Testing Significance of Group Differences with Non-Normal Data

#1
Hi.. I need help with deciding which statistical technique to use for my data
My research design involves 2 Independent Variables with 2 levels each (Gender and Age group) and I have 11 scores from a single scale as my Dependent Variable (continuous scores). My sample size is very large (Over 5000), but the test for univariate normality (Skewness and Kurtosis) showed that my data is not normally distributed. So can I use factorial MANOVA? or should I look for any alternate method? If yes, then which method?

More information on my DV- the 11 scores are-
1. a total scale score
2. total score on dimension A
3. total score on dimension B
4. 4 sets of Total scores on sub-dimensions of A
5. 4 sets of Total scores on sub-dimensions of B

There is correlation among these sub-dimensions.

Please help!
 

rogojel

TS Contributor
#2
hi,
just bear in mind that for large samples the normality tests are so sensitive that you will basically never see normal data as oer the tests and that actually the residuals need to be normal, not the starting data. So, I would do the MANOVA first and check the residuals.

regards
 

Karabiner

TS Contributor
#4
More information on my DV- the 11 scores are-
1. a total scale score
2. total score on dimension A
3. total score on dimension B
4. 4 sets of Total scores on sub-dimensions of A
5. 4 sets of Total scores on sub-dimensions of B
Does that make sense? You put in the same information 3 times (total score = dimension A + dimension B = subdimensions A1+A2+A3+A4+B1+B2+B3+B4).

With kind regards

Karabiner