On the Conditional Distribution of Goodness-of-Fit Tests