collapse Table of Contents
  1. Profiling and its Problems - Jonathan Pryor's web log
    1. Profiling and its Problems

Profiling and its Problems - Jonathan Pryor's web log

« Reverse Engineering | Main | Taxes »

Profiling and its Problems

Via Schneier on Security, What pit pulls can teach us about profiling.

In short, generalizations are useful, except when they're bad, and they can be bad very easily. You can't profile pit bulls, because the human-killing behavior we hate isn't intrinsic to the species, it's intrinsic to the individual. You can't profile terrorists, because terrorists aren't restricted to any particular race, and even if they were they'd still be a minority of that race. Profiling based on race can't work, you need to "profile" something more useful, such as "suspicious behavior," and even that can be problematic.

It doesn't work to generalize about a relationship between a category and a trait when that relationship isn't stable—or when the act of generalizing may itself change the basis of the generalization.

As for dog attackes, the breed has much less to do than other factors, such as gender (6.2x more likely to be male), neutered (2.6x more likely if not neutered), whether the dog is chained (2.8x more likely if chained), and whether the owner was previously involved in illegal fighting.

So generalizations (profiling) can be useful. The devil is in the details, determining what to look for, as it's frequently not obvious.

Posted on 07 Feb 2006 | Path: /etc/ | Permalink
blog comments powered by Disqus