[Evaluation of the quality of a decision-making rule in discriminant analysis].

Logistic regression (LR) and concept of the beta-confidence allocation rule were applied to predict the survival of dogs within the radiobiological experiment. The biochemical and haematological investigations in 86 dogs on the 2nd, 5th and 8th day post-external gamma-neutron irradiation (4.8-7.2 Gy) were used as explanatory variables for the prediction of more than 64 days survival. The influence of experimental conditions was monitored. Correctly predicted cases with LR models were proportioned as 81%, 85% and 93% on 2nd, 5th and 8th day respectively. Using beta-allocation rule, those correctly allocated with 0.50 confidence on the individual days were 75%, 81% and 85% of cases, that adds valuable information on the stability of the estimated classification rule. Method of beta-allocation also allows testing of whether individual observations are correctly assigned at a given significance level.