oversight

Highway Safety: Fatalities in Light Trucks and Vans

Published by the Government Accountability Office on 1990-11-14.

Below is a raw (and likely hideous) rendition of the original report. (PDF)

          GAO

l~--l-_-“l”l_~l.l-~l-




     Novt~rtlht*r        I.
                        1tN0
                               HIGHWAY SAFETY
                               Fatalities in Light
                               Trucks and Vans



                                                 IllI Ml
                                                     142642
--   --.-
          United States
GAO       General Accounting Office
          Washington, DC. 20548

          Program Evaluation    and
          Methodology  Division

          B-236722

          November 14,1999
          The Honorable Frank R. Lautenberg
          Chairman, Subcommitteeon Transportation
            and Related Agencies
          Committee on Appropriations
          United States Senate
          The Honorable William Lehman
          Chairman, Subcommitteeon Transportation
            and Related Agencies
          Committee on Appropriations
          Houseof Representatives
          In a letter dated November 21,1988, and through our subsequentdis-
          cussionswith your staff, you asked us to analyze the National Highway
          Traffic Safety Administration’s (NHTSA'S) Fatal Accident Reporting
          System (FAR@ to
      l   compare passenger-carfatality rates to those for standard pickup
          trucks, small pickups, standard vans, small vans, and multipurpose
          vehicles (for example, all-terrain and 4-wheel-drive vehicles); and
      l   compare the fatality experience for these vehicle types in two, more
          policy-relevant ways: (1) after statistically controlling (that is,
          adjusting) for non-vehicle-related factors (for example, driver and
          roadway variables), and (2) when only those accidentsinvolving roll-
          overs or side-impact collisions are considered.
          The analysis in this report complementsour earlier assessmentof
          NHTSA’S overall strategy for determining if certain Federal Motor Vehicle
          Safety Standards (FMVSS)  should be extended to pickup trucks, vans, and
          multipurpose vehicles.’ This report also provides data relevant to the
          question of whether the risks associatedwith rollover and side-impact
          accidents warrant proposed regulations to require crush-resistant roofs
          and side-impact protection.




          ‘U.S. General Accounting Office, Motor Vehicle Safety: Passive Restraints Needed To Make Light
          Trucks Safer, GAO/RCED-90-66 (Washington, DC.: November 1989).



          Page 1                                      GAO/PEMDSl-S      FataUties   in Light Trucks and Vans
                                         B-220722




                                         Fatality rates for different vehicle types do differ, as table 1 indicates.
                                         During 1986 and 1986, there were 206 fatalities for every million pas-
                                         sengercars registered.2The fatality rate was lower for small and stan-
                                         dard vans and standard pickups, and higher for small pickups and
                                         multipurpose vehicles.
Table 1: Fatality Rate8 Combining 1985
and 1986 Data                                                                                          Fatality rate per million registered
                                         Vehicle type                                   Fatalities                                 vehicles
                                         Standard pickup                                     6,799                                         200
                                         Small pickup                                        3,928                                         308
                                         Standard van                                        I,51 1                                        140
                                         Small van                                              293                                        151
                                         Multipurpose vehicle                                1,639                                         217
                                         Passenger car                                      47,789                                         206
                                         Total                                             61,959                                          207


                                         However, differences in fatality rates may not be solely attributable to
                                         vehicle type. While vehicle type probably contributes to differences in
                                         fatality rates, so do non-vehicle-related factors, such as whether the
                                         victim was wearing a seat belt or whether the crash occurred in an
                                         urban or rural setting. Unadjusted fatality rates are difficult to interpret
                                         becadsewe cannot tell how much of the difference between two rates is
                                         due to vehicle-type differences and how much is due to non-vehicle-
                                         related factors such as gender of the driver. For example, if men have
                                         higher accident rates than women, and if men are more likely to be
                                         drivers of certain types of vehicles than others, then a difference in
                                         fatality rates may be attributable partly to vehicle type and partly to
                                         the gender of the drivers3More policy-relevant information can be pro-
                                         duced by statistically controlling for such non-vehicle-related factors.
                                         Existing research indicates that a disproportionate share of single-
                                         vehicle-accident fatalities involves occupants of light trucks and multi-
                                         purpose vehicles. For example, the research suggeststhat the rollover
                                         tendenciesof light trucks and multipurpose vehicles may be higher as a

                                         ‘For our analysis of highway fatalities we used 1986-86 information from the Fatal Accident
                                         Reporting System (FARS), the latest FARS information available at the time we began our study.

                                         3Appendix I discusses the relationship between vehicle type and eleven variables representing char-
                                         acteristics of drivers, roadway conditions, and accident circumstances. Highly significant differences
                                         exist among vehicle types in the likelihood of their involvlng a drinking driver, a driver under 26
                                         years old, a male driver, a victim being ejected or wearing a safety belt, an accident occurring on
                                         weekends, involving multiple vehicles, occurring on rural or wet or curved roads, or off the road. (See
                                         appendix I.)



                                         Page 2                                        GAO/PEMDWS        Fatalitiea   in Light Truclrs and Vana
B22iv22




result of inherent vehicle characteristics, such as a higher center of
gravity in relation to vehicle track width. However, becausethe studies
we reviewed have not controlled for the influence of different driver
characteristics or vehicle uses,it has not been possible to conclude that
higher fatality rates are due to the characteristics of the vehicle, inde-
pendent of the foregoing confounding influences. Appendix II summa-
rizes the existing research in this area.

In our analysis, we controlled for, or held constant, two sets of non-
vehicle-related factors: driver/victim characteristics and roadway/acci-
dent characteristics. The driver/victim characteristics included restraint
use by fatality (yes or no), sex of driver, age of driver (younger than 26
or 26 and older), and alcohol use by driver (drinking or not). The
roadway/accident characteristics we controlled for included multiple-
vehicle involvement (whether this factor was present), accident location
(on or off the roadway), setting (urban or rural), roadway curvature
(straight or curved), and pavement condition (wet or dry).
Becauseof the Committee’s interest in rollover and side-impact colli-
sions, we examined the fatalities associatedwith each separately. For
each type of collision, we separately estimated the effect of driver/
victim characteristics and roadway/accident characteristics, and then,
having controlled for these effects, we estimated the likelihood of fatali-
ties occurring in each of the six vehicle types. Our full technical report,
included as appendix I, describesthe statistical analyses we performed
and our more detailed findings.

Unfortunately, the information necessaryto calculate fatality rates,
which are adjusted for each of our control factors, doesnot exist. While
we know the number of registered vehicles within each vehicle type,
and therefore can derive general fatality rates as we did in table 1, we
do not have the necessarylevel of detailed “exposure” data to adjust
these rates for driver or roadway conditions. We do not know, for
example, how many miles small vans are driven by men, or on wet pave-
ment, or by drinking drivers. Without such information, it is impossible
to calculate fatality rates for different types of vehicles adjusted for
non-vehicle-related fators.

For this reason, we have expressedthe results of our analysis not as a
comparison of fatality rates for different vehicle types, but rather as the
relative odds of a fatality occurring in one particular type of vehicle as
opposedtoother. For example, 6,401 passenger-carfatalities in our
sample involved rollovers, and 28,493 did not. The odds, therefore, of a


Page 2                            GAO/PFiMbBM   Fatalities   in Light Trucka and Vaau
                                        0286722




                                        fatality occurring in a passenger-caraccident involving a rollover are
                                        .19 (6,401 divided by 28,493). By contrast, 610 fatalities in multipur-
                                        pose vehicles involved rollovers, while 668 did not. The odds for rol-
                                        lover fatalities in multipurpose vehicles, therefore, are 1.09 (610 divided
                                        by 668). By forming an odds ratio between the results of these two cal-
                                        culations, we can conclude that a fatality in a multipurpose vehicle is
                                        6.74 (1.09 divided by .19) times more likely than a fatality in a pas-
                                        sengercar to involve a rollover.
                                        These calculations, however, do not account for the possible con-
                                        founding effect of other variables. For example, they do not take into
                                        account the possibility that drivers of multipurpose vehicles may be
                                        younger than passengercar drivers, or more liable to have been
                                        drinking, or more likely to be male, or less likely to be wearing a safety
                                        belt. They also do not consider the possibility that fatal accidents
                                        involving multipurpose vehicles may be more likely to take place on dry
                                        pavement, or in a rural area, or off the road, or on a curve, or involve
                                        only one vehicle. Any of these non-vehicle-related factors, or somecom-
                                        bination of them, could account, in whole or in part, for the greater like-
                                        lihood of rollover fatalities in one type of vehicle than in another.

                                        For this reason, we constructed statistical models which allowed for the
                                        possible influence of these factors and recalculated the odds for each
                                        vehicle type after adjusting for the non-vehicle-related factors. For
                                        these calculations, we used passengercars as the criterion (or reference)
                                        group. Table 2 presents the results of these analyses for fatalities
                                        involving rollover accidents- that is, the likelihood, relative to pas-
                                        sengercars, of a fatality occurring in each of five vehicle types. The
                                        data are presented (1) before adjustment for non-vehicle-related factors,
                                        (2) after adjustment for driver/victim characteristics, and (3) after
                                        adjustment for accident/roadway characteristics.
Table 2: Fatallty Likelihood Ratio in
Rollover Accidenta, Non-Passenger-Car                                   Multip;;xww&    Standard        Small         Small   Standard
Vehlcler Versus Parsenger Cars          Variables controlled for                             van         van         pickup     pickup
                                        None                                     5.74          1.89           1.21     2.73        2.47
                                        Driver/victim characteristics            5.83          2.11           1.74     2.58        2.35
                                        Accident/roadway
                                          characteristics                        4.59          1.99           1.88     2.25        1.76


                                        We can conclude that, in all of the special vehicle types we examined,
                                        fatalities are more likely than those occurring in a passengercar to have
                                        involved a rollover. This tendency is most pronounced for multipurpose


                                        Page 4                                 GAO/PEMD-91-S     Fatalities     in Ught Trucks and Vane
                                              E-286122




                                              vehicles. While adjustments for the influence of driver/victim character-
                                              istics or for accident/roadway characteristics affect the magnitude of
                                              our estimates somewhat, the differential effect of vehicle type, and the
                                              lower likelihood of a fatality occurring in a passengercar than in any of
                                              these other vehicle types, persist.
                                              Table 3 presents parallel statistics for fatalities occurring in side-impact
                                              accidents. Fatalities in all the non-passenger-carvehicles in our analysis
                                              are less likely to have involved a side impact than those occurring in
                                              passengercars. This tendency persists even after adjustment for driver/
                                              victim characteristics and for accident/roadway characteristics. Multi-
                                              purpose-vehicle, standard-van, and pickup fatalities are approximately
                                              half as likely, and small van fatalities slightly less than two thirds as
                                              likely, to have involved a side impact.
Table 3: Fatallty Likelihood Ratio in Side-
Impact Accidents, Non-Passenger-Car                                              Muitip;;wx          Standard     Small         Small     Standard
Vehicles Versus Passenger Cars                Variables controlled for                                    van      van         pickup       pickup
                                              None                                            0.39       0.52           0.65       0.46        0.46
                                              Driver/victim characteristics                   0.42       0.53           0.60       0.50        0.50
                                              Accident/roadway
                                                characteristics                               0.49       0.54           0.60       0.53        0.54


                                              The results of our analysis suggestthat the increasedlikelihood of fatal
                                              rollover accidents-including fatalities in all five light truck and van
                                              vehicle types-may be attributable to the vehicles themselves.This
                                              increased likelihood may be due to differences in vehicle configuration
                                              (for example, higher center of gravity), as well as to the absenceof spe-
                                              cific safety standards required for passengercars. Therefore, in the case
                                              of rollovers, our results provide somesupport for the proposed exten-
                                              sion and strengthening of federal standards concerning crush-resistant
                                              roofs for all five non-passenger-carvehicle types considered.4
                                              Our results do not provide similar evidence for the extension of side-
                                              impact standards to those samevehicles. Here our results indicate that
                                              fatalities in non-passenger-carvehicles are less likely to involve side
                                              impacts than are passenger-carfatalities.
                                              Somecautions need to be applied in interpreting the results of our anal-
                                              ysis. First, while we have found that non-passenger-carfatalities are
                                              more likely than passenger-carfatalities to involve rollovers, and less

                                              4NHTSA is now reviewing comments received from the notice of proposed rule-making.



