ABSTRACT
However, this common practice is difficult to defend since-unlike the usual binary response model-here the "latent" variable is fully observed. So there is no need for a binary response estimator if one wants to test impacts on poverty of household characteristics. The parameters of interest can be estimated directly by regressing C; on x;. The relevant information is already contained in the levels regression which is consistently estimable under weaker assumptions about the errors. Measurement errors at extreme Cs may prompt the use of probits, though there are almost certainly better ways of dealing with such problems, which do not entail the same loss of information, such as by using more rnhnst estimation methods for the levels regression.
