You are on page 1of 7

Accident Analysis and Prevention 40 (2008) 167–173

Effects of electronic stability control (ESC) on accidents:


A review of empirical evidence
Alena Erke ∗
Institute of Transport Economics, Department of passenger transport, Grensesvingen 7, 0602 Oslo, Norway
Received 7 October 2006; received in revised form 22 March 2007; accepted 5 May 2007

Abstract
This study summarizes evidence from empirical studies on the effects of electronic stability control (ESC) on accidents in a meta-analysis.
Large reductions of single vehicle accidents have been found (−49%; 95% confidence interval: [−55%; −42%]), and smaller but still significant
reductions of head-on collisions (−13%; 95% confidence interval: [−17%; −8%]). Multi-vehicle fatal accidents are also significantly reduced
(−32%; 95% confidence interval: [−43%; −20%]). The effects can be explained with the potential of ESC to improve driving dynamics and
to reduce the probability of loss of control. However, there are significant amounts of heterogeneity in the results, especially for single vehicle
accidents, and a sensitivity analysis shows that the results for single vehicle accidents are likely to be affected by publication bias. The results for
single vehicle accidents are in excess of what might be expected based on studies that have estimated the total amount of accidents that may be
affected by ESC. Consequently, the proportions of accidents that can be avoided by ESC is assumed to be somewhat smaller than suggested by
most empirical studies. Properties of the vehicles, time trends, and driver behaviour may have contributed to the large empirical effects.
© 2007 Elsevier Ltd. All rights reserved.

Keywords: Electronic stability control; ESC; Accident; Meta-analysis; Single vehicle accident; Head-on collision

1. Introduction where the vehicle does not wholly follow the drivers steering
input (understeering). Both may be caused by too high speed in
The objective of this study is to summarize empirical evi- curves, collision avoidance manoeuvres, low friction conditions,
dence of the effects of electronic stability control (ESC) on or combinations of these (Sferco et al., 2001). ESC systems dif-
accidents by reviewing studies that have estimated proportions fer with respect to how they regulate driving parameters (yaw
of different types of accidents that may be affected by ESC, and characteristics, and sideslip), how they counteract deviations
by calculating summary effects from studies that have inves- (e.g. by braking individual wheels and reducing engine power),
tigated empirical effects of ESC on accidents by means of and in what way drivers take notice of the activities of the
meta-analysis. The results are compared in order to assess the systems (Alliance of Automobile Manufacturers, 2005). ESC
size of the effects and to provide an indication of the presence cannot overrun physics, and may not always prevent the vehi-
of factors that may affect the size of the effects. cle from sliding or spinning (Insurance Institute for Highway
Electronic stability control is an active safety device for motor Safety, 2005).
vehicles which aims at improving driving dynamics and at pre- ESC was first introduced as optional safety equipment in
venting accidents which result from loss of control. There exists passenger cars on the European market in 1995. Its public-
a large variety of ESC systems. They have in common that ity increased in 1997 after an event with a rollover of a new
they enhance the controllability of vehicles and that they can vehicle model in a double lane change manoeuvre on a driv-
prevent skidding and loss of control in cases of oversteering ing course (“moose-test”). The proportion of new cars that are
or understeering. These are movements of the vehicle which equipped with ESC increased steadily. In 1995, the proportion
go beyond the steering input by the driver (oversteering) or of new cars equipped with ESC was only 5% (in Germany);
in 2004, the proportion was 36% in Europe, and much higher
in some European countries, e.g. 67% in Germany (Deutscher
∗ Tel.: +47 22573863. Verkehrssicherheitsrat, 2006), and 70% in Sweden (according
E-mail address: Alena.Erke@toi.no. to lists published by the Swedish insurance company Folksam).

