Announcement

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

  • Which regression method should I use with panel data?

    I am trying to run regressions to determine the effect of Financial Fair Play on the Premier League. I have panel data for 17 seasons of the Premier League, 42 clubs and 26 variables. FFP is a dummy variable taking either 1 or 0. Should I be using fixed effects? How can I determine which is the best method to use?

    Also how do i test for serial correlation and heteroskedasticity with panel data in stata? Varying sources are telling me different things. How would I then resolve these issues? Will it affect whether I can used e.g. fixed effects?

    Any help would be greatly appreciated.

  • #2
    Joe:
    welcome to this forum.
    If your dependent variable is continuous, you can go -xtreg-. If you suspect heteroskedasticity and/or autocorrelation, just invoke cluster/robust standard errors (they do the same job under -xtreg-.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      Carlo,

      Thank you very much for your response! My dependent variables that I will use are all continuous. Is there no need to even test for heteroskedasticity or autocorrelation? I suspect that they will be present.

      So using -xtreg-, what model will this be? A pooled model using cluster robust standard errors? Also is the correct command for this -vce(robust)-?

      Many thanks,

      Joe

      Comment


      • #4
        Joe:
        you may find what follows useful: https://www.stata.com/support/faqs/s...tocorrelation/.
        -xtreg- allows you to run a panel data regression for continuous dependent variables, with -fe- or -re- specification;
        -vce(robust)- is correct.
        Kind regards,
        Carlo
        (Stata 18.0 SE)

        Comment


        • #5
          Carlo,

          Okay that's great thank you. I have tried using the commands in that stata support page however it results in iterations being run endlessly?

          How do I know if I want to be using an -fe- or -re- specification?

          Kind regards,

          Joe

          Comment


          • #6
            Joe:
            I can't reply to your first question.
            As far as your second one is concerned, if you use default standard errors you can use -hausman-; for robust/clustered standard errors, type -search xtoverid- from within Stata.
            Kind regards,
            Carlo
            (Stata 18.0 SE)

            Comment


            • #7
              Carlo,

              I have tried the Hausman method. It tells me to use random effects. For this whole method I must use default standard errors? How would I then resolve issues of heteroskedasticity/autocorrelation?

              Additionally do I need to reject using OLS first before testing for whether to use FE or RE?

              Many thanks again,

              Joe

              Comment


              • #8
                If you read the documentation for xtreg, fe, you'll see it automatically includes the test whether the fixed effects are zero.

                Comment


                • #9
                  It tells me to use random effects. For this whole method I must use default standard errors?
                  Not true. You can use cluster robust standard errors with random effects, too.

                  Comment


                  • #10
                    But I cannot run the Hausman test with -vce(robust)- [stata error: "hausman cannot be used with vce(robust)...]", so if I run the Hausman test without robust errors, and it indicates that random effects is the appropriate model, can i then use -re vce(robust)- to estimate a regression? Despite that fact that random effects was decided upon using default standard errors.
                    Last edited by Joe Tollet; 19 Feb 2018, 10:50.

                    Comment


                    • #11
                      you can either use pooled or random effects as the fixed effects will just ignore your dummy variable with 0 and 1 value. further use vce(cluster cluster variable name) for heteroskedasticity and auto-correlation robust std errors.

                      Comment


                      • #12
                        Joe:
                        as per my previous reply, yoiu can replace -hausman- with the user-written programme -xtoverid- if you plan to use robust/cluster standard errors.
                        Just type -search xtoverid- from within Stata, read the help file and install it.
                        As an aside, performing the -hausman- test with default standard errors, follow -hausman- indication and then replace the default standard errors with cluster/robust standard errors is methodologically flawed.
                        Kind regards,
                        Carlo
                        (Stata 18.0 SE)

                        Comment


                        • #13
                          Carlo,

                          Okay thank you. Sorry for the ongoing questions!

                          I did try the xtoverid method however I receive the message "error - saved RE estimates are degenerate (sigma_u=0) and equivilent to pooled OLS". How should I proceed? I am still unsure how to determine whether I should use random effects, fixed effects etc.

                          Your help is greatly appreciated!

                          Joe

                          Comment


                          • #14
                            Joe:
                            as per -xtoverid- outcome, you should go pooled OLS, with standrd errors clustered on -panelid-.
                            Kind regards,
                            Carlo
                            (Stata 18.0 SE)

                            Comment


                            • #15
                              Carlo,

                              Thats great thanks. So that would just be simply the -xtreg- command?

                              Joe

                              Comment

                              Working...
                              X