Microsoft Word - Statistical Analysis of the Electronic Stability Control _ESC_ Systems

U.S. Department

of Transportation

National Highway

Traffic Safety

Administration

____________________________________________________________________________

DOT HS 810 794 July 2007

NHTSA Technical Report

Statistical Analysis of the Effectiveness of

Electronic Stability Control (ESC) Systems –

Final Report

_____________________________________________________________________

This document is available to the public at the Docket Management System of the U.S. Department of Transportation.

The United States Government does not endorse products or manufacturers. Trade or

manufacturers’ names appear only because they are considered essential to the object of

this report.

Technical Report Documentation Page

1. Report No.

DOT HS 810 794

2. Government Accession No. 3. Recipient’s Catalog No.

5. Report Date

July 2007

4. Title and Subtitle

STATISTICAL ANALYSIS OF THE EFFECTIVENESS OF

ELECTRONIC STABILITY CONTROL (ESC) SYSTEMS –

FINAL REPORT

6. Performing Organization Code

7. Author(s)

Jennifer N. Dang

8. Performing Organization Report No.

10. Work Unit No. (TRAIS)

9. Performing Organization Name and Address

Evaluation Division; National Center for Statistics and Analysis

National Highway Traffic Safety Administration

Washington, DC 20590

11. Contract or Grant No.

13. Type of Report and Period Covered

NHTSA Technical Report

12. Sponsoring Agency Name and Address

Department of Transportation

National Highway Traffic Safety Administration

Washington, DC 20590

14. Sponsoring Agency Code

15. Supplementary Notes

16. Abstract

Electronic Stability Control (ESC) is a safety technology designed to enhance a vehicle’s stability and control in all driving situations. ESC first became available in the United

States in 1997. Statistical analyses of 1997-2004 crash data from the Fatality Analysis Reporting System (FARS) and 1997-2003 crash data from the State data files estimate

reductions with ESC for various types of crash involvements.

• ESC reduced fatal run-off-road crashes by 36 percent for passenger cars and 70 percent for light trucks and vans (LTVs). The reductions are statistically significant.

• Police-reported run-off-road involvements were decreased by 45 percent in passenger cars and 72 percent in LTVs. The decreases are statistically significant.

• Fatal single-vehicle crashes that did not involve pedestrians, bicycles, and animals decreased (due to ESC) by 36 percent in passenger cars and 63 percent in LTVs. The

decreases are statistically significant.

• ESC reduced police-reported single-vehicle crashes (excluding pedestrian, bicycle, animal crashes) by 26 percent for passenger cars and 48 percent for LTVs. The

reductions are statistically significant.

• Rollover involvements in fatal crashes were decreased by 70 percent in passenger cars and 88 percent in LTVs. The decreases are statistically significant.

• Police-reported crashes involving rollovers were reduced by 64 percent in passenger cars and 85 percent in LTVs. The reductions are statistically significant.

• ESC reduced culpable fatal multi-vehicle crashes by 19 percent for passenger cars and 34 percent for LTVs. Only the reduction involving LTVs is statistically

significant.

• Culpable involvements in police-reported multi-vehicle crashes were decreased by 13 percent in passenger cars and 16 percent in LTVs. The decreases are

statistically significant.

• Overall, ESC reduced all fatal crashes by 14 percent for passenger cars and 28 percent for LTVs. Only the reduction in LTVs is statistically significant.

• Overall, police-reported crash involvements decreased by 8 percent in passenger cars and 10 percent in LTVs. The decreases are statistically significant.

17. Key Words

NHTSA; FARS; ESC; statistical analysis;

evaluation; benefits; effectiveness; fatality

reduction; electronic stability control systems;

18. Distribution Statement

Document is available to the public at the Docket

Management System of the U.S. Department of

Transportation, http://dms.dot.gov

, Docket Number

28629.

19. Security Classif. (Of this report)

Unclassified

20. Security Classif. (Of this page)

Unclassified

21. No. of Pages

22. Price

Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

iii

TABLE OF CONTENTS

LIST OF ABBREVIATIONS ……………………………………………………….iv

ACKNOWLEDGEMENTS ………………………………………………………….v

EXECUTIVE SUMMARY ……………………………………………………..….. vi

BACKGROUND …………………………………………………………………......1

ESC INFORMATION …………………………………………………………..……5

CRASH DATA (STATE AND FARS) …………………………………………….. 11

ANALYSIS DATABASES ………………………………………………………… 12

VIN decode …………………………………………………………………………..12

Crash involvements ……………………………………………………..................... 13

2X2 CONTINGENCY TABLE ANALYSIS ………………………………………...18

Analysis of fatal crashes (FARS) ………………………………………..................... 20

Analysis of crash involvements (State Data) …………………………………………25

LOGISTIC REGRESSION ANALYSIS ………………………………………….…. 46

COMPARISON OF 2-CHANNEL AND 4-CHANNEL ESC SYSTEMS……………48

CONCLUSIONS………………………………………………………………….….. 52

LIST OF ABBREVIATIONS

ABS Antilock brake systems

ASE Asymptotic standard error

CATMOD Categorical models procedure in SAS

ESC Electronic Stability Control

FARS Fatality Analysis Reporting System

FMVSS Federal Motor Vehicle Safety Standard

GENMOD Generalized estimating models procedure in SAS

GM General Motors

IIHS Insurance Institute for Highway Safety

LR Likelihood-ratio

LTVs Light trucks and vans

MEANS MEANS (simple statistics) procedure in SAS

NCSA National Center for Statistics and Analysis

NHTSA National Highway Traffic Safety Administration

OMB Office of Management and Budget

SAS Statistical analysis software produced by SAS Institute, Inc.

SE Standard error

SUV Sport utility vehicle

TCS Traction control systems

VIN Vehicle Identification Number

ACKNOWLEDGEMENTS

I owe special thanks to the two researchers who peer-reviewed a draft of the report:

1) Dr. Charles M. Farmer, Director of Statistical Services, Insurance Institute for

Highway Safety, Arlington, Virginia

2) Professor Claes Tingvall, Dr Med Sc, Director of Traffic Safety at Swedish Road

Administration

This study estimates the effectiveness of Electronic Stability Control (ESC) in reducing

crashes, specifically crashes where ESC is likely to have made a difference in the

vehicle’s involvement, based on statistical analyses of crash data. The National Highway

Traffic Safety Administration (NHTSA) published a draft of this report in support of a

proposed rulemaking to establish a new Federal Motor Vehicle Safety Standard, FMVSS

No. 126, which requires ESC systems on passenger cars, multipurpose vehicles, trucks,

and buses with a gross vehicle weight rating of 10,000 pounds or less (See NHTSA

Docket Number 25801-02 at http://dms.dot.gov). Because of the potential impacts of the

proposed regulation, the report contains “highly influential scientific information” as

defined by the Office of Management and Budget’s (OMB) “Final Information Quality

Bulletin for Peer Review” (available at

www.whitehouse.gov/omb/inforeg/peer2004/peerbulletin.pdf). Therefore, the report had

to be peer-reviewed in accordance with the requirements of both Sections II and III of

OMB’s Bulletin.

The peer-review process differed from the type used by journals. The effort by Dr.

Farmer and Professor Tingvall was essentially assessment of the scientific adequacy of

the draft to identify weaknesses and help NHTSA strengthen the report. These two

reviewers were selected by NHTSA staff. The publication of this report does not

necessarily imply that the reviewers supported it or concurred with the findings. You may

access their comments on the draft along with the revised report and the entire review

process in the NHTSA docket (Number 26415) at http://dms.dot.gov

. We have tried to

address all of the comments in our revised report (but we did not send it back to the

reviewers for a second round of review). The text and footnotes of the report single out

some of the reviewers’ comments that instigated additions or revisions to the analyses.

EXECUTIVE SUMMARY

Electronic Stability Control (ESC) is a safety system designed to recognize adverse

driving conditions by 1) continuously measuring and evaluating the speed, the steering

wheel angle, the yaw rate, and the lateral acceleration of a vehicle from various sensors

and 2) using those measured data to compare a driver’s steering input with the vehicle’s

actual motion. If an unstable situation is detected, then ESC automatically intervenes to

assist the driver and stabilize the vehicle by applying the brakes to individual wheels as

needed and possibly reducing engine torque. This technology is expected to reduce the

number of crashes due to driver error and loss of control, because it has the potential to

anticipate situations leading up to some crashes before they occur and the capability to

automatically intervene to prevent them. A major benefit should be the reduction of

single-vehicle crashes that involve losing control and running off the road.

In September, 2004, the National Highway Traffic Safety Administration (NHTSA)

issued an evaluation note on the Preliminary Results Analyzing the Effectiveness of

Electronic Stability Control (ESC) Systems. The data suggested that ESC was highly

effective in reducing single-vehicle run-off-road crashes. The study was based on

Fatality Analysis Reporting System (FARS) data from calendar years 1997-2003 and

crash data from five States from calendar years 1997-2002. The data were limited to

mostly luxury vehicles because ESC first became available in 1997 in luxury vehicles

such as Mercedes-Benz and BMW.

NHTSA has now updated and modified its 2004 report, extending it to model year 1997-

2004 vehicles – and to calendar year 2004 for the FARS analysis and calendar year 2003

for the State data analysis. Nevertheless, even as of 2004, a large proportion of the

vehicles equipped with ESC were still luxury vehicles. Moreover, only passenger cars

and SUVs had been equipped with ESC – no pickup trucks or minivans.

The FARS database included fatal crash involvements from calendar years 1997 to 2004.

The State databases included crash cases from California (2001-2003), Florida (1997-

2003), Illinois (1997-2002), Kentucky (1997-2002), Missouri (1997-2003), Pennsylvania

(1997-2001, 2003), and Wisconsin (1997-2003).

The basic analytical approach was to estimate the reduction of crash involvements of the

types that are most likely to have benefited from ESC – relative to a control group of

other types of crashes where ESC is unlikely to have made a difference in the vehicle’s

involvement. Crash involvements in which a vehicle 1) was stopped, parked, backing up,

or entering/leaving a parking space prior to the crash, 2) traveled at a speed less than 10

mph, 3) was struck in the rear by another vehicle, or 4) was a non-culpable party in a

multi-vehicle crash on a dry road, were considered the control group (non-relevant

involvements) – because ESC would in almost all cases not have prevented the crash.

The types of crash involvements where ESC would likely or at least possibly have an

effect are:

vii

• Single-vehicle crashes in which a vehicle ran off the road and then hit a fixed

object and/or rolled over.

• Involvements as a culpable party in a multi-vehicle crash on a dry or wet road.

• Collisions with pedestrians, bicycles, or animals.

The principal findings and conclusions of the statistical analyses are the following:

RUN-OFF-ROAD CRASHES

• ESC reduced involvements in all types of single-vehicle run-off-road crashes the

following percentages:

Crash reduction by ESC (%)

Cars

LTVs

Fatal crash involvements 36 70

Police-reported crash involvements 45 72

• All four of these reductions are statistically significant.

SINGLE-VEHICLE CRASH REDUCTION

• ESC reduced all single-vehicle involvements (excluding pedestrian, bicycle,

animal crashes) by the following percentages:

Crash reduction by ESC (%)

Cars

LTVs

Fatal single-vehicle crash involvements 36 63

Police-reported single-vehicle crash involvements 26 48

• All four reductions are statistically significant.

ROLLOVER CRASHES

• ESC was especially effective in preventing single-vehicle first-event rollovers.

Crash reduction by ESC (%)

Cars LTVs

Fatal rollovers 70 88

Police-reported rollovers 64 85

• All four of these reductions are statistically significant.

CULPABLE INVOLVEMENTS IN MULTI-VEHICLE CRASHES

• ESC likely reduces involvements as a culpable party in multi-vehicle crashes.

Crash reduction by ESC (%)

Cars LTVs

Fatal culpable multi-vehicle crash involvements 19 34

Police-reported culpable multi-vehicle crash involvements 13 16

viii

• Only the reduction in fatal crash involvements in passenger cars is not statistically

significant.

COLLISIONS WITH PEDESTRIANS – BICYCLES – ANIMALS

• There are no consistently significant results in either direction for crashes that

involve pedestrians, bicycles, or animals

Crash reduction by ESC (%)

Cars LTVs

Fatal pedestrian, bicycle, animal crashes -36 -6

Police-reported pedestrian, bicycle, animal crashes 26 -11

• Only the reduction in police-reported crash involvements in passenger cars is

statistically significant. We will continue to monitor the effect of ESC on this

particular type of crash involvements in the future – because we do not have

enough data (at the moment) for any conclusions.

OVERALL CRASH REDUCTION

• ESC reduced all crash involvements by the following percentages:

Crash reduction by ESC (%)

Cars LTVs

All fatal crash involvements 14 28

All police-reported crash involvements 8 10

• Only the reduction in fatal crash involvements in passenger cars is not statistically

significant.

4-CHANNEL VERSUS 2-CHANNEL ESC SYSTEMS IN PASSENGER CARS

The passenger car sample includes certain make-models that had 2-channel ESC systems

and others that had 4-channel systems. Separate analyses were performed to analyze the

difference in effectiveness (if any) between 2-channel and 4-channel systems. We found:

Greater

fatal run-off-road reduction with 2-channel systems – but the larger observed

fatality reduction with 2-channel systems is not statistically significantly different from

the observed reduction with 4-channel systems. The reductions were most certainly

influenced by the small samples.

Larger

reduction with 4-channel systems when police-reported crash involvements were

included (as expected with larger samples). The larger reduction with 4-channel systems

in all run-off-road involvements (mostly non-fatal crashes) is statistically significant.

BACKGROUND

Automotive braking technologies have evolved from very simple systems (i.e., block

brakes) to more sophisticated systems (i.e., cable-operated four-wheel brakes, hydraulic

four-wheel brakes, drum brakes, disc brakes with front-rear split, etc.). Today, drivers

rely on much more technologically-advanced systems to help them not only to decelerate

and accelerate but also to stabilize their vehicles while in motion, such as:

Antilock Brake Systems (ABS) are the first of a series of three braking technology

developments. They are four-wheel systems that prevent wheel lock-up by automatically

modulating the brake pressure when the driver makes an emergency stop.

Traction Control Systems (TCS) are the second technology. They deal specifically with

front-to-back loss of friction between the vehicle’s tires and the road surface during

acceleration.

Electronic Stability Control (ESC) systems are another important technology evolving

from and incorporating the first two technologies – ABS and TCS – with additional

capabilities. They are stability enhancement systems designed to improve vehicles’

lateral stability by electronically detecting and automatically assisting drivers in

dangerous situations (e.g., understeer and oversteer) and under unfavorable conditions

(e.g., rain, snow, sleet, ice). ESC systems have sensors that monitor the speed, the

steering wheel angle, the yaw rate, and the lateral acceleration of the vehicle. Data from

the sensors are used to compare a driver’s intended course with the vehicle’s actual

movement to detect when a driver is about to lose control of a vehicle and automatically

intervene in split seconds by applying the brakes to individual wheels and possibly

reducing engine torque to provide stability and help the driver stay on course. For

example, if an ESC system detects that the rear wheels have begun to slide to the right

and the vehicle is yawing counter-clockwise, it may momentarily brake the right front

wheel, imparting a clockwise torque to counteract the excessive counterclockwise yaw

and stabilize the vehicle. Depending on the driving situation, these brake interventions

may also be used to slow down the vehicle to a speed more appropriate for the operating

conditions.

The reasons ESC appears to provide safety benefits are twofold: (1) it can anticipate

situations leading up to some loss-of-control crashes before they occur, and (2) it has the

capability to mitigate these crashes via automatic intervention. Hence, the potential

benefit should be primarily a reduction of single-vehicle crashes that involve losing

control and running off the road. These crashes include rollovers and collisions with

fixed objects.

In a first published study

analyzing the effectiveness of ESC on three Toyota passenger

car make-models, Aga and Okada reported in 2003, a 36 percent reduction in single-

vehicle crash rates (single-vehicle crashes per 10,000 vehicles per year) when they

compared make-model vehicles with ESC and those without ESC. The study also

Aga, M. and Okada, A. (2003) Analysis of Vehicle Stability Control (VSC)’s Effectiveness from

Accident Data, Paper Number 541, Proceedings of the 18

International Technical Conference on the

Enhanced Safety of Vehicles.

showed a reduction (28 percent) in the rates of head-on collisions in studied vehicles that

are equipped with ESC relative to those that are not equipped with ESC.

In 2004, the National Highway Traffic Safety Administration (NHTSA) initiated an

evaluation to assess the effectiveness of ESC in reducing single-vehicle crashes in

various domestic and imported passenger cars and Sport Utility Vehicles (SUVs). The

preliminary results from that study

showed that ESC is highly effective in reducing

single-vehicle crashes. In fact, single-vehicle crashes were reduced by 35 percent in

passenger cars, and 67 percent in SUVs. Similarly, fatal single-vehicle crashes were

reduced by 30 percent in passenger cars, and 63 percent in SUVs.

