you can measure the global correlation of one variable given another(Y/X) or
(Y/X) . This measure is the percentage of variance due to a funcional
relation between the variables. The squared linear correlation r^2 is the
percentage of variance due to a linear dependence, thus its value is always
inferior than the two global correlation values.
To calculate the global correlation you have first to classify de data
building a bidimensional table then compute the condional means and
variances to and with those calculate
global C. Y/X = e.V./Vtotal(Y)
where
Total variance of Y = mean Variance of the condicional variances +
variance of the condicional means and
e.V. - explained variance = variance of the condicional mean (with respect
to the global mean)
After that
global correlation coeficiente Y/X - R^2 = % of the variance of Y dur to
non-linear relation with X.
Good Work!
--
Rui Jorge de Almeida Soares Gonçalves
[EMAIL PROTECTED] ou [EMAIL PROTECTED]
SMF - Dep. de Engenharia Civil
Faculdade de Engenharia - FEUP
Tel.\Fax 351 2 2041921
R. dos Bragas, 4050-123 Porto Codex
PORTUGAL
