METHODS

The study design is based on a specific application of the induced exposure technique ^10,11 according to a method similar to that used in a 1991 study by Perneger and Smith.¹² It is a retrospective, paired case-control study with a variable number of controls per case. The data we analysed were from the DGT registers of traffic accidents with victims. These registers contain information from the traffic accident reports that the Spanish police are required to file for every crash with victims. For the period of study this register contains data from 895 949 traffic accidents with victims, 46 473 (5.19%) of which resulted in at least one death. The accidents involved 1 527 351 drivers, 79 162 (5.18%) of whom were non-Spanish or for whom no nationality was recorded. One of the variables we recorded was which of 20 possible traffic violations (if any) was committed (see appendix 1) by each of the drivers involved in the accident. The type of infraction was used to classify drivers as infractors or non-infractors. We selected only those collisions that involved two or more vehicles in motion with four or more wheels, and in which only one of the drivers involved was considered an infractor. Therefore, of the total of 519 227 collisions involving at least two vehicles registered in the DGT database, we excluded those that involved a stopped or parked vehicle or two wheeled vehicle, those in which none or more than one of the drivers was considered an infractor, and those in which the nationality of one or more of the drivers was not stated.

The rationale of the study was as follows: in those collisions in which only one driver committed an infraction, this driver can be considered responsible for the collision. The involvement of the other drivers (non-infractors) can be assumed to be passive, and these drivers can be considered a representative sample of all drivers on the road at the time of the collision. Hence the responsible drivers (infractors) constituted the case group. For each case the non-responsible driver(s) involved in the same collision were used as paired controls. (Many cases had only 1 control, others had 2 or more.) The final sample comprised 226 168 responsible drivers involved in as many collisions (representing 25.24% of all accidents registered in the DGT database, and 43.56% of all collisions), and 263 841 non-responsible drivers. Of all collisions studied, 11 694 (5.17%) resulted in at least one death.

For each driver we recorded the following variables, which were coded in the DGT database: nationality (Spanish, Portuguese, French, Moroccan, German, British, Italian, Swiss, Belgian, Dutch, American (USA), other), sex, age, psychophysical circumstances (normal, under the influence of alcohol as documented by a positive breath test, under the influence of other drugs, sudden illness, sleepiness or drowsiness, other), administrative infraction (none, driving licence expired, motor vehicle inspection certificate expired, other), speed related infractions (none, inappropriate speed for the road or whether conditions, speeding, slow driving that interfered with traffic), physical disabilities (none, sight, hearing, upper limbs, lower limbs, other), years in possession of driving licence, years since the vehicle was registered for driving on public roads, type of driver (professional or non-professional), and safety belt use at the time of the accident (included as a surrogate measure of safety awareness). The month in which the accident occurred, and whether the accident occurred within city limits or on the open road was also recorded. Age was stratified into five year intervals with the exception of the 18 year old to 24 year old group (18 years is the minimum legal driving age in Spain) and the >74 year old group (the sample contained very few women drivers aged >74 years).

To estimate odds ratios with a 95% confidence intervals (CI) for each nationality (with Spanish nationality as the reference), we analysed the findings with conditional logistic regression,¹³ using responsible versus non-responsible driver as the dependent variable. Crude (cOR) and adjusted odds ratios (aOR) were calculated, the latter by including in the model all of the remaining driver related and vehicle related variables at once. None of the variables could be removed from the model according to the pre-established p to remove value (p>0.05). The results were stratified according to the period in which the accident occurred (from June to September inclusive or during the rest of the year), and whether the collision occurred within city limits or on an open road. All analyses were done with the STATA statistical package (version 5.0).¹⁴

RESULTS

Of all drivers involved in the sample of collisions we studied, 4.06% were foreign drivers. For this group, the cOR for causing a collision, in comparison with Spanish drivers, was 1.55 (95% CI: 1.50 to 1.60). After adjustment for all other driver characteristics, the aOR was slightly lower: 1.42 (95% CI: 1.35 to 1.49). Table 1 shows the distribution of drivers involved in collisions according to responsibility and nationality. The most frequent non-Spanish nationality was French (0.72%), followed by German (0.55%), Moroccan (0.52%) and Portuguese (0.51%). Table 1 also summarises the estimated cOR and aOR for each nationality. In the crude analysis all ORs except the estimate for Belgian drivers were significantly greater than 1, with values ranging from 2.06 (British) to 1.19 (Portuguese). Adjusted ORs were slightly lower than crude estimates except for British drivers; in this subgroup the aOR was 2.16. Despite the slightly lower associations in comparison with cORs, the aORs still showed significant associations between nationality and risk of causing a vehicle collision for all nationalities except Italian, Belgian, and American (USA).

Table 2 shows the aORs stratified for type of road where the collision took place (urban area or open road). The aORs for collisions that occurred in cities or towns were higher than the estimates for crashes on open roads for all nationalities except Swiss and Dutch, although the reduction of sample size resulted in a loss of statistical significance for several aOR in urban areas. The highest aORs on open roads were found for British drivers (2.15). For accidents within city limits, high aORs were also found for British drivers (2.35), and for drivers from France (2.24), Morocco (2.38), and especially from the USA (4.78).

Table 3 shows the aORs stratified for the period when the collision occurred (summer versus rest of the year). The magnitude and statistical significance of the aORs for each country were similar for both periods, and no general pattern of excess risk in association with the summer season was evident. Only Dutch drivers had an aOR that was appreciably higher during the holiday season (1.73) than during the rest of the year (1.24).

