It doesn't appear that this applies to a one-way random effects model.   However, I think you can get what you want in a 3 step process.  PANEL supports a variety of LAG functions.  What about: Run PROC PANEL with the appropriate LAG and output the dataset.  No MODEL statement in this run. In a DATA step, calculate the differenced values using the original value and the lagged value in this output dataset. Run PROC PANEL with a RANONE option for the MODEL statement on the differenced values.   Now, I have never done this, so take it with a great big shovel of salt.  However, the LAG statement in PROC PANEL says that it is capable of handling panel data correctly.  It would then be a matter of selecting the right version of the LAG statement to handle missing values.   SteveDenham ... View more