                                              Page 5                                     GAO/PEMD-91-9     Fatalities      in Light Trucks and Vans
likely to involve side impacts, we cannot conclude that these differences
are the result of differing protections afforded occupants in these types
of accidents.(Indeed, our study was not designedto measure such
effects.) Such an interpretation is consistent with our findings. How-
ever, in the absenceof information on nonfatal accidents to parallel our
fatality data, this linkage cannot be established.

Second,as we noted earlier, we do not have “exposure” data at the level
of detail neededto compute adjusted fatality rates. We do not know, for
example, how many miles small vans are driven by men, or on wet pave-
ment, or by drinking drivers. Consequently, we cannot estimate the like-
lihood that a given number of miles traveled in one type of vehicle by a
given driver type will result in a rollover fatality.
Finally, our data do not allow us to estimate the effects of proposed
safety features--so that although safety features such as crush-resis-
tant roofs might reduce rollover injuries, without data on vehicles so
equipped we cannot estimate the effectiveness of such features.
While the foregoing limitations do not allow us to demonstrate conclu-
sively that changesin specifications for certain vehicles would result in
fewer highway fatalities, we believe our analysis offers persuasive evi-
dencethat rollover fatalities are more likely, and side-impact fatalities
less likely, to occur in non-passenger-carvehicles, and that these tenden-
cies are vehicle-specific and cannot be attributed simply to driver,
roadway, or accident characteristics.

We conducted our analysis in Washington, D.C., and Kansas City, Mis-
souri, between August 1988 and July 1990 in accordancewith generally
acceptedgovernment auditing standards. Dr. Probir Roy of the Univer-
sity of Missouri at KansasCity and Dr. Douglas Sloaneof the Catholic
University of America assistedus in the development and application of
our statistical model.
We provided draft copies of this report to officials of NHTSA’S National
Center for Statistics and Analysis and discussedwith them the study
results. We incorporated their suggestionsas appropriate. We are
sending copies to the Secretary of Transportation and other interested
parties and will make copies available to others upon request,




Page 6                           GAO/PEMD-61-6   F&ditkm   in LI&t   Trucka and Vam
B-226722




If you have any questions or would like additional information, please
call me at (202) 276-1864.Major contributors to this report are listed in
appendix III.




Eleanor Chelimsky
Assistant Comptroller General




Page 7                           GAO/PJ!MD91-9   FataUtiea   in Light Tmck~ and Vana
Contents


Letter
Appendix I                                                                                               12
The Analysis of FAIRS   Overview                                                                         12
                        Bivariate Relationship Between Rollovers, Side Impacts,                          12
Data on Rollovers and       and Vehicle Type
Side Impacts:           Bivariate Relationships Between Vehicle Type and                                  14
Methodology and             SelectedDriver, Victim, and Accident Characteristics
                        Hierarchical Models With Simultaneous Controls for                               24
Detailed Findings           Driver/Victim and Roadway/Accident
                            Characteristics
                        Higher Order Interactions                                                        36

Appendix II                                                                                              38
Results of Prior        Relative Risk to Car and Light Truck Occupants,1987
                        Light Truck Safety: A Literature Review and Research
                                                                                                         39
                                                                                                         39
Research                    Outline, 1986
                        A Further Look at Utility Vehicle Rollovers, 1984                                40
                        Estimates of the Potential Benefit of Extending Selected                         41
                            PassengerCar Safety Standards to Light Trucks,
                            Vans, and Multipurpose Vehicles, 1982
                        Side Impacts: An Analysis of Light Trucks, Intrusion, and                        42
                            Injury in FARSand NCSSData, 1982
                        Comparison of Truck and Passenger-CarAccident Rates                              42
                            on Limited-Access Facilities, 1981
                        RecentTrends in Van and Small Truck Safety, 1979                                 42
                        Unwarranted Delays by the Department of                                          43
                            Transportation to Improve Light Truck Safety, 1978

Appendix III                                                                                             44
Major Contributors to
This Report
Tables                  Table 1: Fatality RatesCombining 1985 and 1986 Data                                2
                        Table 2: Fatality Likelihood Ratio in Rollover Accidents,                          4
                            Non-Passenger-CarVehicles Versus PassengerCars
                        Table 3: Fatality Likelihood Ratio in Side-Impact                                  6
                            Accidents, Non-Passenger-CarVehicles Versus
                            PassengerCars



                        Page 8                           GAO/PEMD91-8   Fatalities   in Ught Trucks and Vans
Table I. 1: ObservedFrequenciesof Motor-Vehicle                                   13
    Fatalities Involving Rollovers and Side Impacts, by
    Vehicle Type, and Odds and Odds Ratios Derived
    From Them
Table 1.2:ObservedFrequenciesof Motor Vehicle                                     15
    Fatalities Involving SelectedCharacteristics, by
    Vehicle Type and Odds and Odds Ratios Derived
    From Them
Table 1.3:ObservedFrequenciesin the Cross-                                        17
    Classifications of Rollovers, With Selected
    Characteristics and Odds and Odds Ratios Derived
    From Them
Table 1.4:ObservedFrequenciesin the Cross-                                        18
    Classifications of Side Impacts, With Selected
    Characteristics and Odds and OddsRatios Derived
    From Them
Table 1.6:Likelihood-Ratio Chi-Square(L2) Values                                 20
    Associated With Several .Hierarchical Models Fitted
    to Three-Way Tables in Which Rollovers Are Cross-
    Classified by Vehicle Type and Selected
    Characteristics
Table 1.6:Likelihood-Ratio Chi-Square(L2) Values                                 21
    Associated With Several Hierarchical Models Fitted
    to Three-Way Tables in Which Side Impacts Are
    Cross-Classifiedby Vehicle Type and Selected
    Characteristics
Table 1.7:Odds Ratios Indicating the Association of                              22
    Vehicle Type With Rollovers and Side Impacts,
    Before and After Controlling for Selected
    Characteristics
Table 1.8:Likelihood-Ratio Chi-SquareValues and Other                            24
    Characteristics Associated With Hierarchical Models
    Fitted to the Six-Way Tables Formed by Cross-
    Classifying Rollovers and Side Impacts With Vehicle
    Type and SelectedDriver/Victim Characteristics
Table 1.9:Expected FrequenciesUnder the Main-Effect                              28
    Model for the Six-Way Table in Which Rollovers Are
    Cross-Classifiedby Vehicle Type and Driver/Victim
    Characteristics
Table I. 10: Odds Derived From the Main-Effect Model for                         30
    the Six-Way Table in Which Rollovers Are Cross-
    Classified by Vehicle Type and Driver/Victim
    Characteristics


Page 9                         GAO/PEMD-91-9   FataIitiea   in Ltght Trucks and Vans
contents




Table I.1 1: Odds Ratios Derived From the Main-Effect                             32
    Model for the Six-Way Table in Which Rollovers Are
    Cross-Classifiedby Vehicle Type and Driver/Victim
    Characteristics
Table I. 12: Likelihood-Ratio Chi-SquareValues and Other                          34
    Characteristics Associated With Hierarchical Models
    Fitted to the Six-Way Tables Formed by Cross-
    Classifying Rollovers and Side Impacts With Vehicle
    Type and SelectedAccident/Roadway Characteristics
Table I. 13: Odds Ratios Describing the Associations of                           36
    Vehicle Type With Rollovers and Side Impacts, After
    Controlling for Driver/Victim Characteristics and
    Accident/Roadway Characteristics
Table I. 14: Odds Ratios Describing Interactions of Vehicle                       36
    Type With Other Characteristics on Rollovers
Table I. 16: Odds Ratios Describing Interactions of Vehicle                       37
    Type With Other Characteristics on Side Impacts




Abbreviations

our        Department of Transportation
FARS       Fatal Accident Reporting System
FMVSS      Federal Motor Vehicle Safety Standards
GAO        General Accounting Office
NCSS       National Crash Severity Study
NHTSA      National Highway Traffic Safety Administration
UMTRI      University of Michigan Transportation ResearchInstitute


Page 10                          GAO/PEMD91-9   Fatalities   in Light Trucks and Vans
Page 11
Appendix I

The Analysis of FARSData on RAlovers and
SideImpacts: Methodology and
Detailed Findings
                         Our analysis of these data involved four interrelated steps,
Overview
                         1. We examined the data for bivariate relationships between vehicle
                         type and rollovers, and between vehicle type and side impacts. We
                         found strong relationships in both cases.
                         2. We then looked at a series of two-way tables to explore whether cer-
                         tain other variables that measured characteristics of drivers, fatality
                         victims, accidents, and roadways were simultaneously related to both
                         vehicle type and the outcomesof interest (rollovers and side impacts) in
                         such a fashion that they could account for these bivariate relationships.
                         We discovered strong relationships between several of these measures
                         and both rollover and side-impact accidents.

                         3. We then considereda series of three-way tables that permitted us to
                         control for the relationship between vehicle type and these characteris-
                         tics, and between these characteristics and rollover and side-impact
                         accidents,before reestimating the relationship between vehicle type and
                         rollovers, and between vehicle type and side impacts.
                         4. When, in these three-way analyses,the associationsof vehicle type
                         with both rollovers and side impacts persisted, we attempted finally to
                         control for certain of these interrelated characteristics simultaneously.
                         The persistenceof the initial vehicle type/rollover and vehicle type/
                         side-impact associationseven after the introduction of these mul-
                         tivariate controls convincesus that they are not of a spurious nature, or
                         at least are not readily accounted for by the set of control variables we
                         have considered.
                         The nature and magnitude of these associations,and the techniques we
                         used to test and describe them, are discussedin the following sections.

