Research Commentary - Using Income Accounting as the Theoretical Basis for Measuring IT Productivity

W use the under-recognized income accounting identity to provide an important theoretical basis for using the Cobb-Douglas production function in IT productivity analyses. Within the income accounting identity we partition capital into non-IT and IT capital and analytically derive an accounting identity (AI)-based CobbDouglas form that both nests the three-input Cobb-Douglas and provides additional terms based on wage rates and rates of return to non-IT and IT capital. To empirically confirm the theoretical derivation, we use a specially constructed data set from a subset of the U.S. manufacturing industry that involve elaborate calculations of rates of return—a data set that is infeasible to obtain for most productivity studies—to estimate the standard Cobb-Douglas and our AI-based form. We find that estimates from our AI-based form correspond with those of the Cobb-Douglas, and our AI-based form has significantly greater explanatory power. In addition, empirical estimation of both forms is relatively robust to the assumption of intertemporally stable input shares required to derive the AI-based form, although there may be limits. Thus, in the context of future research the CobbDouglas form and its application in IT productivity work have a theoretically and empirically supported basis in the accounting identity. A poor fit to data or unexpected coefficient estimates suggests problems with data quality or intertemporally unstable input shares. Our work also shows how some returns to IT that do not show up in output elasticities can be found in total factor productivity (TFP)—the novel ways inputs are combined to produce output. The critical insight for future research is that many unobservables that have been considered part of TFP can be manifested in rates of return to IT capital, non-IT capital, and labor—rates of return that are separated from TFP in our AI-based form. Finally, finding that the additional rates of return terms partially explain TFP confirms the need for future IT productivity researchers to incorporate time-varying TFP in their models.