Simulations for Causal Modeling

[Jared Greathouse]

This may seem like an odd question, but how would one do simulations for causal modeling?

I know it's possible to stimulate raw data in general, but how would I write a simulation with a pre-definded data generating process? That is, how would I simulate a panel dataset with a "true" treatment effect that a panel data estimator is meant to replicate?

I apologize if I'm not explaining or asking this well. My mentors both advised me to write code with synthetic data for a synthetic control estimator I'm writing. They said that real data analysis is good, but MC simulations would be a great way to validate a panel-data treatment effects estimator, and as far as I'm aware that's pretty standard in the statistics literature.... well, I've never done any kind of Monte Carlo simulation before, and I'm not familiar with it in a causal inference context, so could anyone give thoughts on the commands I might look into for this issue?

I know simulate exists, but are there others I might find useful?

[Leonardo Guizzetti]

There is a really clear tutorial from Alan Feiveson published in the Stata Journal that outlines the mechanical steps and broader contact for designing and conducting simulations, complete with examples. The focus there is ostensibly for power analysis, but it is only just a little more work to look at bias, efficiency, confidence internal coverage, error rates from above and below (the confidence interval excluding the true parameter value by being too large or too small), etc. you could use -simulate-, which has some very nice conveniences, but you can readily produce a simulation using one (or more nested) for loops and posting results using frame post/postfile.

[Jared Greathouse]

Okay thank you so much, I'll look at this!