significance test for qq plots

[Tony Josipovic]

Hi everyone,

I have made qq plots for my data. Now I am wondering if there is a test for significance of the qq plots.

Thanks.

[Nick Cox]

What hypothesis do you want to test?

[Jackson Monroe]

Tests for normality, what I assume you will answer to #2, are not very good and often performed for no good reason whatsoever.

[Tony Josipovic]

What hypothesis do you want to test?

Hi Nick,

one parameter has been estimated for each user two times, e.g. time between the two sessions was e.g. two monthes. Now I have made qq plots of the variable/parameter from one month and the other month. I want to test if the distributions are the same. If the dots in the qq plot are on the diagonal, then they are the same. But since they deviate from the diagonal, I would like to conduct a significance test for the two distributions.

[Nick Cox]

Thanks for the further detail, but I can't easily follow what you want or need. What parameters you estimate has no obvious bearing on whether distributions are the same or different, unless you are looking at results that depend on a previous estimation. Setting aside the latter, the Kolmogorov-Smirnov test is classical, but I agree with the advice quoted in {R] diagnostic plots from Rupert G. Miller that such tests add little or nothing to what the plot shows. (Full disclosure: I sent in the quotation to StataCorp for inclusion in the manuals.)