First, a descriptive analysis of head-on crashes was carried out using frequency distribution of road characteristics- and crash circumstances-related variables. Second, a descriptive analysis of the drivers involved on head-on crashes was carried out of driver characteristics and driver risk behaviour-related variables. Finally, crash-related factors associated to head-on crashes and, driver-related factors associated to be involved in a head-on crash, versus other type of crashes, were studied estimating Proportion Ratios (PR) and their 95% Confidence Intervals (95%CI). The most appropriate method to estimate PR would be Log- Binomial Regression method, however this model does not converge due to the presence of discrete explanatory variables with many categories. An alternative to estimate PR would be adjusted Robust Poisson regression models using generalized linear models (GLM) with a Poisson distribution and a log link. In addition, we used a robust method for estimating the variance of the errors, to avoid overestimation of this variance and obtain better estimates of PR.9,10 Models for men and women drivers were fitted separately.

The correlations between the explanatory variables were assessed using Spearman bivariate correlations and multicollinearity was analyzed by estimating the Variance Inflation Factor. We excluded from the final model traffic flow, post delineator and road markings because they have high multicollinearity with other variables of infrastructures such paved shoulder, medians and safety barriers. We removed also variables related to the cause of the crash, as some of them have more than 90% missing values and some of them refer to attitudes of the driver such alcohol or drug use, or inadequate speed, that were included in the models of drivers. Missing categories are included in the model in order to maintain statistical power, as some variables have a quite proportion of missing values. Only significant variables are shown in the final model.

Between 2007 and 2012, 130,454 crashes occurred on Spanish two-way interurban roads, 7% of which were head-on crashes (n=9,192) in roads with a lane width equal to or greater than 3.25 meters. Two vehicles were involved in 90.6% of head-on crashes, 3 vehicles in 8% and between 4 and 8 in 1.4%. In 14.4% of head-on crashes there was a death and in 36.5% a severely injured, while in non head-on crashes these proportions were 3.5% and 16.6% respectively.

With regard to the road infrastructure, 58.7% of head-on crashes occurred in roads 7 meters wide or more, 42.1% occurred on a straight lane, 19.5% on an intersection, 19.3% on a smooth curve, and 10.2% on a strong curve with a speed signal (Table 1). 77.8% of head-on crashes occurred with fluid traffic density, 68.4% with daylight, 77.1% on a dry and clean surface, 80.2% with good weather conditions. In addition, 71.8% occurred on roads where road markings only defining lanes and edges, compared with 1.5% of crashes that occurred on roads where the marks were erased or nonexistent. Most head-on crashes occurred on roads without any medians (98.7%), safety barriers (65.6%), or delineator posts (61.4%). At least a car was involved in 93.6% of head-on crashes, a truck or a van in 29.0%, a motorcycle in 8.6% and a moped in 4.1% (Table 1).

This paper shows, as far as the authors know, for the first time an analysis of head-on crashes on two-way interurban roads in Spain, including national data. Factors related to road infrastructure such as roads of 7 meters width or more and the presence of curves, slope changes, road narrowing, a side safety barrier, twilight, wet and snowy surface conditions were identified as characteristics with higher probability of head-on crashes comparing to other type of crashes. On the other hand, when there are medians to separate opposite lanes, intersections and paved shoulder there is a lower likelihood of head-on crashes with respect to the other types of crash. Men driver patterns, driving in the opposite direction or incorrectly overtaking, employed, going to or returning from work were associated with an increased probability of head-on crashes compared to the other crashes. Women driver patterns, during working hours, driving in the opposite direction or incorrectly overtaking were associated with a higher likelihood of head-on crashes.

The major strength of the study is that due to the availability of a comprehensive database from the National Traffic Authority for six years, which includes an important number of crashes occurred at national level. It allowed studying not only factors related to the environment and conditions of the crash but also those related to the drivers involved.

However it must be highlighted as a limitation that the design of the study allows knowing the factors associated to head-on crash compared to factors associated to other types of crashes. There could be some factors associated to head-on crashes that are not seen because they are associate as well with other types of crashes. However the results observed are consistent with the literature. It has not been possible to identify factors related to head-on crashes comparing with no crashes because there is no available data when there is no crash. In any way, this design is commonly used to study disease mortality patterns by cause in settings where population denominators are not available.10

Regarding road infrastructure, our results coincide with others in the literature revealing that the presence of curves, whether soft or marked, the narrowing of the road, slope changes and side barriers are factors associated with an increased risk of head-on crashes.2,6 On the other hand, the existence of paved shoulders, and intersections reduces the likelihood of head-on crashes with respect to the other types of crash. Some authors found that paved shoulders and intersections are factors that contribute to a lower number of head-on collisions, but on the other side are factors associated to higher severity.6,7

Matena et al.2 reported that slippery roads due to rain or winter time are related to head-on crashes. Our study confirms this relationship, as a wet or snowy surface has been associated with head-on crashes compared to other type of crashes. In addition, driving during twilight was identified as a factor associated with an increased risk for head-on crashes. This could be due to the decreased visibility at that hour of the day, when factors related to an increased likelihood of a head-on impact such as glare from the sun or presence of fog may be present.11

Regarding driver related factors, several studies have associated head-on crashes to factors such as inattention of drivers resulting from fatigue or drowsiness, or inappropriate speed, especially on road stretches with curves, and alcohol or drug use.2,7 In our study, crashes caused by distraction or inappropriate speed occurred in almost half of head-on crashes; however, they are associated with a lower probability of head-on crash comparing to other types of crash. We found other counterintuitive results such lower risk of head-on crash if not respecting regulations or use of alcohol. This may be due to the fact that those studies analyzed factors related to head-on crashes, regardless of factors related to other types of crash. The fact that some tracks favour quick driving or distraction is a factor related to any type of crash and not specifically to head-on crashes. On the other hand incorrectly overtaking or driving in the inappropriate direction are obvious factors specifically related to head-on crash.

It must be highlighted that only among men drivers, being a professional driver and travelling to and from work are factors related to an increased probability of head-on crashes compared to other types of crashes. Shariat-Mohaym et al. reported a conflicts of head-on crashes occurred more often on workdays than on holidays.5 In Spain according to a report from DGT,12 men were injured three times more likely than women going to and from work, and twelve deaths of men occurred for each of the women in crashes in the normal working day. This points to men workers as a work related crashes risk group, and could be due to the different occupational exposure by gender, being men those who are more likely to work in transportation sector. This is also coherent with our results in the sense that only among men being a bus or truck driver is a factor associated to head-on crashes.

Finally, our results showed a lower likelihood of head-on crash when there is a median. This reinforces that implementing systems to separate lanes in two-way interurban roads could avoid severe crashes such are the head-on. A number of studies in other countries have demonstrated the effectiveness of the installation of measures of central separation between opposing lanes: the installation of rumble strips between opposite lanes of rural roads has been associated with a 25% reduction of frontal crashes,13 the installation of flexible barriers (cable type) has been associated with a 56% reduction of head-on crashes with serious injuries14 and installing concrete medians, guardrails or raised medians has been linked to a reduction of overall crashes15,16 and specifically to head-on crashes.17,18 As recommended by Hosseinpour et al.6 the installation of centreline median barriers on undivided roadways seems to be the most effective counter-measure; however, due to the great expenses associated with the installation and maintenance of continuous median barriers, for lower risk locations, centreline rumble strips could be effectively installed to reduce crossover collisions.