In a study published in 2004 by the Insurance Institute for Highway Safety (IIHS)

Farmer compared per vehicle crash involvement rates of vehicles with model years that

had ESC as standard equipment with identical make-model year vehicles that did not.

Farmer found that ESC reduced single-vehicle crash involvement risk by approximately

41 percent and reduced single-vehicle injury risk by the same amount. Overall, crash

involvement risk was reduced by 7 percent for all crashes and 9 percent for injury

crashes. For fatal crash involvements, it was estimated that ESC reduced single-vehicle

crash involvement risk by 56 percent and overall fatal crash involvement risk by 34

percent. The effect of ESC on multi-vehicle crashes was found to be minimal and not

statistically significant. Unlike the NHTSA evaluation, this study combined the results

for passenger cars and SUVs.

In another study published in 2004

, researchers in Sweden (Tingvall et al.) reported a 32

percent crash reduction with ESC on wet roads and a 38 percent reduction on roads

covered with ice and snow. For all crashes except rear-end impacts, the study found an

overall 22 percent reduction in crashes involving vehicles with ESC versus those without

ESC. In this study, rear-end crashes on dry surfaces, assumed to be unaffected by the

presence of ESC, were used as the control group. The data set used in the analysis

consists of Swedish police reported crashes where at least one occupant was injured. In

2006, Tingvall et al. conducted a follow-up study analyzing the effectiveness of ESC

specifically by injury severity.

For serious and fatal loss-of-control type crashes, it was

estimated that ESC reduced those types of crashes by 56 percent on wet roads and 49

percent on roads that were covered with ice and snow.

Dang, J. (2004) Preliminary Results Analyzing the Effectiveness of Electronic Stability Control (ESC)

Systems, NHTSA Evaluation Note No. DOT HS 809 790, Washington, D.C.

Farmer, C. (2004) Effect of Electronic Stability Control on Automobile Crash Risk, Traffic Injury

Prevention, Vol 5, pp. 317-325.

Lie, A., Tingvall, C., Krafft, M., and Kullgren, A. (2004) The Effectiveness of ESP (Electronic Stability

Program) in Reducing Real Life Accidents, Traffic Injury Prevention, Vol 5, pp. 37-41.

Lie, A., Tingvall, C., Krafft, M., and Kullgren, A. (2006) The Effectiveness of ESC (Electronic Stability

Control) in Reducing Real Life Crashes and Injuries, Traffic Injury Prevention, Vol 7, pp. 38-43.

The latest study from the University of Michigan Transportation Research Institute

confirmed the results of earlier studies worldwide – ESC is highly effective in preventing

single-vehicle crashes, especially rollovers in SUVs. Similar to the NHTSA evaluation,

this study analyzed the effectiveness of ESC separately for passenger cars and SUVs.

The study found that for single-vehicle crashes, ESC reduced the risk of a fatal crash

involvement by 31 percent for passenger cars and 50 percent for SUVs. These analyses

were repeated, for both passenger cars and SUVs, after restricting the data to vehicles that

were three years old or newer at the time of the crash – to control for the effect (if any) of

the age of the vehicle. The results showed that the vehicle age factor did not compromise

the significant reductions in the risk of single-vehicle crash involvements associated with

ESC-equipped vehicles. The risks of fatal rollover involvements were also decreased for

passenger cars and SUVs – 40 percent and 73 percent, respectively. For the analysis of

fatal crash involvements, multi-vehicle crash involvements were used as the control

group. The study also analyzed a nationally representative probability sample of crashes

that included mostly non-fatal crashes. It found significant reductions in run-off-road

crashes for passenger cars and SUVs – 55 percent and 70 percent, respectively. For the

analysis of these mostly non-fatal crashes, struck vehicles involved in rear-end crashes

were used as the control group.

The 2004 NHTSA study used crash data from five States from calendar years 1997 to

2002 and Fatality Analysis Reporting System (FARS) data from calendar years 1997 to

2003, because the study was limited to the years when ESC was offered as standard

equipment on certain vehicle models. Mercedes-Benz and BMW were the first two

manufacturers that installed ESC as standard equipment in certain models in 1997 and in

all models by 2000 and 2001, respectively. Their vehicles constituted 61 percent of the

passenger car sample used in the 2004 study. The passenger car sample also includes

some luxury GM cars, which constituted 23 percent of the sample, and a few luxury cars

from other manufacturers. As for the sample used in the analysis for SUVs, Toyota and

Lexus models constituted 78 percent of that sample. In other words, the data samples

used in the 2004 study are limited to mostly imported luxury vehicles, and thus, they are

not well represented across the entire fleet.

Sales of vehicles equipped with ESC are gradually increasing as shown in Table 1. In

2003, merely 10 percent of that year’s vehicle sales are from vehicles that had ESC. By

2006, nearly one-third

of the new vehicles are expected to have ESC. Thus, future

analysis data samples should consist of a more representative cross-section of the fleet

that will include not only non-luxury vehicles but also a wider variety of manufacturers.

The availability of such data is likely to take at least a few more years. Until then,

NHTSA will continue to evaluate the effectiveness of ESC with limited analysis samples

of crash data from selected make-models of vehicles that had ESC and earlier versions of

similar make-models that did not. The make-models used in this study consist of not only

those used in the 2004 study but also other make-models and the model years extended to

one more year.

Green, P. and Woodrooffe, J. (2006) The Effectiveness of Electronic Stability Control on Motor Vehicle

Crash Prevention, Report Number UMTRI-2006-12, University of Michigan Transportation Research

Institute, Ann Arbor, MI.

Table 1: Percent of the Vehicle Sales that are from Vehicles Equipped with ESC

By Vehicle Group and Model Year

Vehicle Group

Model Year

2003

Model Year

2004

Model Year

2005

(expected)

Model

Year 2006

(expected)

Domestic cars*

3.5% 7.7% 8.2% 10.0%

Domestic trucks*

4.9% 10.2% 15.2% 26.6%

Imported cars

36.6% 45.5% 40.5% 54.3%

Imported trucks

31.8% 48.4% 63.4% 73.5%

All new vehicles 9.7% 15.9% 19.2% 28.6%

* Includes transplants

With one more year of crash data recently available, NHTSA has embarked on this

update and modification of its 2004 analysis, extending it to calendar year 2004 for the

FARS analysis and calendar year 2003 for the State data analysis. The primary objective

of this study is to assess the effectiveness of ESC in reducing crashes, specifically crashes

where ESC is likely to have made a difference in the vehicle’s involvement, while

controlling for others in which ESC is unlikely to have been a factor. The study consists

of a series of analyses of crash data of various domestic and imported passenger cars and

LTVs (light trucks and vans, including pickup trucks, SUVs, minivans, and full-size vans

with Gross Vehicle Weight Rating up to 10,000 pounds) from model years 1997 to 2004.

As its principal analytic tool, the agency utilized 2x2 contingency tables of crash data

from specific make-models equipped with ESC versus earlier versions of similar make-

models not equipped with ESC, using non-relevant crash involvements as a control

group. As a check, a logistic regression analysis of the effectiveness of ESC in reducing

relevant crash involvements (specifically single-vehicle crashes which involve a vehicle

running off the road, while controlling for other non-relevant crash involvements) was

also performed using FARS data in order to adjust for the imbalances in various make-

models of vehicles with ESC versus those of similar make-models without ESC and to

account for the confounding effects from other external factors. We also looked

separately at crash reduction with 2-channel and 4-channel ESC systems. The reason for

doing this is that GM make-models used in the analyses (except for the Corvette) all had

2-channel ESC systems, whereas non-GM make-models all had 4-channel systems.

The next several sections discuss in detail: (1) the availability of ESC (particularly,

which make-model passenger cars and LTVs had ESC and which did not); (2) the

analysis databases (specifically, what and how various data files were obtained and used

in the analysis); and (3) the methods of analyzing crash data and estimating the

effectiveness of ESC in reducing crashes – followed by the results from a series of

statistical analyses of crash data. A summary discussion of the findings concludes the

report.

ESC INFORMATION

Due to the fact that (1) ESC was first introduced in the U.S. on certain luxury vehicles in

1997 and (2) it was not until recently that ESC was offered on non-luxury vehicles,

information is not readily obtainable for earlier model years. For example, ESC was not

listed in Buying a Safer Car

until 2001, and Ward’s Automotive Yearbook

did not

begin reporting ESC information in its yearbook until 2004. The yearly-published

Buying a Safer Car booklet lists safety feature information for the current year’s

production of vehicles, whereas Ward’s yearbook contains a list of factory-installed

equipment for the previous year’s production of vehicles. For instance, Ward’s

Automotive Yearbook 2001 contains a list of factory-installed equipment for vehicles

produced in calendar year 2000, whereas Buying a Safer Car 2001 lists safety

information for vehicles produced in calendar year 2001. Hence, information such as the

availability of ESC in vehicles with model years prior to 2000 had to be obtained

elsewhere (e.g., www.edmunds.com, www.cars.com). Furthermore, since ESC is a fairly

new technology for the U.S. market, information on its availability can vary from one

source to another. In such cases, information had to be verified with the manufacturers

for accuracy.

For the purpose of analysis, make-models of passenger cars and LTVs that are equipped

with ESC as standard equipment were compared with earlier versions of similar make-

models. Thus, only vehicles in which 1) the percentage of factory-installed equipment

for “stability control” systems increased from 0 to 100 percent (Wards’ Automotive

Yearbook) or 2) ESC is listed as standard in all vehicles in that model line for particular

model years and not available at all for previous model years (Buying a Safer Car

Booklet, etc.) were included in the analysis. Vehicles with ESC as optional equipment

were excluded from the analysis because we could not determine (from the VIN) which

vehicles had ESC and which did not.

Tables 2 through 4 show ESC availability on selected make-model passenger cars. In

some cases, models that had factory-installed ESC as a standard feature were compared

with earlier versions of identical

models because those models remained the same (e.g.

same chassis generation) during the pre- and post-ESC years – as shown in Table 2. In

other words, for each vehicle listed in Table 2, exactly the same model – before and after

ESC was introduced – was used for comparison. The conversion from not having ESC at

all to having ESC as standard equipment should occur in consecutive model years as

shown in most models except for the Chevrolet Corvette. ESC was introduced as

optional equipment on the Chevrolet Corvette for model years 1998-2000. Those model

years were eliminated from the study.

This book is published yearly by NHTSA.

This book is published yearly by Ward’s Communication, Inc.

We did examine those make-models that had ESC as optional equipment, under the assumption that all

these vehicles did not have ESC, and these additional data did not make a big difference in the overall

effectiveness.

Table 2: "Identical" Make-Model Comparison of Vehicles (except Mercedes-Benz) In

Model Years - With ESC and Without ESC - for Passenger Cars

Make Model

Model Years

with

No ESC

Model Years

with

ESC (Standard)

Acura 3.5RL 1997-1999 2000-2004

Audi A6 (2.7L) Sedan 2000 2001-2004

Audi TT (1.8L) 180hp Coupe 2000 2001-2004

BMW 740I 1997 1998-2001

BMW 740IL Sedan 1997 1998-2001

Buick Park Avenue Ultra 1997-1999 2000-2004

Cadillac DeVille Concours 1997 1998-1999

Chevrolet Corvette 1997 2001-2004

Lexus GS400 1998 1999-2000

Lexus LS400 1997-1998 1999-2000

Saab 9-5 Aero Sedan/Sport Wagon 2000-2001 2002-2004

In other cases, similar make-models (except for the Mercedes-Benz vehicles which will

be discussed separately) were used for the comparison since the original models were

redesigned, not available, or not of the same chassis generation during the transitional

years (Table 3). For instance, the 2000 Audi A4 (1.8L), 2000 A4 (1.8L) Quattro, and

2000 A4 (1.8L) Avant Quattro models were not of the same chassis generation as their

2002-2004 models. ESC was offered as optional feature on the model year 2001 for

those vehicles, and thus, that model year was not included in the study. The 2004 Audi

A8 4.2L Quattro was not available; thus, the 2004 A8 L was included in the analysis

instead. The 1997 and 1998-2004 Cadillac Seville SLS and STS models were also on a

different chassis, and so did the 1997 and 1998-2004 Lexus GS300, 1997-1999 and 2001-

2003 Oldsmobile Aurora (4.0L), and 1999 and 2002-2004 Volkswagen GTI VR6 models.

As previously mentioned and currently illustrated in Table 3, BMW installed ESC in

certain 5, 7, and 8 series models as early as 1997 and had made it standard in all their

models by 2001, but the installation did not necessarily occur in the same year for all the

sub-series of a make-model. For example, the 3 series coupe and sedan models had ESC

in 2000 whereas the 3 series convertible had it in 2001. Therefore, comparison of

vehicles with ESC and those without ESC was done at the sub-series level. Furthermore,

some manufacturers often changed their sub-series models; thus, it is very difficult to

make an exact model-to-model comparison since the analysis databases consist of crash

data of various make-models ranging from model years 1997-2003 for the State data

analysis and 1997-2004 for the FARS data analysis. In such cases, the best approach

would be to compare similar sub-series models as shown in Table 3. Since BMW’s 3 and

5 series models were completely redesigned and/or grouped into different sub-series at

some point during model years 1997-2004, different versions of the 3 and 5 series were

used depending on which model years had ESC and which did not. Similar to BMW,

General Motors changed their Pontiac Bonneville SSE model to SSEi in 2000 and

installed ESC in the new model as well. The reason that model year 2000 was not

included in the analysis for the BMW 3 series convertible is because that model was not

available in 2000. Model year 1999 was excluded for the BMW 525i and 530i sedan for

the same reason. The BMW 528i sedan was available in 1999; however, it did not

include ESC as standard feature – only as an option. Hence, the 1999 BMW 528i, 525i,

and 530i sedans were not included in the study. For the Volkswagen Passat, the 2000

GLS and GLX models were compared with the 2004 GLX model because the 2003

model year for both make-models had ESC as an option. The 2004 GLS model had

optional ESC as well. Likewise, the 2002 Volkswagen New Beetle (1.8L) was compared

with the 2002-2004 Turbo S models because the model years 2003-2004 of the former

had optional ESC.

Table 3: "Similar" Make-Model Comparison of Vehicles (except Mercedes-Benz) In

Model Years - With ESC and Without ESC - for Passenger Cars

Make Model

Model Years

with

No ESC

Make Model

Model

Years with

ESC

(Standard)

Audi A4 (1.8L) Sedan 2000 Audi A4 (1.8L) Sedan 2002-2004

Audi A4 (1.8L) Quattro

Sedan

2000

Audi A4 (1.8L) Quattro Sedan 2002-2004

Audi A4 (1.8L) Avant

Quattro Wagon

2000

Audi A4 (1.8L) Avant Quattro

Wagon

2002-2004

Audi A8 (4.2L) Quattro

Sedan

1998-1999 Audi A8 (4.2L) Quattro Sedan

Audi A8 L Sedan

2001-2003

2004

BMW

318/323/325/328is/M3

Coupe (6cyl)

1997-1999 BMW

323CI/325Ci/328Ci/330/M3CI

Coupe (6cyl)

2000-2004

BMW 323iC/328iC/M3

Convertible

1997-1999 BMW 325Ci/330Ci/M3Ci

Convertible

2001-2004

BMW 318/323i/328iSedan;

BMW 318TI 2Dr.

1997-1999 BMW 323/325i/328i/330i/Xi

Sedan

2000-2004

BMW 528i Sedan 1997-1998 BMW 528i/525i/530i Sedan 2000-2004

BMW 540i (AT/MT) Sedan 1997 BMW 540i (AT) Sedan 1998

BMW 540i (MT) Sedan 1998 BMW 540i/545i (AT/MT) Sedan 1999-2004

BMW Z3 (2.8L) 1997-1999 BMW Z3 (2.3L,2.5L,2.8L,3.0L) 2000-2002

Cadillac Seville SLS 1997-1998 Cadillac Seville SLS 1999-2004

Cadillac Seville STS 1997-1998 Cadillac Seville STS 1999-2004

Lexus GS300 1997-1998 Lexus GS300 1999-2004

Oldsmobile Aurora (4.0L) 1997-1999 Oldsmobile Aurora (4.0L) 2001-2003

Pontiac Bonneville SSE 1997-1999 Pontiac Bonneville SSEi 2000-2004

Volkswagen GTI VR6 1999 Volkswagen GTI VR6 2002-2004

Volkswagen Passat

GLS/GLX Sedan/Wagon

2002 Volkswagen Passat GLX

Sedan/Wagon

2004

Volkswagen New Beetle

(1.8L)

2002 Volkswagen New Beetle Turbo S

(1.8L)

2002-2004

Like BMW, Mercedes-Benz began installing ESC in certain sub-series of the S class

model in 1997 and had made it a standard feature in all their make-models (except one)

by 2000. For this reason, not all the sub-series (even within the same make-model – i.e.

