Acrid. Anal & Pro Vol. 24. No. 2. pp. ?Ol-210. Printed in Great Britain.
OtXll-4575192 S5.W + .oO 0 1992 Pergamon Press Ltd.
1992
THE RELATIONSHIP BETWEEN BLOOD ALCOHOL CONCENTRATION AND CRASH RESPONSIBILITY FOR FATALLY INJURED DRIVERS NANCY H. MOUNCE and OLGA J. PENDLETON Texas Transportation (Received
Institute, Texas A&M University, College Station, TX 77843-3135 U.S.A.
16 October 1990; in revised form 25 February 1991)
Abstract-Data for 59.5 fatally injured drivers in Texas were analyzed using logistic regression to determine the probability of crash responsibility as a function of blood alcohol concentration (BAC). Three independent panelists rated crash responsibility on a three-point scale, based on information contained in traffic accident reports. High inter-rater reliability was noted. Panel members were in agreement 98% of the time, unanimously assigning full responsibility in 61% of the cases and no responsibility in 31% of the cases. In addition to BAC. time of day and day of week were considered as variables in the regression model. A highly significant (P < .Ol) relationship was found between BAC and the probability of crash responsibility. Furthermore, results of the analysis support previous studies that noted a high rate of responsibility among fatally injured drivers, in general. even those with negative BAC test results. Such findings are of importance in evaluating driver impairment at various BACs and for determining other factors that interact with alcohol in driver fatality crashes.
INTRODUCTION
Beginning in 1983, the role of alcohol involvement in Texas driver fatalities has been analyzed by Texas Transportation Institute (TTI) for the State Department of Highways and Public Transportation (SDHPT). Each year, blood alcohol concentration (BAC) test results for fatally injured drivers are obtained from the medical examiners’ offices in the state. The BAC data are subsequently matched to accident reports submitted to the Texas Department of Public Safety (DPS). Analyses of these matched cases provide information on the extent of alcohol involvement in the available sample of driver fatality accidents, as well as descriptive statistics regarding the fatally injured drivers and the crashes in which they were involved. Results of the study have shown that legally intoxicated drivers in Texas (i.e. those with BACs Z- .lO) have accounted for close to half of the fatally injured drivers in the sample each year, although the proportion has decreased from 51% in 1983 to 45% in 1988 (Mounce et al. 1989). One question that has not been addressed previously, however, is the relationship between alcohol consumption and crash responsibility. When systematic assessments of crash responsibility are not available, drivers involved in single vehicle crashes have frequently been used as a surrogate for drivers considered to be “at fault” in their accidents. The previously referenced TTI study found that in 1988, over 70% of the fatally injured drivers involved in single-vehicle crashes had measurable alcohol in their systems (BAC > 0). By comparison, alcohol was detected for only 37% of the drivers killed in multiple vehicle collisions, while 55% of the total driver fatalities in the sample had positive BACs. Based on this simplistic definition of crash responsibility, it would appear that alcohol involvement among fatally injured responsible drivers was significantly higher than among the fatally injured drivers, in general, or among those involved in multiple vehicle collisions. REVIEW
OF THE
LITERATURE
Numerous studies have attempted to determine the extent to which intoxicated drivers are responsible for crashes resulting in injuries to themselves and others. Borkenstein et al. (1964) conducted a study in Grand Rapids, Michigan in which the relative risk of crash involvement was calculated for varying levels of alcohol consumption. Their 201
202
N. H. MOUNCEand 0. J. PENDLETON
study analyzed BAC distributions for crash-involved drivers and drivers in a comparable exposure sample who were not involved in collisions. Based on their findings, a driver with a BAC of .14 was approximately 20 times as likely to cause a crash as a driver with no alcohol in his system (Borkenstein et al. 1964). Other studies that benefitted from available exposure data confirmed the direct relationship between relative crash risk and alcohol consumption (Mayhew et al.. 1986; Farris, Malone, and Lilliefors 1976: Hurst 1970). Smith and Popham (1951) are credited with the first study in which estimates of crash responsibility were used to assess the relationship between the probability of accident involvement and blood alcohol concentration. In their research, responsibility was assigned on a ten-point scale in which factors beyond the driver’s control (e.g. environmental conditions or other drivers) were taken into consideration in addition to the actions of the respective driver. While acknowledged as a more precise approach, a continuous scale for assigning responsibility was not typically utilized in subsequent research. Instead, later studies (e.g. McCarroll and Haddon 1962; Baker and Spitz 1970; Perrine, Waller, and Harris 197 1; and Waller 1972) categorized responsibility as a dichotomy in which only one driver in a multiple-vehicle crash was determined to be at fault, and drivers involved in singlevehicle crashes were automatically assigned full responsibility. Various criteria were developed as guidelines for assessing crash responsibility, but the “fault vs. no fault” dichotomy was generally accepted, despite the fact that some sensitivity was lost due to the simplification (Terhune 1982). Crash responsibility ratings have been used to estimate relative crash risks, as well. Because exposure data are not readily available, drivers involved in multiple-vehicle collisions who were assigned no responsibility for their collision (i.e. those who were innocent victims) have been used to represent the non-crash-involved drivers on the road (Thorpe 1967). With this approach, accident data were used to estimate exposure data. Studies that relied on inferred exposure measures used analyses of crash responsibility to determine the relative risk of crash involvement for various driver groups. Despite variations in methodology, previous studies have produced fairly consistent results regarding the relationship between blood alcohol concentration and crash responsibility. There is still considerable interest, however, in determining the magnitude of the drinking and driving problem in this country. The extent to which alcohol-imparied drivers are responsible for injury-producing crashes must be established if the societal costs associated with drinking and driving are to be fully realized. METHODOLOGY
