The Origins of Economic Voting: Investigating Where and When Vote Swings Come From

To accurately understand strategic and economic voting, statistical models of voting behaviour must account for the dynamic and compositional nature of voting data. Using Tandon’s (2012) analysis of voter responses to tariff reforms in India, we illustrate that statistical models that fail to take account of these properties draw inaccurate conclusions about the pattern and motivations behind voting. First, we argue that in panel data voting is a dynamic decision; vote swings in one period affect swings in subsequent periods. The inclusion of lagged values of voting outcomes incorporate these time-dependencies and capture dynamic voting processes. Second, we argue that the adding-up constraint implicit in voting data, showing that a vote swing towards one party must always be offset by a swing away from another party, must be explicitly incorporated into voting behaviour models to avoid bias. We make an important methodological contribution by adapting existing compositional models- intended for modelling vote shares in a single election- to capture changes in vote share between elections and developing diagrammatic tools that support intuitive analysis of vote swings. Once the dynamic and compositional nature of the data is accurately modelled, we show that patterns of economic voting seen in India from 1991 to 2004 are starkly different to those reported in Tandon (2012). Voters exposed to tariff reductions ignored both the social protection offered by the Far Left and the anti-reform sentiments of the BJP, turning instead from the INC to Regional parties.