A New Method for Assessing the Construct Validity of Accounting Measures: Looking Where the Action is

We propose and demonstrate a new method for validating proxies for accounting constructs. Our method consists of first identifying high-value observations, which are those statistically most responsible for generating a coefficient of interest, and second determining whether those observations exhibit accounting attributes consistent with theory. This method allows researchers to validate proxies for key theoretical constructs by testing whether high-value observations contain the accounting attributes implied by the theory and lack attributes that run counter to the theory. The method also allows researchers to identify potentially confounding factors and to address alternative explanations for primary results. We demonstrate the method by assessing the validity of two measures for conditional accounting conservatism. However, the method applies generally to large-sample settings and is flexible in its application to key coefficients. We also demonstrate how researchers can compare alternative measures of the same construct to select the best measure for their research context.