The present study sample included all fatally injured motor vehicle drivers who were involved in crashes in 1988 and who were autopsied at one of the twelve medical examiner’s offices in Texas (Mounce et al. 1989). This sample of 59.5 individuals represented approximately 31% of the statewide driver fatalities in 1988. Traffic accident reports for these individuals. obtained from the Texas Department of Public Safety* provided the basis for the assessments of crash responsibility. The accident report narrative, which represents the investigating officer’s interpretation of the crash events, was determined to be the most reliable and consistent data source for purposes of assigning responsibility ratings. The possibility that these accident reports may be biased should not be overlooked. For example, the officer’s interpretation of the crash may be consciously or unintentionally influenced by any evidence of alcohol consumption on the part of one or both drivers. Furthermore, investigating officers may be predisposed to identify human behavior, as opposed to vehicular or environmental factors. as having caused or contributed to a particular crash. While these potential biases are acknowledged, the traffic accident reports were still felt to contain the most consistent and reliable source of data for purposes of this study. The information contained in each accident narrative was reviewed by a panel of three researchers. These individuals assigned independent ratings of crash responsibility
BAC and crash responsibility
203
Table 1. Rater reliability Panelist 1 Not responsible
Panelist 2
Partially responsible 1 17 2
90 1 0
Not Responsible Partially responsible Fully responsible
Fully responsible 2 I 180
Panelist I Not responsible
Panelist 3
Partially responsible 1 17 2
91 0 0
Not Responsible Partially Responsible Fully Responsible
Fully responsible 0 3 180
Panelist 2 Not resoonsible
Panelist 3
Partially resoonsible 1 18 I
91 1 I
Not Responsible Partially Responsible Fully Responsible
Fully resoonsible 0 I 180
using an approach introduced by McCarroll and Haddon (1962), and later adapted by Perrine et al. (1971), Waller (1972), and Baker and Spitz (1970). Specifically, all fatally injured drivers involved in single-vehicle crashes were assigned full responsibility. Drivers involved in multiple-vehicle collisions were coded on the following three-point scale: 012-
not responsible partially responsible (where more than one driver in the collision was judged to be at fault) fully responsible
The driver of the vehicle that initiated the crash was generally assigned full responsibility. However, the partial responsibility category was included in this analysis to handle instances when more than one driver clearly contributed to a particular crash. RESULTS
Rater reliability The three panel members were in substantial agreement in assessing crash responsibility. Table 1 illustrates this uniformity by showing the frequency of cases for which each panelist gave a score of 0 (not responsible), 1 (partially responsible), or 2 (fully responsible). The panel members were in agreement 99% of the time, unanimously assigning full responsibility in 61% of the cases and no responsibility in 31% of the cases. Because of the high agreement among panelists, a single responsibility score was assessed by taking the majority rating of the three panel members. The small number of cases for which partial responsibility was assigned led to a decision to combine the partial and full responsibility classifications. Consequently, alf 294 multiple-vehicle drivers were subsequently reassigned to either the not responsible or fully responsible Table 2. Descriptive statistics for fatally injured drivers
Single-vehicle crashes Multiple-vehicle crashes
n
Average BAC
Percentage BAC>O
Percentage BAC 2 .I0
Average BAC of BACs > 0
301 294
.073 .050
3x 27
33 22
.20 .19
204
N. H. MOUNCE and 0. J. PENDLETON
BAC Fig. 1. Distribution
of non-zero
BACs
for all fatally
injured
drivers
group. The assignment of full responsibility to all drivers involved in single-vehicle crashes was an arbitrary decision. Hence, results of the data analyses are reported separately for the multiple-vehicle drivers and the total sample. Descriptive statistics
Table 2 provides descriptive statistics for the 595 motor vehicle driver fatalities included in the study sample. The proportions of fatally injured drivers involved in single-vehicle and multiple-vehicle crashes were quite comparable (51% vs. 49%, respectively). However, the single-vehicle drivers had a noticeably higher average BAC, as well as higher percentages of drinking drivers (BACs > 0) and legally intoxicated drivers (BACs 2 .lO). Among the drivers who had measurable alcohol in their systems, the average BACs for single- and multiple-vehicle drivers were roughly equivalent. A frequency distribution of the positive BACs (i.e. BACs > 0) for all crash-involved
m
Single
Vehicles
m
Multiple
Vehicles
BAC Fig. 2. Distribution
of non-zero
BACs
for fatally
injured
drivers
in single- vs. multiple-vehicle
crashes
BAC and crash responsibility
205
Table 3. Descriptive statistics for responsible vs. not responsible fatally injured drivers in multiplevehicle crashes
Responsible Not resoonsible
n
Average BAC
201 93
.062 ,024
Percentage BAC>O 32 16
Percentage BAC 2 .lO
Average BAC of BACs > 0
26 13
.19 .15
drivers is provided in Fig. 1. Comparable distributions for drivers involved in singlevehicle and multiple-vehicle crashes are provided in Fig. 2. While the BAC data were fairly normally distributed in each instance, the distribution for drivers involved in singlevehicle crashes appeared to be skewed somewhat in the direction of higher BACs. It is noteworthy that the vast majority of drivers in each group had BACs well in excess of the legal intoxication limit. In fact, the highest frequencies were found in the .20-.25 BAC range across crash types. Furthermore, BACs above .35 were not uncommon among these fatally injured drivers. These findings support the notion that the majority of drivers killed in alcohol-related motor vehicle crashes are not merely social drinkers. Descriptive data on the 294 drivers involved in multiple vehicle crashes are provided in Table 3. As shown, the average BAC was considerably higher for the responsible drivers than for those determined not to be responsible (.06 vs. .02, respectively). Similarly, both the percentage of drivers with positive BACs and the percentage of drivers who were legally intoxicated were twice as high for the drivers judged to be responsible as those deemed not responsible for their collisions. Among those individuals with measurable alcohol in their system, the average BAC was somewhat higher for the responsible drivers than for the not responsible group (. 