Accident Analysis and Prevention

Accident Analysis and Prevention 45 (2012) 50 60 Contents lists available at SciVerse ScienceDirect Accident Analysis and Prevention j ourna l ho me pa ge: www.elsevier.com/locate/aap The effects of studded tires on fatal crashes with passenger cars and the benefits of electronic stability control (ESC) in Swedish winter driving Johan Strandroth a,e,, Matteo Rizzi b,d, Maria Olai b, Anders Lie a,c, Claes Tingvall a,e a Swedish Transport Administration, Sweden b Vectura Consulting, Sweden c Karolinska Institutet, Department of Public Health Science, Sweden d Monash University Accident Research Centre, Australia e Chalmers University of Technology, Department of Applied Mechanics, Sweden a r t i c l e i n f o Article history: Received 13 January 2011 Received in revised form 3 November 2011 Accepted 8 November 2011 Keywords: Studded tires Winter tires Electronic stability control ESC Fatal crashes Fatal accidents a b s t r a c t This study set out to examine the effects of studded tires on fatal crashes on roads covered with ice or snow in Sweden and also to investigate the extra benefits of electronic stability control (ESC) during the winter months. Two different studies are presented in this paper. Both studies used an induced exposure approach. In the main study, 369 in-depth studies of fatal crashes with passenger cars were analyzed to determine whether loss-of-control (LOC) had been a major component or not. Only crashes involving cars without ESC and equipped with approved studded or non-studded winter tires were analyzed. The additional study used police-reported crashes that occurred during the winter seasons 2003 2010, involving passenger cars with and without ESC. While police records in Sweden do not include any tire information, it was assumed that most cars involved in crashes during the winter period would be equipped with studded tires. Findings in the main study showed that in 64% of the fatal crashes on roads covered with ice or snow LOC had been a major component. Furthermore, in 82% of LOC crashes, the passenger car over-steered prior to collision. Studded tires were found to have a statistically significant effect of 42% in terms of fatal crash reduction on roads covered with ice or snow, compared to non-studded winter tires. The effect on dry or wet roads in the winter was negative, although statistically non-significant. In the additional study, it was found that ESC further reduced crashes with injuries by 29%. The benefits on severe and fatal crashes were slightly greater (32%), although the lower 95% confidence limit was lower. Although studded tires were shown to reduce the risk of fatal crash involvement, compared to nonstudded winter tires, the proportion of LOC and over-steering among cars with studded tires was large (59% and 49%, respectively). It was therefore concluded that studded tires do not prevent all LOC crashes, while ESC has benefits in those crashes since this technology mostly addresses over-steering. This is also supported by the fact that the share of LOC fatal crashes is considerably lower for ESC-equipped cars. This study recommends that non-esc cars should be fitted with studded tires if they are to be driven on roads covered by ice or snow. If the proportion of studded tires is to be decreased on Swedish roads to reduce the about of hazardous particulates especially in built up areas, from a road safety point of view it is recommended that this should be done in phase with the implementation of ESC on all passenger cars. 2011 Elsevier Ltd. All rights reserved. 1. Introduction The importance of tires for road safety has been studied over a long period of time. In particular, the importance of winter tires and the influence of studs on winter tires have been subjected to many studies Elvik (1999). This is of extra interest in the some Corresponding author at: Swedish Transport Administration, Sweden. E-mail address: johan.strandroth@trafikverket.se (J. Strandroth). geographical areas, i.e. Sweden, where long and cold winters result in a large proportion of traffic and crashes on roads covered with ice or snow. Approximately 20% of all fatal crashes with passenger cars occur on snow and ice throughout the year in Sweden, while during the winter period the share is about 50%. Naturally, this share varies from year to year (Fig. 1) and between the different geographical regions (Fig. 2). The effectiveness of studded versus non-studded tires became an issue already in the early seventies, showing large effects on the risk of crashes under icy or snowy road conditions. In a 0001-4575/$ see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2011.11.005

J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 51 Fig. 1. Proportion of fatal crashes in Sweden on snow or ice during the winter periods 2000 2010 (Swedish Transport Administration 2010). meta-analysis Elvik (1999) however found, that a more comprehensive evaluation shows a limited effect of studded tires, in relation to non-studded winter tires. The reduction of crashes was found to be 5% and non significant for roads covered with ice or snow. On dry or wet roads the effects of studded tires was 2%, also non-significant. More recent studies and calculations have been done in response to problems with air quality related to studded tires. In these studies and calculations, Gustafsson et al. (2006) estimated the typical added safety effect of studded tires compared to non-studded winter tires to 0 10% reduction of passenger car crashes over a winter period and 25% crash reduction on roads covered with ice and snow. Studies from Norway have though shown that there is no significant increase of crashes as a result of reduced Fig. 2. Proportion of fatal crashes in Sweden on snow or ice during the winter periods 2000 2010 per geographical region. The most northern region on top (Swedish Transport Administration, 2010). fitment of studded tires. However, these studies analyzed the number of crashes inside urban areas before and after the reduction of studded tire fitment, and not on the overall effects, even in rural areas. In general these studies however have not taken crash type or injury severity into account. Findings of added safety effects in different studies could also vary due to different study objectives. The effect in crash reduction of studded tires will vary greatly if it