Partisan dealignment and the personalisation of politics in West European parliamentary democracies, 1961–2018

Abstract

Partisan dealignment is recurrently presented in the literature as one of the main drivers of the ‘personalisation of politics’. Yet, on the one hand, the claim that leader effects on voting behaviour are increasing across time is short on comparative evidence. On the other hand, there is limited empirical evidence that such an increase is due to dealignment. This article explores the longitudinal relationship between partisan dealignment, leader effects and party choice, through a novel dataset pooling 109 national election surveys collected in 14 Western European parliamentary democracies across the last six decades. The results show that leader effects increased over time as a function of the decline of party identification. Additional panel evidence from selected countries shows that partisan dealignment is responsible for increasing leader effects on party choice at the individual level. The longitudinal dimension of this study contributes to the most contested aspect of the personalisation of politics debate.

Keywords:

• Dealignment
• leader effects
• party identification
• personalisation of politics
• voting behaviour

Acknowledgements

The authors wish to thank Russell J. Dalton, Alexander H. Trechsel, and the two anonymous reviewers of West European Politics for their comments and suggestions. We are also grateful to Federico Vegetti for his assistance with data harmonisation and management. An earlier version of this manuscript has been developed during Diego Garzia's research stay at the University of California, Irvine. That version has been presented at the 2018 Southern California Political Behavior Conference at the University of California, Riverside, and published in the CSD Working Paper Series of the University of California, Irvine.

Disclosure statement

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

Data availability statement

The data that support the findings of this study are openly available. Our dataset is composed of 109 merged National Elections Studies (NES), all available online with the national producing institutions.

Notes

1 We follow Elgie and Passarelli’s (2019) discussion of the theoretical and conceptual differences in the use of the term ‘presidentialisation’ between Poguntke and Webb (2005) and Samuels and Shugart (2010). As the focus of our contribution is on the determinants of voting behaviour at the individual level, we decided not to rely on Samuels and Shugart’s notions, which ‘have a narrow focus on constitutional presidentialisation and party presidentialisation’ (Elgie and Passarelli 2019: 116). Instead, we favour Poguntke and Webb’s conceptualisation as ‘a more general idea of presidentialisation that results from a much broader process of social and political change’ (ibid.). Our ‘personalisation’ notion can thus be equated to the ‘electoral face’ of presidentialisation. The theoretical link between partisan dealignment and growing leader effects also fits within Poguntke and Webb's (2005: 15–16) theoretical framework, placing the ‘erosion of traditional social cleavage politics’ as a cause for presidentialisation.

2 An important exception is a recent study by Dalton (2013), including data until 2010 for ten Western European parliamentary democracies.

3 Leadership traits might potentially allow for a more exogenous estimation of leader effects on vote choice, and thus be a preferable baseline option. However, recent research has noted that trait evaluations are also permeable to partisan stereotypes, ideological predispositions and previous voting habits (Bittner 2011; Garzia 2017; Holian and Prysby 2014). More importantly, though, as leader thermometers are more widely available across national election studies, they enable a broader longitudinal and cross-country comparative analysis of the relationship between dealignment and personalisation. Based on Bittner’s (2011) extensive survey of election studies featuring measures of party leader trait batteries, the use of the latter instead of leader thermometers in the current study would imply a drop in our sample from 109 to 17 election studies, rendering impossible a longitudinal and comparative analysis.

4 Partisan loyalty is indeed above 90 per cent throughout the whole time series.

5 In particular, varying choice sets threaten the Independence of Irrelevant Alternatives assumption that is invoked in conditional logit models. Relaxing this assumption requires modeling the variations in the choice sets. Therefore, we experimented with different conditional logit configurations that allow for variation in voters’ choice sets by interacting a choice-set indicator with each alternative-specific covariate (i.e. leader evaluations, partisanship, ideological proximity). This produces estimates that are choice-set-specific, but this comes at the cost of misleading generalisations. For instance, the choice set configurations usually involve one or two specific countries and therefore convey the effect of idiosyncratic context rather than the absence or presence of certain party families. Moreover, these models involve unbearable complexity. In fact, our argument of the diachronic increase in the relative importance of leaders and parties would also demand a triple interaction to model the change of the coefficients of interest over time. While this would represent a feasible option for analyses of the dynamics of a single party system, or in the synchronic variation across party systems, in our long-term comparative setting a more pragmatic approach is preferable.

6 We acknowledge that this approach is not without caveats, but it has the undeniable merit of offering a feasible measure of ideological proximity across electoral studies spanning over six decades in multiple countries.

7 Due to the lack of relevant questions – either self- or party- placement – it was impossible to compute ideological proximity for the respondents featured in the following studies: Denmark 1973; Germany 1961, 1972, 1980 and 1994; Portugal 2011; United Kingdom 1970 and 2010. All these studies are thus excluded from the multivariate analyses that follow.

8 Admittedly, our model specification is parsimonious, which results from the constraints of the stacked data matrix framework. The inclusion of further controls at the respondent level poses a problem since socio-demographic variables are constant at the party-respondent level. However, since the effects of these variables can be claimed to be indirectly accounted in the sense that they are largely subsumed into party identification and ideological self-positioning, we are confident that such a parsimonious model is nonetheless satisfactory.

9 These include contextual differences as well as inconsistencies in question wording across countries and elections.

10 The difference in the N of party*respondent combinations and the N of respondents is due to the stacked data matrix framework. In our sample, on average, each respondent observation was repeated 5,27 times, that is, the mean number of parties included in each election study.

11 The party families were coded according to the Comparative Manifesto Project classification.

12 Note that the results hold regardless of institutional-level variation in electoral systems (proportional vs. majoritarian/mixed) and regime type (parliamentary vs. semi-presidential).

13 Leader effects are also found to increase significantly for de-aligned voters in a panel model with fixed-effects at the individual level. However, this model would reduce the pool of voters considered in the analysis to only those experiencing a change in the variables of interest, resulting in a very small effective sample. For this reason, we present the more comprehensive evidence provided in Figure 4.

14 Noticeably, our results are on the conservative side. With leaders increasingly responsible for shaping partisan identifications – as previous studies have demonstrated (Garzia 2013; Rapoport 1997) – it is likely that leaders matter even more overall.

Additional information

Funding

This work was supported by the Swiss National Science Foundation under Grant PZ00P1167997.

Notes on contributors

Diego Garzia

Diego Garzia is Assistant Professor of Political Science at the University of Lausanne. His current research project on ‘negative voting’ in comparative perspective is funded by an Eccellenza Professorial Fellowship of the Swiss National Science Foundation, 2020–2024. [[email protected]]

Frederico Ferreira da Silva

Frederico Ferreira da Silva is currently an SNSF Senior Researcher at the University of Lausanne, Switzerland. He obtained his PhD in Social and Political Sciences from the European University Institute in 2019. He works on elections, public opinion and voting behaviour. [[email protected]]

Andrea De Angelis

Andrea De Angelis is Senior Researcher and Lecturer at the Department of Political Science, University of Lucerne. His research interests include political behaviour, political methodology, political communication and political psychology. [[email protected]]