somehow combine
condition measures. Equal weight was assumed among the measures. Other assumptions are
independence of features (condition of one features does not influence the condition of another)
and equal number of segments sampled from each feature. Examples of one-to-many
relationships listed in Table 28 include the unpaved shoulders, drainage structures, and safety
appurtenances.
Finally, many-to-many relationships between HMS cost codes and Compass features require
more assumptions. Many-to-many relationships have more than one HMS cost code mapped to
several Compass features. Assumptions include independence of features, equal number of
segments sampled from each feature, and equal distribution of costs among the features. Two
examples Table 28 in are asphalt and concrete pavements where multiple HMS costs codes
relate to multiple features in the Compass elements. For these elements condition data is
complete (not sampled). Therefore, the sampling assumption can be ignored.
46
Table 26. Ohio’s maintenance quality survey (MQS) criteria.
Maintenance
Feature Deficiency Standard Deficiency Count
Category
50% crushed, 50% torn, wrong height, or missing Each rail panel
Rail
Cable touching the ground Every 100 linear feet of cable rail system
2 continuous rotten or missing posts excluding cable rail
Post Every two continuous posts
posts
Spacer block Rotten, missing, or rotated block-out Every 3 continuous block-outs
Guardrail
Concrete median
144 sq. inches of cross section missing Every 12.5 feet of barrier
and bridge parapet
Anchor assembly,
Each or 2 deficiencies per terminal or
end terminal, crash Damaged or missing
attenuator
attenuator
Pothole > 2 inch deep and 144 sq. inches in area Each 6' x 6' area of adjacent paved surfaces
Pavement
Rutting > 2 inch deep within wheel tracks Each 1/10th of a lane mile
deterioration
Shoving > 2 inch deep and 4' x 6' in area Each 6' x 6' area of adjacent paved surface
Pavement Blow-out > 2 inch in height and 6 feet in width Each
Pavement Sag/slip > 2 inch deep, 15 linear feet, and 6 feet wide Each
obstruction Manhole or inlet > 2 inch above or below the paved
Each 6' x 6' area of adjacent paved surface
surface
Vegetation Each sign, guardrail, end treatment, or 100
Vegetation Vegetation obscuring signage or guardrail
Obstruction linear feet of guardrail, or each guardrail run
Litter Exceeding 10 countable litter items Every 1/10th of a mile
Litter
Large litter 8 inches high and 3' x 3' in area Each
≥ 50% filled, standing water ≥ 1 inch deep, covering ≥ 6
Drainage feet of the paved surface for 10 linear feet, or standing Every 100 linear feet of ditch, paved surface,
Ditch
Ditch water ≥ 1 inch deep covering the wheel track within a or highway lane
Obstruction traveled lane for 10 linear feet
Driveway pipe 50% or more of driveway pipe is silted in Each driveway pipe
Sign Damaged sign Sign cannot be clearly read Each sign
Deterioration Missing sign Missing sign Each sign
Edge lines and lane
Pavement > 150 linear feet missing or significantly faded Every 1/10th of a mile
lines
Marking
Auxiliary marking Missing or significantly faded Each stop bar or multiple element marking
47
Table 27. WisDOT compass elements, features, and thresholds (Compass 2005).
Element Feature Thresholds for BACKLOGGED
Pavement, Alligator cracking 10% or more of the surface has unsealed alligator cracking (within a mile)
asphalt Block cracking 10% or more of the surface has unsealed block cracking (within a mile)
Edge raveling Visible cracking is present for 10% or more of the mile
Flushing Flushing is present in more than small, isolated areas (within a mile)
Longitudinal cracking Any unsealed longitudinal cracking (within a mile)
Longitudinal distortion Significant distortion affects 1% or more of roadway (within a mile)
Patch deterioration Any patch is deteriorated enough to affect ride quality (within a mile)
Rutting Ruts are ¼ inch or deeper (within a mile)
Surface raveling The aggregate and/or asphalt binder has worn away and the surface texture is rough or pitted (within a mile)
Transverse cracking Any unsealed transverse cracks at least 6’ in length (within a mile)
Transverse distortion Significant distortion affects 1% or more of roadway (within a mile)
Pavement, Distressed joints/ cracks Distress in wheel path greater than 2 inches wide (within a mile)
concrete Longitudinal joint distress Faulting or signs of distress are present (within a mile)
Patch deterioration Any patch is deteriorated enough to affect ride quality (within a mile)
Slab breakup Slab is divided into at least 2-3 large blocks, affecting 10% or more of the slab (within a mile)
Surface distress Any measurable surface distress is present (within a mile)
Transverse faulting Any measurable faulting (within a mile)
Traffic Centerline/edgeline markings Line with > 20% paint missing (within a mile)
Control & Delineators Missing OR not visible at posted speed OR damaged (by delineator)
Safety Protective Barriers Not functioning as intended (linear feet of barrier)
Other signs (emergency) Missing OR not visible at posted speed (by sign)
Other signs (routine) Beyond service life (by sign)
Raised Pavement Markings Missing OR damaged (by RPM)
Regulatory/ warning signs Missing OR not visible at posted speed (by sign)
Reg./warn. signs (routine) Beyond service life (by sign)
Special Pavement Markings Missing OR not functioning as intended (by marking)
Shoulders Cracking 200 linear feet or more of unsealed cracks > ¼ inch (by mile)
Cross-slope 200 linear feet or more of cross-slope at least 2x planned slope with the maximum cross slope of 8% (by mile)
Hazardous Debris Any items large enough to cause a safety hazard (by mile)
Drop-off/ buildup 200 linear feet or more with drop-off or build-up > 1.5 inches (by mile)
Erosion 200 linear feet or more with erosion >2 inches deep (by mile)
Potholes/ raveling Any potholes OR raveling > 1 square foot by 1 inch deep (by mile)
Drainage Culvert Culverts that are >25% obstructed OR where a sharp object-e.g., a shovel-can be pushed through the bottom of the
pipe OR pipe is collapsed or separated (by culvert)
Curb & gutter Curb & gutter with severe structural distress OR >1 inch structural misalignment OR >1 inch of debris build-up in the
curb line (by linear feet of curb & gutter)
Ditches Ditch with greater than minimal erosion of ditch line OR obstructions to flow of water requiring action (by linear feet of
ditch)
48
Element Feature Thresholds for BACKLOGGED
Flumes Not functioning as intended OR deteriorated to the point that they are causing erosion (by flume)
Storm sewer system Inlets, catch basins, and outlet pipes with >=50% capacity obstructed OR <80% structurally sound OR >1 inch vertical
displacement or heaving OR not functioning as intended (by inlet, catch basin & outlet pipes)
Drains Under- and edge-drains with outlets, endwalls or end protection closed or crushed OR water flow or end protection is
obstructed (by drain)
Roadsides Barriers Noise barrier or retaining wall not functioning as intended (by LF of barrier)
Fences Fence missing OR not functioning as intended (by LF of fence)
Litter Any pieces of litter on shoulders and roadside visible at posted speed, but not causing a safety threat. (by mile)
Graffiti Any graffiti and non-natural encroachments visual at posted speed. (by mile)
Mowing Any roadside has mowed grass that is too short, too wide or is mowed in a no-mow zone (by mile)
Mowing vision Any instances in which grass is too high or blocks a vision triangle (by mile)
Noxious weeds Any visible clumps (by mile)
Woody vegetation Any instances in which woody vegetation blocks a vision triangle (by mile)
Woody vegetation vision Any instances in which a tree is present in the clear zone OR trees and/or branches overhang the roadway or shoulder
creating a clearance problem (by mile)
49
Table 28. HMS cost activity codes mapped to Compass elements/features.
Relationship HMS
(cost-to- CostC
condition) Partial Element Compass Element: Features ode HMS Cost Code Description
Noxious Weed Roadside: noxious weeds 44 Control of unwanted vegetation
One-to-One
Ditches Drainage: ditches 52 Maintain roadside drainage
Unpaved Shoulders Unpaved Shoulders: cross-slope, drop-off/build-up, erosion 21 Gravel shoulders
Paved Shoulders Paved Shoulders: cracking, potholes/raveling 22 Paved shoulders
Mowing Roadside: mowing, mowing for vision 41 Mowing
Roadside: litter
Litter Pickup 42 Litter pickup
Shoulders: hazardous debris
Woody Vegetation Roadside: woody vegetation, woody vegetation control for vision 43 Woody vegetation
Drainage: culverts, curb/gutter, flumes, storm sewer, under/edge
Drainage Structures 51 Clean/repair drainage structure
One-to-Many drains
Traffic: delineators, protective barriers
Safety Appurtenances 55 Maintain safety appurtenances
Roadsides: barriers, fences
Permanent Sign Repair Traffic: routine other signs, routine regulatory/warning signs 81 Permanent sign repair
Temporary/Emergency
Traffic: emergency other signs, emergency regulatory/warning signs 85 Temporary/emergency sign repair
Sign Repair
Traffic: centerline markings, edgeline, raised pavement markers, special 3881 Powerplay traffic program code
Pavement Markings
pavement markings 3882 Powerplay traffic program code
1 Spot repair/pothole repair
Traveled Way, Asphalt: alligator cracking, block cracking, edge raveling, 2 Crack sealing/filling
flushing, longitudinal cracking, longitudinal distortion, patch 3 Seal coating
Asphalt Pavement
deterioration, rutting, surface raveling, transverse cracking, transverse 4 Wedging/rut filling
distortion 5 Milling/bump removal
Many-to-Many 8 Thin resurfacing
Emergency repair of concrete
11
Traveled Way, Concrete: distressed joints/cracks, longitudinal joint pavement
Concrete Pavement distress, patch deterioration, slab breakup, surface distress, transverse Non-emergency repair of concrete
12
faulting pavement
13 Repair of distressed pavement
50
Chapter 5. MODELS RELATING COST TO CONDITION
5.1 Analysis Methodology