is calculated throughout the entire year, the winter period, roads cover with ice or snow, a specific crash type or on different injury severities (Elvik, 1999; Gustafsson et al., 2006). The effects of thread depth of tires have also been studied. Elvik concluded from a number of studies that there is a typical negative effect for thread depth less than 5 mm, while there seems to be no added positive effect over 5 mm (Transportøkonomisk institutt, 2007). The main problems associated with effectiveness studies of studded tires, or any type of tire, are the possible or likely effects of confounders like other risks factors correlated with the choice of tires, if behavioral modification should be included or not, and the classification of crashes. Types of tires are probably not chosen randomly over the population, but rather on a needs basis as well as related to economy that might introduce selective recruitment. Several studies have demonstrated a behavioral adaptation to the type of tire, in that drivers of cars with studded tires drive faster than drivers of cars with other tires. While it is not clear if this is a consequence of the improved road grip or that drivers choosing a higher driving speed tend to use studded tires more frequently. Mäkinen (1996) showed that the improved grip from studded tires lead to higher speed on roads with low friction. In that case it would be natural to include the adaptation in the analysis of the effectiveness of studded tires. Another factor that should be taken into account is that there may be long periods of time during the winter with wet and dry road surface. According to the Scandinavian Tire and Rim Organization (STRO), there are mainly two types of non-studded winter tires on the Swedish market. One suited for Nordic conditions and one for Central European conditions. While Nordic non-studded winter tires are mostly designed for roads covered with ice and snow, Central European non-studded winter tires are designed for milder conditions and higher speeds (STRO, 2010). On wet or dry roads non-studded winter tires might perform better, with regards to stopping distances and stability. This has also been shown in tire tests performed by automobile journals, Auto Motor and Sport (2008). The problem of crash classification is also a reason for varying and not consistent estimates of the effectiveness of studded tires. It is quite clear, that studded tires primarily affect a certain type and amount of crashes, like loss-of-control (LOC) in certain weather or road conditions. Such a classification of crashes is, however, rare and methods based on induced exposure are therefore complex to use. Types of tires classification in mass data sets are also not common. The type of vehicle is also a confounding factor. Forward or rearward drive as well as the presence or absence of ESC might or is known to have a major impact on the risk of certain crash types. In particular, ESC has been shown to have major benefits on wet or icy roads. Lie et al. (2006) have shown a minimum of 35% effectiveness for single/oncoming/overtaking serious and fatal crashes. However, the interaction of ESC and types of tires has not been found in the literature. Studded tires have been a raising issue in Sweden during the last years. Due to their significant contribution to airborne particles (Gustafsson et al., 2006) and the failure in several larger cities to meet the European Union legislation, some municipalities have

52 J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 Fig. 3. The interaction between the main and additional studies. been given the possibility to ban studded tires on some streets were the air quality is particularly poor. These issues motivated a reduction of studded tires in these areas. However, questions are raised about the road safety consequences of this legislation. Hence, there is a need to find the importance for choice of tires for passenger cars, both with and without ESC, in order to give consumers a balanced recommendation taking safety and the environmental effects into account. This project was carried out in two separate studies. The objectives in the main study were to use in-depth data and: analyze the crash dynamics in fatal crashes with passenger cars during the winter period, investigate the effect of studded tires on fatal crashes with passenger cars in different winter road conditions, compared to non-studded winter tires, investigate the influence of tire characteristics (e.g. emplacement of best tires, thread depth, condition) on loss-of-control (LOC) crashes, discuss possible confounders not possible to quantify in the analysis. The objectives in the additional study were to use policereported mass data and: estimate any benefit provided by ESC in reducing crashes during the winter period when many cars are equipped with winter tires. 2. Methods As mentioned above the project was carried out in two separate studies. The main study investigated the effect of studded tires on fatal crashes on different winter road conditions. An additional study was also carried out in order to investigate benefits of ESC on a passenger car population which was reasonably assumed to have a large proportion of studded tires. Fig. 3 reassumes how the main and additional studies interact. Both studies used an induced exposure approach. As mentioned above, the true exposure regarding winter tire (or ESC) is complicated to find and could be associated with many confounding factors. An analysis using induced exposure can then be used when the true exposure is not available or not suitable (Evans, 1998; Lie et al., 2006). Using this approach the crash risk is not calculated by comparing crash involvement to vehicle mileage or number of registered vehicles with and without the system being evaluated. Crash involvement is instead compared to a factor considered nondepended and non-affected by the safety system (i.e. injury outcome in the counterpart or a crash type not influenced by the safety system). In this study it is assumed that the ratio between the number of crashes sensitive to the system being analyzed (in this case, studded tires and ESC) and non-sensitive situations should not differ between two groups of vehicles with the same type of tires or with ESC. If the ratio is different for two groups, where the only noteworthy difference is the type of tires or ESC