Report # MATC-KSU: 111

Final Report

Characteristics and Contributory Causes

Related to Large Truck Crashes (Phase I)
®

-Fatal Crashes

Sunanda Dissanayake, Ph.D., P.E.

Associate Professor
Civil Engineering
Kansas State University

2010
A Cooperative Research Project sponsored by the
U.S. Department of Transportation Research and
Innovative Technology Administration

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.
 Characteristics and Contributory Causes Related To Large Truck Crashes (Phase I) –

Fatal Crashes

Sunanda Dissanayake, Ph.D., P.E Nishitha Bezwada

Associate Professor Graduate Research Assistant
Department of Civil Engineering Department of Civil Engineering
Kansas State University Kansas State University

A Report on Research Sponsored By

Mid-America Transportation Center

University of Nebraska-Lincoln

June 2010
 Technical Report Documentation Page
1. Report No. 2. Government Accession No. 3. Recipient's Catalog No.
25-1121-0001-111

4. Title and Subtitle 5. Report Date

Characteristics and Contributory Causes Related to Large Truck
Crashes (Phase I)
6. Performing Organization Code

7. Author(s) 8. Performing Organization Report No.

Sunanda Dissanayake and Nishitha Bezwada 25-1121-0001-111

9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)

11. Contract or Grant No.

12. Sponsoring Agency Name and Address 13. Type of Report and Period Covered
Mid-America Transportation Center
2200 Vine St. 262 Whittier Building
PO Box 830851
Lincoln, NE 68583-0851 14. Sponsoring Agency Code
MATC TRB RiP No. 17131
15. Supplementary Notes

16. Abstract
One-ninth of all traffic fatalities in the United States have involved large trucks in the past five years, although large trucks
contributed to only 3% of registered vehicles and 7% of vehicle miles travelled. This contrasting proportion indicates that
truck crashes in general tend to be more severe than other crashes though they constitute a smaller sector of vehicles on the
road. To study this issue, fatal crash data from the FARS was used to analyze characteristics and factors contributing to
truck-involved crashes. Driver, vehicle, and crash-related contributory causes were identified, and as an extension the
likelihood of occurrence of these contributory causes in truck-involved crashes with respect to non-truck crashes was
evaluated using the Bayesian Statistical approach. Likelihood ratios indicated that factors such as stopped or unattended
vehicles and improper following have greater probability of occurrence in truck crashes than in non-truck crashes. Also,
Multinomial Logistic Regression was used to model the type of fatal crash (truck vs. non-truck) to compare the relative
significance of various factors in truck and non-truck crashes. Factors such as cellular phone usage, failure to yield right of
way, inattentiveness, and failure to obey traffic rules also have a greater probability of resulting in fatal truck crashes.
Among several other factors, inadequate warning signs and poor shoulder conditions were also found to have greater
predominance in contributing to truck crashes than non-truck crashes. By addressing these factors through the
implementation of appropriate remedial measures the truck safety experience could be improved, which would eventually
help in improving overall safety of the transportation system.

17. Key Words 18. Distribution Statement

19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price
77
ii
 Table of Contents

LIST OF FIGURES ......................................................................................... V

LIST OF TABLES.......................................................................................... VI
ACKNOWLEDGEMENTS .............................................................................. VII
ABSTRACT .................................................................................................. IX
CHAPTER 1– INTRODUCTION ................................................................. 1
1.1 Background .................................................................................................... 1
1.2 Objectives ...................................................................................................... 4
1.3 Outline of the Report ..................................................................................... 4
CHAPTER 2 – LITERATURE REVIEW ............................................................. 6
2.1 Characteristic Comparisons, Rates, and Trends ............................................ 6
2.2 Truck Crash Study on LTCCS, TIFA, and GES ............................................ 9
2.3 Contributory Causes for Large Truck Crashes ............................................ 12
2.4 Drowsy Driver Effect and Hours of Service ................................................ 15
2.5 Speed Limit, Urban /Rural Contrast, Rear-End/Angle Collisions, and Roadway
Parameters .............................................................................................. 16
2.6 Bayesian and Other Modeling Techniques .................................................. 18
2.7 Multinomial Logistic Regression ................................................................. 20
2.8 Countermeasure Ideas .................................................................................. 22
CHAPTER 3 – METHODOLOGY ............................................................. 25
3.1 Data .............................................................................................................. 25
3.2 Analysis Methods......................................................................................... 29
3.2.1 Bayesian Statistical Approach ............................................................ 29
3.2.2 Multinomial Logistic Regression ........................................................ 31
CHAPTER 4 – RESULTS AND DISCUSSION................................................... 35
4.1 Characteristics of Fatal Truck Crashes ........................................................ 35
4.2 Truck Striking/Struck Comparison .............................................................. 44
4.3 Comparison of Characteristics of Fatal Truck and Non-Truck Crashes ...... 46
4.4 Bayesian Statistical Analysis: Contributory Causes for Truck and Non-Truck Crashes
................................................................................................................ 54
4.5 Multinomial Logistic Regression Analysis for Truck Crashes .................... 58
4.5.1 Roadway Characteristics ..................................................................... 61
iii
 4.5.2 Crash Characteristics .......................................................................... 62
4.5.3 Environmental Characteristics ............................................................ 63
4.5.4 Driver Characteristics ......................................................................... 63
4.5.5 Other Contributory Factors ................................................................. 63
CHAPTER 5 – CONCLUSIONS AND SUMMARY............................................. 68
5.1 Conclusions .................................................................................................. 68
REFERENCES ............................................................................................. 70

iv
 List of Figures

Figure 1.1 No Zones or Blind Spots around a Large Truck .................................................1

Figure 1.2 Number of Fatal Large Truck Crashes from 1998-2007 ....................................2

Figure 1.3 Number of Vehicle Occupants Killed in Large Truck Crashes ..........................3

Figure 4.1 Point of Impact for Trucks in Fatal Crashes .....................................................36

Figure 4.2 Manner of Collision of Fatal Truck Crashes ....................................................37

Figure 4.3 Fatal Truck Crashes in Different Speed-Limit Ranges ....................................38

Figure 4.4 Age of Truck Drivers Involved in Fatal Truck Crashes ...................................40

Figure 4.5 Proportion of Fatal Truck Crashes on Different Traffic Flowways .................41

Figure 4.6 Level of Deformation of all Vehicles Involved in Fatal Truck Crashes ..........42

Figure 4.7 Truck Driver-Related Contributory Factors in Fatal Crashes .........................43

Figure 4.8 Fatal Truck Crashes by Roadway Type in Truck Striking/Struck Conditions .45

Figure 4.9 Truck Crashes in Different Light Conditions under Striking/Struck Types ...46

Figure 4.10 Initial Impact Point for Truck and Non-Truck Crashes.................................47

Figure 4.11 Driver Age for Truck and Non-Truck Drivers ...............................................48

Figure 4.12 Posted Speed Limit for Truck and Non-Truck Crashes ................................49

Figure 4.13 Manner of Collision for Truck and Non-Truck Crashes ................................50

Figure 4.14 Level of Deformation for Truck and Non-Truck Crash Vehicles ..................51

Figure 4.15 Trafficway Type for Truck and Non-Truck Crashes ......................................51

Figure 4.16 Rural Urban Contrast for Truck and Non-Truck Crashes .............................52

Figure 4.17 Type of Roadway for Fatal Truck and Non-Truck Crashes ...........................53

Figure 4.18 Number of Lanes on Roadways Where Truck/ Non-Truck Crashes Occurred ..

........................................................................................................................53

v
 List of Tables

Table 4.1 Conditional Probabilities and Likelihood Ratios for Crash-Related Factors ....55

Table 4.2 Conditional Probabilities and Likelihood Ratios for Vehicle-Related Factors .....

........................................................................................................................56

Table 4.3 Conditional Probabilities and Likelihood Ratios for Driver-Related Factors ...57

Table 4.4 Description of the Variables Used in the Model................................................59

Table 4.5 Parameter Estimates and Odds Ratio of Fatal Truck Crashes in the Model ......64

Table 4.6 Model Fit Statistics of the Multinomial Logistic Regression Analysis .............65

Table 4.7 Tests of Independence .......................................................................................66

Table 4.8 Associations of Predicted Probabilities and Observed Responses ....................67

vi
 List of Abbreviations

National Highway Traffic Safety Administration (NHTSA)

Fatality Analysis Reporting System (FARS)
Federal Motor Carrier Safety Administration (FMCSA)
Commercial Driving License (CDL)
Large Truck Crash Causation Study (LTCCS)
University of Michigan Transportation Research Institute‘s (UMTRI)
Trucks Involved in Fatal Accidents (TIFA)
Motor Carrier Management Information System (MCMIS)
Portable Overhead Surveillance Trailers (POSTS)
MoDOT Transportation Management System (TMS)
Lane Change/Merge (LCM)
National Automotive Sampling System, General Estimates System (NASS GES)
Large Truck Causation Study (LTCCS)
Drowsy Driver Warning System (DDWS)
Hours-of-Services (HOS)
Highway Safety Information System (HSIS)
Psychomotor Vigilance Task (PVT).
Vehicle Dynamic Models Roadway Analysis and Design (VDM RoAD)
Light Truck Vehicles (LTV)
General Estimates System (GES)
Blood Alcohol Concentration (BAC)
Florida Department of Transportation (FDOT)
Center for Urban Transportation Research (CUTR)
Commercial Driver‘s License Information System (CDLIS)
National Center for Statistics and Analysis (NCSA)
Police Accident Reports (PARS)
Fatality Analysis Reporting System file transfer protocol (FARS FTP).
Gross Vehicle Weight Rating (GVWR)
Akaike Information Criterion (AIC)
Schwarz Criterion (SC)

vii
 Acknowledgements

The authors would like to thank Mid-American Transportation Center for funding this project

and providing all the help necessary by remaining true to their theme of “improving safety and

minimizing risk associated with increasing multi-modal freight movement on the U.S. surface

transportation system.‖ The authors would also like to thank the National Highway Traffic Safety

Administration (NHTSA) for providing the Fatality Analysis Reporting System (FARS) data,

which has been crucial to conducting this study. Special thanks also go to Steven Buckley at the

Kansas Department of Transportation who served as the Project Monitor.

viii
 Abstract

One-ninth of all traffic fatalities in the United States have involved large trucks in the

past five years, although large trucks contributed to only 3% of registered vehicles and 7% of

vehicle miles travelled. This contrasting proportion indicates that truck crashes in general tend to

be more severe than other crashes though they constitute a smaller sector of vehicles on the road.

To study this issue, fatal crash data from the FARS was used to analyze characteristics and

factors contributing to truck-involved crashes. Driver, vehicle, and crash-related contributory

causes were identified, and as an extension the likelihood of occurrence of these contributory

causes in truck-involved crashes with respect to non-truck crashes was evaluated using the

Bayesian Statistical approach. Likelihood ratios indicated that factors such as stopped or

unattended vehicles and improper following have greater probability of occurrence in truck

crashes than in non-truck crashes. Also, Multinomial Logistic Regression was used to model the

type of fatal crash (truck vs. non-truck) to compare the relative significance of various factors in

truck and non-truck crashes. Factors such as cellular phone usage, failure to yield right of way,

inattentiveness, and failure to obey traffic rules also have a greater probability of resulting in

fatal truck crashes. Among several other factors, inadequate warning signs and poor shoulder

conditions were also found to have greater predominance in contributing to truck crashes than

non-truck crashes. By addressing these factors through the implementation of appropriate

remedial measures the truck safety experience could be improved, which would eventually help

in improving overall safety of the transportation system.

ix
 Chapter 1 Introduction

1.1 Background

Of the 41,059 fatalities related to motor vehicle crashes in 2007 12% or 4,808 deaths

involved large trucks, and 17% of those fatalities involving the large trucks were the occupants

of said trucks. Though large trucks contribute to only 8% of vehicles involved in fatal crashes

over the last five years their impact in terms of severity warrants major concern.

Large trucks of gross body weight greater than 10,000 pounds have different performance

characteristics than smaller vehicles. The large size of the vehicles makes it difficult for drivers

to maneuver smoothly on roadways. Drivers might face vehicle control challenges while

operating large trucks on interstate or state highways at high speeds or at intersections while

making turns. Also, the element of blind spots, as shown in figure 1.1, makes it even more

challenging for the truck driver and surrounding vehicle drivers to avoid the heavy crash risk.

Figure 1.1. No Zones or Blind Spots around a Large Truck

When considering the past 10 years of data, it can be observed that the frequency of fatal

truck crashes varies between 4400 and 4800 crashes per year (Fig. 1.2). Each of these crashes

results in major destruction of human life and property, which in most cases is many times worse

1
than other passenger car crashes. As it is evident that the frequency of these crashes is remaining

consistent, it becomes crucial to identify methods to mitigate this issue.

5000

4900

4800
No. of Truck Crashes

4700

4600

4500

4400

4300

4200

4100
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Year

Figure 1.2. Number of Fatal Large Truck Crashes from 1998-2007

When involved in crashes trucks can rollover or jackknife at high speeds and

exponentially increase the severity of the crash as a result. Many factors such as roadway

geometry, environmental conditions, driver mental and physical conditions, and vehicle

conditions affect the possibility of crash occurrence.

Research has also shown (Fig. 1.3) that large trucks cause more fatalities to other non-

truck vehicle occupants than those in trucks. On average 84% of fatalities related to large truck

crashes in the country are not the occupants of trucks. This reinforces the threat large trucks

impose on other motor vehicles, pedestrians, and pedal cyclists.

2
 6000

5000
Number of Fatalities

4000 Truck
Occupants
3000 Non-Truck
Occupants
` Total
2000

1000

0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Year

Figure 1.3. Number of Vehicle Occupants Killed in Large Truck Crashes

Even though the number of fatal truck crashes has generally been decreasing with some

fluctuations over the past 10 years, the amount of truck travel is increasing. Consequently,

continued attention to this issue is required in order to find ways of reducing truck crash risk.

The Federal Motor Carrier Safety Administration (FMCSA) has set a goal of a 50% reduction in

commercial truck-related fatalities by the year 2010 (1). The FMCSA wants to reduce the

number and severity of large truck- and bus- involved crashes through several inspection and

enforcement procedures. These procedures include more commercial motor vehicle and operator

inspections and compliance reviews; stronger enforcement measures against violators; expedited

completion of rulemaking proceedings; utilization of scientifically sound research; and effective

Commercial Driving License (CDL) testing, recordkeeping, and sanctions. All these measures

3
can be assisted by analyzing large truck crashes. Accordingly, it is important for the safety

community to identify characteristics related to large truck-involved fatal crashes.

1.2 Objectives

Truck crashes can be mitigated by identifying the characteristics and contributory causes

involved and institute suitable countermeasures as rectification. Hence, the primary objectives of

this study are as follows:

 To analyze and evaluate various crash characteristics that prevailed at the time of the

occurrence of fatal truck crashes.

