| Abstract: | This study adds to research which examines the construct validity of coefficients of cue importance in studies concerned with how decision-makers use accounting information in formulating judgments (see Larcker & Lessig, The Accounting Review, January, 1983, pp. 58–77; Selling & Shank, Accounting, Organizations and Society, 1989, pp. 65–77). Historically, accounting studies have modelled cue importance with almost exclusive reliance upon linear models. But as the Selling & Shank study indicates, inferring the importance of accounting cues through reliance upon only one kind of model can leave “method variance” undetected and raises threats to the construct validity of coefficients (see Cook & Campbell, Quasi-experimentation: Design and Analysis Issues for Field Settings, 1979, pp. 59–70). To the extent that cue importance appears similar across models, then the model coefficients are presumed more valid. While Selling & Shank compare linear models to process tracing models, we compare a linear model to an eigenvector-scaling routine known as the Analytic Hierarchy Process (see Saaty, The Analytic Hierarchy Process, 1980). As with Selling & Shank, we find that the importance of cues is sensitive to model choice, suggesting that more research is needed into method variance before judgments can be made with respect to the construct validity of linear coefficients in accounting studies. |