respectively, it is assumed that the difference is due to them. The effect of the system being analyzed is considered to be zero if R in the equation below is equal to 1. S SYSTEM /N SYSTEM R = (1) S NO SYSTEM/N NO SYSTEM where S SYSTEM, number of crashes sensitive to the system, involving vehicles with the system; N SYSTEM, number of crashes non-sensitive to the system, involving vehicles with the system; S NO SYSTEM, number of crashes sensitive to the system, involving vehicles without the system; N NO SYSTEM, number of crashes non-sensitive to the system, involving vehicles without the system. Thus, the effectiveness in crash reduction can be expressed as: E = 100 (1 R)% (2) The standard deviation of the effectiveness was calculated on the basis of a simplified odds ratio variance, according to Eq. (3). This method gives symmetric confidence limits but the effectiveness is not overestimated. Sd = R 4 1 (3) n i i=1 where n is the number of crashes of each type. The 95% confidence limits are given in Eq. (4). E = 100 R Sd 1.96 (4) One critical factor in this approach is to ensure that the only dissimilarity between the analyzed groups is the type of tires and ESC themselves. Given that the method estimates crash risk, factors concerning injury risk such as vehicle crashworthiness, distribution of restrained occupants, and impact speed, could be confounding the result. The two groups also need to be representative. However, it is possible to control for these factors by comparing the groups

J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 53 with regard to factors known to affect injury risk in a crash and with national statistics (Lie et al., 2006). The controls made in this study are included as tables in Appendix A. When possible confounders are not possible to quantify, they are discussed and judged how they might affect the results. The methods used in the main and additional studies are further described below. 2.1. Main study the effect of studded tires The basic assumption in the main study was that LOC is the relevant crash type that should be used in Eq. (1). Naturally, it could be questioned if LOC is the only crash type influenced by tire properties. The main study, however, set out to investigate the effect of winter tires on fatal crashes in passenger cars only, and there are good reasons to believe that LOC would be the most relevant crash type. It should also be mentioned that, if there were further differences between studded and non-studded winter tires, other than risk for LOC crashes, this would give an underestimation in the calculation of the overall effects of studded tires and the results would therefore be conservative. Earlier studies by Gustafsson et al. (2006) have shown how road surface may affect tire properties and driving behavior. For this reason, the odds ratio in Eq. (1) was calculated for different road surfaces. Data from in-depth studies of fatal crashes with passenger cars collected by the Swedish Transport Administration (STA) were analyzed to determine whether loss-of-control had been a key component. If more than one vehicle was involved in the crash (i.e. a head-on collision between the car and a heavy goods vehicle) it was determined which one initiated the chain of events leading to the crash and then whether LOC had a major contribution or not. Some crashes could immediately be categorized as non-loc, for example if the car was hit by a train in a railway crossing due to unawareness or struck from behind by another vehicle. Other crashes had to be more carefully investigated and LOC-crashes were identified using to a number of parameters that indentifies this type of accident scenario. These parameters were in first hand vehicle trajectories, skid marks, impact location and direction, tire and steering wheel angles in impact as well as other circumstances recorded in the in-depth investigations. These parameters were also investigated when the LOC crashes were classified as over- or under-steering. In the present study, over-steering was considered to be the dynamically unstable condition in which the car rear end skids sideways. On the other hand, a car was considered to have under steered when a front-end skid made the car trajectory not follow the steering wheel input. Since the LOC assessment was crucial for the upcoming analysis, two different crash investigators analyzed each crash separately. A number of crashes were further analyzed in a consensus group. If consensus on the LOC-classification was not reached, it was agreed that the crash should not be included in the analysis. Information about the type of winter tire fitted on the cars involved in the crashes was acquired from the in-depth studies. While studded tires are fairly easy to recognize, it may be more difficult to determine whether anon-studded winter tire is of Nordic or Central European type. In a number of cases experts within the tire industry were contacted to ensure the right classification Information on road conditions is also included in the STA s indepth studies. Crashes were grouped according the LOC-criteria, type of winter tire and road conditions. As mentioned above, calculations were made using induced exposure according to Eq. (1): R = LOC STUDDED /non-loc STUDDED LOC NON-STUDDED /non-loc NON-STUDDED (5) Due to the limited number of crashes with ESC-equipped cars (see below), the influence of ESC was only investigated in a qualitative manner in order to get an understanding of the possible interaction between ESC and different types of winter tires. A statistical analysis was then carried out in the additional study. 