英文原文
The antilock braking system anomaly: a drinking driver
problem
David W. Harless , George E. Hoffer
Department of Economics, Virginia Commonwealth Uni_ersity, Box 844000, Richmond, VA 23284, USA
Received 30 November 2000; received in revised form 19 February 2001; accepted 28 February 2001
An important anomaly with the incorporation of safety appliances into motor vehicles is the ineffectiveness – and even perverse effects – of antilock braking systems (ABS), particularly as measured by fatal crash involvement. Analyses of insurance claims in the US (Highway Loss Data Institute, 1994, 1995) showed no change in claim frequency after a set of GM vehicles adopted antilock brakes in 1992 despite engineering studies showing antilock brakes resulted in much better braking performance, especially on wet roads. Using the same group of GM vehicles, Farmer et al. (1997) found that adoption of ABS resulted in more 1993–1995 crashes resulting in the death of an ABS vehicle occupant. Another group of vehicles that adopted antilock brakes as early as 1985 were found to have been involved in both more fatal crashes resulting in the death of an ABS vehicle occupant and more involvements overall. These perverse effects were most evident in single-vehicle crashes. Updating the earlier paper with 1996–1998 data on the same two groups of vehicles, Farmer (2001) finds that the ABS-equipped vehicles no longer had a worse fatal accident involvement rate[16].
The ABS anomaly has sparked a large program of research at the National Highway Transportation Safety Administration, the many components of which are described by Garrott and Mazzae (1999). Several explanations have been forwarded for the anomaly. First, consistent with the risk compensation hypothesis, the higher incidence of fatal crashes could be due to increased speed and aggressiveness by drivers of ABS equipped vehicles (Evans, 1995; Evans and Gerrish, 1996). Second, there may be problems with improper operation of ABS-equipped vehicles: the higher incidence of single vehicle crashes could be due to abrupt steering changes by
panicked drivers inexperienced with ABS reacting instinctively in emergency situations (Kahane, 1994). With ordinary brakes, such steering changes have no effect if the brakes are locked. The subsequent attenuation of the ABS anomaly (Hertz et * Corresponding author. Tel.: +1-804-8287190; fax: +1-804-8281719.E-mail address: dwharles@vcu.edu (D.W. Harless).0001-4575/02/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.PII: S0001-4575(01)00030-6 D.W. 334 Harless, G.E. Hoffer / Accident Analysis and Pre_ention 34 (2002) 333–341 al., 1998, 2000; Farmer, 2001) may be due in part to drivers accumulating more experience using antilock brakes.1
In this paper we report strong evidence supporting Kahane’s hypothesis that the ABS anomaly occurs because of improper operation of antilock brakes. We examine the sample of vehicle lines adopting ABS used by Farmer et al. (1997) and Farmer (2001) and apply the same risk ratio analysis to show that the perverse effects of ABS are confined largely to drinking drivers.
We find dramatic differences between drinking and sober drivers in the rate of fatal accidents in ABS equipped vehicles compared to control vehicles. For example, in the sample of GM vehicles that adopted ABS in 1992, consider the rate of fatal accident involvements resulting in the death of an ABS vehicle occupant in the years 1993–1995: In the risk ratio analysis we find that the number of such accidents involving drinking drivers was 82% higher than expected (P=1.6E?11). When there is an ABS problem (e.g., the years 1993–1995 for the GM vehicle sample) there are large differences between the risk ratios for drinking and sober drivers; when there is no ABS problem (e.g., the years 1996–1998) the differences between drinking and sober drivers disappear. We also show that the attenuation of the ABS problem after the first four years of vehicle service, as in Farmer (2001), was also evident in the earlier group of vehicles which adopted ABS as early as 1985.
We follow Farmer et al. in comparing the fatal accident involvement rates of ABS-equipped 1992 GM vehicle lines (the first year that antilock brakes were adopted as standard equipment in these vehicle lines) to otherwise similar 1991 model-year vehicle lines. To parallel Farmer et al. (1997) and Farmer (2001), Table 1 provides risk ratios for both 1993–1995 fatal accidents (left side) and 1996–1998 fatal accidents (right side).2
To demonstrate our (very close) replication of the previous findings, the first line of the table shows our results for all fatal accident involvements resulting in the death
of an ABS vehicle occupant. For example, from the FARS files (National Highway Traffic Safety Administration, 1997) we observed 328 accidents involving the selected ABS-equipped, 1992 model-year GM vehicle lines during the years 1993–l995.3 Under a null hypothesis of no change in such fatal accident involvements upon adoption of ABS, we expect 273.7 such involvements given total vehicle registrations of the 1992 model-year vehicle lines and the involvement rate of 1991 model-year vehicle lines without ABS. The resulting risk-ratio (=observed/expected) equals 1.20. One can present a test of the null hypothesis of no change in the rate of fatal accident involvement (i.e., that the risk ratio is not significantly different from 1.0) either through a 95% confidence interval for the risk ratio or a Chi-squared statistic (=(observed?expected) 2/expected) with one degree of freedom (Breslow and Day, 1987). As in Farmer et al. (1997), the number of fatal accidents resulting in the death of an ABS vehicle occupant is 20% higher than expected during the 1993–1995 period (risk ratio=1.20, X2=10.79, P=0.001). And as in Farmer (2001), the risk ratio is not significantly different from one during the 1996–1998 period (risk ratio=0.95, X2=0.97, P=0.32) [16].
Bifurcation of these fatal accidents by whether the driver of the vehicle was drinking alcohol has a tremendous impact on the risk ratios for accidents occurring from 1993 to 1995. Drinking drivers in ABS-equipped vehicle lines have 82% more involvements resulting in the death of an occupant of the ABS vehicle than expected given vehicle exposures and the number of drinking drivers involved in such fatal accidents in pre-ABS vehicles (risk ratio=1.82, P=l.6E?11). But for sober drivers the risk ratio, 0.99, is neither practically or statistically different from a risk ratio of one. In contrast to the 1993–1995 period, bifurcation by whether the driver was drinking makes little difference during the 1996–1998 period. The risk ratio is 0.92 for drinking drivers and 0.96 for sober drivers, neither significantly different from one. Hence, when the ABS anomaly is observed, the problem appears to be largely confined to drinking drivers; when the ABS anomaly disappears, so too do the dramatic differences in risk ratios between drinking and sober drivers.4
A similar pattern exists in the analysis of all fatal accident involvements. There are 64% more fatal accident involvements than expected for drinking drivers during 1993 through 1995 (P=4.4E?9). During this 3 Farmer et al. report finding four more such accidents during the 1993–1995 period and one more such accident during the 1996–1998 period. These very slight differences should have no impact on the