Study reveals flaws in test aimed at identifying repeat sex offenders

Princeton University student’s study finds defects in Arizona risk assesment system.

   Some elements of the actuarial tests that predict the percentage of convicted sex offenders who relapse into crime are based upon false assumptions and prejudices, according to the thesis work of a recent Princeton University graduate.
   These tests help determine which sex offenders will, in accordance with Megan’s Law, be announced to the community upon their release from prison.
   Megan’s Law ensures that law enforcement agencies notify the regional community when a dangerous sex offender is released from prison. To determine whether a sex offender is still a danger to the community, about a dozen states use actuarial tests. Actuarial tests assess personality characteristics and details about prior offenses to mathematically predict the likelihood of a person committing repeat offenses.
   Alison Weingarden, a June graduate of Princeton, used Arizona’s actuarial test — the Arizona Risk Assessment — as her case study. The ARA is comparable to other state tests. The state of Arizona granted her access to an earlier version of the ARA, which the state re-evaluated and revised in 2000.
   Ms. Weingarden used survival analysis, a statistical method used to assess the actual recidivism of released criminals, to evaluate the ARA’s predictive accuracy.
   The item "gender of victim" was highly weighted on the ARA, assigning a higher risk score to offenders with a history of male victims. Ms. Weingarden found that, in fact, offenders with a history of female victims were significantly more likely to commit future sex crimes than offenders with a history of male victims.
   "I thought this result was particularly striking because it emphasizes the need for statistics in jurisprudence," said Ms. Weingarden. "The statistical validation studies uncovered a terrible prejudice in Arizona’s previous sentencing practice. Namely, offenders who had homosexual tendencies were being penalized over and above what could have been merited by the facts."
   Arizona’s 2000 study turned up similar results, and amended the ARA accordingly.
   Ms. Weingarden also found that the ARA heavily weighed four items that were of little or no importance to predicting future offenses. The items "use of weapon," "use of force," "paraphilias" — abnormal sexual interests — and "aggravating characteristics" — torture or mutilation of victims — were weighted highly in both the ARA and the recent revision.
   "I would agree that these are important items to include, but they predict the magnitude of future offenses, not the likelihood," said Ms. Weingarden. "I did not propose to eliminate these items from the test, but I did suggest that they be separated from items that predict the likelihood of re-offense."
   Ms. Weingarden used survival analysis to create a revised version of the ARA.
   "I found that my suggested improvements dramatically improved the predictive ability of Arizona’s actuarial test," she said.
   She suggests that survival analysis be used in more of these validation studies.
   Ms. Weingarden earned a bachelor of science in engineering degree in operations research and financial engineering.
   Her thesis is titled "Improving the Lots: Actuarial Testing, Survival Analysis and Megan’s Law."