“class”) had ESC in the same model year. Furthermore, most sub-series were changed

during calendar years 1997-2004 – either redesigned or replaced. Table 4 lists all the

available Mercedes-Benz models and sub-series from 1997 to 2004, but only with the

model years that had ESC as standard equipment and those that did not. In Table 4,

various sub-series within the Mercedes-Benz C, E, and S class that are equipped with

ESC were compared with earlier versions of similar

sub-series, for reasons already

discussed in the previous section. Unlike other manufacturers, quite a few Mercedes-

Benz sub-series had ESC as optional equipment for one, two, or even three years; hence,

those model years were not included in the analysis. For example, the C280, E300DT

sedans and the CL500C coupe all had ESC as an optional feature from 1998 to 1999.

Table 4

: Mercedes-Benz Make-Model Comparison of Model Years With ESC and

Without ESC - for Passenger Cars

Make Model No ESC Make Model

ESC

(Standard)

C36AMG

C220, C230 Sedan

C280 Sedan

1997

1997-1999

1997

C43 Sedan

C230ML Sedan

C280 Sedan

C Sedan

CL Coupe

CL600 Coupe

CLK55AMG Coupe/Cabriolet

CLK320 Coupe/Cabriolet

CLK320/CLK500

Coupe/Cabriolet

CLK430 Coupe/Cabriolet

1998-2000

2000

2001-2004

2000-2004

1998-1999

2001-2002

2000-2003

2003-2004

1998-2003

E300D Sedan

E320 Sedan

1997

E320 Sedan/Wagon

E320W/E500 Sedan

E430/E55 AMG Sedan

2000-2003

2003-2004

1999-2002

S320 SWB Sedan

S320 LWB Sedan

SL320 Roadster

SLK230 Kompressor

1997-1999

1997

1998-2000

S430/S500/S600 Sedan

S600 Sedan

S600 Coupe

SL500 Roadster

SL500/SL55 Roadster

SL600 Roadster

SLK230/SLK320/SLK32AMG

2000-2004

1997-1999

1997

1999-2002

2003-2004

1997-2002

2001-2004

Tables 5 through 7 show the availability of ESC by make-model for selected LTVs. As

shown in these tables, the make-models used in the analysis consist of SUVs and

passenger vans – because most manufacturers had not yet installed ESC in their pickup

and minivan models as of 2004 – the most recent and available fatal crash data used in

this study. With the exception of the Infiniti, Land Rover, Toyota 4 Runner, Mercedes-

Benz, and one Lexus model, none of the models used in the analysis went through a

redesign in the year that ESC became standard equipment or even in the year

immediately before or after it (Table 5). Thus, we were able to compare several model

years of identical make-models for the vehicles with ESC versus those without ESC.

Furthermore, these vehicles went from not having ESC at all in one year to having ESC

as standard equipment in the following year.

Table 5: "Identical" Make-Model Comparison of LTVs In Model Years

With ESC and Without ESC

Make Model

Model Years

with

No ESC

Model Years

with

ESC (Standard)

Acura MDX 2001-2002 2003-2004

Cadillac Escalade 2WD 2002 2003-2004

Chevy Express 3500 Extended Wheelbase 2003 2004

GMC Yukon Denali AWD 2001-2002 2003-2004

GMC Yukon Denali XL AWD 2001-2002 2003-2004

GMC Savana G3500 Extended Wheelbase RWD 2003 2004

Lexus LX470 1999 2000-2004

Mitsubishi Montero Limited 2001-2002 2003-2004

Toyota Land Cruiser 1999-2000 2001-2004

Toyota RAV4 4x2 2003 2004

Toyota RAV4 4x4 2003 2004

As for the Infiniti vehicles listed in Table 6, since there are no previous versions of the

FX35 AWD, FX45AWD, and QX56 models – which had ESC starting in 2003 – the

model years 2002-2003 of the QX4 model were used as comparison vehicles that did not

have ESC. The 2001-2002 and 2003-2004 Land Rover Range Rover models were not of

the same chassis generation. The Toyota 4Runner had a design change in 2003 for both

the 4x2 and 4x4 models. The static stability factor was higher on the 2003-04 4Runners.

Hence, the later model-year vehicles (post 2002 model year) are more stable than the

earlier model-year vehicles. Thus, one would expect the 2003-04 4Runners to be

involved in fewer rollover crashes than the earlier models, even if the newer models are

hypothetically not equipped with ESC. For that reason, the 2003-04 model years were

not included in the analysis.

Lexus changed their RX300 model to RX330 in 2004, the

new model was included in the analysis.

The Mercedes-Benz ML320 – Mercedes’ first ever U.S. SUV model – came on the

market in 1998 but was not equipped with ESC. A year later, Mercedes-Benz had made

Charles Farmer, in his peer review of this report, recommended the exclusion of 2003-2004 Toyota

4Runners.

ESC standard equipment in all their SUV models. We could have included only the

ML320 model in the analysis – since the model was still available in 2004; however, we

decided to include other Mercedes SUV models in the analysis to increase our sample

size.

Table 6: "Similar" Make-Model Comparison of LTVs In Model Years

With ESC and Without ESC

Make Model

Model Years

with

No ESC

Make Model

Model Years

with

ESC (Standard)

Infiniti QX4 2WD

Infiniti QX4 4WD

2002-2003

Infiniti FX35 AWD

Infiniti FX45 AWD

Infiniti QX56 4x2

Infiniti QX56 4x4

2003-2004

2004

Land Rover Range

Rover

2001-2002 Land Rover Range Rover 2003-2004

Lexus RX300 1999-2000 Lexus RX300

Lexus RX330

2001-2003

2004

Toyota 4Runner 4x2 1999-2000 Toyota 4Runner 4x2 2001-2002

Toyota 4Runner 4x4 1999-2000 Toyota 4Runner 4x4 2001-2002

Table 7: "Similar" Make-Model Comparison of Mercedes-Benz LTVs In Model Years

With ESC and Without ESC

Make Model

Model Years

with

No ESC

Make Model

Model Years

without

ESC (Standard)

ML320

1998 ML320

ML350

ML430

ML500

ML55

1999-2004

2003-2004

1999, 2001-2002

2002-2004

2000,2004

CRASH DATA (STATE AND FARS)

Once specific make, model, and model year of vehicles – equipped with ESC and those

not equipped with ESC – were selected for passenger cars and LTVs, those vehicles can

then be identified in the crash data for analysis. In general, a large sample of crash-

involved cars from the selected models is desirable to statistically estimate the effect of

ESC in reducing crashes. The best source of data available to NHTSA for this analysis is

the State data files maintained by the agency’s National Center for Statistics and Analysis

(NCSA). NCSA currently receives crash data from 28 States and maintains these data

files for calendar years 1989 and onward. Since ESC is currently not required safety

equipment, its presence in crash-involved vehicles is not encoded as a separate field in

the police reports. However, we can identify the make, model, and model year of each

crash-involved vehicle from the State data files by decoding the Vehicle Identification

Number (VIN) – if such information is available.

Twenty-one States were excluded from the analysis for various reasons. For instance,

Alabama, Arkansas, Colorado, Delaware, Indiana, Minnesota, Montana, South Carolina,

Texas, and Virginia do not have VIN information in their files. Georgia, Michigan, North

Carolina, New York, and Washington were excluded because NHTSA does not have

their files beyond calendar year 2000, a critical year in the analysis because the majority

of our passenger car sample consists of Mercedes-Benz and BMW vehicles – the first two

manufacturers that offered ESC on all their models by 2000 and 2001, respectively.

Thus, without the post 2000 data, our data sample would consist of mostly crashes from

non-ESC vehicles. Kansas, New Mexico, and Ohio were not uniform in reporting VIN,

having a low percentage of VIN in some years and a high percentage in others. Utah and

Wyoming are States with small numbers of crashes. Maryland data in nearly half of the

reported cases do not indicate the contributing circumstances for the vehicles involved in

the crashes. We must have this information to 1) determine which vehicle was

responsible for the crash and which was not and 2) classify each vehicle’s involvement as

either the control group or response group (relevant) involvement.

As a result, data from seven populous States (California, Florida, Illinois, Kentucky,

Missouri, Pennsylvania, and Wisconsin) were used in the analysis. These States (except

California) consistently have a high percentage of VIN information in their data files

(Table 8). California does not have VIN information in its data files. Thus, we were not

able to identify (from the VIN) specific make-models that had ESC and those that did

not, but we included California in the analysis because it is a large State with large

numbers of crash data. We know that Mercedes-Benz first installed ESC in certain S

class models and had made it standard equipment in all their passenger car models by

2001. Mercedes-Benz did have a few models that had ESC as optional equipment prior

to 2001. As previously mentioned, we did examine vehicles with ESC as optional

equipment under the assumptions that those vehicles did not have ESC, and the initial

results indicated that the additional data did not make a big difference in the overall

effectiveness. Thus, for the analysis of crash data in California, we assumed that all

Mercedes-Benz passenger car models prior to 2000 did not have ESC at all. As for their

SUV models, Mercedes-Benz installed ESC in all their models from model years 1999

and onward. Thus, only the 1998 Mercedes-Benz SUV models did not have ESC. As a

result, in the California analysis, we were able to analyze crashes that involved only

Mercedes-Benz vehicles – because we were able to identify those vehicles using the

variables MAKE, MOD_YR (model year), and CHP_TYP (California Highway Patrol

vehicle type – to differentiate cars from SUVs).

Although the number of vehicles involved in fatal crashes is relatively small, ESC is

believed to be highly effective in reducing fatal crashes. If so, even a relatively small

sample size will suffice for statistically significant effects. Currently available crash data

from Fatality Analysis Reporting System (FARS) from calendar years 1997-2004 were

used for the fatal crash analyses. FARS data files consistently have high percentage of

VIN information in their data files (97% or better). Thus, the make, model, and model

year of fatal crash-involved vehicles can be decoded from the VIN information.

Table 8: Percentage of VIN Information by State and Calendar Year

State

Calendar

Year

1997

Calendar

Year

1998

Calendar

Year

1999

Calendar

Year

2000

Calendar

Year

2001

Calendar

Year

2002

Calendar

Year

2003

Florida 90% 90% 90% 89% 88% 90% 90%

Illinois 86% 88% 87% 77% 80% 87% -

Kentucky 95% 95% 92% 97% 97% 97% -

Missouri 84% 87% 87% 88% 87% 93% 93%

Pennsylvania 93% 94% 94% 94% 95% - 95%

Wisconsin 91% 91% 90% 91% 90% 90% 90%

Since NCSA typically receives crash data from the States two years after or even longer

in some States, our State data analyses consist of crash data from calendar years 1997-

2003 for Florida, 1997-2002 for Illinois, 1997-2002 for Kentucky, 1997-2001 and 2003

for Pennsylvania, and 1997-2003 for Wisconsin – Table 9. Currently, the 2003 Illinois,

2003 Kentucky, and 2002 Pennsylvania crash files are not available. Only data from

calendar years 2001-2003 were included in the California analysis because we could not

distinguish (from the “vehicle type” variable in the data files prior to 2001) between

crashes involving passenger cars and those involving SUVs. Post-2000 California data

files differentiate passenger car and SUVs crash involvements. As for the fatal crash

analyses, FARS data files are generally available one year after. Thus, our FARS data

analysis consists of fatal crash data from calendar years 1997-2004.

Table 9: Calendar Years of Crash Data By State

State Calendar Years

California 2001-2003

Florida 1997-2003

Illinois 1997-2002

Kentucky 1997-2002

Missouri 1997-2003

Pennsylvania 1997-2001, 2003

Wisconsin 1997-2003

ANALYSIS DATABASES

A) VIN Decode

The analysis databases are crash files initially obtained from the FARS and State data

files and then decoded from the VIN to include only the selected make-models listed in

Tables 2-7. In other words, in Florida, Illinois, Kentucky, Missouri, Pennsylvania, and

Wisconsin State data files, vehicle make, model, and model year were decoded from the

variable VIN and matched with the VIN information obtained from the Passenger

Vehicle Identification Manual

for the selected make-models. To ensure that there are

no discrepancies in reporting VIN information of crash-involved vehicles, the decoded

model year from the variable VIN had to match with the variable MOD_YR (model year)

taken directly from the State data files. Since California does not have the VIN variable,

the selected Mercedes-Benz make-models were obtained by using the variable MAKE in

the data files. Like other States, California does have the variable MOD_YR in its data

files; thus, the model year of crash-involved vehicles was obtained directly from that

variable.

Like the State data files (except California), FARS data files include both the variables

VIN and MOD_YEAR (model year). Thus, we were able to extract information that

identified the vehicle make, model, and model year from the VIN variable and

established an analysis database consisting of crash data of make-models listed in Tables

2-7. Also, the model year taken from the VIN variable had to correspond with the

variable MOD_YEAR.

The resulting analysis databases – whether they are FARS or State files – are vehicle-

oriented files, with one record for each vehicle that was involved in a crash. Since these

files only contain the selected make-models from Tables 2-7, only vehicles with model

years 1997-2004 for the FARS files and model years 1997-2003 for the State files were

included in the study. One of the critical parameters used in the databases is the

parameter that identifies the presence or absence of ESC. Each record in the databases

will have a variable ESC with a value of 1 for a vehicle that has ESC and a value of 0 for

the one that does not.

B) Crash involvements

Certain crash involvements in which a vehicle 1) was stopped, parked, backing up, or

entering/leaving a parking space prior to the crash, 2) traveled at a speed less than 10

miles per hour (mph), 3) struck in the rear by another vehicle, or 4) was a non-culpable

party in a multi-vehicle crash on a dry road, were considered the control group (non-

relevant involvements) – because ESC would not have prevented the crash. Since ESC

has the ability to detect when a driver is about to lose control of a vehicle and the

capability to automatically intervene to assist that driver and ultimately reduce the

likelihood of a roadside departure, it may provide considerable benefits to crashes that

involve a vehicle running off the road or running out of the lane and hitting other

vehicles. Most run-off-road crashes are single-vehicle involvements, and these

involvements include rollovers and collisions with fixed objects. Other relevant crash

involvements – where ESC could be a factor in the vehicle’s involvement – include

collisions with one or more vehicles due to the driver’s error or the vehicle’s

unsatisfactory performance (known as culpable involvements in multi-vehicle crashes).

These crashes are relevant because they may (but not necessarily) have involved loss of

This manual is published yearly by the National Insurance Crime Bureau (NICB).

control. Hence, these relevant involvements were included in the analysis and considered

the response group. We also analyzed crashes that involved a pedestrian, bicycle, or

animal – although the vehicles involved in these crashes were not typically at fault. We

would still be interested in knowing whether or not ESC is effective in reducing these

types of crashes.

An important concept in setting up a control- and response-group experiment is that not

every vehicle in the response group necessarily went out of control or stood to benefit

from ESC. For example, a careless driver could steer a vehicle off the road or into the

path of another vehicle with full directional control. Moreover, these crash data generally

do not specify if a vehicle “went out of control” or not. Instead, it is the converse that is

true: if a vehicle did go out of control and get into a crash, it will be in a response group

and not in the control group. The vehicles that could have benefited from ESC are a

subset of the response-group involvements and are not control-group involvements. As

mentioned previously, vehicles that most likely would not benefit from ESC are

considered the control-group involvements – and while this group ideally should be as

unaffected as possible by ESC, the effectiveness estimates are quite dependent on the

choice of the control group. For example, if crashes that were affected by ESC were

included in the control-group involvements, the estimate of ESC effectiveness would be

lower. On the other hand, if crash involvements that were unaffected by ESC were

included in the response-group, the effectiveness estimate would not be compromise as

long as the set-up of a control- and response-group experiment was correct and valid.

The next several sections discuss how these crash involvements were obtained. The

control-group involvements (non-relevant) will be discussed first followed by the

relevant involvements – the response group.

Control group – non-relevant crash involvements

Ideally, only crash involvements where a vehicle was standing still prior to the crash,

moving less than 10 miles per hour, or backing up should be considered the control group

– since we are almost certain that these vehicles were moving too slowly to trigger ESC

and make it a factor in the vehicle’s involvement. But such fatal involvements are few in

number – when compared to the relevant involvements. Thus, a larger sample of vehicles

is needed for the control group. Other non-relevant involvements where ESC is unlikely

to provide any benefit include situations where 1) a vehicle was struck in the rear in a

rear-end collision or 2) a vehicle was involved in a multi-vehicle crash on a dry road but

was not at fault. In both situations, the motion of this vehicle is not what precipitated the

crash, and this vehicle definitely did not hit something else because it had gone out of

control.