2.2. Additional study the benefits of ESC This study used police-reported casualty crashes with passenger cars from the Swedish national accident database, STRADA. As the material in the main study did not include sufficient ESC-equipped passenger cars for statistical analysis (see below), a larger dataset based on national statistics was needed. However, STRADA does not record any information about tires. According to the tire industry, the market share of studded tires among winter tires is today roughly 70% and was even greater during the last few years (SRA, 2009). The assumption was thereby made that a large proportion of passenger cars included in STRADA would have been fitted with studded tires. It was also assumed that the share of studded tires in the ESC group is, in the worst case scenario, as large as in the non-esc group. Newer vehicles in the non-esc groups would be an indication of that. Lie et al. (2006) have earlier used police-reported crashes with passenger cars with a induced exposure approach to show the effectiveness of ESC, using the assumption that rear-end crashes on dry roads are non-sensitive to ESC. This study used the same approach and also assumed that head-on crashes on roads covered with snow and ice as well as single-vehicle crashes (H + S) are sensitive to ESC. The STRADA database does not include any record on ESC fitment on vehicles. However, this information may be acquired through vehicle model and code, as shown in previous studies (Lie et al., 2006). Calculations were made according to Eq. (1), as follows: (H + S) R = ESC /Read-end ESC (6) (H + S) NO ESC/Rear-end NO ESC Furthermore, an attempt was made to calculate the effectiveness of ESC in LOC police-reported crashes. However, crashes in the STRADA database are not automatically classified according to the LOC-criteria. As each crash includes a brief description written by a police officer attending the crash scene, text search was used to find keywords related to LOC-crashes. Crashes containing the word skid ( sladd in Swedish) in the crash description were classified as LOC. In the same way, crashes without the word skid were classified as non-loc. Since this approach had never been tested before, crosschecks were carried out with LOC-crashes in the main study. A sample of 75 fatal LOC crashes from STRADA was identified and compared with the classification done with in-depth studies in the main study and vice versa. Crashes were then grouped according to the LOC-criteria, ESC fitment and road conditions. Calculations were made according to Eq. (1), as follows: LOC ESC /non-loc ESC R = (7) LOC NO ESC/non-LOC NO ESC Finally, the overall effectiveness was then calculated by multiply the effectiveness on LOC crashes with the total share of LOC crashes. 3. Materials 3.1. Main study the effect of studded tires The Swedish Transport Administration (STA) has been carrying out in-depth studies for each fatal road crash since 1997. Crash investigators at STA systematically inspect the vehicles involved

54 J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 Table 1 Number of cars used in the calculations, per road conditions. Ice or snow Dry or wet asphalt Cars with ESC 935 (32%) 3 855 (41%) Cars without ESC 2 010 (67%) 5 346 (58%) ESC fitment unknown 35 (1%) 69 (1%) Sum 2 980 (100%) 9 270 (100%) Table 2 Distribution of LOC in fatal crashes with passenger cars without ESC, per road condition. n LOC crashes Over-steering among LOC Fig. 4. Distribution of cars used in the calculations for the main study, per year of manufacture. in fatal crashes and record direction of impact, vehicular intrusion, seat belt use, airbag deployment, tire properties, etc. The crash site is also inspected to investigate road characteristics, collision objects, etc. Further information about injuries is provided by forensic examinations, questioning and witness statements from the police and reports from the emergency services. These in-depth studies therefore represent a uniquely detailed material that is well suited for detailed analysis. In-depth studies of fatal crashes with at least one killed car occupant were analyzed. Only crashes matching the following criteria were included: The crash was caused by a passenger car (active part). The car was fitted with winter tires especially developed for driving in winter conditions, excluding M + S tires (mud and snow) not approved by STRO for Nordic winter conditions. Thread depth met the legal requirement of minimum 3 mm, on all tires. The car was equipped with the same type of tires. For instance, a crash involving a car with studded tires on the front axle and nonstudded winter tires on the rear axle was excluded. However, a number of crashes involving a car with different types of nonstudded winter tires were included. Finally, 369 in depth-studies were included in the analysis. This material included: Crashes that occurred on road covered with ice and snow during 2000 2009 (n = 242). Crashes that occurred on dry or wet asphalt during 2004 2009. The period November-March was used as temperatures are most likely to be low on the Swedish roads (n = 127). Ice or snow 242 64% 82% Snow 48 60% 76% Ice 120 59% 83% Slush 74 74% 84% Dry or wet asphalt 127 18% 96% Dry asphalt 44 25% 100% Wet asphalt 83 14% 92% in the analysis. Table 2 below shows the distribution of crashes per ESC fitment and road conditions (Table 1). Fig. 5 shows the distribution of crashes per year of manufacture. The average year of manufacture was 2004 for ESC-equipped cars and 2003 for cars without ESC. Police reported crash data are generally known to suffer from a number of data quality problems. As it was assumed that this limitation would equally affect the ESC and non-esc groups, it was not expected to affect this analysis. 4. Results 4.1. Main study the effect of studded tires As shown in Table 2, analysis of the in-depth studies showed that 64% of fatal crashes with passenger cars without ESC on roads covered with ice or snow involved LOC. Loss-of-control crashes were relatively less frequent on dry or wet asphalt (18%). It was also found that, when LOC occurred, the distribution of crashes involving over-steering ranged from 76% to 100%. The analyzed material showed the same distribution of crash type as the national statistics for fatal crashes with passenger cars during the same period, suggesting that the material is representative. Crashes involving ESC-equipped cars were excluded from both groups (n = 10 and n = 16, respectively) since the material was not sufficient for statistical analysis. Fig. 4 shows the distribution of fatal crashes per year of manufacture for cars with studded and non-studded winter tires. The average year of manufacture was 1993 for cars with studded and 1996 for cars with non-studded winter tires. 3.2. Additional study the benefits of ESC Police-reported crashes that occurred during the winter seasons (November-March) between 2003 and 2010 with at least one injured car occupant were used in the additional study. Only passenger cars manufactured between 2001 and 2006 were included Fig. 5. Distribution of cars used in the calculations, per year of manufacture.

