1

I'm collection reaction time (RT) data on participants from the same group tested in 4 conditions (A, B, C, D).

Since I have a directional hypothesis (RT of A>B>C>D), I'm using three one-tailed t-test (A vs B, B vs C, C vs D).

My question is: if I'm performing these three comparisons, should I correct the one-tailed p-value for example using a Bonferroni correction, multiplying that value by 3?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
giorgio-p
  • 47
  • 4
  • You should use a correction for multiple comparisons when you do multiple comparisons, whether the comparisons are one-sided or two-sided. Bonferroni is one of the weakest ways to do this, though that weakness might not be so devastating with just three comparisons. However, this sounds like an $XY$ problem where you have issue $X$ that you want to solve and will have a solution if you could just figure out $Y$, so you ask about $Y$ instead of $X$. Is your real question about testing $H_0: \mu_A > \mu_B > \mu_C > \mu_D?$ – Dave Nov 19 '20 at 16:47

1 Answers1

0

You should use a correction for multiple comparisons, even if some (all) of the comparisons are one-sided, if you would do a correction for two-sided tests. (Not everyone does, and multiple comparison corrections are a contentious issue.) After all, whether the test is one-sided or two-sided, you are risking a false positive over and over.

Bonferroni is one of the weakest ways to do this, though that weakness might not be so devastating with just three comparisons.

Dave
  • 28,473
  • 4
  • 52
  • 104