3

Here is a question on how to test for differences in distribution between three samples of multimodal distributed data.

I have conducted a dictator game (http://en.wikipedia.org/wiki/Dictator_game) where respondents have been randomized into three treatment groups. All samples show bimodal/multimodal distributions. The confirmatory hypotheses where to test for differences in donation behaviour across the treatment groups, more specifically test for means and proportionate differences. However, since the samples turned out to be multimodal, I fear that tests for differences in donation means may not be as useful as I thought. Instead, I intend to test for differences in distributions. If I had only had two samples, I would have used a Kolgomorov-Smirnov two sample test to compare the distributions. Do you know if there is a version of the K-S test that tests for differences in distributioins across >2 samples?

Other suggestions how to interpret the data are welcome as well.

Additional info.

  • I am working in SPSS.
  • Good understading of statistics, but not a pro.
Learner
  • 1,528
  • 3
  • 18
  • 34
Persson_Sweden
  • 31
  • 3
  • Well, [yes](http://projecteuclid.org/euclid.aoms/1177706261) -- but I'm not sure that multimodality would necessarily invalidate a question about means. – Glen_b Apr 17 '15 at 11:38
  • Thank you for your response. Maybe not invalidate, but by all means (pun intended) make it less relevant. I would compare differences in values but there is almost no mass in that region. Wouldn't you agree? – Persson_Sweden Apr 18 '15 at 09:22
  • The mean outcome on a six-sided die is 3.5; there's exactly zero mass around that value... yet people sometimes value in inference about the mean of it. But what matters if you see value in it (which it doesn't sound like). If that paper looks of any value I could write an answer. – Glen_b Apr 18 '15 at 09:23
  • I understand your example, but not your point. Are you saying such test could be useful, or not? – Persson_Sweden Apr 18 '15 at 09:35
  • 1
    A test for change in mean for a multimodal distribution *may be useful for some people or some circumstances*, even when the population mean doesn't have much of the distribution in its neighborhood. But it may not have value for you. – Glen_b Apr 18 '15 at 09:38
  • OK, fair. Thank you for your responses. I will look into the paper. – Persson_Sweden Apr 18 '15 at 09:51

0 Answers0