Comparison of logistic regression models for choice behavior using AIC weight

Since choice is no more than the voluntary substitution of one particular response to another, experimental procedures and data on choice behavior are important for analyzing the distribution of behavior. The analysis of choice behavior data in behavior analysis has been performed by three main types of methods. They are logit analysis, probit analysis and complementary log-log analysis. These analyses are often discussed in comparison with the predictions of the generalized matching law (Baum, 1974, 1979). In this study, nine types of data were artificially generated that follow the generalized matching law, and three types of logistic regression analyses (logit, probit, and complementary log-log function model) were performed. A comparison of the results of these analyses with AIC weight revealed that there are appropriate variable transformations depending on the statistical characteristics of the data being analyzed and that a particular transformation model does not always make good predictions.