19 vs. .15). When drivers involved in single-vehicle crashes were added to the data set, the number of responsible drivers increased to 502 (84%) (Table 4). However, the addition of the single-vehicle drivers did not appreciably change the average BAC for the drivers determined to be responsible (.069 vs. .062 for multiple-vehicle drivers only). Similarly, the percentage of drivers with positive BACs and the percentage with BACs above the legal intoxication limit were increased by only a few percentage points with the addition of the single-vehicle drivers. The average BAC for drivers with measurable alcohol in their systems was unchanged by the addition of the drivers involved in single-vehicle crashes. The fact that the inclusion of single vehicle drivers did not appreciably distort the BAC distribution for drivers determined to be at fault in their crashes was somewhat surprising, given the arbitrary assignment of full responsibility to all individuals in the single vehicle group. Table 5 shows the percentage of drivers who had been drinking and those who were legally intoxicated, who were assigned responsibility for their collisions. Sixty-eight percent of all fatally injured multiple vehicle drivers were assigned full responsibility for their collisions. This percentage increased to over 80% for those drivers involved in multiple-vehicle crashes who had consumed alcohol. Finally, among the legally intoxicated multiple-vehicle drivers, 87% were determined to be at fault. For total crashes (single- and multiple-vehicle), 84% of the fatally injured drivers, in general, were assigned responsibility. Over 90% of the drivers with positive BACs. as well as those with BACs over the legal intoxication limit, were assigned full responsibility for their crashes. Table 4. Descriptive Statistics for Responsible vs. Not Responsible Fatally Injured Drivers (Single and Multiple Vehicle Crashes Combined)
Responsible Not responsible
n
Average BAC
Percentage BAC > 0
Percentage BAC 2 .lO
Average BAC of BACs > 0
502 93
.069 ,024
35 16
30 13
.19 .15
N. H. MOUNCE and 0. J. PENDLETON
206 Table
5. Crash
teponsibility
by blood alcohol injured drivers Percentage
concentration
of responsible BAC
BAC > 0 Multiple-Vehicle Total crashes
crashes
of fatally
drivers
2 .I0
Total
x7 93
68 x4
Relationship between crash responsibility and BAC
The relationship between crash responsibility and blood alcohol concentration was analyzed using logistic regression. With this statistical model, the probability of responsibility is related to the BAC value of a fatally injured driver as follows:
P(C) = (emamh’ + l)-’
where P(c) is the probability of responsibility (ranging from zero to one), and x is the blood alcohol concentration of the driver. As in ordinary least squares regression, the model coefficients are estimated from the data, were a represents the intercept of the curve and b represents the slope. Table 6 lists the results of these models fit to all drivers in the data set with measurable blood alcohol concentrations (BACs > 0). The model coefficients and significance level of the fitted model (p-value for the model Chi-square) are reported. There was a highly significant relationship between BAC and the probability of crash responsibility for all drivers (p < .Ol), as well as for the drivers involved in multiplevehicle crashes (p < .Ol). The probability of a driver’s being judged at fault in a crash increased as BAC increased for both groups, and the rate of increase (i.e. the slope of the curve) was similar for all drivers and for the multiple vehicle drivers (6.68 and 6.81, respectively) (Fig. 3). One of the advantages of logistic regression modeling is that it provides predictions of the probability of responsibility as a function of the BAC of a driver killed in a crash. For example, we would predict from Fig. 3 that a driver with a BAC = .lO killed in a multiple-vehicle collision has a 72% probability of being at fault. When single- and multiple-vehicle crashes are combined, the probability that a driver with a BAC = .lO is at fault in the crash is 88%. In previous studies of fatally injured drivers (McCarrol and Haddon 1962; Baker and Spitz 1970; Perrine et al. 1971; Waller 1972), those who had consumed alcohol generally had higher responsibility rates than did their sober counterparts. However, the relationship between BAC and crash responsibility was not as strong as might have been expected. Even at low BACs, a high proportion of the drivers in each study were determined to be responsible (Fig. 4). This finding was thought to be due, in part, to the fact that only fatally injured drivers were included in the analyses. Furthermore, the dichotomous classification of responsibility was believed to have reduced the ability to statistically detect a relationship between crash responsibility and BAC values.
Table 6. Logistic
Multiple-vehicle Total crashes
crashes
regression
models
a
b
.281 1.320
6.81 6.68
Significance .W23 .OWl
207
BAC and crash responsibility
l.O-
0.5
I .oo
I .05
I
I .lO
--o-
All Vehicles
-Cl-
Multiple
I
Vehicles
I
I .15
I ‘20
BAC Fig. 3. Probability of crash responsibility as a function of driver BAC.
By comparison, the proportion of responsible drivers broken down by BAC category in the present sample is provided below: BAC .OOO .Ol-.04 .os-.09 .lO-.14 .X5-.19 .20-.24 .25-.29 .30-.34 .35-.39 .40+
% Responsible 80.6
80.0 100.0
84.6 88.6 92.0 100.00 100.00 100.00 100.00
% Not responsible 19.4 20.0 0.0 15.4 11.4 8.0 0.0 0.0 0.0 0.0
Analysis of temporal factors
While alcohol is acknowledged as a major determinant of motor vehicle crashes, the fact that a Large proportion of sober drivers were determined to be at fault in their colhsions suggests there may be other factors that are causally related to crash involvement. For example, previous ‘IT1 studies have shown that alcohol-related crashes are overrepresented at certain times of the day and days of the week (Mounce et al. 1989). Table 7 shows the average BAC and sample sizes for the data classified by time of day, weekend vs. weekday, and crash responsibility. Time of day was broken down into four periods: 6 A.M. to 6 P.M. (to reflect commuting hours), 6 P.M.-11 P.M., 11 P.M.-~ A.M., and 3 A.M.-~ A.M. Weekend was defined as 6 P.M. Friday through 6 A.M. Monday. Among the drivers involved in multiple vehicle crashes, the average BAC was consistently higher for drivers who were responsible than those who were not responsible, regardless of the day of week. The only exception was the 6 P.M.-II P.M. time period on weekends, when the average BAC of the not-responsible drivers was .lI2, compared to an average BAC of ,084 for the responsible drivers. It should be noted, however,
