Hello to everyone,
I have collected personality perceptions on a semantic differential scale (1 (one extreme) to 9 (the other extreme) for 5 traits) from one sample (n=53). I have actually asked them to rate their Farsi personality as well as their English personality (not a personality but a self-perception test). I aim to compare means for L1 and L2 and see if there are any differences between their perceptions. Is test language a condition that makes this design a repeated measures or should I consider L1 and L2 personalities as separate entities that I'm testing? The data aren't normal or symmetric, please guide me on which statistical test I must use for this scenario. Thanks a million for your help and time.
Regards
I don't think you should consider it a repeated measure. The words are different, after all! Whether the personality test is equivalent in English and Farsi is an open question - may even be what you are studying. (The complicated piece for you, interpretively, is whether any difference you find is owing to language alone, that is, words that aren't really 100% synonyms, vs. actual differences in how people perceive themselves when they place themselves in one language/culture or another. It's a very interesting question, though!)
The fact that your data are not normally distributed suggests that you might want to use nonparametric statistics. So instead of a paired t test (if you're comparing people on just this variable and only considering language as an IV) you might use a paired Wilcoxon signed-rank test.
Thanks a ton for your answer, much appreciated, words aren't different actually the constructs I'm asking about are, one is L1 personality one is L2 personality, I've attached one page of the survey to make sure. But the sample is dependent so the test is a paired one right? Does Wilcoxon signed rank have an assumption of symmetry? My data was not symmetrical so will I be left with sign test? Again thanks for sharing your information. It's of so much value to me.
So the survey is completed in the same language both times, with the only difference being the 'manipulation' of asking them to think of how they self-perceive when in different cultural contexts? Then yes, it is a repeated-measures design.
The sample is dependent, yes, which gains you a lot of statistical power by controlling for across-subject variability.
The Wilcoxon, like other nonparametric tests, does not assume normally distributed data. These are useful significance tests when data violate the assumptions underlying the more common procedures, such as t-tests and ANOVA. So it is up to you and your advisors whether to go ahead with a repeated-measures ANOVA (ignoring the fact that the data aren't normally distributed) or use an analogous nonparametric procedure - though I would recommend the latter, unless a transformation would render the data more normal. (Sometimes a square-root or logarithmic transformation does the trick.)
I wish someone else would chime in here, though. I do not pretend to be a serious expert in these matters, and usually questions like yours get several answers. If a consensus emerges it is very comforting.
- Badges
- Science topic
- Similar topics
- Psychology