Life-Course Event History of Migration

Adding a time-varying covariate captures association but cannot tell whether the link is causal or driven by a shared unobserved trait. The same restless, opportunity-seeking disposition might cause both early partnering and frequent moving, so a positive coefficient on marital status could reflect selection rather than a marriage effect. Life-course event-history analysis goes further by modeling the processes jointly and letting their unobserved drivers be correlated, so that shared selection is absorbed into the frailty covariance and the remaining cross-effect can be interpreted more credibly as interdependence. The single-equation covariate model is a special case that assumes away this problem.

Multi-state means each career can occupy several distinct states with transitions among them — for partnership, single to cohabiting to married to separated; for residence, one place to another. Multi-episode means individuals can experience the same kind of transition repeatedly: people move many times, partner more than once, have several children. Ordinary first-event survival analysis truncates a life to its first transition, which throws away the recurrence that defines real biographies. The life-course model keeps all episodes and all states, so it can study repeat migration and the sequencing of events across careers, which is exactly what Courgeau and the later literature argue is essential.

Because the joint likelihood integrates over the correlated unobserved frailties shared across processes, which has no closed form and must be evaluated numerically or by simulation, and because identifying both the causal cross-effects and the heterogeneity correlation simultaneously can be delicate. The model can be sensitive to the assumed distribution of the frailties and to functional-form choices, and convergence is slower with many transitions and parameters. The reward, as Kulu and Milewski stress, is the ability to separate interdependence from selection; the cost is computational burden and reliance on assumptions that are difficult to verify, which is why analysts weigh whether the substantive question truly requires the joint approach.