Hi everyone

I hope someone can help me with this statistical analysis - I've been staring at it for so long now that I seem to have lost any ability to think this through, and have managed to completely confuse myself!

I'm trying to compare self-reported behaviour by two different groups of people (one group went through a pilot behaviour intervention programme, the other group didn't). The data is ordinal. I think I should be using a Mann-Whitney U test to identify any differences, and I've calculated the results of this, but when I look at the example in Andy Field's book of how to report the results, I can't see how this fits with the things I've been looking at. So now I'm wondering if this was the wrong test after all!

Does anyone have any helpful suggestions? Or am I just being a complete idiot driven blind by staring at SPSS for too long?

Thanks in advance!

Hi Bat,

Just looking at the intervention program, I would have thought you had a parametric data? In which case your test statistics should not be using Mann-Whitney U?

Anyway, Mann-Whitney U is the same as Wilcoxon-rank-sum test, this is used for non parametric data. As your data is non-parametric and examining the difference between two groups, then it sounds liket the right one to use.

D

Yes, the data is non-parametric (if you count this kind of scale data as ratio rather than ordinal). So maybe I am okay then. Reassurance is good so thanks for that! (Have a star...)

Bat, if it is ratio it is parametric, not the other way round!

D

======= Date Modified 20 Jan 2011 16:38:03 =======
No, parametric is to do with whether it's normally distributed, isn't it? Ratio data isn't always normally distributed, so can be non-parametric. Ordinal data - well, isn't there disagreement about whether it can really be parametric at all? Anyway, either way, mine is non-parametric, even if you think that scales formed from ordinal data provide ratio results....

I think I know what I mean, even if nobody else does...!

I think Mann-Whitney U test sounds right. You basically want a non-Parametric version of a Independent Samples t-test, right?

Out of interest what doesn't sound right from Andy Field's example?

Yes, I think that is what I want.

Andy Field's example of reporting included things changing over time - so he's saying at first there's no difference between two groups (and reports the non-significant findings) then says that after x amount of time, there was a difference between the two groups (and reports the significant findings). Looking at it again now, I think I can see how my results would fit the way he's explained it - I think maybe lack of confidence with stats just caused me to think I was missing something. I think my results would be that for most of the things I'm measuring (different behaviours) there was no difference between the two groups (and I'd give those figures in a table) but in one behaviour there was a difference, and this difference was y.

Thanks, just thinking it through on here has really helped!