Highway Safety: Reliability and Validity of DOT Crash Tests (Letter
Report, 05/08/95, GAO/PEMD-95-5).

Pursuant to a congressional request, GAO reviewed the National Highway
Traffic Safety Administration's (NHTSA) crash test programs, focusing on
whether they provide valid and reliable indicators of occupant safety in
real world crashes.

GAO found that: (1) the probability of sustaining a serious injury has
declined since the inception of NHTSA test programs; (2) cars marketed
in the United States have become more crashworthy; (3) the consistency
of New Car Assessment Program (NCAP) test results is questionable and
its unreliable data may lead to misinformed purchasing decisions; (4)
NCAP ability to predict a vehicle occupant's protection in real world
crashes is limited, since NCAP results can only be applied to frontal
collisions; and (5) there is a statistically significant relationship
between fatality rates and NCAP predicted injuries, however, these high
fatality rates are associated with the poorest NCAP performers.

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

 REPORTNUM:  PEMD-95-5
     TITLE:  Highway Safety: Reliability and Validity of DOT Crash Tests
      DATE:  05/08/95
   SUBJECT:  Motor vehicle safety
             Automobile industry
             Transportation statistics
             Traffic accidents
             Motor vehicle standards
             Highway safety
             Statistical methods
             Statistical data
             Consumer protection
             Accident prevention
IDENTIFIER:  NHTSA New Car Assessment Program
             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 - RELIABILITY AND
VALIDITY OF DOT CRASH TESTS

GAO/PEMD-95-5

Department of Transportation Crash Tests


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

  AAAM - Association for the Advancement of Automotive Medicine
  DOT - Department of Transportation
  FARS - Fatal Accident Reporting System
  FHWA - Federal Highway Administration
  FMVSS - Federal Motor Vehicle Safety Standards
  HIC - Head injury criterion
  ISO - International Standards Organization
  NASA - National Aeronautics and Space Administration
  NASS - National Accident Sampling System
  NCAP - New Car Assessment Program
  NHTSA - National Highway Traffic Safety Administration
  SAE - Society of Automotive Engineers
  SUDAAN - Survey Data Analysis

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


B-250144

May 5, 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

Between 1980 and 1992, the annual death toll on America's highways
dropped from more than 50,000 to less than 40,000.  One factor
contributing to this decline may have been the increasing concern
among consumers about the safety of the vehicles they purchase.  The
crash test programs performed by the Department of Transportation
(DOT) are the major source of the safety information available to
today's automobile purchaser. 

You asked us to supplement our earlier testimony and report to your
Committee on the relationship between automotive design factors and
safety by responding to a number of questions concerning automobile
safety.  As one component of this research, you requested that we
review the automobile crashworthiness program of the National Highway
Traffic Safety Administration (NHTSA) to determine whether NHTSA's
crash test programs provide valid and reliable indicators of occupant
safety in real-world crashes. 


   BACKGROUND
------------------------------------------------------------ Letter :1

The Department of Transportation has four offices that conduct
automobile crash tests:  three within NHTSA and one in the Federal
Highway Administration (FHWA).  The activities of two programs run by
NHTSA are the focus of this report.  NHTSA's Office of Vehicle Safety
Compliance performs a compliance testing program of 30-mile-per-hour
full-frontal crashes of automobiles, light trucks, and vans into a
fixed rigid barrier.  This program was created under section 103 of
the National Traffic and Motor Vehicle Safety Act of 1966, and it is
designed to ensure that vehicles meet minimum safety requirements as
specified in Federal Motor Vehicle Safety Standard No.  208 -
Occupant Crash Protection (FMVSS 208).\1

Also under the authority of NHTSA is the New Car Assessment Program
(NCAP), conducted by the Office of Market Incentives.  This program,
mandated under title II of the Motor Vehicle Information and Cost
Savings Act of 1972, was created to provide information to consumers
on the relative crashworthiness, or safety, of automobiles.  This
charge differs from the compliance test in that vehicles tested in
NCAP are not required to meet specified safety standards, while the
purpose of compliance tests is to ensure that vehicles meet a level
of safety required by law.  The NCAP test also differs from the
compliance test in two important aspects:  NCAP crashes its vehicles
at 35 miles per hour, which translates to over one-third more energy
than compliance tests, and NCAP engages all manual and automatic
restraints, while the compliance test employs only passive
restraints.  By using all restraint systems, NCAP assesses the
maximum crashworthiness of a vehicle in high-speed frontal crashes. 

In addition to the two programs described above, NHTSA's Office of
Crashworthiness Research conducts a variety of tests to study a wide
range of individual safety issues that arise from specific crash
configurations.  FHWA conducts crash tests to study the interaction
between automobiles and roadside obstacles and devices such as guard
rails, telephone poles, and bridge abutments. 


--------------------
\1 Code of Federal Regulations, 49 C.F.R.  Part 571:  Federal Motor
Vehicle Safety Standards (FMVSS), Standard No.  208 - Occupant Crash
Protection.  The Office of Vehicle Safety Compliance also uses
dynamic crash tests to determine vehicular compliance with FMVSS 212,
Windshield Mounting; FMVSS 219, Windshield Zone Intrusion; and FMVSS
301, Fuel System Integrity. 


   OBJECTIVES, SCOPE, AND
   METHODOLOGY
------------------------------------------------------------ Letter :2

To respond to your request, we examined data from tests conducted for
compliance with FMVSS 208 (compliance tests) as well as those
conducted under the New Car Assessment Program.  We chose to focus on
these programs because both conduct tests that are similarly
configured, employ standardized procedures, and have been assessing
vehicle crashworthiness over a period of years.  The two other crash
test programs run by DOT are largely research based and, although
important, have different purposes from those of our study. 

Our analysis consisted of three parts:  (1) an examination of trends
over time in crash test results of both programs, (2) an assessment
of the reliability of NCAP results, and (3) a review of the
relationship between NCAP results and real-world traffic injuries and
fatalities.  We first reviewed the background, sample selection, and
testing procedures of both NCAP and the compliance program.  (See
appendix I).  We then examined what it is that crash tests measure,
as wellas how well measurement devices used in crash tests simulate
human biomechanics and physiological response by reviewing
biomechanic, human tolerance, and automotive safety literature and by
interviewing experts in those fields.  (See appendixes II and III.)
Next, we analyzed changes in crash test results by year for both the
compliance program and NCAP.  (See appendix IV.)

To address the reliability of crash test results, that is, the degree
to which consistent results are obtained through repeated trials, we
examined research conducted by NHTSA and compared NCAP results with
those obtained in crash tests conducted by manufacturers.  (See
appendix V.) Finally, we conducted analyses using two national
databases that allowed us to relate real-world fatality rates for
drivers with the predicted injury risks derived from NCAP results. 
(See appendix VI).  For this analysis, we used Poisson regressions to
assess the relationship between fatality rates, derived from the
Fatal Accident Reporting System and the R.L.  Polk Vehicle
Registration System, and the combined injury risk calculated from the
NCAP measurements that assess the potential for skeletal injuries to
the head and chest.  Analyses were conducted for restrained drivers
in one- and two-car frontal crashes. 

We did not include information from the compliance program in the
analyses we conducted on either the reliability or the predictive
validity of crash test results.  In our assessment of the reliability
of crash test results, we did not uncover a quantity of data
sufficient enough to compare the results of two or more trials of
vehicle models.  In the case of the predictive validity of crash test
results, we did not use compliance test data for two reasons.  First,
the compliance program had conducted only 145 tests between 1987 and
1992.  Second, the variation among compliance results was relatively
narrow and scores tended to cluster far below the ceiling values for
the compliance tests.  These two items resulted in a dataset that was
insufficient for conducting detailed statistical analyses. 

We conducted our review in accordance with generally accepted
government auditing standards. 


   RESULTS IN BRIEF
------------------------------------------------------------ Letter :3

Recent trends in test results indicate that, first, the probability
of sustaining a serious injury, as it is measured by the NCAP and
compliance tests, has decreased substantially since the inception of
these test programs.  In addition, the variation in scores between
automobile models has shrunk, indicating that cars marketed in the
United States have become more uniformly crashworthy.  Some of this
improvement, we concluded, could appropriately be attributed to DOT
crash test programs initiated by the National Traffic and Motor
Vehicle Safety Act of 1966 and the Motor Vehicle Information and Cost
Savings Act of 1972. 

Second, the reliability, that is, the consistency of results derived
from NCAP tests is questionable.  Existing evidence suggests that
large differences between crash test scores most likely reflect true
differences between vehicles.  However, we cannot be sure that even
moderately large score differences between two vehicles might not
disappear or be reversed if they were tested again.  This is because,
in general, only one unit of a specific vehicle line is tested, and
this is not enough to say with confidence that the results are
indicative of how other units of that same vehicle line would
perform.  The new star rating system incorporated by NCAP in recent
years, a system that places vehicles into one of five categories on
the basis of potential injury risk, could exacerbate this problem in
some cases. 

Third, we found that the ability of NCAP to predict a vehicle's
occupant protection in real-world crashes is limited.  By their
nature, NCAP crash test results can be validly applied only to
frontal collisions, which account for slightly more than half of all
injury-producing accidents.  We found a statistically significant
relationship between fatality rates and NCAP predicted injury risk;
however, this relationship derives from the high fatality rates
associated with the poorest performers in NCAP (the 20 percent of
vehicles with the highest potential injury risks derived from NCAP
results). 


   GAO'S ANALYSIS
------------------------------------------------------------ Letter :4


      AUTOMOBILES HAVE BECOME MORE
      CRASHWORTHY
---------------------------------------------------------- Letter :4.1

In NCAP crashes, nearly all cars now meet the head and chest injury
standards of the compliance tests, although they are 36-percent more
violent than compliance crashes.  The average probability of
sustaining a serious injury in a 35 mile-per-hour crash as measured
by NCAP has declined from over 0.5 in 1980 to less than 0.2 in 1993. 
(See figure 1.) Differences among the crashworthiness scores of
vehicles tested in this program have experienced similar declines. 
The introduction of air bags has contributed significantly to this
improvement.  (For a complete discussion of the trends in crash test
results, see appendix IV.)

   Figure 1:  Mean NCAP Injury
   Risk by Model Year\ a

   (See figure in printed
   edition.)

\a For all passenger cars.  Does not include light trucks and vans. 

A causal linkage between improved crash test scores and declining
highway fatalities cannot be asserted with certainty because of both
the many variables involved in a crash and the increased emphasis on
traffic accident and injury prevention over the past decade. 
Nonetheless, it seems reasonable to conclude that manufacturers'
successful efforts to improve their products' performance in NHTSA
crash tests, particularly in NCAP, have contributed to improved
occupant protection in real-world crashes, although we were unable to
quantify that contribution.  These improvements to performance have
derived from a variety of efforts, with two examples being modernized
manufacturing techniques and an increased emphasis on safety systems
and designs. 

In addition, in recent years, automotive designers have turned more
to computer-based simulations to assist in the design of vehicles
that meet crash test standards.  Although we did not evaluate the
state of the art in computer-based crash models, we learned from
industry personnel that such modeling appears to accurately predict
the results of actual crash tests.  Indeed, one computer specialist
informed us that the industry uses crash tests in part to validate
their computer models. 

Although simulated crashes are costly as they currently require
access to supercomputers, they do allow the manufacturer to assess
the crashworthiness of a vehicle in more trials, more quickly, and at
impacts points other than the front (or side) of a car.  These
benefits over actual crash testing permit the identification of crash
forces upon an occupant in a time frame that offers immediate
redesign implications. 


      CRASH TEST RESULTS HAVE
      QUESTIONABLE RELIABILITY
---------------------------------------------------------- Letter :4.2

To determine whether the result of any test is reliable, consistent
results must be obtained through repeated trials of a specified
procedure.  In the case of crash tests, this means that consistent
results of repeated tests of a specific vehicle model are required. 
This is particularly crucial when comparing the safety ratings of
different vehicles.  Both the NCAP and compliance programs generally
conduct only one trial of a specific vehicle model; thus,
insufficient data exist to accurately define the reliability of crash
test results.  That is, the ability to predict with confidence the
likelihood of a tested model's receiving similar scores if tested
again is low. 

We found only two sources of information on which to assess the
reliability of crash test results:  a study conducted by NHTSA in
1984, which examined the variations in test results of 12
consecutively manufactured Chevrolet Citations, and our own analysis
of the differences between results for vehicle models tested in NCAP
and the results for those vehicles in corresponding tests conducted
by automobile manufacturers.  Our analysis of the data derived from
NHTSA's 1984 study revealed wide variations in the head injury
criterion (HIC) results, the measurement taken to assess potential
skeletal head injuries.  (See appendix V.) Although NHTSA ascribed
the variation in results to a number of sources, including the test
itself, it failed to discuss the implications of the combined effect
of these sources on crash test results; namely, that even within a
specific vehicle line, the result of one test may not be indicative
of the model's performance from trial to trial, and large differences
in the resultant HIC may occur. 

