Logisticre gression models were used to explore.possible relationships among the sets of variables presented above: demographic and historical factors, clinical characteristics, offense characteristics, and judicial outcomes. Models were first constructed using only demographic and historical variables as predictors of two offense characteristics indicating potentially different types of homicide-multiple victims and use of a firearm. Clinical variables were then added to each model (Table 4). In a next set of models (Table 5), two judicial outcomespleading guilty and receiving a guilty verdictwere examined using only demographic and historical variables. Either clinical or offense characteristic variables were then added to the model.