Highway Safety: Causes of Injury in Automobile Crashes (Chapter Report,
05/09/95, GAO/PEMD-95-4).

Pursuant to a congressional request, GAO reviewed highway safety,
focusing on: (1) the most important predictors of injury in an
automobile crash; (2) how the risk of injury in a crash is affected by
the severity and type of crash, automobile size, safety belts and
airbags, and the occupants' age and gender; and (3) areas for further
reducing automobile occupants' crash injury risks.

GAO found that: (1) the most important determinants of driver injury in
car crashes are speed at impact, the type of crash, safety belt use,
driver age and gender, and automobile weight and size; (2) injury is
more likely in high-speed crashes, one car crashes, frontal crashes, and
rollovers; (3) occupants of heavier and larger cars are less likely to
be injured, but those cars pose a greater danger to persons in
multivehicle crashes; (4) heavier cars offer more protection in one-car
nonrollover and multivehicle crashes, but occupants of these cars are
subject to more injury in rollovers than are occupants of lighter cars;
(5) although safety belts reduce injury risks overall, they are most
effective in rollovers, single car crashes, and frontal crashes; (6) air
bags are only effective in frontal crashes and are less effective than
safety belts alone; (7) although they are involved in fewer crashes
overall, female and older drivers are more often injured than male and
younger drivers are in similar crashes; (8) safety belts are not as
effective for women as they are for men; (9) female and older drivers
are involved in more multivehicle crashes and male and younger drivers
are involved in more single car crashes; (10) older drivers tend to be
involved in more side impact crashes; and (11) the government and
manufacturers are working to improve automobile safety for each category
of driver.

--------------------------- Indexing Terms -----------------------------

 REPORTNUM:  PEMD-95-4
     TITLE:  Highway Safety: Causes of Injury in Automobile Crashes
      DATE:  05/09/95
   SUBJECT:  Highway safety
             Demographic data
             Motor vehicle safety
             Motor vehicle standards
             Comparative analysis
             Transportation statistics
             Traffic accidents
             Safety standards
             Human factors engineering
             Safety regulation
IDENTIFIER:  NHTSA National Accident Sampling System
             NHTSA Fatal Accident Reporting System
             
**************************************************************************
* This file contains an ASCII representation of the text of a GAO        *
* report.  Delineations within the text indicating chapter titles,       *
* headings, and bullets are preserved.  Major divisions and subdivisions *
* of the text, such as Chapters, Sections, and Appendixes, are           *
* identified by double and single lines.  The numbers on the right end   *
* of these lines indicate the position of each of the subsections in the *
* document outline.  These numbers do NOT correspond with the page       *
* numbers of the printed product.                                        *
*                                                                        *
* No attempt has been made to display graphic images, although figure    *
* captions are reproduced. Tables are included, but may not resemble     *
* those in the printed version.                                          *
*                                                                        *
* A printed copy of this report may be obtained from the GAO Document    *
* Distribution Facility by calling (202) 512-6000, by faxing your        *
* request to (301) 258-4066, or by writing to P.O. Box 6015,             *
* Gaithersburg, MD 20884-6015. We are unable to accept electronic orders *
* for printed documents at this time.                                    *
**************************************************************************


Cover
================================================================ COVER


Report to Congressional Requesters

May 1995

HIGHWAY SAFETY - CAUSES OF INJURY
IN AUTOMOBILE CRASHES

GAO/PEMD-95-4

Highway Safety:  Causes of Injury in Automobile Crashes


Abbreviations
=============================================================== ABBREV

  AIS - Abbreviated Injury Scale
  DOT - Department of Transportation
  FARS - Fatal Accident Reporting System
  GAO - U.S.  General Accounting Office
  IIHS - Insurance Institute for Highway Safety
  JAMA - Journal of the American Medical Association
  NASS - National Accident Sampling System--Crashworthiness Data
     System
  NHTSA - National Highway Traffic Safety Administration
  SUDAAN - Survey Data Analysis

Letter
=============================================================== LETTER


B-250039

May 9, 1995

The Honorable Ernest F.  Hollings
Ranking Minority Member
Committee on Commerce, Science,
 and Transportation
United States Senate

The Honorable Richard H.  Bryan
United States Senate

We are pleased to send you this report on the causes of injury in
automobile crashes in response to your request that we expand our
earlier research on automobile safety.  In this study, we reviewed
the research literature and conducted analyses of traffic accident
data to examine the effects of crash characteristics, automobile
size, safety belts, air bags, and occupant age and gender on the risk
of injury in automobile crashes.  The report also discusses promising
areas for further reducing injury risk. 

We will be sending copies of this report to the Secretary of
Transportation, and we will also make copies available to others upon
request.  If you have any questions or would like additional
information, please call me at (202) 512-3092.  Other major
contributors to this report are listed in appendix II. 

Kwai-Cheung Chan
Director for Program Evaluation
 in Physical Systems Areas


EXECUTIVE SUMMARY
============================================================ Chapter 0


   PURPOSE
---------------------------------------------------------- Chapter 0:1

In recent years, automobiles have been associated with nearly 29,000
traffic fatalities annually in the United States, including the
deaths of both automobile occupants and others involved in collisions
with automobiles.  The Senate Committee on Commerce, Science, and
Transportation asked GAO to study the independent effects of
important crash-related factors on the likelihood of injury once an
automobile crash has occurred.  This report addresses these
questions:  What are the most important predictors of injury in a
crash?  How is the risk of injury in a crash affected by the severity
and type of the crash, automobile size, safety belts and air bags,
and the occupant's age and gender?  What are promising areas for
reducing further the crash injury risk of automobile occupants? 


   BACKGROUND
---------------------------------------------------------- Chapter 0:2

In Highway Safety:  Have Automobile Weight Reductions Increased
Highway Fatalities?  (GAO/PEMD-92-1, October 1991), GAO presented a
number of findings regarding the relationship between automobile size
and safety.  GAO noted there that the safety consequences of
automobile weight or of any other automobile design factor are
confounded by many other factors, chief among them driver attributes. 
For example, automobiles that attract risky drivers may have high
fatality rates solely because they have dangerous drivers, not
because the cars themselves are unsafe. 

Three further GAO reports examine the safety effects of automobile
and driver characteristics within the larger context of other factors
that influence traffic safety.  One, Highway Safety:  Factors
Affecting Involvement in Vehicle Crashes (GAO/PEMD-95-3, October
1994), examines the influence of driver and automobile
characteristics on the likelihood of crash involvement.  Another,
Highway Safety:  Reliability and Validity of DOT Crash Tests
(GAO/PEMD-95-5, May 1995), looks at the extent to which results from
the automobile crash test programs conducted by the National Highway
Traffic Safety Administration (NHTSA) accurately predict injury risk
in crashes. 

This, the third report, focuses exclusively on the safety of
automobile occupants (in contrast to that of occupants of other
passenger vehicles, such as vans, minivans, and pick-up trucks).  To
study this issue, GAO reviewed technical reports from NHTSA and other
sources and consulted auto safety experts and representatives of
automobile manufacturers.  GAO also conducted its own statistical
analyses of automobile crash data obtained from NHTSA.  GAO's
statistical analyses looked at the experience of drivers of
relatively new cars in three types of crashes:  one-car rollover
crashes, one-car nonrollover crashes, and collisions with cars, vans,
pick-ups, and other light trucks. 


   RESULTS IN BRIEF
---------------------------------------------------------- Chapter 0:3

Not surprisingly, speed at impact and crash type are the most
important determinants of the risk of injury to drivers.  Driver age
and safety belt use are also important, while automobile weight and
the gender of the driver have less influence.  Injury is more likely
in high-speed crashes than in crashes with lower impact velocities,
and one-car crashes, particularly one-car rollovers, are more
dangerous than two-vehicle collisions.  Compared to light cars, heavy
cars both generally provide their occupants with more crash
protection and pose a greater danger to other roadway users in
multivehicle crashes.  However, the protective effects of automobile
weight differ by crash type.  Heavy cars offer comparatively more
protection to their occupants in one-car nonrollover and multivehicle
crashes, but once a rollover has occurred, occupants of heavier cars
are more likely to be hurt than are occupants of lighter cars. 

Safety belts greatly reduce injury risk, but the effectiveness of
safety belts is not the same in all crashes; they are more effective
in single-car crashes than in multivehicle collisions.  Air bags
reduce injury risk in frontal impacts, but air bags alone are less
effective than safety belts alone.  In equivalent automobile crashes,
women drivers and older drivers are more likely to be injured than
men and younger drivers.  GAO found evidence that safety belts are
less effective overall for women drivers than for men drivers. 


   PRINCIPAL FINDINGS
---------------------------------------------------------- Chapter 0:4


      CRASH SEVERITY AND CRASH
      TYPE ARE THE MOST IMPORTANT
      PREDICTORS OF INJURY
-------------------------------------------------------- Chapter 0:4.1

In GAO's statistical analysis, speed at impact was the most important
predictor of driver injury, followed by the type of crash, driver
age, safety belt use, automobile weight, and gender of the driver. 
The risk of driver hospitalization or death was 25 times greater in
very severe crashes (as measured by speed at impact) than in
relatively mild crashes.  The risk of driver injury was 9 times
greater in dangerous types of crashes (one-car rollovers) than in
relatively benign crashes (typical two-vehicle collisions).  The risk
of driver injury was 4.5 times greater for drivers age 65 and older
than for drivers 16 to 24 years of age, 3 times greater for unbelted
drivers than for drivers wearing manual lap and shoulder safety
belts, 63 percent greater for drivers of light cars than for drivers
of heavy automobiles, and 29 percent greater for women drivers than
for men drivers (see chapter 4). 


      THE EFFECTS OF AUTOMOBILE
      WEIGHT DIFFER BY CRASH TYPE
-------------------------------------------------------- Chapter 0:4.2

Considering all crash types together, GAO estimates that each
500-pound increase in car weight reduces the risk of driver injury by
14 percent in tow-away crashes.  However, the protective effect of
automobile weight differs by the type of crash.  Compared to light
cars, heavier cars offer more occupant protection in collisions with
cars and light trucks, and in one-car nonrollover crashes, but
drivers of heavier cars are more likely to be injured in one-car
rollover crashes, once a rollover has occurred.  One explanation for
this is that it takes more energy to roll over a heavier car than a
lighter one, meaning that rollover crashes involving heavy cars are
typically more severe than those involving light cars.  GAO estimates
that in multivehicle collisions, each 500-pound increase in
automobile weight decreases a driver's injury risk by 23 percent but
increases the probability of injury to the driver of the other car by
13 percent (see chapter 2). 


      THE EFFECTS OF SAFETY BELTS
      AND AIR BAGS DIFFER BY CRASH
      TYPE
-------------------------------------------------------- Chapter 0:4.3

Safety belts greatly reduce the risk of driver injury in all crashes,
but they are more effective in single-car crashes than in collisions
with cars and light trucks.  By comparison, air bags are effective
only in frontal impacts, not in side impacts or rollovers.  In
addition, air bags offer additional protection to drivers already
wearing lap and shoulder safety belts, reducing their risk of
suffering a serious injury by about 10 percent.  Safety belts alone
are much more effective than air bags alone-- that is, drivers
wearing safety belts in cars without air bags are much less likely to
be seriously injured than drivers not wearing belts in air
bag-equipped cars (see chapter 2). 


      WOMEN AND OLDER DRIVERS ARE
      MORE LIKELY TO BE INJURED
-------------------------------------------------------- Chapter 0:4.4

In equivalent crashes, women drivers and older drivers are more
likely to be injured then men drivers and younger drivers.  The
relative risk of injury for women drivers compared to men drivers
differs by the type of crash.  More specifically, GAO estimates that
women drivers have an injury risk approximately 50 percent greater
than men drivers in statistically equivalent two-vehicle collisions
but that the injury risk for women drivers and men drivers does not
differ in one-car crashes.  This is particularly important because
women drivers are involved in more multivehicle collisions than
one-car crashes, while men have more one-car than multivehicle
crashes.  Thus, women drivers are especially likely to be hurt in the
type of crash that they experience most often.  In addition, GAO
found evidence that heavy automobiles and safety belts offer somewhat
less protection for women drivers than for men drivers.  In contrast,
older drivers are more likely to be injured in a crash than younger
drivers in almost all circumstances (see chapter 3). 


      IMPROVING THE SAFETY OF
      AUTOMOBILE OCCUPANTS
-------------------------------------------------------- Chapter 0:4.5

Increasing safety belt use and effectiveness would reduce injury risk
for all automobile occupants.  It is particularly important to
increase safety belt use among drivers involved in serious crashes,
because they use safety belts less than other drivers. 

Women drivers and older drivers are in fewer crashes than men drivers
and younger drivers, and, because they have more multivehicle than
one-car crashes, the crashes that they are involved in are, on the
average, less severe.  Nonetheless, once a crash has occurred, women
and, especially, older drivers are more likely to be injured than men
and younger drivers.  For this reason, women and older drivers would
benefit substantially from improvements in automobile
crashworthiness.  Efforts to improve the protection of occupants of
all sizes, and to improve protection in side-impact crashes, would be
beneficial (see chapter 5). 


   RECOMMENDATIONS
---------------------------------------------------------- Chapter 0:5

This report contains no recommendations. 


   AGENCY COMMENTS
---------------------------------------------------------- Chapter 0:6

In written comments on a draft of this report, the Department of
Transportation (DOT) generally agreed with GAO's analytic methods and
findings.  In addition, DOT had several comments concerning the
methods and findings of particular analyses.  Those comments, and
GAO's responses to them, are presented at the appropriate places in
the report.  A number of DOT's specific technical comments have been
incorporated in the report where appropriate. 


INTRODUCTION
============================================================ Chapter 1


   PURPOSE
---------------------------------------------------------- Chapter 1:1

In recent years, automobiles have been associated with nearly 29,000
traffic fatalities annually in the United States, including the
deaths of both automobile occupants and others involved in collisions
with automobiles.\1 The National Highway Traffic Safety
Administration (NHTSA), a unit of the Department of Transportation
(DOT), has the lead role in federal government efforts to reduce the
number of traffic crashes and to minimize their consequences.  Some
types of automobiles have higher fatality rates than others.  (For
example, small cars generally have higher rates than large cars.) In
addition, some categories of drivers are more likely to be involved
in serious crashes than others.  (Young drivers, for example, have
higher involvement rates than other drivers.) However, as we
previously reported, one cannot conclude from differences in fatality
rates that some types of cars, or some types of drivers, are in fact
more dangerous than others, because driver and automobile
characteristics are highly related (GAO, 1991).  For example, since
small cars have a disproportionate percentage of young drivers, do
the high fatality rates for those cars stem from vehicle
characteristics or the recklessness with which they are operated? 

The goal of this report is to isolate the independent effects of
important crash-related factors on the likelihood of injury in a
collision.  The report focuses on the most important predictors of
occupant injury in a collision:  crash type and crash severity,
automobile size, safety belt use, and occupant age and gender.  The
report is concerned with both crashworthiness (protecting automobile
occupants) and aggressivity (protecting other roadway users struck by
automobiles).  The report also considers prospects for improving the
safety of automobile occupants. 

This report is one of three GAO reports examining automobile safety. 
One of these, Highway Safety:  Factors Affecting Involvement in
Vehicle Crashes (GAO, 1994), examines the independent effects of
driver characteristics and automobile size on crash involvement. 
Another, Highway Safety:  Reliability and Validity of DOT Crash Tests
(GAO/PEMD-95-5), looks at the extent to which results from the crash
test programs conducted by NHTSA accurately predict injury in actual
automobile crashes. 


--------------------
\1 We use the terms "automobile," "auto," and "car" interchangeably
in this report.  We define automobiles as convertibles, sedans,
hatchbacks, and station wagons.  This definition specifically
excludes other types of passenger vehicles, such as vans, minivans,
multipurpose vehicles, and pickup trucks. 


   BACKGROUND
---------------------------------------------------------- Chapter 1:2

Since the mid-1960's, both the number of traffic fatalities and the
fatality rate per registered vehicle have sharply decreased in the
United States.  The fatality rate for automobile occupants has
declined by 36 percent since 1975.  The continued emphases on
reducing drunk driving and increasing safety belt use, along with the
introduction of antilock brakes, air bags, and other safety enhancing
features, have increased the chances that this favorable trend will
continue.  Nonetheless, in 1991 automobiles were associated with
nearly 29,000 traffic deaths in the United States.\2 About 22,000 of
the automobile-related fatalities were automobile occupants (about
10,000 killed in single-car collisions and 12,000 in multiple-vehicle
collisions), about 2,500 were occupants of other types of vehicles
(for example, light trucks, vans, or motorcycles) involved in
collisions with automobiles, and approximately 4,000 were pedestrians
or cyclists hit by automobiles.  (See table 1.1.)



                                    Table 1.1
                     
                           Roadway Fatalities Involving
                                Automobiles, 1991


Crash           Automobile      Occupants of       Pedestrians
type\a           occupants    other vehicles      and cyclists             Total
--------  ----------------  ----------------  ----------------  ================
Single-              4,519                 0                17             4,536
 car
 rollove
 r
Single-              5,495                 0             3,529             9,024
 car
 nonroll
 over
Two cars             4,638                 0               170             4,808
Car and                 20               717                 6               743
 motorcy
 cle
Car and              3,466               955               124             4,545
 light
 truck
Car and              2,277                59                37             2,373
 heavy
 truck
Car and                 72                60                19               151
 other
 vehicle
Three or             1,744               686               112             2,542
 more
 vehicle
 s
================================================================================
Total               22,231             2,477             4,014            28,722
--------------------------------------------------------------------------------
\a The crash type categories are mutually exclusive.  We defined
automobiles as Fatal Accident Reporting System (FARS) body type codes
1-9 (including convertibles, two- and four-door sedans, three- and
five-door hatchbacks, and station wagons but not vans or minivans). 
Light trucks are FARS body type codes 10-39 (including auto-based
pickups, vans, minivans, and utility vehicles).  Heavy trucks are
codes 40-79 and 93; motorcycles are codes 80-82, 88, and 89; and
other vehicles are codes 90-92, 97, and 99 (including all-terrain
vehicles, snowmobiles, and farm equipment). 

