Understanding bootstrap sample size requirements for bcajack2
Opened this issue · 3 comments
Since bcajack2 internally runs a regression of the bootstrap estimates onto the bootstrap sample matrix, is it correct to say that the minimum number of bootstrap samples needed for this routine to work is roughly (1/pct) * nrow(X)? The (1/pct) factor is to account for the fact that only a subset of the bootstrap samples get used.
additionally, for documentation purposes, its worth pointing out that even when one has constructed Blist <- list(Y = Y, tt = tt, t0=t0), the documentation's stated function call of bcajack2(Blist) does not seem to work, while bcajack2(B = Blist) does.
Thanks for developing this package - aside from these questions above I have found it to be awesome!
Yes, your remark on min sample size is correct. However, the jacksd values of internal accuracy take this into account.
bcajack2 needs more work and is sort of beta. Thanks for your report. We'll fix in the subsequent versions.
Again, if you are interested in bug reports, in the problem I am using bcajack2 with, N = 2280, I'm doing 7000 bootstraps, and I get NaN for the jacksd values of the BCa points. As far as I can tell, none of the bootstrap values in tt are NaN, so my guess is that this may have something to do with N vs. the number of bootstraps I am running.
Can you post a reprex so that I may check?