Hidden Under the ROC. A Critique of Dangerousness Assessment Models Evaluation
Aix Marseille Université; University of Chicago
June 23, 2014
The growing role of dangerousness assessment in the penal system has been critiqued from several perspectives in the last two decades. Apart from the way dangerousness assessment modifies our criminal justice system, the construction of dangerousness assessment models has been criticized for its use of empirical data. This paper critiques models evaluation – the type of scientific research that compares different dangerousness assessment models, and in the end, evaluates recidivism evaluation. The main tool of models evaluation is the receiver operating characteristics (ROC) methodology that was developed for radar transmissions during World War II. This paper emphasizes what ROC curves – that academics have to rely on to rank different dangerousness assessment models due to the inherent logical undecidability of multicriterion decisions – do to the social production of knowledge: they hide ethical choices between order and liberty, substitute an arbitrary acceptable type I/type II error ratio and create an illusion of scientific progress in dangerousness assessment. Finally, thoughts on what data should be made clear in a recidivism prediction meta-analysis for it to be meaningful are discussed.
Number of Pages in PDF File: 25
Keywords: Dangerousness, Actuarial Methods, ROC curves, Incapacitation, Recidivism
JEL Classification: K14working papers series
Date posted: May 31, 2014 ; Last revised: June 25, 2014
© 2014 Social Science Electronic Publishing, Inc. All Rights Reserved.
This page was processed by apollo3 in 0.891 seconds