This paper is motivated by our observation that many researchers have confused the interrelated, but analytically separable, problems of data aggregation, of model specification, and of statistical bias in parameter estimation.