The effectiveness of antilock braking systems in reducing accidents in Great Britain

The effectiveness of antilock braking systems in reducing accidents in Great Britain

Accident Analysis and Prevention 34 (2002) 347– 355 www.elsevier.com/locate/aap The effectiveness of antilock braking systems in reducing accidents i...

127KB Sizes 0 Downloads 34 Views

Accident Analysis and Prevention 34 (2002) 347– 355 www.elsevier.com/locate/aap

The effectiveness of antilock braking systems in reducing accidents in Great Britain Jeremy Broughton *, Chris Baughan Transport Research Laboratory, Crowthorne, Berkshire RG45 6AU, UK Received 23 August 2000; received in revised form 3 January 2001; accepted 16 January 2001

Abstract Antilock braking systems (ABS) are fitted to many new cars with the aims of improving their ability to steer while braking heavily and of reducing stopping distances on some road surfaces. This paper presents the findings of a project that assessed the effectiveness of ABS in reducing accidents in Great Britain. A large postal survey was carried out of the owners of modern cars, asking for details of any accidents in which they had been involved during the previous year as well as factors that might influence their likelihood of being involved in an accident. Questions were also asked to test respondents’ knowledge of ABS. ABS cars differed in several respects from non-ABS cars, as did the two groups of drivers. Consequently, a sophisticated statistical analysis was required to provide unbiased estimates of the effectiveness of ABS. Various results did not achieve statistical significance, but the overall stability of the results indicates that real effects have been measured. They confirm that ABS does have the potential to reduce the number of accidents, but show that this has not been fully achieved. One reason may be that many drivers have little or no knowledge of ABS. © 2002 TRL. Published by Elsevier Science Ltd. All rights reserved. Keywords: Antilock braking systems; Accident analysis; Accident reduction

1. Introduction Antilock braking systems (ABS) are fitted to many new cars with the aims of improving their ability to steer while braking heavily and of reducing stopping distances on some surfaces. Evidence from track studies indicates that they can “ reduce stopping distances, especially on low adhesion surfaces; “ improve lateral stability and ability to keep within lane during braking on curves; and “ improve steering, stability and average braking decelerations in lane-change and braking manoeuvres on ice. Such evidence encourages the expectation that ABS will reduce accidents. The research began with an extensive review of the technical literature to see whether the expectation had been realised, and this found that * Corresponding author. Tel.: + 44-1344-770879; fax: + 44-1344770356. E-mail address: [email protected] (J. Broughton).

studies of accident data had reached conflicting conclusions about the influence of ABS upon accident risk. Farmer et al. (1997) provide a good summary of the evidence. The literature indicates that ABS leads to an decrease in fatalities among cyclists, pedestrians and others not in the vehicle of up to a fifth, but an increase in vehicle occupant fatalities by perhaps a quarter – largely as a result of more single vehicle and rollover crashes. The net effect of ABS on fatalities appears to be either zero or, for older vehicles, an increase. The evidence is inconclusive for non-fatal accidents – the most optimistic indication being that there is a 10% reduction overall, with larger reductions on wet roads than on dry: some types of single-vehicle accidents, especially rollovers, may increase when ABS is fitted. Much of this evidence comes from the USA and may not be applicable to British conditions because of the different traffic conditions and vehicle types, but it does show that the potential benefits of ABS have not been demonstrated in practice to date. This paper describes research that has recently been carried out to investigate the effectiveness of ABS in reducing accidents in Great Britain.

0001-4575/02/$ - see front matter © 2002 TRL. Published by Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 1 ) 0 0 0 3 2 - X