 To identify various crash, vehicle, and driver related contributory causes prevalent to

fatal truck crashes.

 To evaluate the relative significance of various contributory causes in fatal truck crashes

as compared to fatal non-truck crashes through the calculation of likelihood ratios.

 To model the type of fatal crash (truck vs. non-truck) in terms of crash characteristics and

other factors to compare the relative significance of these factors in truck and non-truck

crashes.

1.3 Outline of the Report

This report consists of five chapters with the background and objectives of this research

having been covered in the first. The second chapter consists of a review of prior research related

to the study area, and the third chapter presents the methodologies used in the analysis along

with descriptions of data used in the study. The fourth chapter reports the results of the

characteristic and comparative studies conducted between truck and non-truck crashes derived

by using statistical analysis, and a detailed discussion is presented. In the final chapter summary

and conclusions are presented and a discussed in further detail. By addressing these issues

4
through this report the overall truck crash rate can be reduced, which can help in improving the

overall safety of the transportation system.

5
 Chapter 2 Literature Review

Fatal truck crash-related research studies have an extended history in addressing different

safety aspects using a variety of databases and surveys. Past researchers have used various

statistical modeling techniques to predict or explain the nature of truck crashes, and many

findings are listed under this area. Furthermore, different types of crashes have been examined

by these researchers, thereby narrowing down the study to identify more specific factors related

to selected states. In this chapter, a thorough discussion of past studies is presented under the

following subsections: truck crash characteristics, rates and trends, contributory factors involved,

crash types and related maneuvering difficulties, intersection-related crashes, human factors, risk

to self and risk to others, countermeasure evaluations, medication and risk of injuries, decision to

stop driving, vehicle design, and statistical methodologies.

2.1 Characteristic Comparisons, Rates, and Trends

Blower conducted a study by collecting detailed data on the causes of truck crashes in the

country and developed suitable countermeasures that would be effective in reducing the number

and severity of the crashes (2). The Large Truck Crash Causation Study (LTCCS), used in this

research project, was developed by the FMCSA in cooperation with the NHTSA. The study took

three years and involved investigation teams at 24 locations around the country. Each crash was

investigated on the field and detailed analysis was conducted by experienced crash investigators.

The second study was conducted by the University of Michigan Transportation Research

Institute‘s (UMTRI) Trucks Involved in Fatal Accidents (TIFA) project (3). In contrast to the

LTCCS, a telephone survey was conducted relevant to fatal truck crashes in the country. Also,

police reports were acquired for all crashes as a part of the survey. The variables coded in each

study (2 and 3) were compared by developing an algorithm to analyze the most significant

6
factors in truck crashes and their accuracy. Though studying both sets of data and referring to the

―parent‖ FARS file, some cases could not be matched when defined as per the search protocol.

As a result, the LTCCS proved to be the most elaborate database in truck crash reporting.

Using this very database Krishnaswami et al. analyzed the causes of heavy truck

aggressiveness in two-vehicle truck/light vehicle crashes and also derived detailed models to

help propose the required truck structural countermeasures to mitigate collision severity (4). In

this study, three years of data, from 1996 to 1998, was used from the FARS, TIFA, and GES, or

General Estimates System, databases. Collision and injury models were constructed using

lumped parameter models in a two-stage manner. The first stage was a physical representation of

the collision process using collision variables as inputs; acceleration levels, total velocity change,

and the crush levels experienced by the vehicle occupants are examples of the variables used. In

the second stage, the previous outputs were used in the injury models to predict occupant injury

outcomes. From the results of the collision and injury models it was consistently shown that by

reducing peak vehicle deceleration, injury risk can be decreased. Another important observation

from the simulations was that for a particular deceleration level almost constant injury criterion

could be seen irrespective of the change of velocity.

As an extension of his earlier work, Blower identified the issues that contribute most to

commercial motor vehicle crashes, fatalities, and injuries in the state of Michigan (5). This was

accomplished by conducting a detailed analysis of the available data, which included the

Michigan vehicle crash files, the trucks involved in fatal accidents file, and Motor Carrier

Management Information System Inspection and Carrier files for the period of 2001-2005. From

the analysis it was evident that angle, rear-end, and head-on crashes appeared to be the most

predominant crash patterns among commercial motor vehicles. Also it was observed that in

7
almost all cases, brake defects were associated with fatal rear-end, head-on, and angle collisions,

while lighting defects were associated with fatal rear-end crashes. Hence, it was concluded that

brake and lighting system violations were the most frequent causations. To address these issues,

countermeasures such as preventive maintenance programs, training, consultation, and public

information and education programs were proposed.

Another report by the United States Government Accountability Office to the FMCSA

addressed the importance of reducing the number of commercial vehicle crashes and identifying

carriers that pose a high risk for crashes (1). Presently FMCSA decides which carriers to inspect

primarily by using an automated data-driven analysis system called SafeStat. This system uses

data on crashes, vehicle and driver violations, and other information to develop a priority list of

high-risk-posing carriers. Though this has proved to be highly useful compared to the

conventional random inspection of carriers, a recent study suggested a better and a more accurate

way of analysis. For this purpose, a number of regression methods have been developed using

crash data from the Motor Carrier Management Information System (MCMIS) for the year 2004.

The accident, driver, vehicle, and safety management sections have been taken as independent

variables to predict crash risks. Results were compared to those obtained from the SafeStat

system and were found to be 9% more accurate.

Daniel et al. proposed an accident prediction model which had been built for analyzing

factors affecting truck crashes on roadways with intersections (19). Truck crash data for this

project was developed by including all crashes in 1998 and 1999 in the state of New Jersey from

police accident report files. This database was used to conduct an initial analysis of truck crashes

at signalized intersections along Route 1 in New Jersey. Poisson regression and negative

binomial models were applied using LIMPID software to obtain the analysis results. Variables

8
considered in modeling the crashes were segment length, AADT, degree of curve for horizontal

curves, length of horizontal curve, crest curve grade rate, length of vertical curve, posted speed

on main road, number of interchanges within the segment, and pavement width. From the

analysis, it was concluded that signalized intersections have a significant impact on truck crash

rate. By incorporating this feature it was concluded that a better model on crash involvement can

be designed by accounting for both intersections and adjacent geometric features.

Vap and Sun analyzed truck and passenger car interactions for the state of Missouri on its

urban and rural freeways (24). The urban data was collected from the Portable Overhead

Surveillance Trailers (POSTS) and the rural data was obtained from digital videos set up at the

desired locations. Apart from these data collectors, the MoDOT Transportation Management

System (TMS) was also used. Using this data, an analysis of trucks-at-fault crash rates versus

passenger vehicles-at-fault crash rates, or RSEC ratios, were estimated. These results showed

that on urban freeways the percentage of trucks-at-fault ratio was considerably high. By contrast,

the rural data, in general, showed that truck crashes were not disproportional to the crash rates of

passenger vehicles. Hence, it was concluded that a greater safety concern coefficient is attributed

to truck-passenger vehicle interactions on urban freeways.

Apart from these studies, which particularly focused on truck-involved crashes, many

more reports on general fatal crash data were reviewed (please refer to sources 8-18 for the

reports reviewed for this study) to acquire a larger idea of what these crashes have in common.

2.2 Truck Crash Study on LTCCS, TIFA, and GES

Blower explained the significance of mirror-relevant crash types which occur due to a

driver‘s restricted direct field of view in trucks (3). Mirror relevant crashes are those in which the

truck driver would have needed to use mirrors to maneuver safely. For this purpose, a study has

9
been conducted to evaluate the types of crashes which could have occurred due to insufficient

field of view for drivers. These crash types include lane change/merge, or LCM, left, LCM right,

and left and right turn with conflict vehicles approaching from rear. The observational fatal data

for this purpose was taken from the LTCCS and the TIFA study compiled by the UMTRI. Injury

and property damage data files were taken from the National Automotive Sampling System,

General Estimates System (NASS GES), which is a nationally representative sample of police-

reported crashes compiled by NHTSA. From the results it was summarized that mirror-relevant

crashes account for almost 20% of truck crash involvements and serious measures need to be

taken to minimize these by providing better facilities that provide drivers with a broader view of

their surroundings.

As mentioned briefly in the previous section, the second study was undertaken by the

UMTRI TIFA project (2). In contrast to the LTCCS, a telephone survey was conducted on fatal

truck crashes in the country. Also, police reports were acquired for all crashes as a part of the

survey. Another project was conducted with an objective to identify the issues that contributed

most to commercial motor vehicle crashes, fatalities, and injuries in the state of Michigan (5).

This was done by conducting a detailed analysis of the available data, which included the

Michigan vehicle crash files, Trucks Involved in Fatal Accidents file, and Motor Carrier

Management Information System Inspection and Carrier Files, for the period 2001-2005.

Another study analyzed the causes of heavy truck aggressiveness in two-vehicle truck/light

vehicle crashes and also derived detailed models which will help propose the required truck

structural countermeasures to mitigate collision severity (4). Three years of data from 1996 to

1998 were used from FARS, TIFA, and GES for this project.

10
 Another report was submitted to Congress on the Large Truck Crash Causation Study

conducted by the FMCSA and NHTSA (19). This study has a unique database which not only

covers the descriptive data of the crashes occurring but also incorporates pre-crash factors such

as driver fatigue and distraction, vehicle condition, weather, and roadway conditions. Unlike the

FARS, which only deals with fatal crashes, and NHTSA‘s GES, which considers only

probability-based sample data, the LTCCS focuses on a larger spectrum of variables—

approximately 1,000 per crash—in a crash case. Coding of the events surrounding each crash is

categorized as ―critical event,‖ ―critical reason‖ for the critical event, and ―associated factors‖

present. This study involved three crash severity levels: fatal, capacitating injury and non-

incapacitating injury. The primary protocol for the truck body type is the same as in FARS. The

data has been categorized into 12 different crash types. From the analysis, it was concluded that

rear-end crash type is one of the most predominant cases observed among truck crashes. The

LTCCS database has been made electronically available to the public since 2006. However, this

data does not contain information from interviews. The full database, inclusive of interview data,

will be made available to researchers, private groups, universities and others upon request.

Blower et al. conducted yet another study by applying the NHTSA‘s definition of trucks

to the TIFA project with one exception. Trucks in the TIFA file include all of those with a Gross

Vehicle Weight Rating (GVWR) over 10,000 pounds, but emergency vehicles, such as

ambulances or fire trucks, are excluded. As will be seen below, exclusion of fire trucks and

ambulances accounts for only a small part of the difference between FARS and TIFA. Apart

from this exclusion, both FARS and TIFA count the same types of vehicles as trucks. The

comparison of data files is based on the 1999 data years for both FARS and TIFA. Using

NHTSA'a definition of large trucks in FARS, the 1999 FARS file identified 4,898 trucks

11
involved in fatal accidents in 1999. The TIFA file for that year has 5,233 trucks, a difference of

335 trucks or about 6.8% more trucks in the TIFA file than in FARS. The difference of 335 is the

result of 40 cases that were counted in FARS as trucks but do not qualify as trucks in the TIFA

file, and 375 cases identified as trucks in TIFA but which were classified as some other type of

vehicle in the FARS file

2.3 Contributory Causes for Large Truck Crashes

Rau conducted a study about drowsy driver detection and the effects of employing a

warning system for commercial vehicle drivers (21). The research has been conducted by

NHTSA and its partners since 1996 in order to quantify the loss of alertness among commercial

vehicle drivers. Drowsiness is measured using a three minute running average of slow eyelid

closures, as assessed by the Drowsy Driver Warning System, DDWS, during nighttime driving.

It depends on the capability of the camera to detect infrared light reflected back to the source at

the camera from the driver‘s retina. By this detection system, the measures of performance at

braking, closing, lane changing, lane keeping, and speed maintenance were observed. The first

objective was to find drowsiness-level distributions and the differences between the distributions

with and without the DDWS. The second objective was to see the variations in drowsiness with a

number of independent factors like age, nights of sleep, and so forth. From the experimental

analysis, it was concluded that further understanding was needed about highway safety benefits,

fleet acceptance, operational utility, and fatigue management practices so that the fatigue crash

problems can be minimized.

Garber et al. compared the safety effects of a uniform speed limit, USL, for all vehicles

as opposed to a differential speed limit, DSL, for cars and heavy trucks (22). Crash and volume

data were synthesized from 17 states to obtain the sample of interstate highways used in this

12
study. A modified empirical Bayes framework was used to evaluate crash frequency variations

with changes in speed limit. The basic methodology of the modified Bayes approach was

conducted in four steps. Initially, the number of crashes at each site within a certain state as a

function of related independent variables (in this case traffic volume and segment length) was

created. Then, the number of expected after-period crashes at each site was determined and their

summation ‗π‘ was calculated. Next, the sum of the actual crashes that did occur at each site was

computed as ‗λ‘. Afterwards, the ratio of total actual crashes to the total expected after-period

crashes was determined and checked to see if the ratio of effectiveness ‗Ѳ‘ was significantly

different from unity by using appropriate confidence intervals. From the experimental analysis, it

was concluded that the modified Bayes approach showed no consistent safety impacts

attributable to differential or uniform speed limit policies for rural interstate highways. In most

cases, it was found that the actual number of crashes for the after period was larger than the

predicted expected after-period crashes

Daniel et al. described the use of Poisson regression and negative binomial accident

prediction models for truck accidents on an urban arterial with heavy truck volumes and a large

number of signalized intersections (23). The research had a twofold objective. The first goal was

to identify the factors that impact the occurrence of truck crashes on urban arterials with

signalized intersections. This was achieved by developing an accident prediction model. The

second objective was to conclude on an approach which would account for signalized

intersections in one unified prediction model. For these objectives, a prediction model was

developed for truck crashes on a truck route in New Jersey on Route 1. A truck accident database

for the state of New Jersey from 1998-2000 was collected for the study. Two models were

developed for the selected roadway: unified, including both intersection and non-intersection

13
locations, and separate models. For both models, the goodness of fit between the expected

number of accidents and explanatory variables was evaluated based on both Pearson Rp2and

deviation RD2 values. It was concluded from the model that horizontal and vertical curvature

were critical factors in determining the safety of the roadway. A reduced model derived from the

above two models proved to be more efficient in both types of roadway segments.

Dick et al. presented a comprehensive evaluation of the federal interstate commercial

driving hours-of-services, HOS, rules implemented in January 2004 (24). The rules that had been

largely unchanged for more than 65 years were revised by the FMCSA. The new HOS rules

included a number of prominent changes designed primarily to promote greater daily sleep and

to encourage more regular daily work-rest cycles. Some of the changes included a daily

minimum off-duty requirement of 10 hours, maximum hours of driving prior to going off duty as

11 hours, and also maximum tour-of-duty (beyond which driving is not permitted) as 14 hours.