J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 55 Table 3 Distribution of under- and over-steering and non-loc in crashes on roads covered with ice or snow, with cars (without ESC) fitted with studded or non-studded tires. Cars with studded tires Cars with non-studded winter tires Over-steering 93 (49%) 34 (65%) Under-steering 20 (11%) 6 (12%) Unknown 0 2 (4%) Non-LOC 77 (40%) 10 (19%) Sum 190 (100%) 52 (100%) Table 4 The effect of studded tires in reducing fatal LOC crashes with passenger cars without ESC, in different road conditions. Estimates are compared to non-studded winter tires. Non-significant results with confidence intervals larger than 100% are indicated with ns. LOC on ice or snow 65% ± 9% Snow ns Ice 77% ± 6% Slush 49% ± 32% Fig. 6. Ratio between LOC and non-loc crashes on roads covered with ice or snow, per position of most worn out pair of tires on cars without ESC. LOC on dry or wet asphalt 32% ± 165% Dry asphalt ns Wet asphalt 24% ± 72% Table 3 shows the distribution of over- and under-steering among all crashes on roads covered with ice or snow, with cars fitted with studded or non-studded winter tires. Further analysis of in-depth studies also suggested that non- LOC crashes on roads covered with ice or snow were not likely to be influenced by different types of winter tires. Out of 87 non-loc crashes, only 3 cases were rear-end crashes and further 14 occurred in intersections. In the remaining 70 non-loc crashes, the passenger car slowly deviated from its course without intervention. In the rear-end crashes, investigators found strong evidence that no braking has occurred. In the crashes in intersections, it was estimated that 2 crashes out of 14 possibly could have been influenced by tire type since better grip in acceleration or braking could have prevented a fatal crash. However, every car involved in a crash in an intersection was equipped with studded tires which would indicate that the potential of studded tires on those crashes was none. Calculations were carried out to ensure that the two groups were comparable in terms of injury risk. The results showed similar distributions of factors know to affect injury risk in a crash, such as unrestrained car occupants (studded 27%, non-studded 31%), collisions with heavy good vehicles (studded 30%, non-studded 17%) and over-speeding (studded 12%, non-studded 14%). Also speed limit and direction of impact showed similar distribution. As mentioned above, the average year of manufacture was 1993 for cars with studded tires and 1996 for cars with non-studded winter tires. This ensured that the effect of studded tires should not be overestimated. Driver characteristics in the two groups such as age, gender and driving under influence of alcohol or drugs were also checked. Table 4 shows the results of analysis based on the odds ratio calculated in Eq. (5). Studded tires were found to reduce the risk of a fatal LOC crash without ESC on ice or snow by 65%. These results were statistically significant, as the lower and upper 95% confidence limits were 56% and 74%, respectively. No statistically significant results were found for crashes on dry or wet asphalt. The effect of studded tires in reducing fatal LOC crashes ranged from 51% to 79%, compared to Nordic and central European nonstudded winter tires, respectively. Further checks within the two groups were also carried out to verify if possible confounders affected the results on roads covered with ice or snow. Calculations showed a small risk increase risk for tires with thread depth below 5 mm. However, cars equipped with the most worn out pair of non-studded winter tires on the rear axle Fig. 7. Ratio between LOC and non-loc crashes on roads covered with ice or snow, for front, rear and four wheels driven cars without ESC. were found to have a greater LOC to non-loc ratio on ice or snow, see Fig. 6. A similar pattern was found for RWD cars with winter tires, see Fig. 7. Furthermore, the condition of studs in studded tires was also found to influence the LOC to non-loc ratio on ice or snow. Fig. 8. Ratio between LOC and non-loc crashes per judged condition of studs in studded tires.

56 J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 Table 5 The overall effect of studded tires in reducing fatal crashes with passenger cars without ESC, per road conditions. Estimates are compared to non-studded winter tires. Ice or snow 42% ± 6% Dry or wet asphalt 6% ± 30% Table 6 The benefits of ESC in reducing head-on and single crashes, on different road conditions. All estimates are reductions in relation to rear-end crashes. Non-significant results with confidence intervals larger than 100% are indicated with ns. Injury crashes Severe and fatal crashes Head-on and single on 54% ± 4% 49% ± 18% ice or snow Snow 40% ± 10% 30% ± 41% Ice 66% ± 3% 70% ± 8% Slush 46% ± 6% 33% ± 34% Head-on and single on 22% ± 11% 18% ± 93% dry or wet asphalt Dry asphalt 15% ± 15% ns Wet asphalt 26% ± 11% 8% ± 81% Calculations showed that crashes involving cars fitted with worn out studded tires without any studs did have a greater proportion of LOC (Fig. 8). Even though rear wheel drive (n = 13) and most worn tires on the rear axle (n = 12) in the non-studded group, as well missing studs in the studded group (n = 15) may affect the risk for LOC, it was calculated these factors were not large enough to influence the overall result in any significant manner, due the total number of crashes. While the ratio LOC to non-loc was also calculated and compared for different age groups, drivers under the influence of alcohol or drugs, drivers under the influence of alcohol or drugs, different speed limits, no major discrepancies were found. Results in Table 5 show that the estimated overall effect of studded tires for cars without ESC was 42% on roads covered with ice or snow. The effect is given by multiplying the calculated 65% LOC reduction with the total share of LOC crashes, 64%. Type of winter tire is assumed to have no effect on other crashes than LOC crashes. This result was statistically significant, as the 95% lower confidence limit was 36%. The total effect on studded tires on ice or snow varied between 31% and 49% compared to Nordic or central European non-studded winter tires, respectively. The overall results for crashes on dry or wet asphalt, calculated in the same way, were not statistically significant. Although studded tires were shown to have a large effect on LOC crashes compared to non-studded winter tires, there was still a large proportion of LOC crashes among cars with studded tires (59%). The qualitative analysis of fatal crashes involving cars with ESC showed that, out of 10 crashes on roads covered with ice and snow, LOC occurred in 3 cases. In those crashes, 2 cars were equipped with non-studded winter tires and 8 with studded tires. Only one LOC crash occurred on dry or wet asphalt (out of 16 crashes with ESC). Accordingly, the share of LOC crashes with ESC was 30% and 6% for ice/snow and dry/wet asphalt respectively, compared to 64% and 18% for cars without ESC. Furthermore, in only one case an ESC-equipped car over-steered. 