|
PureBytes Links
Trading Reference Links
|
I recall someone posting in the past few days about using regression
analysis with Omega products.
The following just arrived from the epidemiology e-mail list (not too
many other trades track diseases, eh?):
"[T]he use of classical algorithms (e.g., stepwise regression) to
obtain a "best" regression model can lead to highly misleading
interpretations of the data being analyzed, since their use does
not _automatically_ ensure that confounding and effect modification
features will be properly addressed. Such readily available variable
selection procedures are designed to include independent variables
in a regression model solely on the basis of precision considerations
(i.e., according to which factors explain the most variablility in the
dependent variable, as reflected, for example, by partial correlation
coefficient values) and _not_ on validity grounds. The control of
extraneous factors is primarily a validity issue involving the
assessment of factor relationships with the exposure variable(s) as
well as with the disease variable, and such validity considerations
must take precedence over those concerned with precision."
This is WHY the system I developed does NOT use classic statistical methods ...
and DOES measure the risk/uncertainty of each prediction.
Cheers,
Rob Lake
rbl@xxxxxxxxxxx
P.S. HTML scripting tutorial being developed for all those who asked ...
|