Introduction
Economic voting theory stands as one of the most researched theories in voting behaviour (Lewis-Beck and Stegmaier Reference Lewis-Beck, Stegmaier, Congleton, Grofman and Voigt2019). Economic voting has been primarily developed through a focus on valence: when considering which party to vote for, voters evaluate the state of the national economy and their personal finances, rewarding incumbents in economic good times (Lewis-Beck and Stegmaier Reference Lewis-Beck, Stegmaier, Congleton, Grofman and Voigt2019) and punishing them during bad times. The economy can also be considered a positional issue whereby economic policy preferences guide decision-making at the ballot box (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013). In recent years, economic voting research has expanded to incorporate patrimony or the ownership of assets. The patrimonial voting theory argues that the ownership of various types of assets shapes political interests and thus vote choice (Nadeau et al. Reference Nadeau, Foucault and Lewis-Beck2010). The main idea is that parties of the centre-right usually support economic policies that benefit asset-rich individuals, such as lower taxation and market deregulation. Consequently, guided by material self-interest, those who own assets are more likely to vote for parties of the centre-right (Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011). Importantly, patrimony, which directly captures property ownership and assets, influences voting behaviour above and beyond social class (Nadeau et al. Reference Nadeau, Foucault and Lewis-Beck2010; Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011; Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013). This explanation of the vote is particularly important to contemporary democratic politics given the uneven patterns of wealth accumulation that have been observed over past decades (for example, Hoffmann et al. Reference Hoffmann, Lee and Lemieux2020; Piketty Reference Piketty2014).
Patrimonial economic voting captures individuals’ material interests based on how many assets they own, their type, and value (see also Page and Jacobs Reference Page and Jacobs2009). This relates to tangible assets – thereby not including debt (Nadeau et al. Reference Nadeau, Foucault and Lewis-Beck2010). The first measure concerns overall wealth, that is, the accumulated assets one owns. Second, to capture the nature of wealth, the theory differentiates between high-risk (for example, stocks, investments, or running one’s own business) and low-risk asset types (for example, savings accounts or physical assets – Nadeau et al. Reference Nadeau, Foucault and Lewis-Beck2010). Owners of riskier financial patrimony are considered more likely to opt for policies and parties that improve their chances to further grow their assets, whereas the more risk-averse owners of savings and/or physical assets tend to seek financial security (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013). A third approach accounts for levels of wealth directly in terms of asset value, focusing on the monetary value of patrimony (Foucault et al. Reference Foucault, Nadeau and Lewis-Beck2013; Persson and Martinsson Reference Persson and Martinsson2018). Taken together, this literature expects that increases in patrimony – especially high-risk assets – are associated with the centre-right vote, considering that these parties tend to adopt policies that improve levels of wealth accumulation and capital income returns. Findings across a number of national contexts have provided empirical support for this hypothesis, indicating that patrimony is positively associated with centre-right voting either directly (for example, Costa Lobo Reference Lobo2013; Fraile and Lewis-Beck Reference Fraile and Lewis-Beck2013; Okolikj and Quinlan Reference Okolikj and Quinlan2021; Persson and Martinsson Reference Persson and Martinsson2018; Stubager et al. Reference Stubager, Lewis-Beck and Nadeau2013) or indirectly, mediated by ideology or party identification (for example, Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013; Quinlan and Okolikj Reference Quinlan and Okolikj2022).
The bulk of existing studies, however, while offering important evidence on the correlation between economic circumstances and voting, often fall short of accounting for alternative mechanisms that could be driving this relationship. Indeed, there are reasons to expect that asset ownership is likely correlated with a number of additional factors influencing voting behaviour. First, parents provide their offspring with assets or the financial means to acquire assets. However, parents also transmit all sorts of political orientations to their offspring during childhood or adolescence, as demonstrated by research showing a correlation between parents and offsprings’ political attitudes and vote choices either as a product of social learning (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960; Coffé and Voorpostel Reference Coffé and Voorpostel2010; Durmuşoğlu et al. Reference Durmuşoğlu, de Lange, Kuhn and van der Brug2023; Fitzgerald and Bacovsky Reference Fitzgerald and Bacovsky2022; Rico and Jennings Reference Rico and Jennings2016), or as a result of genetic influence (Alford et al. Reference Alford, Funk and Hibbing2005, Hatemi et al. Reference Hatemi, Medland, Klemmensen, Oskarsson, Littvay, Dawes and Martin2014). In addition, attitudes and traits such as risk-taking preferences and interpersonal trust, which are involved in financial decision-making and are known to predict socio-economic outcomes, are also inherited from parents (Alan et al. Reference Alan, Baydar, Boneva, Crossley and Ertac2017; Zumbuehl et al. Reference Zumbuehl, Dohmen and Pfann2021).