348

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

A wide variety of factors may influence the risk of a car being involved in an accident, including the age, sex and driving experience of the driver, also the distance travelled and the types of road used. This emphasises the need for a multi-6ariate approach that examines simultaneously the influence of several factors on accident risk, including whether the car was equipped with ABS. If an important detail is lacking from a particular accident data set, there must be concern that its effect might mask the actual effects of ABS. Most previous studies have been essentially uni6ariate, comparing the number of accident-involvements of paired groups of cars, one group being ABS-equipped and the other not. The results of such analyses could mislead if the groups are not matched in other respects, such as mileage and type of drivers (a recent paper by Hertz et al. (1998) has gone some way to address this concern, using a version of the statistical model that will be used below). Furthermore, no car sold in large numbers in Great Britain has followed the pattern found with several US models, and switched directly from ‘ABS not available’ to ‘standard fitting of ABS’. Thus, it would not have been feasible to carry out such a paired comparison of accident risk in this country, even if it had appeared to be desirable. The alternative approach adopted for this project was to survey car owners, enquiring about their driving experience and any accidents in which they had been involved. A postal self-completion survey was designed to collect sufficient information to assess the influence of ABS upon accident risk and the factors that may be preventing ABS from achieving its potential for accident reduction. It was recognised that most of the accidents reported by respondents would be relatively minor, involving no personal injury, so the results would relate principally to ‘damage-only’ accidents. Most previous studies have focused on fatal accidents, i.e. at the other end of the spectrum of accident severity, so the results from this study would not be directly comparable with their findings. Full details of the survey and analyses of the survey data are presented in a Transport Research Laboratory (TRL) Report (Broughton and Baughan, 2000). This paper focuses on the central issues of whether ABS is effective in reducing accidents in Great Britain, and whether there may be factors that limit its effectiveness.

2. The survey of drivers Based on experience of previous studies of this general type, it was judged that a minimum of 1000 accidents would be required to test the influence of ABS; the survey should include both equipped and unequipped cars. The rate of accident-involvements per car per year is low in Great Britain, so a large survey

was required, and questionnaires were sent to 80 000 car owners. A total of 20 973 useable responses were received (a 26% response rate), including details of 1684 accidents during the previous year – only 198 of which led to personal injury. The response rate, although low, was at least as good as in comparable surveys conducted by TRL in the recent past. The response rate for such surveys has undoubtedly dropped since the original accident liability survey of 1987–1988, but the declining willingness to respond to postal surveys appears to have been general, not selective, so should not lead to bias. The questionnaire was designed to encourage a positive reaction from its recipients, and was well received in preliminary tests. Its particular interest in ABS only became apparent in the final sections. It is difficult to see how differences between the characteristics of respondents and non-respondents might have biased the results that relate to the effectiveness of ABS. The survey needed to collect information about relatively new cars, to maximise the number of ABSequipped cars. On the other hand, the cars should have been in use for some time to have been exposed to a significant level of accident risk. The vehicle registration mark (VRM) system in use in Great Britain at that time allowed a car’s ‘year of first registration’ to be readily identified. Any car with a VRM prefix of P was first registered between August 1996 and July 1997. The survey was carried out in March 1999, by which time any P-registered car would have been in use for 19–30 months. It was judged that this group of cars offered a good compromise between ‘newness’ and ‘exposure’. Another advantage of this sampling strategy is that cars would be equipped with modern types of ABS, so the results should not be affected by the performance of older and possibly less effective equipment. Previous studies (Maycock et al., 1991) have demonstrated that the expected number of accident-involvements is influenced by factors such as a driver’s age, sex and experience, as well as the annual mileage and the types of road condition under which the journeys are made. Hence, the questionnaire included a range of questions on these topics. When the car had been owned for less than 12 months, the car-related data were for the period since the car was acquired. Otherwise, respondents were asked about their experiences during the previous 12 months: this limit was applied to minimise problems with respondents’ memories. When analysing the accidents reported by any group of respondents, a crucial item of data is the exposure of that group, defined as the number of car–years covered by their responses. A car that has been owned for more than 1 year contributes 1.0 to the exposure of the group, while a car that has been owned for xB 1.0 years contributes x.

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

349

Table 1 ABS equipment Is car equipped with ABS?

Exposure (%)

Relative accident rate (injury accidents)

Relative accident rate (all accidents)

