Talk:Predictive validity

The just-added section on significance correlations equaling predictive validity

For example, the validity of a cognitive test for job performance is the correlation between test scores and, for example, supervisor performance ratings. Such a cognitive test would have predictive validity if the observed correlation were statistically significant.

I'll see if I can dig up a reference, but statistical significance is not enough for this aspect of validity because a statistically significant correlation only indicates a non-zero relationship. A tiny correlation can still be significant, though not necessarily sufficient to support an instrument's predictive validity. (E.g. A racial profiling instrument would be "predicatively valid" for predicting SAT scores, criminal record, and a host of other things.)

R^2 (percentage of variance explained) is a better indicator of the strength of the relationship. If I can't find a ref on the methodology, I'll ignore it.

Nevermind... Someone took care of it.

Jmbrowne (talk) 15:56, 20 February 2009 (UTC)