On the adequacy of the sargan distributionas an approxihation to the normal

The Sargan distribution has been suggested as an alternative to the normal distribution in models Eor which the likelihood function contains the c.d.f. of the errors. However, in such models an incorrect assumption on the error distribution causes the maximum likelihood estimators to be inconsis tent. We calculate this inconsistency, for three simple models. In models with a moderate degree of censoring or truncation, the inconsistency is large e3?0>!gh to he bothersome.