The flattening of road accident risk: How best to describe it?

The variation in risk associated with well-known risk factors, like driver age or speed limits on public roads, has become smaller over time. For driver age, this can be shown as a "flattening" of risk: The U-shaped curve showing the variation in accident rate by driver age is getting flatter.

This paper asks how this trend can best be described. It is shown that two measures of variation in risk, relative risk and population attributable risk, do not capture the trend towards less variation in risk. It is shown that the trend towards less variation in risk is associated with a reduction of the relative contribution of systematic variation in the number of accidents to the distribution of accidents between the units (drivers, road sections, etc.) of a population at risk.

Based on this, it is proposed that accident rates can be interpreted as expected values of Poisson distributions. One may then generate synthetic mixed Poisson distributions and by comparing these for different points in time show that a reduced variation in risk implies that the distribution of accidents is becoming more random.

The distribution of accidents in populations at risk (drivers, road segments, etc.) is becoming more random (i.e., it contains less systematic variation than earlier).

Randomness does not imply unpreventability. Examples are given of how accidents can be prevented in a population where the distribution is random. The paper is based on data for Norway, and no claim is made that the trends examined in the paper apply to all countries.