Pooled-OLS estimation for panel data?

[Anusha Singh]

Hi,

I am working with panel data, and I already have the final results for my fixed-effects regression which I shall be using for my project.
However I also wish to give the reader a background on why pooled-OLS estimation might give biased results. Hence I would like to include a table with the pooled-OLS results as well to facilitate comparison.

Hence, my question is, do I simply use the -- regress depvar independentvar1 independentvar2 -- command? Also, I think I should not tell Stata that the data is panel by using xtset id year.
Is that correct?

Thank you.

[Marcos Almeida]

Yes. You may also see the F test so as to check differences between both models.

[Carlo Lazzaro]

Ahusha:
as an aside to Marcos'helpful advice, please note that:
- if you're dealing with panel data with a continuous dependent variable, -regress- outperforms -xtreg- only if the F-test appearing at he foot of -xtreg- outcome table fails to reach statiostical significance (for more detils, see -xtreg- entry, Example 2, Stata .pdf manual);
- if -regress- were the way to go, no -xtset- is required. However, you should -cluster- your standard errors on -panelid-; otherwise, Stata will interpret your observations as independent (whereas they are not, due to the panel structure of your data).

[Phil Bromiley]

To add a minor point, regress will do ols even on data that has been xtset. regress ignores that panel structure (unless you use for example the cluster option for your standard errors as Carlo suggests).

[Anusha Singh]

Yes. You may also see the F test so as to check differences between both models.

Hi,
Thank you for your replies.

Could you please tell me how can I run a F-test between the pooled-OLS and fixed-effects model? For instance, my models are:

xtset id year
regress depvar indepenvar1 indepenvar1
xtreg depvar indepenvar1 indepenvar1, fe

Also, how should I interpret the results from the F-test, as in, how do I decide which is a better model?

I am not using clustered standard errors as advised by one of the forum members from my previous posts, since my number of clusters is 18 which is not very large.

[Carlo Lazzaro]

Anusha:
look at the F-test appearing at the foot of the -xtreg,fe- outcome table: its lack of statistical significance means: go pooled -regress-; otherwise: go -xtreg,fe-.

[Anusha Singh]

Anusha:
look at the F-test appearing at the foot of the -xtreg,fe- outcome table: its lack of statistical significance means: go pooled -regress-; otherwise: go -xtreg,fe-.

Thank you very much, Carlo!