For nearly a century, investigators in the social sciences have used regression models to deduce cause-and-effect relationships from patterns of association. Path models and automated search procedures are more recent developments. In my view, this enterprise has not been successful. The models tend to neglect the difficulties in establishing causal relations, and the mathematical complexities tend to obscure rather than clarify the assumptions on which the analysis is based. Formal statistical inference is, by its nature, conditional. If maintained hypotheses A, B, C,... hold, then H can be tested against the data. However, if A, B, C,... remain in doubt, so must inferences about H. Careful scrutiny of maintained hypotheses should therefore be a critical part of empirical work-- a principle honored more often in the breach than the observance.

Centre de recherche et développement en économique (C.R.D.E.) / Centre interuniversitaire de recherche en économie quantitative (CIREQ) and Département de sciences économiques, Université de Montréal

Binary dose-response data often exhibit extra-binomial variation when the responses arise naturally in groups or "litters. " This paper investigates the use of generalized Wald and score statistics for robustifying the standard inference methods based on the binomial likelihood. Special attention is given to the probit analysis of a parallel assay of the teratogenic effects on mice of several dioxins.

Département de sciences économiques and CIREQ

Binary response index models may be affected by several forms of misspecification, which range from pure functional form problems (e.g. incorrect specification of the link function, neglected heterogeneity, heteroskedasticity) to various types of sampling issues (e.g. covariate measurement error, response misclassification, endogenous stratification, missing data). In this paper we examine the ability of several versions of the RESET test to detect such misspecifications in an extensive Monte Carlo simulation study. We find that: (i) the best variants of the RESET test are clearly those based on one or two fitted powers of the response index; and (ii) the loss of power resulting from using the RESET instead of a test directed against a specific typeofmisspecification is very small in many cases.