In FARS, vehicle “maneuver” and “travel speed” prior to the crash were used to define

one portion of the control-group involvements in which a crash-involved vehicle was

either at a complete stop or moving at a low speed – less then 10 miles per hour. In

addition, the maneuvers “backing up”, “parking”, or “leaving a parking space” indicate

the vehicle was moving slowly (most likely less than 10 mph) even if FARS does not

specify a travel speed. The other portion of the control group includes vehicles that are

considered a non-culpable party in a fatal multi-vehicle crash on a dry road. To

determine whether or not a vehicle was at fault, related factors such as the driver’s

physical and mental condition, attitude, and driving actions as well as the condition of the

vehicle and roadway (prior to the crash) were taken into consideration. These factors are

coded in FARS and State data files. Essentially, a vehicle is in the control group if none

of these factors were present. Multi-vehicle crash involvements will be discussed

separately in the next section.

In theory, “not at fault” involvements on wet roads should be in the control group.

However, given that many involvements follow loss of control, and given some

uncertainty as to the accuracy of the “fault” determination, those involvements were not

assigned to the control group.

Finally, in a front-to-rear collision (neither vehicle backing up), we will always assign the

rear-impacted vehicle to the control group and the frontally impacting vehicle to the

response group.

As for the States, every State has its own unique way of coding vehicle “maneuver”,

“speed”, and initial “point of impact”, and not every State includes these variables in their

data files. Similar to FARS, contributing circumstances relating to the driver, vehicle,

and roadway were used to determine which vehicle was responsible for the crash and

which was not. If a State has a variable that indicates whether or not the vehicle’s driver

was charged with a moving violation, then we also used that variable to classify the

vehicle’s involvement (culpable or non-culpable) in a multi-vehicle crash. Again, not all

variables are included in every State, and even the same variables are not coded exactly

the same in each State. Thus, depending on what crash information is available in each

State, the control group used in each of the State data analyses includes different types of

non-relevant crash involvements. But at least the definitions for each of these types of

involvements are made as similar as possible.

The definitions used in all the State data analyses for “stopped”, “parked”, “backing”, and

“parking” maneuvers are self-explanatory and were taken directly from the variable

VEH_MAN1 in the State data files. If a State also reports the speed of a vehicle (variable

SPEED) prior to the crash, then the crash-involved vehicle with a speed ranging from 0 to

10 mph would be included in the control group as well. Similar to the control group used

in the FARS analysis, the multi-vehicle crash involvements had to occur on dry roads and

be non-relevant (non-culpable) – in order for those involvements to be included in the

control group for the State data analyses. The next section will discuss in detail the

different types of multi-vehicle crash involvements used in the FARS and State analyses.

Classification of multi-vehicle crash involvements (control and response group)

Multi-vehicle crash involvements were classified as either culpable or non-culpable based

on what information is available in the data files. In many crashes that involve two motor

vehicles in transport, one vehicle can be identified as responsible for the crash, and the

other one not. In some State files this is almost a hard-and-fast rule, with one culpable

and one non-culpable vehicle in nearly every crash, but in other States there may be many

crashes where none or sometimes even all the vehicles are judged culpable. In multi-

vehicle crashes other than front-to-rear collisions, the action of the driver, the

characteristic of the vehicle, the condition of the roadway, or simply the traffic violation

charged were taken in account to determine which vehicle(s) (if any) contributed to the

crash and which did not. In general, the vehicle’s involvement – where the driver in that

vehicle was distracted or under the influence of drugs and/or alcohol, drove carelessly or

aggressively, failed to yield right-of-way, performed an improper turn, backing, passing,

or lane change, followed too closely, disregarded a traffic signal, sign, or other traffic

control, or exceeded the stated speed limit – was considered the most likely to cause the

crash. Other contributing circumstances include the condition of the vehicle’s tires (i.e.,

worn, blowout, puncture, etc.) or the malfunction of the vehicle’s brake systems or the

steering mechanism prior to the crash. Although certain roadway conditions (i.e., wet,

slush, snow, icy, sand, dirt, oil, etc.) might have caused the driver to lose control of the

vehicle and run off the road or hit another vehicle, the striking vehicle would still be

considered the culpable party. Furthermore, if a crash-involved vehicle in which the

driver were charged with a traffic violation, then that vehicle would be considered the

offender. Other contributing factors – whether they are driver, vehicle, or roadway

contributing circumstances – were not considered evidence of culpability. As an

exception to this approach, in a two-vehicle front-to-rear collision, neither vehicle

backing up, the collisions are identified by MAN_COLL=”rear-end” on the accident level

and IMPACT1 (or its equivalent) on the vehicle level: one vehicle has frontal damage

and the other, rear damage. The vehicle with frontal damage (known as rear-end striking)

will always be classified as the culpable party, and the one with rear damage (known as

rear-end struck) will be non-culpable. The criteria for determining culpability in the

FARS analysis are similar to those used in the State analysis. In addition to the ideal

control-group involvements (i.e., vehicle was parked, stopped, or traveled at very slow

speed), the control-group also included non-culpable involvements in multi-vehicle

crashes on dry roads.

Up until now, we have discussed all control-group involvements in which a crash-

involved vehicle was 1) standing still, 2) moving less than 10 miles per hour, 3) backing

up, 4) parking or leaving a parked position, 5) being struck in the rear by another vehicle,

or 6) non-culpable in a multi-vehicle crash on a dry road. We have also talked about

certain relevant involvements (i.e., response group) where 1) a crash-involved vehicle

was considered a culpable party in a multi-vehicle crash on any roads (dry or wet) and 2)

ESC could play a critical role in preventing those crashes.

Categories of single-vehicle crashes

Now, let us discuss further other relevant involvements – mainly crashes that involve

only one motor vehicle in transport – single-vehicle crashes. In most single-vehicle

involvements – whether they are fatals or non-fatals – the case vehicle either rolled over

or hit a fixed object. Those incidents will most likely occur when the vehicle

unintentionally

leaves the travel lane (runs off the road), which is an indication of a

driver’s error due to loss of steering and/or directional control of the vehicle, because it is

very unlikely for a vehicle (even for the high center-of-gravity vehicle) to tip over or hit a

fixed object while it is still in the travel lane and under control. In other single-vehicle

involvements where the driver intentionally

steered the vehicle away from the incoming

object (i.e., pedestrian, bicycle, or animal) to avoid hitting the object but ended up hitting

it, the driver is typically not at fault

in this case, but the involvement is still considered

relevant because ESC could perhaps play a role in the vehicle’s involvement.

Run-off-road crashes involve vehicles that travel out of the lane – where they may or may

not yaw out of control, leave the roadway, and eventually roll over and/or hit one or more

natural or artificial objects along or off the roadside – such as trees, guardrails,

embankments, etc. They can also occur on the median of a divided roadway or on the

other side of a non-divided roadway. Run-off-road crashes typically involve a single

motor vehicle in transport – although a moving vehicle hitting a parked vehicle could also

be considered a multi-vehicle involvement. For the purpose of analysis, those

involvements are treated as single-vehicle involvements.

We know from our preliminary study, as well as studies by other researchers that ESC

appears to provide significant benefits related to single-vehicle crashes – especially those

that are likely to have involved yawing: rollovers and side impacts with fixed objects.

“Rollover” generally involves a vehicle that went out of control, started to yaw while it

was still on the road, and eventually rotated 90 degrees or more, side-to-side or end-to-

end, as a result of a tripping mechanism when it left the roadway. “Side impact with a

fixed object” usually indicates a vehicle that left the roadway, possibly after yawing out

of control, and collided with a fixed object. We were able to define these crash

involvements using the “first harmful event” variable at the accident level and the “initial

point of impact” variable at the vehicle level. These variables are coded in the FARS and

State data files.

In the 2004 NHTSA’s report, we studied ESC by looking at its effectiveness in reducing

single-vehicle crashes as a whole and excluding crashes that involve a vehicle colliding

with a pedestrian, bicycle, or animal. In other words, the response group (used in the

2004 study) excluded all crash involvements with a pedestrian, bicycle, or animal and

included all other involvements with one motor vehicle in transport as single-vehicle

crashes. In this follow-up study, we not only included pedestrian, bicycle, and animal

crashes in our study but also looked at other single-vehicle crashes separately. In other

words, we analyzed separately – the effectiveness of ESC in reducing 1) rollovers, 2) side

impacts with fixed objects 3) other run-off-road crashes, 4) pedestrian, bicycle, and

animal crashes, and 5) other single-vehicle crashes (such as collisions with parked cars,

impacts with thrown or falling objects, undercarriage scrapes, or first-event fires).

Rollovers, side impacts with fixed objects and other run-off-roads are considered “all

run-off-road” crashes, and they are a subset of the single-vehicle involvements used in

the 2004 study. “All run-off-road” and “other single-vehicle” involvements used in this

study are what we have used in the 2004 study as “single-vehicle crashes” – since we

excluded pedestrian, bicycle, and animal crashes in that study. However, a direct

comparison of the relative effectiveness of ESC should be made with caution since the

control groups used in the two studies are not the same – as already discussed in the

previous section. In this study, we also included culpable involvements in multi-vehicle

crashes on all roads as part of the response group – because ESC could be a factor in the

vehicle’s involvement. We also calculated the effect of ESC on all crash involvements,

including the control-group involvements as well as the relevant involvements.

2X2 CONTINGENCY TABLE ANALYSIS

This study consists of a series of analyses of crash data from currently available State and

FARS databases. Crash data from calendar years 1997-2003 from seven States as well as

FARS data from calendar years 1997-2004 were used in the analysis. The analysis then

compares specific make-models of passenger cars and LTVs with ESC versus earlier

versions of similar make-models, using non-relevant crash involvements as a control

group, essentially creating 2x2 contingency tables of crash involvements (control group

and response group) of vehicles with ESC and those without ESC. A general notation of

a 2x2 contingency table is shown below in Table 11.

Table 11: General Notation for a 2x2 Contingency Table

Type of crash

involvement

Vehicles

Without ESC

Vehicles With

ESC

Totals

Control-group

(non-relevant crashes)

a b

a + b

Response group

(relevant crashes)

c d c + d

Total a + c b + d a + b + c + d = N

To detect if ESC is effective, we considered and tested the null hypothesis that the ratio

of d to c is the same as the ratio of b to a. To calculate the effectiveness of ESC, first, we

compute the ratios of relevant crash involvements to control-group involvements. Then,

we compute the percentage reductions in these ratios in vehicles with ESC versus earlier

versions of similar make-models without ESC. These calculations are illustrated below

by the formula:

essEffectiven

(%)

100)]}//()/[(1{

−

abcd

To test for statistical significance of the effectiveness of ESC, we used the chi-square

statistic )(

for the 2x2 table.

For a 2x2 contingency table, the degrees of freedom are always 1. For a two-sided 95%

confidence level (a conventionally accepted significance level) and degrees of freedom

equal to 1, a chi-square statistic has to exceed 3.843 – which indicates that the two

distributions are not the same – to reject the null hypothesis and accept our hypothesis

that the ratio of relevant to control-group crashes is lower in the ESC-equipped vehicles

than in the comparison vehicles without ESC. We will also accept as “statistically

significant” a one-sided 95% confidence level, when chi-square exceeds 2.71. NHTSA

evaluations of safety equipment customarily employ the more lenient one-sided test when

there is a clear expectation that the effect of the equipment will be in the “right” direction

(saving lives, preventing crashes) or, at worst, zero. It is unlikely to be negative. Only in

situations where there is no realistic a priori expectation of an effect in either direction do

we rely exclusively on the more stringent two-sided test.

The next several sections will show 2x2 contingency tables of different types of crash

involvements (FARS and State files) for vehicles with ESC versus those with no ESC.

As previously discussed, the control group includes crash involvements in which a

vehicle was standing still prior to the crash, moving less than 10 mph or backing up,

struck in the rear, or non-culpable party in a multi-vehicle crash on a dry road. The

tables will also show different types of relevant crash involvements such as single-vehicle

run-off-road, other single-vehicle collision, culpable involvement in a multi-vehicle

collision, and collision with a pedestrian, bicycle, or animal. Run-off-road involvements

such as side impacts with fixed objects, rollovers, and other run-off-roads will also be

included in these tables. All run-off-road involvements discussed in this study are single-

vehicle involvements. In addition, the effectiveness of ESC and the chi-square statistic of

each type of relevant involvement will be calculated and illustrated in these tables.

Lastly, the overall effectiveness of ESC on all crash involvements (including the

culpable, non-culpable, and control-group involvements) will be shown in these tables –

to demonstrate the overall effect of ESC since these relevant involvements consist of only

a fraction of all the involvements. Thus, the actual effectiveness will always be less

when considering all crash involvements.

For instance, if the effectiveness of ESC is

estimated to be 30 percent for the relevant involvements (i.e., all run-off-roads, other

single-vehicle collisions, collisions with pedestrians, bicycles, or animals, and all

culpable multi-vehicle collisions), which are 50 percent of all crash involvements, then

the effectiveness on all crash involvements is reduced to

(0.30 x 0.50) x 100 = 15 percent

As previously mentioned, the control group includes not only crash involvements in

which a vehicle was standing still prior to the crash, moving at a very low speed, or

backing up, but also other involvements where a vehicle was considered a non-culpable

party in a multi-vehicle crash on a dry road. For completeness, these tables will also

show other multi-vehicle involvements for vehicles with ESC and those without ESC –

so that the sum of all the involvements (control group, all run-off-road, other single-

For the analysis of fatal crashes obtained from the FARS database, all crash involvements refer (in this

study) to all types of fatal crash involvements. For the analysis of police-reported crashes obtained from

the State data files, all crash involvements

also refer to all types of crash involvements including not only

fatal

but also non-fatal crash involvements (i.e., property damage, possible injury, non-incapacitating and

incapacitating injury).

vehicle, pedestrian, bicycle, animal, culpable involvements as well as other involvement

in multi-vehicle crashes) equals the “total” involvements, which are shown prior to the

last row in each of the tables. But the effectiveness and chi-square statistic will not be

calculated for other multi-vehicle involvements because interpretation of the

effectiveness estimates on those involvements is not clear – since these involvements also

included the non-relevant involvements in multi-vehicle crashes on non-dry roads. As for

the “total” involvements, the effectiveness calculation is also not suitable because they

also include the control-group involvements.

Fatal crashes in passenger cars

Now let us first look at the effectiveness of ESC in fatal crashes in passenger cars. As

illustrated in Table 12, ESC reduced fatal run-off-road crashes by 36 percent

[1 – {(154/217)/(183/166)}]. The reduction is statistically significant with a chi-square

statistic of 8.62 for the 2x2 table consisting of these four cells. It is also evident that fatal

rollover risk is substantially lower with ESC – a statistically significant 70 percent

reduction relative to the control group. Fatal side impacts with fixed objects were

significantly reduced as well – as shown by the 49 percent effectiveness and a 5.89 chi-

square statistic. The negative effectiveness (-36 percent) relative to the control group in

pedestrian, bicycle, and animal crashes suggests possible increase, but the increase is

statistically non-significant. On a more positive note, culpable involvements in multi-

vehicle crashes were also decreased (19 percent), but the reduction is not statistically

significant. Overall, ESC is still very beneficial (14 percent effectiveness) if all crash

involvements were considered – which included culpable, non-culpable, and the control-

group involvements. To calculate the overall effectiveness, the sum of all run-off-roads,

other single-vehicle, pedestrian, bicycle, and animal involvements, as well as culpable

and non-culpable involvements in multi-vehicle crashes must first be calculated for

vehicles with ESC and those with no ESC [(154+3+69+157+76=459) and

(217+6+46+176+65=510), respectively]. The effectiveness in these non-control group

involvements relative to the control-group involvements can then be calculated [1-

{(459/183)/(510/166)} = 18 percent]. The effect of ESC on all fatal crash involvements

for passenger cars can finally be determined by multiplying the effectiveness in non-

control group involvements by the proportion of the total crash involvements that were

relevant [18 x (510/676) = 14 percent]. The effect of ESC on all crash involvements is

not statistically significant because the effect on all non-control group crash involvements

is not.

As previously discussed, “all run-off-road” and “other single-vehicle” involvements

used in this study are what we have categorized in the 2004 study as “single-vehicle

crashes” – which did not include pedestrian, bicycle, and animal crashes. For

comparison purpose (though the control groups used in the 2004 study and in this study

are not the same), we also included in the analysis tables the effectiveness estimates of

ESC in reducing single-vehicle crashes except those involving pedestrians, bicycles, or

animals. The results from this study (shown in Table 12) suggest that ESC reduced fatal

single-vehicle crash involvements of passenger cars by 36 percent – if pedestrian,

bicycle, and animal crashes were excluded – and the reduction is statistically significant.

