Extending RBesT to negative binomial data

[cbaayen2]

Thank you for a great package. Would it be possible to extend RBesT such that it can handle negative binomial data? I can support by providing a case study if helpful.

[weberse2]

Thanks for the suggestion. Would a normal approximation to the log transformed count rate and the assumption of a known overdispersion phi be possible in your case?

See https://www.rpact.org/vignettes/planning/rpact_count_data_planning/

The log count rate has a variance for one observation of phi + 1/(lambda * t) ... so the sigma in RBesT becomes sqrt of that. It is a bit more tricky in this case given then sigma depends on phi and lambda... but one can still work things out.

I leave the issue open for now as I guess it would be nice for the user to be allowed to specify a sigma which depends on the mean (insted of having to call oc2S instead with respective sigma values again and again whenever one wants to try out another log(lambda)). As a note to myself - I should investigate if adding a family argument to the normal oc2S function would be a good option here. Not sure yet; I would need to look into this in more detail.

Does that help already?

[cbaayen2]

Thank you, yes, I think that will work.

[weberse2]

@cbaayen2 ... finally it did happen that RBesT 1.10-0 now supports negative binomial endpoints under the assumption of a known overdispersion parameter. This was a fairly big implementation for RBesT. An intro to the feature is also written up as an article:

https://opensource.nibr.com/RBesT/articles/negbin_interim_futility.html

This is a new feature - so feedback is much appreciated.

[cbaayen2]

Wonderful news! Looking forward to trying it out.