Thursday, July 08, 2004

The correlations of crime

While reading around the subject of crime statistics over the past few days, I stumbled across an interesting report from the Ministry of Justice: Interpreting Trends in Recorded Crime in New Zealand. It performs a multiple regression analysis on 33 years of data to extract social, economic and demographic factors that are statistically correlated with the overall crime rate. To avoid the problem of any old upward trend correlating with the upward trend in crime, it looks at the fluctuations - whether changes in those variables (lagged by a few years, even) correlate with changes in the overall crime rate.

(The aim of the study was to use those factors to forecast future crime rates, so you can see how this appeals to my inner geek...)

Unfortunately, lack of data on some of their possible variables (such as household income levels) and worries about reporting rates make the study less useful than it could have been, but it still managed to extract some rather interesting conclusions.

  • Increases in dishonesty offences (fraud, burglary, and theft) were strongly correlated with decreases in business confidence.
  • Violence and property crime are correlated with GDP growth, increasing two - three years after a trough.
  • The raw unemployment rate does not appear to be a significant factor (this seems to vary from country to country; arguably its underlying variables such as social exclusion and the desperation of people on the bottom of the heap which are the real problem, but some of them are hard to measure)
  • Increased female employment was the most powerful factor correlated with dishonesty offences, particularly with burglaries (there's a very obvious explanation for the latter).
  • "Family factors" (as measured by the divorce rate, numbers on the DPB or births out of wedlock) did not seem to be significant.
  • "The political party in power was not significant in any model" (apparently it matters in Australia)
  • Clearance and conviction rates matter for theft, but not for much else.
  • And finally, for Don Brash's benefit: "no significant deterrent effect was associated with the severity of punishment ... No relationship was found between changes in the number of people in prison and the recorded crime rates."

There's plenty more there, and it's quite interesting reading.

As for their forecasts, they suffer from the expected problems with trying to predict multiple input variables five years in advance. As a result, the forecasts rapidly begin to diverge from reality. It would be interesting to see how the models performed on more recent historical data, and they may be useful for short-term (one - two year) predictions.