0001-4575/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.aap.2007.05.002
168 A. Erke / Accident Analysis and Prevention 40 (2008) 167–173

The proportion of all vehicles that are equipped with ESC in sion with oncoming traffic is replaced by road departure, and
Europe has been estimated by the European Commission (2005) when impact speed is reduced. Collisions in junctions, over-
to be 9% in 2005, and is predicted to reach 50% by 2025 (if no taking accidents, and rear-end collisions are less likely to be
legal requirements for ESC are introduced). prevented by ESC. Accidents which involve loss of control, but
In recent years, the increase in the number of vehicles for in which the driver was driving extremely fast, very sleepy, or
which ESC is available seems to have slowed down, but the driving under the influence of alcohol or drugs, are also less
proportion of vehicles for which ESC is standard equipment likely to be affected by ESC.Proportions of different types of
(instead of optional equipment) is still increasing according to accidents that might be affected by ESC have been estimated in
lists published by the US-government (www.safercar.dot.gov). several studies based on accident reports and in-depth accident
The proportion of new sport utility vehicles (SUVs) with analyses. These studies are summarized in Table 1 in order to
ESC is increasing faster than the proportion of cars (Farmer, provide a basis of comparison for results from empirical acci-
2006). In 2004, nearly 70% of all new SUVs on the US- dent studies that are presented in the next section. The summary
market have been equipped with ESC (Motor and Equipment of the results in Table 1 gives a rough estimate of the magnitude
Manufacturers Association, 2006). The proportion of smaller of the potential effects. Accidents are usually not classified in
cars that are equipped with ESC is much lower, and ESC official accident statistics according to whether or not skidding
is mostly only optional, not standard equipment (Deutscher or loss of control has been a contributing factor. The study of
Verkehrssicherheitsrat, 2006). The costs for the installation of Campbell et al. (2003) is based on a total of ca. 73,000 acci-
ESC depend on the standard of the technical equipment of a dents from two US databases. Accidents of all severities are
car (e.g. antilock brakes, antislip regulation, brake assistant). included in their analysis. In about half of all single vehicle off-
Estimations of costs for the installation of ESC as standard road accidents, skidding has been a major contributing factor,
equipment vary therefore widely, between ca. $100 and $800 per and in about half of these accidents speeding (driving above
vehicle (Insurance Institute for Highway Safety, 2006; NHTSA, the speed limit) has been an additional contributing factor. The
2006), or between D 150 and D 500 per vehicle (European study of the Insurance Institute for Highway Safety (2005) has
Commission, 2005). ESC is not mandatory in any European estimated the proportions of different types of accidents that
country, but recommended by the European new car assessment may be prevented by ESC. Langwieder et al. (2003) have ana-
programme (EuroNCAP). In the US there are plans to make ESC lyzed reports of ca. 1500 injury accidents in Germany in order
mandatory from 2011. to estimate the proportions of accidents which involved skid-
ding. Sferco et al. (2001) have investigated ca. 2700 accidents
2. Potential effects of ESC on accidents in the European accident causation survey (EACS, based on data
from five European countries). They have estimated quite large
ESC aims at reducing the probability of loss of control. Loss amounts of accidents which involve loss of control and which
of control often results in accidents with severe consequences may be affected by ESC. The proportion of accidents which
(Langwieder et al., 2003; Unselt et al., 2004): road departure, actually may be avoided by ESC is assumed to be much smaller.
collisions with objects or vehicles, or turnover. These collisions The proportion of accidents which would definitely have been
are relatively likely to be side collisions, and/or collisions with avoided is estimated to be 2% for injury accidents, and 3% for
oncoming traffic. Side collisions usually have severe conse- fatal accidents. The proportion of accidents which would proba-
quences because vehicles provide only little protection. Head-on bly have been avoided is estimated to be 9% for injury accidents,
collisions often involve high (relative) speed. Accidents may not and 18% for fatal accidents. Unselt et al. (2004) have analysed
always be avoided, but the severity may be reduced, e.g. when data from a representative sample of over two million accidents
a side collision is replaced by a front collision, when a colli- in Germany in 2002 and estimated how many accidents involved

Table 1
Proportions of accidents in which contributing factors may be affected by ESC
Proportion of accidents involving contributing factor (%)

Study Accident contributing factor Single vehicle Multi-vehicle All

Campbell et al. (2003) Skidding 50 (single vehicle off-road) 1 (rear-end)