Relationships between cost and condition were investigated using statistical analysis and
there was no evidence of continuous function relationships. Instead, the researchers used a
regression tree approach that looks for a set of piecewise constant or linear estimates of
regression functions by recursively partitioning the data and sample space (6). Within each
tree, the models capture differences among the data observations depending upon
responsible lane miles, vehicle miles traveled, or other relevant variables.
After the best fit models were determined, the statistical program MINITAB was used to
identify and remove outliers in order to determine a better fit model. We defined outliers
among residuals as points that lay three or more standard deviations from the mean of all
residuals (8). Cook’s statistics, a methodology used for the detection of unusual observations
in a dataset, was used to determine and reject outliers.
The modeling effort explored the relationship between the change in cost and the change in
condition between two years. The analysis looked at county cost in consecutive years, cost
per lane mile, change in cost and change in cost per lane mile between two consecutive
years, condition, and change in condition between two consecutive years.
The regression tree analysis used the GUIDE modeling tool (2). Before running the
regression analysis, the researchers analyzed the correlation between each variable used to
understand how the variables influence each other. Each predictor variable in GUIDE is
assigned either categorical (c) used only for splitting, dependent (d), numerical used for both
splitting and fitting (n) or numerical used only for splitting (s).
5.2 Modeling Cost to Condition For Michigan