We also examined the differences between the results of NCAP and
manufacturers' tests provided to us by NCAP officials.  The tests
conducted by the manufacturers essentially duplicated the NCAP test
procedures.  We compared the results of the two tests using the star
rating system recently developed by NCAP, hypothesizing that if the
manufacturer test were considered a second trial for a model line,
its results should be consistent with the NCAP, or first trial.  The
star rating system ranks cars from 1 to 5 stars, with 5 stars being
the best rating, or safest car, and 1 star being the worst rating, or
least safe car.  These ratings are based on the risks of serious
injury for vehicles, which are calculated from the head injury
criterion and chest acceleration scores from NCAP tests.  (For a
discussion of crash test measurements and the star system, see
appendix II.)

We found that in only about one-half of the paired comparisons would
NCAP- and manufacturer-tested vehicles have received the same star
rating.  In 32 percent of the comparisons, the results of the second
trial would have changed by 1 star, while in 8 percent of the cases,
the ratings of the vehicles would have changed by 2 or more stars. 
When we compared the risks of serious injury (the base unit
categorized into the five star ratings) derived from the manufacturer
and NCAP data, we found that each star category was associated with a
wide band in which the resultant risk scores of subsequent tests
might fall.  For example, the results of a second test of a vehicle
rated as 4 stars by its first test could fall between 5 stars and 2
stars. 

The analyses described above are based on the only two sources of
information we could find.  The quantity of data in each analysis was
not enough for us to fully quantify the reliability of crash test
results; however, we were able to determine that NCAP scores, whether
reported in raw HIC and chest acceleration scores or as categories of
injury probability, have associated levels of imprecision and that
seemingly large differences in crash test results may not necessarily
reflect true differences in a vehicle's safety potential.  By not
properly defining and publishing the degree of reliability, consumers
may be misled into purchasing a vehicle purported to be more
crashworthy than another when, in fact, it may be no more safe, or
even less safe, than the comparison vehicle. 


      NCAP RESULTS AND REAL-WORLD
      FATALITY RATES
---------------------------------------------------------- Letter :4.3

Since NCAP crash tests are designed to simulate full-frontal
collisions, we restricted our analysis to those types of crashes and
found the results of NCAP crash tests are generally reflected in
real-world fatality rates.  That is, on the whole, a statistically
significant relationship exists between real-world highway fatality
rates associated with vehicles tested in the NCAP program and their
scores in crash tests.  However, we concluded that this relationship
derives mainly from the high fatality rates of vehicles with the
worst NCAP scores.  When we divided vehicles into NCAP score
quintiles--that is, placed the vehicles into one of five
20-percentile categories based on their location in the distribution
of NCAP results--we found that the quintile with the worst NCAP
scores (those vehicles in the highest 20-percentile category) had
significantly higher fatality rates than the remaining 80 percent of
NCAP-tested vehicles.  The remaining four quintile categories,
however, had associated fatality rates that were not significantly
different from one another.  (See figure 2 and appendix VI.)

   Figure 2:  Mean NCAP Injury
   Risk Scores and Fatality Rates\
   a

   (See figure in printed
   edition.)

\a By quintile, per 100,000 vehicles. 


   CONCLUSIONS
------------------------------------------------------------ Letter :5

Over time, the mean risk of injury in frontal crashes, as measured by
NHTSA crash tests, has declined and indeed has mirrored a similar
trend in the annual number of highway fatalities.  While we cannot
state with certainty that NHTSA crash tests are a causal factor in
improved crashworthiness, we believe that efforts on the part of
automobile manufacturers to produce vehicles that score well on these
tests have contributed to the improvement of the overall safety of
vehicles.  At the very least, the results of NCAP and compliance
tests provide indications that the vehicle fleet, on the whole, has
become safer over the past 15 years. 

These trends in the mean score of crash tests, however, do not
necessarily suggest that individual vehicles have well-defined levels
of safety, nor do they suggest that the relative rankings of two
vehicles would be the same if subsequent trials were conducted.  They
also do not suggest that differing test results are reflected in data
derived from real-world traffic collisions.  Indeed, only the poorest
performers in NCAP had associated fatality rates that were
significantly different than other NCAP vehicles. 


   RECOMMENDATIONS
------------------------------------------------------------ Letter :6

On the basis of our findings, we make two recommendations to the
Administrator of NHTSA.  First, we recommend that information on NCAP
reliability be updated and made available, in clear language, to the
general public.  Such an effort would require an update of the
repeatability study the agency conducted in 1984 and could result not
only in a better understanding of the reliability of crash tests for
predicting injury risk, but also in discovering ways in which NHTSA
can limit the error that derives from sources under its control. 

We also recommend that NHTSA explore the feasibility of alternative
means of testing the crashworthiness of new vehicles.  Computer
simulations may provide one such alternative.  It may be possible to
better assess the safety potential of a vehicle through
computer-based modeling as this allows more trials, more quickly, and
modeling is capable of simulating impacts at all points of a vehicle. 
In addition, this rapidly emerging technology has the added
capability of providing immediate insights into redesigning vehicles
in which the crashworthiness may not yet be optimal. 


   AGENCY COMMENTS
------------------------------------------------------------ Letter :7

We received written comments on a draft of this report from the
Department of Transportation.  The Department concurred with our
recommendation that it update its information on NCAP reliability. 
We are concerned, however, that the agency might believe it has
already complied with this recommendation by developing the star
rating format.  As noted above and explained in detail in appendix V,
this new format does not resolve our questions concerning NCAP
reliability. 

The Department interpreted our second recommendation as a
recommendation to augment or replace "live" crash tests with computer
simulations and did not concur with us.  The Department cited
concerns about the costs and predictive limitations of such
simulations.  We share these concerns, but we believe the Agency has
misinterpreted the recommendation.  We avoided recommending the
adoption of any particular substitute for the current crash test
procedures at this time.  Rather, we urged the Agency to explore all
possible means of reliably defining vehicle crashworthiness. 
Computer modeling is a potential alternative that deserves
exploration and monitoring as the technology matures.  Other
alternatives could include extending testing programs to include
side, rear, and frontal-offset impacts to gain a better understanding
of the total safety of a vehicle or seeking greater sharing of crash
test data developed by automotive manufacturers either through crash
tests replicating NHTSA's or through their individual component
testing programs. 

The Department provided a number of other specific comments.  They
are reproduced in appendix VII, together with our response.  We have
also made modifications to the report as we deemed appropriate on the
basis of these comments.  After responding to our draft report, the
Department also provided us with additional data relevant to NCAP
reliability.  The results of our analysis of these data can be found
in appendix V. 

We are sending copies of this report to the Secretary of
Transportation, the Administrator of the National Highway Traffic
Safety Administration, and to other interested parties.  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.  Major contributors to this report are listed in
appendix VIII. 

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


DEPARTMENT OF TRANSPORTATION CRASH
TEST PROGRAMS
=========================================================== Appendix I


   INTRODUCTION
--------------------------------------------------------- Appendix I:1

The Department of Transportation has four offices that conduct
automobile crash tests:  three within NHTSA and one in the Federal
Highway Administration.  A compliance test program conducted by
NHTSA's Office of Vehicle Safety Compliance consists of full-frontal
crashes of automobiles, light trucks, and vans into a fixed rigid
barrier to ensure that vehicles meet certain minimum safety
requirements.\1 Also under the authority of NHTSA is the New Car
Assessment Program, conducted by the Office of Market Incentives. 
NCAP tests are similar to compliance tests, but they are performed to
provide consumer information on the relative crashworthiness of
automobiles.  The NHTSA Office of Crashworthiness Research conducts a
variety of tests to study a wide range of individual safety issues
that arise from specific crash configurations.  Finally, FHWA
conducts crash tests to study the interaction between automobiles and
roadside obstacles and devices such as guard rails, telephone poles,
and bridge abutments. 

In this study, we focused on the crash tests run under the compliance
program and the New Car Assessment Program because both tests are
similarly configured, employ standardized procedures, and have been
assessing vehicle crashworthiness over a period of years.  The tests
conducted by the Office of Crashworthiness Research and those of
FHWA, although important, have different purposes from those of our
study and could not provide a quantity of data sufficient for us to
assess relationships between test results and real-world performance. 


--------------------
\1 Because these tests are required by NHTSA under standard 208 of
the Federal Motor Vehicle Safety Standards, they are commonly called
FMVSS 208 tests.  NHTSA also uses dynamic crash tests to determine
vehicular compliance with FMVSS 212, Windshield Mounting; FMVSS 219,
Windshield Zone Intrusion; and FMVSS 301, Fuel System Integrity. 


   COMPLIANCE TEST PROGRAM
--------------------------------------------------------- Appendix I:2

Section 103 of the National Traffic and Motor Vehicle Safety Act of
1966 charged DOT with establishing standards for vehicle safety. 
These safety standards are codified by NHTSA in the U.S.  Code of
Federal Regulations at 49 C.F.R.  Part 571:  Federal Motor Vehicle
Safety Standards (FMVSS), Standard No.  208:  Occupant Crash
Protection.  Beginning with the 1987 vehicle model year, FMVSS 208
required that passenger cars, light trucks, and vans sold in the
United States be certified as meeting minimal safety levels as
measured by anthropomorphic dummies in dynamic crash tests.\2 The
purpose of this standard is

     "to reduce the number of deaths of vehicle occupants, and the
     severity of injuries, by specifying vehicle crashworthiness
     requirements in terms of forces and accelerations measured on
     anthropomorphic dummies, and by specifying equipment
     requirements for active and passive systems."

These crash tests are conducted under the guidance of NHTSA's Office
of Vehicle Safety Compliance. 


--------------------
\2 Self-certification in a 30 mile-per-hour barrier collision and
passive restraint requirements were phased in between 1987 and 1990
for passenger cars.  Light trucks, vans, and sport utility vehicles
were required to meet dynamic crash requirements beginning with the
1992 model year. 


      SAMPLE SELECTION FOR
      COMPLIANCE TESTS
------------------------------------------------------- Appendix I:2.1

The current compliance program relies, for the most part, on a
certification process in which the manufacturer of a specific make
and model vehicle states that the vehicle meets all safety
requirements set forth in FMVSS 208.  In addition, each year NHTSA
selects a number of vehicles to test to ensure that the
manufacturer's certification is justified.  The criteria used to
determine which specific makes and models to test are based on
whether a vehicle is in its first or second model year, whether
safety features have been added or redesigned, and how many units are
on the road.  In selecting models for testing, NHTSA also includes
any evidence of poor crashworthiness derived either from consumer
complaints filed about specific models or from other crash test
programs (in particular, NCAP).  Through these criteria, NHTSA
compiles a preliminary list of about 50 candidate vehicles for
testing and requests information on crash test performance from the
manufacturer of each candidate model to determine the final list of
vehicles to be tested.  Though they are under no obligation to do so,
manufacturers will normally provide one or two sets of results from
their tests of the model NHTSA specifies.  NHTSA uses these data not
only as an input for determining the final list of test vehicles, but
also as a baseline with which to compare its own results. 


      TEST CONDITIONS
------------------------------------------------------- Appendix I:2.2

The compliance test consists of a full-frontal collision of a vehicle
into a fixed rigid barrier at a velocity of 30 miles per hour. 
Anthropomorphic test dummies, fitted with instrumentation to measure
forces and accelerations acting on the head, chest, and both femurs,
are placed in the driver and front passenger seats.  Only passive
restraint systems--those that require no effort on the part of an
occupant--are engaged.  Examples of these are air bags and automatic
seat belts.  Seat belts that require active participation by the
occupant are not used.\3 The underlying assumption is that if a
vehicle meets the standards for those occupants who do not make use
of all available safety restraint systems, it will also meet the
requirements for those who do. 

The test conditions further specify the forward placement of the
seat, the angle of the seat back, the angle of the steering column
(where the vehicle has tilt steering), and a number of other
components.  Some of these, such as adjustable backs for seats, are
placed in the manufacturer's nominal design riding position--that is,
the position the manufacturer says is the proper one for the average
adult male (5 feet 9 inches, 167 pounds). 


--------------------
\3 This requirement applies to all passenger cars and those light
trucks certified by the manufacturer as meeting the automatic
occupant protection requirement and will be phased in for light
trucks beginning with the 1995 model year. 


   NEW CAR ASSESSMENT PROGRAM
--------------------------------------------------------- Appendix I:3

A second crash test program we studied is the New Car Assessment
Program conducted by NHTSA's Office of Market Incentives.  This
program was mandated under title II of the Motor Vehicle Information
and Cost Savings Act of 1972 to provide consumers with an
understanding of the relative crashworthiness of passenger motor
vehicles.  Since 1979, NCAP has conducted almost 500 crash tests of
passenger cars, light trucks, and vans.  From 1979 to 1986, NCAP was
considered an indicant test for vehicle compliance with Federal Motor
Vehicle Safety Standards 212, Windshield Mounting; 219, Windshield
Zone Intrusion; and 301, Fuel System Integrity.  That is, if a
vehicle performed reasonably well on these tests, which required
dynamic testing, then it would likely meet compliance test
requirements because the NCAP test involves a more violent crash than
the one required for the compliance test.  If a vehicle performed
poorly, the information would be transmitted to the Office of Vehicle
Safety Compliance for testing its compliance with the safety
standards.  NCAP has not been an indicant program since the
implementation of dynamic crash tests in the FMVSS 208 program in
1987; however, poor performance on the NCAP test typically leads to
compliance testing of the same model. 