Source:  Our analysis of 1991 FARS data from National Highway Traffic
Safety Administration, Fatal Accident Reporting System 1991,
DOT-HS-807-954 (Washington, D.C.:  1993). 

Different models of automobiles appear to make very different
contributions to this fatality toll.  For instance, the Insurance
Institute for Highway Safety (IIHS) has reported that the most
"dangerous" automobile models have occupant fatality rates more than
nine times higher than the "safest" models.\3 Further, some of the
automobile characteristics associated with this variation in fatality
rates are well known.  For instance, sports cars have higher fatality
rates than station wagons, and small cars have higher fatality rates
than large cars.\4

However, this does not mean that types of automobiles with high
fatality rates are necessarily more dangerous than those with low
fatality rates.  This is because different types of drivers prefer
particular types of automobiles, affecting both the number of
collisions involving particular autos and, perhaps, the probability
of serious injury in the event of a collision.  For example, young
drivers are much more likely to be involved in fatal accidents than
others--drivers age 16 to 20 are involved in fatal accidents at a
rate three times higher than that for drivers age 45 to 54--thereby
inflating the fatality rate for types of cars preferred by young
drivers.  Similarly, some types of automobile occupants are more
likely to be seriously injured in a collision than others.  For
example, in collisions in which at least one vehicle was towed from
the accident scene, NHTSA (1992a) recently estimated that women
automobile occupants are about 36 percent more likely to be hurt than
men occupants in similar collisions.  This suggests that types of
cars with a disproportionate number of women occupants may have
higher fatality rates than other cars. 


--------------------
\2 In 1991, the total number of U.S.  traffic fatalities was 41,462
(including deaths not related to automobiles). 

\3 For example, IIHS (1991) reported that the automobile model with
the highest fatality rate in the late 1980's was the Chevrolet
Corvette Coupe, with 4.7 deaths per 10,000 registered vehicles.  The
Volvo 240 had the lowest fatality rate for that period at 0.5 deaths
per 10,000 registered vehicles. 

\4 For example, according to the IIHS (1991) report of fatality rates
in the late 1980's, the fatality rate for midsize sports cars was 2.5
times that of midsize station wagons and vans.  For four-door cars,
the fatality rate for small cars was 40 percent greater than the rate
for large cars. 


      INJURY MECHANISMS
-------------------------------------------------------- Chapter 1:2.1

As automobiles abruptly stop or change direction in a collision,
occupants continue moving in the original direction of travel.  This
independent movement of an occupant within a rigid vehicle that is
decelerating more quickly than its occupant provides several
opportunities for injury.  First, some occupants are injured by being
ejected (either partially or totally) from the vehicle.  Ejection
substantially increases the risk of serious injury--ejected occupants
are three to four times more likely to be killed in a collision than
occupants who do not leave the vehicle.  Second, occupants can
collide with the interior of the vehicle or other objects intruding
into the passenger compartment.  This "second collision" (following
the "first collision" of the automobile striking an object) is
understandably worse if it occurs at high speed, involves impact with
a sharp or unyielding portion of the car's interior, or involves
contact with part of another vehicle or a roadside object that has
penetrated the passenger compartment.  Finally, because different
portions of an occupant's body decelerate at different rates,
internal injuries can be caused by the "third collision" of soft
tissues against hard, bony structures.  For example, in high-speed
collisions, the skull decelerates more quickly than the brain,
potentially causing injury to the brain as it strikes the hard skull. 


      AUTOMOBILE SAFETY DESIGN
-------------------------------------------------------- Chapter 1:2.2

Automobiles, and federal automobile safety regulations, are designed
to protect their occupants from these dangers in several ways.  One
way is to attempt to reduce the deceleration forces acting on
occupants.  Deceleration forces can be reduced by designing the
structure of a vehicle to absorb as much energy as possible before
the crash forces are transmitted to the passenger compartment or by
giving the occupant more time to slow down, thereby reducing the
maximum force level the occupant is subjected to.  The latter can be
accomplished by starting the deceleration period more quickly (for
example, by designing safety belts that begin holding back the
occupant sooner) or by increasing the total deceleration period (for
example, by lengthening the front end of the vehicle). 

Another way cars may protect their occupants is by encasing them in a
protective compartment that preserves a living space and prevents the
intrusion into the passenger compartment of striking vehicles or
other objects (such as light posts or trees) that a car hits.  Third,
automobiles are designed to keep their occupants both in the vehicle
and away from interior surfaces.  This is most obviously accomplished
through the use of safety belts, but a number of other components are
also intended to keep the occupant in the vehicle, including door
latches and windshields that are reinforced to eliminate potential
ejection routes.  In addition, automobile interiors are designed to
absorb energy from the occupant and to limit the occupant's movement
rather than serve as a rigid barrier.  Energy absorbing steering
columns are one example. 

There is little doubt that cars from recent model years, as a group,
are safer than automobiles from past model years and that some of
this improvement can be attributed to federal government safety
regulations.  For instance, in comparing the crash test results of
cars from model years 1980 and 1991, NHTSA researchers found that one
set of scores measuring injury potential had declined about 30
percent during the intervening years (Hackney, 1991).  Similarly,
Evans (1991b) estimated that the total effect of nine federal motor
vehicle safety standards enacted by 1989 had been to reduce the
occupant fatality rate by about 11 percent. 


   OBJECTIVE, SCOPE, AND
   METHODOLOGY
---------------------------------------------------------- Chapter 1:3


      OBJECTIVE
-------------------------------------------------------- Chapter 1:3.1

The objective of this report is to examine the independent effects of
a number of factors--crash type, crash severity, automobile weight
and size, safety belt use, and occupant age and gender--on the risk
of injury in an automobile crash.  We are concerned with both
crashworthiness and aggressivity.  Crashworthiness refers to the
extent to which automobiles protect their occupants in a collision,
and aggressivity refers to automobile characteristics that affect the
safety of the occupants of the other vehicles in a collision. 


      SCOPE
-------------------------------------------------------- Chapter 1:3.2

We restricted the scope of the study in several ways in order to
obtain a clear picture of the most important phenomena.  First, we
looked only at the safety of automobile occupants and the dangers
automobiles pose to occupants of other vehicles.  We were not
directly concerned with factors affecting the safety of occupants of
other types of passenger vehicles, such as pickup trucks, vans,
minivans, and multipurpose vehicles; we considered these vehicles
only as they affect the safety of automobile drivers in two-vehicle
collisions.  In our judgment, concerns about the safety of light
trucks and other passenger vehicles differ significantly from that of
automobiles.  Light trucks and other passenger vehicles are less
stable and thus roll over more frequently than automobiles, and they
have been subject to less stringent safety regulations than
automobiles. 

Second, our statistical analysis focused on model year 1987 and later
cars, because the safety experiences of those cars are more likely to
apply to today's new cars than are the safety experiences of older
ones.  Finally, we considered in our analysis only the injury
experiences of drivers, not those of automobile passengers or factors
specifically affecting the safety of child occupants.  (Roughly half
of the automobiles on the road have no occupants other than the
driver.)


      METHODOLOGY
-------------------------------------------------------- Chapter 1:3.3

To meet our objective, we reviewed technical reports from NHTSA and
other sources and consulted auto safety experts and representatives
of automobile manufacturers.  We also conducted our own statistical
analyses of traffic safety databases obtained from NHTSA.  Our
primary data set was compiled from the National Accident Sampling
System--Crashworthiness Data System (NASS) for 1988 through 1991. 
NASS is a nationally representative probability sample of all
police-reported crashes involving a passenger car, light truck, or
van in which at least one vehicle was towed from the scene.  In
addition, all the automobiles included in NASS were towed from the
crash site.  Thus, the automobiles in NASS, as a whole, are much more
likely to have injured occupants than are cars involved in typical
crashes.  Not only are police-reported crashes more severe than those
not reported to the police, but tow-away crashes on the whole are
also more severe than those not involving tow-aways.  Indeed, almost
all serious occupant injuries occur in police-reported tow-away
collisions.\5

For our analysis, we selected a subset of cases from the NASS data
for 1988 through 1991.  We included all one-car collisions involving
a 1987 or newer model year automobile and all collisions between a
model year 1987 or later automobile and any other car, van, pickup
truck, or other light truck.  These crash types, taken together,
accounted for about 81 percent of all automobile occupant fatalities
in 1991.  The remaining 19 percent occurred in types of crashes that
we did not include in our data set because of a lack of cases,
principally collisions with medium and heavy trucks (about 10 percent
of the 1991 total). 

In our statistical analyses, we used logistic regression to look at
the independent contributions of a variety of factors on the
probability of driver injury.\6 Driver injury was indexed with a
dichotomous outcome variable coded "1" if the driver was hospitalized
or killed in the crash and "0" otherwise.  The regression analysis
allowed us to isolate the effects of one factor (for example,
automobile weight) while statistically holding constant the other
factors (for example, collision severity as well as driver age and
gender).  Regression analysis answers the question:  If there were no
differences among these drivers except for the factor of automobile
weight, for example, how would that factor predict the probability of
driver injury?  (The data sets and analyses are described in appendix
I.)

The studies that we reviewed from the traffic safety literature
differed in several ways that increase the difficulty of comparing
their results and of relating their conclusions to our own findings. 
For instance, some studies focused on injuries to automobile drivers,
as we did, while others examined injuries to all automobile
occupants, not just drivers.  Similarly, different studies looked at
slightly different sets of automobile crashes--at tow-away crashes
(as we did) or at all police- reported crashes or only at crashes in
which a fatality occurred.  In addition, the studies employed
different outcome measures.  Our analysis concerned driver
hospitalizations or deaths, while other studies looked only at
fatalities or at injuries considered serious or worse or at injuries
categorized as moderately severe or worse, for example.  While these
and other differences mean that the studies we cite rarely produced
precisely equivalent findings, in most the findings were roughly the
same.  In particular, the direction of the findings was almost always
the same (that is, whether a factor increases or decreases the risk
of injury), and there was usually approximate agreement about the
size of the effect (that is, whether a factor has a large effect on
injury risk or only a minor influence).\7

Our work was performed in accordance with generally accepted
government auditing standards. 


--------------------
\5 For example, in a separate analysis not detailed here, we examined
all police-reported crashes for model year 1987 to 1989 cars in North
Carolina for calendar year 1990.  Injuries to automobile drivers
categorized as "serious" by North Carolina law enforcement personnel
occurred about 39 times more often in crashes in which at least one
vehicle was towed away than in other crashes, and 97 percent of all
"serious" driver injuries were in tow-away crashes. 

\6 For the logistic regressions, we used the Survey Data Analysis
(SUDAAN) statistical package.  SUDAAN takes into account the
stratification and unequal selection probabilities inherent in the
sampling design of NASS.  Failure to consider the sampling design in
the regressions is likely to produce artificially low standard
errors, biasing the analysis in favor of finding relationships that
appear to be statistically significant but that stem, in fact, from
chance.  For more information about the sampling plan for NASS, see
NHTSA (1991c).  For more information about SUDAAN, see Shah et al. 
(1992). 

\7 In general, because our data set included only very severe
crashes, the effect sizes from our analyses were somewhat larger than
those reported from analyses using a wider range of crashes.  The
reason for this is that almost all serious injuries occur in
police-reported tow-away crashes, and data sets with a wider range of
crashes thus necessarily include a high proportion of cases without
serious injuries.  Therefore, any statistical relationships are
diluted by the presence of many cases in which there is little chance
of injury regardless of the values of the independent variables. 


   ORGANIZATION OF THE REPORT
---------------------------------------------------------- Chapter 1:4

Chapter 2 looks at the effects of crash type and crash severity.  It
also discusses the effects of automobile size and safety belt use in
three different configurations:  one-car rollover crashes, one-car
nonrollover crashes, and collisions between cars and other light
vehicles.  Chapter 3 examines the influence of driver age and gender
on injury probability.  Chapter 4 discusses the relative
contributions of driver and automobile factors to driver injury. 
Chapter 5 discusses the potential for improving automobile safety. 
Appendix I describes our data set and statistical analyses. 


THE EFFECTS OF AUTOMOBILE WEIGHT,
AUTOMOBILE SIZE, SAFETY BELTS, AND
AIR BAGS IN DIFFERENT TYPES OF
CRASHES
============================================================ Chapter 2

This chapter discusses the safety consequences of crash
characteristics, automobile weight and size, and safety belts and air
bags.  The chapter begins with a discussion of crash characteristics
that are related to occupant injury, including the injury risk
associated with one-car rollover crashes, one-car nonrollover
crashes, and collisions with an automobile, a van, or a light truck. 
It then examines the safety consequences of automobile weight and
size as well as of safety belt use in the different crashes.  Each
section summarizes relevant findings from the literature and then
presents the results of our analyses of the NASS data.  The chapter
ends with a look at the effects of air bags. 


   CRASH CHARACTERISTICS
---------------------------------------------------------- Chapter 2:1


      PREVIOUS FINDINGS
-------------------------------------------------------- Chapter 2:1.1

The great majority of traffic crashes do not involve serious injury. 
A large proportion of all traffic crashes are not reported to the
police (Evans, 1991b).  And NHTSA has estimated that about one third
of crashes reported to the police involve personal injury (two thirds
having property damage only) and that just 6 percent involve a severe
or fatal injury (NHTSA, 1991b). 

Nonetheless, some crashes are much more likely to lead to serious
injury than others.  First, some types of crashes are more severe
than others.  Overall, single-car crashes are more likely to
seriously injure occupants than are multiple-vehicle collisions. 
Single-vehicle crashes account for about 30 percent of
police-reported crashes annually, yet in 1991 about 45 percent of all
automobile occupant fatalities were in these collisions.  Single-car
rollover crashes are particularly dangerous, accounting for only
about 2 percent of all police-reported accidents but about 20 percent
of occupant fatalities (NHTSA, 1991b).  A major reason for the
relative severity of single-car crashes is that most crashes
involving drunk drivers are single-car incidents, and crashes
involving drunk drivers tend to be more severe than other collisions. 
For example, NHTSA reported that 53 percent of drivers killed in
single-vehicle crashes in 1991 were intoxicated, compared with only
21 percent of the drivers killed in multiple-vehicle collisions
(NHTSA, 1993b). 

Second, for nonrollover crashes, some points of impact on the
automobile are more dangerous than others.  The preponderance of
fatal crashes other than rollovers involve frontal impacts, followed
at a distance by left- and right-side impacts.  For example, table
2.1 indicates that 43 percent of all automobile occupant fatalities
in 1991 occurred in frontal impacts, or more than half of all
fatalities that did not occur in single-car rollovers. 



                          Table 2.1
           
              Automobile Occupant Fatalities by
           Rollover or Principal Impact Point, 1991

                                                      Fatali
Type                                                      ty
----------------------------------------------------  ------
Nonrollover: principal impact point\a
Front                                                    43%
Left side                                                 16
Right side                                                15
Rear                                                       3
Other                                                      3
Single-car rollover                                       20
============================================================
Total                                                   100%
------------------------------------------------------------
\a Frontal impact points are clock positions 11:00, 12:00, and 1:00;
left-side impact points are clock positions 8:00, 9:00, and 10:00;
right-side, 2:00, 3:00, and 4:00; and rear, 5:00, 6:00, and 7:00. 

Source:  Our analysis of 1991 FARS data from National Highway Traffic
Safety Administration, Fatal Accident Reporting System 1991,
DOT-HS-807-954 (Washington, D.C.:  1993). 

Third, crashes at high speeds are more dangerous than others.  For
example, Joksch (1993) estimated that the risk to drivers of fatal
injury in two-car collisions was about 1 percent for a 20 mph
collision, 10 percent for a 35 mph collision, and 44 percent for a
collision involving a change in velocity of 50 mph.\1 NHTSA (1993e)
recently reported similar findings for restrained vehicle occupants,
noting, for instance, that the probability of fatal injury is about
nine times as great in frontal collisions with a change of velocity
of 40 mph as in those with a change of 30 mph.  That the probability
of death increases sharply with impact speed is one reason for the
predominance of frontal impacts in fatal crashes, since frontal
impacts are likely to involve cars that are moving forward. 


--------------------
\1 Change in velocity refers to the nearly instantaneous change in a
vehicle's speed in a crash.  For example, a vehicle that was abruptly
stopped from a travel speed of 30 mph would have a change in velocity
of 30 mph (from 30 to 0).  In contrast, a parked vehicle struck from
the rear and moved sharply forward at 10 mph would have a change in
velocity of 10 mph (from 0 to 10). 


      THE RESULTS OF OUR ANALYSIS
      OF CRASH DATA
-------------------------------------------------------- Chapter 2:1.2

Here we report how injury risk and driver and automobile
characteristics vary by crash type.  Our data, from NASS, were for
model year 1987 and later cars in three types of police- reported
tow-away crashes in 1988-91:  one-car rollovers, one-car
nonrollovers, and collisions with other cars, vans, and light trucks. 

The pattern of injury risk by crash type that we found in the NASS
data set reflects the pattern described in the literature.  As table
2.2 indicates, one-car crashes are more dangerous than multivehicle
collisions.  One-car rollover crashes, in particular, are much more
dangerous than other crash types, with a rate of driver
hospitalization or death that is double that of one-car nonrollover
crashes and four times that of collisions with cars and light trucks. 



