Longitudinal Analysis of Employment and Unemployment Based on Matched Rotation Samples

ABSTRACT: The purpose of this paper is to provide methodology to fit longitudinal data on employment and unemployment generated by the rotation sampling schemes of national labour force surveys. The proposed methodology, referred to as infinite-lag Markov models, is a generalisation of autoregressive Markov models developed for application in stochastic reservoir theory (Pegram 1980, Raftery 1985). Infinite-lag Markov chains have infinite memory and, therefore, can usefully serve to model labour supply behaviour taking into account, in principle, the complete past work experience of individuals, and not just the most recent past or the most recent spell. After a brief review of the rotation sampling schemes of 20 national labour force surveys, the different types of longitudinal sequences that can be obtained from the rotation schemes are examined. A review of various models proposed in the literature for analysing longitudinal data on employment and unemployment, expressed under simplified assumptions and in discrete forms, set the stage for the formulation of the proposed infinite-lag Markov model. The method is illustrated using matched longitudinal data derived from the US Current Population Survey.