MULTILEVEL STUDY OF ACCIDENT CAUSATION AND...MULTILEVEL STUDY OF ACCIDENT ...

190 Pages·2005·9.03 MB·English

MULTILEVEL STUDY OF ACCIDENT CAUSATION AND AVOIEANCE
Phase-IA Planning Report
Lyle D. Filkins
Robert E. Scott
James 0 'Day
Michael Kubacki
The University of Michigan
HIGHWAY SAFETY RESEARCH INSTIl'UTE
September 1978
Prepared for the Department of Transportation, National
Highway Traffic Safety Ad~inistration under Contract
No. DOT-HS-7-01669. The opinions, findings, and conclusions
expressed in this publication are those of the authors and
not necessarily thbse of the National Highway Traffic Safety
Administration.
Twhnical Report Documntoti'~P age
1. @.port No. 2. lbomamt Acc*ssia He. 3. Ruipimt's Cocolog No, 1
UM-HSRI-78-38
4. Title d Subtitlo 5. R-rc Doce
MULTILEVEL STUDY OF ACCIDENT CAUSATION September, 1978
6. Puforrinq Ot~irwionCo d.
ALUD AVOIDANCE: Phase IA Planning Report
'8. Pniardng Orporirotion R-t No.
*Vwa'~YDl.e Filkins, Robert E. Scott,
7. UM-HSRI-78-38
James 0' Dav. Michael Kubacki
9. Pwl.cning Orgmixdon Na~aod A ddress 10. Wad Unrt No. (TRAIS)
Highway Safety Research Institute
The University of Michigan 11. Contract a, Gmnt NO.
, DOT-HS- 7-(31669
Ann Arbor, Michigan 48109
'3. f YP. of @.port and Perled Corqrd
12. Spanswine Agmcy Nlw and Addtea*
Interim Report
U.S. Department of Transportation
National Highway Traffic Safety Administration
I
1'6. Abshoct
This report presents results of Phase IA--the planning phase
preparatory to undertaking field work--of the t'Multilevel Study of
Accident Causation and Avoidance."
The case study approach, focusing on the causes of individual
accidents, has been the basis of most previous accident-causation
research. Epidemiology provides an alternative in the statistical
in£e rence approach, a probabilistic model.
We recommend the statistical inference approach. It is able to
deal with complex, interactive phenomena by isolating the influence of
particular factors. Also, the method is compatible with the NASS.
The approach may be formulated i n terms of multidimensional
contingency tables, illustrated in the report by a 2x2 table. There is
a growing body of theory concerning such tables and computer programs
are available to assist in the analysis of data.
Projects were designed for three research topics using the
in
statistical inference approach. These are accident experience stolen
tire
vehicles, field dependence, and characteristics (vehicle handling).
Sampling techniques from epidemiology are described.
L
17. K- Wwds 18. Disir~butionJ t.tm.nt
Accident-causation research
Case study approach
Unlimited
Statistical inference apprcach
L 1
19. karrity Clusif. (04 this -) 1D. kwrb CIasmiL (04 this p w ) 21- ;Yo. of Pqes 22 Pncm
Unclassified Unclassified
The followinq members of the HSRI research staff have
participated in this study. The authors gratefully
acknwledge their contributions to the study and to this
report.
Kenneth L. Campbell, Ph.D.
Charles P. Cornpton
Jairus D, Flora Jr., Ph.D,
Richard J. Kaplan, Ph.D.
Paul L. Olson, Ph.D.
Yichael Sivak, Ph.D.
Arthur C. Wolfe, Ph.D.
In addition, a panel of experienced scientists not
affiliated with HSRI reviewed earlier drafts and made many
valuable suggestions for improvement, A l l suggestions were
evaluated carefully and incorporated as judged appropriate.
No endorsement of the final report was sought from the panel
as a whole or from its individual members, and none is
implied by our appreciation for and recognition of their
participation.
Michael R. Appleby
Automobile Club of Southern California
William L. Carlson, Ph.D.
St. Olaf College
Gerald J. Driessen, Ph.D.
Hans C. Joksch, Ph.D.
The Center for the Environment & Man, Inc.
Leonard B. West, Jr.
The University of Oklahoma
TABLE OF CONTENTS
. . . . . . . . . . . . . . . . . . . . . . . . . . .
LIST OF TABLES vi i
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
SUMMARY
. . . . . . . . . . . . . . . . . . . . . . . . . . .
1 INTRODUCTION
.
2 DEFINITIONS OF ACCID.EN.T .CA.U.SAT.IO.N .AN.D .AP.PR.OA.CH.ES. T.O .AC.CI.DE.NT.- . .
CAUSATION RESEARCH . . . . . . . . . . . . . . . . . .
