2

What to do when qqplots aren't as nice as they should be?

Do you disregard the model and go look for another one? If so, how? How do you figure out from a qqplot what model you should look for?

Or do you continue with the model but stress that it may not be the appropriate one?

I am doing some exercises in my stat class and all our model tests work out perfectly, which I think is annoying, because how is that going to teach me what I should do in real life when things don't always work out neatly?

Chill2Macht
  • 5,639
  • 4
  • 25
  • 51
Mah
  • 21
  • 1
  • 2
    Are you referring to a qq-plot of the *residuals* from a linear (OLS) regression model for assessing normality? – gung - Reinstate Monica Mar 22 '17 at 12:28
  • I've edited your title. "staticians" was clearly a typo, but as there are many people here who aren't statisticians (I'm one), I've tried to preserve the main question without specifying who should answer. – Nick Cox Mar 22 '17 at 13:09
  • 1
    There are many threads here that may help. http://stats.stackexchange.com/questions/111010/interpreting-qqplot-is-there-any-rule-of-thumb-to-decide-for-non-normality/ includes excellent material and also an answer from myself emphasising an obvious but sometimes overlooked principle that you know that a quantile plot can be improved on when you find a better one. – Nick Cox Mar 22 '17 at 13:14

0 Answers0