A neophyte's question regarding fixed/random effect models for panel data analysis...

[Nina Lasek]

Dear all,
I am an absolute beginner in using „fixed effect“/”random effects” models for panel data analysis. And I am wondering why are these model called fixed/random effects? In multilevel modelling a fixed effect is the average effect of a predictor why may or may not vary across units (“random effect”). But for panel data models á la Allison the term fixed/random mean something different, or do I err here?
Many thanks for your suggestions!!!
N.

[Carlo Lazzaro]

Nina:
you can find an useful introduction about -fe- and -re- specifications in -xtreg- entry of Stata .pdf manual, that also reports many useful references in addition to Paul Allison's valuable textbook on fixed effect.
The help file of a famous user-written programme named -xtoverid- (type -search xtoverid- from within Stata to spot and install it) is also an useful reading to understand the difference between -fe-and -re- panel data regression in terms of overidentifying restrictions.
As far as -mixed- models are concerned, they're similar to -xtreg, re mle- (that performs random effect panel regression via maximum-likelihood random-effects estimator instead of generalized least square method).

[Marcos Almeida]

Carlo gave an insightful reply.

Going astray from technical whereabouts, and calling back the nominalism from the medieval times, we could say: what’s in a name?

In short, terms may differ and convey the same meaning, and terms may be the same and convey different meanings.

For the ‘fixed’ part of - mixed - regression is different from the -fe - option of the panel data, which, by instance, is different from the ‘fixed effect’ meta-analysis.

As said in Latin by a medieval monk (and used by Umberto Eco in The Name of the Rose, stat rosa pristina nomine, nomina nuda tenemus.