Accident Analysis and Prevention 81 (2015) 24–29

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Effectiveness of low speed autonomous emergency braking in

real-world rear-end crashes
B. Fildes a, * , M. Keall b , N. Bos c , A. Lie d , Y. Page e, C. Pastor f , L. Pennisi g, M. Rizzi h ,
P. Thomas i, C. Tingvall j
a
Monash University Accident Research Centre, Australia
b
University of Otago, New Zealand
c
SWOV, Institute for Road Safety Research, The Netherlands
d
Swedish Transport Administration (Trafikverket), Sweden
e
European Automobile Manufacturer’s Association, Renault, France
f
Federal Highway Research Institute (BASt), Germany
g
Automobile Club d'Italia (ACI), Italy
h
Folksam Insurance, Sweden
i
Loughborough University, UK
j
Swedish Transport Administration (Trafikverket), Sweden

A R T I C L E I N F O A B S T R A C T

Article history: This study set out to evaluate the effectiveness of low speed autonomous emergency braking (AEB)
Received 29 August 2014 technology in current model passenger vehicles, based on real-world crash experience. The validating
Received in revised form 20 March 2015 vehicle safety through meta-analysis (VVSMA) group comprising a collaboration of government, industry
Accepted 22 March 2015
consumer organisations and researchers, pooled data from a number of countries using a standard
Available online 6 May 2015
analysis format and the established MUND approach. Induced exposure methods were adopted to control
for any extraneous effects. The findings showed a 38 percent overall reduction in rear-end crashes for
Keywords:
vehicles fitted with AEB compared to a comparison sample of similar vehicles. There was no statistical
AEB technology
Effectiveness
evidence of any difference in effect between urban (
60 km/h) and rural (>60 km/h) speed zones. Areas
Meta-analysis requiring further research were identified and widespread fitment through the vehicle fleet is
Real-world recommended.
Rear-end crashes ã 2015 Published by Elsevier Ltd.

1. Introduction There are at least two versions of these systems, namely low-
speed or “City” systems or high-speed “Inter-Urban” systems that
Advanced crash avoidance technologies are increasing rapidly operate at different speed thresholds. These systems commonly
in passenger and commercial vehicles as industry, government and consist of an automatic brake function and a forward collision
the community focus on improved vehicle safety systems. One of warning sensor and vehicles may offer either single of both
the more promising safety technologies that is starting to appear as functionalities (Euro NCAP, 2014). The AEB “City Safety” system
standard equipment on modern passenger cars and sport utility was first introduced by Volvo cars in their XC60 sport utility vehicle
vehicles (suv) is autonomous emergency braking (AEB). Autono- around 2009 (ITS International, 2013) and more recently, extended
mous emergency braking systems apply the vehicle brakes when a the technology as standard equipment in all its passenger vehicles.
collision is eminent in spite of any reaction by the driver. In some In recent years, other manufacturers, primarily in European and
technologies, the system forewarns the driver with an acoustic Japanese models, also offer versions of similar systems in their
signal when a collision is still avoidable, but subsequently applies modern vehicles. The technology operates for vehicle speeds up to
the brakes automatically if the driver fails to respond. 30 km/h or 50 km/h in some vehicle models.
It is claimed that autonomous emergency braking systems offer
substantial reductions in crash avoidance or injury mitigation as
* Corresponding author at: Building 70, Monash University, Wellington Rd.,
shown in Table 1. It should be noted with some caution that many
Clayton, Victoria 3800, Australia. Tel.: +61 399054369. of these studies used a range of different technology functionalities
E-mail address: brian.fi[email protected] (B. Fildes). and assessment methods.

http://dx.doi.org/10.1016/j.aap.2015.03.029
0001-4575/ ã 2015 Published by Elsevier Ltd.
 B. Fildes et al. / Accident Analysis and Prevention 81 (2015) 24–29 25

Table 1
Published studies of benefits of AEB technology.