2.1 The Case-Study Approach . . . . . . . . . . . . . .
2.1.1 Northwestern Univers.ity. . . . . . . . . . . . . . .
2.1.2 Calspan Corporation . . . . . .
2.1.3 Institute for Research in Public Safety
2.2 Definition of Accide.n.t C.a.us.ati.on.: .Th.e .S.ta.tis.ti.c.al. . . . .
Inference Approach
. . .
3 (X-UUWCTERISTICS OF THE CASE-STUDY AND STA.TIS.TI.CA.L .AP.PR.OA.CH.ES. . .
3.1 Features of the Case-Study A.p.pro.ac.h . . . . . . . . . . . .
3.1.1 The Case-Study Model . . . . . . . . . . . . . .
3.1.2 The Study of .T.ra.ns.ie.nts. . . . . . . . . . . . . . .
3.1.3 Limitations . . . . . .
3.2 Features of the Statistical Infe.re.nc.e .A.pp.roa.ch. . . . . . .
3.2.1 Non-Accident Information . . . . . . . . . . . . .
3.2.2 Accident-Causation F.ac.to.rs . . . . . . . . . . . . .
3.2.3 Stability of Factors . . . . . . . . . . . .
3.2.4 Subsetting Compl.ex .Ph.en.om.en.a . . . . . . . . . . . .
3.2.5 Field Personnel
. . . . . . . . . . . . .
4 THE STATISTICAL APPROACH IN GREATER DEPTH . . . .
4.1 Accident Causation and a Mult.ip.le . R.e.gr.ess.io.n .M.od.el . . . .
4.2 Contingency Table Formulation . . . . . . . . . . . . . .
4.2.1 Cross-Sect ional Samp.lin.g . . . . . . . . . . . . . .
4.2.2 Prospective Studies . . . . . . . . . . . . . . .
4.2.3 Ret respective Studies . . . . . . . . . . . . . .
4.2.4 The Cross-Product Ra.tio. . . . . . . . . . . . . . .
4.3 Sample Size Conside.ra.tio.n.s . . . . . . . . . . . . . . . .
4.4 PotentialProblem . . . . . . . . . . . . . . .
4.4.1 Control-Group Problems . . . . . . . . . . .
4.4.2 Interaction between. V.a.ria.b.les. . . . . . . . . . . .
4.4.3 Collapsing Tables
. . . . . . . . . . . . . .
5 CHOICE OF APPROACH AND RES.EA.RC.H .TO.P.ICS. . . . . . . . . . . . . .
5.1 Choice of Approach . . . . . . . . . . . . . . . . .
5.2 Choice of Research Topics . . . . . . . . . . . . . . . .
5.2.1 Survey of Us.er .N.eed.s . . . . . . . . . . . . . . . .
5.2.2 Other Inpu.ts . . . . . . . . . . . . . . . . . . . .
5.2.3 Assessment
. . . . . . . . . . . . .
6 PRELIMINARY DESIGNS FOR PHASE-I1 PRQIECTS . . . . . . . .
6.1 Accident Experience among Stolen V.e.hi.cle.s;. . . . . . . . .
6.1.1 Choice of Savling Yethod
. . . . . . . . . . . . .
6.1.2 Choice of Sampling Fram.es. . . . . . . . . . . . . .
6.1.3 Required Sample Sizes
. . . . . . . . . . . . .
6.1.4 Data Collect ion Techniques . . . . . . . . . . . . .
6.1.5 Data Collect ion Protocols
. . . . . . . . . . . . . . . . . .
6.1.6 Estimated Costs
. . . . . .
6.2 Accident Experience of Field-Dependent Drivers
. . . . . . . . . . . . . .
6.2.1 Review of the Literature . . . . . . . . . . . .
6.2.2 Choice of a Sampling Method. . . . . . . . . . . . .
6.2.3 Choice of a Sampling Fr.am.e . . . . . . . . . . . . .
6.2.4 Required Sample Sizes . . . . . . . . . . . . .
6.2.5 Data Collection Technique.s . . . . . . . . . . . . .
6.3 Vehicle Handling and Stability
. . . . . . . . .
6.3.1 Vehicle-Handling Study Obje.ct.iv.es. . . . . . . . . .
6.3.2 Choice of Sampling Method. . . . . . . . . . . . . .
6.3.3 Choice of Sampling Frame . . . . . . . .
6.3.4 Accident Data Collection Te.ch.ni.qu.e.s . . . . . . . .
6.3.5 Accident Data Requirements
. . . . . .
6.3.6 Control Group Data Collection Tec.hn.iq.u.es. . . . . .
6.3.7 Control Group Data Requirements
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
REFERENCES 83
. .
APPENDIX A: ACCIDENT CAUSATION BIBLICGRAPHY AND LITERATURE REVIEW 87
. . . . . . . . . . . . . . .