Scatter plots for cost per lane mile versus average condition for drainage, pavement and
shoulder elements showed no obvious trends between condition and cost or any trends in
condition over the 3 year period of 2004-2006. However, a regression tree analysis that
explored the relationship between cost and condition relative to other variables revealed
relationships for shoulder maintenance and culvert maintenance worthy of mention.
The following model equation predicts shoulder condition on a scale of 1 to 5 (according to
the rating system in Table 24) with an R2 of 0.61. R2, the coefficient of determination, is the
proportion of variability in a data that is accounted for by the statistical model.
ShdrCondC = 0.513 +0.124 (SubCondC) + 3.63(∆AADT) + 0.64(ShdrCondP)
ShdrCondC and ShdrCondP are the current and previous years’ shoulder condition. SubCondC is
the current years’ condition of the shoulder subbase. ∆AADT is the change in annual
average daily traffic between the current and previous year.
The model equation for predicting drainage condition has an R2 of 0.933. DrainCondC and
DrainCondP are the current and previous years’ average drainage condition. AADT is the
annual average daily traffic, and CPLMP is the previous year’s cost per lane mile for drainage
maintenance in 2006 dollars.
51
DrainCondC = – 0.0305 + 1.17E-3 (AADT) – 1.58 CPLMP + 1.01 DrainCondP
The above shoulder and drainage models were developed for the state of Michigan using
county level averages. Each state defines maintenance costs and conditions in slightly
different ways. The fit for Michigan is satisfactory. The models were not tested in other
locations and there is no reason to expect the models to be applicable outside of Michigan
where cost and condition are defined differently.
5.3 Modeling Cost to Condition For Ohio

Maintenance condition in Ohio is defined as total deficiency on a highway section. Total
deficiency per section was normalized to deficiency per lane mile for each county and cost
was adjusted for inflation then normalized by lane miles.
Parameters of the analysis are listed in Table 29. The analysis looked at relationships between
cost and conditions for each of the maintenance categories. We found model equations for
relationships between change in cost and change in condition for pavement, guardrail, and
litter maintenance. Reliable models could not be determined for the other maintenance
categories.
Table 29. Model parameters for guide analysis of cost and condition in Ohio.
Parameter Description and Units (each county)
TLMc Total Lane Miles
DEFc Total deficiencies
Current Cost &
DPLMc Deficiencies per lane mile
Conditions
COSTc Maintenance cost ($1000 in 2006)
CPLMc Maintenance cost per lane mile ($2006)
δCOST Change in total cost from previous year ($1000)
Change From δDEF Change in Total Deficiency (current year - previous year)
Previous Year δCPLM Change in cost per lane mile
δDPLM Change in Deficiency per lane mile from previous year
COSTp Change in cost per lane mile from previous year
DEFp Total deficiencies in previous year
Previous Costs &
TLMp Total Lane Miles (previous year)
Conditions
DPLMp Total Deficiency per lane mile for the previous year
CPLMp Total Cost per lane mile from the previous year
The models are summarized in Table 30. For pavement, guardrails and litter maintenance,
the fitted models estimate reductions in deficiency as expenditures increase. Figure 23 shows
the observed versus fitted change in deficiency per lane mile for these models.
Table 30. Models that relate maintenance cost to condition in Ohio.
Valid Sample
Element Equation R2
Conditions size
δDPLM = -0.0021 + 2.74E-8 δCOST + 2.83E-
Pavement All 0.94 176
8 COSTP – 0.818 DPLMP
Guardrail δDPLM = 0.018 – 0.64 DPLMP – 4.62E-5 CPLMP All 0.72 176
δDEF = -7.063 + 5.96E-4 δCOST – 0.704 DEFP +
Litter All 0.72 176
0.28 TLMP
52
0.10
0.05
0.00
-0.50 -0.40 -0.30 -0.20 -0.10 0.00 0.10
-0.05
Fitted δDPLM
-0.10
Pavement
Maintenance -0.15
-0.20
-0.25
-0.30
-0.35
-0.40
Observed δDPLM
-0.60 -0.50 -0.40 -0.30 -0.20 -0.10

-0.10
Fitted δDPLM
Guiderail
Maintenance -0.20
-0.30
-0.40
-0.50
Observed δDPLM
500
0.00
-2500 -2000 -1500 -1000 -500.00 0 500
-500
-1000
Fitted δDEF
-1500
Litter
Maintenance -2000
-2500
Observed δDEF
Figure 23. Observed deficiencies versus fitted models for Ohio
53
Ohio uses maintenance deficiency ranges from the MQS survey. Because the number of
deficiencies are grouped into ranges, the models for elements could predict that condition
deteriorate as maintenance costs increase. Maintenance OPI Index for Ohio is shown in
Table 31. For example, to obtain an OPI score of 4, in the pavement deficiency category, the
scale ranges from 0.65 to 0.86 deficiencies per lane mile. For any given year, the deficiency
could increase or decrease slightly and still fall under the OPI score of 4. Although the
deficiency counts change from year to year, the OPI scores may be consist.
Table 31. Deficiencies per lane mile ranges for Ohio’s Maintenance OPI index
Obstruction
Obstruction
Deficiency
Vegetation
Drop-Off
Sign Scale
Pavement
Pavement
Pavement
Guardrail
Drainage
Marking
Ditch
Litter
OPI
Scores
6 0.05 0.2 0.05 0.01 1.35 0.02 0.03 0.06
5 0.13 0.42 0.13 0.03 2.95 0.04 0.06 0.15
4 0.21 0.65 0.21 0.05 4.65 0.06 0.1 0.24
3 0.28 0.86 0.28 0.07 6.2 0.08 0.13 0.32
2 0.35 1.07 0.35 0.09 7.75 0.1 0.16 0.4
1 0.42 1.28 0.42 0.1 9.3 0.12 0.18 0.48
5.4 Modeling Cost to Condition For Wisconsin

