Tom Steichen has supplied the code for Grubbs' test.
However, if your underlying idea is -- as I understand
is implied by Grubbs' test --
it's Gaussian (except that there are outliers)
then this isn't really a problem soluble by
transformation. Any transformation that pulls in outliers
will give you something that's not Gaussian, because of its
effect on the Gaussian lump. Conversely, if the underlying
idea is
it's skew (and that includes some outliers)
then a transformation is one natural solution, and
Grubbs' test looks irrelevant.
On detection, I suggest that boxplots are fairly lousy
graphical detectors, given the amount of information
they discard. You are better off with -dotplot- or
something more like -symplot- or -skewplot- from SSC.
-extremes- from SSC is also available. It is
a tabulation program designed to allow focus on
extremes (which may or may not be outliers).
Nick
[email protected]
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of
> Janet Oliver
> Sent: 20 January 2004 15:22
> To: [email protected]
> Subject: st: Re: Outlier detection
>
>
> Can I detect outliers in Stata. A boxplot of my data shows
> extreme values
> and I cannot find a transformation to normality. I am
> unhappy at just
> discarding results because they are "extreme" and was
> wondering if there is
> an implimentation of Grubb's or Dixon's test, or indeed any more
> satisfactory test.
*
* 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/
