This paper introduces the concept of narratives into a model of political competition to explain some stylised facts related to populism, and provides a way to measure such narratives to empirically analyse the theory. In the model, I define narratives as subjective causal models that map observable signals (for instance, wages) into unobservable latent variables (for instance, intelligence) that voters care about. Narratives differ in the informational content they ascribe to signals: \textit{narratives of agency} posit that signals are very informative about latent variables of interest, while narratives of blame posit the opposite. In order to make sense of their social as well as economic standing, some voters source narratives from parties. The model illustrates that (1) unpolarized equilibria in which both the economically right and left party is populist, in the sense that it offers narratives of blame, exist if the inequality in the social dimensions is limited; (2) only polarized equilibria, where one party is populist while the other is not, can generate an electoral split along social issues, instead of economic issues; (3) such polarized equilbria exist for a strictly larger set of conditions for a populist right party rather than a populist left; and (4) the economically right party will force a transition towards such an equilibrium by becoming more redistributive if (a) low earners who also feel socially left behind are a relatively larger electoral group than high earners who face positive social signals, or (b) the negative dimension of the social issue is sufficiently pronounced such that the turnout of the voters is high. To test the model’s predictions, I use textual analysis tools to classify speeches made in the UK House of Commons as either being migrant-blaming or migrant-affirming, which generates a measure of narratives of blame across political parties at a high frequency. Using this measure, I empirically validate several of the model’s main predictions.