I've read multiple posts/papers citing Tabachnick and Fidell's cut off of +/- 1.5 as the acceptable range for skewness and kurtosis to determine normality; however, I cannot find it in their book. Can someone please tell me where Tabachnick and Fidell have stated this cut off? Book and page number please.
Asked
Active
Viewed 343 times
0
-
3The deeper question is why on earth is this taken seriously? 1.4999999.... is fine, 1.5 requires remedial action? In any case, plenty of distributions that aren't normal at all will satisfy this criterion. (Detail: presumably the kurtosis here is so-called excess kurtosis, kurtosis $-$ 3.) – Nick Cox May 21 '23 at 09:13
-
2A normal is often a pertinent reference distribution, but the best check is a normal quantile plot (aka normal probability plot, normal scores plot, probit plot). I don't have access to this book to check if they say that, but it's very common advice here on CV. Besides, having marginal normal distributions is often not essential for most techniques. – Nick Cox May 21 '23 at 09:25
-
1In this post https://stats.stackexchange.com/a/481022/102879 you will find examples of discrete distributions (similar to Likert scale distributions), which are highly discrete and obviously non-normal for that reason alone. One is more "peaked" than a normal distribution and has excess kurtosis -.54. The other is flat-topped, unlike a normal distribution, and has excess kurtosis 0.45. Both numbers are well within the $\pm 1.5$ bounds. – BigBendRegion May 27 '23 at 12:54