The probability of death in road traffic accidents. How important is a quick medical response? by Guia Buenas Practicas Motocicletas

G Model AAP-2049;

No. of Pages 9

ARTICLE IN PRESS Accident Analysis and Prevention xxx (2009) xxx–xxx

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

The probability of death in road trafﬁc accidents. How important is a quick medical response? Rocío Sánchez-Mangas ∗ , Antonio García-Ferrrer, Aranzazu de Juan, Antonio Martín Arroyo Dpt. Quantitative Economics, Universidad Autónoma de Madrid, Avda. Tomás y Valiente, 5, Campus de Cantoblanco, 28049 Madrid, Spain

a r t i c l e

i n f o

Article history: Received 20 August 2009 Received in revised form 20 November 2009 Accepted 9 December 2009 Keywords: Road trafﬁc accidents Probability of death Emergency services Medical response time Probit model

a b s t r a c t The number of deaths in road trafﬁc accidents in Spain exceeds three thousand people each year. Public authorities have implemented some policies with the aim to reducing this number. Among them, the improvement of road quality standards and some legal changes encouraging careful driving behavior. However, less attention has been focused on one of the issues that may be critical to reducing the number of fatalities caused by trafﬁc accidents: a quick emergency medical care. In this paper, we use a sample of more than 1400 accidents occurred on Spanish roads in May 2004. Our objective is to analyze to which extent a reduction of the time interval between the crash and the arrival of the emergency services to the crash scene is related to a lower probability of death. Our results suggest that a 10 min reduction of the medical response time can be statistically associated with an average decrease of the probability of death by one third, both on motorways and conventional roads. © 2009 Elsevier Ltd. All rights reserved.

1. Introduction Traffic accidents on Spanish roads are responsible for more than 3000 fatalities per year. With the aim to reducing this number, public authorities have acted in three different ways. First, using the mass media as channels to periodically broadcast traffic campaigns promoting careful driving. Second, there has been a considerable improvement in road conditions in the last few years. For instance, from 1998 to 2004 the number of motorway kilometers available in the national road network raised from 3600 to 12500. Additional efforts have been made to identify and reduce the number of accident black spots. Third, in the last several years some new rules got into force, for example, the mandatory use of safety belts for all occupants of a motor vehicle and the prohibition of using cell phones while driving. These and other related legal norms have been established in the new Score Driving License, which is into force in Spain since July 2006. Italy, France or Germany have similar schemes. All these steps show the concern about road safety and its economic and social impact. Apart from the legal changes aimed at encouraging careful driving behavior and other changes focused on improving road quality standards, one of the issues that may be critical to reducing the number of fatalities caused by traffic accidents has to do with a quick and efficient emergency medical care. For example, in UK, according to UNECE (2000), about 50% of deaths occur at the site

∗ Corresponding author. E-mail address: rocio.sanchez@uam.es (R. Sánchez-Mangas).

of the crash or during transport to hospital, within less than 1 h after the accident (Haegi, 2002; Coats and Davies, 2002; Charlton and Smith, 2003). Actually, most of the prehospital deaths from accidental injury are due to road traffic accidents, and a portion of them could be preventable with better emergency medical care (Hussain and Redmond, 1994). The Spanish Traffic General Directory (DGT hereafter) is the institution in charge of all aspects related to road safety in Spain. Let us consider the fatalities/casualties ratio (FCR hereafter), where casualties includes fatalities as well as seriously and slightly injured victims.1 In 2004, according to the DGT, the FCR in road accidents was 3.3% in Spain, a figure higher than the EU-15 average of 2.3% (UNECE, 2004). However, international comparisons of fatality ratios should be taken with caution, for at least two reasons. First, as it is stated in Elvik and Mysen (1999), the definitions of injury severity vary across countries. Besides, there is evidence of incomplete reporting, especially in the case of injuries, but also to some degree in the case of fatalities. Second, the comparisons should be made once we control for exogenous and endogenous factors affecting fatality ratios, such as the age composition of the population or the vehicle fleet, among others (Page, 2001). Then, a crude comparison of fatality ratios across countries is always debatable.

1 DGT deﬁnes fatalities as those victims who die within 24 h after the crash. Seriously injured victims are those who require hospitalization for more than 24 h. The remaining victims are considered as slightly injured. The information used by the DGT to establish to severity of the victim is provided by hospital records.

Please cite this article in press as: Sánchez-Mangas, R., et al., The probability of death in road trafﬁc accidents. How important is a quick medical response? Accid. Anal. Prev. (2010), doi:10.1016/j.aap.2009.12.012

G Model AAP-2049;

