Multinomial Logit Latent‐Class Regression Models: An Analysis of the Predictors of Gender‐Role Attitudes among Japanese Women

This article describes the method and application of multinomial logit latent‐class regression models in sociological research. This group of models is the categorical analogue of latent‐variable regression for continuous latent variables such as those employed in LIS‐REL or LISCOMP; therefore, though never yet employed in sociological research, the potential usefulness of these models in such research is very high. These models are regression extensions of log‐linear latent‐class models with group variables. An application focuses on predictors of three latent classes of gender‐role attitudes among Japanese women. These classes are labeled “traditional gender‐role supporters”, “prowork gender‐equality supporters”, and “antiwork gender‐equality supporters”. The analysis illustrates the usefulness of the method by identifying the characteristics of each class compared with the others. Other possible applications of the models are also discussed.