                         Our analysis began with a consideration of the fatal-accident data in
Bivariate Relationship   table I. 1, where the type of vehicle in which the fatality occurred-a six
Between Rollovers,       category variable contrasting multipurpose vehicles, standard vans,
Side Impacts, and        small vans, small pickups, standard pickups, and passengercars-is
                         cross-classifiedby whether the fatality involved a rollover or a side
Vehicle Type             impact. The numbers given in the first two rows in table I. 1 represent
             *           the number of fatalities, within each vehicle category, that did or did
                         not involve a rollover, or that did or did not involve a side impact. For
                         each of the two accident categoriesshown in the table, a likelihood ratio
                         chi-square statistic (L2) is given. The large value of this test statistic for


                         Page 12                            GAO/PEMD91-9   Fatalitbs   in Light Trucks and Vane
                                           The An&Ma       of FARS Data on Itollovem      and
                                           Side Impacta: Methodology    and
                                           D&ailedFindlngs




                                           both tables indicates that the hypothesis that, among fatalities, rollovers
                                           and side impacts are unrelated to vehicle type can be easily rejected.
                                           Rollovers and side impacts, in other words, are strongly associatedwith
                                           vehicle type.

Table 1.1:Observed Frequencier of Motor-Vehicle Fatalities Involving Rollovers and Side Impacts, by Vehicle Type, and Odds and
Odds Ratios Derived From Them
                                     MultiP;#‘$i                                                         Standard    Passenger
Accldent category, oddr, and ratio                      Standard van      Small van    Small pickup         pickup          car
Rollo”er~                                        610               275           40              934          1,560       5,401
No  rollover
- ----~.--___                                     558                   763               171                 1,806              3,345           28,493
-~-~.-- on rollover
   Odds                                          1.09                  0.36              0.23                  0.52               0.47             0.19
-~ Ratiob                                        5.74                  1.89              1.21                  2.73               2.47
Side impa@                                        180                   202                49                   484                857           10.708
No side impact                                    988                   836               162                 2,256              4,068           23,186
  Odds on side impact                            0.18                  0.24              0.30                  0.21               0.21             0.46
  Ratiob                                         0.39                  0.52              0.65                  0.46               0.46
                                           aL2 = 1,685.41,5 df, P < .OOOl

                                           bAll ratios are expressed relative to passenger cars-that       is, 1.09/0.19 = 5.74,0.36/0.19 = 1.89, and so
                                           on. These odds ratios can be interpreted in a reasonably straightforward fashion: Multipurpose-vehicle
                                           fatalities are 5.74 times as likely to have involved a rollover as passenger-car fatalities, standard-van
                                           fatalities 1.86 times as likely as passenger-car fatalities to have involved a rollover, and so on.

                                           ‘L* - 816.30,5 df, P < .OOOl


                                           The nature of these associationscan be described by calculating odds
                                           and odds ratios from the observed frequencies in the table. The odds of
                                           rollovers (or side impacts) having been involved in these fatalities were
                                           calculated for each type of vehicle. For multipurpose vehicles, for
                                           example, there were 610 fatalities that involved a rollover and 668
                                           fatalities that did not, so the odds on fatalities involving a rollover in
                                           that vehicle type were 610/568 = 1.09. For every multipurpose-vehicle
                                           fatality that did not involve a rollover, in other words, there were 1.09
                                           that did. Thus, for every 100 that did not, there were 109 that did. The
                                           odds on fatalities involving rollovers in other types of vehicles can be
                                           similarly calculated, and the values that result are given in the third
                                           row in each of the two accident categoriescontained in the table. The
                                           odds on fatalities involving a rollover were 0.36 for standard vans, 0.23
                                           for small vans, and so on.
                                           To determine how strongly the odds on rollovers or side impacts are
                                           associatedwith vehicle type, we chosepassengercars as the criterion
                                           vehicle type and calculated the odds ratios, or relative odds on rollovers


                                           Page 13                                          GAO/PEMD-91-9       Fatalities   in Light Trucks   and Vans
                                 APpandirI
                                 TbeAmlyebofFARSlhitaonMlovenand
                                 sldeImjMct8:Metllodologyand
                                 &-Flndinlp




                                 being involved in other-vehicle-type fatalities versus fatalities involving
                                 passengercars. Theseodds ratios are provided in the last row in each of
                                 the two accident categoriesin table I. 1. For example, among rollover
                                 fatalities, the odds ratio between multipurpose vehicles and passenger
                                 cars is 6.74 (1.09/O.19). Similarly, the odds ratios comparing standard
                                 vans, small vans, and standard pickups to passengercars are 1.89 (0.36/
                                 0.19), 1.21 (0.23/0.19), 2.73 (0.62/0.19), and 2.47 (0.47/0.19), respec-
                                 tively. These odds ratios can be interpreted directly to mean that, for
                                 example, fatalities in multipurpose vehicles are 6.74 times as likely to
                                 have involved a rollover as fatalities in passengercars, fatalities in stan-
                                 dard vans 1.89 times as likely to have involved a rollover as passenger-
                                 car fatalities, and so on.
                                 The two full sets of odds ratios provided in table I.1 indicate that all
                                 non-passenger-carfatalities are more likely than passenger-carfatalities
                                 to have involved a rollover (by factors ranging from 1.21 to 5.74), and
                                 all non-passenger-carfatalities are less likely than passenger-carfatali-
                                 ties to have involved a side impact (by factors ranging from 0.66 to
                                 0.39). Differences in the odds on rollovers are most pronounced between
                                 multipurpose vehicles and trucks versus passengercars, as are differ-
                                 encesin the odds on side impacts. The value of the chi-square statistics
                                 associatedwith these differences assuresus that they are due to more
                                 than sampling variability or chance.


                      The fact that these strong associationsof vehicle type with rollovers
Bivariate             and side impacts are not attributable to chanceor random fluctuations
Relationships Between doesnot necessarily imply that they are not spurious, or that they
Vehicle Type and      cannot be accounted for by other variables with which both vehicle type
                      and rollovers or side impacts are jointly associated.It may be, for
SelectedDriver,       example, that the more pronounced tendency for multipurpose-vehicle
Victim, and Accident  fatalities (relative to passenger-carfatalities) to involve rollovers results
Characteristics       from   the fact that multipurpose vehicles are more apt to be driven by
                      males, and males are more likely to be involved in rollovers. Alterna-
                                 tively, drinking drivers may be more likely to be involved in rollovers or
                                 side impacts than nondrinking drivers, and certain vehicle-type fatali-
                                 ties may be more apt to involve drinking drivers.
                                 To gain a preliminary impression of the extent to which certain charac-
                                 teristics may be jointly related to vehicle type, and to rollovers and side
                                 impacts, we examined the simple paired associationsbetween a number
                                 of driver, victim, and accident characteristics and vehicle type, and



                                 Page 14                           GAO/PEMD-91-9   Fatalities   in Light Trucke and Vans
                                                 The Au&d       af FAUS Data on Rolloveru   and
                                                 side Impacts: Methodology   and
                                                 DetauedFIndIngfi




                                                 between these characteristics and rollovers and side impacts. The cross-
                                                 tabulations between these characteristics and vehicle type are presented
                                                 in table 1.2,while the crosstabulations with rollovers and side impacts
                                                 can be found in tables I.3 and 1.4.

Table 1.2:Obrerved Frequencler of Motor Vehicle Fatalltleb Involving Selected Characteristics, by Vehicle Type and Odds and
Odds Ratloa Derived From Them
                                     MultlP~lJ TN$                                                        Standard   Passenger
Characterlrtlc, odds, and ratlo
---                                           R          Standard van     Small van     Small pickup        pickup           car
Male’
-_-                                             984                884          152             2,348         4,478       23,395
Female
-----.____.                                            184                 154                59               392              447          10.499
Oddsonmale
---~-_.---                                            5.35                5.74              2.58              5.99            10.02            2.23
Ratio                                                 2.40                2.57              1.16              2.69             4.50
Under25b
~---___                                                428                 238                26             1,041            1,523          12,363
25andover                                              740                 800               185             1.699            3,402          21,531
Oddsonunder25
~~-.-~                              -_..-_--.-        0.58                0.30              0.14              0.45             0.61            0.57
Ratio
-~-I---                                               1.01                0.53              0.25              0.79             1.07
DrinkingC ..-__-.--_.-.---_                            602                 378                55             1,304            2,464          13,398
No drinkina                                            566                 660               156             1,436            2,461          20.496
Odds     on drinking
-~-.~______-                                          1.06                0.57              0.35              0.91              1.00           0.65
Ratio
I_--        .-..-----_-                               1.63                0.88              0.54              1.40              1.54
No Restraintd
__-II--.__.    - -_.. -_--.-^__II__                  1,026                 935               150             2,533            4,706          28,989
                                                                                                                                                -
Restraint                                              142                 103                61               207              219           4,905
Odds on-.---.--_
-_..          no restraint                            7.23                9.08              2.46             12.24            21.49            5.91
Ratio
---_-_.---._-...                  ..-- .---           1.22                1.54              0.42              2.07             3.64
Ejection0
-----      _______    -- _.__.     ---1                726                 374                62             1,142            1,918        -- 8,066
No   eiection
--..----..-.--.-.-...-.---                             442                 664               149             1,598            3,007          25,828
Odds on.__.
-"~.__-."     ejection
                -.--_         --.--                   1.64                0.56              0.42              0.71             0.64            0.31
Ratio                                                 5.29                1.81              1.35              2.29             2.06       -__I
Multi-vehicles'                                        354                 533               143             1,161            1.883          19.345
Single
I--_--.- vehicle                                       814                 505                68             1,579            3,042          14,549
Odds     on multi-vehicles
-~---_---_                                            0.43                1.06              2.10              0.74             0.62            1.33
Ratio
---.----                                              0.32                0.80              1.58              0.56             0.47
WeekendQ
-__-.I__.-..~                                          707                 529               102             1,478            2,691          17,919
Weekday
---"-...       .----        ...-.-__                   461                 509               109             1,262            2,234          15,975
Oddsonweekend
-----____-_-~.                                        1.53                1.04              0.94              1.17              1.20           1.12
Ratio _-.---
~I-                                                   1.37                0.93              0.84              1.04              1.07
Rural"
_-_ll_._.._.l__-     "-._-.--                          880                 719               135             2,094            3,924          22,274
Other _.----____-.                                     288                319                 76               646            1,001          11,620
Odds     on rural
--_-...----.-                   u                     3.06                2.25              1.78              3.24             3.92            1.92
Ratio                                                 1.59                1.17              0.93              1.69             2.04
                                                                                                                                        (continued)




                                                 Page 15                                      GAO/PEMDSl-8    Fatalities   in Light Trucks and Vana
                                          Appemdix I
                                          The Andyh        of FABS Data on Rollovers   and
                                          Side Impacts lbbthodology     and
                                          Det.alledFindlnga




                                      Multipvul, pc:x                                                                  Standard     Passenger
Characterlstlc, odds, and ratio               5:           Standard van        Small van      Small pickup               pickup            car
Off-road’                                        663                   433               62             1,328              2,668         14,154
--_----
Other                                            505                    605             149             1,412              2,257         19,740
----      ----
Odds on off-road                                1.31                   0.72            0.41              0.94               1.18           0.72
-____-.-.._--..-        _-.___. -
Ratio                                           1.82                   1.oo            0.57              1.31                1.64
Curved roadi                                     411                    262              54               971              1,654          9,878
--
Other                                            757                    776             157             1,769              3,271         23,996
___.__-.-- ---
Odds on curved road                             0.54                   0.34            0.34              0.55               0.51           0.41
-.~              ....--           -
Ratio                                           1.32                   0.83            0.83              1.33                1.23
              _-...__
Wet road”                                        195                    232              40               466                838          7,389
---__
Other                                            973                    806             171             2,274              4,087         26,505
-__“I---~
Odds on wet road                                0.20                   0.29            0.23              0.20               0.21           0.28
~---~.--
Ratin                                            071                   1.04            0.82              0.71               0.75
                                           aL2 - 1,660.49,5 df, P < .OOOl
                                           bL2 - 203.46,5df,P     c .OOOl