No. of Pages 9

ARTICLE IN PRESS R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

The objective of this paper is to determine to which extent a faster medical response may be associated with lower fatality risk in Spanish road accidents. To this aim, we link two different databases. One of them contains information on crashes, vehicles and victims. The other one offers, for the first time in Spain, information on the timing of the emergency medical assistance. After the linkage process, our dataset consists of a cross-section of more than 1400 road accidents occurred in Spain in May 2004. There are several papers in the literature focusing on the relationship between the outcome of a crash and the provision of medical care. From an aggregate perspective, Noland (2003) and Noland and Quddus (2004) point out the improvements of medical care and technology as contributing factors in the reduction of traffic fatalities in industrialized countries in the last several years. At a micro-level, most of the papers analyzing this issue have focused on two factors that can help to reduce fatalities. On the one hand, a quick and accurate crash notification to the emergency services (Brodsky, 1990, 1992, 1993; Evanco, 1999; Clark and Cushing, 2002). On the other hand, the distance from the crash site to a medical centre. This distance can explain geographical variations in traffic mortality rates in areas with different emergency services accessibility (Bentham, 1986; Muelleman and Mueller, 1996; Durkin et al., 2005; Zwerling et al., 2005; Li et al., 2008). Our paper is a contribution to this literature. We focus on the medical response time, that we define as the time interval between the crash and the arrival of the emergency services to the crash scene. We analyze the potential relationship between this response time and the probability of death. From the methodological perspective, we estimate probit regressions on the outcome of the crash (fatality vs. non-fatality). To isolate the effect of the medical response time from other effects, we include control variables related to the crash and victims’ characteristics. The rest of the paper is organized as follows. Section 2 provides a description of the data sources used in this paper and the medical response time. Section 3 focuses on the methodology, presenting probit models of the probability of death in a road accident. Section 4 presents the empirical results. A discussion of the main findings is provided in Section 5. Finally, Section 6 offers our main conclusions and some suggestions for future work. 2. Data sources and descriptive evidence 2.1. Data sources In García-Ferrer et al. (2007), the authors explain the variability of accidents, injuries and fatalities in Spain using econometric models that include variables representing the evolution of the economic activity, the stock of vehicles and its variation and the available toll and free motorway kilometers. Using annual aggregated data, they show that the percentages of explained variances are quite reasonable in the case of accidents and injuries. However, this percentage is considerably lower for fatalities. They argue that, other things equal, a variable representing improvements in emergency medical care should be included in the analysis of fatalities. Unfortunately, such a kind of medical response variable has not traditionally been available in the Spanish data at a micro-level. The DGT performed a pilot survey in May 2004 that overcame this lack. The study focused on road accidents, both on motorways and conventional roads. It was conducted in all the Spanish regions but Catalonia and the Bask Country, and it represented a coverage of more than 83% of all road accidents occurred in Spain in that period. The study was performed through a questionnaire that was organized into two different parts. In the first one, the information to include was related to basic characteristics of the crash, such as the province, day, time of the day, place, and some details about the emergency services: the person who made the first emergency call,

the time this call was made and the emergency centre that received it. There was also information related to the emergency medical services arriving at the crash scene, basically, the arrival time and the type of emergency unit. Since the person making the emergency call is required to provide some information about the severity of the crash (for example, in terms of the number and type of vehicles involved), the type of emergency unit at the crash scene can vary from firemen unit to MICU (Mobile Intensive Care Unit) ambulance, medical helicopter, medicalized ambulance and non-medicalized ambulance.2 The second part of the questionnaire was devoted to reporting information related to the victims: the role at the crash scene (driver, passenger, pedestrian), the position (inside or outside the vehicle) and the physical condition, among these alternatives: conscious injured, unconscious injured, apparently dead. For each of the crashes occurring in May 2004 in the Spanish regions participating in this study, the traffic agent attending the accident was required to fill in this information. Of course, the medical data provided by the traffic agent in terms of the physical condition of the victims could be inaccurate or even wrong. Thus, the information we use related to the severity of the victim comes from hospital records. The relevance of this pilot study is that, for the first time in Spain, it reports the response time of the emergency medical services. Although this survey is helpful for our purposes, the information on accident and victim characteristics was not thorough. For example, it did not contain any information about road and weather conditions, vehicle age, use of safety measures, etc. Thus, this database needed to be completed with additional information. On an annual basis, the DGT collects data from all accidents occurred in Spain, both on roads and urban areas. These data are very informative on the accident characteristics, as well as the vehicles involved in the crash and the victims. It does not contain information on emergency care issues. However, it offers information about the victims’ condition belonging to one of these categories: fatalities, seriously injured and slightly injured victims.3 This information is provided to the DGT by hospital records, as we mentioned. For those road accidents occurred in May 2004 in those regions that participated in the pilot study, it was necessary to link both databases (the pilot survey with emergency medical care variables, and the information from the DGT 2004 report, with exhaustive information on accident, vehicles and victims, including the severity of the victim). The importance of linking data from different sources with information on crashes and health’s status of the victims has been addressed by Rosman (2001) and Boufous and Williamson (2006), among others. After the data record linkage and dropping out some observations on accidents for which we could not match the information of both databases, we got a final dataset with 1463 crashes.4 Table 1 shows the distribution of victims and that of accidents and fatalities by road type.5 2.2. Descriptive evidence of the medical response time The emergency medical services in Spain get into action, as it happens in the rest of Europe, when a call is received at the 112 emergency number. To the aim of improving the performance of the emergency services, the European Commission has set out an action

2 Medicalized ambulance: with medical equipment, a physician and a nurse. Nonmedicalized ambulance: with basic medical equipment, but in which the main purpose is to convey the victims to hospital, not to provide medical care. 3 See the deﬁnition of victim types in footnote 1. 4 A description of the linkage process can be seen in Appendix A. 5 The 2264 involved vehicles reported in Table 1 include cars (75% of the total) and, in much lower proportion, vans, trucks, buses, motorcycles and bicycles. We do not have in the ﬁnal sample accidents with pedestrians involved. The main features of accidents and victims characteristics are shown in Appendix B.

ARTICLE IN PRESS

G Model AAP-2049;

No. of Pages 9

R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx Table 1 The linked data on May 2004 road accidents. Accidents Vehicles Casualties Fatalities Seriously Injured Slightly Injured

1463 2264 2541 136 630 1775

Distribution of accidents and fatalities by road type Accidents Motorways 450 Conventional roads 951 Others 62 Total 1463

Fatalities 51 78 7 136

Table 2 Medical response time (minutes).

