In my experiment, I hypothesised that individuals in one treatment condition would give higher values on a likert scale than individuals in the other treatment condition. It was a one tailed hypothesis.
Histograms and Q plots show that my data is skewed in the direction I would imagine for the first treatment condition – i.e. all clustering around the higher values in the Likert scale. This isn't the case, again as expected, in the second treatment condition.
Even though I predicted this, is this skew still bad? Should I always use non-parametric tests in this situation (the results of parametric and nonparametric tests are the same – I just want to make sure I am using the right tests!).
Any help would be really gratefully received!
Thanks!
Best Answer
Welcome to the site.
Skewness isn't "bad" (nor is it "good"). Skewness may be a violation of assumptions of a test, in which case either figure out a way to fix it, or use a different test.
If you are comparing two single item Likert scales, then all your values will be integers (say, from 1 to 5 or whatever), in which case the mean is going to be related to skewness: You can't have a high mean on a 5 point scale without having left skew.
It is also arguable that even taking the mean on a single item scale like this is unjustified: The scale is ordinal, not interval (actually, it's in-between).
I suggest a nonparametric test. Which one did you use?