Announcement

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

  • Model specification/commands to check for time-invariant unobserved heterogeneity

    If I am comparing fixed effects and pooled OLS models to test or time-invariant unobserved heterogeneity, would I include a control for year effects in my fixed-effects regression or NOT include such a control?

    And if so, why would we include controls for year effects in fixed effects models but not pooled OLS?

  • #2
    Ella:
    I fai to get your query.
    By definition, -fe- specification gets rid of both observed and unobserved time-invariant heterogeneity.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      Hi Ella,

      You can test pooled OLS vs RE with the Breusch and Pagan LM test for random effects (xttest0).
      Regarding controling the year fixed effects, you have to do it in Pooled OLS (and RE) but don't need to do it in FE estimation. As Carlo said, FE controls for fixed heterogeneity already, if you include the year dummies, because they are fixed over time they won't be estimated. However, with Pooled OLS you must include them to control for unobserved effects, say omitted variables.

      However, in this line, I have a question myself regarding controlling for municipality fixed effects -as my balanced panel is at the municipality level. For each individual I study the effect of the local unemployment at the time they moved to that municipality on their employment outcomes -hence, the local unemployment is fixed for each individual over the 6 periods I observe them. I also control for the population and immigrant concentration. When I include municipality fixed effects all the constant characteristics' estimators become insignificant, so I wonder whether it is really necessary to include municipality fixed effects to my estimation? or am I already accounting for it by keeping the local unemployment (population, immigrant concentration, etc.) fixed for each individual? I am using RE for the estimation after having checked that Hausman test.

      Comment


      • #4
        Ruth and Ella:
        as per FAQ, your chances of getting (more) helpful replies is conditional on posting what you typed and whata Stata gave you back. Thanks.
        Ruth: you can check if -i-municipality- is worth keeping via -parmtest-.
        The usual warning about not judging a regression model by the significance of its coefficients obvioulsy applies.
        Kind regards,
        Carlo
        (Stata 18.0 SE)

        Comment


        • #5
          Dear Carlo,

          Thank you for your suggestions. Unfortunately, I am working on a server and I am not allowed to extract/share outputs and I also do not have the permission to install parmtest, is there any other command I could use to test whether i.municipality are jointly significant?

          Thanks again,

          Ruth

          Comment


          • #6
            Ruth:
            my bad!
            The correct name of the official Stata command I meant is -testparm-.
            Sorry for the mishap.
            Kind regards,
            Carlo
            (Stata 18.0 SE)

            Comment


            • #7
              Thank you Carlo!

              Comment

              Working...
              X