Mean Std. deviation Minimum Maximum 5% percentile 25% percentile Median 75% percentile 95% percentile

Motorways

Conventional roads

25.4 16.0 0 215 10 15 23 30 50

26.7 20.3 0 351 10 15 24 30 60

plan with the goal that all new vehicles in Europe are equipped for eCall from 2010. eCall is an emergency call that can be made manually by vehicle occupants or automatically through the activation of in-vehicle sensors in the event of an accident. This system establishes a 112 connection that automatically provides accurate details of the location of the emergency situation. This can result in an important reduction of medical response times (European Commission, 2009). As it was said in the previous subsection, the sample we use in this study provides some information about the emergency services. Although it is well known that the emergency number is 112, only in 43% of the cases in our sample the emergency call was first received in this service. In around 38% of them, the first call was received in the Coordinated Centre of Traffic Accidents (a service depending on the DGT), and in the rest of the cases the first call was received at State Police or Local Police Departments. There were no significant differences in the centre that received the emergency call in terms of the type of road. The information on the time the emergency call was received is missing for a large proportion of observations, around 50%, with some variations depending on the emergency centre. Then, it has not been possible to provide an accurate measure of this variable. Instead, we measure the medical response time as the interval between the accident and the arrival of the emergency medical services to the crash scene, for which the information is more accurate.6 Table 2 shows the main descriptive statistics for this variable by road type. The distribution of the medical response time is quite similar in both types of roads. The main difference appears in the right tail. In 95% of the accidents on motorways, the medical assistance arrived at the crash scene in less than 50 min. On conventional roads, this figure is 60 min. The average medical response time is around 25 min, with only slight differences between both road types.

6 The time of the accident is, thus, referred to the real time of the crash, not the time of the ﬁrst call received by the emergency services.

Fig. 1 provides some insight about the connection between FCR and medical response time. We consider 5 min time intervals for the medical response time in both toll/free motorway and conventional road accidents. We have included the number of accidents and the number of casualties in each interval.7 In motorway accidents, we observe higher FCRs in the first two time intervals, i.e., when the medical response time is ≤10 min. These higher FCRs might be explained by the type of accidents occurring on motorways. Since they tend to be more serious than those on conventional roads, it is more likely that drivers/passengers die almost instantaneously.8 For those accidents in which the medical response time is longer than 10–15 min, we observe that the FCR is higher the longer the medical response time is. If we focus on the last time intervals (> 30 min), FCRs are much lower. According to Fig. 1, the medical assistance delay is not so crucial to saving lives after 30 min. However, it may be extremely important to guarantee better conditions for injured people. The pattern is quite different for conventional road accidents. We do not observe higher FCRs in the first time intervals, which can be interpreted as a lower probability of instantaneous death, in comparison with accidents occurred on motorways. Up to 25 min, the FCR is higher as the waiting time increases. However, if the waiting time is above 25 min, the pattern is just the opposite. Both on motorways and conventional roads, patterns in Fig. 1 support the evidence reported in several studies indicating that the first minutes after the crash are critical to saving lives. According to our data, those “golden minutes” seem to be the first 25–30 ones. Thus, this figure indicates the need of focusing on this time interval to quantify the statistical association between low medical response time and lower probability of death.9 3. Methodology In this section, we formulate probit models for the outcome of a crash. We want to quantify the previously mentioned statistical association between low medical response time and lower probability of death. We include control variables related to accident and victims characteristics. Let Y be a binary variable defined as follows:

Yi =

if individual i dies in the accident

otherwise

and let MRT represents the medical response time in minutes. Let X a be a vector of variables related to the accident characteristics

7 Note that the total number of motorway accidents is 450. This figure is 951 for conventional roads. We let out of the analysis 62 accidents occurred in other type of roads (see Table 1). 8 It is very difficult to provide an estimation of the proportion of victims for which a shorter medical response time will not have an effect due to instantaneous death. In this sample, the number of fatalities on motorways is 51 (see Table 1). We have information about the distribution of these fatalities across the response time intervals. However, a conclusion cannot be inferred from this, since we do not know the time of the death, but only the arrival time of the emergency services. Some useful information to reach a tentative conclusion on this point would be the specific cause of death. For example, in the case of overwhelming injury to the brain, heart of great vessels, death will likely be unavoidable (see Hussain and Redmond, 1994). These authors analyze the type of trauma, classified into an injury severity score from 1 (minor) to 6 (currently untreatable). Using this scale, they provide an estimate of deaths from accidental injury that could be preventable and those that cannot. Unfortunately, our database does not provide information on the type of trauma, and therefore, we cannot make such a kind of analysis. 9 Fig. 1 includes all accidents in our dataset. If we focus on fatal accidents (those with at least one fatality), or serious ones (those with no fatalities but at least one seriously injured victim), the pattern is similar, with the only difference of higher FCRs than those reported in Fig. 1. This evidence is available from the authors upon request.

ARTICLE IN PRESS

G Model AAP-2049;

No. of Pages 9

R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

Fig. 1. FCR and medical response time (minutes).

and X p a vector of personal characteristics of the driver and the passengers. We formulate standard probit models for motorway and conventional road accidents, that can be written as follows: p

Pr(Yi |MRTi , Xia , Xi ) = ˚(ˇ0 + h(MRTi ) + ˇa Xia + ˇp Xi )

(1)

where ˚(·) stands for the standard normal cumulative distribution function and ˇa and ˇp are the vector of parameters associated with accident and personal characteristics, respectively. For conventional roads, we specify a linear function h(MRTi ) = ˇ1 MRTi while for motorways we use a quadratic one, h(MRTi ) = ˇ1 MRTi + ˇ2 MRTi2 . The different speciﬁcation for h(·) for each type of road is driven by the empirical evidence shown in Fig. 1 about the relationship between the medical response time and the probability of death. Our interest is to estimate all the ˇ parameters and obtain the partial effect of medical response time on the probability of death. Considering MRT as a continuous variable, the marginal effect can be computed for conventional roads as: p

∂Pr(Yi |MRTi , Xia , Xi ) ∂MRT

= ˇ1 (ˇ0 + ˇ1 MRTi + ˇa Xia + ˇp Xi )

(2)

and for motorways as: p

∂Pr(Yi |MRTi , Xia , Xi ) ∂MRT

= (ˇ1 + 2ˇ2 MRTi ) (ˇ0 + ˇ1 MRTi + ˇ2 MRTi2 p

+ ˇa Xia + ˇp Xi )

(3)

where (·) stands for the standard normal density function. If we are interested in computing the effect of discrete changes in MRT, e.g., a reduction of m minutes, the partial effect should be computed by differences in probabilities, such as: p

