Michael Cha wrote
Thank you, Nick and Ronan,
My sample size is more than 1000.
If sktest is not that reliable, then is ther any other solution than
graphics?
pnorm and qnorm are very useful but the interpretation is not always
easy unless observed and expected lines exactly overlap.
Though, your comments were useful, Ronan.
>>> There is code for generating a reference portfolio of -qnorm-
plots
embedded in
http://www.stata.com/support/meeting/8uk/fortitude.pdf
See pp.27-28.
The issue is not that -sktest- is unreliable. It does what it claims
to do. But especially with large sample sizes, it rarely answers the
practical
question, do I have important non-normality? And what is "important"
must depend upon why you are interested in non-normality.
summarize v, detail gives the numeric value of skewness and kurtosis.
Can they be more accurate?
>>> They are measures which some people use.
They are very sensitive to outliers, which
some see as a virtue. But that means that they mix sensitivity
to general shape with sensitivity to particular quirks, so that
a unique verbal interpretation is not always easy.
Then, can you guide me one more time if I am wrong?
As far as I know with summarize v, detail values,
Skewness: value 0 = not skewed
negative value = positively skewed
positive value = negatively skewed
>>> Positive = positively skewed, etc.
Kurtosis: value 3 = normal
greater than 3 = slim
smaller than 3 = fat
>>> There are many papers debating precisely what
kurtosis measures, especially from time to time in the
American Statistician. Broadly, tails which are longer
or fatter than those of the normal (Gaussian) mean
high kurtosis.
In practice, skewness and kurtosis are more closely
yoked than some texts imply.
Nick
n.j.cox@durham.ac.uk
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/