Features of the old rule that did not support or promote driver alertness were considered in this

amended version. The results were the opinions expressed by a diverse group of people and there

was a conceded upon positive view of the new rules. They also enabled the drivers to regularize

their work timings more optimally.

Kostyniuk analyzed two-vehicle crashes in the 1995–98 Fatality Analysis Reporting

System (FARS) database to compare car-car crashes with car-truck crashes (25). The research

was conducted in three stages. The first stage sought to identify driving maneuvers or actions of

cars and large trucks that have a higher chance of resulting in fatal car-truck collisions than fatal

collisions with a similar vehicle. The second stage involved discerning patterns associated with

these driving actions through a detailed examination of actual crash reports. The third stage

involved exploring ways that the risks associated with the identified driving actions can be

14
effectively communicated to motorists with special attention to the fit between study findings

and potential instructional approaches. A limitation of the study is that it did not address nonfatal

crashes, single-vehicle crashes, or crashes involving more than two vehicles; these constraints

are important to keep in mind because fatal and injury crashes are not similar in their causes or in

the numbers of people they affect.

2.4 Drowsy Driver Effect and Hours of Service

Khattak and Targa explored the elements of ―injury severity‖ and ―total harm‖ in cases of

truck-involved work zone crashes (26). Their characteristics were empirically compared to those

of non-truck-involved collisions. For this study, a unique dataset from the Highway Safety

Information System (HSIS) with additional variables coded from narratives in police reports was

used. Also, the year 2000 HSIS data for the state of North Carolina was used to develop the

work-zone-related crashes. Using this data, ordered probit models were estimated for the most

seriously injured occupant in the crash, and linear regression models for ―total harm‖ in the

crash were estimated. The linear model contained the variables of frequency and severity of

injuries by transforming them into numerical values. From the results, certain situations which

seemed to enhance the probability of work-zone-truck collisions were observed. The instance in

which the road was completely closed with a detour in the opposite direction seemed to be the

most predominant case for truck crashes in these areas. Also two-way undivided roads and

places where the traffic moved out of normal paths were other scenarios which seem to enhance

the probability of a crash.

Dick et al. presented a comprehensive evaluation of the federal interstate commercial

driving HOS rules implemented in January 2004 (27). The rules that were largely unchanged for

more than 65 years were revised by the FMCSA.

15
2.5 Speed Limit, Urban /Rural Contrast, Rear-End/Angle Collisions, and Roadway Parameters

Dabbour et al. analyzed radius requirements for reverse horizontal curves so as to attain

better vehicle stability for trucks travelling on freeway interchanges (28). For this purpose,

several models developed on vehicle stability were studied and finally, the most advanced

extension of these models, a computer program called vehicle dynamic models roadway analysis

and design (VDM RoAD), was used. This program had a built-in vehicle library that contained

most of the AASHTO-designed trucks. Two different alignment combinations were used; one

with the effect of introducing reverse curvature and the other with the effect of introducing

vertical alignment in the reverse curves. Geometric alignment data of the curves were the data

input for the program. By using the different optimum models suggested by the program, it was

analyzed that an increase is required in the minimum radius of horizontal curves to compensate

for both effects of reverse curvature and vertical alignment. This change was shown to reduce

skidding and rollover accidents on highways.

Miaou and Lum illustrated ways in which the Poisson regression model can be used to

evaluate the effects of highway geometric design on truck accident involvement rates (29). The

model applied in this study can also be applied to any type of interest, such as roadway class,

vehicle configuration, and accident severity. From the model an estimate for reduction in truck

accident involvement caused by improvement in geometric design elements was also calculated.

The percentage of reduction for the model could be specified to estimate the required variations

in the geometric properties. For this analysis, the Highway Safety Information System was used

to gather data from Utah for 1985 to 1989.

Aty and Abdelwahab presented an analysis of the effect of the geometric incompatibility

of light truck vehicles (LTV) on driver‘s visibility of other passenger cars involved in rear-end

16
crashes (30). The objective of this paper was to explore the effect of the lead vehicle‘s size on

rear-end crash configurations. Four types of rear-end crash configurations were taken: car-car,

car-truck, truck-car, and truck-truck. The GES databases were used in this analysis. Nested logit

models were calibrated to estimate the probabilities of the four crash configurations. These were

created as a function of the driver‘s age, gender, vehicle type, vehicle maneuver, light conditions,

driver‘s visibility, and speed. It was concluded from the results that the driver‘s visibility and

inattention in following a vehicle had the largest effect on being involved in a rear-end collision.

Also, the possibilities of a car-truck rear-end crash increased in cases where the lead vehicle

stopped suddenly.

Diener and Richardson studied truck-involved fatalities in Missouri, where nearly 70% of

those who die in traffic crashes are not wearing seatbelts (31). NHTSA determined a vehicle

involvement rate by dividing the number of vehicles involved in fatal rural/ urban crashes by the

vehicle miles traveled. As to laws regarding seat belts, Missouri is a secondary enforcement

state, meaning that drivers and passengers in violation of the law can only be cited when the

vehicle has been stopped by a police officer for a separate offense. In other words, a police

officer in Missouri cannot stop and cite a driver or passenger solely for not wearing a seat belt. A

survey was conducted in several districts and truck drivers were asked to respond to statements

such as ―If I were in a crash, I would want to have my seat belt on,” and the number drivers who

agreed to the questions and their level of agreement was noted and studied.

Burgess studied data from the FARS for the period 1994–2003 to compare characteristics

of fatal rural and urban crashes (32). The study found that there are approximately 42% more

fatal crashes in rural areas compared to urban areas; however, there are fewer vehicle miles

traveled in rural areas than urban areas. In addition, fatal rural crashes are more likely to involve

17
multiple fatalities, rollovers, and trucks. Fatal rural crashes more often occur on curved roadways

and have greater vehicle damage. Head-on crashes are more prevalent in rural areas than in urban

areas. Finally, the length of time for emergency medical services to arrive at the scene is longer

in rural areas than in urban areas.

2.6 Bayesian and Other Modeling Techniques

Majid investigated the effect of heavy commercial vehicles on the capacity and overall

performance of congested freeway section conditions (33). This seems to be an important

situation because the mixed traffic flow on the freeways has different impacts. This lack of

homogeny poses a problem for freeway operations and safety, especially when the truck traffic

percentage is on the higher in comparison to passenger cars. For this purpose, traffic surveys

were performed at two freeway sites in Tokyo and one freeway site in Melbourne. Video data

was filmed using six cameras for six hours at each site with the capacity of tracking a vehicle for

a distance of 700m. The data was microscopically analyzed and variables like the truck‘s

position (lead or lag vehicle), relative speed time gap, and space headways were estimated.

Using this data, various mathematical models were developed and nonlinear regression

techniques were performed in order to calibrate parameters for different T values in the models.

The most optimum models were estimated using optimum response variables, such as the

acceleration of the trucks at different times.

Duncan et al. illustrated the impact of the variable injury severity in truck-passenger car

rear-end collisions (34). For this, two objectives were targeted. The first objective was to

understand the factors that influence the passenger vehicle occupant injury severity in car-truck

rear-end collisions on divided roads. The second objective was to illustrate the use of the ordered

probit model application on particular factors of injury severity levels. For this project, the

18
Federal Highway Administration‘s HSIS database was used along with police reports and

roadway inventory data. The state of North Carolina was chosen for this analysis as it has a large

number of truck routes. The ordered probit model proposed for the given analysis had the

dependent variable (injury severity) coded as 0, 1, 2, 3, 4. The independent variables were factors

such as speed limit, light conditions, weather conditions, age, gender and so forth. From the

model, it was found that environmental and roadway conditions contributed majorly in these

types of crashes. Also factors such as darkness, high-speed differentials, and high-speed limits

were considerably significant.

Pickrell demonstrated in his study that while the overall proportion of passenger vehicle

drivers with alcohol involvement in fatal crashes is lower in older age groups, the median blood

alcohol concentration (BAC) was generally higher for those age groups (35). However, for

motorcycle operators, age groups with the highest levels of alcohol involvement also had the

highest median BAC levels. In order to understand the relationship between alcohol involvement

in fatal crashes and the median BAC levels of the drivers involved, this study examined FARS

data at several different levels, including level of alcohol involvement, median driver/operator

age, median BAC by age group within vehicle type, and median BAC by year and vehicle type

across all age groups. Data from 2004 are presented in the main body of the report, and data from

2000-2003 are included at the end of the report as a comparison of trends. This research work

identifies differences between age groups and within vehicle types, based on the proportion of

drivers with positive BACs: those greater than or equal to 0.01. This study accomplished this

objective by showing differences between passenger vehicles—passenger cars, SUVs, pickup

trucks, and vans—driver‘s and motorcycle operator‘s BAC levels across age groups. Passenger

vehicle drivers in the age groups 20-29 and 30-39 had the highest proportion of drivers with

19
positive BAC levels. However, motorcycle operators in the age groups 30-39 and 40-49 had the

highest proportion of drivers with positive BAC levels.

2.7 Multinomial Logistic Regression

Yan et al. conducted a study on rear-end collisions in trucks using two national crash

databases (2000-2004): the FARS and the GES (36). Overall and fatal truck-involved rear-end

collisions were both investigated in this paper. Three groups were used to classify two-vehicle

rear-end collisions in this study. Using the vehicle's striking/struck role as a basis, crash

categories were car-car (car hitting car), car-truck (car hitting truck), and truck-car (truck hitting

car). There was comparison of occurrence conditions of the three rear-end crash types so that

potential risk factors associated with the truck-involved crashes—such as driver characteristics,

highway designs, and road environments—could be identified. Multinomial logistic regression

results showed a significant association between overall truck-involved rear-end crashes and

factors such as gender, driver age, alcohol use, speed, day of week, interstate, weather condition,

divided/undivided highway, and lighting condition. There was also a significant association

between fatal truck-involved rear-end collisions and gender, driver age, alcohol use, day of

week, divided/undivided highway, and lighting condition. More information regarding effective

crash countermeasures and a better understanding of track-related rear-end crash risk are

provided by this study.

Yan et al. conducted another study by considering data from FARS for the years 2000-

2004 (37). Only two-vehicle angular crashes were considered. The crashes were then divided

into truck-car and car-car categories. The at-fault parameter in these categories was considered

and the truck-truck crashes category was excluded from the analysis. The dataset was further

filtered by citing as two-vehicle crashes those in which only one driver was at fault and the other

20
was not. Multi-logistic regression modeling was used in this project. The dependent variable is y,

which describes the type of crash. Pr(y=m|x) is the probability of observing outcome m given the

set of independent variables x.

Venkataraman and Mannering conducted a research study on motorcycle accident

severity which focused on univariate relationships between severity and an explanatory variable

of interest (e.g., helmet use) (38). The potential ambiguity and bias that univariate analyses

create in identifying the causality of severity has generated the need for multivariate analyses in

which the effects of all factors that influence accident severity are considered. This study

attempts to address this need by presenting a multinomial logit formulation of motorcycle rider

accident severity in single-vehicle collisions. Using 5 year statewide data on single-vehicle

motorcycle accidents from the state of Washington, they estimated a multivariate model of

motorcycle rider severity that considers environmental factors, roadway conditions, vehicle

characteristics, and rider attributes. Their findings show that the multinomial logit formulation

used was a promising approach to evaluate the determinants of motorcycle accident severity.

Moonesinghe et al. conducted a binary response model for rollovers (jackknifes) and

stated that the probability of a rollover (jackknife), given a single-truck fatal crash has occurred,

is a function of selected explanatory variables (39). If Y denotes the dependent variable in a

binary response model for rollovers (jackknifes), Y is equal to 1 if there is a rollover (jackknife)

and 0 if otherwise. The statistical problem was to estimate the probability that Y=1, considered

as a function of the explanatory variables. TIFA data were analyzed using a logit model, which is

a widely used binary-response model. The explanatory variables used in the models were

weather, light, speed limit, curve, weight, length, and width.

21
2.8 Countermeasure Ideas

Samuel et al. conducted a study about drowsy driver detection and the effects of

employing a warning system for commercial vehicle drivers (40). The research has been

conducted since 1996 by NHTSA and its partners, in order to quantify the loss of alertness

among commercial vehicle drivers. By experimentation, it was concluded that a valid measure of

loss of alertness among drivers can be made by the percentage of eyelid closure over the pupil

over time (Perclos). The first objective was to find drowsiness level distributions and differences

between these values with and without the DDWS. The second objective was to see variations in

drowsiness with a number of independent factors like age, nights of sleep, and so forth. From the

experimental analysis, it was concluded that further understanding was needed about highway

safety benefits, fleet acceptance, operational utility, and fatigue management practices so that

fatigue crash problems could be minimized.

Cate et al. presented the results of an evaluation of truck lane restrictions conducted using

the VISSIM microscopic traffic simulation software package as an analysis tool (41). The

objective of this application was to study truck lane restriction at a very detailed level. The

VISSIM traffic simulation model has a number of user-adjustable parameters such as lane usage,

free-flow speeds, lane changing behavior, vehicle power, weight, braking characteristics, and

traffic composition. The focus is on lane restrictions where large trucks are prohibited from using

the far-left travel lane on freeway sections with three or more lanes of travel in a single direction.

In order to make the results of the testing as realistic as possible, field traffic data was utilized to

create volumes and truck percentages representative of actual freeway operations. The

simulations were conducted in two scenarios; firstly, all vehicles were free to travel in any lane

and, secondly, trucks were restricted to the two right lanes of travel. After the simulations were

22
completed, the output files generated were used to calculate the performance statistics on factors

such as vehicle density, level of service, and average travel time. The ―aggressiveness‖ of lane

changes was seen to have increased by reducing the minimum distance and maximum speed

differential between vehicles. Another important measure that allowed for an evaluation of the

safety impact of truck lane restrictions was the frequency of lane changes. As the number of lane

changes decreased, the opportunity for collision was reduced by limiting the interaction between

the vehicles.

Reich et al. proposed an idea of exclusive highway facilities for trucks as a

countermeasure to reduce congestion, enhance safety, and improve free flow of freight (42). The

Florida Department of Transportation (FDOT) contracted with the Center for Urban

Transportation Research (CUTR) to lead this research project. The methodology used involved

selecting sites in Florida that warranted consideration for truckways or reserved truck lanes.

Important factors such as truck crash rates, truck volumes, and percent of trucks in traffic mix

were evaluated based on FDOT data. Then, GIS models were constructed and tested in selected

roadway segments to evaluate the considered parameters. It was concluded that most of the

interstate system is suitable for consideration of exclusive truck facilities. Truck congestion in

some areas appeared to have decreased by 15% by introducing this model. Crashes were also

estimated to decrease considerably.

