paired argument in pairwise_t_test and t_test

[albert-ying]

I'm a little bit confused by the name here. Which one should I use if I want to compare the measurements of the same individuals before and after treatment? And also, how does the function identify the "pairs"? There seems to be no argument to specify the ids of the measurements.

Thank you so much!

[wolass]

you should use t_test(data, formula, paired = TRUE)

the pairwise_t_test is for performing multiple t_tests in many groups like so:

1. group A vs group B
2. Group A vs group C
3. Group B vs group C

[wolass]

regarding the identifier I guess this is a critical issue.

Right now the way (I think) t_test handles paired data is that it expects the order of each observation to be the same in each group. After it splits them into two vectors the first observation in the first vector (group A) would be corresponding to the first observation in second vector (group B). Of course if you have unordered observations according to origin_id (patient_it, sample_id or etc.) then your two vectors may have different order of observations listed in them and they would not correspond to their pair. Therefore this error prone behaviour should be repaired by adding an identifier variable, or sth like that.

[albert-ying]

So in the current version, it is actually critical to pre-sort the table based on the ID. Thank you so much @wolass for clarifying!

[BobMuenchen]

Thanks for all your hard work on rstatix, I love it! One way it could be improved is to accept pairs of variables in some functions rather than requiring a formula. In my experience, it's much safer to enter paired data into two columns, so each row is the same person (or experimental unit). Otherwise, if you miss entering a single cell, the matching between pairs is destroyed. R's built-in function allows this form to handle that common situation:

t.test(mydata$pretest,
mydata$posttest,
paired = TRUE)

It would be helpful if t_test (and wilcox_test) offered that type of variable specification.