A0419
Title: Joint models for longitudinal and time-to-event data in the social sciences
Authors: Sophie Potts - University of Goettingen (Germany) [presenting]
Karin Kurz - University of Goettingen (Germany)
Anja Rappl - Friedrich-Alexander Universitaet Erlangen-Nuernberg (Germany)
Elisabeth Bergherr - Georg-August-Univerität Göttingen (Germany)
Abstract: As joint models for longitudinal and time-to-event data (JM) are a well-established estimation method in biostatistics but do not belong to the standard toolkit of social scientists, the analysis demonstrates its usage and usefulness for an application on marital satisfaction and time to marriage dissolution. The advantages of JMs for social science research questions are highlighted, and the results are compared with classical approaches such as a time-varying covariate (TVC) and the two-stage model. With the separate JMs by gender, the expected negative current value association for marital satisfaction and the risk of marriage dissolution is found. Using a classical TVC model, the effect of marital satisfaction on the risk of marriage dissolution is highly underestimated. Furthermore, applying the JM allows the decomposition of the effect of shared household work into an insignificant ${direct}$ effect and a highly significant ${indirect}$ effect via marital satisfaction for women. The decomposition for men results in both effects being significant, i.e. that a higher share of household work for men is associated with a higher risk of marriage dissolution through both pathways, directly and indirectly via marital satisfaction. The models control for the standard socio-economic variables, premarital cohabitation, and children, as well as for gender-role attitudes.