4.2. Additional study the benefits of ESC The results of the analysis of the benefits of ESC with induced exposure using rear-end crashes on dry roads as non-sensitive to ESC are presented in Table 6. The majority of results for head-on and single-vehicle crashes with injuries were statistically significant. Table 7 The benefits of ESC in reducing LOC crashes, on different road conditions. Nonsignificant results with confidence intervals larger than 100% are indicated with ns. Injury crashes Severe and fatal crashes LOC on ice or snow 51% ± 4% 61% ± 8% Snow 52% ± 10% 78% ± 7% Ice 55% ± 6% 61% ± 13% Slush 46% ± 8% 58% ± 13% LOC on dry or wet asphalt 52% ± 6% 23% ± 40% Dry asphalt 44% ± 14% ns Wet asphalt 56% ± 6% 61% ± 15% Head-on and single-vehicle injury crashes on ice or snow were reduced by 54%, with a lower 95% confidence limit of 50%. Regarding severe and fatal head-on and single crashes the only significant reduction was found on ice or snow, 49% with a lower limit of 31%, and on ice alone, 70% with a lower limit of 62%. Findings from the analysis with induced exposure based only on LOC crashes are listed below. The effectiveness of ESC in reducing LOC crashes with injury on roads covered with ice or snow was 51% and 52% on dry or wet asphalt. The lower 95% confidence limits were 47% and 46%, respectively. All results for injury crashes were statistically significant. When severe and fatal crashes were analyzed separately, even greater LOC-reductions were found on ice or snow. The estimated effectiveness on road covered with ice and snow was 61%, with a lower 95% confidence limit of 53%. No statistically significant results were found for dry or wet asphalt. In the main study, LOC were calculated to amount for approximately 35% of all fatal crashes (in all road surface conditions). In the additional study however, only 16% of all crashes with injuries were found to be contain the word skid and were therefore classified as LOC. The share of LOC among severe and fatal crashes was 20%. This suggested that the method using text search with the word skid could give an underestimation of the proportion of LOC-crashes. This was verified by crosschecking 75 LOC-crashes in the main study with crashes in the additional one. Results showed that the word skid was contained in 42% of LOC crashes, thus providing a 58% underestimation of LOC crashes in police-reported data. While the LOC reduction calculated with this method were comparable with others, it was clear that the overall effectiveness of ESC would be compromised with such an underestimation of sensitive crashes. Therefore the overall effectiveness calculated with this method is not presented. Table 8 shows the overall benefits of ESC according to the calculations based on head-on and single-vehicle crashes. Results for injury crashes were statistically significant. ESC was found to reduce crashes with injuries on road covered by ice or snow by 29%, with a lower 95% confidence limit of 27%. Injury crashes on dry or wet asphalt showed minor benefits, with a 6% reduction. However, severe and fatal crashes on icy or snowy conditions were reduced by 32%. This result was statistically significant, as the lower 95% confidence limit was 20%. No statistically significant reductions were found for severe and fatal crashes on dry or wet asphalt. 5. Discussion In the main study, studded tires on cars without ESC were found to have a statistically significant crash reduction of 42% on road covered with ice or snow, compared to non-studded winter tires. As mentioned above, these road conditions may occur with different frequency depending on geographical region. In urban areas in

J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 57 Table 8 The benefits of ESC in reducing crashes with passenger cars, per road conditions. Injury crashes Severe and fatal crashes Ice or snow 29% ± 2% 32% ± 12% Dry or wet asphalt 6% ± 3% 8% ± 41% southern Sweden only just over 10% of the fatal crashes occur on snow or ice while in rural areas in northern Sweden the same proportion is approximately 90%. Clearly, this issue affects the total effect of studded tires over a winter period in Sweden to a large degree. The effect on dry or wet roads in the winter indicated a negative effect, although statistically non-significant. It was also found that the STA s in-depth fatal crash data did not contain enough crashes involving ESC-equipped cars for a statistical analysis. It was therefore decided that the benefits of ESC during the winter months would be analyzed using mass-data containing cars whose majority was reasonably assumed to have studded tires. Under this assumption, ESC was found to further reduce crashes with injuries by 29%. The effect on severe and fatal crashes was slightly greater (32%), although the lower 95% confidence limit was lower. This is well in line with other studies of ESC (Ferguson, 2007). The basic assumption in the main study was that in fatal crashes on roads covered with ice or snow, LOC crashes are primarily influenced by the type of tires (sensitive crashes). However, LOC crashes may not be the only crash type sensitive to tires, but also rear-end crashes which are related to braking and friction. Also collisions with vulnerable road users (VRU) could of course be affected by type of tires with different friction and braking distances. On the other hand, Strandroth and Persson (2005) showed that LOC crashes with passenger cars have a higher injury risk due to side impacts as an effect of over-steering, while rear-end crashes are more related to non-fatal whiplash injuries with long term consequences. Official statistics in Sweden confirm that, as in the last five winter periods (2005 2009), only three fatalities out of 233 are due to a rear end collision. Given this, it is reasonable to believe that vehicle stability is a far superior factor in preventing fatalities in passenger cars on ice or snow than braking distance. However, further research is needed to investigate the influence of types of tire on rear-end collisions with whiplash injuries and collisions with VRU. It could also be argued that the LOC assumption may not apply to less severe crashes with other scenarios, where