Reiskin studied the proposal made by Stephen Kratzke, NHTSA‘s associate administrator

for rule making, at a truck part makers meeting in Las Vegas in February 2008 (43). In view of

reducing truck-involved fatalities, NHTSA was planning to release rules on brake stopping

distance, brake hose materials, and electronic and roll-stability control. The agency wanted to use

technology for this situation, not by proposing larger drum brakes or disc brakes, but by setting a

23
distance-based standard on the trucks. In April they also published a rule mandating electronic

stability control on all vehicles with gross vehicle weight ratings of less than 10,000 pounds by

2011. That would affect Class 2 trucks. Apart from these rules, the agency was also planning to

release regulations on brake hose standards and upgraded tire standards towards the end of the

year.

Murray et al. conducted a study in collaboration with the American Transportation

Research Institute focusing on driver-specific behaviors and events, and their relationship to

future truck crash involvement (44). Driver-specific data were used by the research team to

design and test a logistic regression model. The data was collected from the MCMIS and the

Commercial Driver‘s License Information System (CDLIS). Initially, statistical tests, including

Chi-square analyses, were done to assess the significant difference between future crash rates

and drivers‘ behavior. The regression model included specific violations discovered during

roadside inspections, driver traffic conviction information and past accident involvement. These

were taken as the independent variables and, through the model, the probability of crash

occurrences were obtained as the dependent variable. The variables named intercept, reckless

driving violation, serious speeding conviction, and hours of service violation seemed to be the

topmost crucial factors in reducing the crash scenario. From the analysis, several

countermeasures were recommended, which when effectively enforced, could bring the required

results.

24
 Chapter 3 Methodology

3.1 Data

Data for the study were procured from the National Highway Traffic Safety

Administration‘s FARS database. FARS database documents detailed data on vehicles, drivers,

roadways, and environmental conditions recorded in police crash reports, emergency medical

service reports, hospital records, and coroners‘ reports of all fatal crashes in the United States. It

contained details of fatal crashes in all 50 states, the District of Columbia, and Puerto Rico. This

database was conceived, designed, and developed by the National Center for Statistics and

Analysis (NCSA) to aid the traffic safety community in identifying traffic safety problems and

providing countermeasures for better driving standards (52). NCSA is a division of the NHTSA

that provides a wide range of analytical and statistical support to NHTSA. NCSA responds to

requests for data from various sources like state and local governments, research organizations,

private citizens, auto and insurance industries, Congress, and the media.

NHTSA has a contract with an agency in each state to obtain information on fatal

crashes. This information is compiled and put into a standard format by FARS analysts who are

state employees specially trained for this job. Fatal motor vehicle traffic crash data obtained from

various state agencies are assembled and coded on standard FARS forms. Various forms used in

assembling the information are Police Accident Reports (PARS), state vehicle registration files,

state driver licensing files, state highway department data, vital statistics, death certificates,

coroner/medical examiner reports, hospital medical records, and emergency medical service

reports. FARS was established in 1975 and data from the establishment date to the present is

available in several formats. FARS data is broadly used within NHTSA to answer many queries

on the safety of vehicles, drivers, traffic conditions, and roadways. Fatal crash reports can be

25
accessed at national and state levels by a FARS analyst acting in response to overall traffic safety

issues.

In order to make an entry into the database, a crash must involve a motor vehicle

traveling on a trafficway customarily open to the public, and must result in the death of an

occupant of a vehicle or non-motorist within 30 days of the crash. The FARS database includes

details of each and every such fatal crash reported. Each crash is characterized in terms of crash,

vehicle, roadway, and people involved with the help of more than 100 coded variables. All these

variables are reported on accident, vehicle, driver, and person forms, respectively. The accident

form contains information such as time and location, first harmful event, weather conditions

under which the crash occurred, number of vehicles, and people involved. Vehicle and driver

forms record details like vehicle type, impact points, most harmful event, and driver‘s license

status. The person form contains details about each individual involved in the crash, such as the

age and gender of the person; whether the person is the driver, passenger, or non-motorist; injury

severity; and restraint use. Individual privacy is maintained by protecting details such as name,

address and any other personal information. Overall alcohol estimates, which describe the

contribution of the alcohol factor in fatal crashes, as well as driver and non-occupant BAC

estimates, are present in the FARS alcohol file, which is an add-on to the data files when no

alcohol information would otherwise be available.

The FARS Encyclopedia is a web-based tool that facilitates downloading the data and

generating results through queries. It also consists of reports and fact sheets drawn from

published FARS data for the relevant year and state. The reports are classified under trends,

crashes, vehicles, and people. The ―trends‖ section covers motor vehicle crashes and fatalities

over a range of years while reports under ―crashes‖ present statistics about motor vehicle crashes

26
based on the injury severity of the person. The ―vehicles‖ section presents details about the kinds

of vehicles involved in fatal motor vehicle crashes. Reports under the people section provide data

on the kinds of people, drivers, passengers, or non-motorists, involved in motor vehicle crashes.

The FARS Query System is a web interface that allows users to perform their own custom

queries such as case listings and univariate and cross tabulations. FARS data files are available in

an archive as a public resource to download in file transfer protocol (FARS FTP). This website

enables users to process the data using their own computer systems.

From this database, truck and non-truck crashes were the two categories examined in the

comparative study. In this study, a truck crash was defined as a crash which involved at least one

truck whose gross body weight was greater than 10,000 pounds. A non-truck crash was defined as

a crash which did not involve a truck. In the FARS database, trucks were divided into different

categories depending on their GVWR. Trucks considered for this study were vehicles with the

following body type codes:

 61: a single-unit straight truck with GVWR greater than 10,000 lbs. and less than

or equal to 19,500 lbs.

 62: a single-unit straight truck with GVWR greater than 19,500 lbs. and less than

or equal to 26,000 lbs.

 63: a single-unit straight truck with GVWR greater than 26,000 lbs.

 64: a single-unit straight truck with unknown GVWR

 66: a truck/tractor with any number of trailing units and any weight

 67: a medium/heavy pickup truck with GVWR greater than 10,000 lbs.

 71: any unknown single-unit or combination unit medium truck with GVWR

greater than 10,000 lbs. and less than 26,000 lbs.

27
  72: any unknown single-unit or combination-unit heavy truck with GVWR greater

than 26,000 lbs.

 78: any unknown medium/heavy truck type

 79: an unknown truck type in the FARS database

These specific body types were considered as they included trucks which had a gross body

weight greater than 10,000 pounds. All other motor vehicles, other than those body types and ones

which had a gross body weight less than 10,000 pounds, were considered as non-truck vehicles.

Files from the database were merged using unique crash, person, and vehicle

identification codes and by employing SAS computing software (53). The merged files were

checked so as to obtain a unique unduplicated crashes, people, and vehicles list to retrieve

frequencies or counts of different characteristics. Various crash characteristics were obtained

using filtering techniques in Microsoft Excel and Access. After suitably merging and filtering

accident, person, and vehicle files, fatal truck crash data for the five-year time period from 2003

to 2007 was combined and truck and non-truck crash cases were separated to obtain consolidated

results with respect to several parameters.

Further, the values obtained were compared at various levels to analyze trends and

patterns of specific crash parameters with respect to time or type of crash, or the extent of fault

for the drivers involved. Also, certain pairs of parameters were selected to observe differences in

the combination of conditions prevailing during higher crash-occurrence levels. Eventually

driver, crash, and vehicle-related factors were extracted to compare the existence of these factors

in both truck and non-truck crashes.

28
3.2 Analysis Methods

3.2.1 Bayesian Statistical Approach

The Bayesian statistical approach is an effective tool in recognizing the predominance of

crash-related factors while comparing truck and non-truck crashes in the given data set. The

computation of likelihood ratios, using Bayesian posterior probabilities, is valid and useful. It

makes good logical sense, while producing significant results from projected analysis of crash

factors.

Equation 3.1 describes the conditional probability of the occurrence of a driver, vehicle,

or crash-related contributory cause (CC), given that it is a truck crash.

P(Truck / CC )  P(CC )
P(CC / Truck )  (3.1)
P(Truck )

where,

P(Truck/CC) = Probability that the crash was a truck crash, given that a

specific contributory cause was reported. As shown in Equation 3.2, this

value is estimated from the data by considering total number of crashes

and those in which a truck crash and its contributory factor were coded

together.

P(CC) = Overall probability of the specific driver, vehicle, or crash-related cause

being reported as a contributing factor, and as shown in Equation 3.4, is

estimated from the numbers of cases in which the CC was reported in

the dataset.

29
 P(Truck) = Overall probability that a crash was a truck crash and was estimated

from the data as shown in Equation 3.3.

Number of Truck Crashes with that Contributo ry Factor

P(Truck / CC )  (3.2)
Number of All (Truck and Non - Truck ) Crashes with that Factor

Number of Truck Crashes

P(Truck )  (3.3)
Number of All (Truck and Non - Truck ) Crashes

Number of Crashes with that Contributo ry Factor

P(CC )  (3.4)
Number of All (Truck and Non - Truck ) Crashes

Similarly, the conditional probability of a contributory cause for a given non-truck crash

is estimated, and the ratio of these probabilities generates the likelihood ratio of that contributory

factor as shown in Equation 3.5.

P(CC / Truck Crash)

Likelihood Ratio  (3.5)
P(CC / Non - Truck Crash)

The likelihood ratio of a given contributory factor being recorded in a truck crash as

compared with a non-truck crash was assessed from crash records. This likelihood ratio is the

probability of a crash being a truck crash when the contributory factor was recorded, as

compared with the probability of a crash being a non-truck crash when the same contributory

factor was identified. The larger the likelihood ratio, the greater the association between the

contributory factor and truck crashes relative to non-truck crashes.

30
3.2.2 Multinomial Logistic Regression

Multinomial logistic regression modeling, which was also used in this study, is an

efficient tool to analyze crash data (36, 37, 38, and 39). The dependent variable in this modeling

technique is denoted as y, which describes the type of crash. Pr(y=m|x) is the probability of

observing outcome m given the set of independent variables x. It is assumed to be a linear

combination xβm.

Pr (y i =m |x i) =exp (x i βm) /∑ J j=1 exp (x i βj) (3.6)

where,

Y=1: Truck Crash

Y=2: Non-Truck Crash

Pi1= exp (x i β1) / (1+ exp (x i β1)+ exp (x i β2)) (3.7)

Pi2= exp (x i β2) / (1+ exp (x i β1)+ exp (x i β2)) (3.8)

where,

Pi1= Probability that the crash type is 1 for observation i.

Pi2= Probability that the crash type is 2 for observation i.

In this study, the SAS LOGISTIC procedure was used to perform the multinomial logistic

regression. The dependent variable was the type of crash which took the binary form depending

on whether it was a truck crash or non-truck crash. The independent variables considered were

driver age, gender, national highway, light condition, weather condition, alcohol use, and 35

other factors. These variables included several crash, driver, vehicle, and environmental factors

using statistical modeling software SAS version 9.1. As the selection criteria of variables to be

included in the model, a 95% confidence level was used in which the probability should be less

31
than 0.05. Colinearity of individual variables was also checked before considering variables in

the model, and if such a relationship existed, one of the two correlated variables was discarded

based on the lowest mean value criterion.

The LOGISTIC procedure used in developing this model fits linear logistic regression

models for binary or ordinal response data by the method of maximum likelihood. The maximum

likelihood estimation is carried out with either the Fisher-scoring algorithm or the Newton-

Raphson algorithm (39).

The LOGISTIC procedure provides four variable selection methods: forward selection,

backward elimination, stepwise selection, and best subset selection. The best subset selection is

based on the likelihood score statistic. This method identifies a specified number of best models

containing one, two, three variables and so forth, until a single model containing all the

explanatory variables is achieved (39).

Odds-ratio estimates are displayed along with parameter estimates. You can also specify

the change in the explanatory variables for which odds-ratio estimates are desired. Confidence

intervals for the regression parameters and odds ratios can be computed based either on the

profile likelihood function or on the asymptotic normality of the parameter estimators.

The Wald Chi-Square and Pr > ChiSq are the test statistics and p-values, respectively, for

the hypothesis test that an individual predictor's regression coefficient is zero given the rest of

the predictors are in the model. The Wald Chi-Square test statistic is the squared ratio of the

estimate to the standard error of the respective predictor. The probability that a particular Wald

Chi-Square test statistic is as extreme as, or more so, than what has been observed under the null

hypothesis is given by Pr > ChiSq.

32
 The ―Model Fit Statistics‖ in Table 4.8 contain the Akaike Information Criterion (AIC),

the Schwarz Criterion (SC), and the negative of twice the log likelihood (-2 Log L) for the

intercept-only model and the fitted model. AIC and SC can be used to compare different models,

and those with smaller values are preferred (39).

Other goodness-of-fit parameters, which the LOGISTIC procedure provides in the output,

are described as follows (39):

 Percent Concordant: A pair of observations with different observed responses is said to

be concordant if the observation with the lower ordered response value has a lower

predicted mean score than the observation with the higher ordered response value.

 Percent Discordant: If the observation with the lower ordered response value has a higher

predicted mean score than the observation with the higher ordered response value, then

the pair is discordant.

 Percent Tied: If a pair of observations with different responses is neither concordant nor

discordant, it is a tie.

 Pairs: This is the total number of distinct pairs.

 Somer's D: A measure used to determine the strength and direction of relation between

pairs of variables. Its values range from -1.0 (all pairs disagree) to 1.0 (all pairs agree). It

is defined as (nc-nd)/t where nc is the number of pairs that are concordant, nd the number

of pairs that are discordant, and t is the number of total number of pairs with different

responses.

 Gamma: The Goodman-Kruskal Gamma method does not penalize for ties on either

variable. Its values range from -1.0 (no association) to 1.0 (perfect association). Because

33
 it does not penalize for ties, its value will generally be greater than the values for Somer's

D.

 Tau-a-Kendall's Tau: A modification of Somer's D to account for the difference between

the number of possible paired observations and the number of paired observations with

different responses. It is defined to be the ratio of the difference between the number of

concordant pairs and discordant pairs to the number of possible pairs (2(nc-nd)/(N(N-1)).

 C: This is another measure of rank correlation of ordinal variables. It ranges from 0 (no

association) to 1 (perfect association).

These goodness-of-fit parameters could be used to evaluate the robustness of a developed

multinomial logistic regression model.

34
 Chapter 4 Results and Discussion

4.1 Characteristics of Fatal Truck Crashes

Analysis of the data showed that large trucks contribute to more fatalities in other (non-

truck) vehicles than in trucks themselves. On average 84% of fatalities occurring in large truck

crashes in the United States are not occupants of trucks. This section elaborates on the

characteristic analysis conducted on fatal truck crashes in the United States, which used five

years of crash data from 2003 to 2007.

