p, r, and the Coefficient of Determination

Many researchers seem to get carried away with the p-levels associated with their correlation coeffcients and thus seem to forget that the estimated strength of a relationship is best assessed by squaring the sample value of r. Discovering that a correlation coefficient is "significant" may not really be very important--even if the results indicate p < .01 or p < .001--unless the value of r2 is reasonably high. The result may be significant in a statistical sense (thus indicating that the sample data are not likely to have come from a population characterized by Ho), but it may be quite insignificant in a practical sense.

(From Chapter 9, p. 217)

Copyright © 2003 Schuyler W. Huck
All rights reserved.