Two types of relationships between cost and condition were explored. The first is the
relationship of fixed cost to fixed backlog. This relationship is based on cost and condition
in a single year. It allows for estimating the cost required to obtain a specific backlog or the
backlog that will result from a specific level of spending. The second relationship is based
on the changes in cost and condition between two years. It allows for estimating of change
in backlog given current backlog and budget constraints. It also allows for estimating the
budget adjustments required to achieve certain maintenance conditions. Finding these
relationships equations would give maintenance engineers and decisions makers the tools
they need for transportation maintenance budgeting.
Maintenance cost data by activity code at the county level along with maintenance condition
data by feature at the county level were provided by WisDOT. The data processing involved
first aggregating conditions and costs for each of the nine elements. The analysis is as
follows:
1. For elements with multiple features, determine if the features are independent of each
other in order to justify aggregating of condition backlog data.
2. For elements with multiple features, if features are independent then aggregate condition
backlog to determine backlog for the element at the county level.
3. For elements with multiple features, aggregate cost code data to determine total cost for
the element at the county level.
4. Analyze the relationship between (aggregate) condition and (aggregate) cost at the county
level using scatter plots and regression tree analysis and draw conclusions.
54
5.4.1 Aggregating Element Condition and Maintenance Costs
A feature is considered backlogged if distresses are more severe than a specific threshold. A
composite %backlog was computed for each element as the aggregate weighted backlog of
the component features. If the presence of one backlogged feature does not influence the
backlog of other features, then the features are independent. This aggregate value would not
be valid the features are highly correlated.
One method to evaluate the independence among variables is to examine the correlation of
backlogging among features of each element. Independence can be concluded if one distress
is not highly correlated to another. The correlation coefficient can range in value from -100
to +100, and tells two things about the linear relationship between two variables. The larger
the absolute value of the coefficient, the stronger the linear relationship between the
variables. A correlation value of 100 indicates a perfect linear relationship, and zero
indicates the absence of a linear relationship. The sign of the coefficient indicates the
direction of the relationship. If both variables tend to increase or decrease together, the
coefficient is positive. If one variable tends to increase as the other decreases, the coefficient
is negative.
The correlation matrix for distressed features in the unpaved shoulders and paved shoulders
in the fiscal years 2004-06 is shown in Table 32. The off-diagonal correlation percentages are
low; the largest is 29% between unpaved cross slope and unpaved drop. This result indicated
the features that comprise the unpaved and paved shoulders elements are independent.
Table 32. Correlation among features shoulder elements
Unpaved Shoulder Paved Shoulder
Drop Off Cross Slope Erosion Cracking Potholes
Drop Off 100
Cross Slope 29 100
Erosion 10 16 100
Cracking 100
Potholes 10 100
The correlation matrices for the remaining elements are shown in Table 33 through Table
35. The 88% correlation between transverse and longitudinal cracking in the element
Asphalt Travel Way is large enough to require attention. Since these two features are so
highly correlated, their backlog values were averaged and the averaged. If either or both
features were backlogged then the combined transverse cracking/longitudinal cracking
feature is backlogged.
Table 33. Correlation among features of mowing, litter pickup and woody vegetation
Mowing Hazardous Woody Wdy Veg. Cntrl
Mowing for Vision Litter Debris Vegetation for Vision
Mowing 100
Mowing for Vision 5 100
Litter 100
Hazardous Debris 18 100
Woody Vegetation 100
Woody Vegetation
Control for Vision 16 100
55
Table 34. Correlation among features of asphalt travel ways
Surface Raveling
Block Cracking
Edge Raveling
Deterioration
Longitudinal
Longitudinal
Transverse /
Transverse
Distortion
Distortion
Cracking
Cracking
Alligator
Flushing
Rutting
Patch
Alligator Cracking 100
Block Cracking -1 100
Edge Raveling 7 3 100
Flushing 0 2 0 100
Longitudinal
Distortion 2 0 -1 0 100
Patch
Deterioration 5 2 12 1 0 100
Rutting 2 2 4 4 0 7 100
Surface Raveling 7 2 6 0 2 4 1 100
Transv/Long
Cracking 13 15 6 2 2 3 8 5 100
Transverse
Distortion 4 1 1 0 0 0 1 0 1 100
Table 35. Correlation among features of concrete travel ways

Dist Joints Long. Joint Patch Slab Surface Transverse
/ Cracks Distress Deterioration Breakup Distress Distortion
Distressed Joints / Cracks 100
Longitudinal Joint Distress 17 100
Patch Deterioration 26 12 100
Slab Breakup 1 14 7 100
Surface Distress 18 2 21 -23 100
Transverse Distortion 48 10 50 0 29 100
After verifying that features are independent, the backlog values were aggregated to derive a
backlog for the element. Two methods of calculating element backlog were considered. They
are referred to as the “average method” and the “extent method.” Both methods compute
county level backlog.
The “average method” uses Compass backlogs for features at the county level to determine
element backlogs. Compass determines backlog by first separating condition data for each
segment into two groups. One group has backlog values of 0 (no backlog) and the other
group have backlog values greater than 0 (backlogged). Feature backlog is then calculated as
Number of Backlogged Segments
Feature Backlog 
Total Number of Segments
Since the features are independent, the aggregate average element backlog is the sum of the
feature backlogs divided by the number of features in the element. This average method is
simple and reasonable when there is approximately the same number of observations for
each of the features in the partial element.
 Backlog
Average %Backlog  Features
Total Number of Features
56
For ditches the %backlog for years 2005 and 2006 was calculated based on the linear feet of
ditch. The %backlog for ditched in 2004 were eliminated because the Compass data was
based on number of segments rather than linear feet of ditch.
Linear feet with more than minimal erosion
of ditch line or obstructions to the flow of
water requiring action
Ditches %Backlog 
Total linear feet of ditch line
Independence of features is not a necessary condition for the extent method. In the “extent
method” if any feature of an element on the segment is backlogged then the entire segment
is backlogged. %backlog is then calculated as
Number of Segments with  1 Backlogged Feature
Extent %Backlog 
Total numberof Segments
The extent method led to higher
90% %backlog values than the average
80% method. This seems reasonable if
70% maintenance crews correct multiple
Extent Backlog
60% distresses when maintenance is