Pr(Yi |MRTi − m, Xia , Xi ) − Pr(Yi |MRTi , Xia , Xi ) p

= ˚(ˇ0 + h(MRTi − m) + ˇa Xia + ˇp Xi ) − ˚(ˇ0 + h(MRTi ) p

+ ˇa Xia + ˇp Xi )

(4)

The partial effects in (2)–(4) are not constant across observations. For each observation, they depend on the values of accident p and personal characteristics, i.e., Xia and Xi . Thus, we compute the partial effect for each observation and we average them for the whole sample. 4. Empirical results Table 3 provides a definition of some of the accident- and individual-related variables included in our dataset. Estimation results appear in Table 4. Based on the patterns shown in Fig. 1, we have focused on those accidents with MRT no longer than 30 min on motorways and no longer than 25 min on conventional roads. Table 3 Accident- and individual-related variables. Accident-related variables Nighttime 1 for accidents occurred during the night, 0 otherwise Outward/return 1 for outward/return journeys from a public holiday, 0 otherwise Work commute 1 for accidents during commuting to work, 0 otherwise Sunday 1 for accidents on Sundays, 0 otherwise Good weather 1 for good weather conditions, 0 otherwise Heavy traffic 1 for high traffic density, 0 otherwise Speed excess 1 for accidents with speed excess, 0 otherwise No. of vehicles Number of involved vehicles in the accident Run-off-the-road 1 for run-off-the-road accidents, 0 for collision with other vehicles Vehicle age Number of years since the vehicle was registered Individual-related variables Male driver 1 for male drivers, 0 for female drivers Alcohol/drugs 1 if driver under alcohol or drugs effects, 0 otherwise Driver’s disability 1 if driver with some degree of disability, 0 otherwise Driving hours Number of non-stop driving hours Not safety 1 if individual did not use any safety measure, 0 otherwise Driver’s age Codified in four intervals: ≤ 24, 25–44, 45–64, > 64 years old

ARTICLE IN PRESS

G Model AAP-2049;

No. of Pages 9

R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx Table 5 Average partial effect of MRT on the probability of death.

Table 4 Probability of death (probit estimates). Motorways −0.144 (0.048) 0.004 *** (0.001) ***

MRT MRT2 Accident-related variables Nigthttime Outward/return Sunday Run-off-the-road Speed excess

0.392 * (0.204) 0.924 ** (0.391) −0.635 ** (0.281) 0.557** (0.242) −0.250 (0.241)

Individual-related variables Not safety Alcohol/drugs

0.945 *** (0.207) 0.784 ** (0.353)

Driver’s age ≤ 24 years old 25–44 years old 45–64 years old Constant No. of observ. Pseudo-R2 Log-likelihood

Conventional roads *

0.023 (0.013) –

0.240 * (0.139) 0.866 *** (0.314) −0.380 ** (0.177) −0.355** (0.141) 0.332 * (0.180)

0.739 *** (0.144) 0.178 (0.334)

−0.816 ** (0.391) −0.889 ** (0.394) −0.347 (0.387)

−0.284 (0.264) −0.277 (0.234) −0.156 (0.260)

−0.522 (0.475)

−1.881 *** (0.296)

557 0.217 −114.60

992 0.104 −202.89

Notes: Standard errors in parenthesis. * Significant at 10%. ** Significant at 5%. *** Significant at 1%.

This selection yields to 886 accidents (315 on motorways, 571 on conventional roads) and 1638 casualties (592 on motorways, 1046 on conventional roads).10 The second column in Table 4 refers to motorway accidents and the third column to conventional road accidents. Since the unit of observation is the casualty and some casualties correspond to the same vehicle and/or to the same accident, this can induce a correlation in the error term, due to non independent observations, that will affect the estimation of the variances of the estimators. One possibility to allow for correlation in the error term is to consider some clustering scheme in the standard errors (Wooldridge, 2003). We have analyzed different clusters, not only at the accident level, but also at the geographical level, considering clusters in terms of provinces or regions. The results are very similar to those reported in Table 4. Thus, the potential lack of independence seems not to be an issue in our case.11 5. Discussion In this section, we present a discussion of the estimation results. We focus on the medical response time and its partial effect on the probability of death. We also discuss the results on the control variables included in the models.

Estimated probability at MRT = 25 Avg. partial effect of 10’ MRT reduction

Motorways

Conventional roads

0.072

0.080

0.024 (0.004, 0.044)

0.026 (0.0007, 0.053)

Notes: 90% conﬁdence intervals of the average partial effect in parenthesis.

shows that the longer the MRT, the higher the probability of death. For motorways, the sign of the coefficients (negative in the linear term and positive in the quadratic one) shows that a similar relationship appears above certain threshold. These findings are in line with the observed patterns in Fig. 1. Our results relate to other micro-level studies in the literature. As it was stressed in the introductory section, most of the papers focusing on the relationship between the outcome of a road crash and the provision of medical care have analyzed the influence of two factors: the crash notification systems and the emergency services accessibility. Both factors explain the time interval between the crash and the arrival of the emergency services. Brodsky (1990, 1992, 1993) analyze the delay in ambulance dispatch in Missouri road accidents, pointing out the importance of improving emergency notification, both in terms of speed and accuracy. Evanco (1999), using US data, shows that the use of automatic notification systems, that reduce delays in the activation of emergency services, can result in substantial reductions of fatalities in rural areas. Clark and Cushing (2002), also using US data, find a similar evidence. With respect to the distance between the crash site and the hospital, Jones and Bentham (1995), using UK data, find no relationship between the outcome of the crash and the estimated time taken for the emergency services to reach the crash scene and to convey the victims to the hospital. However, the opposite evidence is more widely reported. Muelleman and Mueller (1996), using US data, find that one contributing factor to explain the higher road fatality rates in rural regions in comparison with that of urban areas, is the delayed medical care. Similar findings for US data are reported in Zwerling et al. (2005). In the same line, Durkin et al. (2005) analyze crashes in Wisconsin and find that a higher distance from the crash site to an emergency medical centre is associated with a higher fatality risk. Li et al. (2008) report similar evidence using data from crashes in Taiwan. In a recent paper, González et al. (2009) analyze crashes in Alabama. They show that mortality rates are higher in areas with increased emergency services response time, including time to arrive to the crash scene and to convey victims to the hospital. Quick and accurate crash notification systems and good emergency services accessibility contribute to reduce the time interval between the crash and the provision of emergency care. Our results, in line with this related literature, show that there is an association between low medical response time and lower fatality rates.

