Dynamic Analysis of Event Histories

TLDR

Sociologists are interested in studying change, yet research methods rarely reflect this interest. The paper aims to develop dynamic event‑history analysis grounded in continuous‑time stochastic models. The authors employ a continuous‑time Markov model, describe its properties and extensions for heterogeneity and time dependence, and outline a maximum‑likelihood estimation procedure, illustrated with an income‑maintenance experiment on marital status. Dynamic event‑history analysis fully utilizes data, unifies multiple outcomes, and outperforms cross‑sectional, event‑count, and panel approaches, as shown by the income‑maintenance experiment.

Abstract

There is wide interest among sociologists in the study of change but little reflection of this interest in sociological research methods. In this paper we consider the advantages of and procedures of dynamic analysis of event-history data-data givin the number, timing, and sequence of changes in a categorical dependent variable. We argue for grounding this analysis in a continuous-time stochastic model. This permits the data to be fully utilized; it also allows a unified treatment of the various outcomes analyzed in the many approaches that use only part of the information contained in such data. We focus on the familiar continuos-time Markov model, summarize its properties, report its implications for various outcomes, describe extensions to deal with population heterogeneity and time dependence, and outline a maximum-likelihood procedure for estimating the extended model from event-history data. The discussion in illustrated with an empirical analysis of the effects of an income-maintenance experiment on change in marital status. We conclude by contrasting event-history analysis with cross-sectional analysis, event-count analysis, and panel analysis. We find that event-history analysis has substantial advantages over the other approaches.