How to run Pooled OLS?

[Margherita Vinciguerra]

Hello everybody,

I'm somehow newish to Stata, and I'd like to ask for a little bit of advice:

I'm running a regression with an expanded Solow model (Y = GDP growth, Xs = Standard Solow variables + Human Capital + some policy variables and federal funds, whose effect on growth is my main interest). I have 21 regions within the same country, for each of them I have 14 years of data (2000 - 2013).

I have used the command "xtreg" and have obtained some nice results (i.e. in line with my literature) for the basic model, but things change when I expand it with my additional variables. Hoping it is not a data issue, as constructing the datast was a real challenge, I thought it may be because I do not have those many years after all, hence, rather than a panel, I should try and use a cross - sectional model with pooled OLS by country and year.

My question then is: how do I do that? As in, I think I should use the command "reg", however, it is not clear to me how would I pool my data. Should I counstruct some kind of "group" variable, that interacts regions and years?

Thank you all for your help!

[Carlo Lazzaro]

Margherita:
welcome to the list.
As per FAQ your chances of getting more helpful replies is conditional on posting what you typed and what Stata gave you back (with CODE delimiters, please: there's FAQ on this topic).
That said:
-assuming (but you should have specified it in your original post) that you followed the -fe- specification in your -xtreg- attempt and that the standard errors (SEs) are set to default mode (i.e., they are neither robustified, nor clustered, although those options do the same job under -xtreg-, but not under -regression), if the F-test creeping up at the footnote of the outcome table does not reach statistical significance, you are allowed to switch to pooled OLS;
- pooled OLS boils down to a matter of SEs clustering, as you're dealing with non-independent observations;
-grouping (probably via -egen-) together regions and years may give you some hard times in explaining your results;
-if pooled OLS is the way to go, you may want to consider clustering your SEs on panel id and add -i.year- as a predictor.

[Margherita Vinciguerra]

Thank you for your kind reply, Carlo, and apologies for my mistakes. I'll make sure to be more careful next time.

Your answers were very clear and, as your inferences on my code were correct (concerning FE and standard SEs), I see I cannot try Pooled OLS as my F-test is statistically significant.
I got anxious about my code being wrong, and failed to rely on some Econometrics theory.

Thank you again,

Best,

Margherita

[Carlo Lazzaro]

Margherita:
when it comes to different research fields right or wrong may (heavily) depend on what other researchers did in the past when presented with the same research topic.
Hence, I would recommend you to skim through the literature of your research field, hoping that you can find out some enlightening article about the research topic you're interested in.