A General Method for Comparing Probitand Logit-models with Single and Multilevel Data

Abstract

<p>The paper proposes a method for overcoming the so-called latent scale-problem that<br /> prevents nested logistic and probit models from being compared. This allows us to<br /> decompose direct and indirect effects for binary outcomes. Our solution is based on an<br /> explicit construction of a latent propensity behind a given binary variable. The method<br /> is validated based on both simulated and the European Social Survey data. It is more<br /> accurate and easier to interpret than the previously available methods. Furthermore, it is<br /> the only method allowing us to compare mixed binary models: the so-called ystandardisation method, for instance, is not suitable for multilevel data because there is<br /> no global scale parameter applicable to both fixed and random effects. Finally, the paper<br /> concludes that the reason why nested binary regression models are not comparable is<br /> not related to ‘unobserved heterogeneity’, like Mood (2010) suggested, but it reflects<br /> the structure of the observed model.<b></b><i></i><u></u><strike></strike><br /></p>