5.1. Medical response time 5.2. Partial effect of the medical response time According to the results shown in Table 4, MRT appears as a signiﬁcant variable to explain the probability of death for both types of roads. For conventional roads, the positive sign of the coefﬁcient

10 The number of observations in Table 4 is slightly lower than the number of casualties reported here. The reason is that there is missing information for some of the explanatory variables. More speciﬁcally, Not Safety and Run-off-the-road, have around 5% of missing values in each type of road. This yields to a ﬁnal estimation sample of 557 observations in motorway accidents and 992 in conventional road accidents, as it is reported in Table 4. 11 Estimation results under different clustering schemes are available from the authors upon request.

Our probit coefficients offer the sign of the relationship, but not the quantitative estimated decrease in the probability of death associated with a medical response time reduction. From our estimates, we have computed the partial effect of a 10 min response time reduction, from 25 to 15 min.12 Estimated average partial effects and confidence intervals at 90% confidence level are shown in Table 5.

12 The average medical response time in our dataset is around 25 min, with only slight differences by type of road (see Table 2).

G Model AAP-2049;

No. of Pages 9

ARTICLE IN PRESS R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

In motorway accidents, the computed partial effect means that a 10 min response time reduction can be associated with a probability of death 0.024 percentage points lower. Although it seems that this figure is low, it must be taken into account that the estimated probability of death in accidents with MRT = 25 min is 7.2% in our database. Thus, a 0.024 percentage points decrease means that the probability of death goes from 0.072 to 0.048, which represents a 33% reduction. For conventional road accidents, a similar 10 min waiting time reduction is associated with a decrease in the probability of 0.026 percentage points. The estimated probability of death in accidents with MRT = 25 min is 8.0% in our database. Thus, a 0.026 percentage points decrease means that the probability of death goes from 0.080 to 0.054. This represents a decrease of 32%. These results show the relationship between low medical response times and lower probability of death in a crash. In our sample, the estimated average probability decrease for a 10 min response time reduction is approximately one third in both motorway and conventional road accidents. 5.3. Control variables With respect to the coefficients on the control variables included in the models, the following comments are worth mentioning: 1. The most relevant factors explaining the probability of death are associated with individual-related variables, and more specifically, with the lack of use of safety measures (seat belts or other safety elements). In those accidents in which the driver or the passengers do not use any of these safety devices, the probability of death is significantly higher. This finding is consistent with results in García-Ferrer et al. (2007), who found a permanent effect in reducing fatality rates after the introduction of the mandatory use of seat belts in Spain in 1992. The importance of using safety belts and other security measures has also been reported in several works for other countries, like Evans (1987) or Bédard et al. (2002), among others. 2. Driving under the influence of alcohol or other drugs significantly increases the probability of death in motorway accidents. Surprisingly, we did not find a significant effect in conventional road accidents. This lack of statistical significance might be related to the correlation between Alcohol/Drugs and Nighttime.13 This mild collinearity may cause that part of the effect of Alcohol/Drugs is picked up in Nighttime. Since the correlation is higher on conventional roads than on motorways, this could yield to the non-significant effect found on conventional roads. Some studies focusing on the effect of alcohol consumption and fatality risk have found a mixed evidence. See, for instance, Waller et al. (1986) or Bédard et al. (2002). 3. Among the accident-related factors, the most statistically relevant variables are Outward/return journey and Sunday. They show opposite sign effects on the probability of death in the crash. The variable Outward/return journey is related to the purpose of the trip. According to their purpose, accidents can be classified into three main categories in our dataset: work-related crashes (those that occur while working or commuting to work), leisurerelated accidents that occur in the outward trip or the return trip from a public holiday, and other leisure-related accidents. In both types of roads, the probability of death is significantly higher in trips related to outward/return journeys. The day of the week can also play a role. We find that the probability of death is significantly lower in accidents occurred on Sunday. There are

13 The correlation matrices of the variables included in the estimated models are shown in Appendix C, for both motorway and conventional road accidents.