   COMPARISON OF THE TWO TESTS
--------------------------------------------------------- Appendix I:4

The NCAP crash test is generally similar to the compliance test. 
Both are full-frontal collisions into a fixed rigid barrier, and both
use roughly the same criteria when determining which vehicles to
test.  However, three very important differences distinguish the two
test programs.  First, vehicles in the NCAP test are crashed at 35
miles per hour rather than 30 miles per hour, the velocity in the
compliance test.  This 5 mile-per-hour difference results in a
36-percent increase in the amount of energy in the system.\4

Second, all active as well as passive safety belts in the automobile
are used in the NCAP test; that is, the test dummies are restrained
by any manual seat belt furnished with the vehicle as well as any
automatic belt or air bag.  In the compliance test, as noted earlier,
only passive restraints (automatic belts and air bags) are used. 

The third and foremost difference between the two programs is the
underlying purpose of the tests.  NCAP is a market-based program that
disseminates information to consumers on the relative safety of
passenger vehicles.  There are no minimum allowable safety
performance criteria that vehicles must meet, although NCAP collects
the same measurements as the compliance test.  Despite the fact that
NCAP is not a compliance program, industry personnel have expressed
the opinion that the NCAP test has become the de facto regulation. 
That is, failure to meet compliance levels on this more stringent
test involving a more forceful collision than the official compliance
test could imply that a vehicle is unsafe.  Currently, nearly all
vehicles tested under NCAP meet the safety requirements specified in
FMVSS 208. 


--------------------
\4 Kinetic energy is a function of mass and velocity (Ek = 1/2 mv\2
).  The additional energy in the NCAP test over the compliance test
derives from the square of the velocity when the mass of the vehicle
is held constant.  Thus (35 miles per hour)\2 is 36-percent greater
than (30 miles per hour)\2 . 


CRASH TEST MEASUREMENT AND
REPORTING
========================================================== Appendix II


   CRASH TEST MEASUREMENTS
-------------------------------------------------------- Appendix II:1

Both the compliance and NCAP tests use anthropomorphic test dummies
to collect data related to injury potential by measuring
accelerations and forces placed on an occupant's head, chest, and
upper leg.\1 Specific levels for each measure, established under
FMVSS 208, represent upper-bound limits for compliance with vehicle
safety requirements.  These ceilings were designed to correspond to
the level at which there is a one-in-six chance of an occupant's
sustaining an injury that poses a serious threat to life. 

The head injury criterion, the measure used in crash tests to assess
potential head injury, was adopted by NHTSA on the basis of research
conducted to establish the likelihood of skull fractures under
different velocity changes.  HIC is measured as a composite of the
axial accelerations of the head (in three dimensions).  Specifically,
HIC is the product of (1) the 2.5 power of the average of the
resultant head acceleration over a time interval not more than 36
milliseconds and (2) that time interval.  The equation for the
function is



A HIC score of 1,000, the highest allowable score for achieving
vehicle compliance, is associated with a one-in-six chance of
sustaining a serious skull injury. 

For determining potential injury to the chest region, chest
acceleration is measured in gravitational units (g's).\2 The
potential for injury to the chest skeletal structure is measured by
the actual resultant peak tridimensional acceleration of the upper
thorax.  Compliance with FMVSS 208 for this measure is set at 60 g's
over 3 milliseconds, an acceleration level that has been associated
with four or more fractured ribs. 

Depending on the type of crash test dummy used, a second chest
measurement, termed chest compression, is taken.  (See appendix III.)
This measures the amount of reduction in the distance between the
sternum and the spinal column and is determined to assess the
likelihood of injury to internal organs.  Currently, only one of the
two types of crash test dummies (Hybrid III) is capable of measuring
this, and the choice of which dummy type to use in a test (either
compliance or NCAP) is made by the manufacturer of the test vehicle. 
If the Hybrid III is used in a compliance test and the vehicle
exceeds the 3-inch maximum reduction distance allowed (the limit
associated with major lacerations to the spleen or kidneys), the
vehicle is considered not to be in compliance with FMVSS 208. 

The final measure taken in both the compliance and NCAP tests is the
compressive force transmitted axially through the upper legs
(femurs).  The femur tolerance level of 2,250 pounds of force is
based primarily on experimental impacts to the lower limbs and is
associated with a one-in-six chance of sustaining a fracture to that
bone. 

When the results of a compliance test exceed the limit for any of the
measures, an investigation is conducted to determine reasons for the
failure and is typically accompanied by a recall or remedy campaign. 
If a determination of noncompliance is made, the model being tested
may not be sold in the United States.  This differs from NCAP as its
tests are not conducted to assess vehicle compliance with federal
regulations, and therefore, no punitive actions may be taken by NHTSA
should a vehicle exceed any of the limits.  Table II.1 lists the four
measurements made in both test programs and presents the maximum
allowable scores under compliance testing for each measure. 



                          Table II.1
           
            Maximum Allowable Scores on FMVSS 208
              Crash Tests for Vehicle Compliance

                                                Maximum
                                                allowable
Measure                                         score
----------------------------------------------  ------------
Head injury criterion                           1,000

Chest acceleration                              60 g's

Chest compression\a                             3 inches

Femur load                                      2,250 pounds
------------------------------------------------------------
\a This measure applies to the Hybrid III dummy only.  (See appendix
III.)


--------------------
\1 Acceleration is the rate of change in velocity with respect to
time, while force is the rate of change in velocity with respect to
time for a given mass. 

\2 One g is equal to 32 feet/second\2 , or 9.8 meters/second\2 . 


   REPORTING NCAP TEST RESULTS
-------------------------------------------------------- Appendix II:2

In 1978 (for the 1979 model year), NHTSA began testing about 30
vehicles per year through its New Car Assessment Program.  While no
manufacturer is required to exceed 30 mile-per-hour standards, the
program, using a 35 mile-per-hour crash test, is designed to inform
customers of the relative crashworthiness of an automobile. 
Traditionally, NCAP reported the actual HIC, chest acceleration, and
femur load scores with a disclaimer that only vehicles within 500
pounds of each other could legitimately be compared.  Also, NCAP
would cite the compliance ceiling levels (1,000 HIC, 60-g chest
acceleration, and 2,250-pound femur load) as representing a
one-in-six chance of sustaining a severe injury.  Despite NHTSA's
claim of overall success in providing information about how well or
how poorly passenger vehicles protect their occupants in crashes,
some critics argued that NCAP's method of reporting test results left
consumers confused. 

In response to fiscal year 1992 Senate Appropriations Committee
requirements, NHTSA performed a user study and began implementing new
methods of informing consumers of the comparative levels of the
safety of passenger vehicles as measured by NCAP.  This new method, a
star chart rating system, is designed to provide consumers with a
quick, simplified, single point of comparison to evaluate vehicles in
the NCAP test.\3

Based upon analyses of a variety of accident injury studies, NHTSA
developed a scale, known as the "Level of Protection Scale," that
relates the probability of sustaining an injury to the level of
protection a vehicle provides its occupants from receiving such an
injury.\4 This scale forms the basis of NHTSA's star chart method for
releasing NCAP test results to the public.  The star chart, which
NHTSA began using in December 1993, reports a range of 1 to 5 stars,
with 5 stars indicating the best crash protection for vehicles within
the same weight class. 

The number of stars a vehicle may be rated is derived from the injury
probabilities associated with the HIC and chest g scores obtained in
the crash tests.  These probabilities are calculated using the
following formulas: 

  Phead = [1 + exp(5.02 - 0.00351 x HIC)]\-1

  Pchest = [1 + exp(5.55 - 0.0693 x Chest Acceleration)]\-1

  Pcombined = Phead + Pchest - (Phead x Pchest)

A vehicle is then assigned a star rating based on its combined injury
risk, with the specific number of stars determined by the range in
which the combined injury risk lies.  The ranges for each star rating
are shown in table II.2. 



                          Table II.2
           
            Star Ratings and Their Combined Injury
                      Probability Ranges

                                                    Range of
Rating                                           probability
------------------------------------------  ----------------
5 stars                                               0 -.10
4 stars                                           + .10 -.20
3 stars                                           + .20 -.35
2 stars                                           + .35 -.45
1 star                                           + .45 -1.00
------------------------------------------------------------

--------------------
\3 The star chart rating system applies only to vehicles tested in
NCAP.  In compliance tests, the actual HIC, chest acceleration, and
femur load results are reported, as the primary purpose of the test
is vehicular compliance to safety regulations and not a comparative
assessment of the likelihood of occupants' sustaining serious
injuries in different vehicle models. 

\4 In SAE Paper No.  851246, "The Position of the United States
Delegation to the International Standards Organization (ISO) Working
Group 6 on the Use of HIC in the Automotive Environment," P.  Prasad
and H.  Mertz presented an injury risk function curve (which this
scale is based upon) that relates the probability of a severe head
injury to HIC. 


ANTHROPOMORPHIC TEST DEVICES
========================================================= Appendix III


   INTRODUCTION
------------------------------------------------------- Appendix III:1

Currently, two types of anthropomorphic test dummies are used in both
the compliance and NCAP tests:  the Hybrid II and Hybrid III
50th-percentile male dummies.  Requirements for both types of dummies
used in compliance testing are specified in 49 C.F.R.  Part 572: 
Anthropomorphic Test Dummies.\1 The design and performance criteria
specified for each dummy type

     "are intended to describe a measuring tool with sufficient
     precision to give repetitive and correlative results under
     similar test conditions and to reflect adequately the protective
     performance of a vehicle or item of motor vehicle equipment with
     respect to human occupants."\2

In this appendix, we discuss the characteristics and instrumentation
of the Hybrid II and Hybrid III 50th-percentile anthropomorphic test
dummies, under the provisions of the NHTSA standards pertaining to
occupant crash protection.  We also provide a comparison of the two
dummy types' performance in NCAP and the compliance test programs. 
Finally, we summarize the 1993 decision to standardize the test
dummy, requiring the mandatory use of the Hybrid III in all NHTSA
crash test programs beginning in 1997. 

In both the compliance and NCAP test, the manufacturer of the vehicle
being tested has the option to choose which type of dummy will be
used.  While both dummies are designed to represent the physical
characteristics of the average adult male, important differences
between them exist.  Despite the differences, the requirements for
vehicular conformance to FMVSS 208 are not different for the two
instruments, with the exception of the chest compression criterion,
which applies only when the Hybrid III dummy is used. 


--------------------
\1 NCAP is not required to follow the standards in part 572;
nonetheless, it does. 

\2 49 CFR Ch.  V (10-1-91 Edition), Part 572-Anthropomorphic Test
Dummies, p.  581. 


   COMPARISON OF HYBRID II AND
   HYBRID III
------------------------------------------------------- Appendix III:2


      DIFFERENCES IN CONSTRUCTION
----------------------------------------------------- Appendix III:2.1

Part 572 of the Federal Motor Vehicle Safety Standards specifies the
types of anthropomorphic dummies to be used in the FMVSS 208
compliance test.  Currently, two specific types of anthropomorphic
test dummies may be used in a compliance crash test:  the Hybrid II
and the Hybrid III.  As specified in subpart B of 49 C.F.R.  part
572, since 1973 the Hybrid II 50th-percentile male test dummy is 5
feet 9 inches tall and weighs approximately 164 pounds, and until
1986, this dummy was used when determining compliance to FMVSS 208. 
In 1986, 49 C.F.R.  parts 571 and 572 were amended to adopt the
Hybrid III 50th-percentile dummy as an alternative to the Hybrid II
for FMVSS testing.  This gave manufacturers the option of using
either the Hybrid II or Hybrid III test dummy as the means of
determining a vehicle's conformance to NHTSA's performance
requirements.  Like its predecessor, the Hybrid III is 5 feet 9
inches tall but weighs slightly more (167 pounds).  Also, like the
Hybrid II, each Hybrid III used in a compliance test must meet the
specifications and performance criteria of part 572 before and after
each vehicle test in order to be an acceptable compliance tool. 

The Hybrid II and Hybrid III use the same instrumentation in the
head, chest, and femurs.  (See figure III.1.) However, according to
General Motors, developer of the Hybrid III, its 50th-percentile male
dummy was designed to improve on Hybrid II technology and
biofidelity.\3 Most experts regard the Hybrid III test dummy as more
biofidelic than the Hybrid II, having a more human-like seated
posture, as well as head, neck, chest, and lumbar spine designs.  The
Hybrid III's responses to crash conditions more closely approximate
the motions associated with human anatomy in crash situations and,
therefore, more accurately evaluate injury risks.  For example, the
improved flexibility of the Hybrid III's neck over the Hybrid II
allows researchers a greater ability to assess the injury potential
of whipping motions. 

In addition to the greater biofidelity, the experts we interviewed
stated that the Hybrid III is more sophisticated technologically than
the Hybrid II because it has more instrumentation for measuring
potential injuries.  Specifically, the Hybrid III is capable of
measuring nearly four times as many forces and accelerations
throughout the body as the Hybrid II.  For example, not only does the
Hybrid III measure injury potential to the skeletal structures, but
it can also determine injury potential to the soft tissues in the
upper thorax through the chest compression measure.  Further, the
Hybrid III has accelerometers and load cells placed in the neck and
lower legs that can measure the potential for injuries caused to
those anatomical areas.  To date, no criteria have been established
for meeting compliance for the additional measures other than chest
compression.  However, the measures do provide DOT with additional
information on the potential physiological responses associated with
vehicular crashes. 

