Student t's vs. Tukey-Kramer's




I was reading on Each pair, Student's t vs. All pairs, Tukey-Kramer HSD and the noticeable difference I found was that the Student's t is better used for single comparison, while the Tukey-Kramer is better used for multiple comparisons as it is better adjusted to detect false positives (or Type I errors).

Am I correct on my understanding or is student's t also good for multiple comparisons?

Also, is there a computation or a rule to determine which test to use?

Thank you!

P.S. I was using JMP and was reading their basic analysis guide.


Your understanding is correct—to control the family-wise error rate (i.e. the probability of at least one Type I error) several student's t-tests can be performed in conjunction with the Bonferroni adjustment, but this is overly conservative compared to the Tukey method.

Statistical software documentation often contain guides on which multiple comparisons procedures are recommended for which situations (e.g. Dunnett's test for comparisons against a control group), but I think that Tukey's test is generally recommended for "all pairwise comparisons" situations.