8 (lane change)
Insurance Institute for ESC-relevant 56 17 34
Highway Safety (2005)
Langwieder et al. (2003) Skidding 39 12
Sferco et al. (2001) Loss of control, ESC-relevant 42 (injury accident)
67 (fatal accident)
Unselt et al. (2004) Loss of control 21 (injury accident)
43 (fatal accident)
Zobel et al. (2000) Skidding 44 (very severe injury accident)
Summary ESC-relevant 40–50 10 20–40
A. Erke / Accident Analysis and Prevention 40 (2008) 167–173 169

loss of control. Zobel et al. (2000) have analyzed 10,000 acci- dard equipment in these cars. In two of these studies (Aga and
dents in the database of the Hannover Medical School (MHH). Okada, 2003; Farmer, 2006) involvement in different types of
They have estimated the proportions of accidents with at least accidents has been compared between vehicles with and without
one very severely injured (MAIS 5+) that involved skidding. ESC. Bahouth (2005) has additionally included involvement in
According to Zobel et al. (2000), the probability of at least one accidents that are not assumed to be affected by ESC as a con-
very severely injured (MAIS 5+) in accidents, which without trol. Case-control studies have compared accident involvement
ESC would be side collisions, would be reduced by 50% if all between different types of vehicles with and without ESC. All
vehicles were equipped with ESC. The proportion is larger in of these studies have included involvement in accidents that are
curves (53%) that in straight sections (40%). not assumed to be affected by ESC as a control. The types of
The results show quite consistently large proportions of sin- accidents that have been used as controls are shown in Table 2.
gle vehicle accidents which involve contributing factors that may There are large differences between the studies as regards the
be affected by ESC, and smaller proportions for multi-vehicle types of accidents that are assumed not to be affected by ESC.
accidents. The proportions are consistently larger for fatal acci- The range is from all multi-vehicle accidents (Dang, 2004) to
dents than for injury accidents. These results do not take into selected types of accidents in the other case-control studies
account potential effects on driver behaviour. and in the study of Bahouth (2005). The accidents that are not
assumed to be sensitive to ESC in the study of Kreiss et al. (2006)
3. Meta-analysis are mainly accidents involving turning or crossing roads and
accidents involving stationary vehicles. Page and Cuny (2006)
3.1. Studies included in the meta-analysis include all accidents not involving loss of control or guidance
problems in the group of non-ESC sensitive accidents.
The proportions of different types of accidents that are pre-
vented by ESC are examined in a meta-analysis. It includes 3.2. Statistical analysis in meta-analysis
studies of accident involvement of vehicles with and without
ESC. In order to identify studies, a computer search was made of The results of the studies listed in Table 2 are combined by
the Silverplatter TRANSPORT literature database which com- means of the log-odds method of meta-analysis (Christensen,
bines different international transport related databases. Studies 2003; Elvik, 2005; Fleiss, 1981). Estimates of effect were com-
were also searched on the Internet. Studies were retrieved when puted as odds ratios, based on the numbers of accidents reported
in the studies (Formula (1)).
Study acc. with ESC/Study acc. without ESC
Estimate of effect = (1)
Control acc. with ESC/Control acc. without ESC

they were, based on title and abstract, likely to be empirical stud- Study accidents are accidents for which the effect of ESC is
ies and to have investigated effects on accidents. Studies were estimated, e.g. single vehicle accidents. In most studies, types
included in the meta-analysis if they gave sufficient information of accidents that are not assumed to be affected by ESC were
to compute estimates of effect, based on numbers of accidents. used as control.
Studies that have estimated how likely certain types of accidents The study of Aga and Okada (2003) has used the number of
are to be avoided or affected by ESC were not included in the cars as a denominator. The study of Farmer (2006) has used the
meta-analysis. accident involvement of different types of vehicles as a control
Eight studies were included in the meta-analysis. They group.
are listed alphabetically in Table 2. Three of the studies are For combining estimates of effects from different studies,
before-and-after studies, five of them are case-control studies. each estimate is assigned a statistical weight and the weighted
Before-and-after studies have investigated accident involvement mean of the estimates is computed. The statistical weight
of one or more types of cars before and after ESC became stan- assigned to each estimate of the effect is inversely proportional to

Table 2
Evaluation studies included in the meta-analysis
Authors (year) Country Number of estimates Statistical weighta Study design Control