   Figure III.1:  Instrumentation
   in the Hybrid II and Hybrid III
   and the Cartesian and
   Anatomical Coordinates\a

   (See figure in printed
   edition.)

\a Reprinted with permission from SAE publication R-103, copyright
1990, Society of Automotive Engineering, Inc. 


--------------------
\3 The biofidelity of a test dummy is mainly assessed by comparisons
of its response to those of cadavers under the same test conditions. 


      DIFFERENCES IN TEST RESULTS
----------------------------------------------------- Appendix III:2.2

Currently, the determination of which dummy to use in a test is made
by the manufacturer of the vehicle being tested.  Through 1993,
manufacturers chose to use the Hybrid III in 36 of the 133 tests of
passenger cars in the compliance program and in 30 of the 86 tests of
passenger cars in NCAP since the dummy became available for use in
the two programs (1988 and 1990, respectively).  One expert
hypothesized that the reason so few compliance tests involve Hybrid
III dummies is that they tend to receive higher HIC scores,
especially in noncontact situations and that there is no guarantee
that a car designed around the Hybrid II will pass a test using a
Hybrid III. 

The differences between the dummies used in NHTSA's tests, described
above, led us to compare the driver-side HIC and chest acceleration
scores from passenger-car compliance and NCAP tests in an attempt to
quantify the potential effects on test reliability such differences
could create.  In this analysis, we controlled for the presence of a
driver-side air bag in the car.  In general, we found the Hybrid III
dummy scores were lower than the Hybrid II scores, but that the
presence of air bags strongly affects the relative performance of the
two dummy types.  (See tables III.1 and III.2.)



                         Table III.1
           
            Mean HIC and Chest Acceleration Scores
           From NCAP Tests for Hybrid II and Hybrid
                        III Dummies\a


                                      Number          Number
                                          of              of
                               Score   tests   Score   tests
----------------------------  ------  ------  ------  ------
HIC                              793      56     637      30
Air bag                          695      27     511      24
No air bag                       903      29   1,141       6
Chest acceleration (g's)        50.7      56    47.1      30
Air bag                         50.5      27    46.5      24
No air bag                      50.9      29    49.3       6
------------------------------------------------------------
\a For passenger cars from 1990 to 1993, with and without air bags. 

Specifically, we found that

  Vehicles tested with Hybrid III dummies had lower HIC and chest
     acceleration scores than those tested with Hybrid II dummies in
     both compliance and NCAP tests.  In NCAP tests, Hybrid III
     dummies averaged 156 HIC and 3.6 g's less than Hybrid II
     dummies.  (See table III.1) Similarly, in compliance tests, the
     mean head injury criterion score for cars tested with Hybrid III
     dummies was 97 HIC lower than the score for tests that used the
     Hybrid II, while the mean chest acceleration score was about 2
     g's less for test cars that used the Hybrid III.  (See table
     III.2)\4

  In both the NCAP and compliance tests, Hybrid III had significantly
     lower HIC scores than Hybrid II dummies in vehicles equipped
     with air bags.  In vehicles without air bags, Hybrid IIIs had
     significantly higher HIC scores than Hybrid IIs.  The difference
     could occur because of the greater flexibility of the Hybrid
     III's neck.\5

  In both the NCAP and compliance tests, Hybrid III dummies had
     significantly lower chest acceleration results than Hybrid II
     dummies in cars with air bags.  There was little difference
     between the chest scores of Hybrid III and Hybrid II dummies in
     cars without air bags. 



                         Table III.2
           
            Mean HIC and Chest Acceleration Scores
           From Compliance Program Tests for Hybrid
                 II and Hybrid III Dummies\a


                                      Number          Number
                                          of              of
                               Score   tests   Score   tests
----------------------------  ------  ------  ------  ------
HIC                              487      97     390      36
Air bag                          482      47     268      25
No air bag                       492      50     667      11
Chest acceleration (g's)        45.4      97    43.3      36
Air bag                         48.4      47    43.8      25
No air bag                      42.6      50    42.3      11
------------------------------------------------------------
\a For passenger cars from 1988 to 1993, with and without air bags. 

Manufacturers have been reluctant to use Hybrid III dummies for tests
of cars not equipped with air bags because these dummies tend to
produce higher HIC results, especially in cases where the dummy's
head did not contact the interior components of the car.  (Only 18
percent of compliance tests of cars without air bags from 1988 to
1993 and 18 percent of NCAP tests of cars without air bags from 1990
to 1993 used the Hybrid III.) Industry representatives stated that
because HIC was developed to determine potential skull injuries--a
condition that will not occur if the head does not contact the
vehicle's interior--it should be applied only to cases in which the
head actually makes contact.  Although they agree that brain injuries
can occur when the head does not contact the interior, they contend
that the instrumentation in the dummy's head does not measure the
potential for these types of injuries.  Therefore, they conclude that
in cases of "noncontact" HICs, the results are meaningless and
misleading.  Thus, rather than risk a spuriously higher HIC score in
either the compliance or NCAP tests, manufacturers have tended to use
the Hybrid II for vehicles that do not have air bags. 

While these complex interactions of dummy type, safety equipment, and
test conditions can be explained by biomechanical differences between
the Hybrid II and Hybrid III dummies, they may also be explained by
differences in the test vehicles themselves.  As we noted earlier,
the manufacturers specify which dummy type to use in NHTSA crash
tests, and one may assume that, in the absence of other motivations,
they would choose the dummy they anticipate will yield more favorable
results. 


--------------------
\4 All differences except for chest scores in compliance tests were
statistically significant (p < .05). 

\5 Because the neck of the Hybrid III is more flexible, it offers
less resistance to motion.  Assuming two identical crash tests with
the same amount of energy in each system, the Hybrid III's head will
move through its forward arc of motion (flexion) at a greater
velocity than the Hybrid II until it reaches the end of that arc,
when it rapidly decelerates. 


      EXCLUSIVE USE OF THE HYBRID
      III
----------------------------------------------------- Appendix III:2.3

As noted above, each manufacturer undergoing a compliance test may
specify either the Hybrid II or the Hybrid III test device.  But in
recent years, NHTSA has become more convinced that using the Hybrid
III will help ensure that all new vehicles are designed with the
benefit of the most human-like test dummy available.  NHTSA regards
the Hybrid III as more representative of human responses in frontal
crashes, and it can monitor more types of potential injuries as well. 
Further, NHTSA has come to recognize that exclusive use of the Hybrid
III for compliance testing under FMVSS 208 would result in greater
comparability of test results among vehicles produced by different
manufacturers. 

For these reasons, the agency recently issued a Notice of Final Rule
that requires the exclusive use of the Hybrid III for all compliance
testing under standard no.  208.\6 The final rule takes effect
September 1, 1997, to coincide with the date at which all passenger
cars and 80 percent of light trucks must be equipped with air bags
and all light trucks must have passive (automatic) restraint systems. 
NCAP will also switch to exclusive use of the Hybrid III test dummy
beginning with the 1996 model year.  These modifications to the two
programs will create a greater degree of standardization of crash
tests, thereby, in NHTSA's view, increasing the "comparability of
test results among vehicles produced by different manufacturers,
particularly those that now use different dummy types."\7


--------------------
\6 On Nov.  8, 1993, NHTSA published the final rule (Notice 83, 58
Fed.  Reg.  59189), which requires the use of the Hybrid III test
dummy for all compliance testing under standard no.  208, to be
effective Sept.  1, 1997. 

\7 DOT, NHTSA, 49 C.F.R.  Part 571:  Federal Motor Vehicle Safety
Standards; Occupant Crash Protection; 58 Fed.Reg.  214, Nov.  8,
1993, p.  59189. 


TRENDS IN CRASH TEST RESULTS
========================================================== Appendix IV


   CHANGES IN NCAP INJURY RISK
   OVER TIME
-------------------------------------------------------- Appendix IV:1

We conducted analyses of the trends in NCAP test scores from 1979
through 1993 and found that scores have both improved and become more
uniform during the period.  We have expressed NCAP results in terms
of the combined injury risk scores to which NCAP now translates its
HIC and chest scores to produce its new "star system".\1 Figure IV.1
shows the mean injury risks for the driver position, by year, for
model years 1979 through 1993. 

   Figure IV.1:  Mean NCAP
   Combined, Head, and Chest
   Injury Risks by Model Year\ a

   (See figure in printed
   edition.)

\a For all passenger cars.  Does not include light trucks and vans. 

The mean combined injury risk decreased significantly from a high of
0.507 in 1980 to a low of 0.190 in 1993.  The figure also indicates
that the significant reduction in the combined risk derives from a
significant and consistent decrease in the mean head injury risk
probability.  While the mean chest injury risk declined significantly
during the period, it has been relatively stable since 1983.  The
variation between the individual test results has also decreased over
the years.  For example, NCAP head injury criterion scores for
vehicles in 1979 ranged from 521 HIC to 4,513 HIC, whereas in 1993,
the range was between 273 HIC and 1,459 HIC. 

One reason for the decline in the mean combined injury risk and its
accompanying variation over time is the increasingly widespread
installation of air bags.  Cars equipped with air bags had
significantly lower head injury risk probabilities than cars without
air bags.  (See figure IV.2.) Since the first NCAP test of cars
equipped with air bags in 1987, these vehicles have scored an average
head injury risk of 0.063, while cars without air bags have averaged
0.216. 

   Figure IV.2:  Mean NCAP Head
   Injury Risk, With and Without
   Air Bags\a

   (See figure in printed
   edition.)

\a For passenger cars with and without driver-side air bags.  Does
not include light trucks and vans.  Before 1987, no test vehicle was
equipped with an air bag, and only one 1987 model-year test vehicle
was equipped with an air bag. 

There is little difference, however, between the mean chest injury
risks for passenger cars equipped with air bags (0.108) and those
that did not employ this type of restraint (0.120).  (See figure
IV.3.) Given the relatively flat chest injury risk shown in figure
IV.1, it appears that this risk factor, regardless of the type of
restraint, has contributed little to the declining trend for the
combined injury risk. 

   Figure IV.3:  Mean NCAP Chest
   Injury Risk, With and Without
   Air Bags\a

   (See figure in printed
   edition.)

\a For passenger cars with and without driver-side air bags.  Does
not include light trucks and vans.  Before 1987, no test vehicle was
equipped with an air bag, and only one 1987 model-year test vehicle
was equipped with an air bag. 


--------------------
\1 For a complete discussion of the derivation of risk scores and
star ratings see appendix II. 


   CHANGES IN FMVSS 208 INJURY
   RISK OVER TIME
-------------------------------------------------------- Appendix IV:2

We also conducted similar analyses for passenger cars tested in the
compliance program.  Despite fluctuations from year to year, the
combined injury risk did not change significantly from 1987 to
1993.\2 (See figure IV.4.)

   Figure IV.4:  Mean FMVSS 208
   Injury Risk by Model Year\ a

   (See figure in printed
   edition.)

\a For all passenger cars.  Does not include light trucks and vans. 


During the same period a steady, though not statistically
significant, decline in the mean head injury risk was offset by a
significant increase in the mean chest injury risk.  These opposing
trends are associated with the increased installation of air bags,
which are associated with lower head injury risk probabilities and
higher chest injury risk probabilities for compliance tests.  (See
figures IV.5 and IV.6.)

   Figure IV.5:  Mean FMVSS 208
   Head Injury Risk, With and
   Without Air Bags\a

   (See figure in printed
   edition.)

\a For passenger cars with and without driver-side air bags.  Does
not include light trucks and vans.  Before 1988, no test vehicle was
equipped with an air bag. 

   Figure IV.6:  Mean FMVSS 208
   Chest Injury Risk, With and
   Without Air Bags\a

   (See figure in printed
   edition.)

\a For passenger cars with and without driver-side air bags.  Does
not include light trucks and vans.  Before 1988, no test vehicle was
equipped with an air bag. 

The contrasting chest injury risk results between NCAP and compliance
programs may have occurred because of the differences in the
configuration of the two tests--largely from the determination of
which restraint systems are used.  In an NCAP crash that makes use of
all available passive and manual restraint systems, the seat belt
absorbs the dummy's kinetic energy over a gradual period (for a crash
event) before the dummy contacts the air bag.  However, cars with air
bags are not required to have automatic seat belts.  And since manual
seat belts are not engaged in compliance tests, the dummy in the
driver position is not likely to be restrained by a safety belt in
compliance tests of cars with air bags.  The dummy, therefore, is
likely to move forward without a reduction in its kinetic energy,
resulting in a more forceful collision with the air bag than if a
seat belt was also in use.  Over time, as more of the vehicles tested
in the compliance program came equipped with air bags, the mean
compliance chest score increased. 

