Free R package idea!!

JSM was this week, but I’m sad to report that I didn’t get to watch as many talks as I would have liked. For a couple reasons, the conference was only a so-so experience for me because:

1. Loads of technical problems. All of the talks that I attended had some kind of technical problem, either audio, video, or both. This usually resulted in thumb twiddling while waiting for tech support to sort things out. I’m guessing this is because the ASA cheaped out and decided to roll there own video conferencing solution through a webportal, instead of holding the conference on Zoom.
2. I didn’t take much time off from work to attend the conference. So I only got to watch a few talks on Monday, Tuesday, and Thursday. Wednesday was filled with too many meetings, and I don’t have much latitude to use a virtual conference as an excuse to miss meetings of a certain level of importance. The result is that I don’t feel refreshed or energize by this conference. Oh well.¹

I did get to see a talk from Jim Berger, where he outlined some practical changes we could make to significance tests that could help to fix some problems with significance testing. Link to the paper..² The talk was part of The American Statistician session, a session and publication that I recommend to all statisticians.

The three recommendations are simple, but importantly, they are changes they can be made now, with very little changes to the current research workflow, and with very little mental overhead.

And so here’s my R package idea: I would like an R package that implements at least the first two recommendations in a wrapper around the summary function. That way, I can load the library, and every time I do summary(fitted_model), I at least get some visual reminder that my \(p\)-values aren’t so large (maybe a change to the significance stars system?). Next to the stars or \(p\)-values, I could also have the \(1/(-ep\log(p))\) upperbound on the data odds (Recommendation 0.2 from the linked paper).

This would be a cool (and I think easy to write!) R package. Implementing recommendation 0.3 of the wrapper would be more challenging, we would have to figure out how to prompt the user for the prior odds of the hypotheses.