Kwallis2 insignificant - need for comparisons between the groups? I have a question concerning the Kruskal-Wallis test. I compared 4 groups with the kwallis2 test (in stata) and got insignificant results (the multiple comparisons are aborted). When I look at the medians of the variable in two of the groups, one is double of the other. So I did a kwallis test for these two groups and saw it was significant. How do I deal with this? Is there a reason why the multiple comparisons are insignificant that I would still find significant results when comparing two of the groups?
 A: The normal way of proceeding is to check for differences with an omnibus test, and then if if the null is rejected to use post-hoc testing to identify the important differences (multiple comparisons).
If the null is not rejected by the omnibus test one normally stops at that point.
If you're going to go to multiple comparisons no matter what, what was the point of doing the Kruskal-Wallis in the first place? Just say "I'm going to test all possible pairwise differences".
You don't say what significance level the subsequent test was conducted at. It sounds like it was at the same level as the original test, but based on choosing which pair to test from looking at the data. Your nominal type I error rate is thereby rendered nonsensical (your results don't mean what you suggest they do, as your actual overall type I error rate is approximately that of doing all possible pairwise comparisons).
You decide your sequence of actions and consider the subsequent properties before you have data. If you're working out what to test based on what you see in the data, your tests don't have the claimed properties.
This is, in effect, significance hunting. It would be (or should be) treated as such by any decent referee.

Is there a reason why the multiple comparisons are insignificant that I would still find significant results when comparing two of the groups?

You'd need to say which multiple comparisons procedure compared to a test done at what significance level, but if the KW test was at the overall significance level one obvious explanation would be that multiple comparisons would be done at a lower significance level.
But you say the multiple comparisons were 'abandoned'. If they weren't done, you won't know if any would have been significant if carried out. It's possible that a KW could fail to reject even though one of the multiple comparisons would reject (even at a lower significance level). 
