Announcement

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

  • Problem with small number of clusters using reghdfe and vce suboptions

    Hi,

    I am using the command reghdfe for two way clustering at the state and time levels. I have two types of models - one with fixed effects (both state and time fixed effects) and one without fixed effects. Both the models need to include two way clustering for standard errors. The problem I am facing is that the number of clusters in my models is small - 7 for states and 36 for the time period. So, reghdfe is not effective in this case by itself.

    I came across a post that mentioned the use of suite(mwc) or suite(avar) within vce(cluster). However, when I try to use either of these, Stata shows an error: VCE options not supported and I'm not sure why this is the case and how can it be fixed?

    Also, as mentioned, one of my models does not include fixed effects. What is the correct way of using reghdfe with two way clustering when fixed effects are not included in the model?

    Thank you!

  • #2
    You have *way* too few clusters if you only have 7 states. See any of Colin Cameron's papers on this:

    http://cameron.econ.ucdavis.edu/rese...5_February.pdf

    (As a simple rule of thumb, you need at least 50 levels per each cluster variable)

    Comment


    • #3
      Certainly agree with that paper and the use of two few clusters, however, if I am using reghdfe, can the use of this command be justified for the results I get in this particular case? And if so, how? And if not, would including suite(mwc) or suite(avar) be the way to make the results more robust to the small number of clusters?

      If I do include these vce suboptions, Stata shows an error. How can this be fixed?

      Thanks!

      Comment


      • #4
        Changing the suite() was an old option to reghdfe (not needed in new versions), and doesn't change the results, so there are no gains there. There is no easy solution to what you want to do, except for going through the things Cameron et al recommend (which are not perfect either)

        Comment


        • #5
          Thanks for clearing this! I am facing an additional problem where one of my variables is being omitted due to collinearity by Stata when I include that variable in the reghdfe regression. The regression I'm running includes two way fixed effects and two way clustering.

          Any suggestions on how this problem of collinearity can be fixed?

          Thank you!

          Comment

          Working...
          X