N. H. MOI.NCE and 0. J.
208
PENDL~KJN
McCarroll & Haddon ( 1962) Baker & Spitz (1970) A Perrine & Wailer (197 1) n Waller (1972) 0
0
0% ;
I
I
0
.OS
BLOOD
I
.lO
ALCOHOL
.15
I
.25
1 .30
CONCENTRATION-PERCENT
(Midpoints Fig. 4. Responsibility
I
.20
w/w
of BAC Groupings)
rates and BACs of fatally
injured
drivers
(Source:
Terhune
1982)
that only five individuals were included in the category of not-responsible drivers who were involved in crashes during that time period. The same general trend was noted when the group of responsible drivers was expanded to include both single- and multiple-vehicle crashes. In virtually every category, the responsible drivers had higher average BACs than their not-responsible counterparts. Among the responsible drivers. time of day variations were quite common during the weekdays, when average BACs increased dramatically during the late night and early morning hours. There was much less discrepancy in the BACs noted throughout the weekend period. This pattern was observed for responsible drivers in multiple-vehicle collisions, as well as those in the combined sample of single- and multiple-vehicle crashes. For the drivers who were not assigned crash responsibility, weekend crashes more closely resembled the pattern observed for weekday collisions. In other words, the average BACs noted during the evening and early morning hours were noticeably higher than the average BAC observed during the daytime (6 A.M.6 P.M.). These results are depicted graphically in Figs. 5 and 6. There appeared to be no
Table 7. Avrragc
BACs
for drivers
Multiple-vehicle Responsible
killed on weekends
vs. weekdays Total crashes
crashes:
Responsible
Not responsible
( 51
Ave. BAC ,077 ,110
( 47) ( 38)
,053 .076
9) ( 5)
,076 ,064
1’:;‘;
.004 .03 1 ,017 ,000
(32) (18)
,024 ,085 ,092 ,070
Weekend 6AM-6PM hPM-1 1PM
Ave. BAC .070 ,084
(12) (23) (20)
Ave. BAC ,000 ,112
(II) (13)
13AM-6AM IPM-3AM
,089 ,098
(28) ( 8)
Weekday 6AM-6PM 6PM-11PM 1 lPM-3AM 3AM-6AM
,022 ,074 ,124 ,048
(68) I::; (11)
( 9 ( 2)
(11)
K:l
( 66) ( 26)
BAC and crash responsibility
-o-+-
0.16
Responsible Not Responsible
6pm - llpm
6am - 6pm
209
1 lpm - 3am
3am - 6am
TIME Fig. 5. Average BACs by time of day: weekdays.
interaction between time of day and crash responsibility for weekday collisions (Fig. 5). The average BACs for the responsible drivers had the same time of day trend as that observed for the not responsible drivers. With the exception of the 6 P.M.-11 P.M. time period, the same finding was noted for crashes that occurred on weekends (Fig. 6). The weekday average BAC for the responsible and not responsible drivers was .09 vs. .02, respectively. During the weekend, however, the average BACs for the two groups of drivers were equivalent (.08). The small number of individuals in the not-
0.20 1 -oe
0.16 -
Responsible Not Responsible
0.12-
0.08 -
0.04 -
0.00
I 6am - 6pm
I 6pm - llpm
llpm
I - 3am
TIME Fig. 6. Average BACs by time of day: weekends. AAP24:2-H
I 3am - 6am
N. H. MOUNCE and 0. J. PENDLETON
210
responsible analysis.
group of drivers
should
CONCLUSIONS
be kept in mind when interpreting
AND
the results
of this
IMPLICATIONS
There was a very high proportion of individuals in this sample of fatally injured drivers who were determined to be at fault in the crashes in which they were killed. This may be an indication that drivers who are at fault in multiple-vehicle crashes tend to have more severe collisions. Conversely, those drivers who are not at fault in multiplevehicle collisions tend to be involved in less severe crashes, or may exhibit behaviors (e.g safety belt use) that would reduce the likelihood of fatal injury. The fact that surviving drivers were not represented in this data set may explain the high percentage of responsible drivers in the analysis. As previously discussed. results of this study were consistent with results of other studies of fatally injured drivers in that high rates of responsibility were evidenced among even the sober drivers in the sample (McCarrol and Haddon 1962; Baker and Spitz 1970; Perrine et al. 1971; and Wailer 1972). A relatively small difference was found in the proportion of responsible drivers among those who were sober and those who had been drinking. This finding emphasizes the need to identify other factors that, alone or in combination with alcohol, may increase the risk of crash involvement. A brief examination of temporal factors that were felt to interact with alcohol revealed a distinct relationship between crash responsibility and BAC for weekday crashes. Specifically, as BAC increased, the probability of responsibility for the crash increased. and the average BAC was considerably greater among the drivers determined to be at fault in weekday crashes than among those who were not at fault. However, this relationship was not as apparent for weekend accidents. Alcohol use was quite evident in crashes that occurred throughout the designated weekend period, even among the not responsible group of drivers. There are clearly other factors that need to be examined to more fully explain the relationship between alcohol impairment and crash responsibility. It is recommended that future studies include analyses of driver age, gender, driving experience, and other risk-taking behaviors to determine the extent to which these variables, either alone or in combination with alcohol, increase an individual’s likelihood of contributing to a crash.
REFERENCES Baker, S. P.; Spitz. W. U. Age effects and autopsy evidence of disease in fatally injured drivers. JAMA 214:1079-1088; lY70. Borkenstein. R. F.; Crowther, R. F.; Shumate. R. P.; &cl. W. 6.; Zylman. R. The role of the drinking driver in traffic accidents. Bloomington, IN: Indiana University. Department of Police Administration; 1964. Farris, R.; Malone, T. B.; Lilliefors. H. A. Comparison of alcohol involvement in exposed and injured drivers. Alexandria, VA: Essex Corp.; 1976. Hurst, P. M. Estimating the effectiveness of blood alcohol limits. Behav. Res. Highw. Saf. 1(2):87-YY: lY70. Mayhew, D. R.; Donelson. A. C.: Beirness, D. J.; Simpson. H. M. Youth, alcohol and relative risk of crash involvement. Accid. Anal. Prev. 18:273-287; 1986. McCarroll. J. R.: Haddon. W., Jr. A controlled study of fatal automobile accidents in New York City. J. Chronic Dis. 15:811-826: 1962. Mounce, N. H.; Acuff. A.; Peart. S.; Croasdell, D. Alcohol involvement in Texas driver fatalities. College Station, TX: Texas Transportation Institute. Texas A&M University; 1989. Perrine. M. W.; Wailer, J. A.; Harris. L. S. Alcohol and highway safety: Behavioral and medical aspects. Washington, DC: U.S. Department of Trasportation; 1971. Smith. H. W.; Popham, R. E. Blood alcohol levels in relation to driving. Can. Med. Assoc. J. hS:325-328; 1951. Terhune, Kenneth W. An evaluation of crash culpability to assess alcohol and drug impairment effects. 26th Annual Proceedings. AAAM; Arlington Heights, IL: American Association for Automotive Medicine: 1982:329-348. Thorpe, J. D. Calculating relative involvement rates in accidents without determining exposure. Traffic Saf Res Rev 11:3-g; 1967. Wailer, J. A. Factors associated with alcohol and responsibility. Q. J. Stud. Alcohol 33: lhil-170: lY72.
