I am testing the OLS assumptions and I got the following normal P-P plot. Does it look normally distributed or not?

enter image description here

  • $\begingroup$ How many data points is that, 140? 150? How many predictors? $\endgroup$ – Glen_b -Reinstate Monica Sep 13 '15 at 0:10

This looks fine to me, as far as that goes. What you are most concerned about is the tails of your distribution, and pp-plots don't show those very well. QQ-plots do better; see: PP-plots vs. QQ-plots.


There are formal tests for normality- in particular the Jarque-Bera/Skewness-Kurtosis type tests.

  • 1
    $\begingroup$ Jarque-Bera as I have seen it implemented uses asymptotic results for every sample size, a very bad idea as the convergence of sample skewness and kurtosis to normality is very slow indeed. $\endgroup$ – Nick Cox Nov 3 '15 at 19:21
  • $\begingroup$ is a K-S test better? $\endgroup$ – ChinG Nov 3 '15 at 19:29
  • 2
    $\begingroup$ Not to me. Almost none of these tests can possibly compare with a Q-Q plot, which shows you how much and where data depart from normality. A K-S test is arguably most sensitive in the middle of the distribution, precisely where departures are least important. Cf. @gung's comments. $\endgroup$ – Nick Cox Nov 3 '15 at 19:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.