AEB report AEB type Assessment method Crash reductions Injury reductions

Fatalities Serious Slight Injuries

Sugimoto and Sauer (2005) CMBS Simulation rear-end crashes 38% 44%
Page et al. (2005) EBA Case analysis 7.50% 11%
Forward crashes
Najm et al. (2006) ACAS FOT responses 6–15%
Breuer et al. (2007) BAS+ Simulation ped/rear crashes 44%
Kuehn et al. (2009) CMBS Case analysis front/rear crash 40.80% n.a. n.a. n.a. n.a.
GDV (2011) EBA2 Case analysis rear-end crashes 13.90% 2.20% 9.40% 35.70%
Grover et al. (2008) AEB Case analysis sensitive crashes 30%
Kusano and Gabler (2012) AEB Case analysis rear-end crashes 7.70% 50%
HLDI (2011) AEB Insurance claims 22–27% 51%
Döcke et al. (2012) AEB Case analysis rear-end crash 25–28%
Chauvelet al. (2013) AEB Case analysis pedestrians 4.30% 15% 37% n.a. n.a.

While many of these evaluations claim substantial benefits, or VVSMA group) to evaluate the effectiveness of low speed
most are based on desk-top evaluations of expected crash and autonomous emergency braking technology (AEB city), using this
injury outcomes. Unfortunately, there is only limited evidence of new approach. The objective was to measure the likely reductions
their real-world effectiveness in reducing crashes or injuries. in important rear-end injury crashes for vehicles fitted with this
Single country crash databases, traditionally used for conducting safety technology. Case and control vehicles were agreed upon by
real-world evaluations, are limited by the slow take-up rates of the whole group and these are listed in Appendix A.
these new technologies, limited crash data, and lower crash rates
by owners of new safer vehicles. New systems are also commonly 1.2. Low speed AEB technology
available on only a few car models and sometimes optional which
increases the time needed to assess their real-world effectiveness. CarAdvice (2014) noted that low speed AEB or City Safe
One way of potentially speeding up the evaluation process is to technologies are marketed under a variety of names, including City
adopt a wider approach to collecting and analyzing crash data, Brake Control (Fiat), Active City Stop (Ford), City Emergency Brake
rather than simply relying on one country’s analysis from their (Volkswagen) and City Safety (Volvo). As their names suggest, this
limited crash numbers. type of AEB technology is geared towards low speed situations,
generally under 30 km/h. These systems rely on radar sensors
1.1. Meta-analysis detecting an emergency situation and apply the brakes as needed.
They tend to work most effectively over short distances.
Classic meta-analysis, commonly used by the medical fraterni- Low speed AEB technology, such as the City Safety system fitted
ty, typically combines the findings of various existing published to new Volvo vehicles, operates at speeds between 30 and 50 km/h.
randomized control trials of a common theme to produce a much As the name implies, the system is designed to only offer
larger pool of research data, leading to a more robust assessment protection in rear-end crashes in mainly urban areas. Low-speed
(Cochran Collaboration, 2013). This approach typically relies on AEB systems use sensors to monitor the road ahead, typically
retrospectively published studies that meet established criteria, and 6–8 m. One common technology is a LIDAR (light detection and
while they are very useful for helping establish general trends and ranging) sensor, typically mounted at the top of the windscreen,
outcomes, they are commonly assembled from evaluations which determines whether or not there is an object in front of the
(clinical trials) already published in the scientific literature and car which presents a risk. If there is, the AEB system will, typically,
thus still subject to long delays. pre-charge the brakes so that the car will provide its most efficient
An alternative prospective approach involves a planned braking response, should the driver react. If the driver does not
collaboration of independent aggregate analyses from data respond, the car will automatically apply the brakes to avoid, or in
analysts using a common study design. This brings together a some cases to mitigate, the accident. If, at any point, the driver
much larger pool of data than any one country, has available, intervenes to avoid the accident, by hard braking or avoidance
speeds-up the process of evaluating safety technologies, and steering, the system will disengage (Euro NCAP, 2014).
provides a more internationally relevant assessment of the safety
benefits than any one single country can provide. In a recent 1.3. Induced exposure
published study (MUNDS) it was shown that it is possible to
increase the available relevant crash data by combining data from a An induced exposure approach was used in the present paper,
number of countries using meta-analyses and thus obtain robust as the true exposure with low-speed AEB may be difficult to obtain
statistical evaluations more quickly (Fildes et al., 2013). in different countries, and could be also associated with some
Meta-analysis has the additional advantage of circumventing confounding factors. An analysis using induced exposure can be
the need to work with unit-level data. In ideal circumstances, used when the true exposure is not available or not suitable, as
regression models could be fitted to unit-level data, allowing for argued in Evans (1998),Hautzinger (2003) and Lie et al. (2006).
more efficient estimation and control for potential confounders. In Induced exposure approaches to estimating risk attempt to
practice, road safety agencies and police are reluctant to hand over quantify on-road exposure using counts of crash involvements. The
their data at this level to external parties, but are willing to crash types used for these risk estimates generally focus on events
summarise their data at an aggregated level suitable for a meta- where the driver of a given vehicle could be considered to be
analysis. passively involved in the crash. Such crash events, therefore, are
With Euro NCAP’s initiation, a technical group of researchers conceptualised as a sampling mechanism and the counts of the
from government, industry and research organizations was crashes are assumed to be proportional to the amount of driving
assembled (the Validating Vehicle Safety through Meta-Analysis undertaken by a given driver group or vehicle type. Validation of
26 B. Fildes et al. / Accident Analysis and Prevention 81 (2015) 24–29