OtXll-4575192 S5.W + .oO 0 1992 Pergamon Press Ltd.
1992
THE RELATIONSHIP BETWEEN BLOOD ALCOHOL CONCENTRATION AND CRASH RESPONSIBILITY FOR FATALLY INJURED DRIVERS NANCY H. MOUNCE and OLGA J. PENDLETON Texas Transportation (Received
Institute, Texas A&M University, College Station, TX 77843-3135 U.S.A.
16 October 1990; in revised form 25 February 1991)
Abstract-Data for 59.5 fatally injured drivers in Texas were analyzed using logistic regression to determine the probability of crash responsibility as a function of blood alcohol concentration (BAC). Three independent panelists rated crash responsibility on a three-point scale, based on information contained in traffic accident reports. High inter-rater reliability was noted. Panel members were in agreement 98% of the time, unanimously assigning full responsibility in 61% of the cases and no responsibility in 31% of the cases. In addition to BAC. time of day and day of week were considered as variables in the regression model. A highly significant (P < .Ol) relationship was found between BAC and the probability of crash responsibility. Furthermore, results of the analysis support previous studies that noted a high rate of responsibility among fatally injured drivers, in general. even those with negative BAC test results. Such findings are of importance in evaluating driver impairment at various BACs and for determining other factors that interact with alcohol in driver fatality crashes.
INTRODUCTION
Beginning in 1983, the role of alcohol involvement in Texas driver fatalities has been analyzed by Texas Transportation Institute (TTI) for the State Department of Highways and Public Transportation (SDHPT). Each year, blood alcohol concentration (BAC) test results for fatally injured drivers are obtained from the medical examiners’ offices in the state. The BAC data are subsequently matched to accident reports submitted to the Texas Department of Public Safety (DPS). Analyses of these matched cases provide information on the extent of alcohol involvement in the available sample of driver fatality accidents, as well as descriptive statistics regarding the fatally injured drivers and the crashes in which they were involved. Results of the study have shown that legally intoxicated drivers in Texas (i.e. those with BACs Z- .lO) have accounted for close to half of the fatally injured drivers in the sample each year, although the proportion has decreased from 51% in 1983 to 45% in 1988 (Mounce et al. 1989). One question that has not been addressed previously, however, is the relationship between alcohol consumption and crash responsibility. When systematic assessments of crash responsibility are not available, drivers involved in single vehicle crashes have frequently been used as a surrogate for drivers considered to be “at fault” in their accidents. The previously referenced TTI study found that in 1988, over 70% of the fatally injured drivers involved in single-vehicle crashes had measurable alcohol in their systems (BAC > 0). By comparison, alcohol was detected for only 37% of the drivers killed in multiple vehicle collisions, while 55% of the total driver fatalities in the sample had positive BACs. Based on this simplistic definition of crash responsibility, it would appear that alcohol involvement among fatally injured responsible drivers was significantly higher than among the fatally injured drivers, in general, or among those involved in multiple vehicle collisions. REVIEW
OF THE
LITERATURE
Numerous studies have attempted to determine the extent to which intoxicated drivers are responsible for crashes resulting in injuries to themselves and others. Borkenstein et al. (1964) conducted a study in Grand Rapids, Michigan in which the relative risk of crash involvement was calculated for varying levels of alcohol consumption. Their 201
202
N. H. MOUNCEand 0. J. PENDLETON
study analyzed BAC distributions for crash-involved drivers and drivers in a comparable exposure sample who were not involved in collisions. Based on their findings, a driver with a BAC of .14 was approximately 20 times as likely to cause a crash as a driver with no alcohol in his system (Borkenstein et al. 1964). Other studies that benefitted from available exposure data confirmed the direct relationship between relative crash risk and alcohol consumption (Mayhew et al.. 1986; Farris, Malone, and Lilliefors 1976: Hurst 1970). Smith and Popham (1951) are credited with the first study in which estimates of crash responsibility were used to assess the relationship between the probability of accident involvement and blood alcohol concentration. In their research, responsibility was assigned on a ten-point scale in which factors beyond the driver’s control (e.g. environmental conditions or other drivers) were taken into consideration in addition to the actions of the respective driver. While acknowledged as a more precise approach, a continuous scale for assigning responsibility was not typically utilized in subsequent research. Instead, later studies (e.g. McCarroll and Haddon 1962; Baker and Spitz 1970; Perrine, Waller, and Harris 197 1; and Waller 1972) categorized responsibility as a dichotomy in which only one driver in a multiple-vehicle crash was determined to be at fault, and drivers involved in singlevehicle crashes were automatically assigned full responsibility. Various criteria were developed as guidelines for assessing crash responsibility, but the “fault vs. no fault” dichotomy was generally accepted, despite the fact that some sensitivity was lost due to the simplification (Terhune 1982). Crash responsibility ratings have been used to estimate relative crash risks, as well. Because exposure data are not readily available, drivers involved in multiple-vehicle collisions who were assigned no responsibility for their collision (i.e. those who were innocent victims) have been used to represent the non-crash-involved drivers on the road (Thorpe 1967). With this approach, accident data were used to estimate exposure data. Studies that relied on inferred exposure measures used analyses of crash responsibility to determine the relative risk of crash involvement for various driver groups. Despite variations in methodology, previous studies have produced fairly consistent results regarding the relationship between blood alcohol concentration and crash responsibility. There is still considerable interest, however, in determining the magnitude of the drinking and driving problem in this country. The extent to which alcohol-imparied drivers are responsible for injury-producing crashes must be established if the societal costs associated with drinking and driving are to be fully realized. METHODOLOGY
