Economic evaluation of infrastructure related measures

Deliverable 5.3 of the H2020 project SafetyCube
Daniels, S.; Papadimitriou, E. (eds)

This Deliverable reports on the work in SafetyCube Task 5.3. This addresses one of the main objectives of WP5 by contributing to the evaluation of key infrastructure related road safety measures. This is achieved by collecting, assessing and analysing pertinent information in order to conduct costbenefit analyses (CBAs) for selected infrastructure related road safety measures.

The analyses are based on the methodologies developed within SafetyCube WP3, namely: the methods for priority setting between different road safety measures, and in particular the methodology and tool for conducting CBAs (i.e. the “E3 Calculator) - given that CBA is the recommended method for measures priority setting.

A selection procedure was followed for meaningful candidate topics for a CBA. The selection criteria were as follows: first, among the 48 infrastructure measures examined within Task 5.2 of SafetyCube, the 35 measures that were assigned a green (effective) or light green (probably effective) colour code were considered. For these measures, a literature review was performed, in order to identify existing published CBAs that could be used as a basis for SafetyCube CBAs.

The studies found were analysed to identify usable data elements. The items of interest were those that should serve as input data to the E3 Calculator: target group, unit of implementation and time horizon, measures costs and measures safety effects. Lack of reliable information on these aspects resulted in dropping a measure from the CBA analyses - with the exception of measures costs, for which a “break-even” CBA could be conducted, reflecting the measure threshold cost value at which benefits and costs are equal.

Especially as regards the measures costs, apart from the literature search, resulting in a compilation of measures costs from different sources, dedicated inquiries were made to key infrastructure stakeholders to collect additional information; however, stakeholders mostly pointed towards existing publications that were already considered.

As regards crash costs, the improved SafetyCube estimates for EU countries were used in all CBAs. Eventually, CBAs were carried out on 16 measures. In general, there were two options for conducting a CBA on the selected measures:

Generic CBA: this would be the preferred option when a meta-analysis with confidence intervals of the estimate of the measure was available, as such an estimate is considered highly reliable and transferable. However, in this case no “perfectly matching” measure cost and target group was available. Consequently, a generic unit of implementation and related target group was defined, and measure’s cost information was sought from the available sources and value-transferred to the generic context, as required.

Adjustment of an existing CBA: if no meta-analysis was available giving a generic estimate of the measures safety effect, specific case-studies were sought from the literature, with particular emphasis on existing CBAs. The most important condition for existing CBAs to be selected was the presence of a robust and reliable estimate of the safety effect of the measure, including its confidence interval. The advantage of this case is the “matching” measures cost, implementation conditions and safety effect; which is however at the detriment of transferability of the estimates. The existing casestudy was adjusted in two ways: first, with the improved SafetyCube crash costs estimates, and second, with the update of all figures and estimates to the reference year 2015.

The CBA outputs concern both the measures Net Present Value, and the Benefit-to-Cost ratio (BCR). The results suggest large variations of the BCR, ranging from not cost-effective (e.g. automatic barriers at rail-road crossings, traffic calming) to highly cost-effective (e.g. chevron signs, rumble strips, safety barriers, section control, junctions channelization or conversion to roundabouts).

The results of any cost-benefit analysis are much dependent on the underlying assumptions. Effect estimates are – even in the best known cases – only known within a certain uncertainty margin. It is therefore useful to run a sensitivity analysis based on some alternative assumptions about the effects of the measure. For the vast majority of the CBA we ran sensitivity analyses that use some alternative effect estimates.

If available we used the upper and lower limits of the 95% confidence intervals of the estimates. In the ideal case these estimates were resulting from a meta-analysis, in other cases the used values result from one or two particular studies. The used values represent a (much) lower than expected and a (much) higher than expected effect respectively. Overall, this sensitivity analysis did not change the overall trend of the measures cost-effectiveness, with a few exceptions: dynamic speed limits, 30-zones and traffic signals were found not cost-effective under the low measure effect assumption (lower limit of the 95% CI).

Moreover, in order to reflect the inherent uncertainty of cost estimates we decided to include also two scenarios in which the measure costs vary from a ‘very low’ (-50% of the estimate) level to a ‘very high’ (+ 100% of the best estimate) level. The latter case concerns the, more frequent, case of measures cost under-estimation, while the former case may be most applicable to technology solutions, whose costs tend to decrease with increasing penetration or technology improvements. A few measures were found clearly sensitive to changes in the measure costs as their BCR values change from below 1 to above 1 throughout the different scenarios. Interestingly, it is mostly the same measures that were found sensitive in the safety effect sensitivity analysis (dynamic speed limits, 30-zones and traffic signals). Apart from these, also the BCR value of the installation/improvement of lighting is much dependent on the eventual measure cost.

The results were eventually analysed for two rather extreme scenarios:

  • a ‘worst case’ scenario as a combination of a much worse than expected effect (in principle the lower limit of the 95% confidence interval of the effect estimate) and a higher than expected measure cost (i.e. the estimated cost +100%).
  • an ‘ideal case’ scenario that is a combination of a much better than expected effect (upper limit of the 95% CI of the effect estimate) and a lower than expected measure cost (estimated cost -50%).

Even in these scenarios the measures examined remain consistently efficient (e.g. section control or rumble strips), or never become efficient (e.g. automatic barriers). Some other measures (e.g. 30 km/h zones or traffic signal installations) are clearly more susceptible to varying combinations of measure costs and effects.

The executed examples show that the assumptions on all three elements of a CBA can play a decisive role: the effectiveness of the measures, the costs of the measures and the size of the target group. The fragmentary information available in the literature resulted in several cases for a combination of information sources to be used for a single CBA. Although every effort was made by SafetyCube experts to use as consistent sources as possible, in several cases this was simply inevitable. Even in these cases, particular caution was put on the transparent and substantiated combination of information.

The flexibility provided by the E3 tool, which allows to transfer any cost value from any country to another (EU countries, USA, Canada, Australia) was exploited as much as possible, but with particular care to properly combine related information.

It should be stressed that the dependency on all these assumptions is not as such a weakness of the method but rather a weakness of the data that are usually available. In general, too little published literature is available on economic evaluations of traffic safety measures and more reliable data are needed to allow CBA of more infrastructure measures.

It is recommended to avoid relying on existing CBA results and transfer them to a different context, but in any particular case to complement the available information with the case-specific information on the measures target group, the likely safety effects, the measure costs and the circumstances in which they are applied.

Gepubliceerd door
European Commission, Brussels


Dit is een publicatie van SWOV, of waar SWOV een bijdrage aan heeft geleverd.