Abstract

Four percent of Georgia homicides in the late 1970s, the time frame of the Baldus Study, received death sentences. The rate of death sentences for White defendants was either 4% or 6%, depending on whether the victim was Black or White. In contrast, the rate of death sentences for Black defendants was either 1% or 28%, depending on the same.

Baldus adjusted the data for myriad factors and concluded that there was race-of-victim bias, but not race-of-defendant bias. Yet he added up the victims of each race regardless of who killed them, and added up the defendants of each race regardless of who they killed. This conclusion has been taken as a starting point for many subsequent logistic regression race bias studies.

If instead one asks which of the four unadjusted defendant-victim race categories B-B, B-W, W-B, and W-W have the most deviation from the 4% mean, it is the Black defendant groups, B-B and B-W, that account for almost all the error (96%). For these Black defendants, then, there was an equal protection problem in the Baldus data that went unaddressed in his Study.