The present study sample included all fatally injured motor vehicle drivers who were involved in crashes in 1988 and who were autopsied at one of the twelve medical examiner’s offices in Texas (Mounce et al. 1989). This sample of 59.5 individuals represented approximately 31% of the statewide driver fatalities in 1988. Traffic accident reports for these individuals. obtained from the Texas Department of Public Safety* provided the basis for the assessments of crash responsibility. The accident report narrative, which represents the investigating officer’s interpretation of the crash events, was determined to be the most reliable and consistent data source for purposes of assigning responsibility ratings. The possibility that these accident reports may be biased should not be overlooked. For example, the officer’s interpretation of the crash may be consciously or unintentionally influenced by any evidence of alcohol consumption on the part of one or both drivers. Furthermore, investigating officers may be predisposed to identify human behavior, as opposed to vehicular or environmental factors. as having caused or contributed to a particular crash. While these potential biases are acknowledged, the traffic accident reports were still felt to contain the most consistent and reliable source of data for purposes of this study. The information contained in each accident narrative was reviewed by a panel of three researchers. These individuals assigned independent ratings of crash responsibility
BAC and crash responsibility
203
Table 1. Rater reliability Panelist 1 Not responsible
Panelist 2
Partially responsible 1 17 2
90 1 0
Not Responsible Partially responsible Fully responsible
Fully responsible 2 I 180
Panelist I Not responsible
Panelist 3
Partially responsible 1 17 2
91 0 0
Not Responsible Partially Responsible Fully Responsible
Fully responsible 0 3 180
Panelist 2 Not resoonsible
Panelist 3
Partially resoonsible 1 18 I
91 1 I
Not Responsible Partially Responsible Fully Responsible
Fully resoonsible 0 I 180
using an approach introduced by McCarroll and Haddon (1962), and later adapted by Perrine et al. (1971), Waller (1972), and Baker and Spitz (1970). Specifically, all fatally injured drivers involved in single-vehicle crashes were assigned full responsibility. Drivers involved in multiple-vehicle collisions were coded on the following three-point scale: 012-
not responsible partially responsible (where more than one driver in the collision was judged to be at fault) fully responsible
The driver of the vehicle that initiated the crash was generally assigned full responsibility. However, the partial responsibility category was included in this analysis to handle instances when more than one driver clearly contributed to a particular crash. RESULTS
Rater reliability The three panel members were in substantial agreement in assessing crash responsibility. Table 1 illustrates this uniformity by showing the frequency of cases for which each panelist gave a score of 0 (not responsible), 1 (partially responsible), or 2 (fully responsible). The panel members were in agreement 99% of the time, unanimously assigning full responsibility in 61% of the cases and no responsibility in 31% of the cases. Because of the high agreement among panelists, a single responsibility score was assessed by taking the majority rating of the three panel members. The small number of cases for which partial responsibility was assigned led to a decision to combine the partial and full responsibility classifications. Consequently, alf 294 multiple-vehicle drivers were subsequently reassigned to either the not responsible or fully responsible Table 2. Descriptive statistics for fatally injured drivers
Single-vehicle crashes Multiple-vehicle crashes
n
Average BAC
Percentage BAC>O
Percentage BAC 2 .I0
Average BAC of BACs > 0
301 294
.073 .050
3x 27
33 22
.20 .19
204
N. H. MOUNCE and 0. J. PENDLETON
BAC Fig. 1. Distribution
of non-zero
BACs
for all fatally
injured
drivers
group. The assignment of full responsibility to all drivers involved in single-vehicle crashes was an arbitrary decision. Hence, results of the data analyses are reported separately for the multiple-vehicle drivers and the total sample. Descriptive statistics
Table 2 provides descriptive statistics for the 595 motor vehicle driver fatalities included in the study sample. The proportions of fatally injured drivers involved in single-vehicle and multiple-vehicle crashes were quite comparable (51% vs. 49%, respectively). However, the single-vehicle drivers had a noticeably higher average BAC, as well as higher percentages of drinking drivers (BACs > 0) and legally intoxicated drivers (BACs 2 .lO). Among the drivers who had measurable alcohol in their systems, the average BACs for single- and multiple-vehicle drivers were roughly equivalent. A frequency distribution of the positive BACs (i.e. BACs > 0) for all crash-involved
m
Single
Vehicles
m
Multiple
Vehicles
BAC Fig. 2. Distribution
of non-zero
BACs
for fatally
injured
drivers
in single- vs. multiple-vehicle
crashes
BAC and crash responsibility
205
Table 3. Descriptive statistics for responsible vs. not responsible fatally injured drivers in multiplevehicle crashes
Responsible Not resoonsible
n
Average BAC
201 93
.062 ,024
Percentage BAC>O 32 16
Percentage BAC 2 .lO
Average BAC of BACs > 0
26 13
.19 .15
drivers is provided in Fig. 1. Comparable distributions for drivers involved in singlevehicle and multiple-vehicle crashes are provided in Fig. 2. While the BAC data were fairly normally distributed in each instance, the distribution for drivers involved in singlevehicle crashes appeared to be skewed somewhat in the direction of higher BACs. It is noteworthy that the vast majority of drivers in each group had BACs well in excess of the legal intoxication limit. In fact, the highest frequencies were found in the .20-.25 BAC range across crash types. Furthermore, BACs above .35 were not uncommon among these fatally injured drivers. These findings support the notion that the majority of drivers killed in alcohol-related motor vehicle crashes are not merely social drinkers. Descriptive data on the 294 drivers involved in multiple vehicle crashes are provided in Table 3. As shown, the average BAC was considerably higher for the responsible drivers than for those determined not to be responsible (.06 vs. .02, respectively). Similarly, both the percentage of drivers with positive BACs and the percentage of drivers who were legally intoxicated were twice as high for the drivers judged to be responsible as those deemed not responsible for their collisions. Among those individuals with measurable alcohol in their system, the average BAC was somewhat higher for the responsible drivers than for the not responsible group (. 