Resultados de descriptivos funcionales y competencias Método MPC

Meta-analyses’ results are useful for identifying general data patterns, but are not accurate enough for predicting the effect of a speed limit increase on a particular road network as they cannot take into account the area-specific characteristics (geographic, cultural etc.) that may differentiate the outcomes. Consequently, to predict the impact of a potential speed limit increase on a road network it is necessary to define its accident-speed rela- tionship. As has already been discussed, the majority of the before and after studies report proportional changes in accident frequency following speed limit alterations (e.g. Elvik et al., 2004; Aarts and Van Schagen, 2006). This effect is always attributed to the increase of average speeds on the roadway; however, the individual examination of the relationship of speed with accidents does not always support this idea.

Based on the amount of the kinetic energy that is released during a collision (EKinetic = mV2

2 ), accidents that occur under high speed conditions are definitely more likely to lead

to more serious outcomes (e.g. Joksch, 1993; Kloeden et al., 1997; Aarts and Van Scha- gen, 2006; Pei et al., 2012). High travel speeds are also associated with many accident triggering factors such as lower reaction times, longer decisions, breaking and stopping distances, reduction of manoeuvrability and increased possibilities of manipulation error, loss of control and exceeding the critical speed on a curve (Solomon, 1964; Godwin, 1984; Hale, 1990; Fildes and Lee, 1993; Patterson et al., 2000; Navon, 2003; Aarts and Van Schagen, 2006). On the other hand, higher speeds are also related with more uniform distribution of speeds (i.e. lower speed variance) that is considered to be beneficial for road safety (Lave, 1985; Graves et al., 1993; Navon, 2003).

There is a considerable amount of research on the accident-speed relationship but several points of disagreement between studies. Most of the studies found driving speeds to be linearly or exponentially related to accidents (e.g. Fildes et al., 1991; Baruya and Finch, 1994; Kloeden et al., 1997; Quimbly et al., 1999; Taylor et al., 2000; Kloeden et al., 2002). A few studies contradicted the common belief though, proposing that speed is inversely proportional with accidents (e.g. Baruya, 1998a; Stuster, 2004) and others reported sta- tistically insignificant relationships (Lave, 1985; Garber and Gadiraju, 1989). Some of the most recent papers that explored the impact of speed on accidents using advanced

statistical modelling did not find a statistically significant relationship between speed and accidents (e.g. Garber and Ehrhart, 2000; Kockelman and Ma, 2007; Quddus, 2013). Pei et al. (2012) attempted to explain the results’ inconsistencies suggesting that the esti- mated accident-speed relationship by models strongly depends on the selected measure of exposure; the relationship was shown to be negative for distance-based exposure (i.e. vehicle miles travelled) but positive for time-based exposure (i.e. vehicle hours travelled). The inconsistent results between research papers in fact might be related with a variety of methodological and data limitations that do not permit the accurate evaluation of the speed-accident relationship. This will be further discussed in section 2.4.

The work of Solomon (1964) on the relationship of speed and accidents is one of the widely cited, replicated and criticised studies; maybe more than any other relevant study. This is because one of its main outcomes was that speed dispersion and not speed is related with accident frequency; contrasting what was believed so far and the findings of many subsequent studies. Speed dispersion (or speed variance) is defined as speed differ- ences between or within lanes between individual vehicles or in a road section level (Aarts and Van Schagen, 2006). Speed differences lead to more vehicle passes that are related with more accident prone interactions (Lave, 1985; Navon, 2003) In his case-control study Solomon (1964) employed 10,000 accident reports that included information on the speed before the accident (based on police officers’ estimations and a combination of other rel- evant data in accident reports). The control data included spot speed observations for 290,000 drivers that were measured with concealed, speed measuring devices. Solomon (1964) comparing the two speed distributions that stemmed from the case and the compar- ison groups (defining an accident involvement rate) observed that that accident-involved drivers were travelling at speeds that deviated from the average speed and more specif- ically were considerably lower than it. Accident involvement rates were calculated by dividing the number of accident-involved drivers by the respective vehicle-miles of travel and it was plotted against speed forming a U-shaped curve (see Figure 2.3) that according to Solomon (1964) represents the chance of a driver being involved in an accident as a function of their speed. It can be noticed that the involvement rate reaches a peak at very low speeds (15 mph) and its lowest value is for a speed approximately 5 mph above the average speed that was 50 mph. From that point and onwards, involvement rate increases

constantly. As a consequence, Solomon (1964) suggested that the greater the variation (positive or negative) from the average in a vehicle’s speed, the more likely it is for this vehicle to be involved in a road accident or differently put, relatively high driving speed is safer than either low or excessively high speed driving. His finding for accident severity was different and consistent with literature, as he suggested that accident severity and speed are directly proportional.

Figure 2.3: Accident involvement rate as a function of driving speeds for day (solid line) and night (dashed line) (source: Solomon, 1964).

Solomon’s (1964) work was followed by some similar studies, in terms of methodology and results, that supported the existence of a U-shaped curve between the chance of ac- cident involvement and speed (Cirillo, 1968; Munden, 1967; RTI, 1970). The validity of this result however is questionable if the limitations of these studies are considered. The inclusion of vehicles whose speed was not chosen such as turning vehicles and vehicles in congested conditions (46% of the sample) is one factor that could distort the results of these studies (Hauer, 1971; Frith and Patterson, 2001; RTI, 1970). Also, the use of the

ratio of distributions of data that originate from dissimilar in terms of accuracy measure- ment methods (i.e. vehicle speeds before the accidents and driving speeds) by default leads to a U-shaped curve even if the two distributions were equal (Hauer, 2009).

Speed variance has been examined as a potential contributory factor several times and with different approaches ever since. One of the main challenges is the quantification of speed variance (Wang et al., 2013); speed differences cannot be directly measured like other traffic variables (e.g. volume, mean speed) so the researchers employed several dif- ferent surrogate measures. For instance, Lave (1985) used the difference of the mean speed and the 85th _{percentile of the speed distribution and estimated a positive relationship be-}

tween speed variations and accidents. This finding agrees with the majority of studies where speed variance was defined as the standard deviation of speed (Baruya and Finch, 1994; Baruya, 1998a; Taylor et al., 2000; Quddus, 2013). Pei et al. (2012) however reported that there is not a statistically significant relationship between the standard deviation of speed and accidents. This is consistent with the findings of Kockelman and Ma (2007) who used more complex expressions for defining separately speed variance between and across lanes aiming to reflect disaggregate estimates of instant variation. All these results should be seen with some caution though as they probably reflect default mathematical properties of the data rather than actual causal relationships between speed variance and accident frequency (Davis, 2002). In fact, the mechanism of the impact of speed vari- ance cannot be explicitly explained until individual vehicle-level second-by-second data are available (Kockelman and Ma, 2007).