Affinity models for career sequences

We develop an affinity model for longitudinal categorical data in which the number of categories is large and we apply the technique to 20 years of labour market data for a contemporary cohort of young adult workers in the USA. The method provides a representation of the underlying complexity of the labour market that can then be augmented to include covariate effects. These can be understood as effects net of transition patterns associated with labour market sorting. We include in our model pairwise affinities to relate nominal categories representing types of job. The affinities capture complex relationships between these types of job and how they change over time. We evaluate the role of gender and education to illustrate the different types of questions and answers that are addressed by this methodology.