Initial Point of Impact

One observation made from fatal truck crash data was the direction of impact, which is

the initial point on the truck where the other vehicle collides. As shown in figure 4.1, trucks have

blind spots in all directions, and the initial impact point helps in showing which zone is more

critical for a higher crash risk. By observing the initial point of impact on the truck, the position

of the colliding vehicle with respect to the truck was estimated. From this, the blind spot which

results in a higher crash rate was interpreted. From figure 4.1, it is seen that almost 62.5% of the

cases resulted in trucks having the initial impact on their front side. This might weaken the

argument that the poor visibility ranges on the rear side of trucks leads to a majority of rear-end

crashes in trucks. It is possible that other vehicle drivers need to be more vigilant by driving in

front of rather than by the rear of a truck. Around 15.5% of the crashes were on the left-hand side

of the truck driver. This could be a significant observation because from figure 4.1, it was

observed that the left-hand side of the truck driver has the smallest blind spot zone when

compared to all other directions.

35
 70%
62.50%
60%

50%
% of Crashes

40%

30%

20% 15.50%
9.40%
10% 6.50%
2.70% 3.30%
0%
Collision with Front Side Right Hand Left Hand Rear Side Unknown
Object Side of the Side of the
Driver Driver
Point of Impact for the Truck

Figure 4.1. Point of Impact for Trucks in Fatal Crashes

Alcohol Involvement

The alcohol involvement of drivers is potentially one of the most important contributory

factors resulting in crashes, which could also be the case in truck crashes. Analysis showed that

of all the drunken drivers involved in fatal truck crashes, only 12.7% were truck drivers and the

remaining 87.3% were non-truck drivers with blood alcohol levels higher than the 0.08 mg/ml

limit. This eliminates the misconception that a larger percentage of truck drivers are under

influence of alcohol/drugs leading to fatal crashes. Hence, it can be deduced that in fatal truck

crashes with alcohol involvement, non-truck drivers are more likely to be under the influence of

alcohol than truck drivers.

36
Manner of Collision

The manner of collision of trucks in fatal crashes was observed from the combined

dataset for the period of 2003-2007 and the results are shown in figure 4.2. Angle crashes have

the highest proportion with 34.2%, followed by 23.7% of cases in which the vehicles collided

with a fixed object like a tree, guardrail, or other object. Head-on and rear-end crashes also form

a significant portion of crashes resulting in fatalities.

40%
34.2%
35%

30%
% of Fatal Crashes

23.7%
25%
20.4%
20%
16.8%
15%

10%

5% 2.0% 2.3%
0.4%
0%
Single Rear End Head On Angle Sidesweep Sidesweep Unknown
Vehicle same opposite
Crash Direction Direction
Manner of Collision

Figure 4.2. Manner of Collision of Fatal Truck Crashes

Speed Limit

Trucks are more difficult to maneuver smoothly as compared to smaller vehicles, and at

higher speeds they have a higher risk of losing control. This can also be one of the factors

contributing to a higher risk of crashes involving trucks. The speed limit of the roadway where

37
the truck is traversing before succumbing to a fatal crash can approximately show the speed of

the truck. As seen in figure 4.3, the percentage of fatal crashes increases with increase in speed

limit up to 60 mph. The range of 51-60 mph has the highest number (an average of 5,280 crashes

per year) of fatal truck crashes in the past five years. The sudden drop in the number of crashes

from 51-60 mph to 61-70 mph may be because of the smaller number of roadways with the latter

speed range.

45%
39.5%
40%

35%
29.9%
30%
% of Crashes

25%

20%
14.6%
15%

10% 8.2%

5% 3.4% 2.7%
1.3%
0.1% 0.2%
0%
No Speed 1-20 21-30 31-40 41-50 51-60 61-70 71-80 Unknown
Limits Posted Speed Range (mph)

Figure 4.3. Fatal Truck Crashes in Different Speed-Limit Ranges

Truck Driver Age

A number of driver-related parameters can be responsible for influencing crash risk,

especially for trucks which travel on a commercial basis for longer and more strenuous hours. In

a study by Crum and Morrow, they explain that truck driver fatigue plays a major role in the

38
occurrence of a crash. They investigated and established a driver fatigue model to test various

carrier scheduling practices with other driver parameters (45). Another study was done by

Williams et al., to scale the amount of responsibility in drivers by age and gender for all motor

vehicle crashes (46). Here, they compared the number of drivers at fault in different age groups

and gender. From their analysis, they proved that the element of ―responsibility‖ declined with

age until about age 63, at which point it then increased as a function of age.

From Figure 4.4 it is evident that the number of drivers involved in fatal truck crashes is

higher in the age range of 41-50 years than in other groups. With the highest percentage, 29%,

being in this range the graph has an overall normal distribution. Until the range of 41-50 years,

the percentage of fatal truck crashes has an increasing curve and after that range the percentage

of crashes undergoes a decreasing trend.

39
 35%

30% 29.0%
27.3%
% involved in Fatal Crashes

25%

19.5%
20%
15.3%
15%

10%
6.6%
5%
0.8% 0.9% 0.1% 0.6%
0%
11-20 21-30 31-40 41-50 51-60 61-70 71-80 81-90 Unknown
Truck Driver Age Group (yrs)

Figure 4.4. Age of Truck Drivers Involved in Fatal Truck Crashes

Types of Trafficways

Truck maneuverability may become more challenging depending on the type of roadway.

Depending on roadway characteristics, even actions like lane changing and lane merging can

sometimes become critical factors contributing to a crash. Also, the presence of physical dividers

is likely to affect the number of fatal crashes because they have the potential to reduce the

severity of a crash and sometimes may even prevent fatalities.

A majority of almost 55.2 % of fatal truck crashes, as shown in figure 4.5, have occurred

on two-way trafficways with no physical division. This shows that this kind of roadway has a

greater potential for fatal crash occurrences. Traffic flowing in opposite directions with no

physical division in between can be one high-risk situation in which the smallest of human errors

40
can result in highly severe crash scenarios. Roadways of this type should be improved by

providing the necessary divisions so as to minimize the frequency of fatal truck crashes.

The number of lanes on two-way trafficways with 55.2 % of crashes was analyzed, and it

has been observed that almost 77.3% of those crashes occurred on two-lane two-way roadways.

The difficulty in controlling the large size of the vehicle in narrow or smaller roadways can be

the reason for this high frequency. Two-lane roadways are often congested and cannot be easily

traversed. This situation as well as the two-way trafficway without any physical division can set

the stage for the occurrence of a fatal truck crash.

One way
Trafficway Unknown
Divided Highway (0.68%) (1.23%)
Without Traffic
Barrier
(12.68%)

Two-Way
Undivided Trafficway with
Highway Without no Physical
Traffic Barrier Division
(30.21%) (55.20%)

Figure 4.5. Proportion of Fatal Truck Crashes on Different Traffic Flowways

41
Level of Deformation on Urban and Rural Roadways

As seen in figure 4.6, the level of deformation of the vehicles involved in fatal truck

crashes is severely disabling in most cases, which is consistent in both urban and rural roadways.

As large trucks are heavy in weight and volume and since a majority of fatal truck crashes occur

at high speed levels (as was observed in Fig. 4.3), it is evident that consequences of such

conditions result in severe damages to the collided vehicles. However, the percent of severely

disabled vehicles is proportionally smaller in urban areas when compared to rural areas. The

availability of greater space for maneuvering on urban roads along with lower speeds, a

reduction due to the higher traffic volumes, could probably be the reasons for this observation.

80%
73.7%

70%
61.0%
60%
None
50%
% of Crashes

Other (Minor)
40% Functional (Moderate)
Disabling (Severe)
30% Unknown
18.4%
20% 15.0%
12.8%
8.4%
10% 5.5%
1.9% 1.0% 2.3%
0%
Rural Roadways Urban Roadways
Type of Roadway

Figure 4.6. Level of Deformation of all Vehicles Involved in Fatal Truck Crashes

42
Truck Driver At-Fault Factors

Various types of truck driver-related factors have contributed to fatal crashes as shown in

figure 4.7. Around 28.1% of truck drivers have contributed to fatal truck crashes due to non-

compliance with traffic regulations. Improper driving is another factor, which in 24.6% of cases

has contributed to fatal truck crashes. Categories of improper driving include factors such as

running off the road, erratic lane change, following improperly, failure to keep in lane properly,

and so forth. Also, as the figure shows, 15.8% of truck drivers involved in fatal truck crashes had

some type of mental/physical condition—such as fatigue, drowsiness, inattentiveness, or the

influence of drugs—that contibuted to the occurance of a fatal truck crash.

Mental Condition
Improper Driving (15.8%)
(24.6%)

Illegal Driving (9.7%) Non compliance to

traffic regulations
(28.1%)

Possible Distractions,
(8.2%)

Vision Obscured
Miscellaneous
Conditions (3.2%)
Factors (5.7%) Environmental
Conditions (4.8%)

Figure 4.7. Truck Driver-Related Contributory Factors in Fatal Crashes

43
4.2 Truck Striking/Struck Comparison

Truck Striking or Being Struck on Different Roadways

In this section fatal truck crashes are divided into two categories: truck striking and truck

struck. In the former, the truck strikes another vehicle first in the crash and in the latter the truck

is struck first by another vehicle thus resulting in a crash. The analysis was done by comparing

these two impact categories with one another. A similar framework was adapted to the current

data set, as shown in figure 4.8, to observe the crashes on different types of roadways over the

past five years.

It was observed that the truck striking and truck struck categories have a high number of

crashes on state highways as compared with other crashes which have a high number of crashes

on interstates rather than other types of roadways. A truck striking another vehicle results in a

higher number of crashes than a truck being struck on both interstates and state highways. It is

pertinent to note that this comparison has equal proportions in the case of U.S. highways.

44
 40
37.8%
35
28.9% 29.3%
30 26.9% 25.8% 26.4%
25.3 25.4
% of Crashes

25 Interstate

19.1% U.S.Highway
20 18.5% 17.1 18.1%
State Highway
15 Other roads

10

0
Truck Striking Truck Struck Other Crashes
Type of Crash

Figure 4.8. Fatal Truck Crashes by Roadway Type in Truck Striking/Struck Conditions

Truck Striking/Struck under Variable Light Conditions

When truck striking and truck struck were studied under different light conditions, it was

observed that the proportion of cases where trucks are struck was smaller under daylight

conditions than cases where the truck strikes other vehicles, as shown in figure 4.9. In constrast,

the percentage of trucks being struck is higher in dark or dark but lighted conditions when

compared to cases of trucks striking other vehicles.

45
 80
67.1% 71.7%
70
61.0%
60
% of Crashes

50 Daylight
40 34.7% Dark/Dark Lightened
29.2%
30 24.6% Dawn/Dusk
20
10 4.3% 3.4%
3.7%
0
Truck Striking Truck Struck Other Crashes
Type of Crash

Figure 4.9. Truck Crashes in Different Light Conditions under Striking/Struck Types

4.3 Comparison of Characteristics of Fatal Truck and Non-Truck Crashes

Fatal crash data for the period of 2003-2007 was divided into crashes which involved

trucks and those which did not involve trucks, or non-truck crashes. Different characteristic

factors such as initial point of impact, driver age, posted speed limits, manner of collision, level

of deformation, the rural/urban split, types of traffic flowways, and roadway categories were

compared between truck and non-truck crashes. Percentages in each sub-category were

calculated by taking the total number of truck or non-truck crashes as the base value.

It can be seen from figure 4.10 that the initial impact point for vehicles in both truck and

non-truck fatal crashes was mostly on the front side. Although all other categories had lower

proportions in truck and non-truck crashes, the left-hand side of the driver as the impact point

had a comparatively larger proportion of fatal crashes in trucks than in non-trucks.

46
 70%
62.2% 62.7%
60%
Trucks
50%
% of Crashes

Non-Truck
40% Vehicles

30%

20% 16.2%
9.3% 9.5%10.8%
10% 5.7% 6.6% 6.6%
2.5% 3.0% 4.9%
0%
No Collision Front Side Right Hand Left Hand Side Rear Side Unknown
with the Truck Side of the of the Driver
Driver
Point of Impact for the Truck

Figure 4.10. Initial Impact Point for Truck and Non-Truck Crashes

Furthermore, a larger proportion of truck drivers involved in fatal crashes seemed to be of

the age group 41-50 years, whereas the non-truck drivers were mostly in the 21-30 years age

group. Figure 4.11 shows that starting from the age group of 31-40 truck drivers had a larger

involvement than non-truck drivers in fatal crashes.

When the overall trend lines in both truck and non-truck drivers were observed, there was

a difference in the pattern. Truck drivers had almost a normal distribution with the line, peaking

at the age range of 41-50 years, whereas non-truck drivers had the trend line skewed towards the

younger population with the peak at the 21-30 years. This showed that younger drivers have a

larger proportion of involvement in non-truck crashes and middle-aged drivers have a larger

involvement in truck crashes.

47
 35%

30% 29.0%
27.3%

24.4%
25%
Truck Drivers
% of Crashes

19.5%
20%
Non-Truck
15.5% 15.3% 15.9% Drivers
15.2%
15%
10.8%
10%
6.6% 6.6%
5.2%
5% 2.9% 2.1%
0.8% 0.9% 0.6%
0.1%
0%
11-20 21-30 31-40 41-50 51-60 61-70 71-80 81-90 Unknown
Driver Age Group (yrs)

Figure 4.11. Driver Age for Truck and Non-Truck Drivers

Distribution of truck and non-truck crashes in different speed limit ranges is shown in

figure 4.12. In both truck and non-truck crashes, the maximum number of crashes are in the 51-

60 mph range. In the speed-limit range of 21-50 mph, non-trucks had more fatal crashes than

trucks, whereas between 51-70 mph trucks seemed to have more fatal crashes than non-trucks.

This shows that in lower speeds non-trucks have a higher proportion of fatal crashes, and in

higher speeds trucks have a higher proportion of crashes.

48
 45%
39.5%
40%

35% 33.3%
Truck Crashes
29.9%
30%
Non-Truck
% of Crashes

25% Crashes

19.1%
20% 18.1%

14.6% 14.5%
15%
10.1%
10% 8.2%

5% 3.4%
2.7% 2.7%
1.3% 1.3%
0.1% 0.3% 0.2% 0.5%
0%
No Speed 1-20 21-30 31-40 41-50 51-60 61-70 71-80 Unknown
Limits
Posted Speed Range (mph)

Figure 4.12. Posted Speed Limit for Truck and Non-Truck Crashes

As demonstrated by figure 4.13, a majority of fatal non-truck crashes were single-vehicle

crashes but most of the fatal truck crashes were angle crashes. Also proportionately, there were

more rear-end, head-on and angle crashes involving trucks than non-trucks.

49
 70%
61.6%
60%
Truck Crashes
50%
Non-Truck
Crashes
% of Crashes

40%
34.2%

30%
23.7%
20.4% 20.3%
20% 16.8%

9.9%
10%
5.4%
2.0% 2.3%
1.2% 1.1% 0.4%0.5%
0%
Single Rear End Head On Angle Sidesweep Sidesweep Unknown
Vehicle Same Opposite
Crash Direction Direction
Manner of Collision

Figure 4.13. Manner of Collision for Truck and Non-Truck Crashes

Both truck and non-truck fatal crashes most commonly resulted in disabling vehicle

deformations as shown in figure 4.14. However, fatal non-truck crashes had a higher percentage,

78.4%, of severe and/or disabling vehicle deformations than fatal truck crashes. Also, it was

observed from figure 4.15 that more than half of the crashes in trucks and non-trucks occurred on

two-way trafficways with no physical division. Fatal non-truck crashes had a higher percentage,

69.9%, of occurrence on two-way trafficways with no physical division than fatal truck crashes.

