Different variables - most normally distributed, a couple are not - best test to use?

#1
Hello everyone

So glad I've found this site and hope that someone will be able to point in the right direction!

I am looking at how physical activity scores (mins per week) in children, change pre and post intervention - as part of this I want to see what the affect of parental involvement is (measured by attendance at sessions) and I have categorised attendance into low, medium and high. The dependent variable is the children BMI - and I have calculated this as the change score between the pre and post test measurements. My intention was to do a 3x2 Anova

However, I've run the normality tests on all the variables and both my baseline and post physical activity data and BMI score are kurtosed but only my BMI score does not meet K-S significance levels (p=.032) - is the best thing to do is to run a non-parametric test on the pre and post test physical activity and then do a Anova for the other variables that meet the assumptions of a parametric test?

if I don't include the physical activity scores how can i show which variable had the most affect on childrens BMI?

My sample is only small (n=23)

This is my first attempt at statistics I've read loads but its only confusing me even more!

Thank you very much in anticipation