shapiro.test vs shapiro_test

[Vicki-H]

Some inconsistency in which is used - possible holdover from old version?

E.g. Ch 7.5.1 uses pipe + shapiro.test, in some of the CS1 materials we use pull() but would not be necessary with shapiro_test?

[mvanrongen]

Related to the changes made to t.test(). See https://bugs.r-project.org/show_bug.cgi?id=14359 for background info (reminder to self)

Effectively, since that fix:
You can no longer (or should not be able to) do t.test(y ~ x, data = ..., paired = TRUE, …) without getting an error.
To conduct a paired test, you should use the default interface: either pass two vectors, or perhaps a different formula method or explicitly specify the pairing (if available).
Any old code that uses the formula interface with paired=TRUE, it will produce an error rather than quietly give misleading results.

From Vicki (2023-10-24)

Yes, the t.test function specifically no longer works when using the formula syntax (y~x) in combination with paired=TRUE - the same issue is actually true of wilcox.test as well. I thought I'd added this as an issue on the github but it looks as if I didn't, my bad! There's an easy fix - switching to t_test and wilcox_test - and a longer-winded fix that involves pivoting to wide format and using the $ syntax.

On a related note with paired t-tests - I think we might need to revisit the assumptions a bit in the course materials, as there's been quite a bit of confusion around whether equality of variance is actually necessary (my understanding of it contradicts with some of what's written, as does the idea that a paired t-test is actually a one-sample test of the differences)