The Singer not the Song?: Evidence From a Field Experiment in British Columbia, on the impact of Leaders and the Messages they Convey

Are leaders effective because of some innate qualities-for example, clarity, trustworthiness or focality-or because of the particular arguments they employ? To analyze these effects systematically we need variation in both messages and leaders. Whilst these conditions are satisfied in many political settings, and in particular during political campaigns, observational studies are likely to produce biased estimates of these effects due to correlation, selection-bias and endogeneity. We describe a hierarchical model which allows us to untangle the effects of leader qualities from those of the arguments leaders use. We implement this model using data from a unique field experiment: working together with the BC-STV campaign in the May 2009 British Columbia referendum on electoral reform, we randomly assign canvassers to voting areas, who randomly assign messages to households. We analyze leadership effects in two ways: (i) by estimating the between voting area differences in opinion that are due to the assignment of different canvassers; and (ii) by allowing some of the messages to be endorsed by leading public figures. Surprisingly, neither of these strategies yields any evidence of leadership effects: a fairly substantial overall campaign effect is entirely due to the arguments used by the campaign, not to individuals making the case for reform.