ISSN Online: 2377-424X
ISBN Print: 1-56032-797-9
International Heat Transfer Conference 11
REDUCING THE EFFECTS OF COlliNEARITY IN REGRESSION OF HEAT TRANSFER DATA
Résumé
Collinearity among dimensionless groups of the-independent
variables may often prevent obtaining statistically
valid, reliable correlations. In this paper
commonly used collinearity indicators (such as condition
number of the normal matrix, variance inflation
factor and confidence intervals) and a new indicator,
"truncation error to noise ratio" (TNR) are used to
investigate the level of collinearity and its harmful
effects. It is shown that the dominance of one particular
variable in several dimensionless groups is a
common source of collinearity. Proper experimental
design, which considers the range and precision of
the independent variables and use of several different
fluids can minimize the harmful effects of collinearity.
From among the collinearity indicators tested,
the TNR has proven to be superior in predicting the
cases where collinearity prevents obtaining statistically
significant results.