Yes No Not reported Number of accidents

39.9 43.7 16.4

0.96 0.99 1.13 194

0.95 1.05 1.00 1646

3. Exploratory analyses of the data An initial exploration of the questionnaire responses provided results that guided the full statistical analyses presented later. This section presents some of the principal results, in particular the relative accident rate. The accident rate for a specific group is the number of accidents reported per year of exposure for that group, so the relative accident rate is defined as accident rate for the specific group . overall accident rate Many questionnaires were not completed fully, so the information they provide cannot be used for particular analyses. The minimum requirement is that the period of ownership should have been reported, for otherwise the exposure cannot be calculated. Table 1 includes all questionnaires that meet this requirement, including those that failed to report whether ABS was fitted. It is unfortunate that approximately one-sixth of respondents failed to report whether their cars were fitted with ABS. There is no way of inferring the correct response, so the data that they did supply cannot contribute to the investigation of the effects of ABS. This group was included in a trial analysis as ‘ABS not known’, but this did not provide useful information. If ABS were uniformly distributed across the sample of cars and drivers, there would be little need for more elaborate analyses, but it seems likely that ABS is more frequently fitted to more expensive cars. This is supported by Fig. 1, which shows that the fitting rate increases rapidly with engine capacity until about 2.2 l. Questions were asked about any ABS training that the driver might have received but – not unexpectedly – the proportion of trained drivers was found to be too low for a successful analysis of the effects of ABS training on accident risk. Instead, the modelling will examine whether a driver’s level of knowledge about ABS might be linked to the level of benefit provided by the equipment. The analysis of the data by age and sex of driver yielded relative accident rates that followed the pattern found in other studies, falling until about the age of 70 and then rising. The proportion of exposure accounted for by the youngest and oldest age groups is lower than would be found in a survey of randomly selected driv-

ers, probably because the group of cars used to select the sample is relatively new – and hence expensive. Men are more likely than women to drive ABSequipped cars and 41–70-year olds are more likely than the other age groups to drive ABS-equipped cars. These variations mean that age and sex of driver need to be included in the statistical models. Previous research has shown that a driver’s accident liability increases with annual mileage, but less than proportionately. Men reported driving farther than women, with almost one-fifth driving more than 12 000 miles per year. The percentage of cars with ABS rises with mileage, so mileage also needs to be included. Accident risk has also been shown to fall with the length of time that a driver has been driving, presumably as a result of acquiring experience. This was confirmed by the exploratory analysis, but it has little effect overall as there were relatively few young drivers among the respondents. The questionnaire asked for the colour of the car, and a wide range of colours was reported. These needed to be grouped for analysis – although this inevitably involved some simplification. Exploratory analyses indicated that certain colours are associated with relatively high or low accident rates, so colour was also included in the statistical models. The nature of the influence of colour on accident rate is debatable: it might be physical, linked perhaps to conspicuity, or it might be behavioural (e.g. certain types of driver may tend to choose certain colours). The questionnaire asked for the make and model of the car. The most common car model was found to be

Fig. 1. The percentage of cars equipped with ABS, by engine capacity.

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

350

Table 2 Mean scores for Questions 34 and 35a Male drivers

Question 34 Question 35 a

Female drivers

Cars with ABS

Cars without ABS

Cars with ABS

Cars without ABS

1.66 2.13

1.50 1.89

1.39 1.84

1.28 1.64

Note: maximum score = 4 per question.

the Ford Fiesta, with 5.7% of the exposure (in vehicle– years) and 4.1% (67) accidents; the second most common model was the Ford Escort, with 4.8% of the exposure and 4.7% (77) accidents. The relatively small number of responses for even the most common models meant that it was not feasible to analyse model-specific effects of ABS. However, these two examples demonstrate that the relative accident rate can vary between models: it is 0.71 for the Fiesta and 0.98 for the Escort. The differences may be explained by the characteristics of the drivers or other factors, but it seems desirable to group models according to their raw accident rates in case the differences are genuine. Three groups of models were defined: low-risk medium-risk high-risk

the relative accident rate (all accidents)00.90 the relative accident rate (all accidents)\0.90 and 01.08 the relative accident rate (all accidents)\1.08

The values of 0.90 and 1.08 were selected to yield approximately equal exposure levels in the three groups.