19 vs. .15). When drivers involved in single-vehicle crashes were added to the data set, the number of responsible drivers increased to 502 (84%) (Table 4). However, the addition of the single-vehicle drivers did not appreciably change the average BAC for the drivers determined to be responsible (.069 vs. .062 for multiple-vehicle drivers only). Similarly, the percentage of drivers with positive BACs and the percentage with BACs above the legal intoxication limit were increased by only a few percentage points with the addition of the single-vehicle drivers. The average BAC for drivers with measurable alcohol in their systems was unchanged by the addition of the drivers involved in single-vehicle crashes. The fact that the inclusion of single vehicle drivers did not appreciably distort the BAC distribution for drivers determined to be at fault in their crashes was somewhat surprising, given the arbitrary assignment of full responsibility to all individuals in the single vehicle group. Table 5 shows the percentage of drivers who had been drinking and those who were legally intoxicated, who were assigned responsibility for their collisions. Sixty-eight percent of all fatally injured multiple vehicle drivers were assigned full responsibility for their collisions. This percentage increased to over 80% for those drivers involved in multiple-vehicle crashes who had consumed alcohol. Finally, among the legally intoxicated multiple-vehicle drivers, 87% were determined to be at fault. For total crashes (single- and multiple-vehicle), 84% of the fatally injured drivers, in general, were assigned responsibility. Over 90% of the drivers with positive BACs. as well as those with BACs over the legal intoxication limit, were assigned full responsibility for their crashes. Table 4. Descriptive Statistics for Responsible vs. Not Responsible Fatally Injured Drivers (Single and Multiple Vehicle Crashes Combined)
Responsible Not responsible
n
Average BAC
Percentage BAC > 0
Percentage BAC 2 .lO
Average BAC of BACs > 0
502 93
.069 ,024
35 16
30 13
.19 .15
N. H. MOUNCE and 0. J. PENDLETON
206 Table
5. Crash
teponsibility
by blood alcohol injured drivers Percentage
concentration
of responsible BAC
BAC > 0 Multiple-Vehicle Total crashes
crashes
of fatally
drivers
2 .I0
Total
x7 93
68 x4
Relationship between crash responsibility and BAC
The relationship between crash responsibility and blood alcohol concentration was analyzed using logistic regression. With this statistical model, the probability of responsibility is related to the BAC value of a fatally injured driver as follows:
P(C) = (emamh’ + l)-’
where P(c) is the probability of responsibility (ranging from zero to one), and x is the blood alcohol concentration of the driver. As in ordinary least squares regression, the model coefficients are estimated from the data, were a represents the intercept of the curve and b represents the slope. Table 6 lists the results of these models fit to all drivers in the data set with measurable blood alcohol concentrations (BACs > 0). The model coefficients and significance level of the fitted model (p-value for the model Chi-square) are reported. There was a highly significant relationship between BAC and the probability of crash responsibility for all drivers (p < .Ol), as well as for the drivers involved in multiplevehicle crashes (p < .Ol). The probability of a driver’s being judged at fault in a crash increased as BAC increased for both groups, and the rate of increase (i.e. the slope of the curve) was similar for all drivers and for the multiple vehicle drivers (6.68 and 6.81, respectively) (Fig. 3). One of the advantages of logistic regression modeling is that it provides predictions of the probability of responsibility as a function of the BAC of a driver killed in a crash. For example, we would predict from Fig. 3 that a driver with a BAC = .lO killed in a multiple-vehicle collision has a 72% probability of being at fault. When single- and multiple-vehicle crashes are combined, the probability that a driver with a BAC = .lO is at fault in the crash is 88%. In previous studies of fatally injured drivers (McCarrol and Haddon 1962; Baker and Spitz 1970; Perrine et al. 1971; Waller 1972), those who had consumed alcohol generally had higher responsibility rates than did their sober counterparts. However, the relationship between BAC and crash responsibility was not as strong as might have been expected. Even at low BACs, a high proportion of the drivers in each study were determined to be responsible (Fig. 4). This finding was thought to be due, in part, to the fact that only fatally injured drivers were included in the analyses. Furthermore, the dichotomous classification of responsibility was believed to have reduced the ability to statistically detect a relationship between crash responsibility and BAC values.
Table 6. Logistic
Multiple-vehicle Total crashes
crashes
regression
models
a
b
.281 1.320
6.81 6.68
Significance .W23 .OWl
207
BAC and crash responsibility
l.O-
0.5
I .oo
I .05
I
I .lO
--o-
All Vehicles
-Cl-
Multiple
I
Vehicles
I
I .15
I ‘20
BAC Fig. 3. Probability of crash responsibility as a function of driver BAC.
By comparison, the proportion of responsible drivers broken down by BAC category in the present sample is provided below: BAC .OOO .Ol-.04 .os-.09 .lO-.14 .X5-.19 .20-.24 .25-.29 .30-.34 .35-.39 .40+
% Responsible 80.6
80.0 100.0
84.6 88.6 92.0 100.00 100.00 100.00 100.00
% Not responsible 19.4 20.0 0.0 15.4 11.4 8.0 0.0 0.0 0.0 0.0
Analysis of temporal factors
While alcohol is acknowledged as a major determinant of motor vehicle crashes, the fact that a Large proportion of sober drivers were determined to be at fault in their colhsions suggests there may be other factors that are causally related to crash involvement. For example, previous ‘IT1 studies have shown that alcohol-related crashes are overrepresented at certain times of the day and days of the week (Mounce et al. 1989). Table 7 shows the average BAC and sample sizes for the data classified by time of day, weekend vs. weekday, and crash responsibility. Time of day was broken down into four periods: 6 A.M. to 6 P.M. (to reflect commuting hours), 6 P.M.-11 P.M., 11 P.M.-~ A.M., and 3 A.M.-~ A.M. Weekend was defined as 6 P.M. Friday through 6 A.M. Monday. Among the drivers involved in multiple vehicle crashes, the average BAC was consistently higher for drivers who were responsible than those who were not responsible, regardless of the day of week. The only exception was the 6 P.M.-II P.M. time period on weekends, when the average BAC of the not-responsible drivers was .lI2, compared to an average BAC of ,084 for the responsible drivers. It should be noted, however,