these assumptions has rarely been undertaken. Indeed, at-fault each dataset, as explained above. The formula used by each data
classification is unavailable in some of the countries that provided supplier in the computation was in the form of the estimator of the
data here. preventive fraction:
A study using New Zealand data compared exposure in terms of
ða=bÞ
distance driven with counts of crash involvements to identify E¼1 (1)
ðc=dÞ
which crash configurations might be preferred as an induced
exposure measure (Keall and Newstead, 2009). Two sets of crash where:
types – collisions where the vehicle in question was impacted from a = AEB fitted vehicles as striking vehicle,
the rear and collisions where the driver was adjudged not to be at b = AEB fitted vehicles as struck vehicle,
fault – were found to perform equally well. But even these two c = non-AEB vehicles as striking vehicle,
best-performing crash types when used to estimate risk showed d = non-AEB vehicles as struck vehicle.
systematic biases according to the driver characteristics age and Definition of striking and struck vehicles was determined using
sex, and the size of the vehicle (ibid). Despite these limitations, a range of different methods, appropriate from each of the
induced exposure methods are widely used in road safety research individual database codes. Crash variables include 2-vehicle (car–
as they are often the only form of exposure measure available. car) injury crashes that occurred in years 2009 or more recent,
Using this approach the crash risk is not calculated by using relevant vehicles. Urban (speed limit,
60 km/h) and rural
comparing crash involvement to vehicle mileage or the number crashes (speed limit, >60 km/h) was analysed separately. Non-AEB
of registered vehicles with and without the system being vehicles were selected from an agreed list of equivalent vehicles.
evaluated. Crash involvement instead, is compared to a situation
assumed or known to be not affected by the safety system (i.e., 2.2. Meta-analysis approach
non-sensitive). In other words, the ratio between the number of
crashes sensitive and non-sensitive to low-speed AEB is analyzed The individual analyses were then brought together using
across two different crash populations where the only noteworthy meta-analysis. If the population of crashes analysed consists of
difference should be the fitment of low-speed AEB. sensitive crash types combined with comparison crash types, an
odds ratio (a/b)/(c/d) will estimate the relative rate of sensitive
2. Method crashes for the AEB vehicles compared to the same rate for the
control vehicles (Fildes et al., 2013).
The underlying philosophy adopted here involved combining An estimate of the effectiveness of AEB for the set of sensitive
data from a number of countries to overcome the shortage of data crashes was formed by subtracting this odds-ratio from 1, as shown
and provide earlier results. Unfortunately, database owners are in Eq. (1) above. Hedges and Vevea (1998) showed that the optimal
typically unable to contribute actual case records for reasons of weights for meta-analysis are:
confidentiality and legal restrictions. Thus, each database partici-
1
pant agreed to undertake their own analysis using a common w¼ (2)
se2
format and provide an aggregate analysis for combining overall
using a meta-analysis approach. Potential cases and controls are where “se” is the standard error of the estimate from a given study.
listed in Appendix A. The odds-ratio has a skewed distribution and has a complex
Crombie and Davies (2009) describe meta-analysis as a standard error formula. Therefore it is preferable to conduct all
statistical technique for combining the findings from independent calculations on the natural log of the odds-ratios (Wilson, 2000).
studies. They note that in medical research, it is commonly used to Finally results are converted back into odds-ratios by the inverse
assess the effectiveness of healthcare interventions by combining natural log function. The standard error of a logged odds-ratio can
data from two or more randomised control trials. While many of be approximated by:
these independent studies must meet particular entrance criteria rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 1 1 1
to be included, they are nevertheless independent in that se ¼ þ þ þ (3)
a b c d
researchers do not necessarily set out with a common data format
and thus, some variation naturally exists when combining multiple
studies. As noted above, the approach here sets out with each study Using these weights, the combined effect size, (ES) which is a
adopting a common format in their approach ensuring a closer weighted average of the individual study effect sizes are estimated
match between studies. It has been validated in previous research by:
(Fildes et al., 2013).
Sðw  ESÞ
ES ¼ (4)
Sw
2.1. Agreed data format

