Mixed Modeling + fixed effects as robustness

[Ali Alipour]

Hi. I am running a mixed model with stata mixed command on 174000 firms in 20 years across 60 countries. I include the random intercepts of country, industry and year. What will be the advantage of running the model with panel data fixed effects (xtreg) as robustness check?

[Carlo Lazzaro]

Ali:
I think it might be questionable, if random effect is the right specification.

[Ali Alipour]

Thanks Carlo,

So you think that is not necessary? Why will it be questionable?

[Carlo Lazzaro]

Ali:
if the random effect is the correct specification, -fe- will be consistent but inefficient.
That's the reason why I think it may questionable.
However, it is also true that many publications reports both -re- and -fe- models, especially when the dependent variable is continuous (if feasible).

[Ali Alipour]

Thanks Carlo.

I can show the results of both random and fixed effects, or I can find which one is more efficient using Hausman test. By the way, this is a robustness test of my mixed model.

My question in fact is whether such a thing is appropriate as a robustness for mixed models (whether they are fixed or random effects)???

[Carlo Lazzaro]

Ali:
if you went -mixed-, you should have tested beforehand that -re- specification was the right one for your data, as -mixed- is equivalent to -xtreg, re mle-.

[Ali Alipour]

Thanks Carlo