                          Table 2.2
           
            Driver Injury, Driver Characteristics,
             and Automobile Weight, by Crash Type


                                                   Collision
                                                  with other
                       One-car        One-car  cars or light
Item                  rollover    nonrollover         trucks
---------------  -------------  -------------  -------------
Driver characteristic
------------------------------------------------------------
Injury
Hospitalized or          15.7%           7.0%           4.0%
 killed
Killed                     3.0            0.7            0.2
Male                      62.5           58.0           46.3

Age
------------------------------------------------------------
16-24                     53.6           43.3           33.7
25-44                     36.1           42.1           43.3
45-64                      7.0           10.8           15.0
65+                        3.3            3.8            8.0
Average                  2,502          2,688          2,669
 automobile
 curb weight
 (lbs)\b
------------------------------------------------------------
\a Percentages and means weighted by the National Inflation Factor to
represent population values.  The unweighted numbers of cases are 457
for one-car rollovers, 1,253 for one-car nonrollovers, and 4,393 for
collisions with other cars or light trucks.  The corresponding
estimated population numbers of cases are 104,806, 422,056, and
1,510,653.  (See appendix I.)

\b Curb weight refers to the weight of an unoccupied automobile,
including gasoline and other fluids.  In the NASS data set, curb
weight is recorded only to the nearest 100 pounds.  Therefore,
estimates of the average curb weight of any group of automobiles are
less exact than they would be if curb weight were recorded more
precisely. 

Source:  NASS 1988-91 data for model year 1987 and newer automobiles. 

In addition to having a higher rate of driver injury, one-car crashes
are disproportionately likely to involve men drivers and young
drivers.  (See table 2.2.) We found that about 60 percent of the
drivers in one-car crashes were men, compared with approximately 46
percent in two-vehicle collisions.  In addition, while close to half
of the drivers in one-car crashes were 16 to 24 years of age, only
about one third of the drivers in collisions with other cars or light
trucks were that young.  These findings are consistent with those of
our companion report, Highway Safety:  Factors Affecting Involvement
in Vehicle Crashes (GAO, 1994), in which we found that driver age and
gender are more strongly related to involvement in single-vehicle
crashes than they are to involvement in two-vehicle collisions. 

In addition, as table 2.2 indicates, the average curb weight of
automobiles in one-car rollover crashes was somewhat lower than that
of cars involved in other crashes.  As we noted in Highway Safety: 
Factors Affecting Involvement in Vehicle Crashes, involvement in
rollover crashes increases as automobile weight decreases. 

Just as the literature suggests, we found that the risk of injury to
drivers is significantly affected by impact point.  In particular, we
found that two-vehicle collisions involving head-on impacts between
vehicles moving in opposite directions are much more dangerous than
other two-vehicle crashes.  The risk of driver injury or death is
about five times as great in head-on collisions as in other
two-vehicle collisions.  For crashes other than head-on collisions,
frontal impacts and left-side impacts had higher rates of driver
hospitalization or death than others. 

Finally, we found that high impact speeds are, not surprisingly, more
dangerous than low impact speeds.  Considering the three crash types
together, we estimated that each increase of 10 mph in the change of
velocity at impact increases the probability of driver
hospitalization or death nearly sevenfold. 


   AUTOMOBILE WEIGHT AND SIZE
---------------------------------------------------------- Chapter 2:2


      PREVIOUS FINDINGS
-------------------------------------------------------- Chapter 2:2.1

As we have previously reported, safety experts agree that, in
general, heavier and larger cars are both more crashworthy and more
aggressive than lighter and smaller automobiles (GAO, 1991).  Thus,
in the event of a collision, occupants are less likely to be hurt
when they are in heavier and larger cars and when they are struck by
lighter and smaller cars. 

However, there is some disagreement in the literature about which is
the more important dimension for occupant safety, weight or exterior
size (that is, overall length and width).\2 Proponents of weight as
the important dimension argue that automobile mass protects occupants
from injury because it is aggressive--that is, heavier cars knock
down objects and push other vehicles back, thereby transferring
momentum and energy to the struck object, including other vehicles,
that could otherwise affect occupants of the striking vehicle (see,
for example, Evans and Frick, 1992).  In contrast, proponents of
exterior size as the more important dimension maintain that large
vehicles protect their occupants by absorbing crash energy without
increasing the injury risk of other roadway users (for example, see
Robertson, 1991). 

In most cases, this debate is of little practical significance now,
since weight and exterior size are very highly correlated--that is,
heavy cars are almost invariably also long and wide--but it has
important implications for the design of future automobiles.  If
exterior size is the more important dimension, using lighter weight
materials could make future automobiles lighter without decreasing
exterior size, thus increasing fuel efficiency without exacting a
safety cost.  Conversely, using lighter weight materials would
involve a safety cost if weight is the more important dimension. 

In addition, estimates of the amount of additional protection offered
by a given increase in automobile weight vary considerably.  For
example, consider the effects of a 500-pound increase in the weight
of one automobile in a collision with another, assuming no change in
the weight of the latter.  Klein, Hertz, and Borener (1991) used data
from two states to generate two different estimates of the decreased
risk of serious driver injury from that automobile weight
increase--13 percent and 20 percent.  Other estimates are higher. 
For example, Evans (1982) concluded that this increase in automobile
weight would reduce a driver's risk of fatal injury by about 29
percent. 

Further, the protective effect of automobile size appears to differ
by crash type.  First, it is likely that this effect is somewhat less
pronounced in one-car nonrollover crashes than in multivehicle
collisions (Evans, 1991b).  For example, in the 1991 paper by Klein
and colleagues, NHTSA researchers estimated that a 500-pound increase
in automobile weight reduces the risk of driver fatality by not quite
5 percent in one-car nonrollover crashes, somewhat less than the
estimates of 13 percent and 20 percent for two-car collisions. 

Second, although it is well documented that small cars are much more
likely to be involved in one-car rollover crashes than are large cars
(see, for example, GAO, 1994), the literature is less clear about the
safety consequences of automobile size once a rollover has occurred. 
On the one hand, in examining the effects of reduced automobile
weight and size on safety in rollover crashes, some researchers have
focused on the increased number of rollover crashes among light and
small cars (Evans, 1991b; Kahane, 1990; NHTSA, 1991a).  The
implication of these studies is that automobile weight and size do
not affect crashworthiness in rollovers; otherwise, these researchers
would have included weight and size as factors in their calculations. 
On the other hand, some direct studies of crashworthiness in
rollovers have found that drivers of larger cars are more likely to
be injured than drivers of smaller cars in rollovers (see, for
example, Partyka and Boehly, 1989).  One explanation for this finding
is that it takes more energy to roll over a heavy automobile than a
light one, meaning that the typical rollover crash involving heavy
autos is more severe (that is, occurs at a higher speed) than the
typical rollover involving light cars (see, for example, Terhune,
1991). 


--------------------
\2 Exterior size is measured by wheelbase--that is, the distance
between the front and rear axles. 


      THE RESULTS OF OUR ANALYSIS
      OF CRASH DATA
-------------------------------------------------------- Chapter 2:2.2


         THE EFFECTS OF AUTOMOBILE
         WEIGHT AND SIZE VARY BY
         CRASH TYPE
------------------------------------------------------ Chapter 2:2.2.1

After combining all the crashes in our database (one-car rollovers,
one-car nonrollovers, and collisions with cars and light trucks), we
found that the risk of injury to drivers was significantly reduced as
car weight and wheelbase increased in our sample.  (See figure 2.1.)
We estimated that the risk of driver hospitalization or death
decreases about 14 percent for every additional 500 pounds of
automobile weight and about 13 percent for each additional 5 inches
of wheelbase.  (See tables I.1 and I.2.)

   Figure 2.1:  Estimated
   Probability of Driver
   Hospitalization or Death in
   One-Car Crashes and Collisions
   With Other Cars or Light
   Trucks, by Automobile Weight
   and Wheelbase\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
typical driver:  a 30-year-old man wearing manual lap and shoulder
safety belts.  Injury probability changes by 0.97 for each additional
100 pounds of automobile weight and by 0.973 for each additional inch
of wheelbase. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

Further, considering all crash types taken together, we could not
statistically differentiate the injury reduction effects of curb
weight and wheelbase.  That is, the benefits of increasing weight and
wheelbase were roughly equivalent in reducing injuries, and we were
unable to establish that one had a stronger influence than the other. 
The nearly equivalent slopes of the lines for weight and wheelbase in
figure 2.1 demonstrate this.  The endpoints of the lines in figure
2.1 represent approximately the 5th and 95th percentiles of
automobile weight and wheelbase in this data set.  Thus, 2,000-pound
cars are among the lightest and 3,600-pound cars are among the
heaviest in this database of cars involved in serious crashes;
similarly, cars with a wheelbase of 93 inches are among the shortest,
113 inches among the longest.\3 Figure 2.1 shows that, for all three
crash types taken together, whether measured by weight or wheelbase,
drivers in the heaviest and largest cars had a risk of
hospitalization or death about 40 percent less than the drivers of
the lightest and smallest cars. 

However, these overall effects mask the fact that automobile weight
and wheelbase have very different safety consequences in different
types of crashes.  Figure 2.2 shows the estimated effects of curb
weight separately for the three crash types (see also tables
I.3-I.5); figure 2.3, the estimated effects of wheelbase (see also
tables I.6-I.8). 

   Figure 2.2:  Estimated
   Probability of Driver
   Hospitalization or Death by
   Crash Type and Automobile
   Weight\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
typical driver:  a 30-year-old man wearing manual lap and shoulder
safety belts.  Injury probability changes by 1.097 for each
additional 100 pounds of automobile weight in one-car rollover
crashes, by 0.968 in one-car nonrollover crashes, and by 0.948 in
collisions with other cars or light trucks. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

   Figure 2.3:  Estimated
   Probability of Driver
   Hospitalization or Death by
   Crash Type and Automobile
   Wheelbase\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
typical driver:  a 30-year-old man wearing manual lap and shoulder
safety belts.  Injury probability changes by 1.020 for each
additional wheelbase inch in one-car rollover crashes, by 0.980 in
one-car nonrollover crashes, and by 0.958 in collisions with other
cars or light trucks. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

Most importantly, although increasing weight and wheelbase reduces
the risk of driver injury in one-car nonrollover crashes and in
collisions with other cars or light trucks, drivers in heavier cars
were much more likely to be hospitalized or killed in one-car
rollover crashes than were drivers of lighter automobiles.  For
one-car rollover crashes, we estimated that each 500 pounds of
additional automobile weight increases the risk of driver
hospitalization or death by about 59 percent.  This effect is solely
a function of automobile weight, not of wheelbase; we found that the
relationship between wheelbase and driver injury was not
statistically significant in one-car rollovers.  This finding agrees
with the report of Partyka and Boehly (1989) that drivers of heavier
and larger cars are more likely to be injured in rollovers than
drivers of lighter and smaller cars.  This finding is also consistent
with the explanation that it takes more energy to roll over a heavy
automobile than a light one, meaning that rollover crashes involving
heavy autos occur at higher speeds than rollovers involving light
cars.  However, it is important to keep in mind that the rate of
involvement in one-car rollover crashes is much greater for light
cars than for heavy ones, so this finding does not necessarily mean
that, considering both involvement and crashworthiness, drivers of
heavy cars are more likely to suffer injuries in one-car rollovers. 

Figures 2.2 and 2.3 also show that we found a tendency for the risk
of driver hospitalization or death to decrease with increasing car
weight and size in one-car nonrollover crashes, but neither curb
weight nor wheelbase was a statistically significant predictor of
driver injury in those crashes.  In contrast, we found that in
collisions with other cars and light trucks, both automobile weight
and wheelbase were statistically significant predictors of driver
injury.  In those crashes, we estimate that each additional 500
pounds of automobile weight decreased the risk of driver
hospitalization or death by about 23 percent and each 5 inches of
additional wheelbase lowered the risk of driver injury approximately
19 percent.  These findings reflect the pattern, described in the
literature, that the protective effects of automobile weight and
wheelbase are somewhat greater in multivehicle collisions than in
single-car nonrollover crashes. 


--------------------
\3 The heaviest cars in the data set weighed more than 4,400 pounds,
the lightest less than 1,600 pounds.  The longest cars had wheelbases
of more than 120 inches, the shortest less than 84 inches. 


         CHARACTERISTICS OF THE
         OTHER VEHICLE AFFECT
         INJURY RISK
------------------------------------------------------ Chapter 2:2.2.2

In two-vehicle collisions, the injury risk of an automobile occupant
is affected not only by the characteristics of his or her own
automobile but also by the characteristics of the other vehicle.  We
looked at the effects of the weight and vehicle type of the other
vehicle on the probability of injury for the first driver:  both
factors affect the aggressivity of the other vehicle. 

First, not surprisingly, heavier vehicles pose more of a risk than
lighter vehicles.  (See figure 2.4.) In our analysis, each increase
of 500 pounds in the weight of the other vehicle increased the
probability of hospitalization or death by about 13 percent, holding
other factors constant.  (See table I.5.) It is important to note
that the magnitude of this aggressive effect of vehicle weight is
less than that of the protective effect of weight described earlier. 
(We estimated that each additional 500 pounds of automobile weight
reduces the probability of injury by about 23 percent.) After
statistically controlling for the influence of other factors, we
found that this ratio of the protective effect to the aggressive
effect of automobile weight of 1.77 to 1 is roughly consistent with
the findings of other researchers.  For example, Klein, Hertz, and
Borener (1991), analyzing data from two different states, generated
two estimates of the size of this ratio in two-car collisions:  1.54
to 1 and 1.30 to 1. 

   Figure 2.4:  Estimated
   Probability of Driver
   Hospitalization or Death in
   Collisions With Other Cars or
   Light Trucks by Weight of the
   Other Vehicle\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
typical driver:  a 30-year-old man wearing manual lap and shoulder
safety belts.  Injury probability changes by 1.025 for each
additional 100 pounds in the weight of the other vehicle. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

Second, figure 2.5 shows that driver injury risk is strongly
influenced by the body type of the other vehicle.  We found that
while pickup trucks do not pose more danger than automobiles, vans
and other light trucks are more aggressive than automobiles.  Indeed,
statistically controlling for the weight of the driver's car and of
the other vehicle, we estimate that the risk of hospitalization or
death for the driver is more than twice as great in collisions with
vans and light trucks than with other cars or light vehicles.  (See
also table I.5.) This finding reflects two characteristics of vans
and light trucks.  One is that because vans and light trucks can
carry heavy cargo loads, these vehicles may be, in reality, heavier
than the curb weight measurements available to us indicate.  The
second characteristic is that the structure and design of vans and
other light trucks make those vehicles especially dangerous for
automobile occupants in two-vehicle collisions (National Research
Council, 1992; Terhune and Ranney, 1984). 

   Figure 2.5:  Estimated
   Probability of Driver
   Hospitalization or Death in
   Collisions With Other Cars or
   Light Trucks by Type of the
   Other Vehicle\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
typical driver:  a 30-year-old man wearing manual lap and shoulder
safety belts. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 


   SAFETY BELTS
---------------------------------------------------------- Chapter 2:3


      PREVIOUS FINDINGS
-------------------------------------------------------- Chapter 2:3.1

Safety belts greatly reduce the risk of injury and death in roadway
crashes.  In a recent review of studies of safety belt effectiveness,
we concluded that most studies show that belted vehicle occupants
have a risk of serious injury or death that is approximately 50 to 75
percent less than that of unrestrained occupants (GAO, 1992).  Other
researchers have found safety belts to have slightly smaller effects. 
For example, NHTSA (1993d) estimated that when manual lap and
shoulder safety belts are used in serious crashes, they reduce
fatality risk by 45 percent.  Similarly, Evans (1986) estimated that
three-point lap and shoulder safety belts reduce a driver's risk of
fatality by about 43 percent, with about half of that benefit the
result of eliminating or attenuating impacts with the interior of the
vehicle and about half the result of preventing occupant ejection. 

There are three other important points about safety belt
effectiveness.  First, the effectiveness of safety belts varies by
crash type.  Belts are most effective in rollover crashes because
they largely prevent occupant ejection (Evans, 1990; Partyka, 1988). 
They are also more effective in one-car crashes than in multivehicle
collisions.  For example, Evans and Frick (1986) estimated that
safety belts reduce the risk of driver fatality in one-car crashes by
62 percent but by only 30 percent in two-car crashes.  Second, belt
effectiveness also varies by point of impact.  Belts are most
effective in frontal impacts and least effective in left-side impacts
(Evans, 1990).  Since one-car crashes are more likely to involve
frontal impacts than are two-car collisions, this offers one possible
explanation for the greater efficacy of safety belts in one-car
crashes. 

Third, it is likely that manual lap and shoulder belts are somewhat
more effective than other safety belt configurations.  For example,
Evans (1991a) estimated that lap and shoulder belts reduce fatality
risk in serious collisions by about 41 percent, compared with
estimated risk reductions of 18 percent for lap belts only and 29
percent for shoulder belts only.  Evans speculated that these two
components have somewhat different functions, with lap belts
primarily preventing ejection and shoulder belts mitigating contact
with the interior of the vehicle.  Comparing manual lap and shoulder
belts to automatic belts, NHTSA (1993d) estimated that automatic
safety belts, when used in serious crashes, reduce the risk of
fatality by 42.5 percent, compared with an estimated fatality
reduction of 45 percent for manual lap and shoulder belts. 


