Experiment shortly: Participants pre-learn one sequence which is considered to become "automatic", then on the day of the experiment they learn a new one which is considered "non-automatic". The sequence is performed first by hand and then by foot. ( taps on the keyboard and stomps on the floor). Additionally, the sequence is performed within the context of a "single task" (only perform the sequence) or "dual-task" ( perform the sequence and perform a letter counting task in parallel).

I want to look at the interaction effects between these 3 factors.

My main hypothesis are:

1. Non-Automatic sequence is performed significantly worse in the context of Dual-Task.

2.. Automatic and Non-automatic sequences are equally well performed in the context of the Single Task.

3. Automatic sequence is equally well performed in the context of Dual-Task and Single-task

4. There is no difference in performance between hand and foot.

I thought about doing a 3-way ANOVA, however when I test for normality and equality of variances my p values show that I violate these 2 data assumptions about the DATA required for ANOVA.

I tried to transform my dependable variable (performance_error rate) using sqrt and log, but it did not change anything. My variable expresses the percentage of errors(from 0(no errors) to 1(all wrong)). But most of the times participants have either zero errors or a lot of them, there is little "in-between".

Is there a way I could transform my data in such a way to still be able to run a 3way-ANOVA or what is the other test I could use?

I would highly appreciate any help!