04-15-2017 11:30 PM
i have some data I am looking at characterized as follows:
Predictor: # of exacerbations defined as 0 for control, 1 if between 1-4, and 2 if 5 or more (n= approx 20 per group)
I have a couple of outcomes I am looking at based on questionairres:
Quality of Life
Asthma related QOL
+ 2 more
*Each of these scales is made up of 3-10 subscale measures. They are on the most part, skewed to the left. There is also heterogeneity of variances.
I want to look at differences in exacerbation group for these outcomes for my primary objective and look further into difference by gender, type of asthma med (category), severity of asthma (category), FEV1 (continuous).
What would be the best statistical approx given problems with assumptions, small sample for the primary and secondary (interaction by both the cat and cont vars?) objectives.
04-16-2017 08:02 AM
Certainly, MANOVA could work, assuming you don't mind that your errors won't be normally distributed (although maybe they will be close to normal, you'd have to check)
Another approach is PROC PLS.