      THE RESULTS OF OUR ANALYSIS
      OF CRASH DATA
-------------------------------------------------------- Chapter 2:3.2

Considering all three crash types together, NASS researchers
categorized 73 percent of the drivers in the NASS data set as using a
safety belt at the time of collision, with those involved in one-car
rollovers slightly less likely to be belted than others.  This figure
is higher than might be expected from the results of other estimates
of safety belt use among the general driving population, particularly
given that drivers involved in crashes are less likely to wear safety
belts than others and that all the drivers included in our analysis
had been involved in a crash.  In one point of comparison, NHTSA
estimated a 51-percent safety belt usage rate for all passenger cars
in 1991 (NHTSA, 1992a).  Further, it is well established that
unbelted drivers are more reckless than belted drivers (Evans and
Wasielewski, 1983; Evans, 1987; Preusser, Williams, and Lund, 1991;
Stewart, 1993).  As a result, unbelted drivers have much higher crash
involvement rates than belted drivers:  NHTSA (1992a) estimated that
unbelted drivers have an involvement rate in potentially fatal
crashes that is more than double that of belted drivers.\4

We cannot determine with certainty if, or to what degree, the safety
belt use figures reported in NASS are incorrect, nor can we determine
with certainty the extent to which any potential bias in those
figures affected our analyses.  For that reason, our results should
be interpreted with caution.  Nonetheless, because the results of our
analyses concerning the relative effectiveness of different safety
belt configurations in different types of crashes are consistent with
the findings from the traffic safety literature, we believe that any
potential bias has not seriously affected our findings. 

For each of the three categories of crashes, we examined the
performance of three safety belt configurations:  (1) manual lap and
shoulder belts, (2) automatic and manual belts combined (most
commonly automatic shoulder belts and manual lap belts), and (3)
automatic belts without manual components.  Other safety belt
configurations, including manual lap belts alone, had too few cases
in the data set for us to estimate their effectiveness. 

Statistically controlling for crash severity, driver characteristics,
and other background factors, we found that, compared with unbelted
drivers, drivers using any of the three safety belt configurations
had greatly reduced risks of injury.  We also found that, looking at
the three types of crashes together, manual lap and shoulder belts
were somewhat more effective in preventing driver injury than the
other configurations.  (See figure 2.6.) Compared with unbelted
drivers, the estimated risk of hospitalization or death was reduced
about 70 percent for those using manual lap and shoulder belts, about
63 percent for those using automatic and manual belts combined, and
about 54 percent for those using automatic belts without manual
components.  (See also table I.1.)

   Figure 2.6:  Estimated
   Probability of Driver
   Hospitalization or Death in
   One-Car Crashes and Collisions
   With Other Cars or Light Trucks
   by Safety Belt Use\a

   (See figure in printed
   edition.)

\a Estimated probability of driver injury in an average crash for a
30-year-old man. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

We also found small variations in safety belt performance among the
different types of crashes.  Safety belts were somewhat less
effective in collisions with other cars or light trucks than they
were in single-car crashes.  For example, in our analysis, manual lap
and shoulder belts reduced the risk of driver hospitalization or
death by 83 percent in one-car rollover crashes and by 80 percent in
one-car nonrollover crashes but by only 64 percent in collisions with
other cars or light trucks.  (See tables I.3-I.5.)


--------------------
\4 More specifically, NHTSA (1992a, pp.  20-21) estimated that an
automobile fleet composed entirely of cars equipped with manual
safety belts would have had a 48-percent belt usage rate in 1991
(with 52 percent of drivers unbelted) and that the manual belt use
rate in "potentially fatal" crashes would have been 29.5 percent
(with 70.5 percent of drivers unbelted).  Therefore, by these
estimates, the odds of drivers not using a belt in the general
population would be 1.08 (52/48), the odds of drivers not using a
belt in the crash-involved population would be 2.39 (70.5/29.5), and
the relative risk of involvement in serious crashes of unbelted
drivers would be 2.21 (2.39/1.08). 


   AIR BAGS
---------------------------------------------------------- Chapter 2:4

Evaluations of the effectiveness of air bags are hampered by the
relatively small number of cars now equipped with them (although all
passenger cars, vans, and light trucks will be required to have both
driver- and passenger-side air bags by the 1998 model year).  There
were too few automobiles with air bags in the NASS data set for us to
conduct our own analysis of air bag effectiveness.  Nonetheless, some
of the characteristics of air bag performance have already been
established.  First, air bags are effective only in frontal impacts;
they do not protect drivers in side impacts or other nonfrontal
collisions (see, for example, Zador and Ciccone, 1993).  While
frontal impacts account for by far the greatest proportion of
automobile occupant fatalities, more than half of occupant fatalities
do not involve frontal impacts.  (See table 2.1.)

Second, air bags offer additional protection to drivers already
wearing safety belts.  Researchers have found that belted drivers
with air bags are about 10 percent less likely to be fatally injured
than are belted drivers without air bags (Evans, 1991b; Zador and
Ciccone, 1993).  For example, NHTSA (1993d) estimated that lap and
shoulder safety belts alone reduce automobile driver fatality risk by
about 45 percent.  In that paper, NHTSA also estimated that drivers
with lap and shoulder belts and air bags are about 50 percent less
likely to be killed than unbelted drivers, for a safety increment of
close to 10 percent (50/45 = 1.11, or about 10 percent). 

Finally, safety belts alone are much more effective than air bags
alone.  Estimates of the effectiveness of air bags for drivers who do
not wear safety belts indicate that those drivers are approximately
20 to 30 percent less likely to be killed in a collision than are
unbelted drivers without air bags (NHTSA, 1993d; Zador and Ciccone,
1993).  In contrast, as noted previously, drivers wearing lap and
shoulder safety belts are, by the most conservative estimate, 41
percent less likely to be killed than unbelted drivers.\5


--------------------
\5 A potential concern about air bags that has not yet been
thoroughly examined is that driver ejection rates may be somewhat
higher in cars with air bags.  NHTSA (1992a, pp.  50-51) reported
that, in fatal accidents, drivers in vehicles with air bags are
significantly more likely to be ejected from the vehicles than are
drivers in cars with only manual safety belts.  NHTSA appropriately
noted that this result in a database containing only fatal accidents
is suspect, since drivers "saved" by air bags are not included in the
database unless another person in the crash was killed.  However, the
fact that a statistically significant relationship between air bags
and ejection was found in a data set with a relatively small number
of cases suggests that this may be a strong effect.  At the least,
this question deserves further investigation. 


   SUMMARY
---------------------------------------------------------- Chapter 2:5

One-car crashes have higher injury rates than multivehicle
collisions, and one-car rollover crashes are more dangerous than
one-car nonrollover crashes.  Further, automobile drivers are
especially likely to suffer serious injury in crashes involving a
frontal impact, and the probability of injury is greater at higher
impact speeds. 

The protective effects of automobile weight and wheelbase differ by
crash type.  In one-car rollover crashes, drivers of heavier cars are
more likely to be hospitalized or killed than drivers of lighter
cars.  Automobile wheelbase is not statistically related to injury
risk in rollovers.  In contrast, increasing both weight and wheelbase
reduces the risk of driver injury in one-car nonrollover crashes and
in collisions with other cars or light trucks, with a larger safety
benefit in multivehicle collisions than in one-car nonrollover
crashes. 

Safety belts substantially reduce the risk of driver injury.  Manual
lap and shoulder belts are somewhat more effective than other belt
configurations, and safety belts are more beneficial in one-car
crashes than in multivehicle collisions.  The traffic safety
literature indicates that air bags offer a modest degree of
additional protection to belted occupants and that safety belts
without air bags are much more effective than air bags without safety
belts. 


   AGENCY COMMENTS AND OUR
   EVALUATION
---------------------------------------------------------- Chapter 2:6

DOT had one general comment concerning the topics presented in this
chapter:  it maintained that the subset of the NASS data we used in
the report is inappropriate for studying the effect of car size on
safety and, more particularly, that the sample size is inadequate for
assessing the consequences of changing the weight of both vehicles in
a two-vehicle collision.  We disagree.  As the findings presented in
this chapter demonstrate, the NASS data set we constructed clearly
was adequate for uncovering a number of statistically significant
relationships (the analyses are described in appendix I).  In
addition, our findings are similar to NHTSA's findings from
statistical analyses of state accident databases (particularly
concerning the effects of the weights of both vehicles in two-car
collisions; see Klein, Hertz, and Borener, 1991) and to NHTSA's
findings from statistical analyses of a slightly different NASS
database (see table I.1 and NHTSA, 1992a, p.  72). 


THE EFFECTS OF GENDER AND AGE ON
DRIVER INJURY RISK
============================================================ Chapter 3


   THE VULNERABILITY OF WOMEN AND
   OLDER DRIVERS
---------------------------------------------------------- Chapter 3:1


      PREVIOUS FINDINGS
-------------------------------------------------------- Chapter 3:1.1

Safety researchers have consistently found that women automobile
occupants have a greater risk of injury in a collision than men and
that the risk of injury increases with occupant age.  For example,
NHTSA (1992a) found that women vehicle occupants involved in tow-away
crashes are 36 percent more likely than men to suffer an injury
categorized as moderately severe or worse.  NHTSA also found that the
risk of moderate injury increases about 2 percent for each year of
age, meaning that, compared with 20-year-olds, 30-year-olds have a
21-percent greater risk of injury and 60-year-olds are more than
twice as likely to be injured.  Similarly, Evans (1988b) reported
that 30-year-old women have a fatality risk in traffic crashes about
31-percent higher than 30-year-old men and that the risk of fatality
increases about 2 percent for each year of age. 


      THE RESULTS OF OUR ANALYSIS
      OF CRASH DATA
-------------------------------------------------------- Chapter 3:1.2

Our analysis of the NASS data set of police-reported tow-away crashes
produced similar findings.  For statistically equivalent crashes, we
found that women drivers are about 29 percent more likely to be
hospitalized or killed than men drivers.  We also found that drivers
65 and older are about 4.5 times more likely to be seriously hurt
than drivers 16 to 24 years old in equivalent crashes.  (See table
I.1.)


   POSSIBLE EXPLANATIONS
---------------------------------------------------------- Chapter 3:2


      PREVIOUS FINDINGS
-------------------------------------------------------- Chapter 3:2.1

One explanation for the greater vulnerability of women drivers and
older drivers emphasizes their inherent physical frailty.  This view
postulates that the same degree of physical trauma is more likely to
produce injury in women than in men and in older automobile occupants
than in younger ones, because women and older people are physically
less resilient than men and younger people.  Indeed, there is some
support for the view that women are physically more vulnerable than
men (Evans, 1988b), and that older people are more fragile than
younger ones is well documented (for example, Mackay, 1988; Pike,
1989).  The implication of this view is that the greater
vulnerability of women and older persons is not amenable to
correction through automobile design changes, because weaker
individuals will be hurt more often than stronger ones no matter
what. 

Other possible explanations have not been carefully developed in the
literature, but they tend to involve speculation that some
characteristic of the vulnerable group interacts with automobile
design to cause a safety problem.  For example, because women are
shorter than men, on the average, they may sit closer to the steering
wheel, causing them to hit the steering column more quickly in a
crash.  Similarly, the interaction of lower height and safety belts
designed for average-sized drivers may oblige women, for reasons of
comfort, to wear safety belts incorrectly more than men do, thereby
increasing the injury risk of ostensibly belted women drivers
relative to that of belted men drivers (see, for example, National
Transportation Safety Board, 1988). 


      THE RESULTS OF OUR ANALYSIS
      OF CRASH DATA
-------------------------------------------------------- Chapter 3:2.2

Here, we discuss whether the factors we examined in chapter 2
differentially affect the probability of injury of women and men and
of older and younger drivers.  If the "inherent frailty" view is
correct, women should be injured more than men, and older drivers
more than younger drivers, regardless of crash type, automobile
weight, or safety belt use.  If any of these factors affect the
relationship between gender or age and injury risk, the credibility
of this view would be called into question, as this would mean that
something other than frailty also makes an important difference.  It
would also indicate that the safety of women and older drivers could
be at least somewhat improved by automobile design changes. 


         FACTORS AFFECTING WOMEN
         DRIVERS
------------------------------------------------------ Chapter 3:2.2.1

Crash Type.  The pattern of injury by crash type varies for women
drivers and men drivers.  Multivehicle collisions are a greater
source of injury for women than they are for men.  Figure 3.1 shows
our finding that 67 percent of the women drivers hospitalized or
killed were injured in collisions with cars and light trucks, with
only one third injured in one-car crashes (11 percent in rollovers,
22 percent in nonrollovers).  In contrast, only 45 percent of the men
drivers hospitalized or killed were injured in collisions with cars
and light trucks; most of the men drivers were hurt in one-car
crashes (20 percent of the total in rollovers, 35 percent in
nonrollovers). 

   Figure 3.1:  Distribution of
   Driver Hospitalization or Death
   by Crash Type and Gender\a

   (See figure in printed
   edition.)

\a Driver injuries are weighted by the National Inflation Factor to
produce population estimates.  Columns sum to 100 percent separately
for men and women. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

One reason for these differences in the pattern of injury is that men
and women drivers tend to be involved in different types of crashes,
as described in chapter 2.  Men drivers are involved in one-car
crashes more often than women drivers.\1 In our analysis, 69 percent
of the crash involvements of men drivers were in collisions with cars
and light trucks, with about 31 percent in one-car crashes.  In
contrast, about 79 percent of the crash involvements of women drivers
were in collisions with cars and light trucks, with only about 21
percent in one-car crashes. 

However, another reason is that women drivers are much more likely
than men drivers to be hospitalized or killed in collisions with cars
and light trucks.  That is, women drivers are especially likely to be
hurt in the type of crash that they are also particularly likely to
experience.\2 In statistically equivalent crashes, women drivers are
52 percent more likely than men drivers to be hospitalized or killed
in collisions with other cars or light trucks, but injury risks for
women drivers are roughly the same as those for men drivers in
one-car crashes--4 percent higher in one-car rollovers and 6 percent
lower in other one-car crashes.\3 (See tables I.3-1.5.)

Automobile Weight.  In our data set, women drove lighter and smaller
cars than men.  The automobiles women drove had an average curb
weight of 2,615 pounds and a mean wheelbase of 100.7 inches; for men
drivers, the figures were 2,715 pounds and 101.5 inches. 

We also found that the protective effect of increasing automobile
weight was less evident for women drivers than for men drivers. 
Since increasing weight generally offers protection in a crash, the
average automobile weight for drivers who were hospitalized or killed
should be lower than the average weight for those who were not
injured.  This was true for men but not for women.  The average curb
weight of the cars driven by men who were hospitalized or killed was
2,626 pounds, compared with a greater average curb weight of 2,719
pounds for men who were not injured.  In contrast, the average
automobile curb weight for women drivers who were hospitalized or
killed was 2,611 pounds, compared with an equivalent average curb
weight of 2,615 pounds for women drivers who were not injured.\4

Safety Belts.  Each of the safety belt configurations that we
examined (manual lap and shoulder belts, automatic and manual belts,
and automatic belts only) significantly reduced the injury risk of
both men and women drivers.\5 However, we also uncovered evidence
that, in this data set, safety belts were somewhat less effective for
women drivers than for men drivers. 

Table 3.1 compares men and women automobile drivers hospitalized as
the result of a crash by safety belt use.  For all three types of
crashes, the table separates the percentage of drivers who were
hospitalized or killed from those not hospitalized as well as
separating men and women in each group.  Safety belt use did not
differ by gender for drivers who were not hospitalized:  about three
quarters of both the men and women drivers in that group were belted. 
If safety belts offered equivalent protection to men and women
drivers, the belt use percentages among hospitalized or killed
drivers should reflect the same pattern--in this case, rough
equivalence for men and women.  However, the table shows that among
drivers who were hospitalized or killed, women were more likely to
have been wearing safety belts than men.  In particular, injured
women drivers were about 50 percent more likely to have been wearing
manual lap and shoulder belts than were injured men (36 percent to 24
percent).\6



                          Table 3.1
           
              Safety Belt Use by Driver's Injury
                     Outcome and Gender\a



Safety belt
configuration              Men     Women       Men     Women
--------------------  --------  --------  --------  --------
Total belted               44%       55%       74%       75%
Manual lap and              24        36        48        52
 shoulder belts
Automatic and manual         6         9        12        12
 belts
Automatic belts             14        10        15        11
Total unbelted             56%       45%       26%       25%
============================================================
Total                     100%      100%      100%      100%
------------------------------------------------------------
\a The percentages are weighted by the National Inflation Factor to
represent population values. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 


--------------------
\1 See also Highway Safety:  Factors Affecting Involvement in Vehicle
Crashes (GAO, 1994).  There, we found that men drivers are more
frequently involved in single-vehicle crashes than women drivers but
that men and women drivers do not differ in their involvement in
two-vehicle collisions. 

\2 As noted at the begining of the chapter, combining the three types
of crashes, our overall estimate is that women drivers are about 29
percent more likely to be hospitalized or killed in a collision than
men drivers. 

\3 These estimates were derived from logistic regression analyses
that compared the injury experiences of women and men drivers in
crashes that were statistically equivalent on a number of important
dimensions, including crash severity, impact point, safety belt use,
driver age, and automobile weight.  This means that any differences
that may exist between men and women drivers on these dimensions
cannot account for the finding that women drivers are more likely
than men drivers to be injured in collisions with cars and light
trucks.  Thus, whether or not women and men drivers differ on these
dimensions, women drivers' increased injury risk in these crashes is
not the result of, for example, women being involved in more severe
crashes or of women more often driving smaller cars, among other
possible explanations. 

\4 Using the SUDAAN statistical software, we also conducted logistic
regression analyses separately for men and women drivers.  For men
drivers, the coefficient for automobile curb weight (-0.043) has a
p-value of less than 0.01; that coefficient translates to an odds
ratio of 0.958.  For women drivers, the coefficient for automobile
curb weight (-0.012) was not statistically significant and translates
to an odds ratio of 0.988. 

