Half-Normal plot of residuals interpretation

Question

I fitted a regression model and getting trouble to interpret the Half-Normal plot of Pearson and Deviance residuals.

In general the fit looks good because the residuals are inside the confidence bands. But what more I can say?

I read this topic How to interpret a QQ plot but not sure if the interpretation is the same for Half-Normal.

How did you compute the curve in the second plot? Is that from some form of bootstrapping/simulation or some theoretical calculation? — Glen_b, May 28 '17 at 22:27
Thanks; I was asking about the curve in between -- presumably that's also simulation? — Glen_b, May 28 '17 at 23:25

score 2 · Accepted Answer · edited Jun 11 '20 at 14:32

When looking at regression diagnostics, reading a half-normal plot of residuals is fairly similar to reading the top half of a Q-Q plot of residuals, at least when identifying lighter or heavier tails than anticipated. However if one tail is lighter and the other heavier, that may be less easy to identify in a half-normal plot (though much more or less skewness than expected will still show up as a wiggle)

(Half normal plots are sometimes used on estimated effects to identify "important" effects from "unimportant"; that doesn't seem to be what you have here)
In a GLM of course you don't expect the residuals to be actually normal unless the family was normal. The deviance residuals plot you show, where the expected "line" is actually a curve may be a little harder to judge; at least you have the "curve" to help you.

1 Answers1