Applying the copula approach to sample selection modelling

Abstract

The limited availability of tractable multivariate distributions undermines the validity of the standard parametric approach to sample selection modelling. Copula distributions can be very useful in situations where the applied researcher has a prior on the distributional form of the margins, since the modelling of the latter is separated from that of the dependence structure. The present article first presents an application to female work data. Afterwards, the approach is analysed in an application to contingent valuation data on recreational values of forests. It is shown that the copula approach is especially beneficial in case of strong departures from the hypothesis of normality.

Notes

¹ As opposed, for example, to the bivariate logistic that restricts correlation to a narrow range: .

² The present work only deals with parametric copulas.

³ Other measures of dependence rely on the criterion of dependence between random variables: for a definition, see Nelsen (1999) p. 170.