Announcement

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

  • Philosophy behind robustness check using different specification...

    Hi,

    Let Yi = b0 + b1X1 + bi*Ci, where X is variable of interest and Ci are my control variables, i = 1,2,3..,k

    Often I see papers where they run different specifications to show that the coefficient on the variable of interest is indeed consistent (correct me/add if there are more benefits of it).

    Usually, the way out is to retain X1 in these specifications while omitting some of the Ci in these robustness specifications. So, we check that even when effect of X1 on Y is understated in robustness regressions, the direction of effect stays the same.

    I understand that the idea here is to make a point that even in the event of omitted variable bias, the effect of X1 on Y doesn't change. So in a way we are trying to confirm that our results are not simply an association between Y and X1 but also leads us to define a cause and effect relationship by claiming that any sort of endogeneity does not confound our results. Often robustness checks also involves doing sub-sampling. [please add any more robustness that you're aware of]

    My questions :
    1. How should one decide which variable or combination of variable to omit in these robustness regression?
    2. Are such robustness checks fool proof? Do they really take care of endogeneity?

    Thanks!

  • #2
    Gurpreet:
    first, it would be interesting to give an answer to the following questiom: robustness to what? Heteroskedasticity? Highly influencing observations? Else.
    That said, I woud say that the right specification for a regression model is the one that gives a fair and true view of the data generating process. That implies a good theoretical knowledge of the study subject and includes the ability of spotting endogeneous regression..
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment

    Working...
    X