Announcement

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

  • Does i.year (time dummies) change intercept or slope?

    Hi, guys, thanks for reading my post.

    I work with panel data (N dimension is much larger than T dimension), in sum around 4.500 observations - unbalanced panel.

    I tested for individual effects, both fixed and random, and tests showed that individual effects are not statistically significant. After that I run F test for time fixed effects (testparm) and they turned out to be significant, so I included time dummies in model with "i.year".

    My question is, when I include time dummies this way, do I change intercept in my regression model or slope of regression line?

    King regards.

    Jole

  • #2
    You change only the intercept. You get a different intercept for every year, by including a set of time dummies.

    If you still include your intercept, and omit one dummy, the intercept is the value for the omitted year, and the rest of the dummies are telling you the difference between the omitted year and the year in question.

    Comment


    • #3
      Dear Joro,

      Thank you so much. It is clear to me now.

      King regards,

      Jovana

      Comment


      • #4
        You are welcome Jovana.

        Another useful parametrisation is to omit the intercept, and to include the full set of year dummies.

        Then each dummy tells you the intercept for the give year.

        Originally posted by Jovana Ju View Post
        Dear Joro,

        Thank you so much. It is clear to me now.

        King regards,

        Jovana

        Comment


        • #5

          Yes, I understand. Thank you once again for your time and effort to explain all this to me.

          Comment


          • #6
            Dear Joro,

            I will ask you to help me about one more problem.

            As I said before, I work with panel data, and I am running linear regression model:

            regress depvar indepvar1 indepvar2 i.year (in accordance with the tests for individual and time effects).

            How can I test autocorrelation in this model? I have already tested heteroskedasticity, which is significant, so my plan is to manage it with cluster(id) command. Am I right?

            Thank you so much once again

            Comment


            • #7
              Hi Jovana, we do not test for heteroskedasticity and autocorrelation that much nowadays, in large N, small T datasets.

              If you add the option -,robust cluster(id)- to your regression this takes care of arbitrary heteroskedasticity and arbitrary correlation within id.

              If you still insist on testing for heteroskedasticity and autocorrelation, check out this FAQ

              https://www.stata.com/support/faqs/s...tocorrelation/




              Originally posted by Jovana Ju View Post
              Dear Joro,

              I will ask you to help me about one more problem.

              As I said before, I work with panel data, and I am running linear regression model:

              regress depvar indepvar1 indepvar2 i.year (in accordance with the tests for individual and time effects).

              How can I test autocorrelation in this model? I have already tested heteroskedasticity, which is significant, so my plan is to manage it with cluster(id) command. Am I right?

              Thank you so much once again

              Comment


              • #8
                You change only the intercept. You get a different intercept for every year, by including a set of time dummies.

                If you still include your intercept, and omit one dummy, the intercept is the value for the omitted year, and the rest of the dummies are telling you the difference between the omitted year and the year in question.
                @ Joro Kolev, this is new to me. But couldn't understand when I tried an example. Also my slope of Indep var is different between model with and without time dummies

                Comment


                • #9
                  Good point. There are at least two senses in which we can interpret the question "If I include time dummies do this change the slopes?"

                  I went with the first sense: When you include time dummies, you are allowing for different year specific intercepts. You are not allowing for different year specific slopes. Hence my answer was that No, you are affecting only the intercept. Probably I should have said more precisely: By including time dummies you are modelling different intercepts by year, and a constant slope for each variable.

                  The second sense is what you are placing on the question: If you include time dummies, is your slope going to change? The answer to this question is most probably Yes. The only situation in which the slope will not change, is if the time dummies are uncorrelated with all the variables in whose slope you are interested.




                  Originally posted by lal mohan kumar View Post

                  @ Joro Kolev, this is new to me. But couldn't understand when I tried an example. Also my slope of Indep var is different between model with and without time dummies

                  Comment


                  • #10
                    Thanks for the Help Joro as always

                    Comment


                    • #11
                      Thank you, Joro, I am very grateful.

                      Comment

                      Working...
                      X