I am trying to run a mixed anova with the following factors..

between subject factors: group (3 levels: groups 1,2 and 3) education (2 levels : low education/ high education)

within subject factors: task (2 levels)

so there are 12 conditions.

Shapiro Wilk test indicates that 4/12 conditions are not normally distributed (p=0.006, p<0.001, p=0.040, p=0.002).

I've been told that it might be okay to proceed with this anova seeing as its only 4/12 conditions which are not normally distributed, and of these, one of them is quite close to being normally distributed (p=0.04). Am I right to proceed?

If so - how do I justify/explain this in my write up? I'm finding it quite hard to know how to word, e.g. I am not too sure I can write "Although 4/12 conditions are not normally distributed, I decided to proceed with the anova regardless because my supervisor told me I could". haha

And then when I'm analysing the results of the ANOVA, should I write some sort of disclaimer that my results might be invalid? Also if this is not the case, and I shouldn't proceed, is there a better test that I can use?

Any help or tips would be much appreciated, I'm new at this :)


1 Answer 1


In ANOVA, the residuals should be normally distributed, so you will run the Shapiro Wilk test just one time (and not for each condition you have). But you have a within subject (two levels) and i think you should measure the residual for each task and then use Shapiro Wilk test. So you will have 2 Shapiro Wilk tests for these two residuals.

  • $\begingroup$ Sorry I don't understand your comment. What's a residual? I've already run the Shapiro wilk test, are you saying I should've done it differently $\endgroup$ Jul 29, 2017 at 17:34
  • $\begingroup$ @reckrockaree For ANOVA, you dont run Shapiro Wilk test for each factor. If residuals are normally distributed, your data will be. In a classical ANOVA you have to calculate the unstandardized residuals and runs the Shapiro Wilk test. Because you have a within subject factor i think the test will be a little different (because residuals may be dependent with the within subject factor). Take a look on how to perform ANOVA´s assumptions for split plot or nested experiments. $\endgroup$ Jul 31, 2017 at 10:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.