Unbalanced data occur when not all time values are observed for all cross sections or, if time is not part of the estimation, when the cross sections are not all the same size.
Whether the data are unbalanced by design or because of missing values, almost all the methods that the PANEL procedure supports take proper account of the unbalanced data. The lone exceptions are the Amemiya-MaCurdy, Da Silva, and Parks methods, which are suitable only for balanced data.