Synth2 - Running the in-space placebo test with the LOO test

[Noah Spencer]

I am using the user-writter -synth2- command to conduct a synthetic control analysis.

I would like to run the leave-one-out robustness check with a unit/in-space placebo test run for each leave-one-out specification. So, for example, suppose I have state-level data and Maryland and Colorado contribute to my synthetic control. I'd like to run the estimation without Maryland and then use the in-space placebo test to get a p-value, then run the estimation without Colorado and get a second p-value.

Is it possible to do this automatically with the -synth2- command? Or do I have to set up a loop that runs SCM and gets an in-space placebo test p-value after dropping a different state each time?

[Jared Greathouse]

EDIT: LOO is different from in space placebos. You can do both in synth2, but not together at once. I'm not even sure I'd believe this specific test (doing both at once I mean), and I'd need to see it validated, if I'm understanding you well.

Either way, tell me about your context. Why do this? I'm not saying it's a bad idea, I should say. It just seem... overly convoluted, for whatever goal you want to achieve. So please, talk to me about your data. Let's hear about why we might want a combination of these.

[Noah Spencer]

My main form of inference in the paper I'm working on is to show p-values from the in-space placebo test (i.e. p-values obtained from adding the "placebo(unit)" option to my -synth2- command).

I want to use the LOO test to show that my results are not sensitive to any particular unit in the synthetic control. The standard LOO test shows that my point estimates are quite similar. However, I'd like to also produce p-values from the in-space placebo test after running the LOO test to show that I can exclude any unit from my dataset and still get small p-values.