Affective polarization and strategic voting in Britain

ABSTRACT

Do affective attitudes determine voting behavior? As a part of this broad question, we examine how third-party supporters’ affective attitudes decide their strategic voting behavior. Using seven waves of the British Election Study (2015, 2017, and 2019), we find that third-party supporters whose affective attitudes toward the two principal political parties in their constituency are more polarized are more inclined to vote strategically. The results imply that voting behavior is shaped by both feelings toward as well as cognitive evaluations of political parties. They also imply that British voters may increasingly engage in strategic voting if the level of affective polarization continues to rise. Lastly, and more generally, the results imply that all political parties should pay attention to voter “feelings” if they are to maximize their electoral support.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The Comparative Study of Electoral System module 2 and module 5. Party affect is measured on a 0 (strongly dislike) ∼ 10 (strongly like) scale. The difference in voter affect measures the average distance for voters between their affect for the Labour party and their affect for the Conservative party. This grew from 3.9 in 2005 to 4.8 in 2019. Due to the complexity and relative uniqueness of the Northern Ireland party system, “Britain” is hereafter England, Scotland and Wales only.

2 See for example, Greene (1999) on social identity, and, Johnston (2006) on evaluations of political events. Partisanship is a broader concept than party affect with including not only affective aspects but also other psychological and cognitive aspects (Richardson 1991). We focus only on the affective aspect in this paper. Nonetheless, some studies conflate these two by measuring partisanship only with its affective aspect (e.g., feeling thermometers) (e.g., Maggiotto and Piereson 1977; McGregor, Caruana, and Stephenson 2015; Mayer 2017).

3 Also, though she utilizes the concept and measure of partisanship, not party affect, Bankert (2021) finds a weak correlation between positive and negative attitudes. She finds that while three quarters of American voters hold a “diagonal” partisanship pattern (i.e., either high levels of positive (in-group) and negative (out-group) partisanship or low levels of positive and negative partisanship), one quarter of them demonstrates an “off-diagonal” partisanship pattern (i.e., either a high level of positive partisanship and a low level of negative partisanship or a low level of positive partisanship and a high level of negative partisanship).

4 For evidence of the correlation between ideological and affective polarization see Reiljan 2020; Webster and Abramowitz 2017. For an alternative perspective, emphasizing the relative independence of the two see Mason 2018.

5 Though as Tsebelis (1986) demonstrated, in proportional representation systems, inverse tactical voting is possible, where supporters of major parties may actually have an incentive to cast their vote for third parties.

6 See, for example, Simon 1954 on the earliest articulation of the idea of negative voting, and, Campbell et al. 1960 on positive and negative party identification.

7 The rationale of “avoiding worst” behaviors can be found in the primary process as well. Supporters of a political party may participate in the primary of their rival party in an open primary system to vote for a moderate candidate to decrease the likelihood of success for their least preferred, extreme, candidate in the party particularly when they do not find a critical difference between the moderate candidate and their party’s moderate candidate (Cherry and Kroll 2003).

8 The current British Election Study includes 21 waves from 2014 to 2021. All the waves have questions on party affect, but only these seven waves asked people a question that is critical to our analysis: how likely they think each political party would win the election in their constituency. The wave 4 was executed before the 2015 election (March 2015), waves 10, 11, and 12 were run before the 2017 election (November/December 2016, April/May 2017, and May/June 2017, respectively), and waves 16, 17, and 18 were implemented before the 2019 election (May/June 2019, November 2019, and November/December 2019, respectively).

9 Nonetheless, as we discuss below, we also employ two other alternative ways of identifying third-party supporters. See the supplementary appendix for the wordings of all the survey questions we utilize in this paper.

10 While many of this 26% comprised individuals supporting a party other than Labour or the Conservatives, about half were either Labour or Conservative supporters in a constituency in which they believed their party was placed third or lower in likelihood of winning.

11 In our data, 90.2 percent of people actually voted for a political party that they had intended to vote for before an election. This gives us confidence that the voting intention question is a fairly reliable indicator of voting behavior.

12 Hereafter, when we reference the two principal parties we are referencing the two parties best placed to win the electoral contest in a given constituency. While the Labour and the Conservative Parties are the two largest parties nationally, they are not always the two best placed to win in a given constituency. Either one or even both could be a “third party” if they are not one of the two best placed to win.

13 Some research suggests that a high level of information is needed to engage in strategic voting (Hillygus and Treul 2014). If people with certain backgrounds in their gender, age, education, or income more exposed to, interested in, or have access to political information, they may demonstrate a greater likelihood to vote strategically. Voters with an extreme political ideology may be less likely to strategically vote because principal parties tend to be ideologically moderate. Extremity of political ideology is measured as the distance between a voter’s left/right ideology position (0: left ∼ 10: right) and the central position of the scale (5).

14 Despite the inclusion of many control variables, multicollinearity is not a statistically significant problem (VIF < 5).

15 Imai et al. (2011).