The agreed data format was one that all data providers could The standard error (SE) that can then be estimated from:
achieve. Sensitive crash types for the technology focussed on rear- sffiffiffiffiffiffiffiffi
end crashes where the AEB vehicle made impact with another 1
SEES ¼ (5)
vehicle (striking) or was impacted from behind (struck) in only Sw
2-car crashes. Those vehicles fitted with AEB were compared with
non-AEB crashed vehicles in sensitive crash types. AEB fitted
vehicles were identified from local vehicle knowledge where the The overall estimate of effectiveness (E) then is:
technology was known to be standard equipment. The comparison E ¼ 1  expðESÞ (6)
non-AEB vehicles were similar vehicle types but without the
technology. Induced exposure was the method used to control for
extraneous influences. All data providers agreed to conduct
individual local analysis around a core set of parameters, using 2.2.1. Tests of adequacy of estimation method
their national (police-reported) crash database for rear-end The homogeneity analysis tests whether the assumption that all
crashes from 2009. Induced exposure methods were adopted in of the effect sizes are estimating the same population mean is a
 B. Fildes et al. / Accident Analysis and Prevention 81 (2015) 24–29 27

reasonable assumption. The homogeneity statistic, Q, was calcu- Table 2

Number of crashes by country, crash type, AEB and Non-AEB vehicles, all speed
lated as:
limits.

½Sðw  ESÞ2 Countrya

Q ¼ Sðw  ES2 Þ  (7) Crash type AEB Non-AEB
Sw Country 1 Front to rear 21 202
Impacted from rear 84 435

This statistic takes the Chi-square distribution with degrees of Country 2 Front to rear 1 138
freedom = number of ESs less than 1. A statistically significant Q Impacted from rear 4 246
statistic would suggest that the effect sizes are sufficiently
different to require more sophisticated estimation methods, such Country 3 Front to rear 14 434
Impacted from rear 16 374
as a random effects model (Wilson, 2000).
Logistic regression was used in assessing the speed zone effects Country 4 Front to rear 35 404
to control for extraneous influences across the various databases. Impacted from rear 59 450
The outcome variable was crash type, set to 1 when the vehicle was
Country 5 Front to rear 15 105
striking and 0 for struck (as described above) with explanatory
Impacted from rear 24 95
variables AEB (an indicator as to whether the vehicle was AEB
equipped or not), country, speed limit (rural or urban) and an Country 6 Front to rear 2 85
interaction between AEB and speed limit. This coefficient of this Impacted from rear 1 82
final term was used to assess whether there was differential a
Individual countries, predominantly European, could not be singled out for
effectiveness in different speed limit areas. confidentiality reasons.

