How can we determine which arguments in a referendum are most persuasive? We show that the Bradley–Terry model has several features that make it well-suited to this task, and thus preferable to other, more conventional approaches. Using a survey experiment conducted during an electoral reform referendum in Ontario, Canada in October 2007, we demonstrate how unstructured and structured Bradley–Terry models can be straightforwardly fitted and interpreted. In doing so, we gain insight into the factors which determine support for electoral reform. We identify a status quo bias and find that power varies with mention of fairness, local control over candidate selection, and the role of political parties. We conclude by discussing the limits, extensions and further applications of such models in electoral studies and political science more broadly.