50%
40%
performed. For example, crews would
30% also repair cross slope when drop off is
20% being addressed.
10% Although the magnitude of backlog is
0%
different for each method, the
0% 2% 4% 6% 8% 10% 12% distributions are similar. Figure 24 shows
Average Backlog a scatter plot comparison of %backlog of
asphalt travel way computed using the
Figure 24. %Backlog for asphalt pavements in average and extent methods. Regression
Wisconsin tree analysis using average %backlog and
extent %backlog yielded similar models in
2
terms of the best fit parameters and the R values. The average backlog method was selected
for this study simply because it is easier to calculate.
The method for assigning maintenance costs to elements involved distributing and summing
WisDOT’s HMS activity costs (see Table 28 for HMS activity cost codes). Each activity cost
was broken down into labor, equipment, materials. Overhead costs are included in activities
31 to 34 in the miscellaneous category. Administrative costs are included in activities 91 and
93 to 96 in the administration category. These were distributed to the elements based on the
percent allocations determined by WisDOT. Summing the HMS costs determined the total
cost for each of the nine elements. Costs were adjusted by the urban consumer price index
(CPI) to year 2006 costs. This is the same cost index used by WisDOT (5).
The sign repair and pavement markings elements were not included in the study because the
Compass features cannot be related directly to HMS codes. Furthermore, drainage
structures and safety appurtenances were not included because the features are not expressed
as distresses and a rational way to combine backlogs could not be identified. In addition, the
features for drainage structures have costs that vary significantly and often included in the
cost of repairing or replacing bridges which are out of the study scope.
57
5.4.2 Regression Tree Modeling Approach
FY 2006 Dollars Expenditures normalized by million VMT per year
Cost Year enabled comparisons across years and counties.
 The normalized costs were plotted against
mVMT  Million Vehicle Miles Traveled 
  %backlog to reveal trends or patterns relating cost
 Year 
and condition. Each data point on the scatter plots
represents one county in one year. If observations were available in all counties for all years
the total number of possible data points is 216 (72 counties multiplied by 3 years).
In addition, the change in %backlog relative to change in expenditures from years 2004 to
2005 and from 2005 to 2006 were plotted to
 FY 2006 Dollars   FY 2006 Dollars 
determine trends or relationships. The   
Cost  Yearj   Yeari 
equation for change in cost between years i  
(prior year) and j (current year) normalized mVMT  Million Vehicle Miles Traveled 
 