APPENDIX B: TRAFFIC CONFLICTS TECHNIQUE 151
. . . . . . . . . . . . . . . . . . . .
APPENDIX C: INDUCED EXPOSURE 155
. . . . . . . . . . . . .
APPENDIX D: FJWIEd OF TECHNIW DISCUSSIONS 163
. . . . . . . . . . . . . .
APPENDIX E: ACCIDENT CAUSATION WOWSHEET 173
LIST OF TABLES
. . . . . . . . . . . . .
Illustrative 2x2 Contingency Table
Illustrative 2x2 Contingency Table from Prospective Sampling
Illustrat.iv.e 2.x2. C.on.ti.ng.en.cy. T.ab.le. f.ro.m .Re.tr.os.pe.ct.iv.e . . . .
Sampling
. .
Illustrative Values from a Retrospective Sampling Design
. . . . . . . . . .
Required Group Sample Sizes for 2x2 Table
. . . . . . . . . . . .
Interaction Example for All Drivers.
. . . . . . . . . . .
Interaction Example for Normal Drivers
. . . . . . . . . . . .
Interaction Example for Fast Drivers
. . . . . . . . .
Second Interaction Example for All Drivers
. . . . . . . .
Second Interaction Example for Normal Drivers
. . . . . . . . .
Second Interaction Example for Fast Drivers
. . . . . . . . . . . .
Three-Dimensional. Interactive Array
. . . . . . . . . . . . . . . . . . . . . . . .
Collapsed Array
. . . . . . . .
Illustrative Distribution of Stolen Vehicles

SUMMARY
This report presents the results of Phase IA--the planning phase
preparatory to undertaking field work--of the "Vultilevel Study of Accident
"
Causation and Avoidance. The major task accomplished in this phase was the
selecti on of a methodology for conducting the accident-causation research to
follow. The steps required to achieve this goal were definition and
development of an accident-causation model, identification of research
,
topics for further study and development of experimental projects for
studying the selected factors.
A major focus of our review of accident-causation literature was the
search for a methodology appropriate to our goals and the current state of
knowledge about accident causation. For this reason, we expanded the scope
of our review to include work (primarily medical literature) outside of
highway safety research which might nevertheless suggest models for accident
causation or useful methodological tools.
The majority of the highway-related literature involved the case-study
approach to accident causation. This consists of the determination by
accident investigators and reconstruct ionist s of the causes of individual
accidents by applying a deterministic, test-of-necessity criterion to a
candidate set of causal factors. The medical literature, hwever, suugested
an alternative in a probabilistic model. This model focuses on comparing
the distrihtion of a factor in the accident population with that in a non-
accident ppulation. No attempt need be made to relate the factor to the
dynamics of a particular accident. We have called this alternative to the
case-study approach the statistical inference approach to accident-causation
research.
Thus, we were presented with a choice between two fwdamentally
different methodologies.
The case-study approach has an impressive body of literature behind it.
The basic principle is that the investigators develop, for each accident
under investigation, a model of how the accident happened. The components
of the model include all identified factors contributing to the accident,
The results reflect the prior experience and the considered judgements of
the accident investigators and reconstruct ionists concerning the importance
of possible factors contributing to the accident under consideration.
The statistical inference approach is recommended for the accident-
causation research to be undertaken during Phases I1 and I11 of the present
study. It embodies a probabilistic definition of accident causation adapted
from Greenberg's [l] medical model:
A factor F is said to be a cause of accidents if the conditional
probability of an accident occurring in the presence of F is greater
than the conditional probability of an accident occurring in the
absence of F, as determined under condition set E.
The probability expressions on both sides of the equation require
determination of the relative frequencies of both accidents and "non-
accidentsn in the presence and absence of factor F (under the set of
conditions E). Therefore, an essential feature of this approach is that
information on traffic units not involved in accidents be obtained.
Greenberg notes, in tern of his mltiple regression model adopted for
discussion of smking/lung-cancer issues, that one should look not only a t
main effects if one is interested in a particular variable. Interactive
t e r n mst te examined as well. In accident-causation research, the more
obvious causes--roughly equivalent to the "mi n ef f ectsV--have been
identified and quantified, to some extent, by the case-study approach. Our
conclusion is that consideration of these min effects, and the
countem.easures based on them, is no longer adequate. The study of higher-
order and interactive terms should now be undertaken, and this is better
accomplished by the statistical approach.
It is recanmended that the statistical inference approach to accident-
causation research be forn-ulated in terms of multidimensional contingency
tables. The probabilities in the inequality used to determine whether a
factor is causal are ratios of frequencies taken from such tables, and this
assists in gaining an understanding of the f actor-accident relationship.
Further, there exists a considerable--and growing-body of theory about
estimating the parameters of contingency table rrodels and about drawing
inferences from such models. Another important advantage of this approach
is that computer programs are readily available to assist in the analysis of