The trouble with statistics by Jean-Martin

• To: forest@nic.funet.fi
• Subject: The trouble with statistics by Jean-Martin
• From: Valerie M LeMay <lemay@unixg.ubc.ca>
• Date: Fri, 24 Feb 1995 09:35:50 -0800 (PST)
• Content-Transfer-Encoding: 7BIT
• Content-Type: TEXT/PLAIN; charset=US-ASCII
• Resent-Date: Fri, 24 Feb 1995 09:35:50 -0800 (PST)
• Resent-From: forest-owner@nic.funet.fi
• Resent-Message-Id: <01HNFU0EO6YS8WYZRW@palmu.metla.fi>
• Resent-To: mesinfo@pihta.metla.fi

Jean-Martin:

I concur with Peter Elsasser regarding your concerns about the
affects of multicollinearity on your regression analysis.  There
is "hoots" of literature on this topic in the field of econometrics,
since the interpretation of individual coefficients is of interest
to economists. 

If you are truly interested in interpreting individual coefficients
(in terms of the signs and magnitudes) then the affects of
collinearity must be removed.  There are many tools to do this (come
see me sometime-- I am up the street from you in MCML 289).  If 
you are only interested in prediction of the dependent variability,
collinearity of a mild nature is not a problem.  It can be a
problem if extreme (to the point of not being able to find a
stable solution for the regression equation).  There are ways
of testing how severe the dependence among the predictor variables
is and how it is affecting your regression analysis.

Val. LeMay
Forest Resources Management
Faculty of Forestry
UBC
VAncouver, B.C. CANADA

• Prev: SUMMARY: Programs for analyzing spatial patterns-trees
• Next: The trouble with statistics
• Index: Forest mailing list February 1995
• Thread: Forest list subjects in February 1995