several papers in the literature that analyze the influence of the day of the week on fatal or injury road accidents. The reported evidence is mixed. Doherty et al. (1998) find evidence of higher accident risk in weekends, especially for young drivers. Kumar et al. (2008) do not find significant differences among weekdays and weekends in the fatality risk. 4. Having a run-off-the-road accident increases the probability of death on motorways and decreases that probability on conventional roads. Accidents are mainly divided into two types in our data: collision with other vehicles on the road, and run-offthe-road accidents followed by rollover or collision with a fixed object. Given the characteristics of motorways and conventional roads, a run-off-the-road accident is more likely on motorways, whereas a collision with other vehicles is more frequent on conventional roads. Of course this variable is partly picking up the effect of the number of vehicles involved in the accident. Several works in the literature have addressed the influence of the type of accident on its severity. Examples are Krull et al. (2000), who find that rollovers increase the driver’s probability of death, or Holdridge et al. (2005), who analyze this probability in accidents involving different roadside fixed objects. 5. Driving speed significantly increases the probability of death on conventional roads, but not on motorways. One possible explanation for the non-statistical significance of this variable on motorways is that part of its effect can be picked up by the variable Run-off-the-road. Another possible explanation could be the higher quality standard of motorways, which lowers the effect of speed excess driving. This result partly relates to Aarts and van Schagen (2006). They analyze driving speed focusing on the crash risk. They find the effect of an increase in speed on crash rates is lower in major roads than in minor roads. A third possible explanation for the non significance of the speed variable is its potential lack of informative power. We do not have accurate information on the continuous variable mph (miles per hour), which would allow us to better measure its effect. We just have the binary indicator of speed excess vs. normal speed. 6. The probability of death is higher for those accidents happening at night. It could be argued that at night accidents MRT is higher. However, the correlation between both variables is quite low (0.056 on motorways and 0.076 on conventional roads). The correlation is higher between Nighttime and Alcohol/Drugs, as it was pointed out above. In fact, the influence of Nighttime on fatal accidents might be masking the effect of alcohol/drugs abuse. This is consistent with the findings in González-Luque and Rodríguez-Artalejo (2000), who examined a sample of fatal crashes and found a correlation between nighttime and alcohol consumption. With respect to other accident-related variables, we did not find significant effects on the probability of death. Among these nonsignificant variables, we can cite climate conditions, traffic density and the number of vehicles involved in the accident. All accidents in our dataset occurred in May. Thus, the climate conditions do not show as much variability as if we had data from all seasons. This low variability could explain the non-significance of this variable. As regards the traffic density and the number of involved vehicles, their effect could be partly picked up by the variable referring to the speed excess. With respect to other individual-related variables, we did not find significant effect of being the driver of the vehicle. Neither we found significant effects of driver’s gender, disability or the number of non-stop driving hours. The passenger characteristics, mainly gender and age, also appeared as non-significant in both models. We also investigated interaction terms: we considered drivers in different age intervals, and also being a young driver (< 25 years

G Model AAP-2049;

No. of Pages 9

ARTICLE IN PRESS R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

old) under the effect of alcohol/drugs or being a male driver. None of the coefficients were statistically significant. Our data have some limitations. As it is explained, the presence of collinearity among some of the variables prevents us to isolate their effects. But, interestingly, our main results on the association of the medical response time and the probability of death are robust to these alternative specifications, both in terms of the statistical significance and the magnitude of the estimated coefficients.14 6. Conclusions We have analyzed the probability of death in road traffic accidents in Spain, where more than 3000 people die each year due to a crash. When public authorities or other institutions try to explain the number of fatalities, attention is mainly focused on driving behavior and road conditions. However, there is another factor that may be critical to reducing this number: a quick medical response. In this paper, we have quantified to which extent a reduction in the time interval between the crash and the arrival of the emergency services to the crash scene could potentially be associated with lower fatality ratios. To do so, we have linked two databases on road accidents occurred in Spain in May 2004. The first one contains information on road and vehicle conditions, driving behavior and individual characteristics of the driver and passengers. The second one provides, for the first time in Spain, medical response time information. We have estimated probit models on the outcome of a crash (fatality vs. non-fatality) on both motorways and conventional roads. Besides the medical response time, we have included some control variables related to the crash and victims’ characteristics. Our findings suggest a relationship between low medical response times and lower probability of death. More specifically, a reduction from 25 to 15 min could be statistically associated with a significant average decrease of this probability by one third on both motorways and conventional roads. Our results are in line with other studies that point out the importance of emergency medical care issues to reduce mortality rates. Our research has limitations. Although the dataset we use is the first Spanish microdata with exhaustive information on accident and victims that also reports the medical response time, our results should be taken with caution since they are based on data from May 2004 only. The cross-sectional nature of our data prevents us from inferring any conclusion on a dynamic setting. It would be desirable to have more recent and past data that allow to confirm our findings for the May 2004 sample. There is a planned project in the DGT to creating a new dataset on recent accidents with comparable information with the one used in this paper. Part of our future research agenda is to access and exploit the new data if the DGT project is finally undertaken. It would be of great interest to analyze in a dynamic context whether there have been changes in the medical response time or any other concurring circumstances in a traffic accident and what, if any, has been the impact on the fatality ratios. Acknowledgments We want to thank Laura Romeu-Gordo, Patricia Cubi-Molla and Elías Moreno for their help with the manuscript. We also acknolewdge comments received from the participants at the International Symposium on Forecasting (2008), Simposio de Análisis Económico (2008) and seminar participants at Universidad

Estimation results of alternative speciﬁcations are available from the authors upon request.

de Vigo (2009). We thank the editor, the associate editor and two anonymous referees for helpful comments on a previous version of this paper. This research was suppported by projects CCG07-UAM/HUM-1918 (Universidad Autónoma de Madrid and Comunidad Autónonoma de Madrid) and SEJ2006-04957/ECON (Spanish Ministry of Education). We also thank DGT (Dirección General de Tráfico) for providing us the data we use in this research. Appendix A. The linkage process The data used in this paper are the result of the linkage of two datasets conducted by the DGT. The first dataset contains information on all accidents occurred in Spain in 2004. In that year there were 94,009 accidents with 143,124 victims, including 4741 fatalities, 21,805 seriously injured victims and 116,578 slightly injured victims. Almost half of those accidents occurred on toll/free motorways or conventional roads (43,787 accidents on roads and 50,222 in urban areas). The distribution of fatalities was as follows: 3841 on roads and 900 in urban areas. The information is structured in three different files regarding accidents, vehicles and victims characteristics, respectively. In the accidents file, there are as many records as accidents. In the vehicles file, there is one record for each one of the vehicles involved in an accident. In the victims file, there is one record for each one of the victims involved in each accident. The pair accident-province uniquely identifies accidents in each of the files. This identifier has been used to merge the three files. This has produced a general dataset with all the information on accidents, vehicles and victims on Spanish roads in 2004. Despite the exhaustiveness of these data, there was no information on the medical response time. In May 2004, the DGT conducted a pilot survey to get information on this variable. This study, conducted in all Spanish regions, but Catalonia and the Bask Country, contained information on 2018 road accidents. But in this file, the identifier did not match with the identifier in the general dataset. Thus, it was necessary to match the information from both files (the general dataset and the pilot survey) on the basis of the province, day, mile point, time of the accident and number of victims. As a result of the linkage process and after dropping out some observations for which we could not match the information from both Table B1 Road accidents characteristics (percentages in each category). Characteristics

