I am using coarsened exact matching (CEM) to match my treatment-control groups. I match firms with exact matching on industry, year of founding and region, and coarsened match on firm size. I would like to test both fixed effects model and mixed effects model (since observations within strata are expected to be correlated) next. I’m setting up my mixed effects model as following:
mixed dep indep treatment [fweight = cem_weight] || cem_strata: || firm:
mixed dep indep treatment [fweight = cem_weight] || cem_strata: || firm:
- Is it sensible to use cem_strata as a level in the mixed effects model? I’m aware of people using match_id above firm_id in mixed model after propensity score matching.
- Is it ok to use cem_weight as a frequency weight? I read somewhere that cem_weight, usually used as iweight or aweight, can be used as fweight. Even though xtreg requires fweight to be an integer, mixed accepts non integer fweight. Can I directly feed fweight like I have done above?
