1
$\begingroup$

I've performed a post-hoc Dunn's test with Sidak adjustment using the dunn.test code for R. There are three groups in my comparison and I understand from the R help files that the lower number in the output is the adjusted p-value which I use to compare to alpha (0.05).

My question is: do I divide alpha by three (since I've three comparisons), or does the adjusted p-value generated by R factor in that I have three comparisons and I simply compare the value with alpha?

Thanks for any help (including pointing me to another discussion). Have googled in vain.

enter image description here

EDIT: I first performed a Kruskal-Wallis test and a signficanct difference in the groupings was identified. Also I am very new to R and have read to the best of my ability before posting (am not a statistician). I do understand from the dunn.test manual that the output from holm adjustment and others mark rejection with a asterisk. It is interpretation of the Sidak output I'm confused about

$\endgroup$

1 Answer 1

2
$\begingroup$

Author of dunn.test here. As per the help file for dunn.test under the sidak option for the method argument:

the FWER is controlled using Šidák's (1967) adjustment, and adjusted p-values = max(1, 1 - (1 - p)^m).

So, Yes. dunn.test corrects p-values for multiple comparisons given the argument to method. For the Šidák option, you would reject each null hypothesis if $p\le \alpha/2$ (i.e. dunn.test report one-sided p-values).

Also from the help file:

method adjusts the p-value for multiple comparisons using the Bonferroni, Šidák, Holm, Holm-Šidák, Hochberg, Benjamini-Hochberg, or Benjamini-Yekutieli adjustment (see Details). The default is no adjustment for multiple comparisons.

For the stepwise multiple comparisons adjustment methods, because rejection depends both on adjusted p-values (sometimes called q-values) and the ordering of unadjusted p-values, rejection decisions for a specified level of $\alpha$ are starred in the output.

The advantage of not starring output for the static results is that adjusted p-values can be interpreted by the reader according to their own preference for $\alpha$.

$\endgroup$
1
  • $\begingroup$ thank you sincerley for your answer. As I non-stats person I had tried to understand the method argument but was unable to interpret. Your expertise is much appreciated. $\endgroup$
    – Dory
    Mar 17, 2015 at 23:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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