N. H. MOI.NCE and 0. J.
208
PENDL~KJN
McCarroll & Haddon ( 1962) Baker & Spitz (1970) A Perrine & Wailer (197 1) n Waller (1972) 0
0
0% ;
I
I
0
.OS
BLOOD
I
.lO
ALCOHOL
.15
I
.25
1 .30
CONCENTRATION-PERCENT
(Midpoints Fig. 4. Responsibility
I
.20
w/w
of BAC Groupings)
rates and BACs of fatally
injured
drivers
(Source:
Terhune
1982)
that only five individuals were included in the category of not-responsible drivers who were involved in crashes during that time period. The same general trend was noted when the group of responsible drivers was expanded to include both single- and multiple-vehicle crashes. In virtually every category, the responsible drivers had higher average BACs than their not-responsible counterparts. Among the responsible drivers. time of day variations were quite common during the weekdays, when average BACs increased dramatically during the late night and early morning hours. There was much less discrepancy in the BACs noted throughout the weekend period. This pattern was observed for responsible drivers in multiple-vehicle collisions, as well as those in the combined sample of single- and multiple-vehicle crashes. For the drivers who were not assigned crash responsibility, weekend crashes more closely resembled the pattern observed for weekday collisions. In other words, the average BACs noted during the evening and early morning hours were noticeably higher than the average BAC observed during the daytime (6 A.M.6 P.M.). These results are depicted graphically in Figs. 5 and 6. There appeared to be no
Table 7. Avrragc
BACs
for drivers
Multiple-vehicle Responsible
killed on weekends
vs. weekdays Total crashes
crashes:
Responsible
Not responsible
( 51
Ave. BAC ,077 ,110
( 47) ( 38)
,053 .076
9) ( 5)
,076 ,064
1’:;‘;
.004 .03 1 ,017 ,000
(32) (18)
,024 ,085 ,092 ,070
Weekend 6AM-6PM hPM-1 1PM
Ave. BAC .070 ,084
(12) (23) (20)
Ave. BAC ,000 ,112
(II) (13)
13AM-6AM IPM-3AM
,089 ,098
(28) ( 8)
Weekday 6AM-6PM 6PM-11PM 1 lPM-3AM 3AM-6AM
,022 ,074 ,124 ,048
(68) I::; (11)
( 9 ( 2)
(11)
K:l
( 66) ( 26)
BAC and crash responsibility
-o-+-
0.16
Responsible Not Responsible
6pm - llpm
6am - 6pm
209
1 lpm - 3am
3am - 6am
TIME Fig. 5. Average BACs by time of day: weekdays.
interaction between time of day and crash responsibility for weekday collisions (Fig. 5). The average BACs for the responsible drivers had the same time of day trend as that observed for the not responsible drivers. With the exception of the 6 P.M.-11 P.M. time period, the same finding was noted for crashes that occurred on weekends (Fig. 6). The weekday average BAC for the responsible and not responsible drivers was .09 vs. .02, respectively. During the weekend, however, the average BACs for the two groups of drivers were equivalent (.08). The small number of individuals in the not-
0.20 1 -oe
0.16 -
Responsible Not Responsible
0.12-
0.08 -
0.04 -
0.00
I 6am - 6pm
I 6pm - llpm
llpm
I - 3am
TIME Fig. 6. Average BACs by time of day: weekends. AAP24:2-H
I 3am - 6am
N. H. MOUNCE and 0. J. PENDLETON
210
responsible analysis.
group of drivers
should
CONCLUSIONS
be kept in mind when interpreting
AND
the results
of this
IMPLICATIONS
There was a very high proportion of individuals in this sample of fatally injured drivers who were determined to be at fault in the crashes in which they were killed. This may be an indication that drivers who are at fault in multiple-vehicle crashes tend to have more severe collisions. Conversely, those drivers who are not at fault in multiplevehicle collisions tend to be involved in less severe crashes, or may exhibit behaviors (e.g safety belt use) that would reduce the likelihood of fatal injury. The fact that surviving drivers were not represented in this data set may explain the high percentage of responsible drivers in the analysis. As previously discussed. results of this study were consistent with results of other studies of fatally injured drivers in that high rates of responsibility were evidenced among even the sober drivers in the sample (McCarrol and Haddon 1962; Baker and Spitz 1970; Perrine et al. 1971; and Wailer 1972). A relatively small difference was found in the proportion of responsible drivers among those who were sober and those who had been drinking. This finding emphasizes the need to identify other factors that, alone or in combination with alcohol, may increase the risk of crash involvement. A brief examination of temporal factors that were felt to interact with alcohol revealed a distinct relationship between crash responsibility and BAC for weekday crashes. Specifically, as BAC increased, the probability of responsibility for the crash increased. and the average BAC was considerably greater among the drivers determined to be at fault in weekday crashes than among those who were not at fault. However, this relationship was not as apparent for weekend accidents. Alcohol use was quite evident in crashes that occurred throughout the designated weekend period, even among the not responsible group of drivers. There are clearly other factors that need to be examined to more fully explain the relationship between alcohol impairment and crash responsibility. It is recommended that future studies include analyses of driver age, gender, driving experience, and other risk-taking behaviors to determine the extent to which these variables, either alone or in combination with alcohol, increase an individual’s likelihood of contributing to a crash.
REFERENCES Baker, S. P.; Spitz. W. U. Age effects and autopsy evidence of disease in fatally injured drivers. JAMA 214:1079-1088; lY70. Borkenstein. R. F.; Crowther, R. F.; Shumate. R. P.; &cl. W. 6.; Zylman. R. The role of the drinking driver in traffic accidents. Bloomington, IN: Indiana University. Department of Police Administration; 1964. Farris, R.; Malone, T. B.; Lilliefors. H. A. Comparison of alcohol involvement in exposed and injured drivers. Alexandria, VA: Essex Corp.; 1976. Hurst, P. M. Estimating the effectiveness of blood alcohol limits. Behav. Res. Highw. Saf. 1(2):87-YY: lY70. Mayhew, D. R.; Donelson. A. C.: Beirness, D. J.; Simpson. H. M. Youth, alcohol and relative risk of crash involvement. Accid. Anal. Prev. 18:273-287; 1986. McCarroll. J. R.: Haddon. W., Jr. A controlled study of fatal automobile accidents in New York City. J. Chronic Dis. 15:811-826: 1962. Mounce, N. H.; Acuff. A.; Peart. S.; Croasdell, D. Alcohol involvement in Texas driver fatalities. College Station, TX: Texas Transportation Institute. Texas A&M University; 1989. Perrine. M. W.; Wailer, J. A.; Harris. L. S. Alcohol and highway safety: Behavioral and medical aspects. Washington, DC: U.S. Department of Trasportation; 1971. Smith. H. W.; Popham, R. E. Blood alcohol levels in relation to driving. Can. Med. Assoc. J. hS:325-328; 1951. Terhune, Kenneth W. An evaluation of crash culpability to assess alcohol and drug impairment effects. 26th Annual Proceedings. AAAM; Arlington Heights, IL: American Association for Automotive Medicine: 1982:329-348. Thorpe, J. D. Calculating relative involvement rates in accidents without determining exposure. Traffic Saf Res Rev 11:3-g; 1967. Wailer, J. A. Factors associated with alcohol and responsibility. Q. J. Stud. Alcohol 33: lhil-170: lY72.