by million vehicle miles traveled in each  Year 
 j 
county is shown to the right. The maximum
number of data points in the scatter plots is 144 (2 annual changes for 72 counties).
The scatter plots did not show relationships between maintenance cost and condition. In
analyzing change in cost versus change in %backlog plots, we expected most values to be in
quadrants 2 and 4 indicating a positive change in budget causing a negative change in
backlog and a negative change in budget causing a positive change in backlog. However this
is not the case. Given the profound scatter in the data, a regression tree analysis was used to
split and group the data in an effort to tease out relationships. The regression tree analysis
used the parameters listed in Table 36.
Table 36. Model parameters for guide analysis of cost and condition in Wisconsin.
Parameter Description
Cost Annual maintenance cost 2006 FY dollars
%Backlog Backlog – percentage (calculated using the average method)
mVMT Million annual Vehicle Miles Traveled
LM County Lane Miles
WSI Winter Severity Index – Measure of winter severity per county
PopDensity Population Density - Population per square mile per county
Income Median Household Income per county - 2000 FY dollars
Latitude Latitude Groupings – Numbered 1 to 5: North to South
SoilType Soil Type – Numbered 1 to 5: Silty, Sandy, Loamy, Wetland, Sandstone
SoilpH Soil pH – Ranges from 5.0 – 7.3, average = 6.6
SoilK Soil Potassium Content - parts per million, range is 80–166, average = 134
SoilP Soil Phosphorus Content - parts per million, ranges is 30–153, average = 52
SoilOM Soil Organic Matter – percentage, ranges from 1.2 – 7.0, average = 3.0
AveAge Pavement Average Age – Average years since last resurface or reconstruction
LandArea Land Area - Square Miles per county
mTruckVMT Million annual Truck Vehicle Miles Traveled
LMconcrete Lane Miles of concrete traveled way per county
LMasphalt Lane Miles of asphalt traveled way per county
CLMups Center Line Miles of unpaved shoulders per county
CLMps Center Line Miles of paved shoulders per county
CLM Center Line Miles per county
58
Million vehicle miles traveled (VMT) were based on traffic counts for one-third of the
counties each year and were provided by WisDOT. The exception is Milwaukee County in
which approximately one-third of the county is counted every year. Annual average daily
traffic (AADT) estimates for counties not counted in the current year are estimated using a
growth factor for the statewide seasonal factor group average. Vehicle miles traveled (VMT)
estimates are based on AADT estimates (6). VMT was not available for the year 2006 at the
time of the analysis so VMT from the year 2005 was used as data for 2006.
Lane mile data for each of the counties was also provided by WisDOT and came from the
STN Database. The only model that used total lane miles as a variable was litter pickup
because it was hypothesized that roadways with more lanes would have more litter than
roadways with fewer lanes. Variables for lane miles of asphalt and concrete pavement were
included for analysis of asphalt and concrete travel ways, respectively.
The winter severity index is a measure of the winter storm severity and frequency in
Wisconsin counties and was obtained from Compass (7). It was hypothesized that counties
with high winter severity indexes would behave differently than counties with lower winter
severity indexes in that they would have higher %backlog for the maintenance of asphalt and
concrete travel ways, unpaved shoulders, paved shoulders, and ditches. The winter severity
index variable was used to group data.
The variables population density and land area were obtained from United States Census
Bureau (8) and the variable median household income was obtained from State of Wisconsin
Department of Administration (9). These were used as splitter variables for all nine partial
elements as means to differentiate
even further among the counties. It
was hypothesized that these variables 1
may have some influence over
maintenance procedures are the
county level. 2
Counties were assigned to latitude
groups (Figure 25). This enabled the
creation of a different model for each 3
latitude group. The latitude
groupings were numbered 1 to 5 with
1 being the northernmost and 5
being the southernmost counties. Figure 25. 4
Soil characteristics were included as Latitude
variables in order to distinguish groups
5
between counties on mowing,
control of woody vegetation and
noxious weeds, and maintenance of
ditches. These characteristics include soil type (10), pH, potassium content, phosphorus
content and organic matter content (11). The soil type was broken into five categories: 1-
Silty, 2-Sandy, 3-Loamy, 4-Wetland, and 5-Sandstone.
Average age of pavement was a variable for the asphalt and concrete travel way models.
This information was obtained from the agency’s PIF (pavement information files). The age
of pavement is the length of time, in years, between the year the condition was sampled and
59
the last time the traveled way was resurfaced or reconstructed. It was hypothesized that older
pavements have more deterioration and deteriorate more quickly than new pavements.
Truck vehicle miles traveled was provided by WisDOT’s 2000 statewide travel demand
model. The truck vehicle miles traveled were based on commodity freight estimates and
forecasts and thus do not include urban type trucks (delivery trucks, cement trucks, etc.). Yet
the data does capture the majority of trucks on the highways and is consistent with the scope
of roadways of interest for the analysis. Truck vehicle miles traveled was used as a variable
modeling cost versus condition of asphalt traveled way, concrete traveled way, paved
shoulders, and unpaved shoulders.
The variables prior cost and prior %backlog are simply the previous year values for cost and
%backlog used to calculate change in cost and change in backlog. For example, when
calculating the change in %backlog from the years 2004 to 2005, the prior %backlog is the
%backlog from the year 2004. These variables were used in the models that considered
change in %backlog and change in cost to provide a base number for the change variables.
Centerline miles for each of the counties were provided by WisDOT and came from the
STN Database. The variable total centerline miles was used for modeling mowing, litter
pickup, woody vegetation, noxious weeds, and ditches because it measures the distance of
roadway in each county. Counties with greater centerline miles may behave differently than
counties with less centerline miles. The variables for centerline miles with unpaved and
paved shoulders were used for the analysis of unpaved and paved shoulders, respectively.
Centerline miles were not available for the year 2006 at the time of report so centerline miles
from the year 2005 were used as data for 2006.
Finally, there is a strong negative correlation between winter severity index and latitude
group. As mentioned earlier, the latitude groupings were numbered from 1 to 5, north to
south. As the latitude group increases, and therefore is further south, it follows that the
winter severity index would be lower because southern Wisconsin has less severe winter
weather than northern Wisconsin.
The regression tree modeling algorithm, GUIDE (Generalized, Unbiased, Interaction,
Detection and Estimation), was used to create equations that model the relationships
between maintenance cost and condition. For each element, both fixed and marginal models
were investigated. The exceptions are asphalt and concrete travel way which have only fixed
models because condition observations are biennial rather than annual.
After finding the best fit models using GUIDE, the MINITAB statistical program was used
to identify and remove outliers and then to refine the best fit model equations based on R2.
Models were compared using the adjusted R2 because it accounts for the number of variables
in the equation. Adjusted R2 increases only if new variables improve the model more than
would be expected by chance.
The regression tree approach did not produce statistically valid relationships between
backlog and expenditures for paved or unpaved shoulders, noxious weeds and ditches. In
the following subsections, we present the results of our analysis of several elements.
60
5.4.3 Regression Tree Models
CONCRETE TRAVEL WAY

A scatter plot of county-level %backlog on concrete travel ways versus expenditures per
million vehicle miles traveled does not show a relationship
between expenditures and backlog. Alternatively, the regression
tree for the best fit model is shown in Figure 26. The model
splits the data by average pavement age. %Backlog for counties
with average pavement age less than 16 years is modeled by the
equation associated with node 2. The average %backlog for this
group is 15.43%. The %backlog for all other counties is
estimated with the equations associated with nodes 6 or 7
depending on whether the county’s VMT is greater than or less
Figure 26. Regression tree than 1332 million. Counties with no concrete pavements were
for maintenance backlog of removed because keeping them in would cause the model to
concrete pavements force a zero intercept. The overall R2 for the model is 75%.
The model equation for node 2 expresses %backlog as a dependent variable of cost. A
graph of the observed versus fitted values of %backlog for the model is shown in Figure 27.
45%
Fitted Concrete Traveled Way Backlog (%)
40%
35%
30%
25%
20%
15%
10%
5%
0%
0% 10% 20% 30% 40% 50% 60%
Figure 27. Observed versus fitted backlog for concrete traveled ways
Observed Concrete Traveled Way Backlog (%)
For counties having concrete pavements with average age ≤ 16 years (node 2),
with average age > 16 years and VMT ≤ 1332 million (node 6),
with average age > 16 years and > VMT 1332 million.
ASPHALT TRAVEL WAY

Similarly the scatter plot of %backlog versus maintenance expenditures on asphalt travel
ways does not reveal relationships between backlog and expenditure. By considering other
61
parameters and using regression analysis, the best fit model equation that relates %backlog
to cost per lane mile of asphalt pavement has an R2 of 87%. For all counties:
MOWING
Figure 28. Regression tree for mowing maintenance expenditures

