Announcement

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

  • Model Selection with Large N and Small T

    Dear all,
    I am doing a pane analysis with more than 200,000 entities for 5 years (2006-2010).
    My dependent variable is the commercial income of nonprofit organizations and independent variables include: the number of the same types of nonprofit organization located in the same county, donations; salary paid to full time employees, urban index, and racial diversification. I used fixed effects (controlling entity and years) with robust standard errors. I did not use clustered standard errors because there was no clusters present (nonprofits were almost evenly distributed among 48 states in spite of most urbanized cities have more nonprofits)

    I ran a few diagnostics and it showed that there was auto correlation. Should I use dynamic model?? I thought I cannot use it since the T is too small.

    Any type of help is greatly appreciated. Thank you in advance.

    JS

  • #2
    Jingran:
    -the difference between clustering/robustifying your standard errors (SEs)it's immaterial under -xtreg-: SEs are the same;
    -besides, with a large N, small T panel dataset, clustered/robustifyied SEs accomodate for both heteroskedasticity and/or autocorrelation.
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment


    • #3
      Carlo Lazzaro
      Thank you Carlo!
      I just thought if my data are not clustered then it would be theoretically necessary to use clustered SEs.

      However, thank you so much! My hunch is to use two-way fixed effects with robust SEs (xtreg) and wait.

      Thank you again,

      JS

      Comment


      • #4
        Jingran:
        as far as two-way fixed effects are concerned, you may want to take a look at Sergio Correia's -reghdfe- at: net describe reghdfe, from(http://fmwww.bc.edu/RePEc/bocode/r).
        Kind regards,
        Carlo
        (Stata 19.0)

        Comment


        • #5
          Originally posted by Carlo Lazzaro View Post
          Jingran:
          -besides, with a large N, small T panel dataset, clustered/robustifyied SEs accomodate for both heteroskedasticity and/or autocorrelation.
          Carlo, can I ask if you have a reference for this? It is easy to implement in Stata but I would like to understand why this is the correct technique.

          Comment


          • #6
            Matt:
            see: https://www.stata.com/bookstore/micr...metrics-stata/
            Kind regards,
            Carlo
            (Stata 19.0)

            Comment


            • #7
              Thanks, Carlo. Much appreciated.

              Comment

              Working...
              X