3. Results
different vehicle type (cars and SUVs for instance). While this could
The subsequent results obtained from combining the individual be worthy following up in future, importantly, it was unlikely to
country analyses, using the meta-analysis approach outlined above have had any influence on the effectiveness calculations for the low
are listed below. For various reasons, it was not possible to identify speed AEB technology as both the AEB-equipped and comparison
these particular countries in the results that followed. Table 2 vehicles were assessed on the same basis in the analysis.
shows the number of AEB crashed vehicles and non-AEB vehicles
used in the VVSMA AEB meta-analysis. While the number of cases 3.1. Urban and rural crashes
varied across these 6 individual countries, they were controlled for
a number of similar characteristics (national police data, Each country also provided counts of crash involvements
2009 crashes and more recent, similar makes and models, etc). according to the speed limit of the crash location: urban
The resultant country-specific AEB effectiveness estimates, (
60 km/h) and rural crashes (>60 km/h); or other means of
along with the pooled weighted average estimate, is provided in defining crash zone. This enabled the analysis to control for the
Table 3. The overall estimate of 38% effectiveness was highly speed limit and adjust for any potential confounding by the
statistically significant (P = 0.0006) and represents a 38% reduction location of the crash. It also allowed a test to be conducted as to
in the sensitive crashes relative to the comparison crash type. A whether AEB might be more effective in lower speed limit areas
negative percentage (as shown as the lower bound of most than high speed limit areas, given the functionality of the system,
country-specific effectiveness estimates) represents an increase in although sample size limitations were likely to provide low power
sensitive crashes relative to the comparison crashes. The only for such a test.
country whose estimates was statistically significant taken in Using logistic regression, it was possible to test whether there
isolation was country 1, although country 4 was too when the was any difference in the effectiveness of AEB between urban and
controls were non-AEB Volvos or general controls. The homoge- rural areas. These findings showed no evidence whatsoever of any
neity statistic, Q, was calculated to be 1.65, with no homogeneity differential effectiveness according to speed limit area.
issues (P = 0.90) (Keall and Newstead, 2009).
Interestingly, there were global differences observed in the 4. Discussion
ratios between front and rear collisions across the 6 countries
listed in Table 2. They varied from around 1:2 for countries 1 and This analysis of crash data from these six countries has shown a
2 to approximately 1:1 for the others. This was unexpected a-priori, clear change in the distribution of crash types for vehicles with AEB
it would be expected to be equal for all countries. The reasons for relative to control vehicles without this technology. The main
this variation are not readily apparent but likely to involve meta-analysis found a significant 38% reduction in crashes where
differences in entrance criteria for the various databases, the AEB vehicle impacted the rear of another vehicle, relative to the
occupants’ age and sexes, crash distribution types, and possibly rate that these vehicles were impacted from the rear by other

Table 3
Country-specific and overall effectiveness estimates with 95% confidence intervals.

Country Odds ratio (95% confidence limits) Point effectiveness (95% confidence limits)
Country 1 0.54 (0.32,0.89) 46% (11%, 68%)
Country 2 0.45 (0.05, 4.03) 55% (303%, 95%)
Country 3 0.75 (0.36, 1.57) 25% (57%, 64%)
Country 4 0.66 (0.43, 1.03) 34% (3%, 57%)
Country 5 0.57 (0.28, 1.14) 43% (14%, 72%)
Country 6 1.93 (0.17, 21.69) NAa NAa
Overall 0.62 (0.47, 0.82) 38% (18%, 53%)
a
The preventive fraction is not defined when an increase in prevalence is estimated.
28 B. Fildes et al. / Accident Analysis and Prevention 81 (2015) 24–29