\5 As we discussed in chapter 2, the percentage of drivers coded as
using safety belts in the NASS data set is higher than expected. 
While we do not believe that any potential bias in the data set has
seriously affected our findings, these results should be interpreted
with caution. 

\6 Using the SUDAAN statistical software, we conducted logistic
regression analyses to discriminate between men and women drivers. 
We did separate analyses for drivers who were hospitalized or killed
and for those who were not injured, and we used our standard set of
crash-related independent variables.  For injured drivers, the only
statistically significant safety belt factor was the variable for
manual lap and shoulder belts.  For drivers who were not injured,
none of the safety belt variables was statistically significant. 


      FACTORS AFFECTING OLDER
      DRIVERS
-------------------------------------------------------- Chapter 3:2.3

Crash Type.  The patterns of injury by crash type are very different
for drivers 65 and older and for younger drivers.\7 Figure 3.2 shows
that, in our analysis, nearly four fifths of the drivers 65 or older
who were hospitalized or killed were injured in collisions with cars
or light trucks, while only about one fifth were injured in one-car
crashes (and almost none were hurt in one-car rollovers--just 3
percent).  Conversely, just over half of the drivers 16 to 64 who
were hospitalized or killed were injured in collisions with cars or
light trucks, while about 29 percent were hurt in one-car
nonrollovers and 17 percent were in one-car rollovers. 

   Figure 3.2:  Distribution of
   Driver Hospitalization or Death
   by Crash Type and Driver Age\a

   (See figure in printed
   edition.)

\a Driver injuries are weighted by the National Inflation Factor to
produce population estimates.  Columns sum to 100 percent separately
for each age category. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

The primary reason for this difference between the age categories is
that drivers in the two groups are involved in different types of
crashes.  Drivers younger than 65 are involved in collisions with
cars and light trucks less often, and in one-car crashes more often,
than are drivers 65 and older.\8 In our analysis, 73 percent of the
crash involvements of drivers 16 to 64 were in collisions with cars
and light trucks, about 22 percent in one-car nonrollover crashes,
and about 5 percent in one-car rollover crashes.  In contrast, about
86 percent of the crash involvements of drivers 65 and older were
collisions with cars and light trucks, with only about 11 percent
one-car nonrollover crashes and just 3 percent one-car rollovers. 

Older drivers are much more likely to be hurt in crashes than younger
drivers in almost all circumstances.  For one-car nonrollover
crashes, we found that, in statistically equivalent crashes, drivers
65 and older were hospitalized or killed about 6.6 times more often
than the youngest drivers, those 16 to 24.  Similarly, for collisions
with cars and light trucks, drivers 65 and older had a probability of
injury more than four times as great as drivers 16 to 24. 

Automobile Weight.  Drivers 65 and older operated heavier and larger
cars than younger drivers.  The automobiles of drivers 65 and older
had an average curb weight of 2,874 pounds and a mean wheelbase of
104.9 inches.  The automobiles of drivers 16 to 64 had an average
curb weight of 2,649 pounds and a mean wheelbase of 100.8 inches. 

We also found that the protective effect of increasing automobile
weight was only slightly less strong for drivers 65 and older than
for younger drivers.  Thus, the average curb weight of the cars
driven by those 16 to 64 who were hospitalized or killed was 2,590
pounds, compared with a larger average curb weight of 2,652 pounds
for those who were not hospitalized.  The average automobile curb
weight for drivers 65 and older who were hospitalized or killed was
2,836 pounds, compared with an average curb weight of 2,878 pounds
for drivers who were not hospitalized. 

Safety Belts.  Although the safety belt use figures in the NASS data
set may be inflated, as we discussed earlier, we found that safety
belts reduced the risk of injury for drivers in both age categories. 
We also found that the effectiveness of safety belts was roughly
equivalent for drivers 16 to 64 and for drivers 65 and older in this
data set.  For example, table 3.2 shows the percentage of belted
drivers separately for those hospitalized or killed and for those not
hospitalized, as well as separating these categories by age.  The
table shows that drivers 65 and older used safety belts more often
than drivers 16 to 64 and that this pattern holds both among those
who were hospitalized or killed and among those who were not
hospitalized.  Thus, while older drivers use safety belts more
frequently, this difference from younger drivers is found across the
board, rather than only among the hospitalized and killed, as it was
for the comparison between women drivers and men drivers. 



                          Table 3.2
           
              Safety Belt Use by Driver's Injury
                      Outcome and Age\a



                                  Age 65              Age 65
Safety belt            Age 16-       and   Age 16-       and
configuration               64     older        64     older
--------------------  --------  --------  --------  --------
Total belted               48%       66%       74%       82%
Manual lap and              29        39        49        55
 shoulder belts
Automatic and manual         8         8        12         8
 belts
Automatic belts             11        19        13        19
Total unbelted             52%       34%       26%       18%
============================================================
Total                     100%      100%      100%      100%
------------------------------------------------------------
\a The percentages are weighted by the National Inflation Factor to
represent population values. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 


--------------------
\7 We chose to compare drivers 65 and older with younger drivers in
order to present the results of our analysis in a straightforward
manner.  However, any division of age categories is arbitrary. 
Driver age has many possible values and the relationship between
driver age and injury risk is roughly linear--that is, for a given
level of trauma, injury risk increases with each additional year of
age. 

\8 See Highway Safety:  Factors Affecting Involvement in Vehicle
Crashes (GAO, 1994).  There, we reported that younger drivers are
particularly likely to be involved in one-car crashes but not
multivehicle collisions; conversely, older drivers are less likely
than others to be involved in one-car crashes but more likely to
experience multivehicle collisions. 


   CONCLUSIONS AND IMPLICATIONS
---------------------------------------------------------- Chapter 3:3

Taken as a whole, the evidence indicates that the "inherent frailty"
hypothesis does not accurately describe the injury experience of
women drivers in automobile crashes but is consistent with that of
older drivers.  This is because the relative injury risk of women
drivers compared with men drivers differs as a function of crash
type, automobile size, and safety belt use, while the relative injury
risk of drivers 65 and older compared with younger drivers is largely
unaffected by those three factors.  Women drivers are more likely
than men drivers to be hospitalized or killed in collisions with cars
and light trucks but not in one-car crashes, and women drivers may be
protected less well by heavier cars and by safety belts than are men
drivers.  In contrast, drivers 65 and older have a greater risk of
hospitalization or death than younger drivers in one-car as well as
multivehicle crashes, and they are afforded roughly the same degree
of protection as drivers 16 to 64 by greater automobile weight and
safety belt use. 

The NASS data set did not allow us to pursue more specific
explanations for differences stemming from gender and age.  For
example, men and women differ in many ways--on the average, women are
shorter than men, weigh less than men, and have bones that are less
strong than men's, among other potentially relevant differences.  It
is difficult to identify the key difference that accounts for women's
greater injury risk.  Our findings about the applicability of the
inherent frailty hypothesis suggest that the concerns of women
drivers are more likely to be ameliorated by automobile design
changes than are those of older drivers.  This means not that it is
impossible to reduce the injury risk of drivers 65 and older but only
that it may be difficult to close the gap between older and younger
drivers.  The implications of our findings for future automobile
safety are discussed in chapter 5. 

Three other points are worthy of mention.  First, it is not
surprising that the injury risk in one-car rollover crashes is
similar both for women and men drivers and for drivers older and
younger than 65.  One-car rollover crashes are very severe events,
meaning that differences between individual drivers are likely to be
overwhelmed by the magnitude of the crash.  Further, few of the
drivers in one-car rollover crashes were either women or 65 or older. 

Second, while our finding that safety belts may not protect women
drivers as well as men drivers is far from definitive, other
researchers examining data from other sources have also reported that
the benefits of safety belts are not as great for women as they are
for men.  (See, for example, Hill, Mackay, and Morris, 1994; Mercier
et al., 1993.)

Third, the types of crashes experienced by drivers 65 and older
reduce the protective influence of automobile weight for them.  Not
only are older drivers much more likely to have multivehicle than
one-car crashes; also, those multivehicle collisions occur
disproportionately in intersections and, therefore,
disproportionately involve side impacts.  (See Viano et al., 1990.)
Automobile weight offers less protection in side- impact collisions
than in frontal impacts. 


THE RELATIVE IMPORTANCE OF CRASH
FEATURES, DRIVER CHARACTERISTICS,
SAFETY BELTS, AND AUTOMOBILE
WEIGHT
============================================================ Chapter 4

As we demonstrated in chapters 2 and 3, crash severity, crash type,
automobile weight and wheelbase, safety belt use, and driver age and
gender, taken separately, each significantly influences the
probability of driver hospitalization or death.  For this chapter, we
also assessed the relative importance of these factors simultaneously
to see which ones are the most important predictors of injury in a
crash and which ones have relatively little influence. 

We found that crash severity is the most important predictor of
driver hospitalization or death, followed by crash type, safety belt
use, driver age and gender, and automobile weight.  Crash severity
refers to the speed of impact, while crash type refers to the number
of vehicles in a crash, whether the car rolled over, and its points
of impact.  If information about only one of these several factors
were available for predicting whether the driver would be seriously
injured, having access to crash severity information would lead to
the greatest number of accurate predictions.  If crash severity
information could not be obtained, information about the crash type
would give the best chance of accurately predicting whether or not
the driver would be injured.  And so on down the list of factors. 

Table 4.1 documents this finding.  It shows a statistical measure of
the "explanatory power" of each factor.  The table shows that the
largest value for this measure is for crash severity, followed by
crash type, and then the other factors in the order previously noted. 
The "explanatory power" of automobile weight is substantially less
than that of all the other factors. 



                          Table 4.1
           
             Relative Importance of Crash-Related
                 Factors in Predicting Driver
                  Hospitalization or Death\a

                       Change in
                             log    Degrees of
Factor                likelihood       freedom   Probability
------------------  ------------  ------------  ------------
Crash severity            110.70             3          0.01
Crash type                 71.75             5          0.01
Safety belts               38.88             3          0.01
Driver age and             26.04             4          0.01
 gender
Automobile weight           2.36             1          0.03
------------------------------------------------------------
\a Larger change in log likelihood values indicates the more
important explanatory factors.  The probability column shows that all
these factors are statistically significant.  The change in log
likelihood values was computed first by estimating the logistic
regression equation without the variables representing each factor
and then comparing the minus log likelihood values of those equations
with the minus log likelihood from the full model.  The minus log
likelihood for the full equation is 1,020.13 (see table I.1).  For
crash severity, two variables measure change in velocity at impact
and one measures the speed limit of the roadway section where the
crash occurred.  For crash type, five categorical variables represent
a one- or two-vehicle crash, a vehicle rollover, a head-on crash, and
front or left-side impact points.  Three categorical variables
measure driver age, and one indicates gender.  Automobile weight is a
continuous variable. 

Source:  GAO analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 

Another way to illustrate the great importance of the crash severity
and crash type factors is presented in figure 4.1.  Each column in
the figure shows the estimated increment in risk of injury associated
with a change in the associated crash-related factor, combining the
three types of crashes in our analysis.  Thus, the "crash severity"
bar in the figure shows that crashes involving a change in velocity
of 23 mph have an estimated risk of driver injury 25 times as great
as crashes with a change in velocity of only 6 mph.\1 The bar for
crash type shows our estimate that drivers involved in one-car
rollover crashes are about nine times more likely to be hurt than
drivers involved in collisions with cars and light trucks that are
not head-on crashes. 

Similarly, figure 4.1 shows that drivers 65 and older are about 4.5
times more likely to be hospitalized or killed than drivers 16 to 24
and that unbelted drivers have an injury risk more than three times
as great as drivers wearing manual lap and shoulder safety belts. 
Drivers of 2,000-pound automobiles have an estimated injury risk in a
crash that is about 1.63 times (or 63 percent greater than) that of
drivers of 3,600-pound cars.\2 Finally, the estimated injury risk for
women drivers is about 1.29 times (or 29 percent greater than) that
of men drivers. 

   Figure 4.1:  Odds of Injury
   Associated With Changes in
   Crash-Related Factors\a

   (See figure in printed
   edition.)

\a Estimates include all one-car crashes and collisions with cars and
light trucks.  Endpoints for each factor:

  Crash severity:  23 mph change in velocity versus baseline 6 mph
     change in velocity.

  Crash type:  drivers in one-car rollovers versus baseline drivers
     in collisions with cars and light trucks that are not head-on.

  Driver's age:  drivers 65+ versus baseline drivers 16-24.

  Safety belt use:  unbelted drivers versus baseline drivers wearing
     manual lap and shoulder safety belts.

  Automobile weight:  drivers of 2,000-lb cars versus baseline
     drivers of 3,600-lb cars.

  Driver's gender:  women versus baseline men. 

Source:  Our analysis of NASS 1988-91 data for model year 1987 and
newer automobiles. 


--------------------
\1 We used change in velocity values of 23 and 6 mph in this example
because those values are near the endpoints of the change in velocity
distribution.  Twenty-three mph is the 95th percentile of the change
in velocity distribution, while 6 mph is the 5th percentile. 

\2 Two thousand pounds represents the 5th percentile of the
distribution of automobile weights in this data set; 3,600 pounds is
the 95th percentile.

It is important to note that this finding about the injury risks to
individual drivers of automobiles of different weights does not mean
that the overall "downsizing" of automobiles over the past 20 years
has led to more total highway fatalities (see GAO, 1991). 


FUTURE AUTOMOBILE SAFETY
============================================================ Chapter 5

Here we discuss the implications of our findings for future
automobile safety.  The first section below reviews the safety
initiatives from NHTSA that have the greatest importance for
automobile crashworthiness.  The next section discusses ways to
reduce the injury risk for particular categories of automobile
drivers.  The last section discusses the most effective uses of
available safety technologies. 

It is important to keep two points in mind when considering
alternative approaches to automobile crashworthiness.  First,
crashworthy automobiles must offer as much protection as possible for
a broad matrix of crash types, crash speeds, and occupant
characteristics that pose very different occupant protection
problems.  For example, we found that one-car crashes, particularly
rollovers, are much more dangerous than collisions with cars and
light trucks.  We also found that men drivers and young drivers are
disproportionately involved in one-car crashes, while women drivers
and older drivers are more likely to be involved in collisions with
cars and light trucks.  Protecting young men in severe one-car
crashes is very different from protecting women and older drivers in
multivehicle collisions.  Second, individual safety features often
affect only one portion of the matrix of crash types and occupant
characteristics.  For example, air bags clearly help protect
occupants in frontal collisions, but they do not contribute to
occupant safety in side-impact collisions or rollover crashes. 


   NHTSA SAFETY INITIATIVES
---------------------------------------------------------- Chapter 5:1


      RECENT REGULATIONS
-------------------------------------------------------- Chapter 5:1.1


         FRONTAL IMPACTS
------------------------------------------------------ Chapter 5:1.1.1

Starting with the 1990 model year, all automobiles sold in the United
States have had to demonstrate driver and right- front-seat passenger
safety with passive restraints in a full-frontal crash at 30 mph into
a rigid barrier.\1 "Passive restraint" means without the use of any
safety device requiring actions by the driver or passenger, such as
manual safety belts.\2 In model year 1987, the first year of the
phase-in period for this regulation, all the automobiles NHTSA tested
met this requirement with automatic safety belts.  By 1993, almost
all the tested cars fulfilled the passive restraint requirement with
air bags rather than automatic belts alone, although many of the cars
with air bags also had automatic safety belts. 

NHTSA has announced major changes in this regulation.\3 All cars
manufactured in September 1997 or later will be required to have both
air bags and manual lap and shoulder safety belts for both drivers
and right-front-seat passengers.  Very importantly, the revised
regulation prohibits automatic safety belts--not just for use in the
compliance tests but as safety equipment.  All cars will have to be
equipped with manual safety belts. 


--------------------
\1 See Highway Safety:  Reliability and Validity of DOT Crash Tests
(GAO/PEMD-95-5). 

\2 Under this regulation, automobiles can be equipped with manual
safety belts, but the safety standards of the compliance test must be
met without benefit of the protection afforded by the manual belts. 

\3 Federal Register, September 2, 1993 (49 C.F.R.  571- 585). 


         SIDE IMPACTS
------------------------------------------------------ Chapter 5:1.1.2

Beginning with the 1994 model year, NHTSA began phasing in a
requirement for automobile occupant protection in side impacts.  By
model year 1997, all automobiles will have to meet safety standards
in crash tests simulating the impact of a 3,000-pound vehicle hitting
the target car in a side-impact collision at 33.5 mph.  Unlike the
frontal impact crash tests, active restraint systems, such as manual
safety belts, must be used in these tests. 


      OTHER ACTIVITIES
-------------------------------------------------------- Chapter 5:1.2

NHTSA is also undertaking a variety of efforts to deal with
particular mechanisms of occupant injury rather than points of
contact on the automobile.  For example, to reduce head injuries,
NHTSA is developing a regulation that would require energy-absorbing
padding in the areas of automobile interiors that occupants' heads
frequently strike in side-impact collisions.  Also, NHTSA is studying
ways to further reduce injuries in rollover crashes, primarily by
reducing the risk of ejection, by improving door latches and
increasing the strength of automobile windows other than windshields,
as well as by considering tougher roof crush standards. 

Other NHTSA activities are concerned with particular types of
automobile occupants, especially children and elderly persons.  It is
important to note that NHTSA is seeking ways to improve protection
for elderly drivers, although a major focus of NHTSA's work involves
programs to improve their driving skills or otherwise reduce their
likelihood of crash involvement.  (See Transportation Research Board,
1992, and NHTSA, 1993a; see also NHTSA 1992b for its activities
priority plan through 1994.)