This is not to say that an air bag-equipped vehicle is less safe than
one that does not have an air bag.  Indeed, the chest g results of
cars with air bags may not be directly comparable to those without
the devices because the distribution of the force loading on the
chest is different for air bags than for safety belts.  Air bags
distribute the load caused by chest contact across a larger surface
area than safety belts.  Nevertheless, the higher chest g result for
an air bag-equipped vehicle is consistent with the view held by many
traffic safety experts that safety belts alone (that is, without air
bags) are much more effective than airbags alone (that is, without
safety belts).\3

The decreasing mean crash test results parallel a similar trend with
annual fatalities, and part of that latter trend can rightfully be
attributed to NHTSA crash tests.  In discussions with industry
representatives, we found that automobile manufacturers attempt to
design vehicles to meet compliance levels for frontal collisions in
the NCAP test, as well as ensure that all other safety criteria are
met.  In addition to "live" crash tests, manufacturers use computer
models of frontal collisions, rear and side impacts, and roof crush
to simulate NHTSA crash tests and to ensure that the cars meet NHTSA
standards.  The types of simulations range from models of specific
components of the vehicle to nonlinear finite-element models, which
incorporate all specifications of the automobile and can predict
interactions between the car and its occupants during a collision. 
These simulations allow manufacturers to gain insight into the
deformation of the vehicle, likely intrusions into the occupant
compartment, and the force loads generated by various structural
components.  Though computer- simulated crashes are expensive because
they generally require access to a supercomputer, they do allow
manufacturers to gain knowledge of how the car will perform and how
to correct problems before building prototypes.  They are also much
less costly and less time-consuming than building and crashing
prototypes.  In addition, one simulation expert stated that the
results of finite-element simulations generally reflect the results
obtained in actual crash tests and that the industry uses crash tests
in part to validate its computer-based crash models. 


--------------------
\2 Though the compliance program does not report scores in terms of
risk, we translated the results to risk probabilities to give the
reader a base scale from which to compare the differences in scores
that might result from the differences in the velocities and
restraint usage in the two crash test programs. 

\3 The relative protection afforded by safety belts and airbags is
discussed in Highway Safety:  Causes of Injury in Automobile Crashes
(GAO/PEMD-95-4; May 1995). 


RELIABILITY OF NCAP RESULTS
=========================================================== Appendix V

Our analyses of the trends in NCAP test results have shown that (1)
injury risk probabilities have declined over time and (2) the
variation between test results has lessened over time.  (See appendix
IV.) It would seem that cars have become more crashworthy--at least
as measured by NCAP--and that this improved crashworthiness is more
uniformly distributed across the passenger car fleet.  Indeed, in
1979 the combined injury risk for NCAP- tested vehicles ranged from
0.106 to 1.0 (rounded), whereas in 1993, the ends of the distribution
ranged from 0.096 to 0.581.  Despite the decreased variation in test
results, however, the difference between the highest and lowest risk
probabilities is still substantial. 

This variation is open to two quite different interpretations.  It
may indicate the sensitivity of crash tests to real differences
between vehicle models, or it may reflect the imprecision of the test
scores.  In classical measurement theory, reliability is defined as
the repeatability of test results.  The reliability of crash tests
would be estimated by comparing the results from repeated crash tests
of the same model vehicle. 

On only one occasion has NHTSA attempted to determine whether NCAP
crash test results are, in fact, reproducible by crashing a single
model on multiple occasions.  In this study, 12 consecutively
manufactured 1982 Chevrolet Citations were crash-tested by three test
facilities, with each facility testing four vehicles, in an attempt
"to quantify the degree of variation, as well as develop generalized
statistical conclusions about test repeatability."\1 The mean HIC
score for the 12 tests was 685, with the scores ranging from 495 HIC
to 954 HIC.  NHTSA identified several sources of variation in results
derived from the test procedure, as well as from the testing
facilities, the test instrumentation, the test dummy used, and the
individual vehicles.  NHTSA could not quantify the amount of
variation attributable to each of these five areas because of the
number of possible sources of error within each. 

Although the amount of variation that can be attributed to any of the
sources of error is incalculable, the confounding interactions of
accumulated error lead to questions about the reliability of NCAP
results.  The variation between different units was artificially
constrained by selecting 12 consecutively manufactured Chevrolet
Citations, yet the head injury results still had a range of 459
HIC.\2 This variation among the HIC scores implies that two scores
with a range of less than 219 are not, in statistical terms,
significantly different and that any score between 781 HIC and 1,219
HIC is not significantly different from 1,000.\3

Table V.1 illustrates how this level of reliability could affect the
interpretation of other crash test scores.  The table displays the
mean HIC score from the 12 Citation crash tests and the HIC scores of
four other vehicles of similar weight.\4 It also indicates whether
the NCAP HIC scores are significantly different from (1) the mean
Citation score and (2) the putative ceiling of 1,000.\5 The HIC
scores received by all the vehicles except the 1990 Lexus are
significantly lower than 1,000.  However, there is no statistically
significant difference between the mean HIC score received by the
Citation and three of the four other vehicles. 



                          Table V.1
           
                 Mean HIC Scores From NHTSA's
           Repeatability Test and NCAP Results for
                 Vehicles of Similar Weight\a


                            Curb            1982
                          weight          Citati       HIC =
Vehicle                 (pounds)     HIC      on       1,000
--------------------  ----------  ------  ------  ----------
1982 Chevrolet             3,260     685      \b         Yes
 Citation
1988 Oldsmobile            3,460     710      No         Yes
 Delta 88
1990 Lexus ES250           3,280     992     Yes          No
1991 Ford Taurus           3,290     480      No         Yes
1991 Chrysler New          3,310     511      No         Yes
 Yorker
------------------------------------------------------------
\a From NHTSA's 12 1982 Chevrolet Citation tests and NCAP results for
vehicles of similar weight. 

\b Does not apply. 

After reviewing a draft of this report, NHTSA provided us with a
second set of data that could shed additional light on the
reliability of NCAP scores.  These data represent the results of
crash tests of model-year 1991 through 1994 vehicles conducted by
automobile manufacturers in tests that essentially duplicate the NCAP
test conditions.  The data were voluntarily submitted to NHTSA before
planned NCAP tests. 

We compared the manufacturer scores with those obtained from NCAP
after translating them into the single injury probability score that
serves as the basis for NHTSA's recently introduced star rating
system.  (See appendix II.) We found that a statistically significant
first-order correlation exists
(r = .72) between the two sets of injury risk probabilities. 

We then compared the distributions of star ratings derived from NCAP
and manufacturers' tests.  Table V.2 compares the star ratings for
the driver position, and table V.3, for the passenger position.  If
agreement between the two sets of tests had been perfect, all events
in the tables would have fallen on the diagonals from upper left to
lower right.  In actuality, star ratings are the same for
approximately one-half of the vehicle models tested (55 percent for
the driver position and 45 percent for the passenger position). 
Differences of one or more stars exist between manufacturer and NCAP
ratings for about one-half of the tests; 8 percent of the vehicle
models have differences of two or more stars. 



                                    Table V.2
                     
                      Number of Vehicles Within Star Rating
                           Categories, Driver Position


                                               2       3       4       5
NCAP star rating                  1 star   stars   stars   stars   stars   Total
--------------------------------  ------  ------  ------  ------  ------  ------
1 star                                 6       0       3       3       0      12
2 stars                                0       1       3       1       0       5
3 stars                                1       1      16      14       0      32
4 stars                                0       1      15      37       6      59
5 stars                                0       0       0       6       5      11
================================================================================
Total                                  7       3      37      61      11     119
--------------------------------------------------------------------------------


                                    Table V.3
                     
                      Number of Vehicles Within Star Rating
                          Categories, Passenger Position


                                               2       3       4       5
NCAP star rating                  1 star   stars   stars   stars   stars   Total
--------------------------------  ------  ------  ------  ------  ------  ------
1 star                                 2       1       1       2       0       6
2 stars                                1       0       3       3       0       7
3 stars                                1       2       8      23       1      35
4 stars                                1       0      10      33       6      50
5 stars                                0       0       0      11      10      21
================================================================================
Total                                  5       3      22      72      17     119
--------------------------------------------------------------------------------
As appendix II explains, each of NHTSA's star ratings represents a
range of injury probability.  For example, a rating of 4 stars
indicates that in a crash situation similar to that tested by NHTSA
the probability of serious injury to an individual is between 1 in 10
and 2 in 10.  The solid bars in figure V.1 depict these probability
ranges for each star rating.  The lines attached to the bars
represent "confidence intervals" that we estimated from the standard
deviation of the absolute difference between the combined injury
risks for drivers derived from manufacturer and NCAP tests.  These
confidence intervals represent the estimated range of injury
probability within which a vehicle with a nominal rating could be
expected to vary if tested again.  For example, a 4-star rating could
be associated with a vehicle with a "true" injury probability between
zero and 0.363.  This range overlaps the confidence intervals
associated with the 5- and 2-star ratings. 

Our estimates of the reliability of NCAP crash tests are based on the
only two sources of relevant information we are aware of:  the
repeated crashes of the 1982 Chevrolet Citation, and a comparison of
manufacturer and NCAP test scores from 1991 to 1994.  Neither of
these sources provides ideal information for precisely quantifying
the measurement error associated with NCAP scores.  We do not know
how well the results of the Citation experiment can be applied to
vehicles manufactured and crash-tested 10 years later.  While the
manufacturer-NCAP comparison applies to a large number of late-model
cars, we cannot be sure how well the manufacturers succeeded in
replicating NCAP crash test conditions in each case or to what extent
the results from other manufacturer tests varied from the ones
reported to NHTSA. 

   Figure V.1:  Risk of Serious
   Injury, With Estimated
   Confidence Interval, by NCAP
   Star Rating\a

   (See figure in printed
   edition.)

\a Confidence interval is based on 1.96 standard deviations of the
absolute difference between the combined injury risk scores derived
from manufacturer and NCAP tests of the same vehicle model. 

Nevertheless, both analyses support the same conclusion that NCAP
scores, whether reported in raw HIC and chest acceleration scores or
as categories of injury probability, have associated levels of
imprecision.  As a result, substantial differences in scores between
two test results (100-200 HIC, or a 1-star--or possibly
2-star--rating difference) may not represent true differences in
crashworthiness. 


--------------------
\1 John M.  Mackey and Charles L.  Gautier, Results, Analysis and
Conclusions of NHTSA's 35 MPH Frontal Crash Test Repeatability
Program, SAE Paper No.  840201 (Warrendale, Pa.:  SAE, 1984), p.  74. 

\2 This constraint was imposed by NHTSA to limit, as far as possible,
sources of error outside the test process itself.  The results of the
12 tests, however, revealed differences in the crush of the test
vehicles, for example.  (The "crush" is the distance by which the car
is collapsed by the collision--from the front of the car moving
rearward.) Thus, error derived from the manufacturing of the vehicles
still influenced the overall variation among tests, but NHTSA could
not quantify its influence. 

\3 These ranges derive from a calculation of a 95-percent confidence
interval using the mean and standard deviation of the results from
NHTSA's repeatability study.  A similar study of seven 1983 Volvo 760
GLEs conducted by the manufacturer revealed a somewhat narrower,
though still substantial, range in HIC scores (from 697 to 1,004) and
indicated that two HIC scores with a range of as much as 147 are not
significantly different. 

\4 NHTSA cautions that scores from vehicles more than 500 pounds
different are not necessarily comparable. 

\5 For these comparisons, we assume that the other models would
demonstrate the same HIC score variations in multiple crash tests as
the Citation.  While this assumption may be subject to debate, we
have no other data from which to form a different assumption.  We use
HIC scores here rather than combined injury risk since the chest g
scores from the test were unavailable. 


METHODOLOGY FOR ANALYZING THE
PREDICTIVE VALIDITY OF NCAP
========================================================== Appendix VI


   INTRODUCTION
-------------------------------------------------------- Appendix VI:1

The overall purpose of both the compliance and NCAP crash tests is to
determine the crashworthiness, or safety, of passenger vehicles. 
This implies, therefore, that a relationship exists between the
results of crash tests and real-world injuries and fatalities.  To
examine this issue, we conducted two analyses comparing results
derived from NCAP tests to those from national accident databases.\1
Specifically, our analysis compared NCAP results with traffic injury
and fatality information from the National Accident Sampling System
(NASS) and the Fatal Accident Reporting System (FARS).  This appendix
details the methodologies and results of both analyses. 


--------------------
\1 For these analyses, we decided to use results only from NCAP for
two reasons.  First, NCAP has been crash-testing vehicles since 1979
and had conducted about 340 tests of passenger cars through 1992. 
The compliance program has been in existence only since 1987 and had
conducted only 145 tests.  Second, presumably because of the higher
velocity used in the NCAP tests, the results of NCAP tend to vary to
a greater extent than those of the compliance program, and a large
portion of vehicles tested in NCAP have HIC and chest acceleration
scores in excess of the ceiling values of 1,000 and 60 g's,
respectively, while these scores tend to cluster far below the
ceiling values for the compliance tests. 


   RELATIONSHIPS BETWEEN NCAP AND
   NASS DATA
-------------------------------------------------------- Appendix VI:2

Data from the National Accident Sampling System for 1988 through 1991
were combined to determine whether the results of New Car Assessment
Program tests are good predictors of serious injuries and fatalities
in real-world automobile crashes.  The NASS is a sample of annual
police-reported accidents involving passenger cars, light trucks, and
vans that had to be towed because they were damaged.  The NASS year
corresponds to the calendar year rather than the automobile
industry's model year, and emphasis is placed on the most recent 5
model-year vehicles.  We chose this data system for two reasons:  (1)
It is a national database that contains information on all types of
automobile collisions, and (2) it is the only national database that
reports a vehicle's change in velocity (delta v)--the best available
indicator of accident severity--resulting from the collision. 