                                           cL2 - 321.95,5 df,P < .OOOl

                                           dL2 = 604.24,5df,P     < .OOOl

                                           eL2 - 1,437.06,5 df, P < .OOOl
                                           'L* = 1,046.03,5 df,P < .OOOl
                                           gL2 = 35.67,5df,     P < .OOOl

                                           hL2 - 540.53,5df,P     < .OOOi
                                           'L* * 393.65,5 df,P < .OOOl

                                           jL2 - 104.96,5df,    P < .OOOl

                                           kL2 = 104.93,5df,    P < .OOOi




                                           Page 16                                       GAO/PEMD-918     Fatalities    in Light Trucks and Vans
                                           The Analysis of FARS Data on Rollovera    and
                                           Side Impacts: Methodology and
                                           DetailedFlndlngE




Table 1.3:Observed Frequencies In the
Cross-Classltlcatlons of Rollovers, With                                                                            Odds on
Selected Characteristics and Odds and      Characteristic                     Rollover         No rollover           rollover    Odds ratio
Odds Ratlos Derived From Them              Malea                                    6,936           25,305              0.27            1.42
                                           Female                                   1,904            9.831              0.19
                                           Under 25b                                4,020           11,599              0.35            1.75
                                           25 and over                              4,820           23,537              0.20
                                           DrinkinaC                                5.183           13.018              0.40            2.35
                                           No drinking                              3,657           22,118              0.17
                                           No restraintd                            8,373           29,966              0.28            3.11
                                           Restraint                                  467            5,170              0.09
                                           EiectioP                                 6,293            5.995              1.05           11.67
                                           No ejection                              2,547           29,141              0.09
                                           Multi-vehicles’                            725           22,694              0.03            0.05
                                           Sinale vehicle                           8,115           12.442              0.65
                                           Weekends                                 5,221           18.205              0.29            1.38
                                           Weekday                                  3,619           16,931              0.21
                                           Ruralh                                   7,100           22,926              0.31            2.21
                                           Other                                    1.740           12.210              0.14
                                           Off -road’                               7,207           12,101              0.60            8.57
                                           Other                                    1,633           23,035              0.07
                                           Curved roadj                             4,021             9229              0.44            2.32
                                           Other                                    4,819           25,907              0.19
                                           Wet roadk                                1,074            8,086              0.13            0.45
                                           Other                                    7.766           27.050              0.29
                                           aL2 - 15503,l     df, P < .OOOl
                                           bL2 = 467.43,l    df,P < .OOOl

                                           cL2 = lJ37.67,    1 df,P < OOOl

                                           dL2 =I 666.95, 1 df, P < .OOOl

                                           "L2 = 9,382.12,1 df, P < .OOOl

                                           'L2 = 10,066.64,1 df,P < .OOOl

                                           QL2 = 149.91,1 df,P < .OOOl

                                           hL2 = 791.57,l    df,P < .OOOl
                                           'L* = 6,599.96, 1 df, P < 0001

                                           IL* = 1,176.16,1 df, P < .OOOl

                                           kL2 = 557.04, 1 df,P < .OOOl




                                           Page 17                                     GAO/PEMD-91-8   Fatalities   in Light Trucks and Vans
                                                                                                                                 L




                                         Appendix I
                                         The Analyda of FAR8 Data on Rollovers   and
                                         Side Impacts: Methodology and
                                         DetailedFindlnga




Table 1.4:Observed Frequencies in the
Cross-Classifications of Side Impacts,                                                        No side               Odds on
Wlth Selected Characteristic8 and Odd8   Characteristic                     Side impact        impact           side impact    Odds ratio
and Odds Ratio8 Derived From Them        Male8                                    8,595         23,646                 0.36           0.73
                                         Female                                   3,885          7,850                 0.49
                                         Under 25b                                4.338         11.281                 0.38           0.95
                                         25 and over                              8,142         20,215                 0.40
                                         DrinkingC                                4,283         13,918                 0.31           0.66
                                         No drinking                              8,197         17,578                 0.47
                                         No restraintd                           10,392         27,947                 0.37           0.63
                                         Restraint                                2;088          3:549                 0.59
                                         EjectiorP                                2,740          9,548                 0.29           0.26
                                         No eiection                              9,740         21.948                 0.44
                                         Multi-vehicles’                          8,542         14,967                 0.57           2.33
                                         Single vehicle                           4,028         16,529                 0.24
                                         Weekend’J                                6,267         17,159                 0.37           0.86
                                         Weekdav                                  6,213         14,337                 0.43
                                         Ruralh *                                 7,801         22,225                 0.35           0.70
                                         Other                                    4,679          9,271                 0.50
                                         Off -road’                               3.982         15,426                 0.25           0.46
                                         Other                                    8,599         161070                 0.54
                                         Curved roadi                             3,066         10,190                 0.30           0.68
                                         Other                                    9,420         21,306                 0.44
                                         Wet roadk                                3,290          5,870                 0.56           1.56
                                         Other                                    9,190         25,626                 0.36
                                         .L2 = 172.67,1 df, P < .OOOl
                                         %* - 4.37, 1 df, P * ,037
                                         cL2 - 363.75,l     df. P < .OOOl
                                         dL2 - 228.29,l     df, P < .OOOl

                                         ‘L2 - 319.99,l    df. P < DO01
                                         ‘L* - 1,493.97,1 df, P < .ooOl
                                         gL2 - 6620,l      df, P < 0001

                                         “L* - 263.39,l     df, P < DO01
                                         ‘L* = 1,185.38, 1 df, P < .CUHN

                                         IL* - 267.09, 1 df, P C .OOOl

                                         kL2 - 312.75, 1 df, P < .OOOl


                                         Nearly all these crosstabulations reveal highly significant relationships,
                                         as indicated by the chi-square statistics associatedwith them. These
                                         tables also present the magnitude of this relationship as odds and odds


                                         Page 19                                   GAO/PJCMD-91-9   Fatalitlw      in Light Tmcka and Vana
The Anai,ysta of FARS Data on Rollovera   and
Side Impacta Methodology   and
Dm4mdFYndlngE




ratios. These ratios can be interpreted as in table 1.1.For example, the
driver of a multipurpose vehicle involved in a fatality was 2.39 (6.35/
2.23) times more likely to have been male than the driver of a passenger
car involved in a fatality.1 A rollover fatality was 1.42 (0.27/O.19) times
more likely to have involved a male driver than a female driver.2 A
detailed interpretation of each of these relationships appears unneces-
sary, but in general these crosstabulations indicate the following:

1. Among vehicle fatalities, each of the following variables bears a sig-
nificant relationship to vehicle type, to rollovers, and to side impacts:
sex of driver, age of driver, whether the driver was drinking, whether
the victim was using restraints, whether an ejection occurred, whether
multiple vehicles were involved, whether the fatalities occurred on
weekends,or on rural roads, or off-road, or on curved or wet roads.
2. Fatalities involving multipurpose vehicles, and both types of pickups,
were more likely than passenger-carfatalities to have involved a male
driver, a drinking driver, no restraint use, and an ejection, and more
likely to have occurred off-road or on rural roads or curved stretches of
roads. They were less likely, at the sametime, to have involved multiple
vehicles and wet roads and, in the caseof pickups at least, a driver
under the age of 25. Fatalities occurring in both types of van were also
more likely than passenger-carfatalities to have involved a male driver
and an ejection, but, unlike multipurpose vehicle and truck fatalities,
they were less likely than passenger-carfatalities to have involved a
drinking driver, or to have occurred off-road, or on a curved road. (See
table 1.2.)
3. Fatalities involving rollovers were more likely to have involved a
male driver, a driver under age 25, a drinking driver, no restraint use,
and an ejection, and they were also more likely to have occurred on
weekends,on rural roads, on curved roads, or off road. Fatalities
involving rollovers were, at the sametime, less likely to have involved
multiple vehicles or wet roads. Fatalities involving side impacts were,
conversely, more likely to have involved multiple vehicles or wet roads,
but less likely to have involved male drivers, younger drivers, drinking
drivers, no restraint use, and ejection, or an accident that occurred off-
road, on the weekend, or on a curved or wet road. (Seetables I.3 and
1.4.)

ISee table 1.2.
%ee table 1.3.



Page 19                                    GAO/PEMD-9143   Fatalities   in Light Trucka and Vans
                                          Appemdix I
                                          The Andy&~ of FARS Data on Rolloven,           and
                                          Side Impacts! Methodology nnd
                                          DetalledFlndlnga




                                          Given the significance and magnitude of the associationsof these char-
                                          acteristics, both with vehicle type and with the outcomesof interest
                                          (rollovers and side impacts), we concluded that it was necessaryto con-
                                          trol for them in analyzing the association of vehicle type with rollovers
                                          and side impacts. In other words, we attempted first to identify the por-
                                          tion of the bivariate relationship between vehicle type and rollover or
                                          side impact that could be accounted for by the relationship between the
                                          personal and accident-related characteristics we considered,and then
                                          we attempted to determine if vehicle type significantly added to our
                                          understanding of the likelihood of rollovers or side impacts.

                                          We did this by constructing and analyzing a series of three-way tables in
                                          which vehicle type was cross-classifiedby rollover and one control vari-
                                          able at a time, and a similar set of tables for side impact and one control
                                          variable at a time. While we do not provide all of these three-way tables
                                          in this report, table I.6 presents the results of fitting selectedhierar-
                                          chical models to the rollover tables, and table I.6 contains the results of
                                          fitting similar models to the side-impact tables.
Table 1.6~Likelihood-Ratio Chi-Square
(L2) Valuer Associated With Several                                                                        Chi-square value8
Hkrarchlcal Models Fitted to Three-Way                                                         [VC] [R] 11 [VC] [CR] 10 [VC] [CR] [VR] 5
Tables In Which Rollovers Are Crorr-                                                            de m;d?mm de rees of          de reesot
Classified by Vehicle Type and Selected   Control variable                                        f            f!eeedom         #eedom
Characteristic8                           Sex of driver                                           1,748.23           1,593.20                31.11 (.98)
                                          Age of driver                                           2,231.33           1,763.90                12.71 (.99)
                                          Drinking                                                2,876.09           1J38.41                 26.72 (.99)
                                          Restraint use                                           2,226.74           1,557.79                 6.05 (99)
                                          Ejection                                               10,097.19             715.07                47.68 (.99)
                                          Multi-vehicles                                         11,066.65             978.02                62.63 (.99)
                                          Weekend                                                 1,819.83           1,669.92                 2.30 l.99)
                                          Rural road                                              2,306.24           1,514.67                11.16j.99)
                                          Off -road                                               8.177.50           1.57754                190.21 t-98)
                                          Curved road                                             2,802.58           1,626.42                11.75 (.99)
                                          Wet road                                                2,211.57           1,654.54                33.33 (.99)
                                          Models are denoted, following convention, by the underlying marginals of the three-way tables they fit:
                                          V = vehicle type; R - rollover; C - the third (control) variable in each table (for example, sex, age, and
                                          so on). All models are described in detail in the text.

