2-way ANOVA or not 2-way ANOVA?

#1
Hi,
I'm currently doing an experiment assessing the effect that 4 different growth mediums have in 3 different O2 concentrations. There are 6 different flasks in each of these conditions - so 60 in total (we lost 12 flasks to infection :(. I am assessing percentage estimated confluency (what area of the flask is covered in cells) on these flasks and comparing the effects that both the different media conditions have on this and then the effect that the different oxygen concentrations have on the different media groups. Please correct me if I'm wrong but the way I see it is:
1) I'm assessing the difference between multiple groups and thus need to use ANOVA? Whether to use one-way ANOVA or two-way ANOVA I'm still not sure of.. 2) I cannot assume that the data are normally distributed - I did a Shapiro test for normality and the outcome was that n (which was 6) was too small. I then compared p values between an ANOVA and a non-parametric test (Kruskal Wallis) of the difference between the media groups alone (if I were to use 2-way ANOVA I guess I'd have to do a Friedmans test?) - so I've concluded that as they are so different it's 'safer' to use a non-parametric test? 3) The values I'm estimating are percentages and thus cannot be used in these tests unless they are modified - or is there a completely different test I need to use?
Many thanks in advance
 

Miner

TS Contributor
#2
Start with a 2-way ANOVA then diagnose your residuals plots. The raw data need not be normal, just the residuals. If you have problems with the residuals, you may then go nonparametric.