Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: testing normalitt of data


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: testing normalitt of data
Date   Mon, 29 Jun 2009 16:22:39 +0100

I don't understand all of this (what are all the standard error
results?) but I have comments nevertheless: 

1. Although normality is a formal requirement for the t test there is a
widespread view that it is not always very crucial. For more discussion,
see for example Rupert G. Miller. 1986. Beyond ANOVA. New York: Wiley. 

2. You cite here results for one set of data, but the whole point of a t
test is that you are comparing two groups. 

3. It is not very safe to base judgments of (non-)normality solely on
measures of skewness and kurtosis. Plot the data too if you are not
doing so already. 

4. Perhaps the easiest extra thing you can do is add a parallel
Mann-Whitney test and see how far the P-value matches that from a
t-test. 

5. The research problem is not stated here, but a -qqplot- or -dotplot-
is often more informative about a comparison of groups than a bare
t-test. 

Nick 
n.j.cox@durham.ac.uk 

El Hawary

can I use the t-test for independent samples when the skewness and
kurtosis of data are as follow
skewness: -1.232 & the std. error of skewness is .616/ .044 and the std.
error of skewness is .427
kurtosis: 2.195 & the std. error of kurtosis is 1.191/ -.505 and the
std. error of kurtosis is .833

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index