Statistics will show whether the New York police are racist

Saturday's shooting by New York police officers of Sean Bell, a soon to be married black man who was unarmed, brings to mind the 1999 death of Amadou Diallo in which an unarmed Haitian immigrant was shot repeatedly by police.

Malcolm Gladwell, in his book Blink, explained that the officers involved had made a number of common errors, each typical of the way we think under stress. I think he was right, but that it is not the whole story. Why? Because he treats the Diallo shooting as an individual incident.

My problem with Gladwell's entire theory in Blink - his idea that snap judgments are often better than detailed work, providing you train yourself to make good ones - is that he uses anecdotes to make a point that only statistics can prove. Snap judgments will sometimes be better than careful examination, but how often? You can train yourself to reduce bias in instant reactions, but by how much?

And this brings me to the New York case. Each shooting is a disaster and cops need to be trained to avoid such things happening. On the other hand, it is difficult to imagine a training scheme that eliminated error altogether. There will always be deviation from best practice, so what is the trade-off between reducing deviation by a few and increasing the training of everyone?

The temptation will be to isolate this incident and investigate it by itself. But this will tell you little. The inquiry into this case needs to look beyond this one incident and ask - how many of these shootings are happening as a proportion of the number of incidents in which armed police are involved? Can the proportion be realistically reduced?

And, incidentally, if the shooting was an example of racist policing, statistical work will show this in a way nothing else will.