10-12-2014 03:27 PM
I have two groups of each gender and I need to compare two sets of scores for one gender each (a and b class for gender 0 & c and d class for gender 1). In comparing these scores, I am looking for the differences between the two scores with a large sample size (~50 samples). My question is should I be attempting to use a nonparametric method (like that of a WIlcoxon ranked sum test) or a parametric method (like that of a T-test). T-test shows the difference of the means as evident from the output file I ran, while Wilcoxon Ranked sum seems to give me variable data on the Wilcoxon scores for each class. Any help would be appreciated.
10-12-2014 04:00 PM
T-test inference assumes normality within classes. Wilcoxon doesn't, at a small loss of power. 50 is not quite large a sample size (sorry) for deciding on a distribution. It usually turns out that both parametric and non parametric analyses give the same inference when parametric hypotheses are true. Disagreement between the two is a sign that parametric assumptions (i.e. normality, homoscedasticity) are not met or that the data includes outliers.