Wednesday, October 13, 2004

Michael Howard - A New Approach to Statistics

Answering heavy criticism of his speech to the effect that he walked around Brixton for two hours without seeing a policeman, Michael Howard yesterday penned an article in the Guardian ranting against what he calls "selective statistics". He is angry about the British Crime Survey, the scheme created by the Thatcher government in 1981 which questions a very large representative sample of the population about their experience of crime. Now, in Britain there are two measures of crime - the BCS, and the count of crimes recorded by the police. The problem with the latter, of course, is that it only counts those crimes that are reported to the police. Also, it is dependent on the policy of the police as to which crimes they consider worth recording. Mr Howard is currently mad keen on this one, precisely because it has gone up. This suits his political aims. The other measure, though, has been falling steadily since 1995. This of course does not.

Mr Howard has a simple answer. He wants to abolish the BCS.

Statistically, this is a remarkable act of stupidity. All statistics are a sample of reality. The ideal statistical measurement would count absolutely everyone or everything involved with total certainty. The purpose of statistics as a discipline is to approach as closely as possible to this impossible ideal while keeping the sample within the bounds of practicality. One principle emerges from this - the larger the sample, the more reliable the results. A second is that the sample must be as accurate a representation as possible of the population. Anything that tends to skew the make-up of the sample is destructive of accuracy. The recorded crime figures, for example, sample only those crimes that get reported (first principle) and are then contaminated by police policy (second principle). Further, the sampling effect of police policy can change. In 1998 the police began recording all common assaults (pushing and shoving) in the crime stats. Unsurprisingly, the figures went up simply because the sample had expanded. All these offences are classified as violent crimes, so the number of violent crimes surged as a whole new category was shoved in. The BCS has used the same methodology over time, so the figures are comparable with each other.

Howard claims that an estimated 12 million crimes are not picked up by the survey. Very true. But most of those 12 million are not picked up by the police count either, by definition, because the whole world of nonreported crimes are not! This is a definitional, arithmetical fact - without time, place or opinion. Howard points out that the BCS does not count murders, but then murders are such a tiny proportion of crime and so likely to be reported that this is insignificant. Does he really think the streets are littered with the literally millions of unreported murder victims needed to reverse the whole thrust of the figures? We are in the presence of a prospective prime minister who believes that it is better to use a less accurate statistical measure than a more accurate one, because the errors agree with him. I am reminded of a Russian joke about Brezhnev that puts him travelling on a train that breaks down. The General Secretary has the answer, though. He orders the curtains closed so the passengers can be told they are moving.

The situation with Howard is still worse, though. He does not want to give a false impression of progress but to give the impression of continuous deterioration. For his personal benefit, a continued supply of crime is necessary. If the rate of crime was not ever-rising, who would buy either he or David Blunkett's product of ever increasing authoritarianism?

No comments:

kostenloser Counter