Is There Any Logic to Using Logit

• DOI: http://doi.org/10.1111/1745-9133.12059
• Published date: 01 August 2013
• Date: 01 August 2013
• Author: Shawn D. Bushway

POLICY ESSAY

FORECASTING CRIMINAL BEHAVIOR

Is There Any Logic to Using Logit

Finding the Right Tool for the Increasingly Important

Job of Risk Prediction

Shawn D. Bushway

University at Albany, State University of New York

Richard A. Berk and Justin Bleich’s (2013, this issue) article is a direct response to

an article in the Journal of the Royal Statistical Society, Series A on risk prediction

(Tollenaar and van der Heijden, 2013).1In the article, Tollenaar and van der

Heijden concluded that “usingselected modern statistical, data mining and machine learning

models provides no real advantage over logistic regression and LDA” (p. 582). Obviously,

Berk and Bleich disagree. Their primary source of disagreement comes with the models

chosen by Tollenaarand van der Heijden to represent “modern” techniques. Most notably,

Tollenaar and van der Heijden did not include random forests, a technique championed

by Berk, among others.2When Berk and Bleich repeat the exercise using their own data,

logistic regression is outperformed by both random forests and stochastic gradient boosting.

Although Berk and Bleich know that there are cases in which the performance differences

between their preferred technique(s) and logistic regression may be small (when there is a

simple decision boundary), they believe that, at the very least, machine learning techniques

should be directly compared with logistic techniques before a decision is made to rely on

the logistic techniques.

Direct correspondence to Shawn D. Bushway, 219 Draper Hall, School of Criminal Justice, University at Albany,

SUNY, 1400 Washington Avenue, Albany, NY 12222 (e-mail: sbushway@albany.edu).

1. As of January 2013, the author has been an associate editor of the

Journal of the Royal Statistical Society,

Series A

. The author was not on the editorial board during the review process for the article by Tollenaar

and van der Heijden (2013). The views and opinions expressed in this article are those of the author and

do not necessarily reflect the official policy, position or opinions of the editorial board of the

Journal of

the Royal Statistical Society, Series A

.

2. Readers concerned that Berk is merely annoyed because Tollenaar and van der Heijden (2013) ignored

his preferred technique should read Ridgeway (2013), where acting director of the National Institute of

Justice, Greg Ridgeway, talks about the strengths of random forests.

DOI:10.1111/1745-9133.12059 C2013 American Society of Criminology 563

Criminology & Public Policy rVolume 12 rIssue 3