vehicles. Such a reduction can occur via a reduction in the rate of 4.1. Study limitations
impacts to the rear of other vehicles by the AEB vehicles, but it can
also occur (or be contributed to) by an increase in the rate that AEB As noted above, it was not possible to disaggregate the data by
vehicles are impacted from the rear by other vehicles, which could occupants’ age, sex and vehicle type as the data provided by the
be an unintended consequence of more effective braking. As noted various databases were not aggregated to such a level. This was a
above, this requires further analysis. function of the need for each data provider to undertake their own
In a forerunner to the current study (Fildes et al., 2013) in which analysis based on their inherent limitations. Thus, this is clearly a
various countries' data were combined to estimate the effective- potential limitation with the analysis reported here. In addition, it
ness of electronic stability control (ESC), the countries provided would have been useful to have used side-impact crashes as an
counts of crashes disaggregated by driver age group, vehicle size, alternative induced exposure measure as recommended by
year of manufacture, speed zone and road condition (dry/wet/ Keall and Newstead (2009). While this was attempted here,
snow/ice). There was a good rationale for such disaggregation: ESC unfortunately, not all of the 6-countries databases had suitable
has been shown to have different levels of effectiveness under coding for side impact collisions and the results reflected this. This
different road conditions and for different vehicle types (ibid). In is clearly another topic for further research.
that analysis, these various disaggregating variables were then able Furthermore, the data provided for the sensitivity analysis did
to be used as explanatory variables in the regression analysis, not separate crash types for which AEB might be effective from
controlling for potential confounding arising from these factors. those where AEB would have little effect. There was some evidence
Little is currently known about ways that low speed AEB may vary of lower effectiveness when side impacts were used as a
in its effectiveness, unlike ESC. Nevertheless, the current measure comparison crash type but as noted above, these results were
of effectiveness may have been influenced by some of the factors not significant. As the means of assessing impacts to the side of the
listed above that were not available to be used in the analysis. vehicle were not consistent across countries, this finding cannot be
Although not reported in the aforementioned paper, an analysis considered robust, and is something worthy of further research.
was conducted to look at the impact on the overall estimated It might be argued that the underlying assumptions of the
effectiveness of ESC of excluding the control of these potential induced exposure approach should be validated in terms of the
confounders from the analysis. A 30% higher estimate was found, not-at-fault assumption to be sure the conclusions are not
indicating the importance – in the case of ESC evaluations – of misleading. In this study, it was not possible to test this assumption
controlling for such factors, particularly road condition and vehicle as at-fault classifications were not available in all of the national
size. It is a limitation of the current study that the impact of not databases involved in the meta-analysis. Thus, care needs to be
controlling for similar factors is unknown. taken in assuming these findings are totally representative.
The figures from previous simulation or case studies listed in Importantly, the findings reported here confirm the advantage
Table 1 for rear-end crashes were highly variable and dependent on of the meta-analysis process adopted here when evaluating safety
the method used to compute them. Those using simulation technology. As seen in Table 2, only one of the six countries
techniques were closest to the real-world benefits reported here reported was able to show significant differences between the AEB
(between 38% and 44%). Other methods from case-by-case and Non-AEB vehicles albeit with broad confidence intervals. This
estimations or insurance claims were much less (between 14% is clear evidence of the benefits of the approach for providing
and 30%). While injury benefits were not computed in this real- timely real-world evidence of technology effectiveness.
world study, the earlier studies confirm that there is potential for
marked reductions in fatal and serious injury reductions from this 5. Conclusions
technology. This needs further research using the VVSMA real-
world approach. The findings showed a surprising 38 percent overall reduction
The findings from the logistic analysis showed that there were in real-world, rear-end crashes for vehicles fitted with low speed
no differences in effectiveness between crashes that occurred AEB compared to a comparison sample of equivalent vehicles.
overall to those that occurred in urban areas. This might be because There was no statistical difference between urban (
60 km/h) and
most of the crashes observed in this analysis predominantly rural (>60 km/h) speed zones. The meta-analysis approach used in
occurred in low speed regions. Thatcham (2009) claimed that this analysis is a unique academic contribution to the evaluation of
these systems are more effective at lower speeds (<25 mph) where vehicle safety technologies internationally and proved to be
more than 75% of accidents occur, based on insurance findings. The reliable with robust findings. Areas requiring further research
low speed AEB technology is designed to work up to around were identified for fine-tuning these findings. Clearly, at this level
50 km/h which shows the suitability of this technology in the of effectiveness, low speed autonomous emergency braking (AEB)
prevention the bulk of rear-end crashes. Whether the alternative is potentially an important active safety technology and wide-
high speed AEB technology is also effective cannot be assumed spread fitment through the vehicle fleet should be encouraged in
from these results. the interest of improved vehicle safety.
The variation between hitting and being hit with and without
low speed AEB technology across the 6 databases was a surprising Acknowledgements
finding. It may be that the proportional differences were a
function of the crash distributions across countries, the ages and The authors acknowledge the support of Michiel van Ratingen
sex differences across countries, or personality variation of the of Euro NCAP and the Australian NCAP in organising and
drivers. The national data available lacked sufficient detail to supporting this research program, and the Department of
address these issues. It could also suggest differences in the data Infrastructure and Regional Development in Australia for funding
records in the countries that participated where police atten- the Australian contribution. In addition, members of the VVSMA
dance at a crash can be quite variable. As each country committee beyond the authors deserve special credit for their
incorporated much the same vehicle mix, it is unlikely that it's contribution, namely Murray Doyle, Thatcham, UK; Sebastian
a function of major variation in vehicle types. Nevertheless, this is Döring, VW, Germany; Anders Kullgren, Folksam, Sweden; Stefan
an interesting finding and one that requires further research Rauscher and Olaf Jung, BMW, Germany; and Johan Strandroth,
using more detailed data. Trafikverket, Sweden; as well as all the contributing organisations
 B. Fildes et al. / Accident Analysis and Prevention 81 (2015) 24–29 29

