Speed Cameras Revisited

Some time ago I wrote about the speed camera partnerships being rubbish at statistics and effectively using regression of the mean to inflate their effectiveness. [link]

Now, The Cambridge Evening News published the statistics for the before and after three years for each camera in Cambridge for which data is available, a massively pro-camera article full of dubious statistics. They provide a breakdown of which cameras are most effective without mention of statistical significance [hint - none at all], and inform us that the total amount of fines issued by Cambridge Camera Partnership is now £860,000 per annum. Intriguingly, these figures aren't on the homepage of The Cambridge Safety Camera Partnership Website which is suprising.

As detailled on my other page we can't draw any conclusions from the KSI data because it will always suffer regession to the mean problems. However, we can proceed with analysis on the slight accident data, assuming that.

  1. Slight injuries and KSI are uncorrelated.
  2. The missing data supplied by the Safety Camera Partnership is randomly missing, and not deliberately missing to cover up bad results.
  3. The accident data is reliable.
  4. The quantity of data is large enough to conlude with a reasonable accuracy.
  5. The data for camera sites is independent of each other to prevent double counting.

Now, clearly (1) isn't completely true, it's very likely that each KSI will have other slight casualties. (2) we have to assume to be true since we have no evidence to the contrary. (3) is questionable, since the definition of a slight accident is somewhat vague. (4) is probably only just true. (5) is unlikely, since after looking at the map the 1km stretches are likely to overlap in places.

We now have a difficulty, in that the amount of data available after camera installation is not always available for the full three year period. We now calculate the average monthly accident rate for each site for before and after camera installation, and then treating each site as equally significant we average those to get the total change in accident rate.

The answer, is about a 26% decrease in accident rate after camera installation, which sounds plausible and is much lower than the camera partnerships usually claim.

You can see the data in this easy to download and play with spreadsheet. Excel Format.

How reliable is this figure

Broadly, not very. The first problem, is that due to some sites only having 1 year of post camera data their post camera rate is three times the significance of the other sites. This means that a single extra slight injury at one of these sights alters the overall improvement by 0.5%, one bad crash involving 20 people at one of these sites would change that from a 25% to a 15% improvement.

If we repeat the calcaulation for sites that have all three years of data before and after we get a different figure, just 13% improvement in accident rate.

What does this tell us?

Absolutely nothing. Unfortunately the problem here is that we have no control data, that is sites chosen by the same criteria without speed cameras at them so we can isolate the effect on the accident rate by speed cameras from the other background effects, such as improving car safety, driver education etc.

A very vague control is that for a long time the fatality rate has dropped at around 5% per year, so over three years we'd expect to see at 14% drop in accident rate, we've observed a 13% drop which is well within the error margin of this extremely crude experiment.

How can I soundbite your statistics to support my obviously biased propaganda campaign?

If you are a Safety Camera Partnership or similar try : An independent study of the Cambridgeshire accident statistics by Peter Stevens avoiding regression difficulties showed a 25% decrease in accident rate at safety camera installations.

If you are the Association of British Drivers or similar try : An independent study of the Cambridge accident statistics by Peter Stevens avoiding regression difficulties showed that the accident rate was unchanged at speed taxation sites.

Home Mythic Beasts, shell accounts, cvs
hosting, co-location, virtual servers