The practice of combining the scores of multiple item scales, by adding, to form a single composite score is ubiquitous in psychological research. This study examines the effect of such practices in relation to a simple regression model by a series of Monte Carlo simulations. A comparison is made between the addition model and the factor model under varying conditions of reliability and collinearity. It is shown that the use of composite scores will generally tend to result in an underestimation of the population regression effect when the reliability of the items comprising the composite score is low. The use of a factor model results in estimates that are closer to the population value but at the expense of increasing standard errors as the reliability decreases. (C) 1997 Elsevier Science Ltd.
|Journal||Personality and Individual Differences|
|Publication status||Published - Oct 1997|