   INJURY REDUCTION FOR DIFFERENT
   TYPES OF DRIVERS
---------------------------------------------------------- Chapter 5:2

Automobile manufacturers understand that different segments of the
consumer market for automobiles prefer different types of cars.  For
example, young men are likely to prefer sports cars over station
wagons, and older drivers disproportionately prefer large cars over
smaller ones.  In other words, in the marketplace for automobiles,
one size does not fit all.  Similarly, one size does not fit all when
it comes to automobile safety:  the crashworthiness problems of
different types of drivers, and of drivers involved in different
types of crashes, require a variety of different solutions.  Here, we
look at the differential safety concerns of segments of the safety
"marketplace" that are defined by safety belt use and driver age and
gender. 


      UNBELTED AND BELTED DRIVERS
-------------------------------------------------------- Chapter 5:2.1

Drivers involved in traffic crashes, on the whole, operate their
vehicles in a riskier manner than drivers who are not involved in
crashes.  For example, the rate of safety belt use for drivers
involved in crashes is less than the use rate for the general driving
population.  Estimates of the degree to which drivers who do not wear
safety belts are overinvolved in roadway crashes vary considerably. 
For example, NHTSA (1992a) estimated that unbelted drivers experience
potentially fatal crashes 2.2 times more than belted drivers, while
Hunter et al.  (1993) found that unbelted drivers had a crash
involvement rate 35 percent higher than belt users. 

Unbelted and belted drivers have very different injury experiences in
a crash.  Unbelted drivers are more likely to suffer severe injuries,
and their injuries are more likely to result from contact with the
steering wheel or windshield (Danner, Langieder, and Hummel, 1987;
Lestina et al., 1991).  These differences are explained by the
mechanisms of safety belt effectiveness.  Safety belts tie the
occupant to the car, helping the occupant decelerate over a
relatively long period.  In addition, by restricting movement, safety
belts reduce the chances of the wearer's striking the interior of the
vehicle and help make his or her course of motion within the car more
predictable.  In contrast, unbelted occupants keep moving within the
automobile in the moments after collision, the direction of their
movement within the vehicle is relatively unpredictable, and it is
likely either that their rapid motion will be abruptly stopped by
contact with a rigid surface within the vehicle or that they will be
ejected from it. 

Therefore, optimally safe vehicle interiors are conceptually
dissimilar for belted and unbelted occupants (Mackay, 1993).  For
belted occupants, the more interior space the better, as increasing
the space reduces the odds of contact with interior surfaces. 
Conversely, for unbelted occupants, the goal is to restrict movement
and provide a soft place to land, so heavily padded interiors that
minimize interior space are preferred. 

How can crash protection be improved for unbelted and belted drivers? 
For unbelted drivers, the obvious answer is to put them in safety
belts.  In practical terms, the best way to do this is to increase
the number of automobiles with automatic safety belts.  As NHTSA
(1992a) and Williams et al.  (1992) have reported, automobiles
equipped with automatic safety belts have much higher belt usage
rates than those with manual belts.  While experimental vehicles have
been designed with substantial protection for unbelted occupants, we
do not believe that any combination of interior padding, air bags,
and other passive restraint systems will be able to rival the
effectiveness of safety belts in production automobiles for the
foreseeable future.  One reason for this is that, as noted above,
designing an optimally safe car for unrestrained drivers may require
abandoning safety belts as the centerpiece of occupant protection
strategies.  And safety belts are extraordinarily effective; alone,
they are much more effective at reducing serious injuries than are
air bags alone. 

For belted drivers, the prospects for dramatic improvements in crash
protection are less obvious.  On the one hand, promising efforts are
under way to reduce much of the residual risk of injury confronting
belted drivers.  These include improvements in safety belt
technology, the greater availability of air bags, and NHTSA's efforts
to improve occupant protection in side impacts.  On the other hand,
the great success of recent occupant protection efforts means that
further crashworthiness improvements are harder to achieve, primarily
because the dwindling proportion of crashes that still cause serious
injury and death to belted occupants are exceptionally severe events. 
For example, Mackay et al.  (1992), reviewing a sample of crashes
involving the death of restrained front-seat occupants in Britain,
found that the deaths occurred in extremely severe crashes.  Fifty
percent of the deaths in frontal crashes were in collisions with
large trucks, and 86 percent involved passenger compartments crushed
so severely as to eliminate the space occupied by the fatally injured
person before the crash.  Similarly, Green et al.  (1994) reported
that most of the fatalities of restrained occupants that they
examined involved severe intrusion into the passenger compartment and
multiple injuries so severe that 90 percent of the victims died
within an hour of the crash. 


      DRIVER AGE AND GENDER
-------------------------------------------------------- Chapter 5:2.2

The "market segments" for automobile safety defined by driver age and
gender require very different strategies for reducing fatalities. 
For men drivers and younger drivers, the problem is crash
involvement, not crashworthiness.  As we demonstrated in chapter 2,
compared to women and older drivers, not only are men drivers and
younger drivers involved in more automobile crashes but also the
crashes they are particularly likely to be involved in have
comparatively severe consequences-- that is, single-car crashes have
much higher driver injury rates than multivehicle crashes.  However,
as we saw in chapter 3, men drivers and younger drivers are
significantly less likely to be hurt in a crash than women and older
drivers.  That is, men and younger drivers benefit from a degree of
occupant protection that is not available to women and older drivers
(we will discuss some of the reasons later).  In summary, the surest
way to improve the safety of men drivers and younger drivers is to
attempt to reduce their crash involvement rates, particularly their
rates of involvement in single-car crashes. 

The situation is exactly the reverse for women and older drivers. 
The problem for them is crashworthiness, not crash involvement. 
Compared to men and younger drivers, women and older drivers are
involved in fewer automobile crashes, and the crashes they are
involved in are, on the average, less severe, since they are less
likely to be involved in single-car crashes than in multivehicle
collisions.  However, once a crash has occurred, women and,
especially, older drivers are more likely to be hospitalized or
killed. 

In our judgment, improving the crash protection offered by
automobiles to women and older drivers so that it approaches the
level enjoyed by men and younger drivers offers the greatest chance
for reducing roadway injuries for them.\4 In the absence of
compelling evidence for the inherent physical frailty of women
compared to men, we are optimistic that crashworthiness for women can
be substantially improved.  In contrast, the evidence we have
reviewed indicates that older drivers are, in fact, more fragile than
young drivers.  Nonetheless, we believe that older drivers can be
afforded better protection by automobiles than they now receive (see
subsequent discussion and Mackay, 1988).  It is important that
occupant protection for women drivers and older drivers be improved
without compromising the crash protection of men drivers and younger
drivers; design changes that merely shift injury risk from one group
of drivers to another will not improve traffic safety in the
aggregate. 

One possible reason for the relatively high degree of crash
protection enjoyed by men drivers and younger drivers is that efforts
at improving automobile crashworthiness have concentrated on the
crash types and occupant characteristics most often experienced by
them.  Current safety regulations and automobile safety designs
emphasize protection in high-speed frontal collisions, and men
drivers and younger drivers are more likely to be in single-car
crashes, which disproportionately involve frontal impacts.  The
automobile crash tests NHTSA currently requires for all cars include
full-frontal crashes into a rigid barrier at 30 mph (although the
introduction of a requirement for side-impact tests is under way). 
Air bags reduce the risk of injury in frontal impacts only, not in
side impacts.  Similarly, safety belts are more effective in frontal
than in side impacts (for example, Evans, 1990), and because of this,
safety belts have a somewhat greater benefit in single-car crashes
than in collisions with cars and light trucks. 

A second possible reason for the crashworthiness deficit of women
drivers compared with men drivers is that current NHTSA regulations
require the use of only one size of crash test dummy--a dummy
representing the 50th percentile of the male population, or 5 feet 9
inches tall, weighing 165 pounds.  Maximizing the safety of persons
with these characteristics may, in a relative sense, compromise the
safety of others. 

Another possible explanation for the greater injury risk for women
drivers is that, on the average, women are shorter and lighter than
men.  Automobiles designed to accommodate taller and heavier men
drivers may not accommodate women as well.  For example, the
Insurance Institute for Highway Safety (1993) recommends that drivers
sit back as far as possible from the steering wheel and dashboard in
order to minimize the risk of hitting those structures in a crash. 
Shorter drivers obviously cannot sit as far back as taller drivers if
they hope to reach the accelerator and brake pedals, and this may
expose them to more risk. 


--------------------
\4 This does not mean that efforts to reduce the crash involvement
rates of women drivers and older drivers should not continue, but it
does mean that the predominant focus should be on improving
crashworthiness.  This is particularly true for women drivers;
notwithstanding recent calls to focus crash prevention programs on
women drivers (for example, Centers for Disease Control, 1992), men
drivers have much higher crash involvement rates than women drivers
(GAO, 1994).  In recent years, men drivers have been involved in more
than three times as many fatal crashes as women drivers.  For older
drivers, efforts to understand, measure, and control the
deterioration of physical and cognitive abilities that can impair
driving skills are obviously important, but the crash involvement
risk of older drivers is already reduced because they drive so many
fewer miles than younger drivers (Evans, 1988a; "As Nation Grows
Older," 1992; NHTSA, 1993a). 


   MAKING EFFECTIVE USE OF
   AVAILABLE SAFETY TECHNOLOGIES
---------------------------------------------------------- Chapter 5:3


      ENHANCED SAFETY BELTS
-------------------------------------------------------- Chapter 5:3.1

All safety belts are not equally effective.  In particular, many cars
on the market today have safety belts with automatic pretensioners or
web locking devices that substantially improve their effectiveness
(IIHS, 1993).  Pretensioners work by reducing the amount of slack in
the belts or by tightening them in a crash a fraction of a second
sooner.  They cause the belted occupant to begin decelerating sooner
in a crash, thereby increasing the total deceleration period.  In
addition, they increase the chances that the occupant's forward
motion will be stopped before he or she contacts the interior of the
automobile. 

To give an idea of the magnitude of the safety increment available
from belts with these features, Viano (1988) compared the performance
of several restraint mechanisms in frontal crash tests.  Depending on
the outcome measure used, lap and shoulder belts with pretensioners
had injury scores about 15 to 40 percent below those of lap and
shoulder belts without pretensioners. 


      NHTSA'S IMPLEMENTATION OF
      THE REQUIREMENT FOR AIR BAGS
-------------------------------------------------------- Chapter 5:3.2

NHTSA has recently announced regulations that would implement the
requirement in the Intermodal Surface Transportation Efficiency Act
of 1991 that all passenger cars and light trucks be equipped with air
bags and lap and shoulder safety belts.  Beginning with all cars
manufactured in September 1997, both drivers and right- front-seat
passengers will have both air bags and manual lap and shoulder safety
belts; automatic safety belts are prohibited. 

We are concerned that NHTSA's implementation of the requirement for
air bags may not achieve the greatest degree of improvement in the
aggregate safety of the population of automobile occupants. 
Automobile occupants who travel in cars with air bags and who wear
manual lap and shoulder safety belts will be well protected. 
However, because safety belts alone offer much more protection than
air bags alone, occupants of air bag-equipped cars who do not wear
lap and shoulder safety belts will be less well protected than if
they were traveling in cars with automatic safety belts.  This is
important because cars with automatic safety belts have higher safety
belt usage rates than cars with manual belts, and individuals
involved in serious automobile crashes have lower safety belt use
rates than others.  If many automobile occupants in serious crashes
do not wear manual safety belts, the aggregate safety of automobile
occupants under NHTSA's proposal would be less than if, in addition
to air bags, automatic safety belts were encouraged or required. 

To examine this question, we compared the average amount of occupant
protection available to all automobile occupants under three
different safety-belt-use scenarios based on a recent NHTSA report
(NHTSA, 1992a).  In that report, NHTSA noted that cars equipped with
manual lap and shoulder belts had a belt usage rate of 56 percent in
1991, while cars equipped with automatic safety belts had usage rates
ranging from 64 to 97 percent, depending on the type of automatic
belt.\5 If all cars had air bags and manual lap and shoulder belts
and a belt usage rate of 56 percent, we estimate that fatality risk
would fall 37.2 percent for the average automobile occupant compared
with unprotected occupants.\6 If all cars had air bags and automatic
safety belts and a belt usage rate of 64 percent, we estimate that
the average automobile occupant would have a 37.7- percent reduction
in fatality risk.\7 If all cars had air bags and automatic safety
belts and a belt usage rate of 97 percent, we estimate that the total
fatality risk reduction would be 46 percent.\8

Thus, from the standpoint of maximizing the aggregate safety of all
automobile occupants, the best proposal may be one that requires both
air bags and automatic lap and shoulder safety belts.  The magnitude
of the fatality risk reduction arising from that configuration
compared to NHTSA's regulation requiring manual lap and shoulder
safety belts depends on the difference between the usage rates of
automatic and manual safety belts for automobile occupants involved
in serious crashes.  As our estimates show, if that difference is
small, the automatic safety belt alternative offers only a very
slight aggregate safety improvement.  Conversely, if the usage rate
difference is high, placing air bags and automatic lap and shoulder
safety belts in all cars would substantially improve the safety of
automobile occupants in the aggregate. 


--------------------
\5 For the following calculations, we used figures reported by NHTSA. 
We assumed that unbelted drivers are 2.2 times more likely than
belted drivers to be involved in serious crashes (NHTSA, 1992a, p. 
21), that manual lap and shoulder belts reduce the risk of fatality
by 45 percent (1993d, p.  II-13), that automatic safety belts reduce
fatality risk by 42.5 percent (1993d, p.  II-13), that air bags alone
reduce fatality risk by 30 percent, and that air bags with manual lap
and shoulder belts reduce the risk of fatality by 50 percent (1993d,
p.  II-13).  Assuming that air bags improve the lot of belted
occupants by about 11 percent (as NHTSA estimates for manual lap and
shoulder belts), we estimated for this analysis that air bags with
automatic safety belts reduce fatality risk by 47 percent (42.5 x
1.11 = 47). 

\6 If 56 percent of all drivers wore manual lap and shoulder safety
belts, then about 37 percent of drivers in crashes would be belted. 
Those drivers would have a fatality risk reduction of 50 percent,
while the remaining 63 percent of crash-involved drivers--that is,
those not wearing safety belts--would have a fatality risk reduction
of 30 percent, for an overall fatality risk reduction of 37.2
percent. 

\7 If 64 percent of all drivers wore automatic lap and shoulder
safety belts, then about 45 percent of drivers in crashes would be
belted.  Those drivers would have a fatality risk reduction of 47
percent, while the remaining 55 percent of crash-involved
drivers--that is, those not wearing safety belts--would have a
fatality risk reduction of 30 percent, for an overall fatality risk
reduction of 37.7 percent. 

\8 If 97 percent of all drivers wore automatic lap and shoulder
safety belts, then about 94 percent of drivers in crashes would be
belted.  Those drivers would have a fatality risk reduction of 47
percent, while the remaining 6 percent of crash-involved
drivers--that is, those not wearing safety belts--would have a
fatality risk reduction of 30 percent, for an overall fatality risk
reduction of 46.1 percent. 


   AGENCY COMMENTS AND OUR
   EVALUATION
---------------------------------------------------------- Chapter 5:4

DOT had two comments regarding the implications of our finding that,
holding constant crash characteristics and automobile weight, women
are more likely than men to suffer serious injury in a crash.  The
first is that NHTSA plans to conduct crash tests with test dummies of
different sizes rather than only the standard dummy that represents a
50th percentile man driver.  The second is that NHTSA has recently
made final a rule requiring improvements in the adjustability of
safety belts that may increase the percentage of vehicle occupants
using belts correctly.  We applaud both these developments. 
Nonetheless, in our opinion, there is no definitive evidence that
either size differences or patterns of safety belt use fully account
for the differences in injury rates between men and women. 

DOT also had two comments on our discussion of NHTSA's implementation
of the requirement for air bags.  First, it contended that the usage
figures for automatic safety belts that we used in our example are
unrealistically high.  More specifically, DOT stated that the usage
rates for complete automatic belt systems are much less than the
97-percent scenario we described, since some drivers use only one
component but not the other (for example, using the shoulder belt but
not the lap belt) and other drivers disconnect the automatic belt
system entirely.  Second, DOT disagreed with our conclusion that
manual and automatic safety belts provide "a roughly equivalent
degree of protection."

For the first point, we understand that it is extremely difficult to
accurately measure safety belt use, especially the use of particular
safety belt components (see chapter 2).  However, NHTSA (1992a) has
concluded that automatic safety belts are used more often than manual
belts, and our 97-percent usage rate scenario was based on a NHTSA
report, not on our own analysis.  Further, our findings would not
differ even if automatic safety belts had usage rates much less than
97 percent; thus, we found that the scenario with a 64-percent usage
rate for automatic belts (NHTSA's lowest estimate) still provided
slightly more total protection than the other scenario we considered,
manual belts with a 56-percent usage rate.  For the second point, our
conclusion that manual and automatic safety belts provide
approximately equivalent protection is based on NHTSA's work, not on
our own analyses of automobile crash data.  For example, NHTSA
(1993d, p.II-13) estimated that, when used in a crash, manual lap and
shoulder safety belts reduce fatality risk by 45 percent and
automatic three-point belts reduce fatality risk by 42.5 percent. 
Similarly, NHTSA earlier reported that it was unable to find any
statistically significant differences between several different
configurations of manual and automatic safety belts (NHTSA, 1992a,
p.66). 

Most importantly, neither of DOT's comments about our discussion of
NHTSA's implementation of the requirement for air bags addressed our
main point--that drivers involved in serious crashes use safety belts
much less than the general driving population.  Our discussion is
aimed at improving crash protection for drivers who have the greatest
risk of involvement in serious crashes.  A comprehensive evaluation
of the best ways to increase safety belt use for those drivers is
beyond the scope of this report.  However, as our analysis
demonstrates, NHTSA's decision to prohibit automatic safety belts may
not achieve the best result from that perspective.  At a minimum,
NHTSA needs to continue to emphasize in its public education efforts
the importance of wearing safety belts even in cars equipped with air
bags. 


