BIASES RESULTING FROM THE USE OF AGGREGATED VARIABLES IN PSYCHOLOGY

Perloff and Persons (1988) have recently stated that the use of equal-weight indexes (aggregated variables) in regression-correlation analysis will result in biased tests of statistical significance unless the variables aggregated are extremely highly correlated. Referring to psychometric theory, this article shows that the size of the intercorrelations among variables has little bearing on the propriety of using unit weights to aggregate them into a composite. What is important in that regard is the form of those intercorrelations. A statistical test by Wilks (1946) is presented that evaluates the presence in a data set of 3 preconditions for unit-weight aggregation.