                                          Numbers in parentheses next to the chi-square values for the third model fitted to each table indicate
                                          the proportion of the variation in the odds on the fatality involving a rollover that is accounted for by that
                                          model.




                                          Page 20                                          0A0/PEMD-9143 Fatal&h            in Light ‘hocka and Vans
                                          The AaJyrir,       of FARS Data   on R4lovere and
                                          Side Impacta; Methodology         and
                                          DetdledFindlnge




Table 1.6: Likelihood-Ratio Chi-Square
(L2) Values Associated With Several                                                                          Chi-squarwaiues
Hierarchical Models Fitted to Three-Way                                                       PC1IS111 PC1PSI 1:a Ml [dcBsH~~]o:
Tables in Which Side Impacts Are Cross-                                                       de rees oi       de rees of
Classified by Vehicle Type and Selected   Control vafiable                                      Breedom          Breedom                    Breedom
Characteristics                           Sex of driver                                            888.21             715.56                 1.90 (.99)
                                          Age of driver                                            830.53             826.15                 5.13 (.99)
                                          Drinking                                               1,117.37             753.62                 3.54 (.99)
                                          Restraint    use                                         991.21             762.92                14.77 i.98)
                                          Eiection                                               1.018.95             698.96               22.86 1.981
                                          Multi-vehicles                                         2IO67.83             573.86                11.54 i.99;
                                          Weekend                                                  .880.29            815.09                 5.09 i.99;
                                          Rural road                                             1 JlO7.87            744.48                 8.20 (99)
                                          Off-road                                               1 s895.79            710.41               23.01 1.99)
                                          Curved road                                            1D63.36              796.27                 9.37 (99)
                                          Wet road                                               1.098.39             785.64                 7.33 I.991

                                          Models are denoted, following convention, by the underlying marginals of the three-way tables they fit:
                                          V - vehide type: S = side impact; C = the third (control) variable in each table (for example, sex, age,
                                          and so on). All models are described in detail in the text.

                                          Numbers in parentheses next to the chi-square values for the third model fitted to each table indicate
                                          the proportion of the variation in the odds on the fatality involving a side impact that is accounted for by
                                          that model.


                                          Three models were fitted to all tables. The first was the logit-specified
                                          model of independence,which assertsthat in each table rollovers or side
                                          impacts are unrelated to either vehicle type or the control variable pres-
                                          ent. This model can be readily rejected in every case,as could have been
                                          anticipated from our two-way results. The independencemodel, more-
                                          over, is substantially and significantly improved upon by the second
                                          model we fit to the data, which assertsthat the control variables (but
                                          not vehicle type) are associatedwith rollovers and side impacts. (Note
                                          the significant reduction in chi-square values from the first to the
                                          secondcolumns of numbers in tables I.6 and 1.6,which correspond to
                                          these two models.) Additionally, the third model fitted to each of the
                                          tables-which allows vehicle type to be related to rollovers and side
                                          impacts after controlling for the association of each control variable
                                          with both vehicle type and these outcomes-significantly improves
                                          upon the secondmodel. This implies that the associationsof vehicle type
                                          with rollovers and side impacts persist after individual controls are
                                          introduced.
                                          While this third model does not, in every case,provide a reasonablefit
                                          to the data (indicating the presenceof significant three-way interac-
                                          tions), this is not surprising given the large sample being used


                                          Page 21                                          GAO/PEMD-9143      Fatalities   in   L&M Trucks and Vans
                                         Appendix I
                                         The Analysis of FARS Data on Rollover   and
                                         Side Impacts: Methodology and
                                         DetailedFindings




Table 1.7:Odds Ratios Indicating the
Association of Vehicle Type With                                                                                Rollover
Rollovers and Side Impacts, Before and                                                 Multiwg;;
After Controlling for Selected           -Control variable                                                      Standard van         Small van
Characteristics (One at a Time)           None                                                     5.74                    1.89            1.21
                                          briver’s sex                                             5.64                    1.85            1.23
                                          Driver’s age                                             5.93                    2.08            1.44
                                         &inking                                                   5.55                    1.99            1.40
                                          Restraint use                                            5.84                    1.85            1.40
                                         Ejection                                                  3.21                    1.53            1.08
                                         Multi-vehicle                                             4.64                    1.92            1.88
                                         Weekend                                                   5.68                    1.92            1.25
                                         Rural road                                                5.59                    1.88            1.25
                                         Off-road
                                         -~                                                        6.08                    2.14            1.74
                                         Curved road                                               5.81                    2.01            1.28
                                         Wet road                                                  5.71                    1.92            1.21


                                         (approximately 44,000). A better indicator of whether this third
                                         model provides an adequate account of the associations present in
                                         the table can be obtained by determining how much of the variation
                                         in the odds on rollovers or side impacts it accounts for. This can be
                                         computed by dividing the difference between the chi-square value
                                         for the baseline model and the chi-square value for the third (main
                                         effect) model by the baseline model chi-square. For the three-way
                                         table involving rollovers, vehicle type, and sex of driver, for
                                         example, this calculation yields 0.98 [(1748.23 - 31.11)/1748.23].
                                         Ninety-eight percent of the variation in rollovers, across the joint
                                         categories of vehicle type and driver sex, is accounted for by this
                                         model, which posits independent main effects of vehicle type on rol-
                                         lovers. In other words, the model stipulates an effect of vehicle type
                                         that is the same for both male and female drivers and an effect of
                                         sex that is the same for all vehicle types. Therefore, there is no com-
                                         pelling reason to take account of the significant three-way interac-
                                         tion that is present in this table. The same is true for the other three-
                                         way tables as well, inasmuch as in every case our main effect model
                                         accounts for better than 98 percent of the variation in each.




                                         Page 22                                   GAO/PEMD-918           Fatalities   in Light Trucks and Vans
                                                      m &slyele        of FABB Data on Rollovers     and
                                                      Side Impacts Methodology      and
                                                      DetailedFindings




                   Rollover                                                                        Side impact
                                        Standard    MultiP;erc$;             Standard                                                            Standard
   Small pickup                           pickup                                  van                Small van         Small pickup                pickup
 ____.
    .__.__”    2.73
         .-....__-.        --.-              2.47             0.39                  0.52                   0.65                    0.46                0.46
                     2.66                    2.41             0.41                  0.54                   0.66                    0.48                0.48
                    2.75                     2.61             0.39                  0.51                   0.64                    0.46                0.45
                    2.63
      ..~..~-__-..“._--.-.-.-.------         2.34             0.41                  0.52                   0.62                    0.48                0.47
..-- ..__....- -. .--2.61
                      ._^_
                         --._-.__-..-___     2.32             0.40                  0.53                   0.62                    0.48                0.47
                 2.09
.- ..-.-...-....-.---..-    ._..--.--        2.01             0.45                  0.54                   0.67                    0.49                0.48
                 2.39                        1.89             0.48                  0.54                   0.60                    0.51                0.52
                 2.73                        2.49             0.40                  0.52                   0.65                    0.46                0.46
                 2.58                        2.31             0.41                  0.53                   0.65                    0.48                0.48
                 2.85
 .._-___...I. ____..__
                     --.- .._-..             2.23             0.43                  0.52                   0.60                    0.48                0.49
                 2.66 .--...---..---
--_. “..-._-__ll_-..---                      2.46             0.40                  0.51                   0.65                    0.47                0.46
                 2.68                        2.44             0.40                  0.52                   0.66                    0.47                0.46


                                                      We can use the expected frequencies under this model for each table and
                                                      derive from them, as before, the odds on rollovers and side impacts, and
                                                      the ratios of these odds (again with passengercars as the criterion
                                                      vehicle type) acrossvehicle type. The results of these calculations are
                                                      provided in table 1.7,which also provides the initial odds ratios (that is,
                                                      those calculated in table I. 1, prior to controls).

                                                      As table I.7 indicates, the odds ratios after adjusting for the effect of the
                                                      individual control variables are not substantially different from the
                                                      ratios derived without controls. Only the control for whether the victim
                                                      was ejected alters appreciably our estimate of the relationship between
                                                      vehicle type and rollovers, and even there sizable differences among
                                                      vehicle types remain. For side impacts, no control variable, taken by
                                                      itself, does much to alter our conclusion about the sizable differences
                                                      between vehicle types.




                                                      Page 23                                         GAO/PEMD-91-8   Fatalities   in Light Trucks and Vans
                                         Appendix I
                                         The An&da        of FAR8 Data on Rollovera and
                                         Side Impacts: Methodology     and
                                         DetailedFindings




Table 1.8: Likelihood-Ratio Chi-Square
Values and Other Characteristics
Asrociated With Hierarchical Models
Fitted to the SIX-Way Table8 Formed by
Crorr-Classifying Rollover8 and Side     Model                      Marginal8 fitted
Impacts Wlth Vehicle Type and Selected   1                          [SADRV]    [Z]
Drlver/Vlctim Characteridlcs             2                          [SADRV]    [SADRZ]
                                         3                          [SADRV]    [SADRZ]    [VZ]
                                         4                          [SADRV]    [SADRZ]    [SVZ]
                                         5                          [SADRV]    [SADRZ]    [AVZ]
                                         6                          [SADRV]    [SADRZ]    [DVZ]
                                         7                          [SADRV]    [SADRZ]    [RVZ]
                                         8                          [SADRV]    [SADRZ]    [SVZ] [DVZ]




                                         It may be that while no control variable greatly attenuates the initial
Hierarchical Models                      relationships we found when we consider them serially, they do so when
With Simultaneous                        we consider them simultaneously. Unfortunately, the contingency table
Controls for Driver/                     approach demandedby the categorical nature of independent variables,
                                         combined with the small number of fatalities for certain vehicle types
Victim and Roadway/                      (especially small vans), doesnot permit us to build and analyze tables in
Accident                                 which all control variables are considered at once. We were able, how-
Characteristics                          ever, to control for certain of these variables in blocks, by exploring two
                                         pairs of six-way tables in which rollovers, and then side impacts, were
                                         cross-classifiedby vehicle type and certain characteristics of drivers
                                         and victims, and then by vehicle type and certain characteristics of acci-
                                         dents and roadways.

                                         Table I.8 provides information about various hierarchical models fitted
                                         to the two six-way tables in which rollovers and side impacts are cross-
                                         classified by vehicle type and by the following characteristics of drivers
                                         and victims: sex of driver, age of driver, whether the driver had been
                                         drinking, and whether the victim was using restraints.3




                                         3After analysis of the interrelationships of the control variables and consultation with NHTSA
                                         researchers, we decided to omit ejection from our control variables for these models. Restraint use is
                                         highly correlated with e&tion. The use of both control variables simultaneously would result in
                                         numerous empty or sparse cells. More substantively, ejection can itself be considered a function of
                                         vehicle type and therefore could introduce a spurious control into our analysis of the effect of vehicle
                                         type.



