The model Y"it = @s"k (@b"k + @d"ik + @c"tk)@g"ikt + @e"it with @d"ik and @c"tk random is considered as a means of pooling the time series of a cross-section sample. The model is placed in a mixed analysis of variance framework. Relationships between various estimation criteria are derived and their asymptotic properties compared. Some implementation problems are also discussed.