I wonder why it is that some fields of research dont seem to be sensitive to
outliers even when using non-robust estimations of covariance.
Is there some special filter, or way the data comes in, that does the outlier
rejection for you?
We are very sensitive to outliers, so are "stuck" doing only univariate (cuboid)
test limits in production,
rather than the ellipsoidal test limits that would both:
* increase our yield (saying good products are good rather than rejecting them
as we do now)
at the "ends" of the ellipsoid
* increase our rejection of bad product (rather than accepting them as we do
at the "sides" of the ellipsoid
Product Test & Validation
Delphi Delco Electronics Systems
KOKOMO IN 46904-9005
[log in to unmask]
"Expressions are my own, not my Company's. No warranty shall be implied."