Work on bug reports related to stats package

[hturner]

Discussed in https://github.com/r-devel/r-project-sprint-2023/discussions/25

^{Originally posted by **hturner** August 19, 2023} This is a list of bugs related to the stats package that would be suitable for sprint participants with a strong statistics background to work on. | Bug ID | Level | Status | Summary | Language | |-----------------------------------------------------------|--------------|---------------|---------------------------------------------|----------| | [18253](https://bugs.r-project.org/show_bug.cgi?id=18253) | Intermediate | patch welcome | family/link specific starting values in glm | R | | [14359](https://bugs.r-project.org/show_bug.cgi?id=14359) | Intermediate | patch welcome | enhance t.test() for paired data | R | Only a couple so far - I hope to add to this list.

[shannonpileggi]

The suggested changes to t.test are multi-faceted. One element of the Bugzilla discussion relates specifically to the examples section for t.test. In order to isolate the issue, I have created a separate PR. https://github.com/r-devel/r-svn/pull/140/files

Depending on the resulting decision of how to approach the paired t-test, the examples section could be further added to.

[hanneoberman]

Update on 18253:

• As Ben Bolker suggests, the name of the link function is not available within the current scope of binomInitialize().
• Changing the default starting values would require a major overhaul to make this available, and therefore not suitable for the R sprint.
• In case there will be a future effort to address this, it might be worth discussing to allow "binomial" with an identity link

[dickoa]

Here's a PR fixing the issue https://github.com/r-devel/r-svn/pull/142

[bastistician]

Here's a PR fixing the issue r-devel/r-svn#142

I have started a check run for the stop("cannot use 'paired' in formula method") proposal in t.test.formula and wilcox.test.formula over ~5600 CRAN/BIOC packages. I expect that to finish at around 9-10 BST tomorrow morning and will report the results here so that you can discuss the impact in Warwick. (Reviewing/Debugging impacts on packages often takes time; I hope the results will be manageable in this case.)

[dickoa]

With @shannonpileggi , @SaranjeetKaur we have submitted a patch to bugzilla

https://bugs.r-project.org/show_bug.cgi?id=14359

@tslumley please, can you take a look

[bastistician]

The check run has finished (took 14.7 hours). I already filtered out some unrelated failures due to broken internet resources. 10 of 5585 checked packages are affected: diff.txt. These now have to be investigated and it needs to be discussed if the breakage is justifiable (for these cases and likely elsewhere in analysis scripts etc). I haven't myself studied the PR yet. Maybe you still have some time to discuss the results as part of the Sprint in Warwick.

[dickoa]

Thank you @bastistician, we have worked with @tslumley to simplify the patch. You'll find an updated version on Bugzilla.

https://bugs.r-project.org/show_bug.cgi?id=14359

We'll still have the same number of packages affected because they used paired in the formula interface of t.test or wilcox.test, and it's still throwing an error. We can take a look at these 10 packages to see if some of them were correct or not.

Thanks

[aitap]

• [x] fix for 14359 committed on 10th of September, PR closed on 11th
• [ ] 18253: a few comments on 22th of September, but that's it

[hturner]

Since Bug 18253 was deemed unsuitable for the sprint I think we can close this issue.