Comparing parametric and semiparametric binary response models

• Autor(a) principal: Proença, Isabel
• Data de Publicação: 1995
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.5/28433
• topic: Logit Model; Link Function; Bias Correction; Restricted Model; Latent Variable Model

Binary response models are frequently applied in economics and other social sciences. Whereas standard parametric models such as Probit and Logit models still dominate the applied literature, there have been important theoretical advances in semi- and nonparametric approaches to binary response analysis (see Horowitz, 1993a, for an excellent and up-to-date survey). From the perspective of the applied researcher, the development of new techniques that go beyond Logit and Probit are important for several reasons: 1. Economic theory usually does not provide clear guidelines on how a parametric model should be specified. Hence, the assumptions underlying Probit and Logit models are rarely justified on theoretical grounds. Rather, they are motivated by convenience and by reference to “standard practice.” 2. Misspecification of parametric models can cause parameter estimates and inferences based on these parameters to be inconsistent. Moreover, predictions made from misspecified parametric models can be inaccurate and misleading.