Table 12: Effect of ESC on Fatal Crashes in Passenger Cars

(FARS: 1997-2004)

Type of fatal crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectiveness

*Chi-

square

Value

Fatal

Crashes

Fatal

Crashes

(%)

Control group 166 183

All run-off-road 217 154 36 8.62

Side impact with fixed object 41 23 49 5.89

Rollover 36 12 70 12.71

Other run-off-road 140 119 23 2.50

Pedestrian/bicycle/animal 46 69 -36 1.99

Other single vehicle 6 3 55 1.28

Culpable multi-vehicle 176 157 19 1.91

Other multi-vehicle*** 65 76

All non-control group involvements 510 459 18 2.64

All crash involvements 14

Total 676 642

Single-vehicle crashes** 223 157 36 9.04

* For the 2x2 table formed by this row and the “control group” row.

** excludes pedestrian, bicycle, animal crashes.

*** includes involvements that are not culpable but are not part of the control-group.

Fatal crashes in LTVs

For LTVs (the majority of the analysis sample consists of SUVs), ESC is highly effective

in reducing fatal run-off-road crashes (70 percent effectiveness, chi-square statistic

29.90), especially rollovers (88 percent) – Table 13. In this analysis, ESC has little

observed effect on crashes that involved pedestrians, bicycles, or animals. Culpable

involvements in fatal multi-vehicle crashes significantly declined (34 percent

effectiveness; chi-square statistic 3.64) in LTVs with ESC. The overall effectiveness in

all fatal crash involvements is a positive 28 percent – which indicates a substantial benefit

with ESC in reducing all fatal crashes in LTVs. Similar to the result from the 2004 study,

ESC is highly effective in reducing single-vehicle crashes involving LTVs – as shown in

Table 13 by the positive 63 percent effectiveness, and 23.32 chi-square statistic.

Table 13: Effect of ESC on Fatal Crashes in LTVs

(FARS: 1997-2004)

Type of fatal crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectiveness

*Chi-

square

Value

Fatal

Crashes

Fatal

Crashes

(%)

Control group 153 95

All run-off-road 191 36 70 29.90

Side impact with fixed object 15 6 36 0.78

Rollover 106 8 88 37.56

Other run-off-road 70 22 49 6.16

Pedestrian/bicycle/animal 56 37 -6 0.06

Other single vehicle 6 9 -142 2.78

Culpable multi-vehicle 108 44 34 3.64

Other multi-vehicle*** 69 40

All non-control group involvements 430 166 38 8.96

All crash involvements 28

Total 583 261

Single-vehicle crashes** 197 45 63 23.32

* For the 2x2 table formed by this row and the “control group” row.

** excludes pedestrian, bicycle, animal crashes.

*** includes involvements that are not culpable but are not part of the control-group.

Effectiveness confidence bounds in various fatal crash involvements

Now that we have discussed ESC effectiveness in various fatal crash involvements in

passenger cars and LTVs, let us discuss further the confidence bounds for certain

effectiveness estimates that are statistically significant. The entries in Table 11 will be

used to produce an approximate sense of sampling errors and confidence bounds. As

previously explained, the two numbers in the table, a and b, are counts of non-relevant

crash involvements (control group) while the other two numbers, c and d, represent

counts of relevant involvements in various crash modes – in vehicles that are equipped

with ESC versus earlier versions of similar make-models that are not. These numbers can

be considered independent Poisson variates.

As discussed, the effectiveness statistic in

various crashes

(

)

(

)

[

]

E ÷−=−= 1

is based sample odds ratio,

The sampling distribution of the odds ratio can be highly skewed even for moderately

large sample sizes. On the other hand, the log transform of the sample odds ratio, log

Agresti, A. (2002) Categorical Data Analysis, Wiley, New York, pp.16-25.

has a less skewed sampling distribution and thus, is more symmetric. For large sample

sizes, the distribution can be approximate to a normal distribution with a mean of log

and a standard deviation often referred to as asymptotic standard error and denoted by

ASE, of

⎥

⎦

⎤

⎢

⎣

⎡

⎟

⎠

⎞

⎜

⎝

⎛

⎟

⎠

⎞

⎜

⎝

⎛

⎟

⎠

⎞

⎜

⎝

⎛

⎟

⎠

⎞

⎜

⎝

⎛

dcba

ASE

1111

)

(log

Taking 1.645 standard deviations on either side of the log

yields approximate

confidence bounds (two-sided α = .10, i.e., 90 percent bounds):

)

(log645.1

log

θθ

ASE±

The confidence intervals can then be transformed back by using the exponential function

to form the confidence interval for the odds ratio and subtracting from 1 to calculate

effectiveness estimates.

Tables 14 and 15 show not only the estimates and chi-square statistics but also the

confidence bounds of the effects of ESC for the following crash involvements: all run-

off-road, rollover, culpable multi-vehicle, single-vehicle (excluding pedestrian, bicycle,

animal crashes), and all crash involvements – in passenger cars and LTVs, respectively.

In Table 14, the effectiveness statistic in all run-off-road crashes is 1 – 0.6438 = 0.3562,

which is based on a sample odds ratio (154/217) / (183/166) = 0.6438. The natural log of

the sample odds ratio equals log (0.64) = -0.4404 and its asymptotic standard error equals

[1/166 + 1/183 + 1/217 + 1/154]

= 0.1503. A 90 percent confidence interval for the

sample log odds ratio equals

),1503.0(645.14404.0

− or (-0.6877, -0.1932). The

corresponding confidence interval for the sample odds ratio is

[]

).8243.0,5027.0()1932.0exp(),6877.0exp(

−

Consequently, the confidence bounds for the effectiveness estimate is

[]

−

− 100*)5027.01(,100*)8243.01( (18 percent, 50 percent).

Using the same approach, 69 percent effectiveness in rollovers yields approximate

confidence bounds (46 to 83 percent fatality reduction). A non-significant 36 percent

increase in crashes involving pedestrians, bicycles, or animals suggests that the

effectiveness of ESC is between -95 and 5 percent. The reduction estimates are between

-4 and 37 percent for all culpable involvements in fatal multi-vehicle crashes. For all

fatal crash involvements in passenger cars, the confidence bounds for the reduction (14

percent) are estimated to be 0 to 26 percent. For a 36 percent reduction in single-vehicle

crashes involving passenger cars (excluding pedestrian, bicycle, animal crashes), the

confidence bounds are between 18 and 50 percent.

Due to higher chi-square statistics for the effectiveness estimates (smaller standard

deviations) in certain fatal crash involvements in LTVs (specifically in all run-off-roads

and rollovers), the confidence bounds are tighter in LTVs than in passenger cars. For

instance, in all fatal run-off-roads, the confidence bounds for the fatality reduction are 56

to 80 percent in LTVs as compared to 18 to 50 percent in passenger cars. Similarly, the

effect of ESC in reducing fatal rollover crashes is so large that the confidence bounds

range from 77 to 94 percent. Since the observed effect of ESC in reducing crashes

involving pedestrians, bicycles, or animals crashes is negligible and non-significant, the

confidence interval is between -60 and 29 percent. The reduction in culpable fatal multi-

vehicle crash involvements in LTVs, on the other hand, is not as large as the reductions in

various run-off-road involvements; thus, the confidence bounds are between 6 and 54

percent. For all types of fatal crashes in LTVs, the 28 percent effectiveness yields

approximate confidence bounds, 13 to 41 percent fatality reduction. Last but not least,

the reduction in single-vehicle crashes (not including pedestrian, bicycle, animal crashes)

for LTVs is expected to be somewhere between 48 and 74 percent.

Table 14: Effectiveness Confidence Bounds

in Certain Fatal Crash Involvements in Passenger Cars

Type of fatal crash

involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Confidence

Bounds

Chi-

square

Value

Fatal

Crashes

Fatal

Crashes

(%) (%)

Control group 166 183

All run-off-road 217 154 36 18 to 50 8.62

Rollover 36 12 70 46 to 83 12.71

Pedestrian/bicycle/animal 46 69 -36 -95 to 5 1.99

Culpable multi-vehicle 176 157 19 -4 to 37 1.91

All crash involvements 14 0 to 26

Single vehicle crashes* 223 157 36 18 to 50 9.04

* excludes pedestrian, bicycle, and animal crashes.

Table 15: Effectiveness Confidence Bounds

in Certain Fatal Crash Involvements in LTVs

Type of fatal crash

involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Confidence

Bounds

Chi-

square

Value

Fatal

Crashes

Fatal

Crashes

(%) (%)

Control group 153 95

All run-off-road 191 36 70 56 to 79 29.90

Rollover 106 8 88 77 to 94 37.56

Pedestrian/bicycle/animal 56 37 -6 -60 to 29 0

Culpable multi-vehicle 108 44 34 6 to 54 3.64

All crash involvements 28 13 to 41

Single vehicle crashes* 197 45 63 48 to 74 23.32

* excludes pedestrian, bicycle, and animal crashes.

Thus far, we have seen that ESC is highly effective in reducing fatal crashes – not only in

passenger cars but also in LTVs. Let us now look at the results from the State analyses to

see if these effectiveness rates still hold for the State data analyzed. The State data

samples include fatal and non-fatal crash involvements although the vast majority of the

involvements are non-fatal (i.e., property damage, possible injury, non-incapacitating and

incapacitating injury). With the State data, separate analyses were performed in each of

the selected States. Let us start with the analysis results in California.

Crashes in California

(Passenger cars)

As previously explained, the California data sample only includes Mercedes-Benz

vehicles, because we could not identify other vehicles due to lack of VIN information.

As shown in Table 16, run-off-road crashes in California were reduced by 66 percent in

passenger cars equipped with ESC, and the reduction (relative to the control group) is

statistically significant with a chi-square value of 125.22. Rollover crashes alone were

significantly decreased (76 percent effectiveness; chi-square statistic 15.54). Side

impacts with fixed objects were included in “other run-off-road” involvements because

we could not identify those crashes based on the available information coded in the data

files. Crashes that involved pedestrians, bicycles, or animals in California were

decreased by 73 percent, and the decrease is statistically significant. The reduction (22

percent) of culpable involvements in multi-vehicle crashes is also statistically significant

with chi-square statistic 19.32. For all non-control group involvements, the decrease is

32 percent, and it is statistically significant (a chi-square value of 61.99). When

considering all crash involvements in passenger cars in California, the result showed that

the benefit is quite strong for ESC – 15 percent.

Table 16: Effect of ESC on Crashes in Mercedes-Benz Passenger Cars

(California: 2001-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 1688 2510

All run-off-road 331 167 66 125.22

Rollover 25 9 76 15.54

Other run-off-road 306 158 65 113.08

Pedestrian/bicycle/animal 10 4 73 5.65

Other single vehicle 48 60 16 0.78

Culpable multi-vehicle 874 1017 22 19.32

Other multi-vehicle* 151 183

All non-control group involvements 1414 1431 32 61.99

All crash involvements 15

Total 3102 3941

*includes involvements that are not culpable but are not part of the control-group.

(LTVs – mostly SUVs)

As shown in Table 17, ESC is even more effective (81 percent) in reducing run-off-road

crashes in LTVs than in passenger cars, and the reduction is statistically significant.

Particularly, LTVs equipped with ESC decreased the risk of rolling over by 78 percent.

Likewise, those vehicles also reduced the risk of hitting a pedestrian, bicycle, or an

animal by 26 percent although the effect is not statistically significant. Culpable

involvements of LTVs in multi-vehicle crashes in California were reduced by nearly one-

fourth (24 percent). The reduction was even more significant in all culpable

involvements as evidenced by the 40 percent effectiveness and 9.66 chi-square value. In

general, ESC is equally as effective in reducing all crash involvements (16 percent) in

LTVs as in passenger cars.

Table 17: Effect of ESC on Crashes in Mercedes-Benz LTVs

(California: 2001-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 106 575

All run-off-road 20 21 81 29.62

Rollover 5 6 78 7.18

Other run-off-road 15 15 82 24.13

Pedestrian/bicycle/animal 1 4 26 0.07

Other single vehicle 6 15 54 2.57

Culpable multi-vehicle 57 235 24 2.29

Other multi-vehicle* 8 48

All non-control group involvements 92 323 35 7.60

All crash involvements 16

Total 198 898

*includes involvements that are not culpable but are not part of the control-group.

Crashes in Florida

(Passenger cars)

All run-off-road crashes in Florida were decreased by 28 percent for passenger cars –

Table 18. The risk of a rollover or a side impact with fixed object was also lowered in

vehicles with a factory installed ESC. In fact, the relative decreases were 14 percent and

43 percent for side impacts with fixed objects and rollovers, respectively. Furthermore,

the effects are statistically significant for all run-off-roads and rollovers. Collisions with

pedestrians, bicycles, or animals and culpable involvements in multi-vehicle crashes also

benefited from ESC as indicated in Table 18 with statistically significant 49 percent and

13 percent effectiveness, respectively. All culpable crashes decreased by 13 percent, and

the decrease is statistically significant. On the whole, ESC is effective in reducing all

crash involvements by 5 percent.

(LTVs – mostly SUVs)

Table 19 shows that ESC reduced all run-off-road crashes in LTVs by 66 percent with

chi-square statistic 50.08. The reduction is equally substantial in rollovers (78 percent

effectiveness and a chi-square value of 16.83). Crashes that involved a pedestrian,

bicycle, or an animal were also decreased (20 percent effectiveness), but the relative

effect is not statistically significant. On the other hand, the effect of ESC in reducing all

culpable involvements in multi-vehicle crashes is statistically significant as evidenced by

the 10 percent effectiveness and a chi-square value of 3.44. Similarly, ESC is quite

effective in reducing all non-control group crashes (a 13 percent reduction and chi-square

statistic of 8.74). The overall effect of ESC in reducing all crash involvements in Florida

is slightly higher for LTVs – 6 percent reduction.

Table 18

: Effect of ESC on Crashes in Passenger Cars

(Florida: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 5104 5441

All run-off-road 422 324 28 18.59

Side impact with fixed object 63 58 14 0.64

Rollover 33 20 43 4.06

Other run-off-road 326 246 29 16.03

Pedestrian/bicycle/animal 82 45 49 13.13

Other single vehicle 401 471 -10 1.88

Culpable multi-vehicle 2259 2105 13 13.96

Other multi-vehicle* 906 856

All non-control group involvements 4070 3801 12 19.71

All crash involvements 5

Total 9174 9242

* includes involvements that are not culpable but are not part of the control-group.

Table 19: Effect of ESC on Crashes in LTVs

(Florida: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 2993 1655

All run-off-road 257 48 66 50.08

Side impact with fixed object 36 13 35 1.74

Rollover 57 7 78 16.83

Other run-off-road 164 28 69 35.93

Pedestrian/bicycle/animal 34 15 20 0.53

Other single vehicle 298 135 18 3.41

Culpable multi-vehicle 1227 609 10 3.44

Other multi-vehicle* 517 314

All non-control group involvements 2333 1121 13 8.74

All crash involvements 6

Total 5326 2776

* includes involvements that are not culpable but are not part of the control-group.

Crashes in Illinois

(Passenger cars)

Statistically significant reductions in all run-off-road crashes (53 percent), especially in

rollovers (80 percent) were also found in Illinois for passenger cars equipped with ESC –

Table 20. The results also showed significant declines not only in pedestrian, bicycle,

and animal crashes (26 percent) but also in crash involvements where the vehicle was

considered the culpable party in a multi-vehicle crash (16 percent). All non-control

group involvements were reduced (13 percent) as well. When we looked at all crash

involvements in Illinois for passenger cars, we found that ESC, by and large, is still fairly

effective – 7 percent reduction.

Table 20: Effect of ESC on Crashes in Passenger Cars

(Illinois: 1997-2002)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectivene

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 3548 2636

All run-off-road 420 146 53 60.63

Rollover 27 4 80 11.15

Other run-off-road 393 142 51 25.53

Pedestrian/bicycle/animal 247 136 26 7.49

Other single vehicle 662 528 -7 1.24

Culpable multi-vehicle 2002 1253 16 17.57

Other multi-vehicle* 1244 899

All non-control group involvements 4575 2962 13 15.56

All crash involvements 7

Total 8123 5598

* includes involvements that are not culpable but are not part of the control-group.

(LTVs – mostly SUVs)

Although the reduction in rollovers in Illinois – which is credited to ESC – was only a

few percentage points higher (87 percent versus 80 percent) in LTVs than in passenger

cars, the decrease in all run-off-road crashes was substantially higher in LTVs than in

passenger cars (80 percent versus 53 percent, respectively) – Tables 20-21. The analysis

results also revealed potential increase (-20 percent) in pedestrian, bicycle, and animal

crashes, but such increase (if any) is not statistically significant since the chi-square

statistic (0.88) does not exceed 3.84. Culpable involvements in multi-vehicle crashes

were also improved (due to ESC) by 28 percentage points, and the improvement is

statistically significant. All non-control group crash involvements were reduced by 22

percent as well, and the reduction is also statistically significant – Table 21. Once again,

ESC is quite successful in reducing all crash involvements (13 percent) for LTVs.