DISCUSSION

Our results show that vehicle collisions in Spain during the period from 1990 to 1999 were more likely to have been caused by non-Spanish than by Spanish drivers. The difference was especially clear for crashes that took place within city limits. The possible explanations for our findings are discussed below.

Influence of driver related characteristics

Drivers from other countries who are at the wheel of motor vehicles on Spanish roads may not be representative of the total population of drivers in their respective countries. If this were shown to be true, it would be difficult to evaluate the influence of possible selection biases, as the baseline risk of causing a collision while driving in their country of origin is unknown. In particular, it seems evident that in Spain, many foreign drivers are tourists whose behaviour behind the wheel, perhaps in association with a greater risk of causing a collision, might be notably different from that of the rest of the drivers from the same country. However, when the data were stratified for season of the year (summer holiday versus rest of the year), no significant difference was found between the two periods.

Drivers may change their driving habits when they are abroad. This factor seems to have been influential for British drivers, who are used to driving in the left rather than the right lane, and for Moroccan drivers residing in central and northern Europe, most of whom drive across Spain from north to south or vice versa with very little time for rest stops on their way to Morocco for their summer holiday or on their way back to their place of habitual residence. It is not surprising that the cORs for British and Moroccan drivers were higher than for all other nationalities, and that the adjusted OR for Moroccan drivers was much lower than the crude OR when the analysis was adjusted for psychophysical circumstances.

If most foreign drivers in Spain are assumed not to reside habitually in this country, unfamiliarity with the Spanish driving milieu can undoubtedly be an important factor in their higher propensity to cause collisions. This assumption is supported by the generally higher aORs for city driving, where unfamiliarity with local driving conditions is greater.

Influence of classification bias

Another possible explanation for the higher risk of causing a vehicle collision among non-Spanish drivers is bias in the ascription of responsibility for the crash: the police agents responsible for preparing the accident report may be more likely to blame foreign rather than Spanish drivers. However, when we stratified the analysis according to type of infraction, the magnitudes of the OR remained mostly unchanged in relation with such straightforward infractions as driving in the opposite lane or disobeying a traffic light (data not shown).

Methodological considerations

Some methodological considerations need to be discussed. Firstly, we assumed that the drivers identified in the DGT register as infractors were responsible for the accident they were involved in. Although committing traffic violations while driving is closely related with the risk of involvement in an accident,^15,16 committing a traffic violation does not usually lead to an accident under actual driving conditions. However, for a given collision involving one infractor and one or more other drivers who committed no infraction, the chances that the infractor was responsible for the accident can safely be assumed to be much greater than the chances that a non-infractor was responsible. We therefore assumed that most of the infractors in our sample were responsible for the accident, and that most of the non-infractors were not responsible. Evidently, the bias arising from the small proportion of misclassified drivers (that is, non-responsible infractors and responsible non-infractors) would tend toward the null, hence the overall effect would be to underestimate the association between being an infractor and the risk of causing an accident. Indirect support for the validity of our design comes from the finding that the associations we found between risk of causing an accident and each of the other covariables were consistent with the results of earlier studies (data not shown).

This study, based as it is on the induced exposure method, is subject to the advantages and disadvantages of this approach ¹⁷ The main advantage of the design developed by Perneger and Smith ¹² is that by pairing drivers according the collision they were involved in, the confounding effect of the main environmental factors is controlled for. With respect to its limitations, the induced exposure method reportedly underestimates exposure in more careful drivers—that is, those who use a particularly conservative driving strategy and are therefore passively involved in collisions less frequently than their actual exposure would warrant.¹⁸ This may be the case for foreign (non-Spanish) drivers who use an especially conservative driving style when they are abroad (or at least when they are in Spain).

A final note of caution is in order regarding the adjusted analysis. Our analysis was done in a manner that controlled for the effect of the main driver related and vehicle related covariables (age, psychophysical circumstances, speeding, etc). This approach was used to reduce, as far as possible, confusion in the association between nationality and risk of causing a collision. However, because some data for some of the covariables included in the model were missing, the adjusted analysis comprised only 64% of the initial sample of drivers. In an attempt to elucidate whether this lack of data might have introduced selection bias in our adjusted estimates, we obtained crude estimates of the effect of nationality in the subsample of drivers not included in the adjusted analysis. The estimates were quite similar to those obtained for the whole sample, with a tendency toward greater cORs. For foreign drivers considered globally, the cOR for the subsample of excluded drivers was 1.69, a figure 9% higher than that obtained for the whole sample (1.55). When the cOR was estimated for each nationality, 8 of the 11 estimates also yielded values slightly higher than the corresponding values obtained for all drivers. This result suggests that the aORs obtained in our study might have been affected by a slight bias towards the null.

In conclusion, the results of this study should not be taken to imply that the intrinsically poor driving habits of foreign drivers in Spain place them at greater risk for causing a vehicle collision than Spanish drivers. The difference in risk is in fact probably attributable to the combined effect of some or all of the factors discussed above. None the less, both the authorities responsible for traffic safety and all other drivers on Spanish roads should consider foreign drivers a special risk group, especially within city limits. On the basis of a simple calculation of the per cent population attributable risk (estimated from a global aOR of 1.42 for foreign drivers, and a proportion of exposed (foreign) cases (responsible drivers) of 4.8%), approximately 1.4% of the collisions we analysed can be attributed to foreign drivers in Spain. On the other hand, it seems obvious that given the same driving environment, the risk of Spanish drivers causing a collision is no greater than that of drivers of other nationalities.