                                         Page 24                                        GAO/PEMDM-l3       FataWes    in Light TN&E    and Vana
                                            The Analyds of FARS Data on RoUovem            and
                                            Side Impscts: Methodology and
                                            DetalledFindlnga




                      Rollover                                                                           Side Impact
                                                  Proportlon oi                                                                         Proportion of
                                                      varlatlon            De &eeyoi                                                        variation
De?rE22              L*                P             explained               r                           L*                  P             explained
          93   3,766s            c.0001                         .oo                   93          1,371.23           < .ooo1                         .oo
          78   1,605.31          c.OOO1                         57                    78            737.02           < .oooi                         A6
          73     128.72           <.OOl                         .97                   73             99.75               .021                        .93
          66     103.09              .004                       .97                   66             98.32               a09                         .93
          66     120.39           <.OOl                         .97                   66             91.34               ,031                        .93
          68      99.90              a07                        .97                   68             97.23               .012                        .93
          66     126.68           <.OOl                         .97                   66             84.41               .086                        94
          63      80.24              .070                       .98

                                            S - driver’s sex; A - driver’s age; D = drinking driver; R = restraint use; V - vehicle type; Z - rollover
                                            or side impact


                                            The first model fitted to both tables was again the lo&it-specified model
                                            of independence,which allows vehicle type to be related to each of the
                                            control variables (that is, driver/victim characteristics) in the table but
                                            assertsthat rollovers and side impacts are independent of both vehicle
                                            type and all of these controls. The large values of chi-square associated
                                            with this model suggestthat it doesnot fit the data acceptably, and
                                            becauseit posits that the odds on rollovers and side impacts are the
                                            same acrossall of the joint categoriesof the factors in the table, it does
                                            not account for any of the variation in those odds. Model 2, which
                                            allows all factors except for vehicle type to be related in an uncon-
                                            strained or interactive fashion with rollovers and side impacts,
                                            improves significantly upon this first model and accounts for 67 and 46
                                            percent, respectively, of the variation in the odds on rollovers and side
                                            impacts.
                                            More importantly, model 3 improves significantly upon model 2. After
                                            controlling for the associationsof the driver/victim characteristics with
                                            vehicle type, and the associationsof driver/victim characteristics with
                                            rollovers and side impacts, model 3 allows an association of vehicle type
                                            with rollovers and side impacts. The significant improvement of model 3
                                            on model 2 implies that the vehicle type/rollover and vehicle type/side-
                                            impact associationspersist even after controlling for these characteris-
                                            tics simultaneously.




                                            Page 26                                          GAO/FEMD-@U3       FataUtien   in Light Trucka and Vanm
The Analyols    of FARS Data on Rollover   and
Side Impacta    Methodology  snd
De!taIledFIn~




Model 3 doesnot, strictly speaking, fit the data acceptably in either of
these two tables (P < .06)-that is, a statistically significant amount of
variation due to interactions among the control variables, vehicle type,
and the outcome variable remains to be explained. Nevertheless,it does
account for the large bulk of the variation in the odds on rollovers (97
percent) and side impacts (93 percent).
Further analysis indicates that securing a model for rollovers that fits
the data acceptably requires the inclusion of interactions between sex of
driver, vehicle type, and rollovers, and between drinking driver, vehicle
type, and rol1overs.4For side impacts, an acceptablefit of model to data
is achieved by allowing an interaction between restraint use, vehicle
type, and side impact.6We will discussthe nature of these interactions
below, after discussingthe implications of model 3. However, it should
be noted here that these interactions do not account for much of the
variation in rollovers or side impacts, nor even much of that variation
which is directly attributable to vehicle type.
To reestimate the association of vehicle type with rollovers and side
impacts after these simultaneous controls, we can calculate the expected
frequencies under the third model fitted to the data in each of the four
tables considered and derive from them the odds and odds ratios as
before.

Tables I.9 through I.1 1 contain an example of this procedure for a model
of the effect of vehicle type on rollovers, after controlling for the effect
of our driver/victim characteristics. In multivariate tables of this
nature, we can calculate odds within categoriesof vehicle type and
within categoriesin the joint distribution of the four other variables
being controlled for. For multipurpose-vehicle fatalities, for example,
among accidents involving male drivers under 25 who were drinking
and not using restraints, the odds on rollovers having been involved
were 2.047 (135.70/66.30). For passenger-carfatalities, the odds on roll-
overs having occurred for that samegroup defined by that specific com-
bination of control categorieswas 0.351(1206.71/3438.29).




4Note that models 4 and 6, which include these interactions one at a time, improve significantly on
model 3, and that model 8, which includes both interactions, improves significantly on both models 4
and 6.

6Model 7 fits the data acceptably (P < .OS)and significantly improves on model 3.



Page 26                                      GAO/PEMDz)l-8     Fatalities   in Light Trucks and Vans
AppcmUx I
The Anelysb of FARS Data on Rdlovem          and
Side Impacts; Methodology and
DetailedFLn~




As in our previous example, the odds ratio indicating the greater likeli-
hood of rollovers for multipurpose-vehicle fatalities relative to pas-
senger-carfatalities is obtained by dividing the former odds by the
latter. For this case,the odds ratio of multipurpose vehicles to passenger
cars is 5.83 (2.047/0.361). Odds on rollovers can similarly be obtained
for other vehicle types, and ratios contrasting those odds can be
obtained by using the passenger-carodds as the criterion.6




‘The odds ratios relative to passenger cars calculated from this model remain the same for each
vehicle type across each category of the control variables, since this constraint is specified in the
model. These odds ratios would vary somewhat across categories in models that allow for vehick-
type/control-variable interactions.



Page 21                                        GAO/P-91-9         Fatalities   in I&ht   Trucks and Vans
                                                                                                                              ,

                                         Appendix I
                                         The Amlysin of FAJB Data on Rollovers   end
                                         Side Impacts; Methodology and
                                         DetailedFlndings




Table 1.9: Expected Frequenclea Under                                                                                  .a
the Main-Effect Model for the Six-Way
Table in Which Rollovers Are Cross-
Clabsitied by Vehicle Type and Driver/   Sex                    Age                    Drinking                  Restraint used
Vlctlm Charscterlstlcs                   Male                   Under 25               Yes                       No

                                                                                                                 Yes

                                                                                       No                        No

                                                                                                                 Yes

                                                                25 and over            Yes                       No

                                                                                                                 Yes

                                                                                       No                        No

                                                                                                                 Yes

                                         Female                 Under 25               Yes                       No

                                                                                                                 Yes

                                                                                       No                        No

                                                                                                                 Yes

                                                                25 and over            Yes                       No

                                                                                                                 Yes

                                                                                       No                        No

                                                                                                                 Yes




                                         Page 29                                   GAO/PEMD9143   Fatalities   in Light Trucks and Vans
                                              Appendix I
                                              The Analyeie   of FM18 Data on Rollovers   and
                                              Side Impacte   Methodology  and
                                              DetsiledFin~




                                                                       Odds on rollover
                         Mult’p;;$yt           Standard                Small            Small              Standard                  Passenger
Rollover   .~                                       van                 van           pickup                 pickup                         car
Ye9                              ’ %t              E:E                   1.89             E:E                  345.92                   1,206.71
No
---             -__--.                                                   3.11                                  420.08                   33438.29
Yes                                11.20            0.48                 0.00                4.47                3.64                      41.21
No
1_~                                12.80             1.52                0.00              11.53                10.36                     274.79
Yes                                                35.13                                  136.36               191.40                     655.46
No
-_--.--    -                       !E              73.87                 %                234.64               362.60                   2,913.54
Yes                                 7.78            1.10                 0.13                4.43                4.10                      36.46
No                                 15.22            5.90                 0.87              19.57                19.90                     415.54
Yes                              it%               93.26                11.57             Z:                   592.67                   1,177.16
No                                                161.75                24.43                                  926.33                   4,316.84
Yes                                                 3.08                                    9.48                10.79                      62.08
No
--~                               : ::i;            8.92                i::;               22.52                28.21                     380.92
Yes                               98.13            63.78                8.56              144.58               284.69                     634.25
No
.--~                             161.87           290.22               47.44              538.42             1,167.31                   6,101.75
Yes                                 8.51            5.11                3.08               10.87                11.44                      81.99
No                  -             29.49            48.89               35.92               85.13                98.56                   1,658.Ol
Yes                                    9.85         2.31                0.83               %:                   26.88                     285.93
No                                 4.15             2.69                1.17                                    28.12                     702.07
Yes                                0.53             i%                  i::i:               1.34                 0.31                      13.56
No                                 0.47                                                     2.66                 0.69                      69.44
Yes                               :%                 6.96                1.45              34.33                36.91                     367.94
No                                                  14.04                3.55              56.67                67.09                   1,569.06
Yes                                2.03              E                   E                  1.44                 0.60                      20.51
No                                 5.97                                                     9.56                 4.40                     352.49
~-
Yes                               18.84              %                   1.58              28.09                26.12                     275.06
No                                12.16                                  3.42              40.91                41.88                   1,034.94
ies                                                  0.21                                   0.49                 0.46                      13.36
No                                 2;                0.79                51:::              1.51                 1.54                     105.64
Yes                               25.70             17.97                5.22              32.94                41.67                     466.50
No                                36.30             70.03               24.78             105.06               146.33                   3,843.50
---
Yes                                4.14              2.20                1.00               2.42                 2.42                      62.83
No                                14.86             21.80               12.00              19.58                21.58                   1,316.17




                                              Page 29                                      GAO/PEMD-91-S   Fatal&lea    in Light Trucks and Vans
                                                                                                                                I




                                          The Anal@     of FABlp Data on Rollovers   and
                                          Side Impacts! Methodology   and
                                          Detalled~




Table l.l& Odds Derived From.the Main-
Effect Model for the Six-Way Table In
Which Rollovers Are Cross-Clnsslfied by
Vehicle Type and Driver/Vlctlm            Sex                      Age                     Drinking                  Restraint used
Characteristics                           Male                     Under 25                Yes                       No

                                                                                                                     Yes

                                                                                           No                        No

                                                                                                                     Yes

                                                                   25 and over             Yes                       No

                                                                                                                     Yes

                                                                                           No                        No

                                                                                                                     Yes

                                          Female                   Under 25                Yes                       No

                                                                                                                     Yes

                                                                                           No                        No

                                                                                                                     Yes

                                                                   25 and over             Yes                       No

                                                                                                                     Yes

                                                                                           No                        No

                                                                                                                     Yes




                                          Page 30                                     GAO/PEMD-91-9   Fatalitier   in Light Trucke and Vans
                                               The Analyeio of FARS Data on Ibllovers   and
                                               Side Impacts Methodology  and
                                               DetailedFinding




                                                                      Odds on rollover
           Multipurpose                         Standard              Small            Small             Standard                    Passenger
Rollover     _.. -vehicle
                   . _.--                            van               van           pickup                pickup                           car
Yes                      2.047                      0.742             0.608              0.907                 0.823                        0.351
No                     .-. -----
Yes                      0.875                      0.316               WA               0.388                 0.351                        0.150
No
Yes                     1.312                       0.476             0.391              0.581                 0.528                        0.225
No
Yes                     0.511                       0.186             0.149              0.226                 0.206                        0.088
No
Yes                     1.590                       0.577             0.474              0.704                 0.640                        0.273
No           _     .      .-- ._.-
Yes                     0.951                       0.345             0.282              0.421                 0.382                        0.163
No
Yf?S                    0.606                       0.220             0.180              0.269                 0.244                        0.104
No
Yes                     0.289                       0.105             0.086              0.128                 0.116                        0.049
No
Yes                     2.373                       0.859             0.709              1.052                 0.956                        0.407
No
Yes                     1.128                         WA              0.333              0.504                 0.449                        0.195
No                                 --..-_I__
Yes                     1.368                       0.496             0.408              0.606                 0.550                        0.234
No           .~         -. ..--.
YC?S                    0.340                       0.124             0.099              0.151                 0.136                        0.058
No                ~~.._ .._-.-- .-_.
Yes                  1.549                          0.563             0.462              0.687                 0.624                        0.266
NO
YCS                     0.736                       0.266             0.220              0.325                 0.299                        0.126
No
Yes                    0.708                        0.257             0.211              0.314                 0.285                        0.121
No                 . .~ .---_-~-
Yes                    0.279                        0.101             0.083              0.124                 0.112                        0.048
No