3.1. Knowledge about antilock braking systems Surveys in other countries have demonstrated that a considerable proportion of drivers is unaware of the correct technique for activating ABS during emergency braking (Williams and Wells, 1994; Hans, 1995). Drivers may also be startled by the feedback (e.g. noise, vibration), causing them to remove pressure from the brake pedal in an emergency (Collard and Mortimer, 1998). This project carried out the first systematic research into drivers’ knowledge and use of ABS in Great Britain, one of the aims being to find whether knowledge about ABS influences the accident rates of cars with ABS. Several questions assessed drivers’ knowledge of ABS and a poor level of knowledge was revealed (Broughton and Baughan, 2000). The answers to two questions proved to be correlated with accident rates and were included in the statistical models:

Question 34 – how do you think stopping distances during emergency braking differ between cars with and without ABS, on the following road conditions? (a) Dry roads, (b) wet roads, (c) icy roads, (d) loose surfaces, e.g. snow/gravel. Question 35 – compared with non-ABS fitted cars, how effective do you think ABS fitted cars are at allowing you to steer while braking in an emergency, on the following road conditions? Conditions (a)– (d) from Question 34 were repeated. There is an element of doubt over the correct answer in some cases, but a consensus was reached by consulting experts. A score was calculated by awarding one mark per correct answer – so the maximum score was four per question. The results are summarised in Table 2 for four groups of driver (the analyses are confined to the drivers who reported whether or not their car was fitted with ABS). These results show a limited level of knowledge about the operation of ABS and its effects. Drivers of ABS cars tend to have slightly more knowledge of ABS than do other drivers. A more detailed analysis shows that younger men (up to 55) tend to be more knowledgeable about ABS than older men (over 55) – who tend in turn to be more knowledgeable than women about ABS. There is a poor correlation between the responses to the two questions, so they appear to be testing different aspects of drivers’ knowledge about ABS.

4. The statistical model The influence of ABS on accident risk will be examined using a suitable statistical model, which can take account of other factors that may influence accident risk. Numerous analyses of the survey data were carried out, following a structured approach; parallel analyses were made for ‘all accidents’ and ‘injury accidents’. Each analysis had the same dependent variable: the number of accidents reported. Accidents are rare and independent events, so the error distribution was assumed to be Poisson. In each analysis, a group of independent variables was tested to see how well the

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

variables might – in combination – explain the dependent variable. X, the exposure, has a special role as independent variable in the analysis. Suppose that, in addition to exposure, the particular set of data contains I independent variables x1, x2,…, xI. The model asserts that ACC, the number of accidents per year, may be expressed as ACC(X,x1,x2,…,xI ) = Xexp[i0 +i1x1 + i2x2 +···+ iIxI ].

(1)

The coefficients i0,…,iI are estimated to provide the best representation of ACC on the basis of the independent variables, using the GLIM program (Francis et al., 1993). The use of generalised linear models (GLMs) to relate the incidence of accidents to a range of potential explanatory factors is well established, for example by Maycock et al. (1991). The explanatory factors are discussed below; they include details of the driver and the car mileage, also features of the car such as whether it is equipped with ABS. The estimated coefficient for the presence of ABS will show how much more or less likely an ABS-equipped car is to be involved in an accident than another car which is in all respects identical except that it is not equipped with ABS. One major advantage of this approach is that the question of control groups is dealt with automatically. The control for the ABS-equipped cars in the survey consists of the non-equipped cars in the survey that are otherwise similar. The two groups of car may not be matched in terms of number, driver and trip characteristics, but the necessary allowances are made automatically when the GLM is fitted. The ABS-equipment variable was included at all stages of the development of the model. The principal driver-related factors were added in succession, and the final variables were: sex and age of driver, annual mileage, percentage of total mileage driven on motorways and percentage of total mileage driven on rural roads. Separate series of models for male and female drivers were developed in parallel. Similar ABS effects were found in the three male age groups up to 55-years old, and another effect in the two older male age groups. Consequently, to obtain clearer estimates of the effects of ABS, separate analyses were made for ‘younger’ men aged 17–55 and ‘older’ men aged at least 56. Equally detailed models were fitted for women but there was no such consistent variation with age and they are treated as a single group in the tables. Driving experience was added to the models and had the expected effect, but this achieved only a small improvement in the fit of the model – presumably because of the relatively small number of drivers with limited experience. Consequently, driving experience was not retained during the further development of the

