Logit and multinomial logit models for discrete-time event-history analysis: a causal analysis of interdependent discretestate processes

This paper introduces a new direction of methodological elaborations in event-history analysis based on discrete-time logit and multinomial logit models. The methodological elaborations address (1) simultaneous analysis of two-way transitions between two states, and (2) its extension for a causal analysis of two interdependent two-state processes. The former elaboration permits a more parsimonious and unambiguous expression for the effects of covariates on the dependent process than can be obtained by separate analyses of each direction of transition. The latter elaboration provides an extension for a log-linear causal analysis of 16-fold tables presenting two-wave-two-variable panel data, and relates the log-linear analysis to event-history analysis. An illustrative application, focusing on the dynamic relationship between premarital cohabitation and marijuana use, demonstrates the usefulness of the new models and methods.