On Tue, 22 Aug 2006, Arnaud Bervas wrote: Yes, you're right. Though you could try substituting missing values for the unacceptable observations. For example, if negative values are not permissible: genr newx = x + (x < 0) * NA That would result in the "bad" observations being dropped from the sample for estimation, without changing the dates. It should operate once for all. I don't understand what you're seeing here. I'd have to see it in the context of a script, or explicit instructions on how to replicate the issue in the gui program. (One thing to note is that the sample actually used in estimating a model can differ from the "official" sample range, if there are missing values to be skipped, and redefining variables can change the pattern of missing values.) Allin Cottrell