Table 21: Effect of ESC on Crashes in LTVs

(Illinois: 1997-2002)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 1252 959

All run-off-road 147 22 80 59.71

Rollover 41 4 87 21.44

Other run-off-road 106 18 78 40.18

Pedestrian/bicycle/animal 60 55 -20 0.88

Other single vehicle 217 165 1 0.00

Culpable multi-vehicle 756 416 28 19.71

Other multi-vehicle* 484 332

All non-control group involvements 1664 990 22 18.52

All crash involvements 13

Total 2916 1949

* includes involvements that are not culpable but are not part of the control-group.

Crashes in Kentucky

(Passenger Cars)

As for Kentucky, Table 22 shows a significant 47 percent reduction in all run-off-road

crash involvements of passenger cars equipped with ESC. Of that, the decrease in

rollovers is 73 percent, although the effectiveness is statistically not significant (1.72 chi-

square statistic) due to limited number of involvements in our data sample (i.e., 7 crashes

total; 6 pre-ESC crashes and 1 post-ESC crash). The effectiveness in passenger car

involvements with pedestrians, bicycles, or animals in Kentucky is negative (-4 percent)

and non-significant (chi-square statistic of 0.02). As for culpable involvements of

passenger cars in multi-vehicle crashes in Kentucky, the effect is positive but non-

significant as well. Likewise, the effect of ESC on non-control group involvements is

positive (10 percent). As a whole, ESC still has a positive influence on all crash

involvements in Kentucky for passenger cars.

Table 22: Effect of ESC on Crashes in Passenger Cars

(Kentucky: 1997-2002)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 771 482

All run-off-road 117 39 47 10.80

Rollover 6 1 73 1.72

Other run-off-road 111 38 45 9.59

Pedestrian/bicycle/animal 37 24 -4 0.02

Other single vehicle 30 19 -1 0.00

Culpable multi-vehicle 409 231 10 1.02

Other multi-vehicle* 165 112

All non-control group involvements 758 425 10 1.68

All crash involvements 5

Total 1529 907

* includes involvements that are not culpable but are not part of the control-group.

(LTVs – mostly SUVs)

As for LTVs, ESC is highly effective in reducing all-run-off road crashes, especially

rollovers – Table 23. The reductions are statistically significant with the effectiveness

equals to 77 percent for all run-off-roads and 92 percent for rollovers. On the other hand,

crashes that involved pedestrians, bicycles, or animals, were increased by 21 percent, but

the increase is once again statistically not significant. Although the effectiveness in

culpable involvements in multi-vehicle crashes is also not significant, it is nevertheless a

strong positive effect (17 percent). The effect of ESC on all non-control group crashes is

also positive (17 percent). All in all, ESC is very effective in reducing all LTV crash

involvements in Kentucky as shown by the 9 percent reduction.

Table 23: Effect of ESC on Crashes in LTVs

(Kentucky: 1997-2002)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectivene

Chi-

square

value

All

crashes

All

crashes

(%)

Control group 463 208

All run-off-road 95 10 77 20.73

Rollover 28 1 92 10.08

Other run-off-road 67 9 70 12.15

Pedestrian/bicycle/animal 22 12 -21 0.28

Other single vehicle 16 6 17 0.14

Culpable multi-vehicle 257 96 17 1.60

Other multi-vehicle* 106 62

All non-control group involvements 496 186 17 2.27

All crash involvements 9

Total 959 394

* includes involvements that are not culpable but are not part of the control-group.

Crashes in Missouri

(Passenger Cars)

All run-off-road crash involvements in Missouri were also decreased (44 percent) for

passenger cars equipped with ESC – Table 24. Similarly, crash involvements with

pedestrians, bicycles, or animals as well as culpable involvements in multi-vehicle

crashes and all non-control group crashes were reduced by 48 percent, 8 percent, and 17

percent, respectively. Except for the reduction in culpable involvements in multi-vehicle

crashes, all other reductions are statistically significant. We found that ESC was 10

percent effective in reducing all passenger car crash involvements.

(LTVs – mostly in SUVs)

A continuing trend in crash reductions – specifically in all run-off-road crashes – is

observed for LTVs in Missouri – Table 25. The reduction is significant with 80 percent

effectiveness and chi-square statistic of 39.51. ESC also reduced crashes that involved

pedestrians, bicycles, or animals, but the reduction is statistically non-significant because

the chi-square statistic is close to 0. The results also suggested that ESC provides little

benefits (if any) to LTVs – particularly, in culpable involvements in multi-vehicle

crashes, but for all non-control group crash involvement, the benefits are significant (a 16

percent reduction). Overall, all crash involvements decreased by 9 percent.

Table 24: Effect of ESC on Crashes in Passenger Cars

(Missouri: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 1141 915

All run-off-road 232 105 44 21.09

Side impact with fixed object 21 20 -19 0.30

Rollover 14 2 82 6.59

Other run-off-road 197 83 47 22.24

Pedestrian/bicycle/animal 58 24 48 7.43

Other single vehicle 161 98 24 4.15

Culpable multi-vehicle 757 560 8 1.28

Other multi-vehicle* 313 225

All non-control group involvements 1521 1012 17 9.65

All crash involvements 10

Total 2662 1927

* includes involvements that are not culpable but are not part of the control-group.

Table 25: Effect of ESC on Crashes in LTVs

(Missouri: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectivene

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 562 283

All run-off-road 150 15 80 39.51

Side impact with fixed object 13 0 100 6.50

Rollover 31 0 100 13.34

Other run-off-road 106 15 72 22.08

Pedestrian/bicycle/animal 27 12 12 0.12

Other single vehicle 71 36 -1 0.00

Culpable multi-vehicle 401 204 -1 0.01

Other multi-vehicle* 158 76

All non-control group involvements 807 343 16 3.04

All crash involvements 9

Total 1369 626

* includes involvements that are not culpable but are not part of the control-group.

Crashes in Pennsylvania

(Passenger cars)

For passenger cars in Pennsylvania, ESC has quite an impact on all run-off-road

involvements (40 percent effectiveness, 21.69 chi-square statistic), especially in rollovers

(66 percent effectiveness, 5.23) – as shown in Table 26. Although it is not statistically

significant, observed ESC effectiveness in pedestrian, bicycle, and animal crashes is

negative – as illustrated in Table 26 by the negative 23 percent effectiveness.

Pennsylvania is the first State whose analysis shows a negative ESC effectiveness in

culpable involvements in multi-vehicle crashes, but the effect is statistically non-

significant. Furthermore, ESC has little effect in reducing all crash involvements (1

percent) in Pennsylvania for passenger cars.

Table 26: Effect of ESC on Crashes in Passenger Cars

(Pennsylvania: 1997-2001, 2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 983 553

All run-off-road 431 146 40 21.69

Side impact with fixed object 53 18 40 3.36

Rollover 26 5 66 5.23

Other run-off-road 352 123 38 16.61

Pedestrian/bicycle/animal 62 43 -23 1.04

Other single vehicle 31 24 -38 1.34

Culpable multi-vehicle 731 454 -10 1.53

Other multi-vehicle* 282 183

All non-control group involvements 1537 850 2 0.06

All crash involvements 1

Total 2520 1403

* includes involvements that are not culpable but are not part of the control-group.

(LTVs – mostly SUVs)

Table 27 shows that ESC significantly reduced all run-off-roads in Pennsylvania by 63

percent for LTVs and was even more effective in reducing rollovers – as evidenced by

the 86 percent effectiveness and 18.26 chi-square statistic. We see possible harm with

ESC for crash involvements with pedestrians, bicycles, or animals; however, the effect is

not statistically significant. This analysis showed that ESC has negligible effect on

culpable involvements of LTVs in multi-vehicle crashes in Pennsylvania, but it has quite

an effect on all non-control group involvements (22 percent) as well as on crash

involvements (14 percent). These effects are statistically significant.

Table 27: Effect of ESC on Crashes in LTVs

(Pennsylvania: 1997-2001, 2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 355 210

All run-off-road 193 42 63 28.64

Side impact with fixed object 19 8 29 0.63

Rollover 48 4 86 18.26

Other run-off-road 126 30 60 17.71

Pedestrian/bicycle/animal 19 19 -69 2.49

Other single vehicle 13 3 61 2.27

Culpable multi-vehicle 226 132 1 0.01

Other multi-vehicle* 129 70

All non-control group involvements 580 266 22 4.97

All crash involvements 14

Total 935 476

* includes involvements that are not culpable but are not part of the control-group.

Crashes in Wisconsin

(Passenger cars)

Similar to all run-off-road involvements of passenger cars in other States, the

involvements in Wisconsin were significantly decreased – 34 percent effectiveness and

9.58 chi-square statistic – Table 28. In addition to involvements with pedestrians,

bicycles, or animals, culpable involvements in multi-vehicle crashes as well as in all

non-control group crashes decreased – for passenger cars equipped with ESC versus

those not equipped with ESC – as shown by the 23 percent, 16 percent, and 17 percent,

respectively. The decreases are significant in the culpable involvements in multi-vehicle

crashes and in all non-control group crashes. ESC is effective overall in Wisconsin for

passenger cars – as demonstrated in Table 28 with an 11 percent reduction of all crash

involvements.

(LTVs – mostly SUVs)

Likewise, LTVs with ESC are much less at risk of being involved in run-off-road crashes

when compared to vehicles with no ESC – as shown in Table 29 by the 68 percent

effectiveness and 32.45 chi-square value. In addition to these single-vehicle crash

involvements, other culpable involvements in multi-vehicle crashes were also

significantly decreased by 27 percent – though the decrease is not as substantial as in the

single-vehicle crashes. ESC had negligible effect on crashes that involved pedestrians,

bicycles, or animals as evidenced by the negative 4 percent effectiveness and 0.02 chi-

square statistic. All in all, ESC is very effective in reducing all crash involvements in

Wisconsin – 23 percent for LTVs.

Table 28: Effect of ESC on Crashes in Passenger Cars

(Wisconsin: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-square

Value

All

Crashes

All

Crashes

(%)

Control group 705 516

All run-off-road 214 104 34 9.58

Rollover 13 4 58 2.42

Other run-off-road 201 100 32 8.19

Pedestrian/bicycle/animal 120 68 23 2.49

Other single vehicle 57 25 40 4.39

Culpable multi-vehicle 645 395 16 4.28

Other multi-vehicle* 178 147

All non-control group involvements 1214 739 17 6.14

All crash involvements 11

Total 1919 1255

* includes involvements that are not culpable but are not part of the control-group.

Table 29: Effect of ESC on Crashes in LTVs

(Wisconsin: 1997-2003)

Type of crash involvement

Vehicles

with no

ESC

Vehicles

with

ESC

Effectiveness

Chi-

square

Value

All

Crashes

All

Crashes

(%)

Control group 359 231

All run-off-road 162 33 68 32.45

Rollover 50 4 88 21.51

Other run-off-road 112 29 60 17.15

Pedestrian/bicycle/animal 42 28 -4 0.02

Other single vehicle 26 10 40 1.85

Culpable multi-vehicle 310 146 27 5.68

Other multi-vehicle* 128 63

All non-control group involvements 668 280 35 15.16

All crash involvements 23

Total 1027 511

* includes involvements that are not culpable but are not part of the control-group.

Crashes in all selected States

At this point, we have comprehensively discussed separate analysis results of the effect of

ESC in reducing relevant crash involvements of passenger cars and LTVs – relative to the

control-group involvements – in each selected State. Let us now try to assess the overall

effect of ESC across these seven States. The easiest way to obtain a single estimate

would be to pool the cases from the States and perform a single analysis. We are

reluctant to analyze the data that way because:

Different States have different crash-reporting thresholds. States with low thresholds

have more reported crash cases per capita, and would account for an unjustifiably large

share of the pooled data.

The make-model mix varies from State to State.

The distribution of crash types varies from State to State, as do the definitions of the

crash types.

Instead, we will use the weighted mean of the sample log odds ratios from the seven

States as the best indicator of the central tendency of the data, and the weighted standard

error of the log odds ratios as a basis for judging statistical significance and estimating

confidence bounds, which will be discussed in the next section. In this section, we will

concentrate only on the overall effects of the systems.

The reason for using the mean of the log odds ratios is due to the facts that it is (1) less

affected by extreme values, (2) useful as a measure of central tendency for certain

positively skewed distributions, and (3) an appropriate measure to use for averaging rates

(e.g., crash reduction/increase). The weighted mean is used to take into account the

difference in the sampling distributions among the seven States. The weighted value for

each State is simply the reciprocal of the sum of the reciprocals of the four frequencies

from the 2x2 contingency table. The weighted mean of the log odds ratios is determined

using the PROC MEANS procedure in SAS from the seven log odds ratios and the

weighted value for each of the seven States.

(Passenger cars)

We will first discuss the effects on passenger cars and then on LTVs. Table 30 shows not

only the individual percentage reductions in the ratios of relevant involvements to

control-group involvements – in passenger cars with ESC versus cars of similar

make-models without ESC – in each State, but also the overall reductions – which are

determined by computing the weighted mean of the sample log odds ratios. For example,

for all run-off-road crashes in California involving passenger cars (Table 16), the odds

ratio equals [(167/331) ÷ (2510/1688)] = 0.3393. The log odds ratio is log (0.3393) =

-1.0809. The weighted value equals

(

)

(

)

(

)

(

)

{

}

00.100

167

331

2510

1688

1 =+++÷

Charles Farmer recommended this approach in his peer review.

Using the same approach, the log odds ratios for Florida, Illinois, Kentucky, Missouri,

Pennsylvania, and Wisconsin, are -0.3282, -0.7595, -0.6289, -0.5720, -0.5073, and

-0.4095, respectively. The corresponding weighted values are 171.36, 101.10, 26.62,

63.28, 83.37, and 56.67. The computed weighted mean of the log odds ratios from SAS

program is -0.5969. The weighted mean of the overall effectiveness is estimated simply

by first taking the antilogarithm of -0.5969 and then subtracting [exp (-0.5969) = 0.5505]

from 1, which yields 45 percent effectiveness.

As illustrated in Table 30 in the gray-highlighted column, ESC is most effective in

reducing all run-off-road crashes by 45 percent, particularly in rollovers by 64 percent.

Other single-vehicle crash involvements – where ESC also had an impact – are those that

involved pedestrians, bicycles, or animals, even though the results are inconsistent across

States. Nevertheless, the impact is still positive: 26 percent effectiveness. The table

shows negative overall effectiveness (-2 percent) of ESC in reducing “other single-

vehicle” crashes (such as collisions with parked cars, impacts with thrown or falling

objects, undercarriage scrapes, or first-event fires), and the results are also inconsistent

among the States. Culpable involvements in multi-vehicle crashes received some benefit

from ESC as evidenced by the 13 percent effectiveness. When all crashes (including

culpable, non-culpable, and the control-group involvements) were taken into account,

ESC is still quite effective – as shown by the 8 percent effectiveness. In the State

analyses, we also analyzed the effectiveness of ESC on single-vehicle crashes that did not

involve pedestrians, bicycles, or animals. As shown in Table 30, the reductions of single-

vehicle crashes were observed in each of the States, and the overall reduction (which is

the weighted mean of the reductions in seven States) is 26 percent.

Table 30: The Effects and Mean of the Effects of ESC on Various Crash Involvements of Passenger Cars - By State

Type of crash

involvement

CA FL IL KY MO PA WI

*Mean of

States

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Control group

All run-off-road

66 28 53 47 44 40 34

Rollover

76 43 80 73 82 66 58

Other run-off-road

65 29 51 45 47 38 32

Pedestrian/bicycle/animal

73 49 26 -4 48 -23 23

Other single vehicle**

16 -10 -7 -1 24 -38 40

-2

Culpable multi-vehicle

22 13 16 10 8 -10 16

All crash involvements

15 5 7 5 10 1 11

Single vehicle crashes

(excluding pedestrian,

bike, animal crashes)

60 9 16 37 36 35 35

* indicates the weighted mean of the effects in all 7 States.

** includes collisions with parked vehicles.

(Light-Trucks – mostly SUVs)

ESC was even more effective in reducing run-off-road crashes (relative to the control-

group involvements) for LTVs, especially in rollover involvements – as shown in Table

31 by the 72 percent and 85 percent effectiveness, respectively. In calculating the

weighted mean effect for the rollover reductions in seven States, the crash involvements

in Illinois and neighboring Missouri were pooled because a 100 percent reduction in

Missouri yields an undefined log odds ratio. ESC had a negative effect on pedestrian,

bicycle, or animal crashes. In LTVs, culpable involvements in other single- and multi-

vehicle crashes were also decreased by 12 percent and 16 percent, respectively. Overall,

ESC was also effective in reducing all crash involvements in LTVs. The reduction in

single-vehicle crashes (excluding those involving pedestrians, bicycles, or animals)

involving LTVs is not as high as the reduction in all run-off-road crashes (48 percent

versus 72 percent.

Table 31: The Effects and Mean of the Effects of ESC on Various Crash Involvements in LTVs - By State

Type of crash

involvement

CA FL IL KY MO PA WI

*Mean of

States

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Control group

All run-off-road

81 66 80 77 80 63 68

Rollover**

78 78 87 92 100 86 88

Other run-off-road

82 69 78 70 72 60 60

Pedestrian/bicycle/animal

26 20 -20 -21 12 -69 -4

-11

Other single vehicle

54 18 1 17 -1 61 40

Culpable multi-vehicle

24 10 28 17 -1 1 27

All crash involvements

16 6 13 9 9 14 23

Single vehicle crashes

(excluding pedestrian,

bike, animal crashes)

74 40 33 68 54 63 64

* indicates the weighted mean of the effects in all 7 States.

** Data in IL and MO were pooled to allow computation of a log-odds ratio.

Effectiveness confidence bounds and statistical significance in various crash

involvements

Now, let us discuss confidence intervals for certain effectiveness estimates that are

considered significant crash reductions in passenger cars and LTVs – Tables 32 and 33.

In the State analysis results, we will calculate confidence bounds using weighted mean of

the log odds ratios, weighted standard errors based on the variation of the seven States,

and take 1.943 standard errors for two-sided

10.0

, i.e., 90 percent bounds, because

1.943 is the 95

percentile of the t-distribution with (7-1)=6 degrees of freedom. A

general notation for calculating the confidence interval for the overall effect is illustrated

below by the formula

Confidence interval = (overall effect)

oddsanofweightedme log

1,05. −N

(SE)

where

(overall effect)

oddsanofweightedme log

is determined using the approach discussed in the

previous section.

1,05. −N

is the t value for a two-sided test (90 percent bounds) and N-1 degrees of

freedom.

N is the number of individual effects.

SE is the weighted standard error of the log odds ratios obtained from the PROC

MEANS procedure in SAS (i.e., the weighted standard deviation of these ratios

divided by

7 , the square root of the number of States).

Let us take one example (all run-off-road crash involvements of passenger cars) and

discuss the statistics in detail. As discussed in the previous section, the weighted mean of

the log odds ratios is -0.5969. The weighted standard error of the log odds ratios

(computed from SAS) is 0.1068. Applying these statistics to the above t-confidence

interval equation produces

90% confidence interval = -0.5969

1.943 (0.1068) = -08044 to -0.3894

which translates to (0.4474, 0.6775) in odds ratio by taking the antilogarithm of the

confidence interval estimates (-0.8044, -0.3894). The confidence interval for the overall

effect of ESC on all run-off-road crashes involving passenger cars is (32%, 55%). Using

the same approach, we found the following confidence intervals for other critical

reductions in passenger cars: 50 to 75 percent in rollover crashes, 8 to 41 percent in

crashes that involved a pedestrian, bicycle, or an animal, 7 to 18 percent in culpable

involvements in multi-vehicle crashes, and 5 to 10 percent in all crashes. For single-

vehicle crash involvements (not including collisions with a pedestrian, bicycle, or an

animal), the expected effectiveness is between 10 and 40 percent. The overall

Handbook of Probability and Statistics with Tables, Second Edition, McGraw-Hill, Inc., USA, 1970,

pp. 244-245.

effectiveness estimates listed in Table 32 are statistically significant because the

confidence bounds are greater than 0.

For LTVs, the confidence bounds for the reductions (listed in Table 33) were also

calculated using the above method. Table 33 shows that the weighted mean of the

effectiveness of ESC in reducing all LTV run-off-crashes is 72 percent, and the 90

percent confidence interval is 65 to 77 percent. As previously discussed, because the

crash involvements in Illinois and neighboring Missouri were pooled for the analysis of

rollovers (Table 31), the number of individual effects (N) is reduced to 6, and the critical

t-value for 6-1 = 5 degrees of freedom is 2.015

. Hence, the confidence interval for the

rollover reduction is (79%, 90%). A negative 11 percent effectiveness in crashes

involving pedestrians, bicycles, or animals implies that the expected effectiveness

estimate is between -31 and 6 percent. The confidence bounds for the decrease in

culpable involvements in multi-vehicle crashes are 7-23 percent. The confidence bounds

for the reduction in all crashes in LTVs are between 6 and 14 percent. When we look at

all single-vehicle crashes except the pedestrian, bicycle, and animal crashes, the

confidence interval for the reduction is (35%, 58%). In general, the spread of the sample

distribution determines the range of the confidence interval. In other words, if the

effectiveness estimates are consistent across all studied States, then the spread of the

sample distribution is expected to be small, and hence, the range of the confidence

interval is expected to be small as well – as evidenced by the estimates and the

confidence bounds listed in Tables 32-33. Similar to the effects on passenger cars, the

weighted mean effects on LTVs for run-off-road, rollover, culpable multi-vehicle, and all

crash involvements are statistically significant because the confidence bounds are all

positive. The overall reduction of single-vehicle crashes excluding pedestrian, bicycle,

and animal crashes is also statistically significant. Only the overall effect on crashes that

involved a pedestrian, bicycle, or an animal is not statistically significant because the

estimates from the individual States are inconsistent.

Handbook of Probability and Statistics with Tables, Second Edition, McGraw-Hill, Inc., USA, 1970, p.

383.

Table 32: The Effects and Mean of the Effects of ESC of Various Crash Involvements of Passenger Cars - By State

Type of

crash

involvement

CA FL IL KY MO PA WI

*Mean of

States

Confidence

bounds

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

All run-off-

road

66 28 53 47 44 40 34

32 to 55

Rollover

43 80 73 82 66 58

50 to 75

Pedestrian/

bike/animal

73 49 26 -4 48 -23 23

8 to 41

Culpable

multi-vehicle

22 13 16 10 8 -10 16

7 to 18

All crash

involvements

15 5 7 5 10 1 11

5 to 10

Single-

vehicle

crashes**

60 9 16 37 36 35 35

10 to 40

* indicates the weighted mean of the effectiveness estimates of 7 States.

** excludes pedestrian, bicycle, animal crashes.

Notes:

1) Bold – weighted mean effect is statistically significant.

2) Plain type – weighted mean effect is not statistically significant.

Table 33: The Effects and Mean of the Effects of ESC on Various Crash Involvements in LTVs – By State

Type of

crash

involvement

CA FL IL KY MO PA WI

*Mean of

States

Confidence

bounds

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

Effectiveness

(%)

All run-off-

road

81 66 80 77 80 63 68

65 to 77

Rollover**

78 78 87 92 100 86 88

79 to 90

Pedestrian/

bike/animal

26 20 -20 -21 12 -69 -4 -11

-31 to 6

Culpable

multi-vehicle

24 10 28 17 -1 1 27

7 to 23

All crash

involvements

16 6 13 9 9 14 23

6 to 14

Single-

vehicle

crashes***

74 40 33 68 54 63 64

35 to 58

* indicates the weighted mean of the effectiveness estimates of 7 States.

** A combined effectiveness calculated from crash involvements in IL and MO as 1 State (Tables 21 and 25). E = {1-[((4+0) / (959+283))

((41+31) /

(1252+562))}=0.92.

Only 6 States (N=6) are included for this calculation because the data in IL and MO data were pooled.

*** excludes pedestrian, bicycle, animal crashes.

Notes:

1) Bold – weighted mean effect is statistically significant.

2) Plain type – weighted mean effect is not statistically significant.

LOGISTIC REGRESSION ANALYSIS

Logistic regression permits a refinement and check of the basic analyses. Logistic regression by

the GENMOD procedure

was used to estimate the effect of passenger car ESC on the

probability that a crash involvement was relevant (specifically the involvement of a vehicle

running off the road) as opposed to being non-relevant (i.e., the control-group involvement),

while controlling for other factors. Estimating the impact of ESC in reducing single-vehicle run-

off-road crashes could be confounded by factors related to the driver, vehicle, roadway, or other

circumstances or by a proportionately different make-model mix in the vehicles without ESC and

the vehicles with ESC. To make certain that the effect of ESC is estimated accurately, variables

were included in the logistic regression to control for those external factors, other than ESC, that

could possibly influence the ratio of run-off-road to control-group crash involvements.

Here is one example of how demographic factors (i.e., driver characteristics) could confound the

effect of ESC. Since ESC is more commonly installed on higher-priced vehicles, ESC-equipped

vehicles are more likely to be driven by a certain group of drivers – perhaps more affluent and

older – than by other segments of the driving population. Although it is unlikely that the

majority of the passenger cars used in our analysis sample were driven by 16-24 year-olds, we

included this group of drivers in our model because they could potentially confound the effect of

ESC in reducing run-off-road crashes. This driver-age category generally has a higher

percentage of drivers considered more aggressive in their driving than others, and thus, their

vehicles are at a higher risk of being involved in run-off-road crashes. Furthermore, based on

historical trends of drivers involved in crashes by sex and crash severity,

it is a common

stereotype and many researchers have attempted to test the hypothesis that – male drivers are

more likely to be involved in crashes because they are generally more aggressive in their driving

than female drivers. Thus, driver gender was included in the regression model as well.

Other external factors include whether or not the crash occurred: (1) at night, (2) on a rural

roadway, or (3) on a freeway. These factors could influence the proportion of run-off-road crash

involvements because a driver, in those environment or situations, is more likely to lose control

of a vehicle, and hence, ESC is more likely to be activated. The vehicle make-model group was

also included in the regression model. The age of the vehicle was initially considered, but we

did not include this factor in the final regression analysis, because its effect was not statistically

significant when (1) all independent variables mentioned above were included in the regression

model and even when (2) only variables for ESC and vehicle age were included.

The data points in the logistic regressions for passenger cars and LTVs are the FARS driver

fatality cases, each given a weight factor of 1. Cases with unreported age, gender, or roadway

function class (rural or urban) are excluded; so are the drivers younger than 14 years or older

than 97 years. The dependent variable ROR (for run-off-road) – a dichotomous variable with a

Allison, P.D. (1999) Logistic Regression Using the SAS System: Theory and Application, Cary, NC: SAS

Institute Inc., pp.82-83.

Traffic Safety Facts, 2004, NHTSA Technical Report No. DOT HS 809 919, Washington D.C., 2004, pp.19-20.

Charles Farmer, in his peer review, recommended analysis (2) as an additional way to check if vehicle age is a

factor. Claes Tingvall also emphasized the potential importance of vehicle age in the current analysis, and

especially in future analyses as the vehicles without ESC get older.

binomial distribution – equals 1 if the crash involvement was a single-vehicle run-off-road and 2

if it was non-relevant (control group). The key independent variable, ESCS indicates whether or

not the crash-involved vehicle had ESC as standard equipment: 1 if ESC was installed and 0 if it

was not. The other independent variables are:

Gender, expressed as DRVMALE (= 1 if the driver is male and 0 if the driver is female).

The driver’s age, expressed as D16_24 (= 1 if the driver is 16-24 years old and 0 otherwise).

Time of day, expressed as NITE (= 1 if 7:00 p.m. – 5:59 a.m., 0 if 6:00 a.m. – 6:59 p.m.)

Roadway function class (RURAL, FREEWAY). RURAL has the value 1 if the crash occurred

on a rural roadway, 0 if the crash occurred on an urban roadway. FREEWAY has the value 1 if

the crash occurred on an interstate, freeway, or expressway and 0 otherwise.

The vehicle make-model group, expressed as a categorical variable (CAR_MFG).

The regression for the analysis of passenger cars is based on 708 FARS fatality cases. The

likelihood-ratio (LR) statistics for TYPE 3 analysis

are shown in Table 34. TYPE 3 analysis

was used because of the inclusion of the CLASS variable CAR_MFG in the model. After we

controlled for the important quantifiable human parameters – driver’s age, gender – plus other

variables that are strongly associated with single-vehicle run-off-road crash involvements, the

effect of ESC on fatal run-off-road involvements (relative to the control-group involvements) in

passenger cars is statistically significant with the LR chi-square statistic being 18.39.

In fact,

the model predicted that ESC reduced fatal single-vehicle run-off-road crashes by 1 - exp(-

0.7838) = 54 percent because the regression coefficient estimate for ESCS is -0.7838. The

results from the 2x2 contingency table analysis (Table 12) showed a 36 percent reduction in all

fatal run-off-road crash involvements relative to the same control-group involvements used in the

regression analysis. Thus, if we compare those two results, we see that the 2x2 contingency

table approach may be more conservative than the regression analysis method. Hence, the

regression analysis gives us increased confidence that the effectiveness estimates obtained from

our 2x2 contingency table analyses are not biased in favor of ESC.

The regression results from the passenger car analysis supported our assumptions that: (1) male

drivers are at a higher risk of being involved in fatal run-off-road crashes than females, and (2)

the 16-24 year-old group of drivers is also at a higher risk of being involved in fatal run-off-road

crashes when compared to other groups of drivers. These factors are statistically significant – as

indicated by the LR statistics shown in Table 34 (chi-square values 35.45 for DRVMALE and

19.65 for D16_24). Other independent variables (specifically those related to the environment)

that also had statistically significant effects on fatal run-off-road crashes are those that indicate

whether or not the crash occurred at night (NITE) or on a rural roadway (RURAL) – as

evidenced by the LR chi-square statistics 46.13 and 19.23, respectively. The effect of another

factor FREEWAY – which indicates whether or not the crash occurred on a freeway, interstate,

or expressway – is borderline-significant (chi-square = 6.46; Pr>chisq = 0.0111). The vehicle

make-model group (CAR_MFG) is, however, statistically significant – as shown by the 23.14

LR chi-square statistic and 0.0008 Pr>chisq.

Allison, P.D. (1999) Logistic Regression Using the SAS System: Theory and Application, Cary, NC: SAS

Institute Inc., p. 27.

Results are statistically significant at the 0.05 level when the LR chi-square statistic is greater than 3.84.

Table 34: Likelihood-ratio Statistics for TYPE 3 Analysis of the Independent Variables

(FARS: 1997-2004) - Passenger Cars

Source

Degrees of

Freedom

Chi-Square Pr > ChiSq

ESCS 1 18.39 <.0001

DRVMALE 1 35.45 <.0001

D16_24 1 19.65 <.0001

NITE 1 46.13 <.0001

RURAL 1 19.23 <.0001

FREEWAY 1 6.46 0.0111

CAR_MFG 6 23.14 0.0008

For the analysis of LTVs, the regression is based on 471 FARS fatality cases. In this analysis,

the model predicted that ESC reduced fatal single-vehicle run-off-road crashes involving LTVs

by 68 percent because the regression coefficient estimate for ESCS is

-1.1320. Similar to the effects of ESC and other independent variables in the passenger car

regression analysis, the effects of those variables in the LTV analysis (shown in Table 35) are

also statistically significant except for the effect of the vehicle make-model group. Again, the

effectiveness estimate obtained from the regression analysis is similar to the observed estimate

from the 2x2 contingency table analysis – 68 percent versus 70 percent, respectively (Table 13).

Table 35: Likelihood-ratio Statistics for TYPE 3 Analysis of the Independent Variables

(FARS: 1997-2004) - LTVs

Source

Degrees of

Freedom

Chi-Square Pr > ChiSq

ESCS 1 21.31 <.0001

DRVMALE 1 11.89 0.0006

D16_24 1 10.75 0.0010

NITE 1 7.51 0.0061

RURAL 1 18.37 <.0001

FREEWAY 1 21.83 <.0001

CAR_MFG 2 1.19 0.5511

COMPARISON OF 2-CHANNEL AND 4-CHANNEL ESC SYSTEMS

We also compared crash reduction with 2-channel versus 4-channel ESC systems for reasons

already explained in the “background” section. The 4-channel ESC systems have a control

algorithm and means to apply all four brakes individually

whereas the 2-channel systems do not.

In other words, the 2-channel systems are capable of applying brake torque only to the two front

wheels. Hence, the latter systems only have the capability to correct oversteer whereas the

former systems are capable of correcting not only oversteer but also understeer.

Table 35 shows passenger car fatalities (single vehicle run-off-road versus control-group

involvements) in vehicles with and without ESC – by system type (2-channel versus 4-channel

systems). We subdivided our FARS data into two groups of make-models: those equipped with

2-channel ESC systems (all GM models selected for the study except the Corvette) and those

equipped with 4-channel ESC systems (all non-GM models selected for the study plus the

Corvette). The make-models used in this analysis (2- versus 4-channel ESC) were identical to

those listed in Tables 2 through 7. There were no make-models that had received both types.

The crash reduction with 2-channel ESC is compared to cars of the same make-models, but in

the model years before they received ESC – and likewise to the make-models that received 4-

channel ESC.

