This post is a reply to the recent post “Normality and Testing for Normality” on Tom Hopper’s Learning As You Go blog. In that post, Tom aims to explore, among other things, in how far normality tests, specifically the Shapiro-Wilk, test are “sensitive to the tails”. I really liked the post, it contains good ideas and good reasoning and I think it’s worth reading first of all. I’m sceptic, however, whether the specific method used there for exploring the “sensitivity to the tails” does really examine what it’s intended to examine. I’m responding to describe this scepticism, and try out an alternative approach.
(This was originally posted on my old wordpress blog and not imported in full text.)