The scatter plot of %backlog versus expenditures per million VMT for mowing maintenance
did not reveal any relationships. The regression tree for mowing maintenance expenditures
in Figure 28 provided a model that relates %backlog and expenditure. Node 13 applies to
counties in latitude groups 4 and 5, with soil phosphorous content less than or equal to
54.50, and soil type 3. The equation for annual cost has an R2 value of 99%.
This equation was derived from 9 data points for three adjacent counties (Calumet,
Manitowoc, and Winnebago) and each of the three years. The equation is an excellent tool to
estimate mowing cost in this area of the state.
LITTER PICKUP
The scatter plot of %backlog versus expenditures for litter pickup shows no relationship.
The regression tree analysis found models but with unsatisfactory fit.
Alternatively, a best subsets analysis Table 37. Best Subsets Regression for
used %backlog, million VMT, lane Litter Pickup Cost
miles, and centerline miles as the
predictor variables for cost. Table 37
lists the subset of variables that
produced the best fit regression
equation along with the Mallows C-p
and R2 goodness of fit measures. All of
the models include %backlog as a predictor variable. The Mallows C-p is 5.0, which is close
62
to the number of predictor variables, and the adjusted R2 is 75.2%. After removing outliers,
R2 for the model equation is 86%. For all counties:
WOODY VEGETATION
The scatter plot of %backlog versus expenditure per million VMT for litter pickup shows
considerable dispersion and a wide range of expenditure.
The regression tree for expenditures for managing woody
vegetation, shown in Figure 29, splits the data by latitude
group.
The model at node 3 is for counties in latitude groups 4 and
5. The model predicts expenditures as a function of
%backlog, centerline miles, and soil properties including pH,
potassium content, and phosphorus content. The model is Figure 29. Regression
based on 96 data points and has an R2 of 65%. tree for managing
woody vegetation
For counties in latitude groupings 4 and 5:
5.4.4 Discussion
The most influential parameter for fitting and splitting the asphalt and concrete traveled way
regression tree models is average pavement age. This indicates that age of the pavement is a
critical parameter for estimating the backlog and maintenance budget pavements. Nodes for
model with high average pavement age have average backlogs that are larger than nodes with
low values for average pavement age.
The parameter that is most influential for splitting in the mowing and woody vegetation
regression trees is latitude group. This indicates that northern counties behave differently
than southern counties when estimating the backlog and budget for mowing and woody
vegetation activities. Counties in the southern latitude groups (4 and 5) have higher costs
than counties in the northern latitude groups (1, 2, and 3). This finding is independent of the
labor rates, because labor rates were adjusted prior to running the models.
Other influential parameters for fitting the models for mowing, litter pickup, and woody
vegetation are million vehicle miles traveled and centerline miles. In all cases the parameters
budget and backlog were used as variables in the equations. These parameters are important
for predicting the backlog and budget for these elements. Large values for million vehicle
miles traveled and centerline miles result in large costs.
A relationship between median household income and cost occurred for two elements:
concrete traveled way and mowing. In both cases median household income was a splitter
variable – not in the models. The node with higher median household income resulted in a
higher average cost. This could indicate that counties with higher median household incomes
have larger budgets for highway maintenance. Again, this finding occurs after labor costs
were adjusted for location.
63
Chapter 6. Conclusions
Maintenance cost and condition are difficult to relate and study. This research explored the
ability to predict maintenance condition given cost trends and to predict cost given the
desired condition level. Equations that predict condition could be used to determine the
effects on condition of increased or decreased funding. Equations that estimate cost could
be used to determine the required cost to achieve certain levels of condition. In this project,
neither of these modeling efforts was particularly successful.
Among the salient results of this research are a set of probabilistic distribution functions for
annual maintenance costs for a wide range of maintenance activities. Confidence intervals
can be constructed around the average using the chosen level of confidence (i.e., 95%). The
functions are useful for sensitivity and simulation analyses.
The GUIDE regression tree analysis found few models for relating condition to costs. This
could occur because budget allocations are based on previous allocations not on condition.
Counties in Ohio, Michigan and Wisconsin receive a fixed amount based on variables such
as vehicle miles traveled and centerline miles which is adjusted according to available funds
rather than current maintenance conditions.
The research identified some important parameters for estimating and relating condition and
cost:
 The most important parameter for estimating the backlog and the required maintenance
budget for asphalt and concrete traveled ways is average pavement age. The regression
tree nodes with high average pavement age also had high average backlogs and higher
estimated costs than nodes with low values for average pavement age. This finding is
statistically significant.
 In Wisconsin, the most influential parameter for estimating mowing and woody
vegetation maintenance costs and condition is latitude group. Counties in the southern
latitude groups (4 and 5) have higher costs than counties in the northern latitude groups
(1, 2, and 3). This finding is independent of the labor rates, because labor rates were
adjusted prior to running the models.
 Other influential parameters for mowing, litter pickup, and woody vegetation are million
vehicle miles traveled and centerline miles. Normalized costs are higher for counties with
large values of million vehicle miles traveled and centerline miles.
Lack of significant findings could be attributed to the data. First, to see trends overtime,
three years of data may not be enough. With biennial state budgets, three years is only one
and one-half budget cycles. The analysis showed little change over the three year period.
Furthermore, even with budget cuts, noticeable deterioration in condition, deficiencies, or
maintenance backlog may take longer than three years.
Second, cost and condition data are often collected and managed to support different
business processes by separate units of the agency. Consequently, the analysis of this
research depended on assumptions to aggregate condition data or cost data so that they can
be related. Similarly assumptions about the activities associated with the cost data were
necessary to assign costs as being spent for maintenance of specific elements. When
examining the impact of this mapping, the many-to-many and one-to-many mappings
65
performed best. One possible explanation is that for the one-to-one mapping of cost to
condition, the costs did not accurately apply for the elements and conditions to which they
were mapped.
Trends and relationships between cost and condition are washed out by the lack of precision
and accuracy of the maintenance activities and associated costs. For these, the condition
data is complete, not sampled. The best results were for asphalt and concrete pavements.
The available cost and condition data are aggregated and generalized over large areas for
many highway miles. Maintenance management and cost records generally do not include
the precise highway location where maintenance was performance, the specific activities that
were performed, nor the precise cost and timing of those activities. To overcome this
problem, maintenance activities and costs must be more clearly tied to specific maintenance
elements and the condition of those elements.
More accurate models could help maintenance programmers justify budgets and show
accountability. A follow up study should use data covering a longer time period. The study
should focus on maintenance features for which the maintenance expenditures and
condition data can be clearly related.
66
REFFERENCES
1. Adams, Sokolowski, Juni, Kannan (2007). “Routine Highway Maintenance: the
relationship between Cost and Condition.” Proceedings of the 2007 Mid-Continent
Transportation Research Symposium, Ames, Iowa, August 2007
2. Loh, W.Y., GUIDE (version 4) User Manual. Department of Statistics University of
Wisconsin Madison
3. Ohio Department of Transportation , Transportation Management System Manual
4. Ohio Ellis Program – Ohio Department of Transportation.
http://www.dot.state.oh.us/. Accessed July 12, 2007
5. PASER Manual, Transportation Information Center, University of Wisconsin-Madison,
November 1996
6. Loh, W. Y. Regression trees with unbiased variable selection and interaction detection.
In Statistica Sinica, vol. 12, 361-386, 2002
7. Adams, T.M., E. Juni, M. Sproul, and L.Xu (2006). “Regression Tree Models to Predict
Winter Storm Costs” TRR Journal of the Transportation Research Board. No 1948,
pages 117-124.
8. Draper, D.R, H. Smith “Applied Regression Analysis. Third Edition pages 75-76, 210-
211
9. Michigan Pavement Management System, Transportation Asset Management – Michigan
Department of Transportation.
http://www.michigan.gov/documents/pavement_16566_7.pdf. Accessed December
17, 2007
10. Allen, Patrick. “Sufficiency Code Descriptions (last updated 5/30/2007)”, Michigan
DOT internal document. [email protected].
11. R. Johnson and D. Wichern (1992). Applied Multivariate Statistical Methods, Third Edition.
Prentice Hall.
12. Bureau of Labor Statistics - Occupational Employment Statistics.
http://www.bls.gov/oes/2007/may/oes_2600004.htm. Accessed on March 18, 2009
13. BLS Metropolitan and Nonmetropolitan Area Definitions.
http://www.bls.gov/oes/2007/may/msa_def.htm#12980
14. Schmitt, R.L. S. Owusu-Ababio, R.M. Weed and E.V. Nordheim.. (2006).
“Understanding statistics in Maintenance quality assurance programs.” TRR: J. of the
Transportation Research Board. No 1948. Pages17-25.
67