We reduced the NASS data sets to single-vehicle and two-vehicle
accidents and then combined them into one data set.  This resulted in
a total of 14,253 vehicles in the final data set.  We then matched
the results of the NCAP crash tests to the vehicles in our NASS file. 
NCAP results from 1983 through 1992 were chosen for the analysis for
two reasons:  (1) We assumed that crash test results are applicable
for a given time (as cars age, their crashworthiness may decrease
owing to wear and tear) and (2) NASS data were available for the
calendar years 1988 through 1991.  Because NASS emphasizes for
inclusion collisions involving vehicles from the 5 most recent model
years, we chose 5 years as the period of time at which a test score
no longer applies; therefore, NCAP scores from 1983 were the earliest
we included in the analysis.  In addition, each NASS year has a
half-year's data from the following model year, as the model year
usually begins in late summer.  Therefore, the 1991 NASS had some
accidents involving 1992 model-year automobiles. 

The NCAP results were matched to the NASS data set on the following
criteria:  the make of the vehicle (that is, the manufacturer), the
model, the model year, and the body type (sedan, convertible, and so
forth).  In cases in which a specific make, model, model year, and
body type were tested on more than one occasion, only the first test
was used.  Results for models with corporate twins (vehicles with
platforms identical to one another but sold under different model
names--for example, Ford Taurus and Mercury Sable) were projected to
the twins.  The resultant data set for our analyses contained 1,985
cases.  When weighted by NASS sampling weights, these represented
more than 9 million accidents. 

We conducted logistic regression analyses to determine whether a
relationship exists between serious injuries and fatalities in actual
automobile collisions and results from crash tests conducted for
NCAP.  We analyzed crashes in which damage to the right front, left
front, or full front of the vehicle occurred.  Single-car and two-car
crashes were examined separately.\2 We conducted two sets of
analyses:  one on the unweighted sample in our dataset and a second,
which incorporated the NASS sampling weights.\3

Our analyses of the relationship between real-world traffic injuries
and fatalities and NCAP injury risk probabilities were limited to
drivers of passenger cars (two-door and four-door sedans, coupes,
hatchbacks, and convertibles).\4 We restricted our analyses to
"restrained" drivers--that is, drivers who made proper use of either
a manual or an automatic seat belt or whose air bag deployed during
the crash. 

The dependent variable used in the analyses was constructed from NASS
injury codes to represent whether the driver of an NCAP passenger car
involved in a crash either died or was hospitalized for at least 1
day specifically because of the crash.  This was coded as a
dichotomous variable, with those who died or were hospitalized
receiving a 1 and all other nonmissing values receiving a zero.  The
independent variable of interest was the combined injury risk score
associated with specific vehicle models as derived from the HIC and
chest acceleration scores from NCAP tests.  (See appendix II.)
However, because characteristics of the driver and the vehicle and,
most importantly, the severity of the crash (as measured by the total
change in velocity, or delta v) are associated with the likelihood of
injuries and fatalities, we included occupant characteristics (age,
gender), vehicle characteristics (curb weight and, in two-car
crashes, the weight of the other vehicle), and crash severity (delta
v) in our logistic regression models.\5

Tables VI.1 and VI.2 present the results of our analyses of the
unweighted sample for one- and two-car crashes.  The predictive power
of delta v dominates both models, but the driver's age and the car's
weight also appear as significant predictors of injury in two-car
crashes.  In these crashes, older drivers and drivers of lighter cars
were more likely to suffer injury or death.  In neither model was the
NCAP injury risk significantly related to hospitalization or death in
either one- or two-car crashes. 



                          Table VI.1
           
                Logistic Regression Predicting
            Hospitalization or Death in Single-Car
             Frontal Crashes, Unweighted Sample\a

                                  Standa    Wald
                                      rd  statis  Significan
Variable                    Beta   error     tic    ce level
------------------------  ------  ------  ------  ----------
Injrisk                        -  5.0594  1.8243       .1768
                          6.8337
Age                       0.0004  0.0245  0.0002       .9883
Gender                    0.2689  0.4369  0.3790       .5381
Curbwgt                        -  0.1404  0.5287       .4672
                          0.1021
Dvtotal                   0.1580  0.0567  7.7579       .0053
Constant                  0.4380  3.9979  0.0120       .9128
------------------------------------------------------------
Legend

Injrisk = Driver injury risk
Age = Age of driver
Gender = Gender of driver
Curbwgt = Vehicle's curb weight
Dvtotal = Total change in velocity (mph)

\a Represents 46 restrained drivers. 



                          Table VI.2
           
                Logistic Regression Predicting
             Hospitalization or Death in Two-Car
             Frontal Crashes, Unweighted Sample\a

                                  Standa    Wald
                                      rd  statis  Significan
Variable                    Beta   error     tic    ce level
------------------------  ------  ------  ------  ----------
Injrisk                   1.2413  1.0302  1.4519       .2282
Age                       0.0430  0.0142  9.1362       .0025
Gender                         -  0.2388  0.1072       .7434
                          0.0782
Curbwgt                        -  0.0611  7.6327       .0057
                          0.1689
Othvehwgt                      -  0.0338  0.0017       .9671
                          0.0014
Dvtotal                   0.2311  0.0516  20.036       .0000
                                               8
Constant                       -  1.8460  2.2231       .1360
                          2.7524
------------------------------------------------------------
Legend

Injrisk = Driver injury risk
Age = Age of driver
Gender = Gender of driver
Curbwgt = Vehicle's curb weight
Othvehwgt = Weight of other vehicle
Dvtotal = Total change in velocity (mph)

\a Collisions with vehicles weighing less than 10,000 pounds. 
Represents 131 restrained drivers. 

Tables VI.3 and VI.4 present the findings from the weighted sample
and show very similar results to the unweighted sample.  The
strongest predictor remains the crash severity, and in two-car
crashes, driver age is related to collision outcomes.  The weighted
sample presents two different conclusions from the unweighted sample,
however.  The curb weight of the vehicle falls short of statistical
significance by traditional criteria, and more to the point, a
significant relationship between NCAP risk scores and death or
hospitalization appears. 



                          Table VI.3
           
                Logistic Regression Predicting
            Hospitalization or Death in Single-Car
              Frontal Crashes, Weighted Sample\a

                                  Standa    Wald
                                      rd  statis  Significan
Variable                    Beta   error     tic    ce level
------------------------  ------  ------  ------  ----------
Injrisk                        -  11.277  1.6075       .2230
                          14.298       0
                               0
Age                            -  0.0332  0.8135       .3804
                          0.0300
Gender                    0.4028  1.5665  0.0661       .8004
Curbwgt                        -  0.1546  1.7118       .2092
                          0.2023
Dvtotal                   0.2672  0.0690  15.040       .0013
                                               9
Constant                  2.3599  6.3191  0.1395       .7137
------------------------------------------------------------
Legend

Injrisk = Driver injury risk
Age = Age of driver
Gender = Gender of driver
Curbwgt = Vehicle's curb weight
Dvtotal = Total change in velocity (mph)

\a Represents 8,401 restrained drivers. 



                          Table VI.4
           
                Logistic Regression Predicting
             Hospitalization or Death in Two-Car
              Frontal Crashes, Weighted Sample\a

                                  Standa    Wald
                                      rd  statis  Significan
Variable                    Beta   error     tic    ce level
------------------------  ------  ------  ------  ----------
Injrisk                   3.2379  1.3351  5.8817       .0275
Age                       0.0635  0.0200  10.093       .0059
                                               3
Gender                    0.9361  0.6075  2.3752       .1429
Curbwgt                        -  0.0702  3.4238       .0828
                          0.1298
Othvehwgt                      -  0.0587  0.5435       .4717
                          0.0433
Dvtotal                   0.2860  0.0891  10.307       .0055
                                               8
Constant                       -  2.6253  7.4875       .0146
                          7.1838
------------------------------------------------------------
Legend

Injrisk = Driver injury risk
Age = Age of driver
Gender = Gender of driver
Curbwgt = Vehicle's curb weight
Othvehwgt = Weight of other vehicle
Dvtotal = Total change in velocity (mph)

\a Collisions with vehicles weighing less than 10,000 pounds. 
Represents 23,514 restrained drivers. 

Some degree of doubt must be associated with these findings because
of the nature of the sample on which they are based.  NASS uses a
highly complex stratified sampling design to achieve national
representativeness for its relatively small sample of observations. 
The NASS database we used contained 21,377 observations, which when
properly weighted, represent more than 9 million accidents. 
Unfortunately, we found only 366 instances of NCAP-tested cars that
met our criteria of properly restrained drivers, and only about
one-third of these could be used because of missing values on one or
more variables.  This drastic reduction in sample size, when combined
with the highly uneven distribution of missing values across sampling
strata, makes the sampling weight associated with any observation of
doubtful validity. 


--------------------
\2 "Two-car crashes" refers to collisions between two passenger cars,
or a passenger car tested in NCAP and either a light truck or van as
defined by the NASS. 

\3 To analyze the weighted sample, we used the Survey Data Analysis
(SUDAAN) statistical package, which takes into account the
stratification and unequal selection probabilities inherent in the
sampling design of NASS. 

\4 Although we included light trucks and vans in determining the
sample for two-car collisions, we analyzed the relationship between
NCAP scores and traffic injuries and fatalities for passenger cars
only. 

\5 See Highway Safety:  Have Automobile Weight Reductions Increased
Highway Fatalities?  (GAO/ PEMD-92-1; Oct.  1991); Highway Safety: 
Factors Affecting Involvement in Vehicle Crashes (GAO/ PEMD-95-3;
Oct.  1994); and Highway Safety:  Causes of Injury in Automobile
Crashes (GAO/ PEMD- 95-4; May 1995). 


   RELATIONSHIPS BETWEEN NCAP AND
   FATALITY RATES
-------------------------------------------------------- Appendix VI:3

To overcome the statistical limitations of our NASS database, we
turned to the Fatal Accident Reporting System (FARS).  By using FARS,
we looked to substantially increase the number of usable cases in the
analysis in that FARS contains information on all accidents in a
given year that involve at least one fatality (about 45,000 cases per
year), while NASS contains only a sample of all accidents (about
3,000 cases per year).  In addition, we reasoned that while FARS
lacks the information on crash severity provided by NASS' estimate of
the total change in velocity, its severity of crashes was relatively
homogeneous because the database is restricted to fatal--presumably
severe--crashes. 

For this analysis, only passenger cars from the 1982 to 1991 model
years were included.\6 In addition to the actual test vehicles, our
analysis included vehicles that had no substantial structural changes
in model years following the tested model year.  That is, if a 1984
model-year vehicle were tested in NCAP and no structural changes were
made to the 1985 version of the vehicle and it was not retested, the
1985 model year was assigned the same combined injury risk score as
the 1984 vehicle. 

We then matched the vehicles to the FARS and the R.L.  Polk Vehicle
Registration System (Polk) databases based on the make (that is, the
manufacturer), model, model year, and body type of the vehicle.  The
FARS database is a compilation of all automobile accidents in the
United States in any given calendar year in which at least one
fatality occurred.  The Polk system is a database that contains
information on the types, numbers and weights of vehicles registered
in a given calendar year.  Data for both systems are for the calendar
years 1987 through 1991.  As with the analysis of NASS data, we
restricted this analysis to one- and two-car frontal collisions in
which the driver of the NCAP-tested vehicle was restrained by either
a seat belt or an air bag. 

Having matched the NCAP vehicles to the FARS and Polk systems, we
then calculated the fatality rates for the vehicles.  This was done
simply by dividing the number of fatalities by the number of
registered vehicles.  The fatality rates in our analysis are
expressed in terms of fatalities per 100,000 registered vehicles. 

We then correlated the driver combined injury risk scores and
fatality rates associated with vehicle models in a number of ways. 
First, we calculated a simple correlation using just information on
those elements.  Next, we regressed fatality rates on additional
characteristics associated with vehicles using a Poisson model, which
allows one to compare rates of individual cases, especially when the
sample size is moderately large and the probability of an event
occurring is either very low or very high.  This type of analysis fit
our needs in that a large number of cases (884) were included in our
analyses while the fatality rates of the vehicles included were low
(overall, there were 1,036 deaths for approximately 19 million
registered vehicles). 

The variables added to the model held information on the model year
and body style of the vehicles in the dataset.  We controlled for the
model year as a proxy for certain driver, vehicle, and roadway
characteristics that could not be included in the model.  We
controlled for the body style for two reasons:  (1) as a surrogate
for the relationships found between specific body styles and certain
driver characteristics and (2) as a rough surrogate for the weight of
the vehicle.\7

As a final analysis, we divided the NCAP injury risk distribution
into quintiles and compared the fatality rates of the different
groups.  Each quintile represented one-fifth of the passenger cars
tested in NCAP from 1982 to 1991. 

We found that a first-order correlation between NCAP injury risk and
fatality rates exists (p = .007).  When information on the body style
and model year of the vehicle was included in the analysis, the
strength of the relationship increased (p = .001).  However, the
relationship appears to be the result of the high fatality rates
associated with the poorest performers in NCAP.  Indeed, vehicle
models within the highest quintile of injury risk (those in the
highest 20 percent of the distribution) had significantly higher
fatality rates than all other quintile categories.  Further, we found
that the worst performers on the NCAP test had injury risk
probabilities approximately eight times higher than the best-scoring
cars, while their fatality rates were almost 28 percent higher.  The
remaining four quintile groups, on the other hand, were not
significantly different from one another.  (See tables VI.5 and
VI.6.) Thus, it seems that the relationship between driver fatality
rates and predicted injury risk stems from the significantly higher
fatality rates associated with vehicles that have very high NCAP
injury risk probabilities. 



