The Christian Science Monitor reports that a computer program designed by a team of criminologists and computer scientists is able to predict the outcome of death penalty cases with better than 90% accuracy. The program considers no law or legally significant facts in making its assessment, instead basing its analysis entirely on factors such as age, race, sex, and marital status of the offender and the date and type of offense.
The implication, says Dee Wood Harper, one of the researchers and a professor of criminal justice at Loyola University in New Orleans, is that “if this mindless software can determine who is going to die and who is not going to die, then there’s some arbitrariness here in the [United States justice] system.”
I was considerably less impressed by the 19-variable model. I can devise a zero-variable model that will have a better-than-80% accuracy rate for predicting the outcome of modern American death penalty cases simply by having a model that will always answer “no” to the question “Will the death row inmate be executed?” Given states like California where the death penalty is on the books, and juries regularly sentence criminals to death, but the Ninth Circuit refuses to let the state perform executions, there’s little surprise that a model that accounts for location and year of conviction can do even better. (As an ACSBlog commenter notes, part of Harper’s “arbitrariness” is reflected in the difference in state laws.) Between 1973 and 2000 there have been about 4500-5000 death row inmates, and fewer than 800 executions. (Susan Llewelyn Leach, “Using software to model death row outcomes”, Christian Science Monitor, Apr. 27)(& letters to the editor, May 10 and Jun. 8).