3
$\begingroup$

I have four individual groups with two controls and two interventions. The second control group is specifically related to the first the intervention group, making the comparison to the second intervention group unnecessary.

Differences between the groups are being analysed by the Kruskal-Wallis test followed by Dunn's test for multiple comparisons (dunntest command in Stata). It's been suggested that I remove the pairwise comparison between the second control group and the second intervention and then correct for 5 multiple comparisons rather than the 6 run by the test. Is this an appropriate thing to do?

$\endgroup$
  • 2
    $\begingroup$ How did you determine that the comparison was unnecessary? Did you know this prior to collecting the data? $\endgroup$ – Michael R. Chernick Feb 18 '18 at 2:29
  • $\begingroup$ It was determined prior to collecting the data. The first control group is no intervention at all. The second control group was added purely to correct for any placebo affect associated with the delivery method of the first intervention group. The second intervention group has no delivery method and thus the appropriate control is the first control group. $\endgroup$ – user195769 Feb 21 '18 at 9:33
2
$\begingroup$

This is an appropriate thing to do. For your example, the number m (number of tests) would equal five, rather than six. However dunntest in Stata (and dunn.test in R) assumes your will be performing all 6 multiple comparisons. Therefore, while dunntest can provide you with the unadjusted p-values for each pairwise test, and you can simply direct your attention to the 5 your care about, you will need to perform the multiple comparisons adjustments by hand or using some other software.

I am the author of dunntest, and can talk to you about how to work out the arithmetic if you want to email me (my contact info is in the help file).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.