                          Table VI.5
           
             Mean Combined Injury Risk Scores and
             Fatality Rates for All Body Styles\a

                                                    Fatality
                                  Mean combined         rate
                                    injury risk       (FARS/
Quintile                                 (NCAP)        Polk)
------------------------------  ---------------  -----------
1                                          .093         .503
2                                          .140         .577
3                                          .192         .573
4                                          .307         .497
5                                          .723         .642
------------------------------------------------------------
\a Fatality rate calculated by dividing the number of fatalities in a
given risk quintile by the total number of registered vehicles in
that quintile.  Fatality rates are expressed as fatalities per
100,000 registered vehicles.  The mean combined injury risk is the
average NCAP injury risk for the quintile.  Body styles include
coupes, sedans, two- and four-door hatchbacks, and station wagons. 



                          Table VI.6
           
             Z-scores of Differences Between Mean
            Fatality Rates of Combined Injury Risk
                         Quintiles\a



Quintile                   1       2       3       4       5
--------------------  ------  ------  ------  ------  ------
1                         \b    .363    .224    .813   2.745
2                      -.363      \b   -.150    .520   2.684
3                      -.224    .150      \b    .635   2.690
4                      -.813   -.520   -.635      \b   2.119
5                          -       -       -       -      \b
                       2.745   2.684   2.690   2.119
------------------------------------------------------------
\a Mean fatality rates as reported in table VI.5.  A z-score of 1.96
or greater shows significant differences between the means at a
probability level of at least p = .05. 

\b Not applicable. 



(See figure in printed edition.)Appendix VII

--------------------
\6 Convertibles, light trucks, vans, and multipurpose vehicles were
excluded from the analysis. 

\7 For a discussion of vehicle and driver characteristics associated
with accident involvement, see Highway Safety:  Factors Affecting
Involvement in Vehicle Crashes (GAO/PEMD-95-3, Oct.  1994). 


COMMENTS FROM THE DEPARTMENT OF
TRANSPORTATION
========================================================== Appendix VI



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)

See comment 10. 



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)

See comment 2. 

See comment 2. 

See comment 6. 

See comment 13. 



(See figure in printed edition.)

See comment 9. 

See comment 14. 



(See figure in printed edition.)

See comment 15. 



(See figure in printed edition.)

See comment 4. 

See comment 16.

See comment 17. 

See comment 17. 



(See figure in printed edition.)


The following are GAO's comments on the letter from the Department of
Transportation dated July 13, 1994. 

GAO COMMENTS

1.  We agree with NHTSA that the terms "reliability" and "validity,"
as used in the report, refer to their statistical meanings, test
repeatability, and predictive validity, respectively.  We share the
concern that common usage of the terms "could have damaging effects"
on NCAP's credibility and mislead a casual user to conclude that the
tests have not had positive effects on the crashworthiness of the
U.S.  passenger car fleet.  For this reason, we have modified the
subtitle of the report. 

NHTSA is correct in asserting that we use the term reliability in its
traditional meaning of reproducibility.  We disagree with NHTSA's
implication that this meaning is not relevant to "the relative safety
performance of vehicles." Indeed, as our report indicates, the band
of uncertainty that surrounds crash test scores (as it does any test
results) can affect the relative ranking of vehicles. 

2.  We recognize that model lines would have NCAP result variations
unique to themselves, and the report clearly states this caveat (see
p.  38, footnote 5).  After completing the Citation experiment, NHTSA
made changes in its test procedures to improve their reliability. 
Unfortunately, no equivalent test of how effective these changes were
in reducing the variability between test scores was subsequently
performed. 

After NHTSA had provided its official comments on our draft report,
the agency also provided us with crash test results from automobile
manufacturers for model-year 1991 through 1994 vehicles.  These data
were provided to NHTSA in preparation for tests conducted under NCAP
and were results of tests that essentially duplicated the NCAP
testing procedure.  The agency had previously declined to provide
this information because they considered it proprietary.  We analyzed
these data and have included our findings in the body of the report
(see pp.  39-41).  Though this information cannot define the
boundaries of NCAP reliability, the difference between manufacturer
and NCAP results reinforces our conclusion that the reliability of
NCAP results is limited. 

3.  We do not disagree with NHTSA that rigorous protocols for crash
testing are followed and that NHTSA verifies the results of the crash
test with high-speed film.  However, this process merely verifies
that the accelerometers placed in the test dummy accurately recorded
the data from the specific trial.  It does not address the issue of
reliability, which in classic statistical theory holds that test data
are reliable if consistent results are obtained through repeated
trials of an experiment using specific procedures.  In the case of
NCAP crash testing, a specified procedure exists, but the model of
the test vehicle changes.  Unless multiple trials of the same model
line are conducted, we cannot determine the reliability of test
results. 

We are sensitive to the costs the Agency could incur in addressing
our recommendation that it update and publish, in clear language, its
knowledge of the reliability of NCAP results.  For this reason, we
suggest that the Agency explore alternative means for accomplishing
this goal, in particular by making use of the knowledge base
developed by manufacturers.  Regardless of the method the Agency uses
to address the recommendation, its purpose will not be, as NHTSA
suggests, to "enhance the scientific reliability of its data" or
"narrow its standard deviation," but to assure American consumers
that they are provided with accurate information about the relative
crashworthiness of vehicles. 

4.  We agree that NHTSA has found a statistically significant
difference between the fatality risk of belted drivers involved in
two-car frontal collisions in cars with "good" NCAP performance and
those that were "poor" performers.  In our analysis of the fatality
rates, though using a different methodology, we found similar
results. 

The analyses we performed shared a common weakness with NHTSA's;
namely, they were both limited to a relatively small proportion of
real-world crashes.  NHTSA's estimate of reduced fatality risk for
better scoring NCAP cars is derived from analyses using only two-car
crashes in which both drivers were belted and at least one occupant
was killed.  These conditions limited their analyses to between 81
and 170 crashes.  (NHTSA's database was drawn from the 1979 through
1991 FARS years, which represent between 40,000 and 50,000 highway
fatalities annually.)

Our analysis was also limited to NCAP cars involved in fatal
accidents with restrained drivers, although we also included
single-vehicle crashes and FARS data from 1987 through 1991.  These
limitations reduced our sample to 884 cars.  Both NHTSA and we agree
that a statistically significant correlation between NCAP scores and
real-world crashes can be found but, to use NHTSA's words, the
correlation is "far from perfect."\1 Our analyses suggested that this
correlation derived from the fatality rates of the worst scoring
cars, and not from crashworthiness differences among relatively good
NCAP performers. 

5.  We generally agree with NHTSA's comment that the improvement in
NCAP scores over time has contributed to an improvement in highway
safety.  However, many other influences unrelated to crash testing,
such as safety belt usage laws, and the toughening of drunk driving
laws, have also contributed to this trend. 

6.  Our report does not address the purported ease of interpretation
associated with the new star rating system.  However, we did
incorporate NHTSA's new reporting system into our analysis of the new
data provided us by NHTSA.  Our findings provide detail to support
the conclusion of our draft report:  that a reporting system can be
no more reliable than the scoring system on which it is based. 

We disagree that the new star rating system "eliminates .  .  . 
[the] implied precision" of HIC, chest, and femur scores.  It is true
that some cars with nonsignificant differences in scores would end up
in the same category under the new system, and thus correctly be
presented to the public as roughly equal in crashworthiness. 
However, it is also true that other cars with nonsignificant score
differences could be placed in different categories, a scoring
artifact that incorrectly implies substantial differences in the
relative levels of crash protection provided by the vehicles.  For
example, while a vehicle with a chest g of 40 and a HIC of 550 will
receive 5 stars, one with a chest g of 40 and a HIC of 555 will
receive 4 stars.  We do not believe that this difference in HIC
scores implies an actual crashworthiness difference. 

7.  NHTSA appears to suggest that it has already complied with this
recommendation through the adoption of its new reporting system.  We
disagree.  The new system, while seemingly clear, does not
communicate to the public the band of uncertainty associated with
star ratings.  Our analyses of the manufacturers' NCAP test results
suggests that this band is sizable and illustrates its potential
effects.  However, additional information needs to be collected and
analyzed before the precision of crash test results can be adequately
defined. 

8.  The recommendation is to explore alternative methods for
determining the crashworthiness of vehicles.  We cited computer
simulations as one promising avenue to explore.  We recognize the
limitations of the current capabilities of computer simulations (the
high cost of supercomputers, the complexity of programming, and so
on), and we agree with NHTSA that this technology could not replace
actual crash tests in the near future.  However, NHTSA's comment
suggests that it has examined the potential benefits of this rapidly
emerging technology and has dismissed them.  We believe that it
should continue to monitor and periodically reassess them. 

It appears to us that as the technology develops and becomes less
costly, the potential benefits of such a system extending to a much
larger set of crashes than NCAP now considers at some point in the
future may outweigh its costs. 

Other possible approaches include ones that NHTSA is already
considering, such as extending the range of tests to include both
side-impact crashes and frontal-offset crashes.  While such tests
would expand the applicability of NCAP tests to a larger portion of
real-world events, they would also substantially increase the costs
of the program.  This consideration reinforces our belief that the
costs and benefits of alternative approaches such as computer
modeling need to be revisited regularly over the next decade. 

9.  We are aware of the limitations of the NASS database, and we
agree with NHTSA that the number of cases that mimic the NCAP
configuration are few.  Indeed, in the 4 years of NASS data we
analyzed for the project, only 46 of over 14,000 cases applied to our
model most closely resembled the NCAP configuration. 

We disagree that the methodology was inappropriate.  The models used
in the analyses were designed not only to simulate the NCAP
conditions, but also to discover the sensitivity of NCAP for
predicting other frontal collisions, and thereby maximize the number
of frontal crash configurations for which the test was applicable and
meaningful.  It seems reasonable to expect that, given enough cases,
NCAP should predict real-world traffic injuries and fatalities in
collisions that essentially duplicate the test conditions; however,
given the small number of actual events that apply to this
configuration, the meaning of any unweighted statistically
significant relationship is questionable. 

10.  There is no contradiction between these statements.  NCAP's
relative rankings of different models may be inaccurate in some
cases.  Nevertheless, it is unlikely that the parallel improvement in
NCAP scores and highway safety statistics over the past 15 years is
totally coincidental. 

11.  We agree with NHTSA's comment about manufacturer participation
in NCAP, and the language has been changed.  With respect to the
quasi-regulatory nature of NCAP, manufacturers repeatedly stated that
they must design automobiles to meet this test as if it were the
standard.  In oral commentary on our draft report, one NHTSA official
pointed out that almost all passenger cars meet compliance standards
in the NCAP test, and that "in effect, it's a de facto standard."

12.  The 220-point reduction in the mean HIC falls short of
statistical significance (p = .171).  This, we believe, is a function
of the low number of cases and high variations in the early years of
compliance testing.  Although the decline is not statistically
significant, we would reiterate that, on average, vehicles tested for
compliance with FMVSS 208 tend to have HIC and chest acceleration
scores that are far below the maximum allowable levels. 

13.  Although "it has not been proven," it seems reasonable to assume
that a 1979 vehicle that was involved in an accident in 1991 no
longer had the same level of structural integrity as it did when it
was new, owing to the rusting of the frame, for example, or the
weakening of welds.  It also seems reasonable to assume that a 1979
model-year vehicle would not perform at its full original safety
potential in a collision that occurred in 1991. 

14.  As our report states, and NHTSA cites, no statistical analysis
can, by itself, establish cause-and-effect linkage, and we do not
demand this result of our analyses. 

15.  We accepted NHTSA's suggestion and used SUDAAN to perform
additional analyses of the NASS data.  The results were inconclusive. 
In one case (two-vehicle collisions), we found a statistically
significant relationship between NCAP scores and serious injury. 
(See table VI.4.) However, this result could easily be spurious since
the application of NASS sampling weights (which vary substantially
and can be quite large) to the small subset of cases that both fit
our criteria and have no missing data can greatly distort the
analysis.  If, as NHTSA suggests, a subset of NASS data is
"insufficient to conduct any type of statistical analyses," applying
sampling weights to a nonrandom selection of variously weighted cases
is potentially misleading. 

16.  We are aware that in some analyses, NHTSA has used the
combination of subject vehicle weight and its ratio to the other
vehicle weight as predictors of injury instead of simply using the
weight of the two vehicles.  We did not feel it necessary to
reanalyze the data using weight and weight ratio since, as NHTSA has
pointed out, they "are mathematically equivalent to the information
provided by the two individual vehicle weights."\2

17.  We used the traditional adjustment, the ratio of fatalities to
the number of registered vehicles, to correct for the variations in
exposure to accident involvement among the NCAP-tested vehicles.  We
agree with NHTSA that other factors, such as driver age and driving
history, are also important predictors of accident involvement and
are not captured by this adjustment.  \3 Our goal here, however, was
to answer the simple question:  Are proportionately more drivers
killed in poor scoring NCAP cars than in better scoring cars?  Our
answer is "yes."

18.  Based on NHTSA's comment, we converted HIC and chest g scores to
the combined injury probability, which forms the basis for NHTSA's
new rating system, and used it as a variable in the analyses
conducted and presented in this report. 

