This article examines challenges of measurement validity in aggregate governance indicators. We focus on three deleterious consequences of aggregating perception-based indicators in the absence of conceptual clarity: (i) the scant attention to content validity; (ii) the conflation of causes, characteristics and consequences of governance; and (iii) the underestimation of uncertainty. As an alternative, we present a Bayesian latent variable framework for measuring governance. This alternative formal statistical model offers several advantages: it is a principled method that allows the researcher to make explicit the conceptual choices in measuring governance, and it provides an assessment of uncertainty by letting measurement error (noise) in governance measures propagate into inferences. We argue that this multidimensional, disaggregated and transparent approach will lead to greater measurement validity and transparency about the uncertainty in governance measures. The conceptual and empirical analysis uses data from the World Bank Governance Indicators.
