Who follows the leader? Leadership heuristics and valence voting at the UK’s 2016 Brexit referendum

Abstract

Recent accounts of British voting behaviour emphasise the importance of voters’ valence judgements on deciding which party to support: significant numbers of voters do not rely on their membership of particular socio-economic groups or ideological preferences; instead, they rely on evaluations of government performance to determine their vote. Part of this evaluation comes from ‘fast and frugal’ heuristics such as their evaluations of party leaders and other senior politicians. Such heuristics have been shown to be influential in the United Kingdom’s 2016 Brexit Referendum for example. We extend this research to show that the influence of leadership heuristics on individuals’ chances of voting for Brexit varied depending on evaluations of leading politicians associated with the Remain and the Leave campaigns. Moreover, feelings toward the party leaders had a larger influence on the referendum votes of those holding middle-of-the-road views on Europe than on those who were themselves either strongly pro- or anti-EU.

Keywords:

• voting
• valence
• heuristics
• leader cues
• EU referendum
• England

Notes

1 A number of studies use survey data to explore patterns of voting at Brexit without incorporating these arguments: see, for example, Alabrese et al. (2018).

2 A related issue, addressed in a pioneering experimental study by Goodwin et al. (2018), is not simply the cues provided by leaders or other campaigners but rather the nature of the messages they present.

3 On the data see https://www.britishelectionstudy.com/data/#.XBzOsme7Jow.

4 As with most recent voter surveys, the BES substantially under-sampled those who did not vote, which was 28 per cent according to the official data (and this was an underestimate since it excluded those who were not registered on the electoral roll.)

5 Further modelling showed that adding variables for party identification – whether the respondents identified with one of four parties (Conservative, Labour, Liberal Democrat, UKIP) or none (used as the comparator) – added little to the statistical explanation, largely because of collinearity with the leader feelings variables. Adding the four party identification variables to Model IV did not change the Cox-Snell and Nagelkerke R² values. If they are the only four variables included in a model the two R² values are only 0.18 and 0.25 respectively, with significant coefficients for each: 0.67 for Conservative, −0.77 for Labour, −1.37 for Liberal Democrat, and 4.40 for UKIP. When those four variables are added to Model IV the coefficients are −0.35 (insignificant), −0.45, −0.35 and 0.69 respectively, indicative of substantial collinearity with the other variables. Further the coefficients for the leadership components hardly change: from 2.31 to 2.27 for the Brexiteer component; from −1.04 to −1.03 for the Cameron-Osborne component; and from −0.25 to −0.22 for the Corbyn-Farron component.

6 This was achieved using a bespoke routine in Excel; Tomz et al. (2003) have provided Stata macros for the CLARIFY procedure at https://gking.harvard.edu/clarify.