A second set of potential confounders concerns social outcomes that simultaneously correlate with levels of wealth and political preferences. One example is neighbourhood choice. In Britain, wealthy families tend to reside in ethnically homogenous White areas (Zuccotti Reference Zuccotti2019). Lack of exposure to diversity during one’s formative years is associated with conservative party voting even decades later (Brown et al. Reference Brown, Enos, Feigenbaum and Mazumder2021). At a broader level, social homophily in human networks – along with sociodemographic characteristics, such as income, education, and social class – solidify social identities, norms, and repertoires of behaviour (for example, McPherson et al. Reference McPherson, Smith-Lovin and Cook2001), including political behaviour (Huckfeldt Reference Huckfeldt2014). Therefore, the differences in political behaviour between those with assets and those without may reflect differences in socialization and social networks rather than be causally related to the accumulation of assets.
However, the vast majority of research on patrimony is cross-sectional (for example, Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011; Persson and Martinsson Reference Persson and Martinsson2018; Okolikj and Quinlan Reference Okolikj and Quinlan2021) and thus cannot address the question of whether voting preferences change when an individual acquires assets. Resolving this question is important to more precisely understand how economic self-interest shapes voter decision-making (vis-à-vis the influence of social learning or inherited characteristics) and, furthermore, to clarify how changing economic circumstances – at both the individual and societal level – might benefit the fortunes of particular political parties. This second concern is particularly significant in an era of rising economic inequality and given the difficulties facing younger generations in acquiring assets. The only studies, to our knowledge, that have attempted to assess the causal effect of patrimony on voting use a discordant twin design in a Swedish sample and find that while there is an association between assets and voting conservative, the actual effect is very weak (Ahlskog and Brännlund Reference Ahlskog and Brännlund2022– see also Ahlskog and Oskarsson Reference Ahlskog and Oskarsson2023). This suggests that patrimony is likely correlated with other factors that predict vote choice.
Against this background, we investigate the association between patrimony and voting in Great Britain and offer a two-fold contribution. First, we simultaneously assess the role of wealth and socialization on voting behaviour. To this end, we draw on Understanding Society (US), a unique panel dataset that covers a time span of eleven years and includes, alongside individual vote choice, detailed measures of both asset ownership and information on parental voting preferences. Second, we employ a longitudinal research design. This is a significant advancement in studies that draw on cross-sectional designs that are particularly prone to omitted variable bias. By including individual fixed effects, we control for time-invariant individual characteristics. We, therefore, account for the influence of stable, between-person differences between those who own assets and those who do not – such as the socialization effects that derive from growing up in an asset-rich environment or parental transmission of political preferences during childhood and adolescence, which might confound the relationship between asset ownership and voting behaviour in prior patrimonial research. Through the inclusion of wave fixed effects, we control for the influence of over-time trends that exert a constant influence across the study population, such as the electoral popularity of parties over the study period. This combination provides a rich set of controls, ensuring that any observed effect of asset ownership on voting behaviour is not the result of unmeasured confounding. Whilst we cannot entirely rule out the possible influence of unmeasured variables, this research design goes much further than past research on patrimony in addressing these concerns.
Britain offers an ideal case for investigating the association between wealth and voting. The existence of patrimonial voting relies on two assumptions: that the mainstream parties of the right and left offer different policy choices on economic matters and voters recognise these differences (Hellwig and McAllister Reference Hellwig and McAllister2019); and that the economic issue is salient enough to influence voting behaviour (for example, Krosnick Reference Krosnick1988). Britain fully satisfies these criteria (see supplementary materials section 1.1).