Motorways

Conventional roads

Commuting a Related to work Related to leisure

27.1 52.9

26.0 60.8

Working day Good road conditions b During the daytime b Good weather conditions b Low trafﬁc density b

47.6 66.9 77.6 65.3 88.4

47.6 75.4 72.2 76.8 96.7

Type of infraction a Alcohol/drugs Speed excess

2.4 30.7

3.4 28.5

Type of accident a Collision with other vehicles Run-off-the-road

35.0 65.0

47.7 51.7

Vehicle age < 1 year 2–5 years > 5 years

17.7 44.5 37.8

15.3 33.2 51.5

a Variables with more than two categories for which only the most frequent values are shown. Thus, the reported percentages do not add up to 100%. b Binary variables, values yes (= 1), no (= 0). Proportion of 1’s is shown.

ARTICLE IN PRESS

G Model AAP-2049;

No. of Pages 9

R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

Table B2 Victims characteristics (percentages in each category.) Characteristics

Motorways

Conventional roads

Drivers

Males

Passengers

Drivers

Passengers

FAT

INJ

FAT

INJ

FAT

INJ

FAT

INJ

89.0

80.1

54.2

42.9

91.2

81.5

38.1

45.5

Age (years) < 18 18–24 25–44 45–64 65–74 > 74

0.0 15.4 46.1 30.7 3.9 3.9

0.2 17.8 55.9 22.3 2.8 1.0

4.0 16.0 36.0 24.0 8.0 12.0

12.0 27.5 30.7 18.8 5.2 5.8

0.0 22.8 47.3 19.3 5.3 5.3

2.5 19.2 49.2 20.6 5.6 2.9

4.8 28.6 19.0 23.8 9.5 14.3

14.3 25.9 31.1 16.9 5.5 6.3

Use of safety measures a

68.0

90.0

48.0

81.9

58.9

85.9

61.9

81.7

Driving experience (years) ≤1 2–4 5–10 > 10

4.4 13.0 30.4 52.2

11.6 14.5 24.0 49.9

3.9 21.1 36.5 38.5

12.8 18.0 19.5 49.7

Good psychophysic conditions a , b

94.1

94.0

85.7

95.5

Non-stop driving hours <1 1–3 >3

41.2 47.0 11.8

53.1 38.4 8.5

87.5 8.3 4.2

85.9 12.0 2.1

Notes: FAT: fatalities; INJ: injured people. a Binary variables, values yes (= 1), no (= 0). Proportion of 1’s is shown. b Good psychophysic conditions: no presence of alcohol, drugs, fatigue or sleepy feeling. Table C1 Correlation matrix for motorway accidents. Obs= 557

MRT

Nighttime

Outward/return

Sunday

Run-off-the-road

Speed excess

Not safety

Alcohol/drugs

Driver’s age

MRT Nighttime Outward/return Sunday Run-off-the-road Speed excess Not safety Alcohol/drugs Driver’s age

1 0.056 −0.032 0.004 0.049 0.052 −0.021 −0.059 −0.009

1 −0.065 −0.097 −0.023 −0.058 0.150 0.107 −0.143

1 0.136 0.145 0.041 −0.104 −0.051 −0.008

1 −0.012 −0.060 0.020 −0.083 −0.009

1 0.200 0.060 −0.189 −0.059

1 −0.104 −0.105 −0.027

1 0.004 −0.041

1 0.021

files, we get a final dataset of 1463 crashes with one-to-one match, where 2264 vehicles and 3287 people were involved (2541 of them were casualties). Appendix B. Accident and victim characteristics The main features of the accidents in our dataset are shown in Table B1. Victims characteristics are reported in Table B2. Both of them offer separate information for motorways and conventional roads. As we can see in last column of Table B1, most accidents were related to leisure and they occurred during the daytime, with good weather and road conditions, and with low traffic density. The

most frequent types of infraction were related to driving under the effects of alcohol or other drugs and exceeding the speed limit. The most frequent accident on motorways was run-offthe-road, with almost twice the frequency of collision with other vehicles. The difference was not so pronounced on conventional roads. As regards to the victims, Table B2 shows that more than 80% of dead or injured drivers were male, while this percentage decreases almost by half when we refer to passengers. With respect to the use of safety measures, it is remarkable that on both types of roads and irrespective of being driver or passenger, the percentage of those victims using safety measures was much lower among the fatalities than among the injured ones.

Table C2 Correlation matrix for conventional road accidents. Obs= 992

MRT

Nighttime

Outward/return

Sunday

Run-off-the-road

Speed excess

Not safety

Alcohol/drugs

Driver’s age

MRT Nighttime Outward/return Sunday Run-off-the-road Speed excess Not safety Alcohol/drugs Driver’s age

1 0.076 0.012 −0.082 0.049 0.009 0.085 0.001 −0.076

1 0.042 0.144 0.124 0.007 0.134 0.141 −0.122

1 −0.015 −0.009 −0.034 0.008 −0.023 −0.020

1 0.038 −0.057 0.060 −0.060 −0.079

1 0.292 0.082 0.015 −0.069

1 −0.006 −0.031 −0.103

1 0.046 −0.075

1 −0.028

G Model AAP-2049;

No. of Pages 9

ARTICLE IN PRESS R. Sánchez-Mangas et al. / Accident Analysis and Prevention xxx (2009) xxx–xxx

