Hi folks on RG, which method would you prefer to use for addressing selection bias in longitudinal cohort studies - Heckman's selection model (for MNAR) or inverse-probability weighting (IPW)-based approach? If the goal is to get an unbiased estimate for the exposure-outcome association, and the variables most predictive of selection have been collected, does it bother to use Heckman's model?
Thanks!
Unlike Heckman, IPW assumes that selection depends only on observables. Unobservable phenomena such as ability, which are of central importance in Heckman are assumed to be ignorable in IPW. So IPW doesn't make sense in the soc sciences.
Thank you Michael Beenstock for your answer. However in many national longitudinal surveys, ability is partly captured by multiple variables indicating interview status (e.g., proxy interview, cognitively impaired, medical conditions). In such context, if I control for those factors in IPW, do you think they are adequate for selection bias?
In terms of Heckman selection model, do you have suggestions for selecting the variables to be included in the first stage?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression