1

I am conducting a study on the relationship between rural/urban living, intentions and barriers to help seeking in adult males. I was planning to conduct a correlation and then pending that result, a MANOVA (no correlation) or a MANCOVA (should there be a correlation present).

I have used the theory of planned behaviour questionnaire (4 scores) and the barriers to help seeking scale (1 score) as my DV and the location as my IV (rural or urban).

When conducting normality testing, the skewness and kurtosis scores for all 5 DV's are within range +/- 1.96, however, my K-S and S-W results are in the majority p<.05.

Should I go on and do non-parametric MANOVA/MANCOVA (some variables correlate with others so I have to get my head around what to do about that, but that's a question for another day), if so, does anyone have any advice? I haven't come across data that fits skewness/kurtosis but not K-W/S-W before.

Thanks

Rachel
  • 13
  • 2
  • You are making 2 kinds of dichotomous decisions (correlations and distribution shape). The data do not have sufficient information to accurately make both of those decisions. Formulate a model that doesn't depend on making perfect decisions. In my [RMS](https://hbiostat.org/rms) book and course notes I discuss reversing the problem to predict the IV from all the DVs using a binary logistic model. – Frank Harrell Mar 03 '21 at 13:02

1 Answers1

0

If you have a very large data set, that could be part of the reason. Slightly non-normal distributions in large samples are more likely to be "significantly non-normal" according to KS/SW tests.

P.P.
  • 201
  • 1
  • 6