In the present study, we are concerned with generalizing our findings at each decision stage to the population of eligible cases at each decision stage.
Cases are not known to be excluded at each stage by anything other than theoretically relevant predictors.
This raises a second point: we would note that the issue of selection bias—absent any guiding theory—does not and should not drive the specification of explanatory models. We have accounted for theoretically relevant explanatory and control variables, as informed by the extant literature.
As such, we have no reason to believe that selection bias (as a form of omitted
variable bias) is at play in the analysis reported herein.
To thoroughly address potential concerns with selection bias, in supplemental
analyses (available upon request), we attempted to incorporate Heckman-style probit-based estimators in our models.
We encountered extreme multicollinearity and unstable estimates(evidenced by “beta flip” for some coefficients); as such, we opted not to include those selection bias correction factors in the analyses reported herein.