Root-mean-squared errors and asymptotic standard errors of ML coefficient estimates are compared for equation systems of different size, using normal and discrete bootstrap. It appears that exploiting normality does not yield any gain. Suggestions are made for correcting the downward bias of...