351

model. Two engineering variables were added to the model for ‘all accidents’: the three risk groups for car models as defined above and three colour groups. Only the risk groups could be added to the injury accident model because of the small number of injury accidents. To summarise: the full models consisted of parallel series of models fitted for all drivers (age and sex of driver included among the variables) and for subsets of drivers specified by age and sex. The other variables in each model were “ annual mileage, “ percentage of total mileage driven on motorways, “ percentage of total mileage driven on rural roads, “ ABS equipment, “ accident risk group of car, and “ colour group of car (but not for injury accident models). The resulting models fit the accident data well, and should provide unbiased indications of the effects of ABS on accident risk. Many of the questionnaires were returned incomplete, yet only those that supplied all the requisite data could be used to fit the models. Consequently, the final ‘all accidents’ models used only 67% of the questionnaires and accidents, and the ‘injury accidents’ models used 68%. It seems unlikely that the omission of approximately one-third of questionnaires could affect systematically the statistical results. Some respondents omitted factors such as age, presumably due to carelessness. The question with the poorest response rate was whether or not the car was equipped with ABS: one in six failed to respond. Most respondents had acquired their cars well over a year earlier, so many would have found it difficult to answer this question – without actually carrying out a road test involving heavy braking. There seems to be little cause for concern that those who failed to respond to this question were in other respects distinctive and that the final results might have been different if it had been possible to use the data that they supplied.

5. Results This section presents the principal results from the full models. The modelling produced a wide range of results, but the ‘non-engineering’ results largely duplicate those already available from earlier research. The theme of this paper is the effectiveness of ABS so the presentation focuses on the ABS-related coefficients.

5.1. Antilock braking systems and accident risk Table 3 presents the final results relating to ABS from this first stage of modelling; each estimated effect of ABS is accompanied by its t-value and 90% confi-

352

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

dence interval. The ‘all accidents’ models included the variables discussed previously; the ‘injury accidents’ models excluded colour group. A negative result in Table 3 indicates that an ABS-equipped car has reduced risk compared to an otherwise identical car that lacks ABS, and a positive effect indicates that it has increased risk. Results are presented for younger and older male drivers because the evidence for the effects in the two groups being different is reasonably strong; the age-related differences are much less for females, so only results for all female drivers are presented. The final row presents results from models in which data for male and female drivers were analysed together. Table 3 shows that there is reasonably strong evidence that ABS reduces the risk of an accident of any severity among younger men by about 16%. There is, however, some evidence that ABS might increase the risk among older men by about 10%, and stronger evidence of an increase among women of about 18%. ABS may reduce risk by about 3% overall, although this result does not approach statistical significance. The small number of injury accidents recorded in the survey makes it more difficult to draw conclusions in this case, as demonstrated by the much wider confidence intervals. The results for younger and older men again appear to differ, and this time the size of the increase among older men is sufficient to be statistically significant. There appears to be no effect among women. Table 4 presents the results that relate to the risk groups for car models and colour groups. The basis for the comparison is now the ‘low-risk’ group of models and colour group 1. Thus, cars in the medium-risk group with male drivers have about 73% more accidents than those in the low-risk group, rising to 112% for cars in the high-risk group. The model-group effects are almost as strong in injury accidents as in all accidents. The colour effects are less strong, and the effects for men and women appear to differ.

5.2. The influence of knowledge about antilock braking systems Table 2 showed that men are more knowledgeable than women about ABS, and that younger men are

more knowledgeable than older men. This suggests a possible explanation of the results of Table 3: drivers who are relatively knowledgeable about ABS are benefiting from the technology, whereas those who are less knowledgeable may often use the equipment incorrectly. In order to test whether the estimated effects of ABS are linked to the driver’s level of knowledge about ABS, the score for each question in turn was introduced into the modelling; i.e. the score was added to each of the models summarised at the end of Section 4. Preliminary tests showed that it would be appropriate to assume a linear response, i.e. that the effect of ABS among a particular group of drivers could be expressed as effect of ABS for driver with score s= A+ Bs.

