Fixed-effects multilevel (FEML)

[Helmut Hansen]

My question should be very simple: What is the Stata command of the Fixed-effects multilevel model (FEML)?

[daniel klein]

Your question is harder than you might realize. There is a multilevel linear, binary logit, ordinal logit, ... model. There is also confusion on what exactly is meant by the term "fixed-effects".

In general type

help xt
help mixed
help me

Best
Daniel

[Helmut Hansen]

That is the Stata command(s) of the Fixed-effects multilevel (FEML) LINEAR model. I am not interested in mixed-effects multilevel models ("almost identical" to REML); for example xtmixed ......, || xxxxxx:, covariance(independent), even if they have a "fixed" part.
Less important: Could I use the Hausman test to determine which of these models, FEML & REML, I should employ in the final model (i don't ask for an interpretation here), in the same way as the "standard" FE and RE models?

[Roman Mostazir]

xtset cluster

xtreg y x, fe // Fixed effect [You will ignore any between subject difference]

For Hausman test:

xtreg y x, fe // Fixed effect

esti sto fixed

xtreg y x, re // Random effect
esti sto random

hausman fixed random

[Helmut Hansen]

I didn't ask how to code for the Hausman-test .....

[daniel klein]

Could I use the Hausman test to determine which of these models, FEML & REML, I should employ in the final model (i don't ask for an interpretation here), in the same way as the "standard" FE and RE models?

Did you read the thread I was pointing you to? What do you mean by "standard" FE and RE models? What do you think the difference is between what you call FEML and FE?

Best
Daniel

[Helmut Hansen]

[QUOTE=daniel klein;n1290853]

Did you read the thread I was pointing you to? Yes I did, still.
What do you mean by "standard" FE and RE models? I was only refering to the syntax really (xtreg, fe,re) (I think I should drop the subquestion regarding the Hausman-test)
What do you think the difference is between what you call FEML and FE? It is not exactly clear to me, really I don't know, that is why I am asking (maybe just a misunderstanding?). To be more pragmatic: There are clusterrobust "versions" of fixed-effects models (,fe vce (cluster..)) and so on, but are there any for fixed-effects multilevel models? Is FE equivalent to FEML?

[daniel klein]

Is FE equivalent to FEML?

The answer probably is yes, these are the same models. But I cannot really be sure about that answer, as you still do not describe what you think the difference might be. What do you mean by FE? What do you mean by FEML?

When I am talking about FE, I mean a with-estimator, that is an estimator that only uses the within unit variance. The mathematical model is documented in xtreg ,fe Methods and formulas section.

If you have a different understanding of a linear fixed-effects model, you need to tell us - perhaps in terms of statistical formulas - what it is you are referring to. It might also help to describe the substantive problem you are trying to solve. What is the research question? What is the nature of your data?

Best
Daniel