braking (or accelerating) could be more crucial for preventing or mitigating the crash. However, the analysis of the in-depth studies showed that very few of the fatalities in the non-loc crashes could have been prevented with studded tires, by either avoiding the crash or mitigating injury severity. While analyzing in-depth studies could suffer from subjectivity issues, this has been done in earlier studies with good results (Sferco et al., 2001; Sporner and Kramlich, 2003; Rizzi et al., 2009). Besides, it was shown that LOC classification of fatal crashes is possible. While this method is rather uncommon, presumably due to its complexity, it is clear that such classification should suit better an analysis on tires on fatal crashes where stability is the main issue. As already shown by earlier studies (Strandroth and Persson, 2005), LOC accounted for a majority (64%) of the fatal crashes with cars without ESC on roads covered with ice or snow, while on dry or wet roads LOC accounted for only 18%. This suggests that countermeasures increasing stability may have greater potentials on roads with low friction. This is verified by several studies on the effects of ESC (Lie et al., 2006; Chouinard and Lécuyer, 2010). This study showed that passenger cars without ESC had a 42% lower fatal crash involvement on snow and ice with studded tires, compared to non-studded winter tires, which is considerably larger than previous studies. However, other studies have not taken crash type or injury severity into account. The large effects could be a result of a severity gradient not investigated in previous studies. It would be logical to believe that something that improves stability would primarily have effect in fatal crashes were stability is crucial. It is also important to notice that the effect of 42% lower crash involvement is a result of large effects on LOC-crashes which are common on roads covered with ice or snow, but not common on wet or bare roads. Thus, the calculated crash reduction of 42% for studded tires compared to non-studded winter tires are not relevant to the entire winter period. Naturally, it can be questioned whether the calculated effect is affected by confounders or not. The ideal would be to have access to more advanced vehicle data, e.g. EDR-data. However, this data was not available and therefore a number of factors related to crash and injury risk within road, vehicle and driver characteristics were controlled, and reported in Appendix A. While the studded and nonstudded groups were matched with regard to the most relevant parameters influencing injury risk, it may be difficult to control all parameters for crash risk. To better understand why studded tires have this positive effect, it could have been valuable to have more detailed information about friction coefficient, horizontal and vertical alignment, stopping sight distance, number and width of lanes, etc. to investigate differences in tire performance in detail. However, reliable information about these parameters is hard to find in almost any accident database. Also, they were not considered crucial for the purpose of this study which was, as mentioned above, to give general consumer advice rather than detailed tire properties results. It should also be noted that the approach based on induced exposure would normally compensate for this, as the result is given by relative differences within the two analyzed groups. Furthermore, if there were major confounders contributing to the result apart from type of tires, the findings would be affected not only on roads covered with ice or snow, but also on other road surfaces. This is, however, not the case. The positive effect is isolated only to roads with ice and snow, ice in particular. On dry and wet road surface the effect is negative ( 6%, not significant) suggesting that the results for snow and ice largely are due to studded tires themselves. Other factors giving conservative results would be if studded tires had positive effects on other types of crashes on roads covers with ice or snow besides LOC. Also, passenger cars equipped with studded tires were in average 3 years older than cars with nonstudded winter tires. While several studies have shown that newer cars generally perform better in terms of crashworthiness (Swedish Transport Administration, 2010; Farmer and Lund, 2006), the material included in this study was based on the injury outcome in the crash (i.e. only fatal crashes were analyzed). Clearly, this would give an underestimation in the calculations, although it was argued that the difference was too small to affect the results substantially. However, behavioral adaptation among drivers using nonstudded winter tires could have affected the results. If some of these drivers were to drive more slowly on ice or snow, this would be likely to affect the injury outcome in the crash and consequently decrease the number of non-loc crashes with non-studded winter tires included in the study. This, on the other hand, would affect the results in favor of studded tires (see Eq. (5)). While this issue is difficult to control when true exposure is not available, it could be argued that this scenario is unlikely. One reason for that is the fact that causality can be seen between studs and stability. An interesting finding in this study was that the risk for a fatal LOC crash increased if the worst tires were placed on the rear axle, regardless of type of winter tire. While this is in line with the common recommendation to keep the best pair of tires on the rear axle, caution needs to be taken, as the material was limited. It was also found that studded tires without any studs left related to greater risk for a fatal LOC crash (almost twice as high as for non-studded