Overall, our findings suggest that the relationship between patrimony and vote choice is driven mainly by pre-existing differences between those who own assets and those who do not, rather than the assets themselves. Patrimony, whether measured by the number of assets one owns or the total value of these assets, is unrelated to support for the Conservative Party. Furthermore, even under what we view as a generous test of the theory – the impact of homeownership on vote choice among first-time buyers – patrimony does not lead to significant changes in party support. Instead, evidence suggests that the influence of parental preference on vote choice exceeds and appears to explain away much of the association between patrimony and the vote.
Data and methods
Data come from Understanding Society (US), a large-N household panel study conducted in the UK (University of Essex 2022): US (formerly the British Household Panel Survey) is an ongoing panel that began in 2009 with 39,802 households and 101,086 individuals participating at wave one (Buck and McFall Reference Buck and McFall2012). Original sample households and members at wave 1 are recontacted at each wave. Respondents are interviewed in person within their households, completing both an individual and household questionnaire (Boreham et al. Reference Boreham, Boldysevaite and Killpack2012). Importantly, US contains detailed measures of asset ownership and vote choice at multiple waves, enabling an assessment of the impact of asset accumulation on voting. Panel attrition is low, with the survey maintaining average wave-to-wave retention rates of 82.8 per cent (ISER 2023).
Vote choice, the outcome variable, is measured through a question which asks respondents which party they would vote for if there was an election tomorrow. Following the logic of patrimonial voting, asset ownership would be positively associated with voting for the Conservatives over all other parties. The vote choice item was therefore coded as a binary measure (1 = Conservative vote; 0 = vote for all other parties).
To measure patrimony, many existing studies use a count measure – assigning an integer value to each individual depending on the number of asset types (such as a home or business) that they possess (for example, Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013; Hellwig and McAllister Reference Hellwig and McAllister2019). We construct a count measure of patrimony using the self-reported ownership of four types of asset: home, savings, business, and second home (that is, an individual that possesses all four asset types at wave T is coded with a value of 4 at that wave). Recognising the additional need to account for the value of assets (Foucault et al. Reference Foucault, Nadeau and Lewis-Beck2013; Persson and Martinsson Reference Persson and Martinsson2018), we also employ a value measure, which represents the combined total of the self-reported value of one’s property, the self-reported value of one’s business assets, and self-reported income from savings and investments, adjusted for inflation (respondents are coded with a value of 0 if they do not possess any of these assets).
The count patrimony items (and subsequent models using these items) are only available at four waves – waves 2, 4, 6, and 10. By contrast, the value patrimony items are available at all waves, and models using only these items are thus estimated across 12 waves of data. As a result, the model size varies from around 4,000 to around 33,000 observations (see supplementary material 1.2 for descriptive statistics on patrimony). Given that sufficient within-person variation in a predictor is essential to minimize bias when using individual fixed effects (Clark and Lizner Reference Clark and Linzer2015), we also examine within-individual variation in patrimony and show that it frequently varies within individuals over time across the sample: fully 21.4 per cent of repeated person-observations for the count patrimony measure in the panel are an increase on the prior wave (see supplementary materials table 1.2.1). We are therefore confident that we can precisely estimate the influence of asset acquisition on vote choice.
Whilst the use of individual and wave fixed effects accounts for potential time-invariant confounding, we take additional steps to address time-varying confounders. The acquisition of assets potentially constitutes a bundled treatment in that changes in assets are often accompanied by other significant changes in life circumstances (for example, the decision to buy a property may also be accompanied or motivated by having children). We therefore control for significant life events: marriage, having children, graduating from University, changes in occupational seniority and household income, and dummies for retirement and becoming unemployed. We also control for sociodemographic characteristics that likely correlate with both patrimony and vote choice (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013) and are necessary to precisely estimate the effect of patrimony: gender, age, ethnicity, and region of residence. We also include a binary indicator of whether the respondent pays a mortgage. Descriptive statistics for all study variables are provided in supplementary materials table 1.3.
