Computer predicts who dies on death row: study

A computer program designed by U.S. researchers can predict with chilling accuracy the very few men among the thousands on America's death row who will actually be executed, according to a new study.

It says the chief factor that determines whether a man will die is neither race nor poverty but education - the less schooling, the higher the chances of a lethal outcome.

There are more than 3,200 men and women in U.S. prisons who have been condemned to death. Some have been on death row for decades, but only a relatively small percentage - 53 in 2006, for example - have been executed.

Previous studies have argued that non-whites are disproportionately sentenced to death in the United States. But with little research as to whether there is any bias in deciding who will actually die, critics say the choice seems arbitrary.

Stamos Karamouzis and Dee Wood Harper of Texas A&M University in Texarkana used a computing tool modelled after the human brain, called artificial neural networks (ANN), to search for patterns linked to executions.

They created profiles for 2000 death-row inmates - half of whom had been put to death - and entered them into the program. Each profile included information on race, sex, age number and type of capital offences, prior convictions, marital status, and level of schooling. The researchers then fed in 300 profiles of other inmates from the same period, and asked the neural network to predict what had happened to them. It correctly predicted the fates of more than 90 % of this 2nd group.

To find out which of the 18 factors best matched these outcomes, Karamouzis and Harper ran the analysis repeatedly, withholding one factor each time.

Being a woman, it turned out, was the best guarantee against having one's sentence carried out - women are rarely executed. But the next most telling indicator was the number of years an inmate had spent in high school.

"The results pose a serious challenge to the fairness of the administration of the death penalty," the pair write.


The paper is published in a British-based journal, International Journal of Law and Information Technology, and features in a report this week by the British magazine New Scientist.

(source: Agence France Presse)