Do I need to "control" for endogeneity in regressions (e.g., run Hackman model) even when the potentially endogenous coefficients are insignificant and I hence reject the "causal effect" hypothesis? Suggestion by reviewer. Is it possible for the coefficients to become significant when I would "exclude" the potentially strengthening effect of endogeneity?
The size and significance of the coefficient of potentially endogenous variable changes when omitted variables are taken into account. Useful to consider behavioral model first and control for essential variables before trying any mechanical tests.
Well, its not me who asks for an explicit test on endogeneity. I think that it is not necessary when my potentially endogenous explanatory variables remain insignificant. However, maybe I am wrong?
Good morning Tom,
The hypothesis tests which determine whether the coefficient is significantly different from zero (had an effect) all involve the estimated coefficient, which--when endogenous--is both biased and inconsistent. Since b^ is biased the significance tests are invalid.
You either need to prove exogeneity or, if the variables are endogenous, use something like IV (instrumental variables) to remove endogeneity before concluding that these variables have no causal effect on the dependent variable.
Best,
Aran
Thanks for the answer. Endogeneity will bias the coefficient towards higher likelihoods of significance, right?
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running