A double-hurdle labour supply model with fallible indicators of labour force state

Abstract

We modify the formulation of a labour supply model with unemployment, to allow for measurement errors in some of the indicators used to operationally identify the labour force state of each individual. The resulting model consists of the combination of two models: (I) the model of interest is a reduced form bivariate Probit model, aimed at explaining jointly the participation/non-participation decision and the employment/ unemployment outcome; (II) to this model, we superimpose a fairly general and flexible measurement model, specifying how discrete indicators - obtained from responses to a relevant set of questionnaire items - measure the labour force states. We establish conditions for model identification, and present a sensible iterative strategy for obtaining ML estimates. Evidence from a sample of married women in the Italian Labour Force Survey corroborates the intuitively plausible notion that parameter estimates are highly sensitive to alternative ways of defining unemployment, and documents the advantages of the proposed model.