07-12 FR PDF

Uploaded by

Copyright:

Available Formats

07-12 FR PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

07-12 FR PDF

Uploaded by

Copyright:

Available Formats

ESTIMATING COST PER

LANE MILE FOR ROUTINE

Midwest Regional University Transportation Center

Authors: Teresa M. Adams, Ph.D.

Principal Investigator: Teresa M. Adams, Ph.D.

4. Title and Subtitle 5. Report Date January 2011

7. Author/s 8. Performing Organization Report No.

15. Supplementary Notes

1.1 Problem Statement

Figure 1. Trend for Miles Traveled and Maintenance Expenditures in Wisconsin

Data Collection Michigan, Ohio, and

Prepare and Format Cost and Environmental,

Synthesize and Characterize Data

Map Cost to Condition Figure 2. Methodology

Analyze Correlation of all Available Data

Tools: Scatter Plots,

1.5 Organization of Report

2.1 Adjustment for Inflation

2.2 Adjustment for Location

2.3 Cost Transformation Functions

2.3.1 Box-Cox Transformation

2.3.2 Johnson Transformation

2.4 Maintenance Activity Costs per Lane Mile (CPLM)

2.4.2 Highway Maintenance in

2.4.3 Highway Maintenance in

2.5 Distribution Functions for Maintenance Expenditures

Crash Attenuator Repair or Johnson: μ=0.072, σ=0.921 1.8658 + 0.4306

6123 Full Depth Repair 102.15 118.62 249.34 78

0.4544 + 1.9991 Ln((X + 294.54) / (1302.4 - X))

Chemical Spray Vegetation

Johnson: μ=-0.0069, σ=0.995,

Overhead Mounted Johnson: μ=-0.0057, σ=0.947,

3.1 Michigan’s Maintenance Costs per Lane Mile

3.1.1 Maintenance Activity Categories and Lane Mile (LM) Groups

range for half of the 600

Figure 13. CPLM trends for LM groups – direct forces at

3.1.3 Annual Cost Trends during the Study Period

2004 2005 2006 2004 2005 2006

500 100 1200 250

2004 2005 2004 2005

Table 15. Categorical CPLM – contracts with Michigan’s counties

and Lane Mile (LM) 900

Responsible Lane Miles

Ohio data included 88 counties 200

3.2.2 Cost Trends for LM

signs and pavement 600

Figure 19. Annual Fiscal Year - Ohio's Counties

3.3 Wisconsin’s Maintenance Costs per Lane Mile

3.3.1 Maintenance Activity Categories and Lane Mile (LM) Groups

1500 2 72 221.68 198.71 377.95

Figure 20. Lane mile (LM) groups – Wisconsin’s counties

450 450 150

300 300 100

LM Group - Wisconsin's Counties

4.1 Michigan’s Maintenance Quality Assurance Program

4.2 Ohio’s Maintenance Quality Survey (MQS) Program

4.3 Wisconsin Maintenance Quality Assurance Program

5.1 Analysis Methodology

5.2 Modeling Cost to Condition For Michigan