Aga and Okada (2003) Japan 4 139.1 Before–after Numbers of cars


Bahouth (2005) USA 14 1910.2 Before–after ESP vehicle hit from rear
Dang (2004) USA 4 589.9 Case-control Multi-vehicle collisions
Farmer (2006) USA 6 7985.2 Before–after Non-ESC vehicles
Kreiss et al. (2006) Germany 1 1318.3 Case-control Accidents not sensitive to ESC
Lie et al. (2004) Sweden 1 50.3 Case-control Rear-end collisions on dry roads
Page and Cuny (2006) France 1 14.6 Case-control Accidents not sensitive to ESC
Thomas (2006) UK 7 7378.5 Case-control Collisions not involving forward
movement of the vehicle
a Sum of statistical weights in fixed effects models of meta-analysis (only effects that are included in the meta-analysis).
170 A. Erke / Accident Analysis and Prevention 40 (2008) 167–173

the variance. How the statistical weights are computed depends low heterogeneity (Christensen, 2003). It is hypothesized that
on the amount of heterogeneity in the estimates of effect. If the effects of ESC are different for different types of accidents
there is only a small amount of heterogeneity, a fixed effects (single vehicle, multi-vehicle), and for different accident sever-
(FE) model of meta-analysis can be applied, in which variance ities (injury accidents, fatal accidents). Therefore, the test for
is assumed to arise from random variation of the estimates of heterogeneity is performed within each group of different acci-
effect around one “fixed” mean effect. In the FE model, the dents and accident severities where the number of estimates of
statistical weights are computed according to Formula (2). effect is larger than one. The results of the heterogeneity test are
summarized in Table 3 for all groups of effect sizes (types of
1
FE weight =  (2) accidents, severities) where more than one effect size is avail-
( 1/Number of accidents) able. There are significant amounts of heterogeneity within the
If there is heterogeneity in the estimates of effect, a ran- effect sizes for single vehicle accidents. For head-on collisions
dom effects (RE) model of meta-analysis is applied. In the RE and rollover accidents, the results are not very far from being
model, the variance of the estimates of effect is assumed to arise significant.
from two sources. One source of variance is random variation Summary estimates are computed with the RE model for all
around the mean effect (as in the FE model). The mean effect groups of accidents where more than one effect size is avail-
sizes are additionally assumed to vary between studies. When able. Each summary estimate is computed by summing up all
there is heterogeneity, RE models yield slightly different esti- products of the logarithms of the estimates of effects and the
mates of effect, and larger confidence intervals than FE models. corresponding RE weights, and dividing this sum by the sum
Statistical weights in the RE model are computed according to of all RE weights. The result is then rescaled from the loga-
Christensen (2003) as a function of the logarithm of the effects rithmic scale in order to obtain the combined estimate of effect
and weights in a FE model, and Cochran’s Q statistic (Formula in terms of odds ratios. This estimate indicates the proportion
(3)). The Cochran’s Q statistic is used to test heterogeneity. It of accidents that are not prevented by ESC. If the estimate is
is a function of the weighted squared differences between study for example 0.87, the reduction in the number of accidents in
effect estimates and the fixed effects summary effect. Under the vehicles with ESC is 13% ([1–0.87] * 100% = 13%).
assumption of no heterogeneity, it follows a Chi-square distribu-
tion where the number of the degrees of freedom is the number 3.3. Results of meta-analysis
of studies minus one.
1 The estimates of effect for different types of accidents and
RE weight = (3) different levels of accident severity are summarized in Table 4.
(1/FE weight + (Q − df)/C)
Additionally, 95% confidence intervals are shown.
where Q refers to the Cochran’s Q statistic, df is the degrees of Large reductions were found for single vehicle, loss of con-
freedom of Q, and C is computed as shown in Formula (4). trol and rollover accidents. The number of head-on collisions
  was also reduced significantly. The effects on fatal accidents are
 Ln(FE effect) ∗ FE weight)2
C= FE weight −  larger than the effects on injury accidents for head-on and for
FE weight all multi-vehicle accidents. For single vehicle accidents there is
(4) no significant difference.
A combined effect for all types of accidents was not com-
In order to examine if a RE model is more adequate, it is puted. Most studies have focused on specific types of accidents;
tested if there is a significant amount of heterogeneity applying the composition of types of accidents would therefore not be
Cochran’s Q-test described in Christensen (2003). If Q is sta- representative in a combined effect.
tistically significant, a RE model should be applied. This test is Three of the studies have investigated effects of ESC under
quite weak (i.e. the probability of detecting heterogeneity when different driving conditions (Farmer, 2006; Kreiss et al., 2006;
there is heterogeneity is not very high), and the RE will not Lie et al., 2004). In all studies, the largest effects have been
be inappropriate, but approaching the FE model when there is found on icy or snowy roads, and the smallest effects on dry

