PSM is a non-parametric analysis of measuring impact. It does not require distributional assumptions. However, the observed covariates entering the probit or logit models for the propensity score would have to satisfy the conditional independence assumption (CIA) and common support assumptions. If 3/4 covariates satisfy these assumptions then can it be appropriate to use the model? Or, should the number of covariates be around 15? How can we determine this?
It is not complete a number of covariates to make the PSM, it is more appropriate to estimate the matching observations according to the likelihood of participation in the program model. Usually required to have a minimum number of variables such as 10 (but reflecting characteristics at the individual household and community) that when estimating the Probit model some out not to be significant.
Hi Shaheen,
There is no "statistical" answer to your question. The covariates that you include in a matching (or any econometric analysis) should begin solely with theory (subject to data constraints). If theory suggests that 3-4 covariates jointly determine the selection process and outcomes, then those are the covariates that should be included. Adding extraneous covariates will only increase your probability of achieving good balance on the covariates that actually matter.
Best,
Merlin
- Badges
- Science topic
- Similar topics
- Social Science