Klein, S. P., & Rolph, J. E. (1991). Relationship of offender and victim race to death sentences in California. Jurimetrics, 32, 33-48.
Data were analyzed for 496 California defendants whose convictions for homicide required sentences of life-without-possibility-of-parole or death. A classification tree analysis with several case characteristics was used to predict which sentence would be imposed. This procedure had accuracy rates ranging from 81% to 91%. There was little or no increase in accuracy when offender race, victim race, and interactions among these variables were allowed to enter the prediction system. Thus, as a group, the sentencing decisions studied did not appear to be either arbitrary or racially discriminatory. “[D]iscretionary use of the death penalty creates two moral and legal dangers: arbitrariness and discrimination. When a handful of cases is selected [for death] from a large mass [of all homicides] there is risk that the selection will not be based on any consistent normative criteria—that those chosen for execution will be indistinguishable from the rest on any legally appropriate basis—or, worse, that they will be distinguished only by legally improper criteria—poverty, powerlessness, or race.”
Used classification tree algorithm to cluster cases; race of defendant and race of victim variables did not improve classification accuracy in either cross-validated or full model. White defendants somewhat more likely than Black defendants to receive death sentence.