Hello

Many times on this list voices against stepwise regression has appeared
unfortunately no substitutes have been proposed for selection of
'independent' variables which should be included in final model. I have data
from network of sample plots from mountain area. On plots indicators of tree
stand damage and tree characteristics were measured. I want to create model
for forest damage depending on plot characteristics like altitude, age etc.
Unfortunately 'independent' variables are not 'independent'. For example
tree age is highly correlated with altitude because older tree stands occupy
higher locations. I want to select variables correlated with tree stand
damage. I took all variables which I suspect can have influance on damage
and created a linear regression model with all variables. Can I use
calculated partial correlations between modelled damage and each
'independent' variable to select variables which I should use in final
model? If it is not wise approach what methods would you suggest?

Thanks for help

Rafal



.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to