Jargon Jousting – Corrected Correlation

Why would anyone need to correct a correlation? Correlations typically don’t run afoul of the law and certainly no correlation has ever been accused of committing a felony. Now it is true that some correlations are so trivial as to be misdemeanors, but certainly they aren’t deserving of correction either.

The only legitimate reason to correct a correlation might be prior miscalculation, in which case the numbers simply should be run again. But what justification could there possibly be for taking correlations that are at worst misdemeanors and pumping them up to look like relationships on steroids? If the correlation is non-existent or trivial to begin with, why not just admit it?

An unfortunate example of ruthlessly corrected correlations occurs in the Technical Manual for a recently released assessment of “Talent”. The report of the validity correlations for the twelve scales on the Talent measure reveals that only four of these twelve scales reach a correlation of .10 or greater with supervisor ratings of talent behaviors. The table reports that correlations of .10 or greater are significant at or beyond the .05 level of significance. Yet after not one, not two, but three different corrections, nine of the twelve scales exhibit significant correlations with supervisor ratings. Most miraculous of all is the way in which the five insignificant correlations below .10 became significant relationships.

If there really were a Psychometric Department of Corrections, it should be the correlation correctors, not the correlations themselves, that should be targeted for correction! Taking an already significant correlation and pumping it up on steroids should be a misdemeanor, but somehow pulling an insignificant correlation out of the hat and employing three corrections to make it significant clearly should be a psychometric felony.