HI! I'm really quite new to stats so I apologise for glaring foolishness :P

Basically in the study that I'm looking at is a within-subjects design. n=12, and each subject was tested before a treatment (no treatment, NT) and with a treatment (T).

In my first analysis the dependent variable was the overall number of errors that each subject made on the task. The distribution cannot be assumed to be normal. So I used a Wilcoxon test comparing the errors before and after the drug for each subject. I found that there was no significant difference between the overall number of errors.

The task involved three phases A, B and C. Before the expt we thought that the treatment T might have a selective effect on phase A. So we split the errors up into the different phases. This meant we had two variables now: T v NT and A v B v C.

I did a two-way ANOVA and found that there was a main effect of phase (p<0.05) and a phase x drug interaction approaching significance (p=0.06). A wilcoxon test was also conducted and found that there was a significant difference in phase A and C (but not B).

I concluded: there is a significant effect of treatment T on performance in phase A and C. The interaction approached significance; perhaps the study is underpowered as the p value approaching significance suggests that there might be a true interaction.

Is this OK? I'm slightly confused about:
1) Using a two-way ANOVA when the data is non-parametric?
2) How do I link the ANOVA and Wilcoxon data? What are the two getting at? Have I captured it above?

Thanks so much!