Other types of traffic flowways, such as divided highways with or without traffic barriers, were

observed to have a larger proportion of truck crashes than non-truck crashes.

50
 90%
78.4%
80%
69.3% Truck
70% Crashes

60% Non-Truck
Crashes
% of Crashes

50%

40%

30%

20% 16.2%
9.9% 12.0%
10% 5.7%
3.2% 2.5%
1.5% 1.5%
0%
None Other (Minor) Functional Disabling Unknown
(Moderate) (Severe)

Level of Deformation

Figure 4.14. Level of Deformation for Truck and Non-Truck Crash Vehicles

80%
69.9%
70%
Truck Crashes
60% 55.2%
Non-Truck Crashes
% of Crashes

50%

40%
30.2%
30%

20% 18.3%
12.7%
10% 8.6%
0.7% 0.9% 1.2% 2.4%
0%
Two-Way Divided Divided One way Unknown
Trafficway with Highway Highway Trafficway
no Physical Without Traffic Without Traffic
Division Barrier Barrier
Type of Trafficflow Way

Figure 4.15. Trafficway Type for Truck and Non-Truck Crashes

51
 Arterial roadways in both urban and rural sectors had a higher predominance of fatal

truck crashes, whereas collector and local roads had a higher predominance of non-truck crashes,

as shown in figure 4.16.

40%
35%
35%
Truck Crashes
30% Non-Truck Crashes
25%
25%
22%
% of Crashes

20% 18%
15%
15% 13%
12% 11%
10% 9% 9%
10%
6%
5% 4% 4% 3%
2%
0%
Rural Rural Rural Rural Urban Urban Urban Urban
Principal Minor Collector Local Principal Minor Collector Local
Arterial Arterial Road Arterial Arterial Road

Rural/ Urban

Figure 4.16. Rural Urban Contrast for Truck and Non-Truck Crashes

Different types of roadways on which truck and non-truck crashes occurred are shown in

figure 4.17. Trucks had a larger proportion of fatal crashes on interstates and highways, whereas

other county and municipality roads had a higher proportion of fatal non-truck crashes. A larger

presence of trucks on these major arterials and roadways might be one of the causes for this high

proportion of fatal truck crashes.

52
 35%
30.3%
30% 28.1% Truck Crashes
25.1%
24.2% Non-Truck
25%
21.3% Crashes
% of Crashes

20%
15.1%
15% 14.5%
10.8%
9.1%
10%
6.8%
5.1% 4.7%
5%
1.5% 2.6%

0%
Interstate U.S. State County Township Municipality Other
Highway Highway Road
Type of Roadway

Figure 4.17. Type of Roadway for Fatal Truck and Non-Truck Crashes

80%
72.0% 75.5%
70%

60%
Truck Crashes
50%
% of Crashes

Non-Truck Crashes

40%

30%

20%
13.5%
12.0%
8.9%
10% 6.9%
1.3% 1.3% 1.5% 1.3% 2.1%1.7% 0.9%1.3%
0%
1 2 3 4 5 6 or more Unknown
Number of Lanes

Figure 4.18. Number of Lanes on Roadways Where Truck/ Non-Truck Crashes Occurred

53
 In addition, factors such as alcohol involvement and cellular usage were also analyzed. Of

all fatal truck crashes which had some alcohol involvement, it was seen that in 87% of cases non-

truck drivers were the ones involved in alcohol consumption and truck drivers were under the

influence of alcohol in only 12% of cases. Also, cellular usage was among the top 10 driver-

related contributory factors for truck drivers involved in fatal crashes.

4.4 Bayesian Statistical Analysis: Contributory Causes for Truck and Non-Truck Crashes

The following section shows the likelihood of contributory factors occurring in fatal truck

crashes when compared to fatal non-truck crashes. If the probability of the factor is greater than

one it indicates the factor was more predominant in fatal truck crashes than fatal non-truck crashes.

Factors in the tables that follow belong to categories of driver-related, vehicle-related, or crash-

related issues. The likelihood ratios are recorded in descending order of predominance in each

category. Each crash might have more than one contributory factor leading to the event since

FARS records up to four driver-related, three crash-related, and two vehicle-related factors per

crash. Hence, the sum of the number of crashes in truck and non-truck categories will not be equal

to the number of crashes that occurred in the time considered.

Table 4.1 shows crash-related contributing factors in 11 different categories as defined by

the FARS database. Crash data for the period of 2003-2007 was considered for this analysis.

Recent previous crash nearby/ vehicle set in motion by a non-driver, work area conditions, poor

shoulder conditions, and inadequate warning signs are the topmost factors which are more likely to

contribute to truck crashes than to non-truck crashes. Providing sufficient signs in work zones and

all other areas, as well as improving shoulder conditions might help reduce fatal truck crashes.

54
 Table 4.1. Conditional Probabilities and Likelihood Ratios for Crash-Related Factors

Conditional
Conditional
Number Number of Probability of
Probability of This Likelihood
Crash-Related Factor(CF) of Truck Non-Truck This CF Given a
CF Given a Fatal Ratio
Crashes Crashes Fatal Truck
Non-Truck Crash
Crash
Recent Previous Crash
Nearby/ Vehicle Set in 416 1025 0.01901 0.00602 3.15
Motion by a Non-Driver
Motor Vehicle Struck by
558 1496 0.02550 0.00879 2.90
Falling Cargo
Construction/ Work Area
122 342 0.00557 0.00201 2.77
Condition
Inadequate Warning of
15 57 0.00069 0.00033 2.04
Exits, etc.
Aggressive Driving or
Road Rage of Non- 102 391 0.00466 0.00230 2.02
Contact Vehicle Driver
Poor Shoulder Condition 22 158 0.00101 0.00093 1.08

Within Designated
6 51 0.00027 0.00030 0.91
School Zone

Poor Roadway Condition 33 443 0.00151 0.00260 0.57

Speed Limit Is a Statutory
61 1004 0.00279 0.00590 0.47
Limit but is not Posted
Police Pursuit Involved 57 1557 0.00260 0.00915 0.28

Vehicle-related contributory factors between fatal trucks and non-truck crashes are listed

in Table 4.2. As most of the utility vehicles are trucks rather than other motor vehicles, this

cannot be considered a contributory cause, but defective brake systems, having the second

highest likelihood ratio, seems to be more predominant in truck crashes rather than other vehicle

crashes. Defective lights, mirrors, and engines also appear to have more likelihood of

contributing to crashes because of the severe wear and tear trucks undergo as a result of long

miles traveled. These factors, recorded as vehicle-related factors, are subjective with respect to

police officers present at crash sites. As officers are not professional vehicle inspectors, these

records might not be absolutely precise.

55
 Table 4.2. Conditional Probabilities and Likelihood Ratios for Vehicle-Related Factors

Conditional Conditional
Number of Number of Probability of Probability of
Likelihood
Vehicle-Related Factor (VF) Truck Non-Truck This VF Given This VF Given
Ratio
Crashes Crashes a Fatal Truck a Fatal Non-
Crash Truck Crash
Vehicle Identified as
Utility/Emergency/Other 188 80 0.00859 0.00047 18.27
Working Vehicle
Defect in Brake System 445 421 0.02033 0.00247 8.22
Defects in
89 260 0.00407 0.00153 2.66
Lights/Horn/Mirror/Wiper
Defects in
Steering/Suspension/Engine/ 77 263 0.00352 0.00155 2.27
Exhaust System
Other Vehicle
Defects(Wheels/Doors/Safety 124 499 0.00567 0.00293 1.93
Belts/Air Bags)
Defective Tires 358 2501 0.01636 0.01470 1.11
Identified Vehicle
65 581 0.00297 0.00341 0.87
Registration as Handicapped
Identified as a Hit-and-Run
306 7727 0.01398 0.04540 0.30
Vehicle
Vehicle Went Airborne
57 1489 0.00260 0.00875 0.29
During Crash
Vehicle Set in Motion by
Another Vehicle/Non- 9 316 0.00041 0.00186 0.22
Motorist

FARS records 94 different driver-related factors which include mental, psychological, vision

obscured, environmental, and other miscellaneous factors. Of these 94 factors, only those which

reasonably reflect the truck driver contributing to the occurrence of the crash were included here. As

shown in Table 4.3, the conditional probability of each driver‘s contributory factor in truck and non-

truck crashes and their likelihood ratios were estimated. Factors having a considerable number of

frequencies were selected, and results were listed in descending order of their likelihood ratios.

56
 Table 4.3. Conditional Probabilities and Likelihood Ratios for Driver-Related Factors

Conditional Conditional
Non- Probability of Probability of
Truck Likelihood
Driver-Related Factor(DF) Truck This DF Given a This DF Given a
Crashes Ratio
Crashes Fatal Truck Fatal Non-Truck
Crash Crash
Stopped or Unattended Vehicle 501 1019 0.02289 0.00599 3.82
Following Improperly 903 1902 0.04126 0.01118 3.69
Starting or Backing Improperly 147 349 0.00672 0.00205 3.27
Overloading or Improper Loading of
111 309 0.00507 0.00182 2.79
the Vehicle
Making Improper Exit or Entry 76 287 0.00347 0.00169 2.05
Erratic Lane Change 525 2129 0.02399 0.01251 1.91
Cellular Telephone in Use in Driving 765 3488 0.03496 0.02049 1.70
Signal Inattention/Unfamiliar
128 643 0.00585 0.00378 1.54
Roadway
Passing with Insufficient Distance or
Inadequate Visibility or Failing to 283 1700 0.01293 0.00999 1.29
Yield to Overtaking Vehicle
Driving on Wrong Side of the Road 557 3379 0.02545 0.01985 1.28
Failure to Yield Right of Way 2968 18801 0.13562 0.11047 1.22
Failure to Obey Traffic Rules 1688 10899 0.07713 0.06404 1.20
Drowsy ,Sleepy, Fatigued 683 4499 0.03121 0.02644 1.18
Tire Blow Out or Flat Tire 134 887 0.00612 0.00521 1.17
Inattentive(Talking, Eating) 2569 17407 0.11739 0.10228 1.14
Driving/Passing in Prohibited or
83 701 0.00379 0.00412 0.92
Wrong Direction
Passing Where Prohibited by Posted
104 900 0.00475 0.00529 0.89
Signs
Failing to Dim Lights or Have
39 338 0.00178 0.00199 0.89
Lights When Required
Other Non-Moving Traffic Violation 745 6690 0.03404 0.03931 0.86
Operating without Required
285 2648 0.01302 0.01556 0.83
Equipment
Failure to Keep in Proper Lane 5921 61914 0.27056 0.36379 0.74
Making Improper Turns 664 7085 0.03034 0.04163 0.72
Non-Traffic Violation Charged-
286 3540 0.01307 0.02080 0.62
Manslaughter or Homicide, etc.
Reckless Driving 1040 13141 0.04752 0.07721 0.61
Driving Over the Posted Speed Limit 4070 54837 0.18598 0.32221 0.57
Driver Inexperienced or Impaired
328 4683 0.01499 0.02752 0.54
Health or Physical Condition
Illegal Driving on Road Shoulder 54 912 0.00247 0.00536 0.46
Over Correcting 657 11656 0.03002 0.06849 0.43
Running Off the Road 587 11815 0.02682 0.06942 0.38
Other Drugs (Cocaine, etc.) 1520 33954 0.06946 0.19951 0.34
Hit-and-Run Vehicle Driver 264 6807 0.01206 0.04000 0.30

57
 Stopped or unattended vehicles, improper following, and starting and backing the vehicle

improperly are factors with the highest likelihood ratios of contributing to fatal truck crashes more

often than fatal non-truck crashes. Erratic lane change, cellular phone usage, and signal inattention

are also factors significantly contributing to fatal crashes. Truck drivers appear to be more fatigued,

drowsy, and inattentive when compared to other vehicle drivers, having a likelihood ratio of greater

than one.

4.5 Multinomial Logistic Regression Analysis for Truck Crashes

The multinomial logistic regression technique was used on a subset of the FARS data in

this study to elaborately analyze factors which have a higher rate of occurrence in fatal truck

crashes than in non-truck crashes. The subset data consists of only single-vehicle fatal crashes

that occurred in the United States from 2003-2007. The dependent variable for this model is

dichotomous, as it can either be a truck crash or a non-truck crash.

There were 35 independent variables which included several crash, driver, vehicle, and

environmental factors using statistical modeling software SAS version 9.1 (53). As the selection

criteria of variables to be included in the model, a 95% confidence level was used in which the

probability should be less than 0.05. Colinearity of individual variables was also checked before

considering variables in the model and if such relationship existed one of the two correlated

variables was discarded based on the lowest mean value criterion.

The independent variables considered in this model are shown in Table 4.4. Additionally,

the odds-ratio values are presented along with parameter estimates in Table 4.5. One can also

specify the change in the explanatory variables for which odds-ratio estimates are desired.

Confidence intervals for the regression parameters and odds ratios can be computed based either

on the profile likelihood function or on the asymptotic normality of the parameter estimators.