Table 3
Test for heterogeneity (Cochran’s Q-test)
Accident severity Type of accident df Cochran’s Q p

Injury/unspecified accidents Single vehicle 13 82.939 <.000


Injury/unspecified accidents Loss of control 2 57.621 <.000
Injury Rollover accidents 2 3.892 .143
Injury/unspecified accidents Head on 8 11.530 .183
Injury/unspecified accidents Multi-vehicle 3 0.721 .868
Injury accidents Single vehicle 1 8.575 .003
Fatal accidents Single vehicle 4 8.379 .079
Injury accidents Multi-vehicle 1 0.050 .823
Fatal accidents Multi-vehicle 1 0.786 .375
A. Erke / Accident Analysis and Prevention 40 (2008) 167–173 171

Table 4
Estimated reductions of accidents for vehicles equipped with ESC (RE models)
Change of number of accidents (%)

Accident severity Type of accidents Best estimate 95% confidence interval

Injury/unspecified accidents Single vehicle −49 (−55; −42)


Injury/unspecified accidents Loss of Control −41 (−62; −7)
Injury accidents Rollover −69 (−82; −45)
Injury accidents All except for rear-end collisionsa −22 (−41; +3)
Injury/unspecified accidents Head on −13 (−17; −8)
Injury/unspecified accidents Side collisionsa −7 (−14; 0)
Injury/unspecified accidents Multi-vehiclea +3 (+1; +5)
Injury accidents Single vehicle −46 (−64; −18)
Fatal accidents Single vehicle −49 (−62; −33)
Injury accidents Head ona −10 (−15; −5)
Fatal accidents Head ona −79 (−97; +61)
Injury accidents Multi-vehicle +2 (+1; +3)
Fatal accidents Multi-vehicle −32 (−43; −20)
a Fixed effects model (too few effect sizes for random effects).

roads. Meta-effects could not be computed because all studies are shown as vertical lines. For other types of accidents, there
have investigated different types of accidents. were too few estimates of effect to create a funnel diagram.
In accordance with the results of the test for heterogeneity,
3.4. Sensitivity analysis there is large variance in the estimates of effects for single vehi-
cle accidents in Fig. 1, and the effects do not seem to cluster
Sensitivity analyses are conducted to test the robustness of the around one common value. The distribution is somewhat asym-
results. Sensitivity analysis focuses on publication bias, outlier metrical. Large negative effect sizes (large accident reductions)
bias, effects of statistical weighting, and study design.

3.4.1. Publication bias


Publication bias exists when studies that do not find signif-
icant effects in the expected direction are not published. When
there is publication bias, the largest effects are found in small
studies, because effects have to be larger in smaller studies in
order to become significant. When there is no publication bias,
studies which do not yield significant results in the expected
direction are also published, and the effects will be uncorre-
lated with study size. More extreme values will be still more
frequent in smaller studies. A simple method to test for publi-
cation bias is to investigate the relationship between the effect
sizes and the weights of the corresponding studies in funnel
plots (Christensen, 2003). A funnel plot is a scatter diagram
Fig. 1. Funnel plot for single vehicle accidents (all severities).
with the logarithm of the effect sizes (Ln(effect)) on the x-axis,
and the statistical weights on the y-axis. Statistical weights are
computed with the FE model. Logarithms of effect sizes are
used instead of effect sizes because the weights refer to the log-
arithm of the effects. A “funnel” is drawn around the average
Ln(effect). Because FE statistical weights are displayed on the
y-axis, Ln(effects) are also computed with the FE model, and
therefore slightly different from the logarithms of the effect size
shown in Table 4. Without publication bias, the Ln(effects) are
expected to be assembled symmetrically in the shape of a fun-
nel around the average, where the largest studies are grouped
around the average effect. Funnel plots are shown in Fig. 1 for
single vehicle accidents and in Fig. 2 for head-on collisions.
Fatal accidents are shown as circles in both figures. The sum-
mary estimates in the FE models for injury/unspecified accidents Fig. 2. Funnel plot for head-on collisions (all severities).
172 A. Erke / Accident Analysis and Prevention 40 (2008) 167–173

