You’re right…the variation in terminology can seem like an effort to exclude. For example, I have experience analyzing TSCS data and am familiar with fixed effects/random effects/whateverBeckandKatzmightbesellinglately. [As a funny aside, my much beloved TSCS instructor who shall remain nameless—though you might be able to guess, and it wasn’t Stimson, he’s much too nice a guy—taught me to be skeptical of whateverBeckKatz&Satanwereselling.] This spring, I’ll begin a new project where I’d like to combine firm-level survey data with country-level characteristics and have been led to believe by the terminology that I’ll need to learn a new class of models: hierarchical linear or multi-level linear models. Of course, these models might not be that different mathematically from the TSCS models I’ve estimated before. But since I don’t usually look under the hood and can’t derive the proof that they are mathematically equivalent in certain ways, they seem different to me. The data structure is different (instead of one set of dummies for each panel [i.e., fixed effects], I’ll have several interval variables), so the naive methodologist in me would believe that the estimation should be different, too.

It seems to me that the only thing that would be different (possibly) is the structure of the error term—i.e. you’d expect errors to be correlated between firms in the same country. But I haven’t sat down with my Hierarchical Bayesian Mixed Effects in Clusters books lately.