58
 Table 4.4. Description of the Variables Used in the Model

Variable Notation Value Description Frequency %

1 Winter 26,571 24.0
Month of the 2 Spring 24,549 22.2
month
Year 3 Summer 29,430 26.6
4 Fall 30,006 27.1
Day of a 1 <14 54,696 49.5
day
Month 2 >=14 55,860 50.5
1 <10 54,583 49.4
Hour in a Day hour
2 >10 55,973 50.6
Road Function 1 Rural 62,965 57.0
road_func
Class 2 Urban 47,591 43.1
Interstate/US and State Highway/County
1 81,311 73.6
Route route Road
2 Local Roads 29,245 26.5
Special 1 No Special Jurisdiction 109,212 98.8
sp_jur
Jurisdiction 2 Under Special Jurisdiction 1,344 1.2
1 Overturn/Rollover 19,783 17.9
2 Pedestrian 20,473 18.5
First Harmful Motor Vehicle in Transport on Same
harm_ev 3 105 1.1
Event Roadway
4 Tree (Standing Tree Only) 15,424 14.0
5 All Other Categories 54,771 49.5
1 Not a Collision with a Motor Vehicle 109,051 98.6
2 Rear End 744 0.7
Manner of
man_coll 3 Head On Collision 133 0.1
Collision
4 Angle 61 0.1
5 Other 628 0.6
Traffic 1 Not Physically Divided 75,299 68.1
traf_flo
Flowway 2 Divided Highway/One way/Ramp/Other 35,257 31.9
1 Two Lane Or Less 88,232 79.8
No. of Lanes no_lanes
2 More than Two Lanes 22,324 20.2
1 <40 27,536 24.9
2 40<=x<50 21,644 19.6
Posted Speed
sp_limit 3 50<=x<60 35,693 32.3
Limit
4 60<=x<70 13,928 12.6
5 >=70 11,755 10.6
Road 1 Straight 73,044 66.1
alignmnt
Alignment 2 Curved/Unknown 37,512 33.9
1 Level 76,923 69.6
Road Profile profile
2 Grade, Hillcrest, Sag, Unknown 33,633 30.4
Pavement 1 Blacktop 95,673 86.5
pave_typ
Type 2 Concrete and Other 14,883 13.5
Light 1 Day Light 44,192 40.0
lgt_cond
Condition 2 Poor Light Conditions/Other 66,364 60.0
Surface 1 Dry 92,610 83.8
sur_cond
Condition 2 Wet/Snow/Slush/Ice/Sand, Dirt 17,946 16.2

59
 Table 4.4. Description of the Variables Used in the Model (cont.)

Variable Notation Value Description Frequency %

Weather 1 No Adverse Condition 98,290 88.9
weather
Condition 2 Rain/ Sleet/ Snow/Fog/Rain/Sleet/Smog 12,266 11.1
Crash-Related 1 No Factor 106,415 96.3
Contributory cf1
Factor 2 Some Factor Present 4,141 3.8
1 One Fatality 104,411 94.4
No. of Fatalities fatals
2 More than One Fatality 6,145 5.6
Day of the 1 Fri,Sat,Sun 58,728 53.1
day_week
Week 2 Mon-Thur/Unknown 51,828 46.9
1 Young 47,267 42.8
Age of the
age 2 Middle 26,392 23.9
Driver
3 Older 36,897 33.4
Sex of the 1 Male 83,345 75.4
sex
Driver 2 Female 27,211 24.6
Ejection Type of 1 Not Ejected 82,586 74.7
ejection
the Driver 2 Totally or Partially Ejected 27,970 25.3
1 NO 38,905 35.2
Alcohol
drinking 2 YES 27,542 24.9
Involvement
3 Not Reported/Unknown 44,109 39.9
Alcohol 1 Test Conducted 31,212 28.2
alc_det
Detection 2 Not Reported 79,344 71.8
1 NO 26,259 23.8
Drugs
drugs 2 YES/Unknown 17,895 16.2
Involvement
3 Not Reported 66,402 60.1
1 No Injury 23,448 21.2
Injury Severity
inj_sev 2 Fatal Injury 68,300 61.8
of the Driver
3 Other Injury 18,808 17.0
1 No Rollover 68,659 62.1
Rollover rollover
2 Happened as a First/Subsequent Event 41,897 37.9
1 Not an Articulated Vehicle 107,554 97.3
Jacknife j_knife
2 No/Other 3,002 2.7
1 Between 0 and 45 mph 15,927 14.4
Travelling 2 Between 45 and 60 mph 22,224 20.1
trav_sp
Speed 3 Above 60 mph 6,161 5.6
4 Not Reported/Unknown 66,244 59.9
Initial Impact 1 Front Side/Other 95,931 86.8
impact1
Point 2 Rear Side 14,625 13.2
Extent of 1 Severe Disabling Deformation 80,688 73.0
deformed
Deformation 2 Functional and Other Deformation 29,868 27.0
Vehicle-Related 1 No Vehicle factor 98,818 89.4
Contributory veh_cf1
Factor 2 Some Vehicle Factor 11,738 10.6
1 None 16,091 14.6
Driver
2 Improper Physical/Mental Condition 29,333 26.5
Contributory dr_cf1
Factor 3 Improper Following of Traffic Regulations 53,919 48.8
4 Other Miscellaneous factors 11,213 10.1

60
 From the output parameters shown in Table 4.5, those response variables which are

significant in the model are identified by setting the alpha level at 0.05 value. For all variables

which have a p-value greater than 0.05, the model fails to reject the null hypothesis which says

that the coefficient of that variable is zero. Hence, all such variables become insignificant in the

model. Therefore, the variables of month, day, sp_jur, harm_ev, no_lanes, alignment, pave_typ,

and drugs become insignificant in the model as they have a p-value greater than the assumed

cutoff value. All 27 other response variables remain in the model as they have a p-value less than

0.05 and hence the null hypothesis is rejected and it is concluded that the regression coefficient

for all these variables has not been found to be statistically different from zero in estimating the

model.

From the sign of the significant variables in the model when the coefficient estimates are

observed the kind of proportionality the response variables have with the type of crash can be

discerned. While analyzing this aspect, it should be noted that the analysis is done with respect to

the occurrence of a fatal single-vehicle truck crash. All variables with regard to their estimate

value and point estimate (odds ratio) are explained in the following categories.

4.5.1 Roadway Characteristics

The negative coefficient for the response variable route would explain that there are a

larger proportion of fatal single-vehicle truck crashes on interstates than on local roads as

compared to non-truck crashes. Similarly, the coefficient of road_fnc explains that fatal truck

crashes are more frequent on rural roads than on urban roads.

The variable traf_flo was also found to be significant in the model. As the indicator that

the estimate value is positive, it shows that truck crashes have 1.98 times greater odds of

occurring on roadways which are not physically divided when compared to non-truck crashes.

61
Also, the profile variable has a positive estimate value. This shows that the type of crash has a

direct relationship with the roadway profile at the crash. Truck crashes tend to have 1.26 times

greater odds of occurring on level roadway profiles than when compared to non-truck crashes.

Similarly, when the surface condition at the crash site was analyzed it had a negative estimate

value in the model. This implies that non-truck crashes have 0.76 times lesser odds of occurring

on dry surfaces when compared to truck crashes.

4.5.2 Crash Characteristics

In the case of the manner of collision, the estimate suggests that fatal single-vehicle truck

crashes have 1.24 times higher odds of resulting in angle crashes than rear-end or head-on

collisions. Furthermore, they have 1.21 times greater odds of occurrence at speed limits greater

than 60 than on roadways with lower speed limits. Similarly, the variable hour shows that truck

crashes have 0.461 times lesser odds of occurring in morning and dawn hours of the day than in

non-truck crashes. The variable fatals in the model which shows the number of fatalities in the

crashes was found to have a negative estimate value in the model. This implies that fatal truck

crashes have 0.647 times lesser odds of resulting in more than one fatality in a crash when

compared to non-truck crashes.

Further, it was observed that truck crashes have 2.096 times higher odds of resulting in

rollover crashes and 1.035 times higher odds of having a travelling speed above 60 mph than

non-truck crashes. Also, it was seen that truck crashes have 1.45 times higher odds of having a

rear side initial impact point in a single vehicle crashes and 1.601 times higher odds of suffering

functional deformation of the vehicle than when compared with non-truck crashes.

62
4.5.3 Environmental Characteristics

The light condition variable explains that truck crashes have 0.44 times lower odds of

occurrence in dark conditions, and in the case of weather variables they have 1.22 times higher

odds of occurrence than in no adverse weather conditions.

4.5.4 Driver Characteristics

The age variable has a positive coefficient which shows that truck drivers are mostly in

the middle and older population, whereas non-truck drivers tend to be mostly in the younger

population. Truck drivers have 1.906 times higher odds of being middle or older aged than being

younger aged. From the alcohol involvement variable, it can also be derived that truck drivers

have 0.88 times lesser odds of involvement in fatal crashes when compared to non-truck drivers.

The variable representing the gender of the driver had a negative estimate value in the

model. This shows that truck drivers are 0.076 less likely to be female than in non-truck crashes.

Also, when the ejection variable was observed it showed that truck drivers had 0.596 times lesser

odds of ejecting out of the vehicle during the crash than when compared to non-truck drivers in

fatal crashes.

4.5.5 Other Contributory Factors

When the overall crash-related factor, cf1, is observed the positive coefficient shows that

truck crashes tend to have some significant factor which has been identified in the police report.

Moreover, the vehicle-related factor shows there are 1.42 times higher odds of a truck having a

significant vehicle contributory factor than a non-truck vehicle.

63
 Table 4.5. Parameter Estimates and Odds Ratio of Fatal Truck Crashes in the Model

Wald 95% Wald

Standard Pr > Odds
Parameter Estimate Chi- Confidence Limits
Error Chi-Sq Ratio
Square For Odds Ratio
intercept -1.5899 0.3494 20.7004 <.0001
month 0.0138 0.0154 0.7958 0.3723 1.014 0.984 , 1.045
day -0.0178 0.0341 0.2716 0.6023 0.982 0.919 , 1.050
hour* -0.7747 0.0357 470.137 <.0001* 0.461 0.430 , 0.494
road_func* -0.3144 0.0442 50.6108 <.0001* 0.73 0.670 , 0.796
route* -0.225 0.0485 21.5353 <.0001* 0.799 0.726 , 0.878
sp_jur -0.2132 0.1807 1.3916 0.2381 0.808 0.567 , 1.151
harm_ev 0.0141 0.011 1.6281 0.202 1.014 0.992 , 1.036
man_coll* 0.2173 0.0404 28.8839 <.0001* 1.243 1.148 , 1.345
traf_flo* 0.6831 0.0418 266.572 <.0001* 1.98 1.824 , 2.149
no_lanes -0.0395 0.0438 0.8136 0.3671 0.961 0.882 , 1.047
sp_limit* 0.1947 0.016 148.621 <.0001* 1.215 1.178 , 1.254
alignmnt 0.0234 0.0418 0.3136 0.5755 1.024 0.943 , 1.111
profile* 0.2384 0.0388 37.86 <.0001* 1.269 1.176 , 1.369
pave_typ -0.0011 0.0487 0.0005 0.9825 0.999 0.908 , 1.099
lgt_cond* -0.8113 0.0357 517.257 <.0001* 0.444 0.414 , 0.476
sur_cond* -0.2735 0.071 14.8527 0.0001* 0.761 0.662 , 0.874
weather* 0.1993 0.0788 6.4037 0.0114* 1.221 1.046 , 1.424
cf1* 0.2813 0.0805 12.2242 0.0005* 1.325 1.132 , 1.551
fatals* -0.436 0.1002 18.9312 <.0001* 0.647 0.531 , 0.787
day_week* 0.8383 0.0362 535.453 <.0001* 2.312 2.154 , 2.483
age* 0.645 0.0219 870.743 <.0001* 1.906 1.826 , 1.989
sex* -2.5808 0.0921 785.063 <.0001* 0.076 0.063 , 0.091
ejection* -0.517 0.0496 108.486 <.0001* 0.596 0.541 , 0.657
drinking* -0.1237 0.0218 32.0713 <.0001* 0.884 0.847 , 0.922
alc_det* 0.2171 0.0436 24.8166 <.0001* 1.242 1.141 , 1.353
drugs -0.0011 0.0218 0.0024 0.9608 0.999 0.957 , 1.042
inj_sev* -1.0785 0.0418 664.734 <.0001* 0.34 0.313 , 0.369
rollover* 0.7401 0.047 247.666 <.0001* 2.096 1.912 , 2.299
trav_sp* 0.0342 0.0149 5.2619 0.0218* 1.035 1.005 , 1.065
impact1* 0.373 0.0504 54.8538 <.0001* 1.452 1.316 , 1.603
deformed* 0.4706 0.0473 99.1199 <.0001* 1.601 1.459 , 1.756
veh_cf1* 0.3514 0.055 40.8397 <.0001* 1.421 1.276 , 1.583
dr_cf1* -0.097 0.0194 25.0645 <.0001* 0.908 0.874 , 0.943

64
* – Significant at 0.05 level

The ―Model Fit Statistics‖ in Table 4.6 contain the Akaike Information Criterion (AIC),

the Schwarz Criterion (SC), and the negative of twice the log likelihood (-2 Log L) for the

intercept-only model and the fitted model. AIC and SC can be used to compare different models

and those with smaller values are preferred.

The AIC value of 34,527 is the smallest value obtained in the repeated trials performed in

this dataset, which shows that this model is the optimum result. The SC and the -2 Log L values

were also observed to be the least therefore reinforcing the above statement.

Table 4.6. Model Fit Statistics of the Multinomial Logistic Regression Analysis

Criterion Intercept Only Intercept and Covariates

AIC 34527.4 27107.59

SC 34537.0 27434.44
-2 Log L 34525.4 27039.59

The three independence tests of likelihood ratio, overall score, and Wald‘s Chi-Square

have a p-value less than .0001 as shown in Table 4.7, therefore showing that results are very

significant.

65
 Table 4.7. Tests of Independence for the Multinomial Logistic Regression Analysis

Test Chi-Square DF Pr > ChiSq

Likelihood Ratio 7485.8133 33 <.0001

Score 6963.1703 33 <.0001
Wald 5397.167 33 <.0001

Table 4.8 shows other goodness-of-fit parameters values obtained from the LOGISTIC

procedure performed on the dataset. Descriptions of those parameters are as follows:

 Percent Concordant: This has a value of 85.1% which shows a high rate of concordance

between the pairs of observations with differences observed, and the observation with the

lower ordered response value has a lower predicted mean score than the observation with

the higher ordered response value.

 Percent Discordant: If the observation with the lower ordered response value has a higher

predicted mean score than the observation with the higher ordered response value, then

the pair is discordant.

 Percent Tied: If a pair of observations with different responses is neither concordant nor

discordant, it is a tie.

 Somer's D: This is used to determine the strength and direction of relation between pairs

of variables. Its values range from -1.0 (all pairs disagree) to 1.0 (all pairs agree). The

value of 0.71 is closer to 1 which therefore shows that all pairs of variables agree to a

large extent.

66
  The Goodman-Kruskal Gamma: This has a value of 0.717 which also signifies the perfect

association of the variables in the model. This method does not penalize for ties on either

variable. Its values range from -1.0 (no association) to 1.0 (perfect association).

 Tau-a: This has the value of 0.05, which is a modification of Somer's D to account for the

difference between the number of possible paired observations and the number of paired

observations with different responses.

 c: Another measure of rank correlation of ordinal variables, c has a value of 0.855 which

reinforces the perfect association between the data variables and the observed variables.

This value usually ranges from 0 (no association) to 1 (perfect association).

Table 4.8. Associations of Predicted Probabilities and Observed Responses

Percent Concordant 85.1 Somers' D 0.71

Percent Discordant 14 Gamma 0.717
Percent Tied 0.9 Tau-a 0.05
Pairs 428,069,684 c 0.855

Hence, multinomial logistic regression provides useful goodness of fit measures which

help analyze the significance of various parameters with truck crashes in comparison with non-

truck crashes.

67
 Chapter 5 Conclusions and Summary

5.1 Conclusions

This study explored the characteristics of trucks involved in fatal crashes and evaluated the

fatality risk posed for them in relation to select driver, vehicle, environmental, and roadway-related

variables. Fatal crash data obtained from NHTSA was used for this analysis.

