The PANEL procedure does not require balanced data for most of the models supported in the procedure. The lone exceptions are the Amemiya-MaCurdy, Da Silva, and Parks methods, which are suitable only for balanced data.

The TSCSREG procedure does not require balanced data for any of the following specifications: FIXONE, FIXTWO, RANONE, and RANTWO. With these options, PROC TSCSREG estimates one- or two-way fixed effects models or one- or two-way random effects models. However, PROC TSCSREG requires balanced data for the Fuller-Battese, Da Silva, and Parks methods. To specify the Fuller-Battese method with unbalanced panel data, use PROC PANEL instead of PROC TSCSREG. The following statements use the Fuller-Battese method, for balanced data only, in PROC TSCSREG and, for balanced or unbalanced data, in PROC PANEL.

      proc tscsreg data=a fuller; 
         id state year;
         model y = x1 x2 x3;
         run;

      proc panel data=a;   
         id state year;
         model y = x1 x2 x3 / rantwo vcomp=fb;
         run;

As an alternative to the Da Silva and Parks methods in PROC PANEL or PROC TSCSREG, you can use the MIXED procedure in SAS/STAT^® to fit very similar models. However, the estimates from PROC MIXED will not be identical to those from PROC PANEL since they use different methods—PROC PANEL uses Seely's method while PROC MIXED uses REML. The following statements fit the Da Silva model using PROC PANEL and PROC MIXED.

      proc panel data=a dasilva m=5;
         model y = x1 x2 x3;
         id state year;
         run;

      proc mixed data=a; 
         class state year; 
         model y = x1 x2 x3 / s; 
         random state year; 
         repeated / type= toep(6) sub=state;      
         run;

PROC MIXED can be used as an alternative to the Parks method as discussed in SAS Note 22115, "Estimating a model using the Parks method with unbalanced panel data".