Road Safety: State Versus State

[Chris Lloyd]

As a soon-to-be-Australian motorcyclist I had a look at some motorcycling related figures for Australian states.

The figure below merges the data from the Australian Fatal Road Crash Database and the Motor Vehicle Census to look at the number of fatalities of motor bike riders (excluding pillions) per 100000 registered bikes for Australia and its major states, for the period 1995-2009. This appears to be the only feasible way of constructing fatality rates for motorcyclists since the available data for distance travelled has a large amount of sampling (and possibly non-sampling) variation, rendering it unusable as a denominator.

States are plotted in (row-wise) population order, so expect the variation to be greater in the last row. The points in red are based on estimated registration numbers as there was no vehicle census in the year 2000. There is also an issue with the census being in different months within the year prior to 2000, but you could easily account for this if you had the time.

I’ll leave the reader to comment. I also wanted to construct a measure of improvement. Taking the (scaled) negative derivative of separately fitted non-parametric curves as an improvement measure gives the following figure showing improvement values for the previous decade.

Perhaps the main point to take from this is the difference between the three states Victoria, New South Wales and South Australia (who are not only improving over time but are doing so at a faster rate) and the two states Western Australia and Queensland (who have not been so successful). For 2009, Western Australia has an improvement figure of roughly zero, and for Queensland it is very low, suggesting that fatality rates have remained stagnant in these states. On the other hand, South Australia has seen fatality rates drop substantially over the last few years. The volatility of the fatality rates in SA is naturally larger than the more populous states, so it will be interesting to see if they can remain at their currently low levels.

Performing the same analysis using all motor vehicles gives very similar results: the ordering of the 2009 improvement values for the five states is identical, suggesting that whatever is bringing down these rates is largely common to both bikes and cars.

Footnote: A nice parametric approach here is to fit the model A-(B/C)*(1-exp(-C*t)) over time t, giving a simple expression for the improvement measure of B*exp(-C*t), which works well and yields essentially the same conclusions.

?



Get More from the original blog...