I totally agree. One should never employ these techniques if the
relationship is NOT linear or until some transformation to achieve
linearity takes place.
Raphael
On 12/18/06, Nick Cox <[email protected]> wrote:
I don't want to deny the users of -pcorrmat- any
benefit they may derive from it, but deeply embedded here
is an elementary statistical fallacy. A variable can be (almost)
uncorrelated with another (meaning that the correlation
is (near) zero) but nevertheless not independent of it.
Thus a near zero correlation does not inevitably carry
no information about a relationship as it could easily
represent some nonlinear structure (e.g. curvature).
This applies whether the correlations are partial or not.
Thus correlations should always be considered in
relation to graphical diagnostics.
Nick
[email protected]
Raphael Fraser
> This is a very useful program especially in multiple regression. When
> used with pwcorr one can decide if certain variables add any
> additional information to the model by comparing the correlation and
> partial correlation coefficient.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
