Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Control for country effect in a pooled estimation

    Hello everybody,

    I have panel data with two time periods (t=2) and n regions as entities. I am considering the possibility of including a variable to control for country effect in the regressions. In the research, I employ both pooled and fixed effect specifications. If I'm correct, it is unnecessary to include a country variable in the fixed effect model since it is time-invariant. However, I would like to control the country effect in the pooled model.

    I generated a factor variable
    Code:
    egen f_country= group(country), label
    and then ran this regression:
    Code:
    reg y x i.f_country, vce(cluster region_id)
    The country effect is not my priority estimation here. I want to somehow control for spatial correlation. My questions are: is it correct to include the country variable in a pooled model? Is there a way to control for a general country effect instead of having the estimates for each country in comparison to the base level?

    Thank you in advance for your help




  • #2
    Duccio:
    1) your pooled OLS code is correct as far as -i.f_country- is concerned.
    I'm not clear why you ran pooled OLS too if you want to go -xtreg,fe- (as pooled OLS without is inconsistent if -xtreg,fe- is the right specification).
    2) if by "general country effect" you mean converting -region- from categorical to continuous, I do think it makes sense.
    Kind regards,
    Carlo
    (Stata 17.0 SE)

    Comment


    • #3
      Carlo, thank you for the answer.
      I am employing both specifications because I'm using the RD design, which in theory should work properly with pooled OLS by restricting the sample around the cut-off. However, I want to include fixed-effect estimation in the analysis for completeness and robustness.

      I do not understand what you mean in the second answer. Why should I convert regions to a continuous variable to assess whether the country of origin significantly affects the regional outcome?

      Comment


      • #4
        Duccio:
        please read my second answer as "I do not think it makes sense". Sorry for the omission.
        Kind regards,
        Carlo
        (Stata 17.0 SE)

        Comment


        • #5
          Ok, I see.
          I will probably just drop the country variable from the pooled since it is hard to read given the high number of countries. Just a few estimates among the country categories are statistically significant. The great majority is not, so I think I can assume that omitting the country variable won't provide a large bias.

          Thanks again!

          Comment


          • #6
            Duccio:
            you may want to test the joint statistical significance of -i.country- via -testparm- after pooled OLS.
            Kind regards,
            Carlo
            (Stata 17.0 SE)

            Comment


            • #7
              You write that you wanna adjust for spatial autocorrelation. Is your dataset spset?

              Comment


              • #8
                Originally posted by Jared Greathouse View Post
                You write that you wanna adjust for spatial autocorrelation. Is your dataset spset?
                No it is not.
                Isn't it possible to get a raw assessment of the country of origin effect even without spset? Maybe spatial correlation is not the right term.

                Comment


                • #9
                  Originally posted by Carlo Lazzaro View Post
                  Duccio:
                  you may want to test the joint statistical significance of -i.country- via -testparm- after pooled OLS.
                  "Unfortunately", they are jointly significant. I have to include the country variable in the regression somehow.

                  Comment


                  • #10
                    Duccio:
                    why unfortunately?
                    You have another predictor that, when adjusted for the other ones, helps explaining variations in the regressand.
                    Kind regards,
                    Carlo
                    (Stata 17.0 SE)

                    Comment


                    • #11
                      Originally posted by Carlo Lazzaro View Post
                      Duccio:
                      why unfortunately?
                      You have another predictor that, when adjusted for the other ones, helps explaining variations in the regressand.
                      I wrote "unfortunately" for personal convenience. It is only good to have checked for an omitted variable bias. It's just that I would have preferred that the control variables were not significant for the scope of the university report I have to submit. Also, I'm not sure how to interpret the fact that some country categories carry significant estimates and others do not.
                      Last edited by Duccio Milani; 15 Jan 2022, 04:20.

                      Comment


                      • #12
                        Duccio:
                        this is often thecase when a categorical variable has many levels.
                        That's why we investigate their joint statistical significance.
                        Kind regards,
                        Carlo
                        (Stata 17.0 SE)

                        Comment

                        Working...
                        X