Announcement

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

  • Compare beta coefficients of different panel regressions

    Dear Statalist-Users,

    Thank you for taking me in. After many hours of research, also in this forum, I decided to open my own thread. Hopefully, you can help me.

    For my recent research project, I want to compare different regression results of my DiD panel regressions. I estimate the following DiD regressions in Stata with two different samples:

    Sample 1: ceo_o==1
    Code:
    xi: xtreg depvar1 $xlistdid  i.year if year >2001 & ceo_o==1, r cluster(gvkey)
    eststo a
    xi: xtreg depvar2 $xlistdid  i.year if year >2001 & ceo_o==1, r cluster(gvkey)
    eststo b
    xi: xtreg depvar3 $xlistdid  i.year if year >2001 & ceo_o==1, r cluster(gvkey)
    eststo c
    Sample 2: ceo_o==2
    Code:
    xi: xtreg depvar1 $xlistdidagent  i.year if year >2001 & ceo_o!=1, r cluster(gvkey)
    eststo d
    xi: xtreg depvar2 $xlistdidagent  i.year if year >2001 & ceo_o!=1, r cluster(gvkey)
    eststo e
    xi: xtreg depvar3 $xlistdidagent  i.year if year >2001 & ceo_o!=1, r cluster(gvkey)
    eststo f
    The interaction term is a little different for both groups, as the treatment differs a little between both groups. All other controls ceo_o==1 = ceo_o!=1.

    My question now is whether there is a way to compare my results from a-d, b-e, c-f,... , and show that there is a significant difference between both sample outcomes.
    I know the suest command, but that one doesn't work for panel regressions and as my sample is really big (>1000 firms in the ceo_o!=1 sample), it's not that easy to run a simple OLS with dummies for all years and firms.


    I am looking forward to reading your recommendations!

    Best,
    Michael


  • #2
    I think the two most important pieces of information needed to answer your question are precisely the information that you have not shown: what is in global macros xlistdid and xlistdidagent, and exactly what you want to compare across the different regressions. ("...and show that there is a significant difference between both sample outcomes. " is far from clear and I can think of at least a half-dozen things it could mean.)

    As an aside, -xi- is at this point an almost entirely obsolete prefix. Not only would you be better off not using it, you really should try to almost forget you ever knew it. Modern Stata has factor-variable notation. Read -help fvvarlist- to learn about it. It makes the use of categorical variables and interactions (including interactions with continuous variables) much simpler, and, best of all, when you use factor-variable notation, post-estimation you can then use the -margins- command to simplify the interpretation of your model. Read the excellent Richard Williams' https://www3.nd.edu/~rwilliam/stats/Margins01.pdf for an exceptionally lucid introduction to -margins-.

    Comment


    • #3
      Dear Clyde,

      Thank you for your reply and your hints in part 2 of your answer . My global varlists are:
      Code:
      global xlistdid tr_mz_post tr_mz post q ln_ppe_emp founding_year
      global xlistdidagent tr_mz_agent_post tr_mz_agent post q ln_ppe_emp founding_year
      and I am interested in the beta coefficients of the interaction terms (tr_mz_post tr_mz_agent_post) and if they are significantly different from each other.

      Best,
      Michael

      Comment


      • #4
        As a further aside, it seems from reading between the lines of your statements, that you think that you re running a fixed effects regression.

        What you re currently doing is a Random Effects panel regression.

        xtreg y x, robust cluster

        is giving you a Random Effects regression because this is what -xtreg- defaults to, and you have not specified otherwise.

        Comment


        • #5
          Otherwise what you want to do is very easy to do manually.

          Say I want to compare whether the coefficient of the regression of wages on experience is the same for males and females. I generate a dummy variable female equal to 1 for females and 0 for males, and I regress for the full sample

          reg wages female experience female*experience

          test female*experience

          the coefficient on female*experience is telling you how much higher/lower is the regression parameter in

          reg wages experience

          for females compared to males.



          Comment


          • #6
            Originally posted by Joro Kolev View Post
            As a further aside, it seems from reading between the lines of your statements, that you think that you re running a fixed effects regression.

            What you re currently doing is a Random Effects panel regression.

            xtreg y x, robust cluster

            is giving you a Random Effects regression because this is what -xtreg- defaults to, and you have not specified otherwise.
            I know that i am using re, but thanks .

            test female*experience

            the coefficient on female*experience is telling you how much higher/lower is the regression parameter in
            Thank you for that, I will try

            Comment

            Working...
            X