STANDARDIZED REGRESSION-COEFFICIENTS - A FURTHER CRITIQUE AND REVIEW OF SOME ALTERNATIVES

Standardized regression coefficients continue to be used, no doubt because they appear to simplify the interpretation of regression results: Unitless quantities such as standardized coefficients seem easier to describe than quantities that depend on units, and, precisely because they lack units, such quantities appear comparable. Nevertheless, as previously explained, 1 this apparent comparability is illusory. Now that a defense of standardized coefficients has been mounted, 2 we can offer some new illustrations of logical fallacies inherent in their use. We begin by repeating the main argument against comparing standardized coefficients: that such comparisons are in fact confounded, rather than standardized. 1 Next, we describe a modification of traditionally standardized coefficients that allows them to be compared across studies when a simple logistic model is adequate, and we describe the more complex procedures necessary when the effects under study cannot be modeled by a single regression coefficient, or when one wants to compare the public health effects of different factors. Finally, we note that all our arguments apply with equal force against the use of correlation coefficients as measures of effect.