Hi Becky, If matching is not exact, failure to control for the variables in the logit (or any other) model on the matched dataset will lead to bias. This is because although we have reduced the potential bias in the matching process, it is not all eliminated since we have to estimate the propensity score (and we don't know the true propensity score). Controling for the covariates will help clean up any remaining bias. This is part of the logic behind Gary's notion of "preprocessing" - first you match, and then you run the same parametric model you would have run before matching. Best, Ian On Sun, 23 Apr 2006, Rebecca Marie Nelson wrote: