In an experiment I exposed each of 4 subject groups to a different condition to observe the impact on dependent variable. Assumptions for parametric tests were violated hence I decided to employ non-parametric Mann-Whitney for group comparisons. What I do not understand is why I have different scores of mean ranks in the output tables depending on comparison. For example, when I compare group 1 and 2, and then in a separate test group 1 and 3 the mean ranks for group 1 are different in output tables. When I run t-tests means remain the same in both tables. The conclusion I have reached is that it is due to the way Mann-Whitney is calculated but I also suspect I may have used the tests incorrectly. I would very much appreciate some help as I am unsure how to proceed. I have searched the forum for a similar query and could not find anything - apologies in advance if I omitted an answer to a similar question.

Best wishes,

  • 1
    $\begingroup$ 1. What assumption do "parametric tests" collectively make? 2. The ranks are calculated by ranking the two samples you're comparing. Obviously that changes if you change either of the groups. 3. Why are you doing two-sample tests pairwise? (rather than say a Kruskal-Wallis followed by suitable multiple comparisons?) $\endgroup$ – Glen_b -Reinstate Monica May 27 '16 at 12:24
  • $\begingroup$ Thanks Glen. To answer your questions 1. The main one violated is a non-normal distribution. 2. Are you saying you agree with my interpretation? I.e. mean ranks are different each time it's calculated. 3. Sorry I didn't provide enough detail, Kruskal-Wallis of course preceded tests I inquire about. Mann-Whitney are used as post-hoc tests with a correction. Hope this addresses your questions - thanks again. $\endgroup$ – George May 28 '16 at 16:27
  • $\begingroup$ 1. An assumption of normality is not a feature of "parametric tests" in general. 2. I was explaining how it happens. The problem is with your choice of test. 3. a number of posts on this site address why Mann-Whitney is not suitable to use for post hoc for Kruskal Wallis (including the reason you raise here) and offer better alternatives. I will dig up a link for you. $\endgroup$ – Glen_b -Reinstate Monica May 28 '16 at 23:29
  • 1
    $\begingroup$ See this answer by Alexis and also this one as well as this answer by gung which relate to the issue. $\endgroup$ – Glen_b -Reinstate Monica May 28 '16 at 23:41

Your Answer

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

Browse other questions tagged or ask your own question.