Crime modeling

Here is a sneak preview of my short presentation tomorrow.

In Crime modeling, Alfred Blumstein discusses attempts by researchers to characterize crime frequency in individual offenders. An individual’s offending frequency, called \lambda, is immensely helpful in modeling the criminal justice system. By the late 1970’s, estimates of these values were crudely derived for specific crime types. However, the values of \lambda were based on aggregated crime data and were insufficient in predicting individual crime patterns. A better way to predict the heterogeneous patterns for individual offenders was needed.

Around 1980, Rand initiated an attempt to improve how individual crime patterns could be predicted. They resorted to a self-report by about 2500 prisoners from California, Michigan, and Texas. Each inmate essentially reported his or her own crime history. Upon analyzing the data, Rand noticed several phenomena. For one, the data was extremely skewed. Blumstein gives the example that among prisoners who engaged in burglary, the median number of robberies was five. However, the 90th percentile of burglaries was around 80, and the mean number of burglaries was roughly 50. In short, the data was being radically pulled by offenders who committed many more burglaries than their peers. Rand’s goal was to be able to utilize this data to better identify these high-\lambda offenders.

In 1982, using this newfound data, Peter Greenwood and Allan Abrahamse proposed a policy known was “selective incapacitation.” Selective incapacitation was a way for an individual’s crime to be correlated with the self-reported data set to identify individuals whose crime patterns suggested a high frequency of offenses. This set of predictors was meant to influence sentencing decisions in court. Naturally, this policy was not well-received. Selective incapacitation was a biased sentencing of individuals for crimes they could commit in the future.

The predictability of selective incapacitation was tested in 1987. A group of released inmates from a California prison were correlated with the self-reported data and assigned low and high offense frequencies. If selective incapacitation truly was a predictive measure of crime patterns, the individuals with high offense frequencies would be the most likely to return to the criminal system quickly. However, no correlation was actually observed. Blumstein outlines two major problems with the experiment. For one, the prisoner population involved in the self-report was not indicative of the general population. One can expect that those in prison are more likely to be high frequency offenders. Blumstein also notes that the experiment measured arrest frequencies rather than crime frequencies.

Blumstein’s vignette leaves us with a snapshot of the criminal justice system’s complexity. Forecasting crime patterns is immensely difficult. Not only is knowledge of individual crime patterns often restricted or unavailable, but even accurately using these data sets can require both care and finesse.

Source: Blumstein, Alfred. 2002. Crime modeling. Operations Research 50(1) 16-24.


2 thoughts on “Crime modeling

  1. Thanks for your interesting post and presentation! I had not previously thought of applying OR methods to issues related to reducing crime. It’s remarkable to me that anyone actually thought it would be a good idea to use selective incapacitation techniques. (Minority Report, anyone? We all know how well THAT went….) Has any more recent work been done in this area? I think it would be interesting to do another study, perhaps on a larger scale, using data other than what is self-reported.

    I know I mentioned this during your presentation, but it still seems odd to me that these studies focus almost entirely on people who are in prison. Aren’t inmates much less likely to commit crimes (at least in the near future) for the simple reason that they are in prison? It seems to me that we should somehow account for the time people are “out of the system” when building models of age vs. criminal activity. (I do suspect such things are built into the model and I’m just overthinking the issue.)


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s