Fixed-Effects Choice

[Stanley Cholly]

Dear all,

I have a statistical-oriented question regarding the choice of time and regional fixed-effects in Stata. I need to pick the level at which I want to introduce fixed-effects in a simple regression design. I have at my disposal a database of survey's answers in around 300 regions in 28 countries spanning 80 months. However, I have some holes in my dataset, which means that a given region hasn't answered each month of the 80 months. For example, a region in Austria might only have answered in January and December and another region in France only answered in 6 months.

Hence my question is... I would like to introduce fixed effects controls at the finest level for both dimensions (i.e monthly and regional FE) but I am afraid this would "vacuum" all the variance in my sample. Do you have any insight on how to check for FE relevance in these kind of cases because I have to admit that my econometrics education has unfortunately omitted this kind of trade-off encountered by researchers.

Hope my question is clear enough and somebody can shed some light on this issue.

Thanks in advance,

Stanley

[Pratap Pundir]

As far as I know Stata FE doesn't require the panel to be unbalanced. You should probably run your model, before you decide that you lose all variance.

[Stanley Cholly]

Already did ! The question being is there a way to determine what is the most relevant level of FE to use but thanks anyway !

[Carlo Lazzaro]

Stanley:
usually, this kind of questions find their solution in the literature of your research field. Especially if you plan to submit your reserach to a tecnical journal, it would be advisable to see what other did in the past when presented with the same research goal and act on accordingly.