Announcement

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

  • Endogeneity test ivreg2

    Hi everyone
    I'm researching a project on the impact of child nutrition on cognitive performance and scholastic attainment. I'm using an instrumental variable strategy using maternal height and drought exposure as instruments for early childhood nutrition . This is based on reading the literature where most papers exploit fixed effects and instrumental variables.

    I performed a first stage regression and did some tests of my instruments using ivreg2. Most of the tests show the right direction eg I have valid , non weak instruments which are relevant to my regressors and uncorrelated with the error term. One problem is I conducted a test of endogeneity (ivreg2). I get a test statistic of 0.13. This is above a critical value of 10% so theory would suggest I use OLS instead.

    However, I strongly believe my variable is endogenous. Is there any reason that the endogeneity test could fail to be rejected and yet I still assume endogeneity and proceed with IV 2sls? Furthermore , the result of this test seems to be very sensitive to one of the controls in the regression, hovering above or below the 10% level dependent on their inclusion . Is there any precedent to proceed with instrumental variable analysis if I just barely fail to reject the null hypothesis ? Or should I always use OLS in this case?

  • #2
    if the test is not significant, you cannot ignore this. You may want to add additional instruments for instance, but even though the theory suggests you something (in this case endogeneity) there is no guarantee that this is always the case.

    Comment


    • #3
      You do not say what test of endogeneity you carried out, and saying these things help clear up the thinking. E.g., if it was a Hausman type of test, the test is about whether the IV estimates are different from the OLS estimates, and "different" is measured in the metric of the standard errors of each of these estimators.

      Two general remarks:

      1. These thresholds like 5%, 10% are not written in stone. It is up to you where you draw the line.

      2. Nobody can stop you from doing IV estimation if you want to. The danger is that you might lose precision.

      3. You can just eyeball how different are the IV and OLS estimates, and make up your mind whether the difference is economically important.

      Comment

      Working...
      X