-1
$\begingroup$

What is the best method for testing for normality? I have a smaller data set today (30 in each group) and on the histograms none of them look normally distributed at all, whereas with the skewness test skewness<2X standard error of skewness, they all appear to be normal. I also did the shapiro wilk test, which told me that 8 out of my nine variables were normally distributed?

$\endgroup$
  • 1
    $\begingroup$ Shapiro-Wilk told you that one of your variables is not normally distributed. It can't tell you if the variables are normally distributed. The opposite of "significant" is "I don't know". $\endgroup$ – Horst Grünbusch Nov 25 '14 at 12:22
  • $\begingroup$ what?im asking which test is the most accurate... $\endgroup$ – Zoe Campbell Nov 25 '14 at 12:29
  • 2
    $\begingroup$ It depends: there are many ways in which a variable can be non-normal and some test are more sensitive to one type of deviation and others more sensitive to other types of deviations. Why do you want to test for normality? The reason I ask is that it is a very common error to test for normality when you should not do so. If that is also the case for you (very likely), then we have a very simple solution... $\endgroup$ – Maarten Buis Nov 25 '14 at 14:21
  • 2
    $\begingroup$ The answer to your previous question validly points out that this issue is irrelevant. $\endgroup$ – whuber Nov 25 '14 at 15:00
  • 1
    $\begingroup$ What kind of answer are you looking for? "Do A and you will get the right answer"? You won't get an answer like that, and you should be glad, because such answers are always wrong. Instead you should understand what you are doing (and @whuber gave you a hint on where to start reading) or get a co-author on the project who understand what (s)he is doing. $\endgroup$ – Maarten Buis Nov 27 '14 at 15:53
0
$\begingroup$

Short answer: there is no BEST method for testing for normality. Different methods are superior than others in (1) detecting different types of departure from normality (2) simplicity and easy to use.

$\endgroup$

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