58 J. Strandroth et al. / Accident Analysis and Prevention 45 (2012) 50 60 winter tires, Fig. 7). While the number of crashes was limited, this may suggest that the main difference between studded and non-studded winter tires is the studs themselves, and not the compound or other characteristics. The effect of studded tires in terms of crash reduction ranged between 31 and 49%, depending on whether they were compared to non-studded winter tires made for Nordic and central European conditions, respectively. The material was too small to investigate the same proportion on dry or wet surfaces. While it may be reasonable to believe that under such conditions central European non-studded tires would perform better, this issue needs to be further investigated. It is also important to further investigate the effect of studded tires on not only fatal crashes. For example rear-end crashes which cause the majority of serious whiplash injuries in Sweden. In this case braking distance would be more crucial than stability and other methods would be needed. If possible the true exposure and crash rate would be suitable. But maybe also induced exposure could be used considering crashes on a wet or dry road surface to be non-sensitive to studded tires. A similar approach has been used by Evans (1998) to investigate the safety benefits of antilock brake systems on passenger cars. Since meaningful statistical analysis was not possible on fatal crashes with ESC-equipped cars, the benefits of ESC were investigated on mass data from police reports. While previous studies based on the same approach have already proven great benefits with ESC (Lie et al., 2006; Chouinard and Lécuyer, 2010), the present study aimed to investigate the same issue but with cars mainly equipped with winter tires, assumedly studded tires. Because of this reason, the study included only crashes that occurred during the period in which winter tires are mandatory in Sweden. ESC showed large effects on LOC crashes regardless of road surface (Table 7), which may not be totally unexpected. Unfortunately, it was difficult to identify LOC crashes in mass data and consequently head-on crashes as well as single-vehicle crashes were used in the calculations. In the additional study it was shown that ESC can reduce LOC crashes during the winter months by approximately 50%. This was also suggested by the fact that the number of fatal crashes involving ESC-equipped cars in the main study was too limited for statistical analysis. Yet, it was still evident that the LOC-ratio is extremely low in those crashes. On roads covered with snow and ice in particular, the LOC rate for ESC-equipped cars was 30%, compared to 64% for cars without ESC. On dry or wet roads, the same ratio was 6% and 18%, respectively. As mentioned above, the additional study was based on the assumption that most cars in these calculations would be fitted with studded tires. Table 6 shows that ESC was found to have the same effectiveness in reducing LOC crashes on roads covered with ice and snow (51%) as well as on dry and wet asphalt (52%), on which studded tires were not shown to have any significant safety benefits compared to non-studded winter tires. This may indicate that the interaction between ESC, different winter tires and LOC crashes cannot be explained by tire grip only. ESC would obviously need a certain amount of grip in order to control the car by braking one wheel individually. However, other factors may also come into play and the findings in this study indicate that ESC can reduce LOC crashes, regardless of the tire grip. Previous studies (Lie et al., 2006; Chouinard and Lécuyer, 2010) as well as the current one have shown that ESC has a large effect on reducing real-life crashes on road with low friction (i.e. covered by ice or snow), which would be contradictory if the effectiveness of ESC only depended on the grip between the tires and road surface. Studded tires have been a raising issue in Sweden during the last years. Due to their significant contribution to airborne particles (Gustafsson et al., 2006) and the failure in several larger cities to meet the European Union legislation, some municipalities have been given the possibility to ban studded tires on some streets were the air quality is particularly poor. These issues motivated a reduction of studded tires in these areas. However, questions are raised about the road safety consequences of this legislation. Based on the findings in the present study, it could be argued that studded tires could prevent many LOC crashes. While this is an important finding, further studies should be carried out to better understand why this is the case. The importance of different tire properties should be investigated and tested in combination with different road geometries. However, it was also shown that a large proportion of crashes involving over-steering would still happen even with studded tires. Those crashes would normally be addressed by ESC. While there are reasons to believe that ESC would prevent those crashes regardless of type of winter tires, future studies will need to confirm this hypothesis. The true interaction between ESC and different types of winter tires, depending on the road surface, needs to be further investigated. In the meanwhile, it is recommended that consumers driving non-esc cars should fit their cars with studded tires if they intend to drive on roads covered by ice or snow. Also, if a legislation is intended to limit the fitment of studded tires on Swedish roads in order to reduce the about of hazardous particulates in built up areas, it is recommended that this is done in phase with the implementation of ESC on all passenger cars. 6. Conclusions There were a number of important major findings in this research: 64% of fatal crashes with passenger cars without ESC on roads covered with ice or snow involved LOC. Loss-of-control crashes were relatively less frequent on dry or wet asphalt (18%). When LOC occurred, the distribution of crashes involving oversteering ranged from 76% to 100%. Studded tires were found to reduce fatal crashes with passenger cars on roads covered with ice and snow with 42% compared to non-studded winter tires. The lower limit with 95% confidence interval was 36%. Studded tires were not found to have any significant effects in reducing fatal crashes with passenger cars on dry or wet asphalt, compared to non-studded winter tires. On roads covered with ice or snow, cars equipped with the most worn out pair of non-studded winter tires on the rear axle were found to have a greater risk for LOC crashes. A similar pattern was found for RWD cars, as well as cars equipped with studded tires without any studs left. Crashes involving cars fitted with worn out studded tires without any studs did have a greater proportion of LOC. Given the effect of studded tires on roads covered with ice or snow shown in this study, the importance of specific tire properties and road geometries should be further investigated. The effectiveness of ESC in reducing LOC crashes with injury on roads covered with ice or snow was 51% and 52% on dry or wet asphalt. The lower 95% confidence limits were 47% and 46%, respectively. All results for injury crashes were statistically significant. While behavioral adaptation among drivers using non-studded tires could have affected the results in favor of studded tires, it was argued that this scenario is unlikely. ESC was estimated to reduce all casualty crashes by 29% on roads covered with ice and snow. The effect on dry and wet roads during