Results
Patrimony
We begin with a cross-sectional assessment of patrimony. In order to provide a baseline for comparison, we validate that patrimony is related to voting preferences as expected when we adopt the cross-sectional approach used in prior literature (for example, Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011; Stubager et al. Reference Stubager, Lewis-Beck and Nadeau2013). We first estimate a pooled logit model using the four waves of data at which both measures of patrimony are available before estimating the same regression at each wave. For this and all subsequent models, we present logit coefficients in the main analysis, with odds ratios provided in the supplementary materials. We also recode the value measure of patrimony so that a one-unit increase corresponds to a half-standard deviation increase in this measure (equating to just over £300,000). Results from the pooled model indicate that the count measure of patrimony is significantly and positively associated with Conservative support (b = 0.21, SE = 0.03, p < 0.001), as is the value measure (b = 0.05, SE = 0.02, p < 0.05). Results at each wave confirm this conclusion (see supplementary materials section 2.1): in short, we replicate the findings of previous cross-sectional studies (for example, Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013; Okolikj and Quinlan Reference Okolikj and Quinlan2021) when we, too, use cross-sectional comparisons with our data.
Next, we assess the influence of patrimony on the vote by drawing on a longitudinal design. We first estimate logit models that include only patrimony, wave fixed effects, and demographic controls (that is, excluding individual fixed effects), with standard errors clustered at the individual level. Unlike a two-way fixed-effects approach, these estimates utilize both between-person and within-person variation in patrimony and thus provide another set of baseline results that enable us to understand the impact of including individual fixed-effects (and thus eliminating the effect of stable between-person differences) in later models.
Figure 1 presents a coefficient plot of vote choice regressed on patrimony, including demographic controls and wave fixed effects (full model available in table 2.2, supplementary materials). Results indicate that both the count measure of patrimony (b = 0.29, SE = 0.04, p < 0.001) and the value measure of patrimony (b = 0.06, SE = 0.03, p < 0.05) are positively associated with Conservative voting. We then add individual fixed effects, meaning that we exclude any potential between-person effects of patrimony and limit our analysis to the influence of within-person changes in assets on vote choice. First, we estimate a baseline model (that is, a two-way fixed-effects model without controls). In contrast to the pooled and wave fixed effects models, results indicate that both count patrimony (b = 0.05, SE = 0.07, p = 0.451) and the value of one’s patrimony (b = −0.01, SE = 0.02, p = 0.531) are unrelated to Conservative voting. Adding covariates does not alter this relationship - both count (b = 0.04, SE = 0.07, p = 0.569) and value (b = −0.01, SE = 0.02, p = 0.562) are unrelated to Conservative voting (see table 2.2 of the supplementary materials for the full model). Whilst the standard errors for the patrimony coefficients are slightly larger in the two-way fixed effects model, indicating a slight loss in the precision of estimates, by far the larger driver of the difference between the results from the wave and two-way fixed effects models is the drop in the size of the patrimony coefficient estimates. Thus, when we account for between-person differences and assess the influence of within-person changes in assets on vote choice, patrimony is unrelated to Conservative Party support. Our findings therefore suggest that the often-identified relationship between patrimony and voting (see Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011; Quinlan and Okolikj Reference Quinlan and Okolikj2022; Stubager et al. Reference Stubager, Lewis-Beck and Nadeau2013) seems to be explained not by the accumulation of assets themselves but instead by pre-existing differences between those who own assets and those who do not.
Coefficient plot of the relationship between patrimony and Conservative voting.
The role of direct parental transmission
As we have argued, parental transmission and socialization potentially confound the relationship between patrimony and vote choice. Two-way fixed effects results provide support for this argument. However, these models can only provide an indirect means of assessing the influence of parental transmission. To complement these results, we employ a more direct test of the potential confounding influence of parental transmission. As US is a household panel, it contains data on parental and child political preferences and measures the evolution of offspring voting preferences over time as these individuals accumulate assets of their own. As a result, we can estimate the influence of patrimony on Conservative voting whilst directly accounting for the influence of parental vote choice.
To do so, we employ a subsample of individuals for whom we also have parental voting preferences and estimate generalized linear mixed effects models (GLMMs), regressing (offspring) vote choice on patrimony and parental vote choice. Whilst in baseline GLMMs (that is, without parental preferences), we find a positive and significant association between patrimony and vote choice; the magnitude of the association between patrimony and Conservative voting is greatly reduced when parental preferences are included in the model, with parental Conservative support strongly predicting offspring Conservative support (see supplementary materials, section 2.3). These results provide further evidence that the apparent relationship between patrimony and Conservative voting is instead driven by parental transmission and the influence of early-years socialization.
