Shapiro wilk effect size

#1
Hi,

Do you know how to calculate the effect size for Shapiro Wilk test?

When using Shapiro Wilk test on a large sample, most likely it will reject H0 even for normally distributed population, since large sample will identify the minor effect as significant.

Thanks :)
 
#2
You might be interested in this Cross Validated thread.

Basically, I think the conclusion in that thread is that there is no uniform definition for the deviance from normality, since it can be defined in many ways.

I think most textbooks advise not to use the Shapiro Wilk test anyway and just use a Q-Q plot and inspect the matter graphically. Obviously, that does raise the question how to decide when the data are too non-normal. But in many cases violation of the normality assumption is not that critical, anyway. What is the context for your question?
 
Last edited:
#3
Thanks Junes,

I read the thread (and again) and they mainly say what is not effect size ...(not W...)

I thought the recommendation is to use a combination of Shapiro-Wilk and QQ plot.

I don't really believe we need to use our eyes instead of the computer ... and I suspect a combination of Shapiro-Wilk test and effect will give a better result than only Shapiro-Wilk test, like in any other test.