A and B are coefficients to be estimated. A is the average effect for drivers with score= 0 and B is the average change in the effect of ABS as the score increases by 1. In the first set of analyses, drivers of ABS cars were compared with all drivers of non-ABS cars, irrespective of how they responded to the specific question. The results suggested that drivers with more knowledge of ABS derive greater benefit from the equipment (i.e. B was negative). It does not necessarily follow, however, that the extra benefit is the result of the extra knowledge; the two questions could actually be assessing something else about drivers. For example, drivers who are well informed about ABS may also tend to be knowledgeable about and interested in driving safely, and this might be the cause of the lower accident risk. In order to test this possibility, linear models were fitted in parallel for the accident risks of drivers of ABS cars and of non-ABS cars. If the drivers of non-ABS cars have similar accident risks irrespective of their score (i.e. if B is approximately 0) then the benefits associated with higher scores among drivers of ABS cars can legitimately be attributed to greater knowledge about ABS. The estimates of B are shown in Table 5. To be consistent with the previous results, a negative value

Table 3 Estimated effects of ABS on accident risk All accidents

Men aged 17–55 Men aged 56 and above All men All women All drivers

(2)

Injury accidents

Estimate (%)

t

90% confidence interval

Estimate (%)

t

90% confidence interval

−16 10 −7 18 −3

−1.81 0.73 −1.01 1.54 −0.47

(−28%, −1%) (−11%, 36%) (−18%, 5%) (−1%, 40%) (−12%, 7%)

−13 150 26 −5 3

−0.48 2.17 0.98 −0.17 0.18

(−47%, 41%) (25%, 403%) (−15%, 85%) (−1%, 40%) (−23%, 39%)

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

353

Table 4 Estimated variation of accident risk with risk groups of car models and coloura Male drivers

Female drivers

Estimate (%)

t

90% confidence interval

Estimate (%)

t

90% confidence interval

63 137 −11 −1

3.3 6.3 −1.02 −0.09

(27%, 108%) (89%, 197%) (−25%, 7%) (−26%, 31%)

81 115

1.57 2.12

(−3%, 240%) (19%, 290%)

All accidents

‘Medium-risk’ ‘High-risk’ Colour group 2 Colour group 3

73 112 11 25

5.4 7.6 1.33 1.92

(46%, 104%) (80%, 150%) (−3%, 27%) (3%, 52%)

Injury accidents

‘Medium-risk’ ‘High-risk’

21 108

0.62 2.54

(−28%, 103%) (29%, 236%)

a Notes: the colour groups are: (1) white, silver, yellow and green, (2) red, blue and gold, and (3) black, grey, brown and multi-coloured. Colour group was not included in the injury accident model.

indicates that a higher score (denoting greater knowledge about ABS) is associated with greater benefit or reduced disbenefit. The results for Questions 34 and 35 are different. The relatively low number of reported accidents combined with the number of coefficients to be fitted leads to rather imprecise estimates, few of which differ significantly from zero. Nevertheless, the differences between the results for the two questions are reasonably clear. For Question 35 (which asks how effectively ABS allows the driver to steer while braking), introducing the split between drivers of cars with and without ABS improves the fit of the model considerably in each case. The results in Table 5 have the expected pattern, i.e. the effect is slight among drivers of cars without ABS (shown by small values of B) but increases with the score among drivers of cars with ABS (shown by negative values of B – although none achieve statistical significance). By contrast, introducing the split for Question 34 (which asks about the effect of ABS on stopping distances) has no effect upon the fit of the model. The results for this question have a counterintuitive pattern, with the effect of increased knowledge among drivers of cars without ABS appearing to be at least as great as the effect among drivers of cars with ABS. This suggests that Question 35 assesses drivers’ knowledge about ABS rather effectively, and is more discriminating than Question 34. The knowledge tested by Question 34 may be associated with general aspects of braking performance, which is as valuable to drivers of non-ABS cars as to drivers of ABS cars. Alternatively, it may be associated with other factors that reduce a driver’s accident liability. Table 6 interprets the estimates of coefficient A to show the relative risk of drivers scoring 0 for Question 35. Fig. 2 then illustrates the relationship between the score for Question 35 and the relative risk that is implied by these results. Fig. 2 has been prepared from the best estimates of the coefficients and does not

indicate the uncertainty involved. Although the results are not exact, the degree of parallelism of the fitted lines suggests that a real effect has been detected. It appears that younger men tend to benefit from ABS with any score over 0, but among women and particularly older men even knowledgeable drivers tend to be at greater risk in ABS cars. Knowledge that ABS allows the driver to steer while braking does appear to help drivers of ABS cars to reduce their accident liability. Presumably, it increases the likelihood that they will use this characteristic of ABS effectively, while not encouraging unrealistic expectations of what the equipment can achieve. Nonetheless, it still appears that only younger men (up to 55-years old) tend to benefit from the technology. Among the two other groups, only the most knowledgeable of women do not tend to suffer a disbenefit from ABS. Various hypotheses can be advanced to explain these differences, such as “ they respond better to any feedback which shows when the ABS is operating; or “ younger men tend to be physically better able to exert the necessary force on the brake pedal; or “ they are more likely to perceive an emergency in sufficient time to be able to benefit from the performance of ABS. Research into these differences is needed in order to allow all groups of driver to benefit from ABS.

