Hello everyone, I was wondering if someone could tell me what happens when you use na.omit in the gls() function in library(nlme) when you assume a correlation structure for the errors: gls(Y ~ X1 + X2 + X3, data=sample.data, na.action=na.omit, correlation=corARMA(p=1)) Is na.omit is an appropriate argument to use in this case? There is considerable information I could find about imputing missing data - which I have done to determine that the autocorrelation is approximately AR(1) - but I have not found any literature on whether it is then acceptable to omit the imputed data and just use observed measurements for generalized least squares after the correlation structure has been determined. Any help would be very much appreciated. Thank you very much! Satoshi ___ Satoshi Takahama Dept. of Chemical Engineering Carnegie Mellon University Pittsburgh, PA 15213