for self-funding their own research effort. We also thank the Chauvel, C., Page, Y., Fildes, B.N., Lahausse, J., 2013. Automatic emergency braking for
reviewers of this paper for their useful and helpful comments. pedestrians effective target population and expected safety benefits, Paper
number 13-0008. The 23rd International Technical Conference on the Enhanced
Safety of Vehicles (ESV) Seoul, Republic of Korea, May 27–30.
Appendix A. Cochran Collaboration, 2013. Cochran Reviews; Cochran Community (beta),
(downloaded September 2013). http://community.cochrane.org/cochrane-
reviews.
Crombie, IK., Davies, H.T., 2009. What is Meta-Analysis? Published by Hayward
Selected case and control vehicles. Medical Communications, a division of Hayward Group Ltd. Covent Garden,
London, Euro NCAP (2014). Autonomous Emergency Braking, http://www.
euroncap.com/testsexplained/aeb.aspx.
Case vehicles Selected controls Döcke, S.D., Anderson, R.W.G., Mackenzie, J.R.R., Ponte, G., 2012. The potential of
Volvo case models Volvo controls Car by car Sports utility vehicles autonomous emergency braking systems to mitigate passenger vehicle crashes.
controls Paper presented at the Australasian Road Safety Research, Policing and
Education Conference, 4–6 October, 2012, Wellington, New Zealand.
Volvo S60 from w Volvo S60 Alfra Romeo Audi Q5 Euro NCAP, 2014. Autonomous Emergency BrakingEuro NCAP, Brussels, Belgium. .
2010 2006 to 2009 159 from 2006 (downloaded February 2015) http://www.euroncap.com/en/vehicle-safety/
Volvo V60 from Volvo V70 from Audi A4 and BMW X3 the-rewards-explained/autonomous-emergency-braking.
w48 2010 w20 2007 A6 from 2005 Evans, L., 1998. Antilock Brake Systems and Risk of Different Types of Crashes in
Volvo XC60 from Volvo XC70 from BMW 3 series (4/ Citroen C- Traffic. In proceedings of the 1998 ESV Conference. Paper Number. 98-S2-O-12.
w20 2008 w20 2007 5 door) Crossover Fildes, B., Keall, M.D., Thomas, P., Parkkari, K., Pennisi, L., Tingvall, C., 2013.
Volvo V70 from Volvo S80 from BMW 5 series Ford Kuga Evaluation of the benefits of vehicle safety technology: the MUNDS study. Accid.
Anal. Prev. 55, 274–281.
w20 2011 w20 2006
GDV, 2011. Advanced Driver Assistance Systems, Research Report FS 03,
Volvo XC70 from Volvo XC90 from Citroen C5 Honda CRV from
Unfallforschung der Versicherer (GDV), ISBN-Nr: 978-3-939163-37-4, German
w20 2011 2006 2006 Insurance Association Insurers Accident Research, Berlin. (Authors: T. Hummel,
Volvo S80 from Ford Mondeo Hyundai Santa Fe M. Kühn, J. Bende, and A. Lang).
w20 2011 from 2006 from 2006 Grover, C., Knight, I., Okoro, F., Simmons, I., Couper, G., Massie, P., Smith, B., 2008.
Volvo V40 from Honda Accord Mercedes GLK Automated Emergency Brake Systems: Technical Requirements, Costs and
w20 2012 from 2006 Benefits, PPR227, TRL Limited, DG, Enterprise, European Commission, April
Mazda 6 from Mitsubishi 2008.
2006 to 2012 Outlander from Hautzinger, H., 2003. Measuring the effect of primary safety devices on accident
2006 involvement risks of passenger cars – some methodological considerations,
Other case vehicles Mercedes C Class Nissan X-Trail from SARAC, Project, European Commission, Subtask Report 12-09-2003.
4/5 door 2006 Hedges, L.V., Vevea, J.L., 1998. Fixed-and random-effects models in meta-analysis.
