I am using propensity score matching to reduce selection bias for a particular condition (‘treatment’). However there is no specific event to differentiate a before and after for that condition (‘treatment’), and so I am not really trying to make any causal inference but rather build a more robust predictive model. Since I have repeated cross-sectional data, is it acceptable to use PSM to create a matched dataset for each year and then pool those subsets and perform a logistic regression? Is there a better way to structure my analysis?
In addition, does it matter if the covariates to estimate the propensity score also enter into my logit? Many thanks for any advice.
In addition, does it matter if the covariates to estimate the propensity score also enter into my logit? Many thanks for any advice.