Several significant characteristics of fatal truck crashes have been observed from this

analysis. Fatal crash frequency was observed to be greater with the initial impact point for the

vehicle in the front side of the truck than anywhere else. All fatal truck crash cases which had

alcohol involvement indicated that in 87% of cases, non-truck drivers were the ones under this

influence. Trucks seemed to have a majority of fatal crashes at higher posted speed levels, which

might also be due to a larger presence of trucks at higher speed ranges. Fatigue, drowsiness, and

inattention were observed to be more predominant in truck drivers than in other motor vehicle

drivers. The majority of fatal truck crashes occurred on two-way two-lane traffic flowways with

no physical divisions. Such roadways could be altered by providing necessary changes in the

roadway design. Improper driving and non-compliance to traffic regulations were observed to be

the main driver-related contributory factors in cases of fatal truck crashes. In comparing the

overlapping effect of two fatal crash characteristics, truck striking and truck being struck, there

seemed to be similar proportions on all roadway types. This proportion remained consistent even

under different light conditions.

From the likelihood ratios, stopped or unattended vehicles or improper following had

greater probabilities of occurrence in fatal truck crashes than in non-truck crashes. Recent or

previous crash nearby and/or vehicle set in motion by a non-driver, work area conditions, poor

shoulder conditions, and inadequate warning signs are the topmost factors which have more

likelihood in fatal truck crashes than non-truck crashes. Other factors like cellular usage, failure

68
to yield right of way, inattentiveness, and failure to obey traffic rules are more likely to

contribute to fatal truck crashes. Additionally, truck drivers appear to be more fatigued, drowsy,

and inattentive when compared to other vehicle drivers with a likelihood ratio of greater than

one.

From the Multinomial Logistic Regression Modeling performed on the single-vehicle

fatal crashes, several conclusions were ascertained. For example, it was concluded that single-

vehicle fatal truck crashes are more frequent on rural roads than on urban roads. The manner of

collision coefficient estimate suggests that fatal single-vehicle truck crashes have 1.24 times

higher odds of resulting in angle crashes than rear-end or head-on collisions. Also, they have

1.21 times greater odds of occurrence at speed limits greater than 60 than on roadways with

lower speed limits. The light condition variable explains that truck crashes have 0.44 times

lower odds of occurrence in dark lighting conditions and in the case of the weather variable they

have 1.22 times higher odds of occurrence in conditions with no adverse weather. Finally, the

overall 85.1% concordance value of the model has shown the level to which it fits the given data,

hence proving to be a decent model fit.

The results provide a deep understanding of the various factors which have greater

association with truck crashes when compared to non-truck crashes. By addressing these issues

the overall truck crash rate can be reduced, which can help in improving overall safety of the

transportation system.

69
 References

1. ―Motor Carrier Safety: A Statistical Approach Will Better Identify Commercial Carriers

that Pose High Crash Risks than Does the Current Federal Approach.‖ Publication

GAO-07-585, United States Government Accountability Office, June 2007.

2. Blower, D. ―Comparison of Certain Variables in the Large Truck Crash Causation Study

with Variables Coded in a Review of the Trucks Involved in Fatal Accidents Case

Materials.‖ Publication UMTRI-2007-28. The University of Michigan

Transportation Research Institute, 2007.

3. Blower, D. ―Truck Mirrors, Field of View, and Serious Truck Crashes.‖ Publication

UMTRI-2007-25. University of Michigan Transportation Research Institute, U.S.

Department of Transportation, June 2007.

4. Krishnaswami, V., D. Blower, L. Schneider., and D. Putcha, ―Heavy Truck

Aggressiveness Reduction: Statistics, Analysis, and Countermeasures.‖ Publication

UMTRI-2005-38. University of Michigan Transportation Research Institute, 2005.

5. Blower, D., and L. P. Kostyniuk. ―Strategies to Reduce Cmv-Involved Crashes,

Fatalities, and Injuries in Michigan.‖ Publication UMTRI-2005-38. University of

Michigan Transportation Research Institute, August 2007.

6. Daniel, J., C., Tsai, and S. Chien. Factors Influencing Truck Crashes on Roadways with

Intersections. CD-ROM. Transportation Research Board of the National

Academics, 2002.

70
7. Vap, D., and S. Carlos. ―Investigating Large Truck-Passenger Vehicle Interaction.‖

Publication RI 07-006. Missouri Department of Transportation (MoDOT), U.S.

Department of Transportation, September 2007.

8. Massie, D. L., E. G. Paul., and Kenneth L. Campbell. ―Crash Involvement Rates by

Driver Gender and the Role of Average Annual Mileage.‖ Accident Analysis and

Prevention 29.5 (1997): 675-685.

9. Patterson, T. M., and L. W. Munnich. ―Five-County Minnesota Case Study: Rural

Roadway Fatal Crash Characteristics and Select Safety Improvement Programs.‖

CTS 08-03. Center of Excellence in Rural Safety, December 2008.

10. Mehta, Y., and C. Gabler. ―Analysis of Fatal Accidents in New Jersey.‖ Publication

FHWA-NJ-2008-05. Federal Highway Administration (FHWA), U.S. Department

of Transportation, June 2008.

11. Alam, B. M., and L. K. Spainhour. Behavioral Aspects of Younger At-Fault Drivers in

Fatal Traffic Crashes in Florida. CD-ROM. Transportation Research Board, 2009.

12. Haworth, N., K. Grieg., and A. Neilson. A Comparison of Risk Taking in Moped and

Motorcycle Crashes. CD-ROM. Transportation Research Board, 2009.

13. Williams, A. F., and V. I. Shabanova. ―Responsibility of Drivers, by Age and Gender, for

Motor Vehicle Crashes.‖ Journal of Safety Research 34 (2003): 527-531.

14. Williams, A. F., and V. I. Shabanova. ―Analysis of the Fatal Crash Involvements of

Fifteen-Year Old Drivers.‖ Journal of Safety Research 28 (2003): 49-54.

71
15. Pickrell, T. M., and M. Starnes. ―An Analysis of Motorcycle Helmet Use in Fatal

Crashes.‖ NHTSA Technical Report. DOT HS 811 011. National Highway Traffic

Safety Administration, Washington DC, August 2008.

16. McCartt, A. T., E. R. Teoh., M. Fields., K. A. Braitman., and L. A. Hellinga. ―Graduating

Licencing Laws and Fatal Crashes of Teenage Drivers: A National Study.‖

Insurance Institute for Highway Safety, May 2009.

17. ―Geo-Demographic Analysis of Fatal Motorcycle Crashes.‖ Report No. HS 809 197.

National Highway Traffic Safety Administration, U.S Department of

Transportation, January 2001.

18. Lerner, N., M. Freedman., P. Zador., M. Cluad., B. Simons., and D. Duncan. ―Passenger

Age and Gender Effects on Adult Driver Fatal Crash Rate.‖ Proceedings of the

Fourth International Driving Symposium on Human Factors in Driver Assessment,

Training and Vehicle Design, Stevenson, Washington, 2007.

19. ―Report to Congress on the Large Truck Crash Causation Study.‘ Publication MC-R/MC-

RRA. FMCSA, U.S. Department of Transportation, March 2006.

20. Blower, D., and A. Matteson. Large Trucks in FARS and in TIFA. Center for National

Truck Statistics, University of Michigan Transportation Research Institute, July

2002.

21. Rau, P. S. ―Drowsy Driver Detection and Warning System for Commercial Vehicle

Drivers: Field Operational Test Design, Data Analysis, and Progress.‖ National

72
 Highway Traffic Safety Administration. U.S. Department of Transportation, May

2005. Paper number 05-0192

22. Garber, N. J., J. S. Miller., X. Sun., and B. Yuan. ―Safety Impacts of Differential Speed

Limits for Trucks and Passenger Cars on Rural Interstate Highways: A Modified

Bayes Approach.‖ Journal of Transportation Engineering 132.1 (January 2006):

19-29.

23. Daniel, J., and S. I. Chien. ―Truck Safety Factors on Urban Arterials.‖ Journal of

Transportation Engineering 130.6 (November-December 2004): 742-752.

24. Dick, V., J. Hendrix., and R. R. Knipling. ―New Hours-of-Service Rules: Trucking

Industry Reactions and Safety Outcomes.‖ Transportation Research Record:

Journal of the Transportation Research Board (January 2006): 103-109.

25. Kostyniuk, L. P., F. M. Streff., and J. Zakrajsek. Identifying Unsafe Driver Actions that

Lead to Fatal Car-Truck Crashes. AAA Foundation for Traffic Safety, April 2002.

26. Khattak, A. J., and F. Targa. Injury Severity and Total Harm in Truck-Involved Work

Zone Crashes. CD-ROM. Transportation Research Board of the National

Academics, 2004. Paper No.04-2458.

27. Daniel, J., R. Rajbhandari., and S. Chien. In Evaluation of Crashes on the National

Network. CD-ROM. Transportation Research Board of the National Academics,

2002.

73
28. Dabbour, E., S. M. Easa., and A. O. Abd El Halim. ―Radius Requirements for Reverse

Horizontal Curves on Three-Dimensional Alignments.‖ Journal of Transportation

Engineering 130.5 (September-October 2004): 610-620.

29. Miaou, S., and H. Lum. ―Statistical Evaluation of the Effects of Highway Geometric

Design on Truck Accident Involvements.‖ Transportation Research Record:

Journal of the Transportation Research Board 1407 (1993): 11-33.

30. Abdel-Aty, M., and H. Abdelwahab. ―Modeling Rear-End Collisions Including the Role

of Driver‘s Visibility and Light Truck Vehicles Using a Nested Logit Structure.‖

Transportation Research Record: Journal of the Transportation Research Board

1840 (2003): 140-147.

31. Diener, J., and L. Richardson. ―Seat Belt Use among Rural and Urban Pickup Truck

Drivers.‖ Report 4-2007. Institute Public Policy, July 2007.

32. Burgess, M. ―Contrasting Rural and Urban Fatal Crashes 1994–2003.‖ Report No. DOT-

HS-809-896. National Highway Traffic Safety Administration U.S. Department of

Transportation, 2005.

33. Majid, S. Measuring and Modelling the Dynamics of Heavy Commercial Vehicles and

Their Interactions with Other Vehicles. CD-ROM. Transportation Research Board

of the National Academics, 2008.

34. Duncan, C. S., A. J. Khattak., and F. M. Council. ―Applying the Ordered Probit Model to

Injury Severity in Truck/Passenger-Car Rear-End Collisions.‖ Transportation

74
 Research Record: Journal of the Transportation Research Board 1635 (2004):

Paper No. 98-1237.

35. Pickrell, T. M. ―Driver Alcohol Involvement in Fatal Crashes by Age Group and Vehicle

Type.‖ Report No. DOT-HS-810-598. National Highway Traffic Safety

Administration, U.S. Department of Transportation, 2006.

36. Yan, X., E. Radwan., and K. Mannila. ―Analysis of Truck-Involved Rear-End Crashes

Using Multinomial Logistic Regression.‖ Advances in Transportation Studies 17

(2009): 39-52.

37. Yan, X., E. Radwan., and K. Mannila. ―Investigation of Truck Involved Angle Crashes

Based on GES and FARS Crash Databases.‖ Open Transportation Journal 3

(2009): 49-59.

38. Venkataraman, S., and F. Mannering. “An Exploratory Multinomial Logit Analysis of

Single-Vehicle Motorcycle Accident Severity.‖ Journal of Safety Research 27

(1994): 183-194.

39. Allison, D.P. Logistic Regression Using SAS System. North Carolina: John Wiley &

Sons, Inc., 1999.

40. Samuel, P., R. W. Poole., and J. H. Veras. ―Toll Truckways: A New Path towards Safer

and More Efficient Freight Transportation.‖ Policy Study No. 294, June 2002.

www.rppi.org/ps294.html.

75
41. Cate, M. A., and T. Urbanik II. ―Another View of Truck Lane Restrictions.‖

Transportation Research Record: Journal of the Transportation Research Board

(2004): 19-24.

42. Reich, S. L., J. L. Davis., A. J. Ferraro., and Martin Catala. ―Exclusive Facilities for

Trucks in Florida: An Investigation of the Potential for Reserved Truck Lanes and

Truckways on the State Highway System.‖ Mid-American Transportation Research

Symposium, Iowa, August 2003.

43. Reiskin, J. S. ―NHTSA Seeks Ways to Use Technology to Reduce Truck-Involved

Fatalities.‖ Transport Topics (February 2008): 5-24.

44. Murray, D., S. Turrentine., B. Lantz., and S. Keppler. ―Predicting Truck Crash

Involvement: Developing a Commercial Driver Behavior Model And Requite

Enforcement Countermeasures.‖ Transportation Research Record: Journal of the

Transportation Research Board 1840 (2006): 40-47.

45. Crum, M. R., and P. C. Morrow. ―The Influence of Carrier Scheduling Practices on Truck

Fatigue.‖ Transportation Journal (2005): 20-41.

46. Williams, A. F., and V. I. Shabanova. ―Responsibility of Drivers, by Age and Gender, for

Motor Vehicle Crash Deaths.‖ Journal of Safety Research 34 (2003): 527-531.

47. Golob, T. F., and A. C. Regan. Truck-Involved Crashes and Traffic Levels on Urban

Freeways. CD.ROM. Transportation Research Board, 2004. Paper No. 04–1503.

76
48. Gabler, H. C. ―The Risk of Fatality in Motorcycle Crashes with Roadside Barriers.‖

Proceedings of the Twentieth International Conference on Enhanced Safety of

Vehicles, Lyons, France June 2007. Paper No. 07-0474.

49. Moonesinghe, R., A. Longthorne., U. Shankar., S. Singh., R. Subramanian., and J.

Tessmer. ―An Analysis of Fatal Large Truck Crashes.‖ Report No. DOT HS 809.

Mathematical Analysis Division, National Center for Statistics and Analysis,

National Highway Traffic Safety Administration U.S. Department of

Transportation, Washington, D.C. March 2003.

50. Zador, P. L., S. A. Krawchuk., and R. B. Voas. ―Relative Risk of Fatal and Crash

Involvement by BAC, Age, and Gender.‖ Report No. DTNH22-97-P-05174.

National Highway Traffic Safety Administration Office of Research and Traffic

Records, Office of Research and Traffic Records, Washington, DC, April 2000.

51. Hedeker, D. ―Mixed-Effects Multinomial Logistic Regression Model.‖ Division of

Epidemiology & Biostatistics, School of Public Health, University of Illinois at

Chicago, 2006.

52. National Highway Traffic Safety Administration. Fatality Analysis Reporting System.

http://www.nhtsa.dot.gov/people/ncsa/fars.html. Accessed January 17, 2010

53. SAS Institute Inc., SAS Online Doc 9.1.3., SAS Institute Inc., Cary, NC, 2004.

77