A latent class approach for estimating gross flows in the presence of correlated classification errors

In this chapter, we use a model-based approach to adjusting observed gross flows for correlated classification errors. It combines a structural sub-model for unobserved true transition rates and a measurement sub-model relating true states to observed states. A convenient framework for formulating our model is provided by latent class analysis. We apply our approach to data on young people’s observed gross flows among the usual three labour force states – Employed (E), Unemployed (U) and Out of the labour force (O) – taken from the French Labour Force Survey (FLFS), March 1990-March 1992. The model is shown to correct flows in the expected direction: estimated true transition rates exhibit higher mobility than observed ones. In addition, the measurement part of the model has significant coefficient estimates, and the estimated response probabilities show a clear, sensible pattern. Our approach provides a means of accounting for correlated classification errors across panel data which is less dependent on multiple indicators than previous formulations of latent class Markov models