0

I am trying to show, that my errors in my multiregression model is normally distributed. I have the values for Skewness and Kurtosis. Can I use Bowman Shelton test to show that?

user265686
  • 1
  • You can use tests for normality, but in my experience qq plots are better – Stochastic May 18 '20 at 19:31
  • Check https://stats.stackexchange.com/questions/2492/is-normality-testing-essentially-useless – Tim May 18 '20 at 19:53
  • Thank you Niel van Zyl and Time. When using the qq plot and pp plot to show normality does it make a different if I square the residuals, because it seems that it is easier to see if I reject the normality condition or not? – user265686 May 18 '20 at 20:08

1 Answers1

0

I would actually suggest a qq plot instead of a test. If you are working in R you can use:

qqnorm(errors)
qqline(errors)

Here is more information on qq plots.

Stochastic
  • 799
  • 1
  • 6
  • 28
  • QQ plots are evocative, but I've never made a decision based on one. Could you give an example where a QQ plot causes you to make one decision and test would suggest that you make a different one? And how did you use the plot? This could be a *very* interesting answer. – Peter Leopold May 18 '20 at 21:50
  • Well, maybe it is just the way that I have been taught. In undergrad lecturers made us use tests for normality, where as in post grad they suggest that we rather use qq plots. I have been told that if the test yields a p-value of 0.1, you will still assume normality (testing at 5% significance). Using a qq plot for that same data one could easily see that it was skewed to the left. Unfortunately can't find the specific example that I am referring to now. – Stochastic May 19 '20 at 08:47
  • To clarify, I believe especially for small samples normality tests can be fairly inaccurate. – Stochastic May 19 '20 at 14:49