Back to basics in spatial criminology

Henk Elffers
Netherlands Institute for the Study of Crime and Law Enforcement

This is an old man’s grumbling complaint: things were much better in my days, folks. All these new and fancy methods for analyzing spatial influence are no good at all: geographical information systems, local spatial autocorrelation analysis, it is not helping us forward. What do we know better about the geography of crime than we did twenty years ago? Maybe pictures for illustrating our presentations are more fancy than they used to be, so we may sell results better, but have we increased our understanding of how spatial processes in crime are functioning? I not only doubt it, I even feel our efforts are misdirected, if we forever try to increase the sophistication of our analytical tools, without having a better grasp of the underlying principles.

In this presentation I will illustrate the above grunts by treating two unrelated cases:

(a) Modeling spatial influence by specifying spatial autocorrelation models is successful if and only if we come up with a model in which all spatial influence is modeled away through spatially distributed explanatory variables, leaving no room whatsoever for spatially autocorrelated processes. I will demonstrate this by discussing the relative merits of three nested types of spatial regression models: the model with spatially autocorrelated explanatory variables, the model with spatially lagged explanatory variables, the model with spatially lagged dependent variables. Grossly formulated: we understand spatial influence only by explaining it away.

(b) The recent trend in spatial criminal simulation research, in which much effort is invested in letting simulation models run against a “real” geographical background, is counter-productive: all questions for which simulation research may be useful can be investigated in the very simple geographical background of a checker board, while many important questions are getting out of reach when the simulation is hampered by a realistic background. I will argue that the “realistic simulation trend” is a regrettable result of confounding the theoretical and applied aims of simulation research.

The presentation finishes with an effort to identify a number of key problems in the field of the geography of crime that we do not understand well enough, and should be studied concisely on a very simple scale, before we are ready to summon the available sophisticated analytical tools for probing into details.

Audio (MP3 File, Podcast Ready) Presentation Files (Zip Archive)

Back to Crime Hot Spots: Behavioral, Computational and Mathematical Models