VW Up – country Mercedes E Class Opel Antra Psychol. Methods 3, 486.
HLDI, 2011. Volvo’s City Safety prevents low-speed crashes and cuts insurance costs,
dependent 4/5 door
Status Report, Vol. 46, No. 6 | July 19, 2011.
VW CC – country Opel Insignia Peugeot 4007
ITS International, 2013. City Safety reduces low speed accidents on Volvo’s XC60 and
dependent
S60, First published in ITS International, May June 2013. http://www.
Mazda CX5 2011- Peugeot Renault Koleos itsinternational.com/sections/nafta/features/city-safety-reduces-low-speed-
some country 407 from 2006 accidents-on-volvos-xc60-and-s60/.
Mazda 6 2013 –- Peugeot 508 Toyota RAV4 from Lie, A., Tingvall, C., Krafft, M., Kullgren, A., 2006. The effectiveness of electronic
some countries 2006 stability control (ESC) in reducing real life crashes and injuries. Traffic Inj. Prev.
Renault Laguna VW Tiguan 7, 38–43.
Skoda Suberb Kuehn, M., Hummel, T., Bende, J., 2009. Benefit estimation of advanced driver
Toyota Avensis assistance systems for cars derived from real-world accidents, Paper Number
VW Passat 09-0317. The 21st International Technical Conference on the Enhanced Safety of
Vehicles Conference (ESV) – International Congress Centre, Stuttgart, Germany,
June 15–18.
Keall, M., Newstead, S., 2009. Selection of comparison crash types for quasi-induced
exposure risk estimation. Traffic Inj. Prev. 10 (1), 23–29.
Notes: speed threshold increased from 30 km/h to 50 km/h in Kusano, K.D., Gabler, H.C., 2012. Safety Benefits of Forward Collision Warning, Brake
Volvo models w20 from 2012. Assist, and Autonomous Braking Systems in Rear-End Collisions, Intelligent
Not all models were available or sold apart from Volvo XC60 in Transportation Systems, IEEE, Transactions, Volume: 13 (4).
Najm, W.G., Stearns, M.D., Howarth, H., Koopmann, J., Hitz, J., 2006. Evaluation of an
every country. Automotive Rear- End Collision Avoidance System (technical report, DOT HS 810
Not all models listed (apart from Volvo) were sold with low 569). Cambridge, MA: John A. Volpe National Transportation System Center, U.S.
speed AEB in all countries. Department of Transportation.
Page, Y., Foret-Bruno, J., Cuny, S., 2005. Are expected and observed effectiveness of
Volvo XC60 was the predominant case model across all
emergency brake assist in preventing road injury accidents consistent? 19th
countries. ESV Conference, Washington D.C., USA.
Sugimoto, Y., Sauer, C., 2005. Effectiveness Estimation Method for Advanced Driver
Assistance System and its application to Collision Mitigation Brake Systems.
References
Paper Number 05-0148, the 19th International Technical Conference on the
Enhanced Safety of Vehicles (ESV), Washington D.C., June 6–9.
Breuer, J.J., Faulhaber, A., Frank And Gleissner, P.S., 2007. Real world safety benefits of Thatcham, 2009. Autonomous Emergency Braking, http://www.thatcham.org/aeb.
brake assistance systems. Proceedings of the 20th International Technical Wilson, D.B., 2000. Meta-analyses in alcohol and other drug abuse treatment
Conference on the Enhanced Safety of Vehicles (ESV) in Lyon, France, research. Addiction 95, 419–438.
June 18–21.
CarAdvice, 2014. Autonomous emergency braking explained, (downloaded
December 2014). http://www.caradvice.com.au/293366/autonomous-
emergency-braking-explained/.