Measuring the democracy–repression nexus

A considerable amount of time and intellectual energy has been expended detailing the nature of the relationship between democracy and a state's use of repressive force against its citizens. This article moves the discussion forward both methodologically and substantively. I use a fully Bayesian structural model with latent variables to mitigate the effects of measurement error in both democracy and repression. Further, I estimate two models, one treating democracy as a single overarching concept, similar to the strategy followed by Davenport and Armstrong (2004), Poe and Tate (1994) and one treating democracy as two-dimensional (voice and veto) as suggested by Davenport (2007b). I find that the two models, though closely related offer substantively different predictions. Statistical measures of fit favor the two-dimensional model. Further, I find that the effects of voice and veto are each strongly conditioned by the other.