bug integrating bayesmh into a bootstrap program

[Christopher Snyder]

I think I discovered a bug trying to bootstrap a program I wrote involving the bayesmh command. Instead of 1000 runs with different subsamples, I got 1000 copies of the same result. To confirm, I wrote a toy program that doesn't use bayesmh, which works as expected, then added an irrelevant bayesmh call to it, which then produced the 1000 copies problem.
I suspect there is a macro keeping track of simulation runs shared by both procedures that gets overwritten.

[daniel klein]

You need to provide code replicating the reported behavior; either here or to Stata tech support. Without reproducible examples, there is not much anyone can do.

[Christopher Snyder]

I don't need anyone to do anything for myself; I've written a version of the program called by bootstrap using closed-form solutions for the Bayesian estimator (beta-binomial model, so well-known formulas), eliminating the call to the bayesmh procedure. Just alerting folks to the issue as soon as I noticed it. I'll write a a do file that illustrates the problem simply when I get a minute.

[daniel klein]

It's great that you found a solution for yourself. It's also nice to alert others to a potential issue. However, I believe others will benefit even more from seeing and replicating the actual problem and the details of your solution. StataCorp is also typically interested in fixing potential bugs so everyone automatically benefits. Thus, I'm sure a do-file would be greatly appreciated.

On a slightly different note, I wonder whether combining a frequentist bootstrap with a Bayesian estimator is theoretically sound. It's an honest question that might well be beyond the scope of this thread.