Measurement error in performance studies of health information technology: lessons from the management literature

Just as researchers and clinicians struggle to pin down the benefits attendant to health information technology (IT), management scholars have long labored to identify the performance effects arising from new technologies and from other organizational innovations, namely the reorganization of work and the devolution of decision-making authority. This paper applies lessons from that literature to theorize the likely sources of measurement error that yield the weak statistical relationship between measures of health IT and various performance outcomes. In so doing, it complements the evaluation literature's more conceptual examination of health IT's limited performance impact. The paper focuses on seven issues, in particular, that likely bias downward the estimated performance effects of health IT. They are 1.) negative self-selection, 2.) omitted or unobserved variables, 3.) mis-measured contextual variables, 4.) mismeasured health IT variables, 5.) lack of attention to the specific stage of the adoption-to-use continuum being examined, 6.) too short of a time horizon, and 7.) inappropriate units-of-analysis. The authors offer ways to counter these challenges. Looking forward more broadly, they suggest that researchers take an organizationally-grounded approach that privileges internal validity over generalizability. This focus on statistical and empirical issues in health IT-performance studies should be complemented by a focus on theoretical issues, in particular, the ways that health IT creates value and apportions it to various stakeholders.