First-time buyers
Results so far suggest that, for the most part, patrimony is unrelated to support for the Conservative Party. To complement these findings, we now turn to an alternative approach, examining the effect of homeownership on vote preferences among first-time buyers (see supplementary materials, section 4). We view this as a generous test of the patrimony hypothesis or a most likely case in which to identify a relationship between asset acquisition and Conservative voting. Homeownership – particularly among first-time buyers – represents a significant change to one’s socio-economic position and self-perception (Fischel Reference Fischel2001; McCabe Reference McCabe2013); homeownership is an important political characteristic in Britain that is associated with historic (Thatcher-era) Conservative policies (Garrett Reference Garrett1992; Heath et al. Reference Heath, Curtice and Jowell1991; Huberty Reference Huberty2011); and self-report bias in homeownership is likely to be minimized. However, employing a two-way fixed effects procedure, we find no evidence to suggest that acquiring a home makes one more likely to vote Conservative. These additional null findings, in what we view as a generous test of the patrimony hypothesis, provide further support for the idea that attitudinal differences between those who own assets and those who do not are the product of pre-existing differences rather than the accumulation of assets themselves.
Robustness tests
To further validate our findings and address potential alternative explanations, we conduct a number of follow-up analyses and robustness tests, which provide further support for our central findings. First, we conduct placebo tests with our data. Using a two-way fixed effects approach, we identify the effect of personal financial evaluations on Conservative voting and of patrimony on turnout (supplementary materials section 2.4). That these placebo tests provide evidence in line with prior studies (for example, Hall and Yoder 2021; Tilley et al. Reference Tilley, Neundorf and Hobolt2018) suggests that the main findings represent truly null results rather than an artefact of the data or estimation procedure. Second, to assess potential measurement error in the self-reported measures of asset ownership and value, we compare self-reported home ownership and house price data from US with objective data (see supplementary materials section 2.5). This analysis suggests that self-reported asset measures in the US dataset are broadly accurate. Third, we conduct a number of replications: the absence of an identifiable relationship between patrimony and Conservative voting is replicated under a number of different model specifications and alternative estimators (see supplementary materials sections 3.1 and 3.2).
Fourth, we employ additional datasets to address potential alternative explanations: using the British Election Study, we examine the relationship between patrimony and vote choice alongside the valence and positional dimensions of economic voting (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013; Lewis-Beck and Nadeau Reference Lewis-Beck and Nadeau2011) and using the British Household Panel Study, we test the relationship between patrimony and vote choice in Britain during the period of Labour government from 1997 to 2008. Presented in supplementary materials section 3.3–3.4, results indicate that even when accounting for alternative dimensions of economic voting and examining the patrimony hypothesis in an alternative political (and temporal) context, within-person changes in asset ownership are unrelated to Conservative voting. Finally, given that the acquisition of assets is unlikely to be surprising, we test the possibility that the absence of a patrimony effect is explained by the anticipation of asset ownership. Testing lead effects, or a change in patrimony at T +n , we find some evidence that the future acquisition of patrimony is positively associated with Conservative voting (see supplementary materials section 5.1). However, further analysis of potential lead effects among first-time homeowners provides little evidence in support of this argument (see supplementary materials section 5.2). The overall evidence for lead effects is thus inconsistent.
Discussion
Does the ownership and value of one’s assets have an impact on voting behaviour? Drawing on eleven years of longitudinal data from Britain, we have shown that the patrimony effect is mainly the product of between-person differences between those who own assets and those who do not, rather than the accumulation of assets themselves. Whilst we find some evidence of lead effects, suggesting that the future acquisition of patrimony exerts a small (albeit inconsistent) influence on voting preferences, results on the whole show that voters often do not update their preferences in line with the assets that they acquire. In addition, when we concurrently estimate the influence of patrimony and parental preferences, parental preference appears to explain away much of the association between patrimony and vote choice.
