Individuals versus aggregates: The pros and cons of each perspective in examining offender choices

In this article, Professor Johnson raises a very interesting, important, and timely set of questions. Chief amongst these is, as Johnson writes, ‘whether there exist (spatial) patterns in sequential choices’ of offenders. This issue has, of course, been touched upon from a certain perspective previously; any method for generating prospective hotspot maps from historical and recent criminal activity is in some sense attempting to answer this question, by predicting future crime locations (both spatial and temporal) using the locations of previous crimes (Bowers, Johnson, & Pease, 2004; Mohler, Short, Brantingham, Schoenberg, & Tita, 2011). However, there is a somewhat subtle difference between the goals of prospective hotspot mapping and what Johnson is referring to here. Prospective hotspot mapping techniques typically place no emphasis on exactly who is committing the predicted crimes, whereas Johnson is really asking how individual offenders make their decisions. Of course, if a ‘good’ model of how individuals choose their targets can be made, such a model can then be used to create prospective hotspot maps; the inverse is not necessarily true, however. That is, suppose that a good model for individual offender choices exists, and it depends on various parameters belonging to each criminal – his home location, his routine activity anchor points and paths between them, his predilection for committing crimes in general, his desires to cluster or spread his crimes, etc. – then, if one also possesses (or can estimate) the probability distributions for these parameters over the entire criminal population, a prospective hotspot map can be made by averaging the hotspot maps of each of the many individual offender types (parameter combinations), weighted by the probability that such a criminal exists and has committed some portion of the recorded prior crimes. On the other hand, for a ‘good’ method of creating a prospective hotspot map, this averaging has in a sense already been done, and it is quite difficult, if not impossible, to invert this and discover all the various criminal types that contribute to this average. As a simplification, one could assume that all criminals are in fact of the same type, so that the average accurately represents the way any given individual will behave. However, this is most likely not true, and this assumption is therefore a perfect example of the ecological fallacy. Unfortunately, most studies of distance-to-crime data, and the inter-event-distance analysis performed in this article,