Proposed statistical methods for comparing road traffic collision and casualty rates

One way that National Highways monitors the safety of the strategic road network is by calculating and monitoring road traffic collision and casualty rates. We do this by using traffic data and STATS19 data (road traffic collision and casualty records collated and managed by the Department for Transport).

We expect these rates to fluctuate from road to road, between different road types and from year to year. This is because, fortunately, road traffic collisions are quite rare given how many miles are driven on the network. But these fluctuations can make it difficult to compare safety on different roads or road types, or over time as they make it difficult to understand whether a particular difference is more than the fluctuation we expect.

To help us make these comparisons more robustly, we've developed some statistical methods that we propose can be used to calculate confidence intervals on road traffic collision and casualty rates. We've also developed methods that we propose can be used to carry out hypothesis tests to compare collision and casualty rates.

In 2022, we published the following document describing our proposed methods:

for transparency
because we hope these methods will be useful to the wider road safety community

Statistical methods for comparing road collision and casualty rates: proposed approach

We’ve rigorously and carefully developed the proposed methods and invited feedback to help improve them. Though feedback to date has been limited, it has been supportive and constructive.

We successfully used the methods in the second and third year progress reports on the smart motorway evidence stocktake and action plan. They allowed us to compare collision and casualty rates between different road types.

Both the Department for Transport (DfT) and the Office of Rail and Road (ORR) have analytically assured these reports. Neither organisation has concerns on either the statistical methods or their application.

Based on this, we have confidence in applying the methods to comparisons of collision rates and all casualty rates.

There are still limitations in the range of safety statistics these methods can be used with. We want to ensure that they are appropriate for use in different scenarios. That's why we'll continue to extend these methods, working with specialist statisticians, DfT and ORR.