In the 2-channel systems (Category 1), there were 56 fatalities in single-vehicle run-off-road

crashes and 56 fatalities in non-relevant crashes (control group) – in vehicles without ESC, a risk

ratio of 1.000. In vehicles with ESC, there were 16 run-off-road and 41 non-relevant fatalities, a

risk ratio of 0.390. That is a significant 61 percent reduction in single-vehicle run-off-road

fatalities in 2-channel systems relative to the control-group. In the 4-channel systems (Category

2), single-vehicle run-off-road fatalities also decreased, but the decrease is 39 percent and

statistically significant. Additional analyses were conducted to refine and properly characterized

these results, as discussed below.

We can test if the fatality reduction is significantly greater in category 1 (2-channel) than in

category 2 (4-channel) by performing a three-dimensional contingency table analysis. The

difference in effectiveness is not statistically significant, as evidenced by the chi-square of the

three-way interaction term when the CATMOD procedure of SAS is applied to the three-way

table.

In fact, the chi-square statistic for the difference in the reduction in fatal single-vehicle

run-off-road crashes between the two ESC systems is 1.35, which suggests that the larger

observed fatality reduction with 2-channel systems (shown in Table 35) is not significantly

different from the observed reduction with the 4-channel system. The reductions in fatal single-

vehicle run-off-road fatalities should be interpreted with caution because of the small number of

fatalities in both samples (2-channel and 4-channel systems) – Table 35. As we have found from

the FARS and State data analyses (discussed in detail in the previous sections), the single-vehicle

run-off-road fatality reduction (due to ESC) for passenger cars is 36 percent (Table 12), and the

run-off-road crash reduction (mostly non-fatal crashes) is 45 percent (Table 30). Based on these

results, both fatal and non-fatal single-vehicle run-off-road reductions are expected to be

somewhere in this range – whether the reduction is associated with 2-channel or 4-channel

systems. Hence, a 61 percent reduction in fatal single-vehicle run-off-road involvements in

vehicles with 2-channel systems is probably an unrealistic estimate and most certainly influenced

by the small sample. Consequently, the difference in the reduction in fatal single-vehicle run-

off-road crashes between the two ESC systems is found to be statistically non-significant (chi-

square value of 1.35) – due to the small samples.

For a CATMOD analysis of crash data with statistically significant three-way terms see Morgan, C., The

Effectiveness of Retroreflective Tape on Heavy Trailers, NHTSA Technical Report No. DOT HS 809 222,

Washington, 2001, pp. 29-37, summarized in Kahane (2004), pp. 43-44; SAS/STAT

User’s Guide, Vol. 1, Version

6, 4

Ed., SAS Institute, Cary, NC, 1990.

Given that the N of FARS cases is insufficient for statistically meaningful results, we strongly

believe, in this instance, that findings based on State data are more relevant and should be given

greater weight.

Table 35: Effectiveness of ESC as the Percentage Reduction in Fatal Single Vehicle Run-Off-

Road Crash Involvements Relative to the Control-Group Involvements - by ESC System Type -

for Passenger Cars

All run-off-

road

involvements

Control-

group

involvements

Risk

Ratio

ESC

reduction

Category 1: 2-channel ESC systems

Vehicles not equipped with

2-channel

56 56 1.000

Vehicles equipped with 2-

channel

16 41 0.390

61%

Category 2: 4-channel ESC systems

Vehicles not equipped with

4-channel

165 105 1.571

Vehicles equipped with 4-

channel

139 144 0.965

39%

Tables 36 shows not only the reduction in all single-vehicle run-off-road crashes – which consist

of mostly non-fatal involvements – in each selected State, but also the weighted mean of the

reductions in all States (similar to the calculation method used in Tables 30 and 31) – in

passenger cars for 2-channel and 4-channel ESC systems. To test for the significance of the

difference in effectiveness in six States, a four-dimensional CATMOD analysis was used. In

theory, a four-dimensional CATMOD analysis involves aggregating data from all the States,

using all possible combinations of the four variables in the model. However, this approach could

create biased results due to: (1) different crash-reporting thresholds among the States, (2)

variation of make-model mix from State to State, as well as (3) variation in the distribution of

crash types from State to State. Thus, to adjust for those biases, a CATMOD analysis was

performed on 6 x 4 x 2 table of State by ESC system by crash involvement. The dichotomous

dependent variable is ROR with values (= 1 for single vehicle run-off-road involvements; = 2 for

the control-group involvements). The independent variables are STATE (a categorical variable

with six categories: 1, 2, 3, 4, 5, and 6 for six States – Florida, Illinois, Kentucky, Missouri,

Pennsylvania, and Wisconsin, respectively), FOUR_CH (a dichotomous variable with values = 1

for 4-channel systems; = 2 for 2-channel systems), and ESC (another dichotomous variable with

values = 1 for vehicles with ESC: 2-channel or 4-channel; = 2 for vehicles without ESC). The

model included the following terms: 1) ROR*FOUR_CH, 2) ROR*ESC, 3) ROR*STATE, 4)

ROR*FOUR_CH*STATE, 5) ROR*ESC*STATE, and 6) ROR*FOUR_CH*ESC.

The chi-

square statistic for the ROR*FOUR_CH*ESC term was 4.69, which is statistically significant at

the 0.05 level but not significant at the 0.025 level. In other words, the difference in

The four-way interaction term is not included because we do not anticipate that the effectiveness gap between 4-

channel and 2-channel systems would vary from State to State.

effectiveness between the 2-channel and 4-channel systems – specifically the larger observed

reductions in single-vehicle run-off-road involvements with 4-channel systems versus the smaller

observed reductions with 2-channel systems – are statistically significant in the State data.

Table 36: Effectiveness of 2-Channel versus 4-Channel ESC Systems as the Percentage

Reduction in All Single-Vehicle Run-Off-Road Crashes in Passenger Cars – By State Run-

Off-Road Crash Reduction (%) in

Florida

Illinois

Missouri Kentucky Pennsylvania Wisconsin *Mean of

the

reductions

in 6 States

With 2-

channel

ESC

systems

0 49 31 37 50 11 30

With 4-

channel

ESC

systems

31 57 48 50 41 47 43

* indicates the weighted mean of the reductions in 6 States.

Another approach that we used to compare 2-channel and 4-channel systems is to pool crash

involvements (specifically single-vehicle run-off-roads and non-relevant involvements) from the

States and estimate the significance of the difference in the overall effectiveness estimate of the

two systems. Table 37 shows the sum of crash involvements from six States and the

effectiveness estimates for the two systems. Again, the 4-channel systems are more effective in

reducing single-vehicle run-off-road crashes than the 2-channel systems as shown by the 48

percent and 33 percent effectiveness, respectively. When the CATMOD procedure is applied to

the three-way table, the chi-square of the three-way interaction term is 5.41, which is statistically

significant not only at the 0.05 level but also at the 0.025 level. Regardless of how we assess the

statistical significance of the difference in the effects of the two ESC systems in the State data

analysis, the results from two different methods of analysis (non-parametric and parametric

approach) consistently showed that the 4-channel systems reduced more single-vehicle crashes

than the 2-channel systems.

Table 37: Effectiveness of ESC as the Percentage Reduction in Single-Vehicle Run-Off-Road

Crash Involvements Relative to the Control-Group Involvements from Six States - by ESC

System Type - for Passenger Cars

All run-off-

road

involvements

Control-

group

involvements

Risk

Ratio

ESC

reduction

Category 1: 2-channel ESC systems

Vehicles not equipped with

2-channel

437 3304 0.132

Vehicles equipped with 2-

channel

145 1638 0.089

33%

Category 2: 4-channel ESC systems

Vehicles not equipped with

4-channel

1399 8948 0.156

Vehicles equipped with 4-

channel

719 8905 0.081

48%

CONCLUSIONS

Based on the analysis results from the 2x2 contingency tables, ESC appears to be extremely

successful in reducing not only fatal crashes but also other crash involvements. Tables 38 and 39

summarize the effectiveness of ESC – as the percentage reduction in the ratios of relevant crash

involvements to non-relevant involvements (i.e., the control-group) – in passenger cars and

LTVs with ESC versus those without ESC, for certain crash involvements that are considered

relevant.

In Table 38, ESC reduced all fatal single-vehicle run-off-road crash involvements by 36 percent

in passenger cars and 70 percent in LTVs. The table also includes effectiveness estimates of

fatal single-vehicle crashes excluding pedestrian, bicycle, animal crashes for passenger cars and

LTVs – 36 percent and 63 percent, respectively. These results are similar to those in the 2004

study – which showed that ESC was effective in reducing fatal single-vehicle crashes by 30

percent in passenger cars and 63 percent in SUVs. The control group used in the 2004 study

included all multi-vehicle crash involvements, whereas this study used non-relevant

involvements – where the crash-involved vehicle was stopped, parked, traveled at a speed of less

than 10 mph, or non-culpable party in a multi-vehicle crash on a dry road – as the control group.

As mentioned in the 2004 study, using multi-vehicle crashes as the control group, when it is

possible that multi-vehicle crashes are being reduced by ESC, actually means that the true

effectiveness of ESC could be higher than what we had estimated in the 2004 study for fatal

single-vehicle crashes. Based on the results from both studies, we may conclude with confidence

that the effectiveness still hold for all fatal single-vehicle crashes especially for all run-off-road

involvements in passenger cars and LTVs. Furthermore, the logistic regression analysis results

showed that ESC reduced fatal run-off-road involvements by 54 percent for passenger cars and

68 percent for LTVs, controlling for certain external factors that could confound the effects of

ESC. Initially, the vehicle “age” effect was included in the regression analysis, but the effect

was not statistically significant even when only independent variables for ESC and vehicle age

were considered and other factors were not. Thus, this factor was not included in the final

regression model. While the regression results showed that the age of vehicles had no significant

effect on crash involvements (specifically fatal run-off-road involvements), it is important that

this factor be evaluated in future analyses as ESC equipped vehicles get older and its influence

on crash involvements might become more significant.

Moreover, in this study, we found that rollover involvements in fatal crashes decreased by 70

percent in passenger cars and 88 percent in LTVs. Hence, ESC is very effective in reducing

fatal single-vehicle run-off-road crashes and extremely beneficial in rollover involvements. The

reductions are statistically significant. Furthermore, multi-vehicle crash involvements were also

reduced – in passenger cars and in LTVs with ESC – as shown by the positive effectiveness of

19 percent (although not statistically significant) and 34 percent, respectively – Table 38. The

observed effects of ESC on pedestrian, bicycle, and animal crashes are not statistically

significant, and they are inconsistent – as evidenced by the negative 36 percent and negative 6

percent effectiveness for passenger cars and LTVs, respectively. If we excluded pedestrian,

bicycles, and animal crashes, ESC is very successful in reducing single-vehicle crashes – 36

percent for passenger cars and 63 percent for LTVs. The results are similar to those found in the

2004 study. Even with the inconsistent results for ESC in crashes that involved pedestrians,

bicycles, or animals, ESC is still highly effective in reducing all fatal crash involvements

(including the control-group involvements as well as other non-relevant involvements) by 14

percent in passenger cars and 28 percent in LTVs.

Table 38: Effectiveness of ESC as the Percentage Reduction in Certain Fatal Relevant Crash

Involvements in Passenger Cars and LTVs

Types of Crash

Involvement

Passenger Cars LTVs

Single-vehicle crashes*

36% 63%

All run-off-road

36% 70%

Rollover

70% 88%

Pedestrian/bicycle/animal -36% -6%

Culpable multi-vehicle 19%

34%

All crashes 14%

28%

* excludes pedestrian, bicycle, animal crashes.

Notes:

1) Bold – effect is statistically significant.

2) Plain type – effect is not statistically significant.

Table 39 shows the weighted mean of the effects of ESC in all selected States (California,

Florida, Illinois, Kentucky, Missouri, Pennsylvania, and Wisconsin) on certain relevant crash

involvements (which consist of mostly non-fatal crashes) in passenger cars and LTVs. The

notations listed in the table below are similar to those in Table 38. Similar to the results in the

fatal crash analysis, ESC is highly effective in reducing all single-vehicle run-off-road crashes,

especially rollovers. In fact, the reductions are 45 percent in passenger cars and 72 percent in

LTVs for all run-off-road crashes – and 64 percent and 85 percent, respectively, for rollovers.

Also, the observed reductions in single-vehicle crashes (not including pedestrian, bicycle, animal

crashes) for passenger cars and LTVs are 26 percent and 48 percent, respectively, which are

comparable (to some extent) to the results from the 2004 study (35 percent for passenger cars

and 67 percent for SUVs). The results for ESC-equipped vehicles in crashes that involved a

pedestrian, bicycle, or animal are inconsistent – as shown by the 26 percent and negative 11

percent effectiveness in passenger cars and LTVs, respectively. This contrasts with the fatal

crashes, where the results were less favorable for cars. Culpable multi-vehicle crash

involvements decreased by 13 percent in passenger cars and 16 percent in LTVs, and the

decreases are statistically significant. When we looked at all crash involvements, we found

significant reductions with ESC – 8 percent reduction in passenger cars and 10 percent reduction

in LTVs.

Table 39: Effectiveness of ESC as the Percentage Reduction in Certain Relevant Crash

Involvemetns in Passenger Cars and LTVs

Types of Crash

Involvement

*Passenger Cars *LTVs

Single-vehicle crashes**

26% 48%

All run-off-road

45% 72%

Rollover

64% 85%

Pedestrian/bicycle/animal

26%

-11%

Culpable multi-vehicle

13% 16%

All crashes

8% 10%

* indicates the weighted mean of the effectiveness estimates of 7 States.

** excludes pedestrian, bicycle, animal crashes.

Notes:

1) Bold – mean effect is statistically significant.

2) Plain type – mean effect is not statistically significant.

All in all, ESC significantly reduced single-vehicle crashes especially run-off-road and rollover

involvements. In all likelihood, ESC may also be helpful in reducing culpable involvements in

multi-vehicle crashes. At this time we do not have enough data for statistically meaningful

results on the effect (if any) of ESC on pedestrian, bicycle, and animal crashes. Thus, we will

continue to monitor the effect on those crashes in the future.

Although there is a fairly large variation in ESC effectiveness (possibly because the estimates

somewhat depend on the choice of the control group) between different studies (as previously

discussed in the “Background” Section), ESC is still highly effective in all studies. ESC is likely

to have the largest effect on crashes involving severe injuries. Minor crashes are less likely to be

rollovers. Thus, including a large number of damage only crashes or slight injury crashes in this

study would likely reduce the effectiveness of ESC. In other words, the effectiveness of ESC

may even be higher than estimated if the large number of damage-only crashes are excluded in

the analysis. Although the sample used in this study is based on mostly luxury vehicles, we

believe that ESC would still be highly effective across the entire on-road fleet based on the

results from other studies (specifically those in Europe) where a large population of various

vehicle classes were used in analyzing the effectiveness of ESC.

REFERENCES

Aga, M. and Okada, A. (2003) Analysis of Vehicle Stability Control (VSC)’s Effectiveness from

Accident Data, Paper Number 541, Proceedings of the 18

International Technical Conference

on the Enhanced Safety of Vehicles.

Agresti, A. (2002) Categorical Data Analysis, Wiley, New York, pp.16-25.

Allison, P.D. (1999) Logistic Regression Using the SAS System: Theory and Application, Cary,

NC: SAS Institute Inc., pp.82-83.

Dang, J. (2004) Preliminary Results Analyzing the Effectiveness of Electronic Stability Control

(ESC) Systems, NHTSA Evaluation Note No. DOT HS 809 790, Washington, DC.

Farmer, C. (2004) Effect of Electronic Stability Control on Automobile Crash Risk, Traffic

Injury Prevention, Vol 5, pp. 317-325.

Green, P. and Woodrooffe, J. (2006) The Effectiveness of Electronic Stability Control on Motor

Vehicle Crash Prevention, Report Number UMTRI-2006-12, University of Michigan

Transportation Research Institute, Ann Arbor, MI.

Handbook of Probability and Statistics with Tables, Second Edition, McGraw-Hill, Inc., USA,

1970, pp. 244-245.

Lie, A., Tingvall, C., Krafft, M., and Kullgren, A. (2006) The Effectiveness of ESC (Electronic

Stability Control) in Reducing Real Life Crashes and Injuries, Traffic Injury Prevention, Vol 7,

pp. 38-43.

Lie, A., Tingvall, C., Krafft, M., and Kullgren, A. (2004) The Effectiveness of ESP (Electronic

Stability Program) in Reducing Real Life Accidents, Traffic Injury Prevention, Vol 5, pp. 37-41.

Morgan, C. (2001) The Effectiveness of Retroreflective Tape on Heavy Trailers, NHTSA

Technical Report No. DOT HS 809 222, Washington, DC, pp. 29-37.

SAS/STAT

User’s Guide, Vol. 1, Version 6, 4

Ed., SAS Institute, Cary, NC, 1990.

Traffic Safety Facts, 2004, NHTSA Technical Report No. DOT HS 809 919, Washington, DC,

pp.19-20.