All existing indices of state fragility suffer from either opaque methodologies that prohibit replication, a weak link between conceptualization and operationalization or limitations in their geographical coverage. Furthermore, except for the Failed States Index, all indices fail to provide classification results, thus rendering the identification of critical borderline cases arbitrary. The present study remedies these shortcomings by developing a statistical model for measuring state failure that is characterized by a strong link between conceptualization and operationalization, highly transparent, global in coverage, and based on a parsimonious set of theoretically valid and statistically significant key indicators of state capacity. In addition, the model does not only allow for the determination of a state’s degree of state capacity, but also for its classification as strong, weak or failed. The model is predicated on two assumptions: first, the term “state failure” describes a situation that is located near the negative end-point of a one-dimensional continuum of state capacity; second, state capacity has essentially five dimensions: the state’s ability to successfully claim the monopoly of violence, generate sufficient revenue to finance its operations, effectively administer its operations, regulate civil society through the provision of public goods, and generate voluntary compliance with its rule. Exploratory factor analysis is used to investigate the proposed dimensionality of state capacity and to identify key indicators of each dimension. Multiple discriminant function analysis is used to develop the classification model. The final model consists of four variables spanning three dimensions of state capacity. It explains 74 percent of variance between strong, weak and failed states and classifies 88 percent of cases correctly when using the Failed States Index as reference. The degree of state capacity is assessed through the calculation of discriminant function scores. 172 states are ranked and classified. The construct validity of the model is demonstrated.