Appendix C. Correlation matrices Tables C1 and C2 show the correlation matrices for the variables included in the estimated models in Table 4. The correlation matrix for motorways is reported in Table C1. For conventional roads, it is shown in Table C2. References Aarts, L., van Schagen, I., 2006. Driving speed and the risk of road crashes: a review. Accident Analysis and Prevention 38, 215–224. Bédard, M., Guyatt, G., Stones, M.J., Hirdes, J.P., 2002. The independent contribution of driver, crash, and vehicle characteristics to driver fatalities. Accident Analysis and Prevention 34, 717–727. Bentham, G., 1986. Proximity to hospital and mortality from motor vehicle traffic accidents. Social Science and Medicine 23 (10), 1021–1026. Boufous, S., Williamson, A., 2006. Work-related traffic crashes: a record linkage study. Accident Analysis and Prevention 38, 14–21. Brodsky, H., 1990. Emergency medical service rescue time in fatal road accidents. Transportation Research Records 1270, 89–96. Brodsky, H., 1992. Delay in ambulance dispatch to road accidents. American Journal of Public Health 82, 873–875. Brodsky, H., 1993. The call for help after an injury road accident. Accident Analysis and Prevention 25, 123–130. Charlton, R., Smith, G., 2003. How to reduce the toll of road traffic accidents. Journal of the Royal Society of Medicine 96, 475–476. Clark, D.E., Cushing, B.M., 2002. Predicted effect of automatic crash notification on traffic mortality. Accident Analysis and Prevention 34, 507–513. Coats, T.J., Davies, G., 2002. Prehospital care for road traffic casualties. British Medical Journal 324, 1135–1138. DGT, 2004. Informe Anual y Estudio Mayo 2004. Dirección General de Tráfico, Min˜ isterio del Interior, Gobierno de Espana. Doherty, S.T., Andrey, J.C., MacGregor, C., 1998. The situational risks of young drivers: the influence of passengers, time of day and day of week on accident rates. Accident Analysis and Prevention 30, 45–52. Durkin, M., McElroy, J., Guan, H., Bigelow, W., Brazelton, T., 2005. Geographic analysis of traffic injury in Wisconsin: impact on case fatality of distance to Level I/II trauma care. Wisconsin Medical Journal 104, 26–31. Elvik, R., Mysen, A.B., 1999. Incomplete accident reporting: meta-analysis of studies made in 13 countries. Transportation Research Record 1665, 133–140. European Commission, 2009. eCall: saving lives through in-vehicle communication technology. Factsheet 49. Information Society and Media. Evanco, W.M., 1999. The potential impact of rural mayday systems on vehicular crash fatalities. Accident Analysis and Prevention 31, 455–462. Evans, L., 1987. Fatality risk reduction from safety belt use. The Journal of Trauma 27, 746–749. García-Ferrer, A., de Juan, A., Poncela, P., 2007. The relationship between road traffic accidents and real economic activity in Spain: common cycles and health issues. Health Economics 16, 603–626.

González-Luque, J., Rodríguez-Artalejo, F., 2000. The relationship of different socioeconomic variables and alcohol consumption with nighttime fatal traffic crashes in Spain: 1978–1993. European Journal of Epidemiology 16, 955–961. González, R.P., Cummings, G.R., Phelan, H.A., Mulekar, M.S., Rodning, C.B., 2009. Does increased emergency medical services prehospital time affect patient mortality in rural motor vehicle crashes? A statewide analysis. The American Journal of Surgery 197, 30–34. Haegi, M., 2002. A new deal for road crash victims. British Medical Journal 324, 1110. Holdridge, J.M., Shankar, V.N., Ulfarsson, G.F., 2005. The crash severity impacts of fixed roadside objects. Journal of Safety Research 36, 139–147. Hussain, L.M., Redmond, A.D., 1994. Are pre-hospital deaths from accidental injury preventable? British Medical Journal 308, 1077–1080. Jones, A.P., Bentham, G., 1995. Emergency medical services accessibility and outcome from road traffic accidents. Public Health 109, 169–177. Krull, K.A., Khattak, A.J., Council, F.M., 2000. Injury effects of rollovers and events sequence in single-vehicle crashes. Transportation Research Record: Journal of the Transportation Research Board 1717, 46–54. Kumar, A., Lalwani, S., Agrawal, D., Rautji, R., Dogra, T.D., 2008. Fatal road traffic accidents and their relationship with head injuries: An epidemiological survey of five years. Indian Journal of Neurotrauma 5, 63–67. Li, M.D., Doong, J.L., Chang, K.K., Lu, T.H., Jeng, M.C., 2008. Differences in urban and rural accident characteristics and medical service utilization for traffic fatalities in less-motorized societies. Journal of Safety Research 39, 623–630. Muelleman, R.L., Mueller, K., 1996. Fatal motor vehicle crashes: variations of crash characteristics within rural regions of different population densities. The Journal of Trauma: Injury, Infection and Critical Care 41, 315–320. Noland, R.B., 2003. Medical treatment and traffic fatality reductions in industrialized countries. Accident Analysis and Prevention 35, 877–883. Noland, R.B., Quddus, M.A., 2004. Improvements in medical care and technology and reductions in traffic-related fatalities in Great Britain. Accident Analysis and Prevention 36, 103–113. Page, Y., 2001. A statistical model to compare road mortality in OECD countries. Accident Analysis and Prevention 33, 371–385. Rosman, D.L., 2001. The Western Australia Road Injury Database (1987–1996): ten years of linked police, hospital and death records or road crashes and injuries. Accident Analysis and Prevention 33, 81–88. UNECE, 2000. Statistics on Road Traffic Accidents. United Nations Economic Commission for Europe, Transport Division. UNECE, 2004. Statistics on Road Traffic Accidents. United Nations Economic Commission for Europe, Transport Division. Waller, P.F., Stewart, J.R., Hansen, A.R., Stutts, J.C., Popkin, C.L., Rodgman, E.A., 1986. The potentiating effect of alcohol on driver injury. Journal of the American Medical Association 256, 1461–1466. Wooldridge, J.M., 2003. Cluster-sample methods in applied econometrics. American Economic Review 93, 133–138. Zwerling, C., Peek-Asa, C., Whitten, P.S., Choi, S.W., Sprince, N.L., Jones, M.P., 2005. Fatal motor vehicles crashes in rural and urban areas: decomposing rates into contributing factors. Injury Prevention 11, 24–28.