REGRESSION ANALYSES FOR DRIVER
INJURY
=========================================================== Appendix I


   THE DATA SET
--------------------------------------------------------- Appendix I:1

We compiled the data set for our analyses from the National Accident
Sampling System--Crashworthiness Data System (NASS) for 1988 through
1991.  NASS is a nationally representative probability sample of all
police-reported crashes involving a passenger car, light truck, or
van in which at least one vehicle was towed from the scene.  For more
information about NASS, including the sampling frame, sampling plan,
and variable definitions, see NHTSA (1991c, 1991d). 

For our data set, we selected the subset of cases from NASS that
included all one-car crashes involving 1987 or later model year
automobiles and all collisions between a model year 1987 or later
automobile and any other car, van, pickup truck, or light truck.  We
discarded automobiles that had no valid curb weight information or no
identifiable driver or a driver younger than 16 years of age.  The
final data set included 6,103 cases of eligible automobiles and their
drivers:  457 in one-car rollover crashes, 1,253 in one-car
nonrollover crashes, and 4,393 in collisions with other cars or light
trucks. 


      THE OUTCOME VARIABLE
------------------------------------------------------- Appendix I:1.1

The outcome variable in our analyses was a dichotomous variable for
driver injury coded "1" if the driver was hospitalized or killed in
the crash and coded "0" otherwise.  We chose this outcome measure
because it is easy to understand and unambiguous.  This variable was
created from the "treatment" variable in the NASS file.  In NASS,
occupants were coded as hospitalized if they were admitted to a
hospital for an overnight stay and coded as killed if they died
within 30 days of the crash from injuries sustained in the crash. 

To relate this measure of driver hospitalization or death to other
measures of injury severity, our measure overlaps with injuries
categorized as moderate to serious by the Abbreviated Injury Scale
(AIS).  (See Evans, 1991b.) Thus, we classified 19 percent of the
drivers in this data set as hospitalized or killed.  Had we used the
AIS scheme, 24 percent of all drivers would have been classified as
suffering a moderate or more severe injury (AIS 2-6) and 10 percent
as suffering a serious or more severe injury (AIS 3-6).\1 For the
drivers we coded as hospitalized or killed, 18 percent were killed,
33 percent had an injury listed as serious or critical on AIS (AIS
3-6), 33 percent had a moderate injury on AIS (AIS 2), and 16 percent
had minor or no injuries according to the AIS scale. 


--------------------
\1 All the percentages in this paragraph are unweighted. 


      THE PREDICTOR VARIABLES
------------------------------------------------------- Appendix I:1.2

For independent variables in the regressions, we used a number of
factors measuring crash type, crash severity, automobile
characteristics, driver safety belt use, and driver age and gender. 
All the independent variables are listed below.  As the following
regression tables show, not all these variables were used in all the
regression analyses.  The estimated population values for the
variables are listed in the tables. 


         CRASH TYPE
----------------------------------------------------- Appendix I:1.2.1

  One-car crash was coded "1" if only one car was involved in the
     crash and coded "0" if the crash was a two-vehicle collision. 

  Rollover was coded "1" if the automobile rolled over (either as the
     primary crash event or as a secondary event) and coded "0"
     otherwise. 

  Head-on collision was coded "1" if the crash was a two-vehicle
     collision between vehicles moving in opposite directions and
     coded "0" otherwise. 

  Front area damage to the car was coded "1" if the crash was not
     head-on yet the car sustained frontal damage and coded "0"
     otherwise. 

  Left side damage to the car was coded "1" if the crash was not
     head-on and the car sustained left-side damage and coded "0"
     otherwise. 


         CRASH SEVERITY
----------------------------------------------------- Appendix I:1.2.2

  Speed limit 40 mph or greater was coded "1" if the speed limit at
     the roadway location of the crash was 40 mph or greater and
     coded "0" otherwise. 

  Has change in velocity value was coded "1" if the automobile had a
     valid change in velocity value in NASS and coded "0" otherwise. 
     (Many automobiles had missing values for change in velocity in
     NASS; in particular, all the cars in one-car rollover crashes
     had missing values.)

  Change in velocity was measured in miles per hour.  (Change in
     velocity refers to the nearly instantaneous change in a
     vehicle's speed that occurs in a crash.  For example, a vehicle
     that was abruptly stopped from a travel speed of 30 mph would
     have a change in velocity of 30 mph.) Automobiles coded "0" on
     the "has change in velocity value" variable were assigned the
     mean value of the change in velocity scores for each analysis. 


         AUTOMOBILE
         CHARACTERISTICS
----------------------------------------------------- Appendix I:1.2.3

  Model year was indexed by three categorical variables for model
     years 1987, 1988, and 1989.  Cars from those model years were
     coded "1" for the appropriate year and "0" for the other years. 
     Automobiles from the 1990 model year and later were coded "0"
     for all three variables. 

  Curb weight was measured in 100-pound increments in NASS.  Curb
     weight refers to the weight of the unoccupied automobile,
     including gasoline and other fluids. 

  Wheelbase was measured in inches.  (Wheelbase is a measure of
     automobile length that is one indication of the exterior size of
     a vehicle.  Wheelbase refers to the distance between the front
     and rear axles.) Many of the cars in the data set had missing
     wheelbase values in NASS.  We were able to assign valid values
     to most of them by matching the missing cases to cars that had
     valid wheelbase values, based on model year, make, model, and
     body type.  Nonetheless, slightly fewer cars had valid wheelbase
     values than had valid curb weight values in our final data set. 


         DRIVER SAFETY BELT USE
----------------------------------------------------- Appendix I:1.2.4

  Manual lap and shoulder belt use was coded "1" if the driver was
     noted by NASS researchers as wearing manual lap and shoulder
     safety belts (a "3-point belt") at the time of crash and coded
     "0" otherwise. 

  Automatic belt use was coded "1" if the driver was noted by NASS
     researchers as wearing an automatic safety belt (a 3-point belt,
     a lap belt only, or a shoulder belt only) at the time of the
     crash and coded "0" otherwise. 

  Both manual and automatic belt use was coded "1" if the driver was
     noted by NASS researchers as wearing both an automatic shoulder
     belt and a manual lap belt at the time of the crash and coded
     "0" otherwise. 


         DRIVER AGE AND GENDER
----------------------------------------------------- Appendix I:1.2.5

  Driver age was indexed with four categorical variables representing
     ages 16 to 24 years, 25 to 44, 45 to 64, and 65 and older. 
     Drivers were coded "1" for the appropriate age category and "0"
     for the other categories. 

  Driver gender was coded "1" for men and "0" for women. 


   RESULTS OF THE REGRESSION
   ANALYSES
--------------------------------------------------------- Appendix I:2

Tables I.1 through I.8 present the results of our regression
analyses.  The eight tables show the analyses for four sets of
crashes (all crashes included in the data set, one-car rollover
crashes, one-car nonrollover crashes, and collisions with cars and
light trucks), using two measures of automobile size (curb weight and
wheelbase). 



                                    Table I.1
                     
                       One-Car Crashes and Collisions With
                       Other Cars or Light Trucks: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                   Curb Weight

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.188     1.207      0.75        0.39        0.31
1988                          -0.024     0.976      0.01        0.91        0.29
1989                           0.205     1.227      0.71        0.40        0.21
Speed limit 40+ mph            0.459     1.582      9.76        0.01        0.50
One-car crash (versus          0.915     2.497     38.37        0.01        0.26
 two-vehicle crash)
Rollover                       1.291     3.637     36.35        0.01        0.06
Head-on collision              1.159     3.188     24.65        0.01        0.03

Area of damage to the car (if not head-on collision)
--------------------------------------------------------------------------------
Front                          0.521     1.684      8.45        0.01        0.60
Left side                      0.690     1.994     17.38        0.01        0.21

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.216     0.297     51.77        0.01        0.49
 belts
Automatic belts               -0.769     0.464     17.14        0.01        0.13
Both manual and               -1.002     0.367     20.48        0.01        0.11
 automatic belts
Male driver                   -0.251     0.778      4.18        0.04        0.50
Driver age in years
 (versus 16-24)
25-44                          0.542     1.720      9.19        0.01        0.43
45-64                          0.940     2.559     33.72        0.01        0.14
65+                            1.509     4.521     54.35        0.01        0.07

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.833     2.300     35.01        0.01        0.35
 value
Change in velocity (mph)       0.191     1.211    116.89        0.01       14.55
Automobile weight (100s       -0.030     0.970      6.98        0.01       26.64
 of pounds)
Constant                      -6.447
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile weight by 100 pounds decreases the odds of hospitalization
or death by a factor of 0.970, or 3 percent.  The odds ratio for
categorical variables indicates the change from the left-out group. 
For example, the injury odds for men drivers are 0.778 that of women
drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.2
                     
                       One-Car Crashes and Collisions With
                       Other Cars or Light Trucks: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                    Wheelbase

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.210     1.233      0.90        0.34        0.31
1988                          -0.029     0.971      0.02        0.89        0.29
1989                           0.254     1.289      1.03        0.31        0.21
Speed limit 40+ mph            0.456     1.577     10.05        0.01        0.50
One-car crash (versus          0.873     2.395     34.11        0.01        0.26
 two-vehicle crash)
Rollover                       1.306     3.693     33.75        0.01        0.06
Head-on collision              1.102     3.010     22.12        0.01        0.03

Area of damage to the car (if not head-on collision)
--------------------------------------------------------------------------------
Front                          0.542     1.719      8.72        0.01        0.60
Left side                      0.675     1.964     16.04        0.01        0.21

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.240     0.289     53.03        0.01        0.49
 belts
Automatic belts               -0.763     0.466     16.77        0.01        0.13
Both manual and               -1.048     0.351     22.28        0.01        0.11
 automatic belts
Male driver                   -0.262     0.769      4.54        0.03        0.50
Driver age in years
 (versus 16-24)
25-44                          0.565     1.760      9.65        0.01        0.43
45-64                          0.966     2.627     34.93        0.01        0.14
65+                            1.609     4.997     51.73        0.01        0.07

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.817     2.264     31.53        0.01        0.35
 value
Change in velocity (mph)       0.192     1.212    115.64        0.01       14.55
Automobile wheelbase          -0.028     0.973      9.24        0.01      101.07
 (inches)
Constant                      -4.456
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile wheelbase by 1 inch decreases the odds of hospitalization
or death by a factor of 0.973, not quite 3 percent.  The odds ratio
for categorical variables indicates the change from the left-out
group.  For example, the injury odds for men drivers are 0.769 that
of women drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Cotingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.) Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.3
                     
                        One-Car Rollover Crashes: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                   Curb Weight

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.468     1.596      0.78        0.38        0.31
1988                          -0.124     0.883      0.03        0.86        0.35
1989                           0.458     1.581      0.55        0.46        0.18
Speed limit 40+ mph            1.111     3.038      8.50        0.01        0.75

Area of damage to the car
--------------------------------------------------------------------------------
Front                          1.202     3.326     18.17        0.01        0.32
Left side                      0.108     1.114      0.05        0.82        0.16

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.775     0.169     18.89        0.01        0.44
 belts
Automatic belts               -0.684     0.505      2.05        0.15        0.14
Both manual and               -0.806     0.447      1.91        0.17        0.10
 automatic belts
Male driver                   -0.040     0.961      0.01        0.93        0.63

Driver age in years (versus 65+)
--------------------------------------------------------------------------------
16-24                          0.114     1.121      0.02        0.89        0.54
25-44                          0.350     1.419      0.17        0.68        0.36
45-64                          0.920     2.508      1.15        0.29        0.07
Automobile weight (100s        0.092     1.097      4.44        0.04       25.02
 of pounds)
Constant                      -5.110
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile weight by 100 pounds increases the odds of hospitalization
or death by a factor of 1.097, not quite 10 percent.  The odds ratio
for categorical variables indicates the change from the left-out
group.  For example, the injury odds for men drivers are 0.961 that
of women drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.4
                     
                      One-Car Nonrollover Crashes: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                   Curb Weight

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                          -0.220     0.803      0.34        0.56        0.32
1988                          -0.366     0.694      0.57        0.45        0.31
1989                           0.050     1.051      0.01        0.90        0.21
Speed limit 40+ mph           -0.006     0.994      0.00        0.98        0.59

Area of damage to the car
--------------------------------------------------------------------------------
Front                          0.502     1.652      1.83        0.18        0.67
Left side                      0.346     1.414      1.17        0.28        0.15

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.624     0.197     35.85        0.01        0.48
 belts
Automatic belts               -0.816     0.442      3.35        0.07        0.12
Both manual and               -1.027     0.358      4.94        0.03        0.12
 automatic belts
Male driver                    0.059     1.060      0.04        0.85        0.58

Driver age in years (versus 16-24)
--------------------------------------------------------------------------------
25-44                          0.777     2.174      7.56        0.01        0.42
45-64                          0.609     1.839      3.03        0.08        0.11
65+                            1.889     6.611     11.19        0.01        0.04
Automobile weight (100s       -0.032     0.968      2.00        0.16       26.88
 of pounds)

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.927     2.527      7.79        0.01        0.18
 value
Change in velocity (mph)       0.200     1.221     23.24        0.01       13.98
Constant                      -5.604
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile weight by 100 pounds decreases the odds of hospitalization
or death by a factor of 0.968, about 3 percent.  The odds ratio for
categorical variables indicates the change from the left-out group. 
For example, the injury odds for men drivers are 1.060 that of women
drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.5
                     
                      Collisions With Cars and Light Trucks:
                     Logistic Regression Analysis Predicting
                      Driver Hospitalization or Death, Model
                                 With Curb Weight

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.263     1.301      0.83        0.36        0.31
1988                           0.021     1.021      0.01        0.93        0.28
1989                           0.191     1.211      0.38        0.54        0.21
Speed limit 40+ mph            0.667     1.948     14.74        0.01        0.46
Head-on collision              1.642     5.166     22.30        0.01        0.05

Area of damage to the car (if not head-on collision)
--------------------------------------------------------------------------------
Front                          0.273     1.314      1.27        0.26        0.55
Left side                      0.874     2.396      9.90        0.01        0.23

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.029     0.357     23.48        0.01        0.49
 belts
Automatic belts               -0.654     0.520      8.24        0.01        0.13
Both manual and               -0.989     0.372     10.27        0.01        0.11
 automatic belts
Male driver                   -0.419     0.658      6.07        0.01        0.46

Driver age in years (versus 16-24)
--------------------------------------------------------------------------------
25-44                          0.365     1.441      2.31        0.13        0.43
45-64                          0.920     2.509     23.02        0.01        0.15
65+                            1.420     4.139     38.59        0.01        0.08

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.969     2.635     35.60        0.01        0.41
 value
Change in velocity (mph)       0.195     1.215     95.11        0.01       13.94
Automobile weight (100s       -0.054     0.948     12.85        0.01       26.69
 of pounds)
Weight of other vehicle        0.025     1.025      3.47        0.06       30.60
 (100s of pounds)

Body type of other vehicle (versus automobiles)
--------------------------------------------------------------------------------
Pickup truck                   0.166     1.181      0.62        0.43        0.13
Van                            0.803     2.231      3.54        0.06        0.05
Other light vehicle            1.000     2.720      6.82        0.01        0.04
Constant                      -6.784
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile weight by 100 pounds decreases the odds of hospitalization
or death by a factor of 0.948, about 5 percent.  The odds ratio for
categorical variables indicates the change from the left-out group. 
For example, the injury odds for men drivers are 0.658 that of women
drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.6
                     
                        One-Car Rollover Crashes: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                    Wheelbase

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.365     1.440      0.52        0.47        0.31
1988                          -0.182     0.833      0.07        0.79        0.35
1989                           0.386     1.471      0.39        0.53        0.18
Speed limit 40+ mph            1.155     3.173     10.12        0.01        0.75

Area of damage to the car
--------------------------------------------------------------------------------
Front                          1.247     3.480     17.72        0.01        0.32
Left side                      0.142     1.152      0.09        0.76        0.16

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.795     0.166     15.93        0.01        0.44
 belts
Automatic belts               -0.683     0.505      2.11        0.15        0.14
Both manual and               -0.915     0.400      2.91        0.09        0.10
 automatic belts
Male driver                   -0.017     0.983      0.00        0.97        0.63

Driver age in years (versus 65+)
--------------------------------------------------------------------------------
16-24                         -0.315     0.730      0.14        0.70        0.54
25-44                          0.071     1.073      0.01        0.93        0.36
45-64                          0.707     2.027      0.74        0.39        0.07
Automobile wheelbase           0.020     1.020      0.34        0.56       98.83
 (inches)
Constant                      -4.398
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile wheelbase by 1 inch increases the odds of hospitalization
or death by a factor of 1.020, or 2 percent.  The odds ratio for
categorical variables indicates the change from the left-out group. 
For example, the injury odds for men drivers are 0.983 that of women
drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.7
                     
                      One-Car Nonrollover Crashes: Logistic
                      Regression Analysis Predicting Driver
                       Hospitalization or Death, Model With
                                    Wheelbase

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                          -0.147     0.863      0.15        0.70        0.32
1988                          -0.298     0.743      0.39        0.53        0.31
1989                           0.162     1.176      0.13        0.71        0.21
Speed limit 40+ mph           -0.020     0.980      0.00        0.95        0.59

Area of damage to the car
--------------------------------------------------------------------------------
Front                          0.574     1.775      2.54        0.11        0.67
Left side                      0.322     1.380      1.00        0.32        0.15

