[R] Outliers

[Marc Schwartz]

On Sun, 2004-07-04 at 19:41, Richard A. O'Keefe wrote:
> Last week there was a thread on outlier detection.
> I came across an article which has a very interesting paragraph.
> 
> The article is
>     Missing Values, Outliers, Robust Statistics, & Non-parametric Methods
>     by Shaun Burke, RHM Techology Ltd, High Wycombe, Buckinghamshire, UK.
> It was the fourth article in a series which appeared in
>     Scientific Data Management
> in 1998 and 1998.
> 
> The very interesting paragraph is this:
> 
>     NB: It should be noted that following a
>     judgement in a US court, the Food and
>     Drug Administration 9FDA) in a guide -
>     Guide to inspection of pharmaceutical
>     quality control laboratories - has
>     specifically prohibited the use of outlier
>     tests.
> 
> Elsewhere, the article recommends the use of outlier tests as a way of
> locating possible transcription errors, but NOT as a way of discarding data.


The FDA Guide referred to in that article is here:

http://www.fda.gov/ora/inspect_ref/igs/pharm.html

If you search that page using the keyword 'outlier' you will note
several references.

The part of the document relevant to the above citation is:

In a recent court decision the judge used the term
"out-of-specification" (OOS) laboratory result rather than the term
"product failure" which is more common to FDA investigators and
analysts. He ruled that an OOS result identified as a laboratory error
by a failure investigation or an outlier test. The court provided
explicit limitations on the use of outlier tests and these are discussed
in a later segment of this document., or overcome by retesting. The
court ruled on the use of retesting which is covered in a later segment
of this document. is not a product failure.



Some of the above and elsewhere in the document, relative to grammar and
punctuation, suggests that the HTML page was converted from another
format, perhaps Word or PDF. Some things do not quite make sense, but
you can get the basic idea.

Note also the use of the word 'limitation' above rather than
'prohibited'.

HTH,

Marc Schwartz