have been found in smaller studies. This is to be expected when tion bias. The effects on fatal single vehicle accidents are also
there is publication bias. A possible alternative explanation is influenced by outlier bias. Actually, each of the four estimates
that studies in which the accident severity is unspecified include might be regarded as an outlier. Additionally, there may be an
property damage accidents as well as injury accidents. For prop- effect of study design. These results are however inconclusive.
erty damage accidents no large effects are expected, and they are These results suggest that the effect on single vehicle accidents
more frequent (produce larger statistical weights) than injury is likely to be smaller than suggested by most accident studies.
accidents or fatal accidents. For fatal single accidents there are No conclusions at all can be drawn from the results for fatal sin-
only four effect sizes, all of which are far away from the sum- gle vehicle accidents. It is also likely that there are factors that
mary estimate. The effects of ESC on head-on collisions are affect the effects of ESC that can not be investigated, based on
quite symmetrically distributed around the summary estimate. the present studies.
They do not seem to be affected by publication bias. For fatal The effects of ESC on head-on collisions are more consistent.
head-on collisions there is only one effect size. They show no evidence of publication bias, outlier bias, or
influence of statistical weighting. A possible explanation is that
3.4.2. Outlier bias most studies focus mainly on single vehicle accidents, and that
The effects of single estimates on the summary estimates effects on head-on collisions are less susceptible to publication
(outlier bias) are investigated by omitting each of the single bias.
estimates from the calculation of the summary estimates, one at
a time. Most of these new summary scores cluster closely around 4. Summary and conclusions
the original summary scores, i.e. within the 95% confidence
interval and not more than 2% above or below. Single vehicle A review of studies that have estimated potential effects of
accidents are an exception. The new summary scores for injury ESC on accidents has been conducted and results of empirical
accidents range from −18 to −46%. The new summary scores studies of the effects of ESC on accidents have been summa-
for fatal single vehicle accidents range from −43 to −52%. rized by means of meta-analysis. The results show a considerable
potential of ESC to reduce the number of accidents, especially
3.4.3. Influence of statistical weighting single vehicle and rollover accidents, head-on collisions, and
In order to test the influence of the statistical weighting of multi-vehicle fatal accidents.
the estimates of effect, all summary effects were compared to The meta-analysis shows large effects on single vehicle acci-
the arithmetic means of the single effect sizes (for all summary dents, and a still larger effect on rollover accidents. These effects
effects which are based on more than one individual effect size). can be explained in terms of the large proportions of these types
No significant differences between the RE summary effects and of accidents that involve loss of control. Among head-on col-
the arithmetic means have been found for any of the accident lisions, the proportion that involves loss of control is lower.
types in the meta-analysis. Accordingly, the reduction of head-on collisions is smaller than
the reduction of single vehicle accidents. For all types of multi-
3.4.4. Effects of study design vehicle accidents, a significant decrease has been found for
Effects of study design on the estimates of effect are exam- fatal accidents only. This is not surprising because the pro-
ined by comparing results between before-and-after studies and portion of multi-vehicle accidents that involve loss of control
case-control studies. In before-and-after studies the types of cars and that might be affected by ESC is mainly limited to head-
with and without ESC are identical, but ESC is not the only on collisions, and these represent only a small proportion of
change that has been made to the vehicles as ESC became stan- all multi-vehicle accidents. The larger reductions of fatal head-
dard equipment. In case-control studies there are different types on and multi-vehicle accidents can be attributed to the severe
of cars with and without ESC. The vehicles have been matched consequences that accidents involving loss of control usually
with respect to properties that are relevant for active and passive have.
safety, which are not controlled for in before-and-after studies. So far, the meta-analysis confirms the results of the studies
Effects of time trends can be assumed to be present in before- that have estimated potential accident reductions. However, a
and-after studies, but not in case-control studies. Vehicle age has closer look at the results casts some doubt on the magnitude
been controlled for in most studies. A comparison can be made of the effects on single vehicle accidents. There is large varia-
for single vehicle accidents. The estimates of effect for injury tion in the results, the results change significantly when single
and unspecified accidents (RE model) are very similar for the estimates are omitted from the analysis, and they are likely to
two types of study design (−46% in the before-and-after studies, be affected by publication bias. In contrast to the results of
−41% in the case-control studies). For fatal single vehicle acci- the studies that have estimated potential effects of ESC and
dents the effect is −54% in before-and-after studies and −40% in contrast to the results of the meta-analysis for head-on and
in the case-control studies. This might indicate an effect of study all multi-vehicle collisions, the effect on single vehicle injury
design. However, none of the differences is significant. accidents is nearly as large as the effect on fatal single vehicle
accidents. Additionally, the reductions of the number of single
3.4.5. Summary of the sensitivity analysis vehicle accidents that have been found in the meta-analysis are
The sensitivity analysis shows that the effects of ESC on almost the same as the estimated proportions of accidents that
single vehicle accidents are likely to be influenced by publica- might be affected by ESC. It seems unrealistic to assume that all
A. Erke / Accident Analysis and Prevention 40 (2008) 167–173 173

