Announcement

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

  • IV Regressions Panel Data

    Hi,

    I am using Stata 14 to run regression analysis on my dataset, which is survey data from the World Bank.

    I am trying to run an IV estimation for an endogenous bribe variable (x variable in my equation). However I am unable to find a suitable test for postestimation that works to see if a) instrument is valid b) to see if 'bribe' is in fact endogenous

    Here is the original equation:

    xtreg ln_invest bribe ib0.missing_year i.size i.year i.Industry i.Govt i.For i.fincons i.Court lnAge, fe vce(r)

    IV regression:

    xtivreg ln_invest lnAge i.size i.year i.Industry i.Govt i.For i.fincons i.Court lnAge ib0.missing_year (bribe = j2), fe vce(r)

    I have tried the Hausman test (does not work with robust s.e.) and also tried xtoverid, both to no avail.

    Please could you guide me in finding a suitable test to find out a) and b) above..many thanks in advance!


  • #2
    Kimara:
    in all likelihood your problem with -xtoverid- rests on the fact that it doesn't support factor variables; give it one more try with -xi- instead of -i.<categorical>-.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      Hi Carlo,

      Thanks for your response!

      I have since tried running the regression using -xi- instead and followed it with the -xtoverid- command and got the following:


      xtoverid

      Test of overidentifying restrictions:
      Cross-section time-series model: xtivreg fe robust cluster(panelid)
      Sargan-Hansen statistic 0.000 (equation exactly identified)

      Having used -help xtoverid- in my stata window it mentions that after running an IV regression, using -xtoverid- is a test of the null hypothesis that the excluded instruments are valid instruments. However, after seeing "equation exactly identified" I wonder if this is the right test to be using to get what I want?


      Best,
      Kimara

      Comment


      • #4
        Kimara:
        as far as I know, you cannot have overidentification if the equation is exactly identified.
        If you Google with the string -equation exactly identified AND Stata- you will find some promising entries (unfortunately, , at the moment I do not have the time to delve into them).
        Kind regards,
        Carlo
        (Stata 18.0 SE)

        Comment


        • #5
          You might want to look at xtivreg2 which includes some diagnostics xtivreg does not.

          Comment

          Working...
          X