Testing difference between BLUPS after mixed

[Erik Nyfjell]

Dear Statalisters,


I'm running a two-level empty mixed model, with levels individuals, city and state. The dependent variable being satisfaction with city services.

mixed satisfaction   || state: || city:

After that I'm retrieving BLUPs and standard errors for the city level:

predict eb*, reffects relevel(city)  
gen beta0 = _b[_cons] + eb1
predict eb_se*, reses relevel(city)  
gen eb_se1_l = beta0 - 1.96*eb_se1  //ci
gen eb_se1_u = beta0 + 1.96*eb_se1

What's the best way to test the statistical difference between spesific citities? Is it possible to use OLS, like this:

regress beta0 i.city
margins city, pwcompare(group sort)

Looking forward to your comments.

[Joseph Coveney]

When you declare city as a random effect, you're declaring the cities to come from the same population. Having made that declaration, "test the statistical differences between specific cities"—that is, an NHST of whether two samples are drawn from two different populations—doesn't make a whole lot of sense.

[Erik Nyfjell]

Thank you for replying. Just to clarify, my approach here is similar to this article, where it seems to me the author is drawing some kind of conlusion based on the BLUPs.