                                               Page 31                                   GAO/PEMD-91-3   Fatalities    in Light Trucks   and Vans
                                          The Analyela of FAR8 Data on Rollovera   and
                                          Side hpach       Methodology and
                                          DetalledFindlnge




Table 1.11: Odda Ratlor Dwlved From the
Maln-Effect Model for the SIX-Way Table
In Which Rollovers Are Cross-Classified
by Vehicle Type and Drlver/Vlctim         sex                     Age                     Drinking                  Restrain‘t used
Characterlstlcs                           Male                    Under 25                Yes                       No

                                                                                                                    Yes

                                                                                          No                        No

                                                                                                                    Yes

                                                                  25 and over             Yes                       No



                                                                                          No                        No

                                                                                                                    Yes

                                          Female                  Under 25                Yes                       No

                                                                                                                    Yes

                                                                                          No                        No

                                                                                                                    Yes

                                                                  25 and over             Yes                       No

                                                                                                                    Yes

                                                                                          No                        No




                                          Page 32                                    GAO/PEMD-91-3   Fatalities   in Light Truclce and Vane
                                                   The AnalysLe of FAN3 Data on Rollovem     and
                                                   slde&mnm~~odology         and




                                                                  Odds ratios relative to passenger cars
                             Multipur 008                    Standard                 Small                    Small                      Standard
Rollover
_--.__-___- ...-_ -I .--____        ve Rlcle                      van                   van                   pickup                        pickup
Yes
No                                      5.83                      2.11                1.74                       2.58                          2.35

No
-..___-._l
        . _._-...----..--                   5.83                  2.11                N/A                        2.58                          2.35
Yes
No
--I ._-._.- _-._..-.-_-._.-.                5.83                  2.11                1.74                       2.58                          2.35
Yes
No                                          5.83                  2.1 1               1.74                       2.58                          2.35
Yes
No
.-._-- ...-._.--_-_.-    _..-
                           - . ..--         5.83                  2.11                1.74                       2.58                          2.35
Yes
No
_..-__--.-_.-.-----            .__.__ ~     5.83                  2.11                1.74                       2.58                          2.35
Yes
No                                          5.83                  2.11                1.74                       2.58                          2.35
Yes
No .-..- ~.-_-
.._l___"~                                   5.83                  2.11                1.74                       2.58                          2.35

Zs                                          5.83                  2.11                1.74                       2.58                          2.35
Yes
No                                          5.83                  N/A                 1.74                       2.58                          2.35

Es
------                                      5.83                  2.11                1.74                       2.58                          2.35
Yes
No                                          5.83                  2.11                1.74                       2.58                          2.35
Yes
No
.--_.-. ._.-..."-l-_..-..- . . --.-   __-   5.83                  2.11                1.74                       2.58                          2.35
Yes
No                                          5.83    -             2.11                1.74                       2.58                          2.35
Yes
No
__.
  ^__
    - ..-_ ._ .-- -.        ____.           5.83        --        2.11                1.74                       2.58                          2.35
Yes
No                                          5.83                  2.11                1.74                      2.58                           2.35




                                                   Page 33                                    GAO/PEMD-91-3   Fatalities   in Light Trucks and Vans
                                         Appendix I
                                         The helyedo of FARS Data on Rvllovem   and
                                         Side Impncta: Methodology and
                                         Dc!tauedFindlngo




hbk 1.12: Llkolihood-Ratio Chl-Square
Values and Other Characterlrtlcs
Aoaociated Wlth Hlerarchlcal Models
Pltted to the Six-Way Tables Formed by
Cross-Clasrlfying Rollovers and Sldo     Model                  Marginale fitted
Impacts With Vehlclo Type and Selected   1                      [RMCWV]    [Z]
Accident/Roadway Characterlrtlcs         2                      [RMCWV]    [RMCWZ]
                                         3                      [RMCWV]    [RMCWZ]     [VZ]
                                         4                      [RMCWV]    [RMCWZ]     [RVZ]
                                         5                      [RMCWV]    [RMCWZ]     [MVZ]
                                         6                      [RMCWV]    [RMCWZ]     [CVZ]
                                         7                      [RMCWV]    [RMCWZ]     [WVZ]
                                         8                      [RMCWV]    [RMCWZ]     [MVZ] [CVZ]


                                         The results of such calculations can be summarized by the odds ratios
                                         provided in table I. 11. After controls were introduced for these driver/
                                         victim characteristics, multipurpose-vehicle fatalities remained more
                                         than five times as likely as passenger-carfatalities to have involved a
                                         rollover. Both types of pickups, and standard vans as well, were more
                                         than twice as likely as passengercars to have involved a rollover, and
                                         small vans were almost twice as likely.

                                         When we fitted a series of models involving roadway and accident char-
                                         acteristics to our rollover and side-impact data, we reached conclusions
                                         about the preferred models similar to those derived from our driver/
                                         victim models. Table I. 12 presents a summary of these hierarchical
                                         models.

                                         As Table I. 12 shows, for both the rollover and side-impact data, model 2
                                         improves significantly on model 1, and model 3 improves significantly
                                         on model 2. Moreover, model 3 accountsfor the great bulk of the varia-
                                         tion in rollovers (99 percent) and side impacts (97 percent) and, in the
                                         caseof side impacts, it fits the data acceptably and is not improved on
                                         significantly by models 4 through 7, which include interaction terms.
                                         For the rollover table, none of the models (including three-way




                                         Page 34                                   GAO/PEMD-91-8     Fatalities   In Light Truck6 and Vana
                                                            Appendix I
                                                            The Analyeb of FARS Data on Rollovers        and
                                                            Side Impacta: Methodology and
                                                            DetaIledFYndlng6




                                       Rollovsr                                                                          Side impact
                                                                      Proportion of                                                                   Proportion of
   De ftft$llol                                                           variation         De rees of                                                    variation
-.-----f        .-.--       -          L*               P                explained            ?reedom                    L*                  P           explained
- ---~~-.-_ 94                  12,148.64         <.OOOl                         .oo                   94         2,716.21           < .OOOl                       .oo

i.--_ _-.----- 79                 986.70          <.OOOl                         .92                   79           597.01           -=c.OOOl                      .78
---------      74                 177.19          -c .OOOl                       .99                   74            93.27                .05                      .97
---..-.---     69                 167.83          < .Oool                        .99                   69            85.29                .05                      .97
_--._-         69                 117.37           < ,001                        .99                   69            82.59                  .lO                    .97
--             69       -         165.11          < .OOOl                        .99                   69            85.87                  .05                    .97
               69
_..._._..._- A-..-                170.17          -K .OOOl                       .99                   69            87.10                  .05                    .97
               64                 106.40           < .OOl                        .99
                                                            Note: R - rural road; M = multi-vehicle accident; C = curved road; W = wet road; V = vehicle type;
                                                            2 = rollover or side impact

                                                            interactions) fit the data, which implies the existence of significant
                                                            higher order-that is, four-way or five-way-interactions.      Given the
                                                            large proportion of variation explained by the models we have con-
                                                            sidered, however, it seems reasonable to assume that such interac-
                                                            tions are substantively trivial in spite of their statistical
                                                            significance.’

                                                            A summary of the odds ratios obtained from each of our four preferred
                                                            models is presented in table I. 13. We have already presented a detailed
                                                            interpretation of the odds ratios for rollover models considering driver/
                                                            victim characteristics. We found similar results when we controlled for
                                                            accident/roadway characteristics, although controlling for this set of
                                                            characteristics doesdiminish our estimate of the differences between all
                                                            type of vehicles (with the exception of small vans) and passengercars.
                                                            These associationspersist, however, and remain quite sizable even after
                                                            introducing these controls.




                                                            ‘Among the models provided in table I. 12 for the rollover data, it would appear that the only three-
                                                            way interaction of any importance is the one involving multi-vehicles, vehicle type, and rollovers,
                                                            inasmuch as mode16, which includes that interaction, improves significantly on mode13 and is not
                                                            itself improved on by mode18, which includes another interaction that had appeared significant in
                                                            the absence of this one. This interaction will be discussed but caution must be applied in interpreting
                                                            it. In spite of its statistical significance--which is achieved rather easily in working with samples of
                                                            this size-it accounts for very little of the variation in the odds on rollovers.



                                                            Page 36                                         GAO/PEMtbBl-8      Fatalities   in Light Truclm and Vana
                                             Appendix I
                                             The Analyeie of FARS Data on Rollovers   and
                                             Side Impacte: Methodology and
                                             DetailedFindjnga




Table 1.13: Odd8 Ratios Dercrlblng the
A88ociatlonr of Vehicle Type Wlth                                                               Ratios of odds on roiioverb, relative
Roiioverr and Side impactr, After                                                               Muitip;;rc+t   Standard
Controiilng for Driver/Victim                Variables controlled for                                                van      Small van
Characteristic8 and Accident/Roadway         Driver/victim characteristics                                 5.83          2.11            1.74
Characteristics (Derived From Main-Effects   Accident/roadway characteristics                              4.59          1.99            1.88
Models)

                                             The association of vehicle type with side impacts also persists after we
                                             control for either the driver/victim or accident/roadway characteristics,
                                             Multipurpose-vehicle, standard-van, and pickup fatalities are only
                                             roughly half as likely as passenger-carfatalities to have involved a side
                                             impact, and small-van fatalities are slightly less than two thirds as
                                             likely.

                                             The odds ratios estimating these associationsof interest that are
Higher Order                                 presented in table I. 13 were derived from models that constrain those
Interactions                                 associationsto be equally large acrossall categoriesof the control vari-
                                             ables employed. As we noted, however, there is someevidenceof certain
                                             interactions present. The nature of those interactions is demonstrated in
                                             tables I. 14 and I. 16, where we have reestimated odds ratios using inter-
                                             action models.

