Dependency Structure between Health, Retirement, and Work Activity: A Log-Linear Analysis with Modulation by Age, Gender, and Education Level

Abstract

This study investigates the dependency structure among three key domains of the labor life course — health (S), retirement status (Ap), and labor activity (At) — using log-linear models applied to data from the first wave of the ELSI-Brazil survey. By comparing independent models, pairwise-interaction models, and conditional models, we formally test whether conditional independence holds between the state variables given the sociodemographic characteristics: age, sex, and education. Results show a substantial reduction in deviance when moving from the independent model to the model with all two-way interactions, indicating strong structural dependence among S, Ap, and At. In the conditional models, although age, sex, and education influence the magnitude of the associations, none fully accounts for the dependencies, leading to systematic rejection of the hypotheses S ⟂ Ap | C, S ⟂ At | C, and Ap ⟂ At | C. The strongest interactions involve age and education, highlighting marked social stratification in health conditions, retirement processes, and labor market participation. The findings underscore the need for integrated policy approaches that jointly address these interconnected life-course dimensions.