0

I am attempting to analyse pre-existing data (using SPSS). I have one group of participants, 4 quantitative, non-parametric outcome measures, and am looking to see if there is a change in outcomes post intervention. I also have qualitative outcomes, but am not worrying about them yet! I understand that t-tests (and the non-parametric equivalents) can be used for comparing several groups (i.e. if I had a control group as well), but don't know what to do with my single group of participants.

kjetil b halvorsen
  • 77,844
AJay
  • 1
  • 1
    Do I understand correctly that you have data on your outcomes for the same participants before & after the intervention? If so, there is a sense in which you do have 2 groups, & you can run a paired / dependent / repeated-measures t-test (different names for the same thing). However, I have no idea what it means to say that you have quantitative outcome measures that are "non-parametric". Can you clarify this? – gung - Reinstate Monica Oct 14 '12 at 15:41
  • Like @gung I am confused by your use of "nonparametric". Data cannot be parametric or nonparametric. Tests or statistical methods can be.

    You should also be aware that a single group pre- post- design is a very poor design, especially if your measures are not extremely reliable. You will be correlating error with change.

    Some have even suggested that such analyses not be done as they are inherently misleading.

     – Peter Flom Oct 14 '12 at 15:44
  • @PeterFlom, I wasn't aware of anyone saying that paired t-tests are unreliable. Do you know who says this / have some reference where I could learn about the issues involved? What is their recommendation in such a case, to run the pre & post as though they were independent groups? Certainly we prefer not to have variables w/ a lot of measurement error, but once you're in that situation, what are you supposed to do? – gung - Reinstate Monica Oct 14 '12 at 15:49
  • @Gung the problem isn't paired t-tests, the problem is pre- post- one sample designs. I think there was a chapter on this in Collins.

    The basic problem is that if observed score = true score + error, then change in observed score will be correlated with error. This is related to regression to the mean.

     – Peter Flom Oct 14 '12 at 16:01
  • I think this is also in the first pages of Hedeker & Gibbons – Peter Flom Oct 14 '12 at 16:01
  • Yes, I have data on outcomes for the same group before and after treatment. I guess I worded this wrongly in saying the data was non-parametric. What I was meaning is that my outcome measures are ordinal and therefore I need to use non-parametric tests (if I've got that right?) sorry, if you hadn't figured, I'm pretty new to all of this and still trying to get my head around it. And yes, I am aware that it is a very poor study design but I didn't have any say in that, I was merely give the data and told to analyse it - this is my dissertation and I didn't get to choose it! – AJay Oct 14 '12 at 16:27
  • So yeah, am I right in saying I should use a dependent t-test using my pre treatment measures as one group, and my post-treatment outcomes as another? – AJay Oct 14 '12 at 16:59

0 Answers0