How to use nonparametric test with repeated measurements and many factors?

[Amana]

Hi,
I am investigating the effect of treatment A and B in two different groups of mice (Group A and group B).
The outcome is continous (ug insulin/ml blood).
I have several factors, they are all present in both treatment groups;
Sex (male, female), fixed factor
Method (concentration was measured in two different settings/methods, A and B, an it is clear that this is a main effect as one of them generally estimates the outcome to be a bit higher than the other one does), fixed factor
Age (5 weeks, 7 weeks, 9 weeks of age) Each mouse was tested at each age (always in the same treatment group), random factor (?)
The problem is that residuals are not normally distibuted. Before, I've used ANOVA (general linear model) and checked for main effect in eg. sex there and could have this as a main effect in the model. But now I have to use nonparametrics, and I've found Kruskal-wallis. But will that take into accout the effect of method?

I want to compare treatment A compared to treatment B, at 5, 7 and 9 weeks of age respectively.
I also want to see if there is a difference in outcome between all ages (if age matters).

[hlsmith]

Are the residuals approximately normal or are there some major issues of heteroskedasticity?

[victorxstc]

It seems that you want a non-parametric alternative for two-way repeated-measures ANOVA? We have discussed it before and this link [click] might help with its nonparametric alternatives.

[Amana]

Are the residuals approximately normal or are there some major issues of heteroskedasticity?

There is a major issue with heteroskedadticity, where p<0.01 with ryan-joiner normality test.
I found how to do a Friedman test in GraphPad prism, with the repeated measures, but I couldn't include the factor of "method" there. Can the results from the two different methods be compared at all? I'm wondering if it is the same at doing two different experiments, you can't include your samples in the same model as they had different conditions.

[victorxstc]

I found how to do a Friedman test in GraphPad prism, with the repeated measures, but I couldn't include the factor of "method" there. Can the results from the two different methods be compared at all? I'm wondering if it is the same at doing two different experiments, you can't include your samples in the same model as they had different conditions.

Please have a look at this post which conveys more details about a two-way Friedman is SPSS. Besides, you might combine a Friedman with a Mann-Whitney U, or try data transformation as stated here.