Table 1.14: Odds Ratloe Describing interaction8 of Vehicle Type With Other Characteristics on Rollovers
                                                                Ratios of odds on rollovers, relative to passenger cars
                                                      MuiWy~~~                                                                     Standard
Categories of interacting variables                                    Standard van       Small van      Small pickup                pickup
Male drivers, drinking                                                                1.50          1.35                 2.27            2.31
Male drivers, not drinking                                        Z:E                 2.43          2.47                 2.60            2.22

                                                                  5.27
                                                                  7.72                2.38
                                                                                      3.87          0.79
                                                                                                    1.45                 2;              %I
-Female
   -~    drivers,
        ..--        drinking
                    not drinking
             ___---..--~.--__
Sir@+vehicle
Multivehicle accident
              accident                                            9.74
                                                                  3.87                4.13
                                                                                      1.68          2.28
                                                                                                    1.82                 3.14
                                                                                                                         2.14            ::Bs:




                                             Page 36                                    GAO/PEMD-91-S   Fatalities   in Light Trucks and Vane
                                                Appendix I
                                                The Analysla  of FARS Data on Rollovers and
                                                Side Impact% Methodology   and
                                                Detailed Findln@




    to passenger cars                                     Ratio of odds on side impacts, relative to passenger cars
     Small       Standard            Mu”‘iyy~                                                                  Small                                Standard
   pickup _. __-_. --_--_--
                     pickup                                         Standard van           Small van          pickup                                  pickup
      _.2.58
          .._      -I       2.35
                        _._--__
                             ~~-__              0.42                             0.53                   0.60                       0.50                  0.50
       2.25                 1.76                0.49                             0.54                   0.60                       0.53                  0.54


                                                Wenoted, for example, that it appeared that the vehicle type/rollover
                                                association interacted with sex of driver, whether the driver was
                                        *       drinking, and whether multiple vehicles were involved. This can be seen
                                                in the odds ratios presented in tables 1.14,and 1.15,as can the interac-
                                                tion of vehicle type with rollovers and side impacts. For all types of non-
                                                passengercars relative to passengercars, differences in the odds on
                                                fatalities involving rollovers appear to be more pronounced when
                                                drivers were female, when drivers were not drinking, and when multiple
                                                vehicles were involved. Also, non-passengercar/passengercar differ-
                                                encesin the odds on side impacts were more pronounced when
                                                restraints were used than when they were not.

Table 1.1s: Odd8 Ratios Describing interactlone of Vehicle Type With Other Characteristics on Side impacts
                                                              Ratio of odds on side impacts, relative to passenger cars
                                                      MWygg                                                             Standard
Categories of Interacting variables                                    Standard van       Small van     Small pickup      pickup
No restraint use                                                       0.43                 0.59               0.56                       0.50           0.51
Restraint use                                                          0.37                 0.22               0.68                       0.47           0.39


                                                As discussedpreviously, however, the improvement in our under-
                                                standing of rollovers or side impacts afforded by these models, while
                                                statistically significant, is slight. In only one case-that of female
                                                drivers of small vans who had been drinking-is the odds ratio to pas- ,.
                                                sengercars reversed. This anomaly should be considered a statistical
                                                artifact of the small number of small vans, and particularly of this sub-
                                                category of small-van fatalities, in our sample.s




                                                *Only ten of the nearly 44,000 cases in our sample fall into this subcategory.



                                                Page 37                                       GAO/PEMD-91-9           Fatalities   in Light Truck   and Vans
                                                                                                    I


Appendix   II

&sub            of Prior Research


                        Researchersappear to agreethat a disproportionate share of the single-
                        car-accident fatalities occurring on our nation’s streets and highways
                        involve occupantsof small trucks and multipurpose vehicles. While
                        study results suggestthat the rollover tendenciesof small trucks and
                        multipurpose vehicles may reflect vehicle characteristics-such as a
                        high center of gravity in relation to vehicle track width-every study
                        cautions that various characteristics of drivers and vehicle use may
                        affect the results. Becausethese factors have not been consideredin
                        previous research,researchershave not been able to conclude that
                        higher fatality rates are due to inherentcharacteristics of these
                        vehicles.
                        To determine what research has been done on the subject of light-truck,
                        van, and multipurpose-vehicle safety, we searchedthe National
                        Highway Traffic Safety Administration (NHTSA), Department of Trans-
                        portation (nor), University of Michigan Transportation ResearchInsti-
                        tute (UMTRI), and GAO libraries for research on highway safety, with
                        specific emphasis on studies dealing with fatalities. These libraries col-
                        lectively contain most of the research in the highway safety area. We
                        identified over one hundred research citations pertaining to small truck
                        and van safety, motor vehicle safety standards, and the extension of
                        those standards to small trucks and vans.

                        We reviewed all identified researchto determine its relevance to our
                        analysis-specifically, to identify those studies that compared fatality
                        experience by vehicle type and/or by vehicle type and type of impact
                        (side impact or rollover). We found eight studies comparing the fatality
                        experience of small trucks, vans, and multipurpose vehicles to that of
                        passengercars, with emphasis on type of impact.
                        Although most of the studies baaedtheir results on the type of vehicle
                        involved in the accident (adjusted to reflect vehicle exposure), none of
                        the studies accounted for the amount and type of use each vehicle type
                        received-that is, annual number of miles driven in total or for specific
                        purposes.Of the eight studies, four used registration data for the expo-
                        sure measurement;one used vehicle production data combined with
                        scrap-rate information as a proxy for estimating the number of vehicles
                        in use; one limited itself to toll roads for which accurate exposure rates,
                        using miles traveled, could be obtained; and two used no measure of
                        exposure, basing their results on investigations of samplesof vehicles
                        involved in accidents.




                        Page 98                           GAO/PEMD-918   Fatalities   in Light TIUC~S ad   Van13
                        Appendix II
                        liewlta of Prior Besearch




                    l light truck occupant-injury rates in multi-vehicle crasheswere generally
                      lower than passengercar occupant-injury rates in multi-vehicle crashes;
                    . light trucks have higher rollover and occupant-ejectionrates than do
                      passengercars, which creates a greater potential for injuries to their
                      occupants as compared to occupants of passengercars; and lastly,
                    . light trucks as striking vehicles were found to have a greater tendency
                      to injure occupants of the struck vehicle-that is, they appear to be
                      more “aggressive” than passengercars.

                        This study was a follow-up to a 1981 report by Reinfurt, et al., that
A Further Look at       analyzed the relative involvement in rollover crashesof utility vehicles
Utility Vehicle         (also referred to asjeeps), pickup trucks, and passengercars, using
Rollovers, 1984         crash data from North Carolina (1973-78), Maryland (1974-78), and the
                        Fatal Accident Reporting System (1978-79). The highlight of the results
                        of the earlier study was that smaller vehicles generally had higher rates
                        of rollover involvement than larger vehicles3
                        This follow-up study examined more recent crash data for North Caro-
                        lina (1979-82) that included several additional utility-vehicle models for
                        which data were previously inadequate or nonexistent.4Also, this report
                        estimated vehicle-specific mileage exposure from newly available data.
                        As was found in the earlier study, rollovers occurred approximately ten
                        times as often in single-vehicle crashesas in multi-vehicle crashes.
                        Among the vehicle groups, utility vehicles had, by a considerable
                        margin, the highest involvement rate in single-vehicle rollover crashes;
                        pickups and cars were equally invqlved in single-vehicle rollover
                        crashes,at a rate of involvement that was considerably lower than that
                        for utility vehicles. In addition, the study found that utility vehicles had
                        serious or fatal driver injury rates that were approximately three times
                        higher than the rates for pickups or passengercars.
                        The authors stressedthat, as indicated by earlier literature, track width
                        and center of gravity of a vehicle are very important factors with
                        respect to rollovers.6Utility vehicles have a higher center of gravity and

                        3D. W. Reinfurt, et al., A Comparison of the Crash Experience of Utility Vehicles, pickup Trucks, and
                        Passen er Cars (Chapel Hill, NC.: The University of North Carolina Highway Safety Research Center
                        an t e as mgton, DC., Insurance Institute for Highway Safety, 1981).
                        -Y3Ti%T
                        4D. W. Reinfurt, et. al., A Further Look at Utility Vehicle Rollovers (Chapel Hill, NC.: The University
                        of North Carolina Highway Safety Research Center, 19i34,1986).

                        ‘See J. W. Garrett, “A Study of Rollover in Rural United States Automobile Accidents,” Society of
                        Automotive Engineers, Paper No. 680772,1969.



                        Page 40                                        GAO/PEMD-91-9      FataMes    in Light Trucks and Vane
                                                                                                                 .

                         Appendix II
                         Re4YuIta of Prior Besearch




                         In this report, side impacts of light trucks were studied using 1979 FARS
Side Impacts: An         data and data from NCSS.’ Vehicle types included within the light-truck
Analysis of Light        category were pickups, small vans, and large station wagons.
Trucks, Intrusion, and   Study results indicated that 12.6 percent of the fatalities in light trucks
Injury in FARS and       resulted from side impacts. Within the light-truck category, pickups
NCSSData, 1982           accountedfor nearly 86 percent of the light truck side-impact fatalities,
                         and small vans accounted for only 13.5 percent.
                         The study further disclosedthat there is a high correlation between
                         intrusion into the passengercompartment and serious injury in side-
                         impacted vehicles. However, the authors stressedthat it is not clear
                         whether the correlation is due to the intrusion by itself or to the greater
                         impact severity associatedwith the intrusion.

                         In this study, vehicles were classified into three broad categories-pas-
Comparison of Truck      sengercars, light trucks, and heavy trucks.a Vehicles were classified into
and Passenger-Car        three broad categories-passenger cars, light trucks, and heavy trucks.
Accident Rates on        The study conducted a nationwide survey of 1976 through 1978 acci-
                         dent rates for 34 limited-accessfacilities. These included 21 toll express-
Limited-Access           ways and turnpikes and 13 bridges and tunnels for which accurate
Facilities, 1981         exposure rates using vehicle miles traveled could be obtained. The
                         results showed that the fatal-accident rate for light trucks on express-
                         ways was significantly greater than that for passengercars. On the
                         average,light trucks were involved in 2.36 times more fatal accidents
                         than were passengercars for the samedistance traveled.

                         This study examined whether small truck and van safety had been com-
Recent Trends in Van     promised due to the exclusion of these vehicle types from certain Fed-
and Small Truck          eral Motor Vehicle Safety Standards (FMVSS).~ For its exposure
Safety, 1979             measurement,the study used vehicle production data combined with
                         scrap-rate information to estimate the number of vehicles in use. Using
                         data from the 1977 FARS data base,the study found that pickup trucks


                         7R. E. Scott, Side Impacts An Analysis of Light Trucks, Intrusion, and Injury in FARS and NCSSData
                         (Ann Arbor:The University of Michigan Transportation Research Institute, 1982).
                         8W.E. Myers, “Comparison of Truck and Passenger-Car Accident Rates on Limited-Access Facilities,”
                         Transport&Ion Research Record, 808 (1981), pp. 48-66.

                         eJ. 0%~ and R. Kaplan, “Recent Trends ln Van and Small Truck Safety,” Society of Automotive
                         meers   Technical Paper, 1979.



                         Page 42                                     GAO/PEMD-91-8     Fatalities   in Light Trucks and Vans
Appendix III

Major Contributors to This Report


                        Michael J. Wargo, Director
Program Evalluation     Richard 2‘. Barnes, Assistant Director
and Methodology         Roy R. Jones,Project Manager
Division, Washington,   ~~e$!~~!t~~@?~~~~or
                                  ,
DC.

Kansas City Regio       DeniseM. Wempe,Deputy Project Manager
Office                  Donald L. Ficklin, Computer Analyst
                        Thomas M. Cook, Project Staff
                        Kimberli S. Hagberg, Project Staff




(878264)                Page 44                          GAO/PEML%Bl-9   Fatalities   in Light Trucka and Vans