Combining treatment units for synthetic control method

Cross-posting this from the Stata forum to increase visibility - but let me know if that's an issue!

I am thinking about using the synthetic control method to evaluate an intervention that took place across several geographic units (so I have multiple treatment units). I saw on the synth command documentation that only a single treatment unit may be specified but if there are multiple units, users can think about combining them.

I'm curious how combining units works in practice. For example, do people take an average or weighted average of their predictors and outcomes across all the treatment units (though how would this work for categorical variables)? Are there other methods for doing this?

A bit late for the reply, but you should check out the paper on "Generalized Synthetic Controls" by Yiqing Xu. It extends the basic synthetic control technique to calculate heterogeneous treatment effects for individual treated units. It also allows for units to be treated at different points in time. Yiqing also has a handy R package called gsynth (vignette). It uses cross-validated factor loadings to combine the individual treatment effects.