19.  The section is no longer in the report. 


--------------------
\1 See NHTSA, Correlation of NCAP Performance With Fatality Risk in
Actual Head-on Collisions, 1994, p.  xviii. 

\2 "NHTSA, A Collection of Recent Analyses of Vehicle Weight and
Safety," 1991, p.  6. 

\3 See Highway Safety:  Factors Affecting Involvement in Vehicle
Crashes (GAO-PEMD-95-3; Oct.  1994). 


MAJOR CONTRIBUTORS TO THIS REPORT
======================================================== Appendix VIII

PROGRAM EVALUATION AND METHODOLOGY
DIVISION

Robert E.  White, Assistant Director
David G.  Bernet, Project Manager
Edward J.  Logsdon, Project Staff
Martin T.  Gahart, Project Staff
Dale Harrison, Project Staff
Beverly A.  Ross, Project Staff
Venkareddy Chennareddy, Referencer


BIBLIOGRAPHY
=========================================================== Appendix 0

Alem, Nabih M., Guy S.  Nusholtz, John W.  Melvin.  "Head and Neck
Response to Axial Impacts." Proceedings of the Twenty-Eighth Stapp
Car Crash Conference (SAE Paper No.  841667).  Warrendale, Pa.: 
Society of Automotive Engineers, 1984.  Pp.  275-82. 

Association for the Advancement of Automotive Medicine and the
International Research Council on the Biomechanics of Impact.  The
Biomechanics of Impact and Motor Vehicle Crash Performance:  A Global
Concern--Conference Materials.  Des Plaines, Ill.:  AAAM, 1993. 

Backaitis, Stanley H., ed.  Biomechanics of Impact Injury and Injury
Tolerances of the Head-Neck Complex (SAE Book No.  PT-93-43). 
Warrendale, Pa.:  SAE, 1993. 

Blalock, Hubert M., Jr.  Social Statistics.  New York:  McGraw-Hill,
1979. 

Dixon, Camille M.  "Automotive Crashworthiness Rating:  Legislation
and Testing." Proceedings of the Thirtieth Stapp Car Crash Conference
(SAE Paper No.  862046).  Warrendale, Pa.:  SAE, 1986. 

Eiband, A.M.  Human Tolerance to Rapidly Applied Accelerations:  A
Summary of the Literature (NASA Memorandum No.  5-19-59E). 
Washington, D.C.:  National Aeronautics and Space Administration,
1959. 

Eppinger, Rolf H., and Susan C.  Partyka.  "Estimating Fatality
Reductions With Safety Improvements." Proceedings of the Eighth
International Technical Conference on Experimental Safety Vehicles. 
Washington, D.C.:  U.S.  Department of Transportation, National
Highway Traffic Safety Administration, 1980.  Pp.  432-38. 

Foster, J.  King, James O.  Kortge, and Michael J.  Wolanin.  "Hybrid
II:  A Biomechanically-Based Crash Test Dummy." Proceedings of the
Twenty-First Stapp Car Crash Conference (SAE Paper No.  770938). 
Warrendale, Pa.:  SAE, 1977.  Pp.  975-1014. 

Fraser, T.M.  Human Response to Sustained Acceleration"
(NASA-SP-103).  Washington, D.C.:  NASA Scientific and Technical
Information Division, 1966. 

Gadd, C.W.  "Criteria for Injury Potential." Impact Acceleration
Stress Symposium, 27-29 November 1961, Brooks AFB, Texas (National
Research Council Pub.  No.  977).  Washington, D.C.:  National
Academy of Sciences, 1962.  Pp.  141-44. 

Gadd, Charles W.  "Use of a Weighted-Impulse Criterion for Estimating
Injury Hazard." Proceedings of the Tenth Stapp Car Crash Conference
(SAE Paper No.  660793).  Warrendale, Pa.:  SAE, 1966.  Pp.  164-74. 

General Motors.  MVSS:  A Guide To Federal Motor Vehicle Safety
Standards & Regulations.  Detroit:  GM, Environmental Activities,
1989. 

Gennarelli, Thomas A.  "The State of the Art of Head Injury
Biomechanics." Proceedings of the Twenty-Ninth Conference of the
American Association for Automotive Medicine.  Des Plaines, Ill.: 
AAAM, 1985.  Pp.  447-63. 

Gillis, Jack.  The Car Book:  The Definitive Buyer's Guide to Car
Safety, Fuel Economy, Maintenance and More, 1993 ed.  New York: 
Harper Perennial, 1993. 

Gurdjian, E.S., et al.  "Significance of Relative Movements of Scalp,
Skull, and Intracranial Contents During Impact Injury of the Head."
Journal of Neurosurgery, 29:1 (1967), 70-72. 

Gurdjian, E.S., V.L.  Roberts, and L.M.  Thomas.  "Tolerance Curves
of Acceleration and Intracranial Pressure and Protective Index in
Experimental Head Injury." Journal of Trauma, 6:5 (1965), 600-04. 

Gurdjian, E.S., J.E.  Webster, and H.R.  Lissner.  "Observations on
the Mechanism of Brain Concussion, Contusion, and Laceration."
Surgery, Gynecology, and Obstetrics, 101:12 (1965), 680-90. 

Hirsch, Arthur E., and Rolf H.  Eppinger.  "Impairment Scaling From
the Abbreviated Injury Scale." Proceedings of the Twenty-Eighth
Conference of the American Association for Automotive Medicine. 
Morton Grove, Ill.:  AAAM, 1984.  Pp.  209-24. 

Hodgson, V.R., L.M.  Thomas, and P.  Prasad.  "Testing the Validity
and Limitations of the Severity Index." Proceedings of the Fourteenth
Stapp Car Crash Conference (SAE Paper No.  700901).  Warrendale, Pa.: 
SAE, 1970. 

Hodgson, Voigt, and L.M.  Thomas.  "Comparison of Head Acceleration
Injury Indices in Cadaver Skull Fracture." Proceedings of the
Fifteenth Stapp Car Crash Conference (SAE Paper No.  710854). 
Warrendale, Pa.:  SAE, 1971.  Pp.  190-206. 

Holbourn, A.H.S., and M.A.  Edin.  "Mechanics of Head Injuries."
Lancet, 245 (1943), 438-41. 

Joksch, Hans C.  "Velocity Change and Fatality Risk in a Crash--A
Rule of Thumb." Accident Analysis and Prevention, 25:1 (1993),
103-04. 

Jones, Ian S., and R.A.  Whitfield.  "Predicting Injury Risk With New
Car Assessment Program Crashworthiness Ratings." Accident Analysis
and Prevention, 20:6 (1988), 411-19. 

Jones, Ian S., R.A.  Whitfield, and Diane M.  Carroll.  "New Car
Assessment Program Results and the Risk of Injury in Actual
Accidents." Proceedings of the Tenth International Technical
Conference on Experimental Safety Vehicles.  Washington, D.C.:  DOT,
NHTSA, 1985.  Pp.  371-80. 

Koch, M., et al.  "Car Model Safety Rating--Further Development Using
the Paired Comparison Method." Proceedings of the Eighth
International Technical Conference on Experimental Safety Vehicles. 
Washington, D.C.:  DOT, NHTSA, 1980.  Pp.  432-38. 

Kornhauser, Murray.  Structural Effects of Impact.  Baltimore: 
Spartan Books, 1964. 

Langwieder, K.  "Passenger Injuries in Collisions and Their Relation
to General Speed Scale." Proceedings of the Seventeenth Stapp Car
Crash Conference.  New York:  SAE, 1973.  Pp.  1-34. 

Langwieder, Klaus, Maximilian Danner, and Walter Schmelzing. 
"Comparison of Passenger Injuries in Frontal Car Collisions With
Dummy Loadings in Equivalent Simulations." Proceedings of the
Twenty-Third Stapp Car Crash Conference (SAE Paper No.  791009). 
Warrendale, Pa.:  SAE, 1979.  Pp.  201-31. 

Mackey, John M., and Charles L.  Gauthier.  Results, Analysis and
Conclusions of NHTSA's 35 MPH Frontal Crash Test Repeatability
Program (SAE Paper No.  840207).  Warrendale, Pa.:  SAE, 1984. 

McCormick, Earnest James.  Human Factors in Engineering and Design. 
New York:  McGraw-Hill, 1982. 

Marquardt, James F.  "Collision Severity--Measured By V." Proceedings
of the Twenty-First Conference of the American Association for
Automotive Medicine.  Morton Grove, Ill.:  AAAM, 1977.  Pp.379-90. 

Newman, James A.  "Head Injury Criteria in Automotive Crash Testing."
Proceedings of the Twenty-Fourth Stapp Car Crash Conference (SAE
Paper No.  801317).  Warrendale, Pa.:  SAE, 1980.  Pp.  703-47. 

Partyka, Susan C.  "A Comparison of AIS and ISS Predictions of
Fatality on NCSS." Proceedings of the Twenty-Fourth Conference of the
American Association for Automotive Medicine.  Morton, Grove, Ill.: 
AAAM, 1980.  Pp.  156-69. 

Pike, Jeffrey A.  Automotive Safety:  Anatomy, Injury, Testing and
Regulation.  Warrendale, Pa.:  SAE, 1990. 

Prasad, Priya, and Harold J.  Mertz.  The Position of the United
States Delegation to the ISO Working Group 6 on the Use of HIC in the
Automotive Environment (SAE Paper No.  851246).  Warrendale, Pa.: 
SAE, 1985. 

Salvendy, Gavriel.  Handbook of Human Factors.  New York:  John Wiley
& Sons, 1987. 

Society of Automotive Engineers.  Crash Avoidance SP-544: 
International Congress & Exposition, 1983 (SAE Pub.  No.  SP-83-544). 
Warrendale, Pa.:  1983. 

Society of Automotive Engineers.  Human Tolerance to Impact
Conditions as Related to Motor Vehicle Design (SAE Pub.  No. 
SAE-J885-APR80).  Warrendale, Pa.:  1980. 

Swearingen, John J.  Tolerances of the Human Face to Crash Impact
(FAA Pub.  No.  AM-65-20).  Oklahoma City:  Federal Aviation Agency
Office of Aviation Medicine, 1963. 

Thomas, C., et al.  "Crashworthiness Rating System and Accident Data
Convergences and Divergences." Advances in Belt Restraint Systems: 
Design, Performance and Usage (SAE Paper No.  840200).  Warrendale,
Pa.:  SAE, 1984. 

U.S.  Department of Transportation.  Appendix AO (A-Optional):  Part
572E Dummy Performance Calibration Test Procedure.  Washington, D.C.: 
NHTSA, 1993. 

U.S.  DOT.  A Collection of Recent Analyses of Vehicle Weight and
Safety (DOT-HS-807-677).  Washington, D.C.:  NHTSA, 1991. 

U.S.  DOT.  Correlation of NCAP Performance With Fatality Risk in
Actual Head-On Collisions (DOT-HS-808-061).  Washington, D.C.: 
NHTSA, 1994. 

U.S.  DOT.  Federal Motor Vehicle Safety Standards and Regulations
(DOT-HS-805-674).  Washington, D.C.:  NHTSA, 1989. 

U.S.  DOT.  Head and Neck Injury Criteria:  A Consensus Workshop
(DOT-HS-806-434).  Washington, D.C.:  NHTSA, 1983. 

U.S.  DOT.  Laboratory Indicant Test Procedure:  New Car Assessment
Program.  Washington, D.C.:  NHTSA, 1990. 

U.S.  DOT.  Laboratory Test Procedure for:  FMVSS 208, Occupant Crash
Protection; FMVSS 212, Windshield Mounting; FMVSS 219, Windshield
Zone Intrusion; and FMVSS 301, Fuel System Integrity.  Washington,
D.C.:  NHTSA, 1993. 

U.S.  DOT.  New Car Assessment Program:  Plan for Responding to the
Fiscal Year 1992 Congressional Directives.  Washington, D.C.:  NHTSA,
1992. 

U.S.  DOT.  New Car Assessment Program:  Response to the NCAP Fiscal
Year 1992 Congressional Requirements--Report to Congress. 
Washington, D.C.:  NHTSA, 1993. 

U.S.  DOT.  The New Car Assessment Program--Status and Effect. 
Washington, D.C.:  NHTSA, 1982. 

U.S.  DOT.  1991 Traffic Fatalities Preliminary Report.  Washington,
D.C.:  NHTSA, 1992. 

U.S.  DOT, National Transportation Safety Board.  Safety
Effectiveness Evaluation of the National Highway Traffic Safety
Administration's Rule-Making Process.  Vol.  11:  Case History of
Federal Motor Vehicle Safety Standard 208:  Occupant Crash Protection
(NTSB-SEE-79-5).  Washington, D.C.:  U.S.  Government Printing
Office, 1979. 

Unterharnscheidt, F.J.  "Translational Versus Rotational
Acceleration-- Animal Experiments With Measured Input." Proceedings
of the Fifteenth Stapp Car Crash Conference (SAE Paper No.  710880). 
Warrendale, Pa.:  SAE, 1971.  Pp.  767-70. 

Versace, John.  "A Review of the Severity Index." Proceedings of the
Fifteenth Stapp Car Crash Conference (SAE Paper No.  710881). 
Warrendale, Pa.:  SAE, 1971.  Pp.  771-96. 