While the use of longitudinal observational data provides a compelling test of the patrimony hypothesis by measuring individual trajectories of asset accumulation and vote choice over time, there are limitations to such a design, which provide innovative avenues for further research. Whilst we control for time-invariant factors through the use of individual fixed effects along with a number of time-varying controls, the self-reported nature of our patrimony measures means that we cannot completely rule out the influence of time-varying omitted variables. Without exogenous variation in assets, we are unable to entirely rule out the possibility that the anticipation of future assets explains the absence of a patrimony effect, although we have addressed this possibility via an analysis of lead effects. To further strengthen our conclusions and assess the true causal effect of asset acquisition on vote choice, future research should test the patrimonial voting hypothesis using a truly exogenous change in asset or via experimental designs and employ objective measures of assets (and changes in assets). Future research should also explore whether patrimony has an impact on voting behaviour in the much longer term (beyond the 3-year lags that we examine). Moreover, while Britain has been the focus of cross-sectional studies on patrimonial economic voting (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013), some studies suggest that this relationship may be weaker in Britain compared to other countries (Hellwig and McAllister Reference Hellwig and McAllister2019; Stubager et al. Reference Stubager, Lewis-Beck and Nadeau2013). Finally, future research could investigate whether the null effect that we report for Britain applies to other contexts as well.
Our research has several implications. First, we cast doubt on the relevance of the patrimonial dimension of economic voting theory and call for a re-evaluation of the ‘compleat’ economic voter (Lewis-Beck et al. Reference Lewis-Beck, Nadeau and Foucault2013). Patrimonial economic voting research typically theorizes and empirically investigates the effects of individual wealth on vote choice. Our results suggest that patrimony conceived and measured in this way falls short of having an effect on the vote, at least in the case of Britain. Even though our results hold under a number of robustness tests, we cannot fully rule out that patrimony may affect the vote choice albeit on a family basis (for example, via inheritance). Future research should consider incorporating family wealth in addition to individual wealth in economic voting models. Second, voters might not necessarily evaluate the implications of government policy (for example, taxation) directly with reference to their personal asset accumulation. Future research could test experimentally if individuals primed with information on the costs and benefits of specific policies on their asset accumulation change their political views.
Further, our findings have implications for our understanding of political behaviour more broadly. First, they shed light on the role of self-interest in vote choice. Research has shown that egocentric economic considerations bear an impact on vote choice (for example, Pop-Eleches and Pop-Eleches Reference Pop-Eleches and Pop-Eleches2012; Tilley et al. Reference Tilley, Neundorf and Hobolt2018; Zucco 2013). Yet, our research paints a more complex picture. A narrow conception of self-interest in terms of the mere accumulation of assets falls short of having an influence on vote choice. This finding lies in tension with some literature showing that lottery winners may update their political preferences in a conservative direction (Doherty et al. Reference Doherty, Gerber and Green2006; Peterson Reference Peterson2016). While we do not directly test this hypothesis, we believe that the discrepancy in the findings may be explained by the difference between investigating sudden and disproportionately large wealth changes in a selected pool of voters with specific traits (for example, a propensity to gamble) and investigating the accumulation of assets in the general population.
Finally, our findings indicate that economic policies that allow people to accumulate assets do not seem to benefit the Conservative Party. This contextualizes why addressing housing policy and supporting business is increasingly becoming a priority across the left-right spectrum, and it is not seen as a right-wing party policy. It also casts doubts on the popular argument among journalists and pundits that British millennials are increasingly left-leaning because of their inability to get on the housing ladder. The findings suggest that this may not be the case and that explanations for the left-leaning turn of younger generations should be sought elsewhere.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123425000146
Data availability
The data and replication instructions can be found at: https://doi.org/10.7910/DVN/KL4UWH.
Acknowledgements
The authors would like to thank Ruth Dassonneville, Michael Lewis-Beck, Patrick Fournier, the participants of the Canada Research Chair in Electoral Democracy research seminar at the University of Montreal, the Political Participation panel at the 2023 ECPR General Conference, and the UK Political Dynamics panel at EPOP 2024 for their helpful comments on earlier versions of this manuscript. We would also like to thank the editors and the three anonymous reviewers for their constructive feedback and useful suggestions.
Financial support
Support for this research was provided in part by the Economic and Social Research Council White Rose Doctoral Training Partnership (grant code: ES/P000746/1).
Competing interests
The authors declare no competing interests.
Open access
