They are consecutive times of observation, and the time span covered was
chosen for substantive reasons. In the social sciences, many of the
effects of interest depend on institutional contexts and settings that
are historically specific. Consequently one would not be interested in
generalizing findings to other time periods. For these reasons, it seems
more sensible to treat the waves of a panel study as fixed effects rather
than random.
I disagree, and stand by my prior post. The subscripts on an effect are not
what is fixed or random; it is the effect on y that is taken as fixed or
random in these models. There are common textbook treatments of 'two-way'
random effects models. See e.g. Hsiao, Analysis of Panel Data, Econometric
Society Monographs, 1986, section 3.6.2.