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.640     0.194     38.01        0.01        0.48
 belts
Automatic belts               -0.798     0.450      3.30        0.07        0.12
Both manual and               -1.058     0.347      5.56        0.02        0.12
 automatic belts
Male driver                    0.025     1.025      0.01        0.93        0.58

Driver age in years (versus 16-24)
--------------------------------------------------------------------------------
25-44                          0.830     2.293      9.19        0.01        0.42
45-64                          0.661     1.938      3.78        0.05        0.11
65+                            1.977     7.220     12.27        0.01        0.04
Automobile wheelbase          -0.020     0.980      1.38        0.24      100.53
 (inches)

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.907     2.475      7.31        0.01        0.18
 value
Change in velocity (mph)       0.200     1.221     23.70        0.01       13.98
Constant                      -4.588
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile wheelbase by 1 inch decreases the odds of hospitalization
or death by a factor of 0.980, or 2 percent.  The odds ratio for
categorical variables indicates the change from the left-out group. 
For example, the injury odds for men drivers are 1.025 that of women
drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Tests for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 



                                    Table I.8
                     
                      Collisions With Cars and Light Trucks:
                     Logistic Regression Analysis Predicting
                      Driver Hospitalization or Death, Model
                                  With Wheelbase

                                                          Probabilit
                          Coefficien      Odds      Chi-           y  Population
Variable                         t\a   ratio\b  square\c     level\d      mean\e
------------------------  ----------  --------  --------  ----------  ----------
Automobile model year (versus 1990 and newer)
--------------------------------------------------------------------------------
1987                           0.284     1.329      1.01        0.31        0.31
1988                           0.001     1.000      0.00        1.00        0.28
1989                           0.242     1.273      0.59        0.44        0.21
Speed limit 40+ mph            0.653     1.921     13.97        0.01        0.46
Head-on collision              1.589     4.900     21.13        0.01        0.05

Area of damage to the car (if not head-on collision)
--------------------------------------------------------------------------------
Front                          0.286     1.331      1.39        0.24        0.55
Left side                      0.848     2.335      9.42        0.01        0.23

Safety belt use (versus unbelted)
--------------------------------------------------------------------------------
Manual lap and shoulder       -1.056     0.348     23.44        0.01        0.49
 belts
Automatic belts               -0.631     0.532      7.39        0.01        0.13
Both manual and               -1.043     0.353     11.01        0.01        0.11
 automatic belts
Male driver                   -0.445     0.641      6.77        0.01        0.46

Driver age in years (versus 16 to 24)
--------------------------------------------------------------------------------
25-44                          0.375     1.455      2.45        0.12        0.43
45-64                          0.902     2.464     20.98        0.01        0.15
65+                            1.499     4.476     37.71        0.01        0.08

Collision severity: change in velocity
--------------------------------------------------------------------------------
Has change in velocity         0.965     2.624     34.19        0.01        0.41
 value
Change in velocity (mph)       0.196     1.217     93.85        0.01       13.94
Automobile wheelbase          -0.043     0.958      9.56        0.01      101.37
 (inches)
Weight of other vehicle        0.026     1.026      3.58        0.06       30.60
 (100s of pounds)

Body type of other vehicle (versus automobiles)
--------------------------------------------------------------------------------
Pickup truck                   0.185     1.203      0.75        0.39        0.13
Van                            0.845     2.327      4.05        0.04        0.05
Other light vehicle            0.957     2.603      6.16        0.01        0.04
Constant                      -3.899
--------------------------------------------------------------------------------
\a Coefficients are from logistic regression analyses conducted with
the SUDAAN software package.  The outcome variable is
dichotomous--that is, "1" indicates that the driver was hospitalized
or killed, "0" that the driver was neither hospitalized nor killed. 

\b The odds ratio is the exponentiated coefficient (e\coefficient ). 
The odds ratio indicates the change in the odds of injury that occur
with a change of one unit in the variable.  For example, increasing
automobile wheelbase by 1 inch decreases the odds of hospitalization
or death by a factor of 0.958, just over 4 percent.  The odds ratio
for categorical variables indicates the change from the left-out
group.  For example, the injury odds for men drivers are 0.641 that
of women drivers, the left-out group. 

\c The chi-square values test the statistical significance of the
coefficients.  The values are calculated from the Satterthwaite
approximation to the chi-square distribution.  This procedure reduces
the chances of a Type I error.  (See J.  N.  K.  Rao and D.  R. 
Thomas, "Chi-Squared Test for Contingency Tables," in C.  J. 
Skinner, D.  Holt, and T.  M.  F.  Smith (eds.), Analysis of Complex
Surveys (New York:  John Wiley and Sons, 1989).)

\d Probability levels are from the Satterthwaite adjusted chi-square
tests.  Probability level refers to the chances that the coefficient
equals zero in the population.  By convention, coefficients with a
probability level less than or equal to 5 percent (0.05) are regarded
as statistically significant.  In this table, 0.01 indicates a
probability less than or equal to 0.01. 

\e The population means are the variable values in this sample
weighted by the National Inflation Factor to approximate population
values. 


MAJOR CONTRIBUTORS TO THIS REPORT
========================================================== Appendix II


   PROGRAM EVALUATION AND
   METHODOLOGY DIVISION
-------------------------------------------------------- Appendix II:1

Robert E.  White, Assistant Director
Martin T.  Gahart, Project Manager
Dale W.  Harrison, Project Adviser
Beverly A.  Ross, Project Adviser
Penny Pickett, Communications Analyst


BIBLIOGRAPHY
=========================================================== Appendix 1

"As Nation Grows Older, Traffic Safety Officials Confront Questions
of Who Should Drive." Journal of the American Medical Association,
268:3 (1992), 307-8. 

Centers for Disease Control.  "Trends in Alcohol-Related Traffic
Fatalities, by Sex--United States." Journal of the American Medical
Association, 268:3 (1992), 313-14. 

Danner, M., K.  Langieder, and Thomas Hummel.  "Experience from the
Analysis of Accidents with a High Belt Usage Rate and Aspects of
Continued Increase in Passenger Safety." Presented at the Eleventh
International Technical Conference on Experimental Safety Vehicles,
Washington, D.C., 1987. 

Evans, Leonard.  Car Mass and Likelihood of Occupant Fatality. 
Technical paper 820807.  Warrendale, Pa.:  Society of Automotive
Engineers, 1982. 

Evans, Leonard.  "The Effectiveness of Safety Belts in Preventing
Fatalities." Accident Analysis and Prevention, 18:3 (1986), 229-41. 

Evans, Leonard.  "Belted and Unbelted Driver Accident Involvement
Rates Compared." Journal of Safety Research, 18:2 (1987), 57-64. 

Evans, Leonard.  "Older Driver Involvement in Fatal and Severe
Traffic Crashes." Journal of Gerontology:  Social Sciences, 43:6
(1988a), 5186-93. 

Evans, Leonard.  "Risk of Fatality from Physical Trauma versus Sex
and Age." The Journal of Trauma, 28:3 (1988b), 368-78. 

Evans, Leonard.  "Restraint Effectiveness, Occupant Ejection From
Cars, and Fatality Reductions." Accident Analysis and Prevention,
22:2 (1990), 167-75. 

Evans, Leonard.  "Motorized Two-Point Safety Belt Effectiveness in
Preventing Fatalities." Accident Analysis and Prevention, 23:2/3
(1991a), 165-74. 

Evans, Leonard.  Traffic Safety and the Driver.  New York:  Van
Nostrand Reinhold, 1991b. 

Evans, Leonard, and Michael C.  Frick.  "Safety Belt Effectiveness in
Preventing Driver Fatalities Versus a Number of Vehicular, Accident,
Roadway, and Environmental Factors." Journal of Safety Research, 17:4
(1986), 143-54. 

Evans, Leonard, and Michael C.  Frick.  "Car Size or Car Mass:  Which
Has Greater Influence on Fatality Risk?" American Journal of Public
Health, 82:8 (1992), 1105-12. 

Evans, Leonard, and Paul Wasielewski.  "Risky Driving Related to
Driver and Vehicle Characteristics." Accident Analysis and
Prevention, 15:2 (1983), 121-36. 

Green, R.N., et al.  "Fatal Injuries to Restrained Passenger Car
Occupants in Canada:  Crash Modes and Kinematics of Injury." Accident
Analysis and Prevention, 26:2 (1994), 207-14. 

Hackney, James R.  "The Effects of FMVSS No.  208 and NCAP on Safety
as Determined from Crash Test Results." Presented at the Thirteenth
International Technical Conference on Experimental Safety Vehicles,
Paris, France, 1991. 

Hill, J.R., G.M.  Mackay, and A.P.  Morris.  "Chest and Abdominal
Injuries Caused by Seat Belt Loading." Accident Analysis and
Prevention, 26:1 (1994), 11-26. 

Hunter, William W., et al.  "Observed and Self-Reported Seat Belt
Wearing as Related to Prior Traffic Accidents and Convictions."
Accident Analysis and Prevention, 25:5 (1993), 545-54. 

Insurance Institute for Highway Safety.  "Death Rates Vary Among Cars
by More than 800 Percent." Status Report, 26:4 (1991), 1-5. 

Insurance Institute for Highway Safety, Shopping for a Safer Car: 
1994 Models.  Arlington, Va.:  1993. 

Joksch, Hans C.  "Velocity Change and Fatality Risk in a Crash-- A
Rule of Thumb." Accident Analysis and Prevention, 25:1 (1993), 103-4. 

Kahane, Charles J.  Effect of Car Size on the Frequency and Severity
of Rollover Crashes.  Washington, D.C.:  National Highway Traffic
Safety Administration, 1990. 

Klein, T.M., E.  Hertz, and S.  Borener.  A Collection of Recent
Analyses of Vehicle Weight and Safety, DOT-HS-807-677.  Washington,
D.C.:  National Highway Traffic Safety Administration, 1991. 

Lestina, Diane C., et al.  "Motor Vehicle Crash Injury Patterns and
the Virginia Seat Belt Law." Journal of the American Medical
Association, 265:11 (1991), 1409-13. 

Mackay, G.M., et al.  "Restrained Front Seat Car Occupant
Fatalities--The Nature and Circumstances of Their Injuries." Accident
Analysis and Prevention, 24:3 (1992), 307-15. 

Mackay, Murray.  "Crash Protection for Older Persons." In
Transportation Research Board.  Transportation in an Aging Society,
special report 218.  Washington, D.C.:  Transportation Research
Board, 1988. 

Mackay, Murray.  "Occupant Protection and Vehicle Design." Presented
at The Biomechanics of Impact and Motor Vehicle Crash Performance:  A
Global Concern, conference sponsored by the Association for the
Advancement of Automotive Medicine and the International Research
Council on the Biomechanics of Impact, Atlanta, May 1993. 

Mercier, Cletus R., et al.  "Age as a Predictor of Injury Severity in
Broad-Side and Angle Vehicular Collisions." Presented at the annual
meeting of the American Statistical Association, New Orleans, 1993. 

National Highway Traffic Safety Administration.  Effect of Car Size
on Fatality and Injury Risk.  Washington, D.C.:  1991a. 

National Highway Traffic Safety Administration.  General Estimates
System 1990, DOT-HS-807-781.  Washington, D.C.:  1991b. 

National Highway Traffic Safety Administration.  National Accident
Sampling System--Crashworthiness Data System:  Analytical User's
Manual, 1991 File.  Washington, D.C.:  1991c. 

National Highway Traffic Safety Administration.  National Accident
Sampling System--1991 Crashworthiness Data System:  Data Collection,
Coding, and Editing Manual.  Washington, D.C.:  1991d. 

National Highway Traffic Safety Administration.  Evaluation of the
Effectiveness of Occupant Protection, Federal Motor Vehicle Safety
Standard 208:  Interim Report, DOT-HS-807-843.  Washington, D.C.: 
1992a. 

National Highway Traffic Safety Administration.  Moving America
Safely:  Priority Plan, NHTSA's Activities Through 1994.  Washington,
D.C.:  1992b. 

National Highway Traffic Safety Administration.  Addressing the
Safety Issues Related to Younger and Older Drivers, DOT-HS-807-957. 
Washington, D.C.:  1993a. 

National Highway Traffic Safety Administration.  1991 Alcohol Fatal
Crash Facts.  Washington, D.C.:  1993b. 

National Highway Traffic Safety Administration.  Fatal Accident
Reporting System 1991, DOT-HS-807-954.  Washington, D.C.:  1993c. 

National Higway Traffic Safety Administration.  Final Regulatory
Evaluation, FMVSS No.  208, Mandatory Air Bag Installation, Docket
74-14-N82-002.  Washington, D.C.:  1993d. 

National Highway Traffic Safety Administration.  New Car Assessment
Program:  Response to the NCAP FY 1992 Congressional Requirements. 
Washington, D.C.:  1993e. 

National Research Council.  Automobile Fuel Economy:  How Far Should
We Go?  Washington, D.C.:  National Academy Press, 1992. 

National Transportation Safety Board.  Safety Study--Performance of
Lap/Shoulder Belts in 167 Motor Vehicle Crashes, NTSB/SS-88/02. 
Washington, D.C.:  1988. 

Partyka, Susan C.  Papers on Adult Seat Belts--Effectiveness and Use,
DOT-HS-807-285.  Washington, D.C.:  National Highway Traffic Safety
Administration, 1988. 

Partyka, Susan C., and William A.  Boehly.  "Passenger Car Weight and
Injury Severity in Single Vehicle Nonrollover Crashes." Presented at
the Twelfth International Technical Conference on Experimental Safety
Vehicles, Goteborg, Sweden, 1989. 

Pike, J.A.  "The Quantitative Effect of Age on Injury Outcome."
Presented at the Twelfth International Technical Conference on
Experimental Safety Vehicles, Goteborg, Sweden, 1989. 

Preusser, David F., Allan F.  Williams, and Adrian K.  Lund. 
"Characteristics of Belted and Unbelted Drivers." Accident Analysis
and Prevention, 23:6 (1991), 475-82. 

Robertson, Leon S.  "How to Save Fuel and Reduce Injuries in
Automobiles." The Journal of Trauma, 31:1 (1991), 107-9. 

Shah, Babubhai V., et al.  SUDAAN User's Manual, Release 6.0. 
Research Triangle Park, N.C.:  Research Triangle Institute, 1992. 

Stewart, J.  Richard.  "Seat Belt Use and Accident Involvement:  A
Comparison of Driving Behavior Before and After a Seat Belt Law."
Accident Analysis and Prevention, 25:6 (1993), 757-63. 

Terhune, Kenneth W.  The Contribution of Rollover to Single-Vehicle
Crash Injuries, Calspan report 7881-1.  Buffalo:  Arvin/Calspan
Transportation Sciences Center, 1991. 

Terhune, Kenneth W., and Thomas A.  Ranney.  "Components of Vehicle
Aggressiveness." Presented at the Twenty-Eighth Annual Meeting of the
American Association for Automotive Medicine, Denver, 1984. 

Transportation Research Board.  Research and Development Needs for
Maintaining the Safety and Mobility of Older Drivers, circular 398. 
Washington, D.C.:  1992. 

U.S.  General Accounting Office.  Highway Safety:  Have Automobile
Weight Reductions Increased Highway Fatalities?  GAO/PEMD-92-1. 
Washington, D.C.:  October 1991. 

U.S.  General Accounting Office.  Highway Safety:  Safety Belt Use
Laws Save Lives and Reduce Costs to Society, GAO/RCED-92-106. 
Washington, D.C.:  May 1992. 

U.S.  General Accounting Office.  Highway Safety:  Factors Affecting
Involvement in Vehicle Crashes, GAO/PEMD-95-3.  Washington, D.C.: 
October 1994. 

Viano, David C.  "Limits and Challenges of Crash Protection."
Accident Analysis and Prevention, 20:6 (1988), 421-29. 

Viano, David C., et al.  "Involvement of Older Drivers in
Multivehicle Side-Impact Crashes." Accident Analysis and Prevention,
22:2 (1990), 177-88. 

Williams, Allan F., et al.  "Use of Seatbelts in Cars with Automatic
Belts." Public Health Reports, 107:2 (1992), 182-88. 

Zador, Paul L., and Michael A.  Ciccone.  "Automobile Driver
Fatalities in Frontal Impacts:  Air Bags Compared with Manual Belts."
American Journal of Public Health, 83:5 (1993), 661-66. 





RELATED GAO PRODUCTS
=========================================================== Appendix 2

Highway Safety:  Reliability and Validity of DOT Crash Tests
(GAO/PEMD-95-5, May 1995). 

Highway Safety:  Factors Affecting Involvement in Vehicle Crashes
(GAO/PEMD-95-3, Oct.  1994). 

Highway Safety:  Safety Belt Use Laws Save Lives and Reduce Costs to
Society (GAO/RCED-92-106, May 1992). 

Highway Safety:  Have Automobile Weight Reductions Increased Highway
Fatalities?  (GAO/PEMD-92-1, Oct.  1991). 

Highway Safety:  Motorcycle Helmet Laws Save Lives and Reduce Costs
to Society (GAO/RCED-91-179, July 1991). 

Highway Safety:  Fatalities in Light Trucks and Vans (GAO/PEMD-91-8,
Nov.  1990). 

Highway Safety:  Trends in Highway Fatalities 1975-87
(GAO/PEMD-90-10, Mar.  1990). 

Motor Vehicle Safety:  Passive Restraints Needed to Make Light Trucks
Safer (GAO/RCED-90-56, Nov.  1989). 

Drinking-Age Laws:  An Evaluation Synthesis of Their Impact on
Highway Safety (GAO/PEMD-87-10, Feb.  1987). 