5.3. Supplementary results The questionnaire asked for details of accidents in which the car owner had been involved in the previous year. Perhaps the most interesting responses were to the question about whether the driver had attempted any avoiding action. The results suggest that drivers of ABS cars are more likely to take avoiding action, in particular ‘positive’ action such as acceleration and swerving. The increase in swerving as an avoiding action is to be

354

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

Table 5 Estimates of B for drivers of cars with and without ABS Cars without ABS

Cars with ABS

Estimate (%)

t

90% confidence interval

Estimate (%)

t

90% confidence interval

Question 34 Men aged 17–55 Men aged 56 and above All men All women

−17 −12 −14 −10

−2.91 −1.39 −3.00 −1.57

(−25%, −8%) (−8%, 2%) (−21%, −7%) (−19%, 1%)

−12 −12 −13 −9

−1.37 −1.09 −1.81 −0.84

(−1%, −25%) (−28%, 7%) (−23%, −1%) (−23%, 9%)

Question 35 Men aged 17–55 Men aged 56 and above All men All women

−3 7 1 −1

−0.49 0.90 0.27 −0.23

(−11%, 6%) (−5%, 21%) (−6%, 9%) (−9%, 7%)

−9 −6 −8 −6

−1.22 −1.22 −1.33 −0.63

(−21%, (−21%, (−17%, (−18%,

expected since improved control during braking is one of the recognised benefits of ABS. The increase in acceleration is more difficult to explain, and may be related to drivers’ beliefs about the improved control and stability provided by ABS.

6. Conclusions ABS are fitted to many new cars with the aims of improving their ability to steer while braking heavily and of reducing stopping distances on some surfaces, thereby avoiding accidents or mitigating their severity. Almost one-half of P-registered cars surveyed during this project were equipped with ABS, so conditions are now favourable for assessing the effectiveness of ABS in reducing accidents in Great Britain. A large postal survey of owners of P-registered cars was carried out. Predictably, most of the reported accidents were relatively minor, involving no personal injury, which means that the results relate principally to ‘damage-only’ accidents. Most previous studies of ABS have focused on much more serious accidents, so the results from this study cannot be compared directly with earlier findings. On the other hand, this surveybased approach has allowed the effects of ABS to be related to drivers’ knowledge of the equipment. The two groups of cars and drivers differed in several respects, so a sophisticated statistical approach was required to make an unbiased comparison of the accident rates of the two groups of car. The results for all accidents suggested that driving an ABS car was associated with “ about 16% fewer accidents among men up to 55years old (the 90% confidence interval is from 1 to 28%); “ about 10% more accidents among older men (the 90% confidence interval is from 11% fewer to 36% more accidents);

4%) 12%) 2%) 8%)

“

about 18% more accidents among women (the 90% confidence interval is from 1% fewer to 40% more accidents); and “ about 3% fewer accidents overall (the 90% confidence interval is from 12% fewer to 7% more accidents). The likely reason for the lack of precision of these results is the relatively low number of accidents reported: the 20 973 useable responses provided details of 1684 accidents during the previous year. Only 198 of these accidents involved personal injury, so the results for injury accidents are even less precise; they are broadly consistent with the results for all accidents, although the increase among older men is more marked (and statistically significant). Although the response rate to the survey was only 26%, there seems to be no reason for concern that this might have biased the results. The survey tested drivers’ knowledge about the operation and effectiveness of ABS, so it was possible to see whether the relative levels of knowledge could explain these results. The results suggested that knowledge of the ability conferred by ABS to steer while braking can help drivers of ABS cars to reduce their accident liability. Although few results achieved statistical significance, their overall pattern suggests that the number of Table 6 Relative risk of drivers of ABS-equipped cars with score 0 for Question 35a

Men aged 17–55 Men aged 56– All men All women

Estimate (%)

t

90% confidence interval

1 45 17 32

0.04 1.13 0.77 1.12

(−35%, (−16%, (−17%, (−12%,

56%) 152%) 65%) 99%)

a Note: a positive relative risk indicates greater risk in an ABS car than a non-ABS car.