accidents which theoretically might be affected actually will be of ESC (e.g. by benefit-cost analysis), and as ESC is becoming
prevented. more common, its effects should be taken into account in studies
Study design (before-and-after versus case-control) may have of other safety measures which aim at reducing the same types
affected the estimated effects. Case-control studies yield smaller of accidents.
effects than before-and-after studies, where more potential con-
founding factors are present. However, the difference is not References
significant. Other methodological aspects may have contributed
to the relatively large effects. These include the types of con- Alliance of Automobile Manufacturers, 2005. Preliminary results of ESC per-
formance metric evaluations. Presentation from the meeting of the Alliance
trol accidents in case-control studies, the (lack of) control for of Automobile Manufacturers, June 23, 2005.
properties of the vehicles in the study or for time trends. The Aga, M., Okada, A., 2003. Analysis of vehicle stability control\s effectiveness
studies of Dang (2004) and Farmer (2006) have found much from accident data. ESV paper 541.18th ESV Conference. Nagoya.
larger effects on accidents with SUV than with passenger cars. Bahouth, G., 2005. Real world crash evaluation of vehicle stability control (VSC)
Differences may also be found between other types of vehicles, technology, 49th Annual Proceedings Association for the Advancement of
Automotive Medicine, September 12–14.
depending on weight, centre of gravity, and driving dynamics. Campbell, B.N., Smith, J.D., Najm, W.G., 2003. Examining of crash contributing
As ESC is mostly standard equipment in higher-class cars, the factors using national crash databases. NHTSA report DOT-VNTSC-
results may not be transferable to lighter vehicles. Time trends NHTSA-02–07.
which may have contributed to overestimated effects may arise Christensen, P., 2003. Topics in meta-analysis. Report 692/2003. Oslo: Institute
from improvements of vehicles (e.g. side-airbags) or infrastruc- of Transport Economics.
Dang, J.N., 2004. Preliminary results analysing the effectiveness of electronic
ture (e.g. improvements of road curvature and speed reductions stability control systems. NHTSA, DOT HS 809 790.
in curves) that aim at reducing the same type of accidents or Deutscher Verkehrssicherheitsrat, 2006. Sicherheitstechnik ESP: Kleinwagen
accident consequences as ESC. Due to the small number of stud- dem Schleudern weiterhin ausgesetzt (Safety equipment ESP: Small vehicles
ies, and partly due to lack of information in the studies, these still prone to skidding). DVR Report 728/30.
aspects could not be tested systematically. Three of the studies Elvik, R., 2005. Introductory guide to systematic reviews and meta-analysis.
Transportation Research Record 1908, 230–235.
in the meta-analysis have found different effects on dry, wet and European Commission, 2005. Cost-benefit assessment and prioritisation of vehi-
icy roads. cle safety technologies. European Commission Directorate General Energy
The results may additionally be affected by differences and Transport, Interim report October 2005.
between drivers buying cars with and without ESC. Drivers buy- Farmer, C.M., 2006. Effects of electronic stability control: An update. Traffic
ing cars with ESC may set a higher value on safety, or they may Injury Prevention 7, 319–324.
Fleiss, J.L., 1981. Statistical methods for rates and proportions. Wiley, New
choose an ESC-car in order to be able to drive faster. ESC may York.
also affect driver behaviour, depending on whether or not the Insurance Institute for Highway Safety, 2005. Electronic Stability Control. Status
