We bridge together the graph-theoretic and the econometric approach for defining causality in statistical models to consider model misspecification problems. By presenting a solution to disagreements between the existing frameworks, we build a causal framework that allows us to express causal implications of econometric model specifications. This allows us to reveal possible inconsistencies in models used for policy analysis. In particular, we show how a common practice of doing policy analysis with vector error-correction models fails. As an example, we apply these concepts to discover
fundamental flaws in a resent strand of literature estimating the carbon Kuznetz curve, which postulates that carbon dioxide emissions initially increase with economic growth but that the relationship is eventually reversed. Due to a causal misspecification, the compatibility between climate and development policy goals is overstated.