J. Broughton, C. Baughan / Accident Analysis and Pre6ention 34 (2002) 347–355

Fig. 2. Relative risk as a function of the score for Question 35.

accidents reported by drivers of ABS cars tended to fall as their level of knowledge rose, but there were differences between the three groups of drivers. “ Among men up to 55-years old, drivers of ABS cars who were ignorant about ABS tended to report about the same number of accidents as drivers of non-ABS cars. The reduction in accidents as knowledge improved meant that, on average, drivers in this group who knew something of ABS benefited from the equipment. “ Among older men, drivers of ABS cars tended to report more accidents than drivers of non-ABS cars. Despite the reduction in accidents as knowledge improved, on average even the drivers with most knowledge of ABS still reported more accidents. “ Among women, drivers of ABS cars tended to report more accidents than drivers of non-ABS cars. The reduction in accidents as knowledge improved meant that drivers with most knowledge of ABS reported on average as many accidents as drivers of non-ABS cars. This raises the possibility that there are aspects of ABS as currently implemented that prevent a major part of the driving population from deriving its benefits – even when they are knowledgeable about ABS. It could be that younger men tend to be physically better able to exert the necessary force on the brake pedal, or that they respond better to the feedback provided by ABS, or that they are more likely to perceive an emergency in sufficient time for the system to act effectively. To summarise, the information about drivers’ knowledge of ABS has shown that the incomplete fulfilment of the potential of ABS may partly be explained by the fact that many drivers have little or no knowledge of ABS: indeed, ignorance of ABS tends to be associated

355

with increased accident risk in cars equipped with ABS. It appears that the way in which many women and older men use the equipment may tend to increase their risk of accidents. These findings may have broader implications. ABS is amongst the first of the high-technology in-car devices that have been or are being developed to assist drivers. It does not require any special action by the driver, so the significance of specific training has not previously been evident. The experience of ABS suggests that when cars fitted with these new devices are sold in future, it will be important to train drivers in their use, rather than simply rely on the traditional methods: ‘word of mouth’ and manufacturers’ handbooks.

Acknowledgements The work described in this paper was carried out in the Safety Department of the Transport Research Laboratory, under a contract from the UK Department of the Environment, Transport and the Regions. The contribution of Gavin Buckle, Nichola Christie, Liz Lowe and Lee Smith in conducting the survey is gratefully acknowledged.

References Broughton, J., Baughan, C.J., 2000. A survey of the effectiveness of ABS in reducing accidents. TRL Report 453: Transport Research Laboratory, Crowthorne. Collard, D.A., Mortimer, N.L., 1998. A survey of Canadian drivers’ knowledge about and experience with anti-lock brakes. (98-S2-O09) Road Safety and Motor Vehicle Regulation Directorate, Canada. Farmer, C.M., Lund, A.K., Trempel, R.E., Braver, E.R., 1997. Fatal crashes of passenger vehicles before and after adding antilock braking systems. Accident Analysis and Prevention 29, 745 – 757. Francis, B., Green, M., Payne, C. (Eds.), 1993. The GLIM System, Release 4 Manual. Clarendon Press, Oxford. Hertz, E., Hilton, J., Johnson, D.M., 1998. Analysis of the crash experience of vehicles equipped with antilock braking systems (ABS) – an update. NHTSA paper 98-S2-O-07. Paper presented at the 16th ESV Conference, February 1998 in Windsor, Canada. Hans, M., 1995. Drivers need to learn the ABCs of ABS. Traffic Safety, November/December 1995. Maycock, G., Lockwood, C.R., Lester, J.F., 1991. The accident liability of car drivers. TRL Research Report 315: Transport Research Laboratory, Crowthorne. Williams, A., Wells, J., 1994. Driver experience with antilock brake systems. Accident Analysis and Prevention 26, 807 – 811.