driver knows that the vehicle is equipped with ESC and how Report, 4(1), 2–4.
ESC works, whether the driver notices when ESC is getting Insurance Institute for Highway Safety, 2006. Electronic Stability Control. Status
active, and how the activity of ESC is perceived (e.g. as “danger” Report 41(5), 2–3.
Kreiss, J.-P., Schüler, L., Langwieder, K., 2006. The effectiveness of primary
versus “fun”). Effects on driver behaviour may reduce or neu- safety features in passenger cars in Germany. ESV-paper No. 05-0145. 19th-
tralize the positive effect of ESC on driving dynamics if drivers ESV-Conference, Washington D.C. (USA).
have too much trust in the system and drive faster, or if they Langwieder, K., Gwehenberger, J., Hummel, T., Bende, J., 2003. Benefit poten-
neglect shifting out worn tyres. On the other hand, ESC may be tial of ESP in real accident situations involving cars and trucks, 16th
informative, warning drivers about slippery driving conditions, International Technical Conference on the Enhanced Safety of Vehicles,
Nagoya (Japan).
and thus contributing to safer driver behaviour. These effects Lie, A., Tingvall, C., Krafft, M., Kullgren, A., 2004. The effectiveness of ESP
may be different for different drivers. In summary, drivers of in reducing real life accidents. Traffic Injury Prevention 5, 37–41.
cars that are equipped with ESC may drive more riskily or less Motor and Equipment Manufacturers Association, 2006. NHTSA Says Rollover
riskily than drivers of cars without ESC. None of the studies has Ratings for SUVs Improved in 2006 (http://www.mema.org/services/govt/
investigated effects on driver behaviour. A comparison of the issuedetail.php?id=50currentOnly=yes).
NHTSA, 2006. DOT Proposes Anti-Rollover Technology for New Vehicles.
estimated proportions of accidents that may be affected by ESC NHTSA Press releases NHTSA 09-06 Thursday, September 14, 2006.
and the results of the meta-analysis of accident studies does not Page, Y., Cuny, S., 2006. Is electronic stability control effective on French roads?
give any clear answer. If ESC were leading to more careful driver Accident Analysis and Prevention 38, 357–364.
behaviour, this might partly explain the large empirical effects Sferco, R., Page, Y., Lecoz, J.Y., Fay, P.A. S, 2001. Potential effectiveness of
on single vehicle accidents. The larger effects of ESC that have electronic stability programs (ESP)–what European field studies tell us, Pro-
ceedings of the 17th international technical conference on the enhanced
been found under slippery driving conditions also suggests that safety of vehicles, Amsterdam.
drivers do not (over-)compensate the effects of ESC on driving Thomas, P., 2006. Crash involvement risks of cars with electronic stability
dynamics by more risky driving behaviour. The small but signif- control systems in Great Britain. International Journal of Vehicle Safety
icant increase of multi-vehicle accidents suggests the contrary 1, 267–281.
effect. More research is needed to investigate the size of the Unselt, T., Breuer, J., Eckstein, L., Frank, P., 2004. Avoidance of loss of con-
trol accidents through the benefits of ESP. FISITA Conference paper no.
effects and the sources of the heterogeneity in the effects. The F2004V295, Barcelona.
aim should be an improved system design of ESC which avoids Zobel, R., Friedrich, H., Becker, H., 2000. Accident research with regard to
negative behaviour adaptation effects. More knowledge on the crashworthiness and crash avoidance, Conference Vehicle Safety 7–9 June
size of the effects is needed to support increased implementation 2000, London, UK.

You might also like