Sage Journals: Discover world-class research

Abstract

As immigration issues waned in salience during the COVID-19 pandemic, populist radical right (PRR) parties repositioned themselves by politicizing various pandemic policies. In light of this changing political landscape, scholars have analyzed what factors are associated with PRR voting. Yet, most studies focus on small sets of covariates that could easily ignore other key determinants. To address this limitation, we use MI-LASSO logistic regression, which is a more inductive data-driven approach that can incorporate a huge number of covariates. Our research analyzes the key determinants of voting for the People’s Party of Canada—a PRR party that rose rapidly during the pandemic. Using the 2021 Canadian Election Study dataset (N = 14,841), we confirm that PRR voters in the pandemic were both protest and policy-oriented voters. They were protest voters since anti-establishment attitudes consistently correlate with their vote choice. On the other hand, PRR voters’ policy concern was about pandemic policies rather than immigration, as nativist attitudes never emerge as key determinants. Additionally, we uncover that the ideological placement of the mainstream right party and the defense of hate speech are strong correlates, while conventional variables like sociodemographics are not. These findings enrich our understanding of PRR voting during the pandemic.

Keywords

Populist radical right COVID-19 pandemic LASSO multiple imputation voting behavior

Introduction

Before the COVID-19 pandemic, the literature on populist radical right (PRR) parties suggested that citizens vote for this party family when they demand nativist policies and are dissatisfied with the establishment (Cohen, 2020). However, since the pandemic occurred, whether these factors are still associated with PRR voting is less certain for two reasons. First, the immigration issue, which many PRR parties and candidates had focused on in the past, declined in salience (Crulli, 2023; Pirro, 2022). Second, the pandemic altered the dimensions of electoral competition: most PRR parties politicized restrictive health policies (Lehmann and Zehnter, 2022; Zanotti and Turnbull-Dugarte, 2022). Hence, what constitutes the key determinants of PRR voting during the pandemic is an important question, as we move past the pandemic and look to understand the “new normal” of political competition.

Recently, there has been a growth in research investigating this topic. These studies find that pandemic-related factors and anti-establishment attitudes correlate with supporting the PRR (Aviña and Sevi, 2021; Kessenich and Van der Brug, 2022; Medeiros and Gravelle, 2023; Miller et al., 2022; Serani, 2023). However, because these studies mostly focus on specific small sets of covariates, they easily overlook other key determinants. For instance, as immigration issues became less salient, do nativist attitudes still correlate with PRR voting? Relatedly, as PRR parties politicized lockdown measures and vaccination programs (Lehmann and Zehnter, 2022; Zanotti and Turnbull-Dugarte, 2022; Zulianello and Guasti, 2023), what is the explanatory power of pandemic-related factors on PRR voting in comparison to nativist attitudes? These questions are fundamental to understanding PRR voters’ characteristics during the pandemic and can prognose whether PRR’s support is transient in the post-pandemic world. However, because recent works include only a limited number of covariates, they cannot provide adequate insight into these questions. To circumscribe this limitation, we study PRR voting using MI-LASSO (multiple imputation-least absolute shrinkage and selection operator) logistic regression, which is a more inductive data-driven approach that can incorporate a huge number of variables.

Our research studies the key determinants of voting for the People’s Party of Canada (PPC)—a PRR opposition party in Canada—in the 2021 general election. Since its foundation in 2018 by a disgruntled Conservative Member of Parliament, Maxime Bernier, the PPC campaigned against immigration policies and multiculturalism (Budd, 2021; Erl, 2021), and promoted a culture of individual freedom from government incursion. Its 2019 platform included statements about preserving freedom of expression and gun rights, “reflect[ing] Bernier’s longstanding libertarian political philosophy prioritizing free market economics and individual freedom” (Budd, 2021: p. 169). After the pandemic occurred, however, the PPC followed PRR parties in other countries by becoming the most vocal critic of pandemic policies. During the 2021 election, its core message challenged vaccine-based restrictions in the name of freedom (Medeiros and Gravelle, 2023), which echoes that PRR parties were “freedom defenders” during the pandemic (Lehmann and Zehnter, 2022; Zanotti and Turnbull-Dugarte, 2022). Although the PPC did not win a seat in Parliament, its national vote share was 4.9%, which was threefold compared to the 2019 pre-pandemic election. In several constituencies, its vote share even reached more than 10%. However, a recent poll shows that the PPC’s support dropped to a mere 2% in November 2023 (Kurl and Korzinski, 2023). The PPC’s rapid rise and fall raises the question of what factors contributed to its significantly stronger support during the pandemic.

Our study uses the 2021 Canadian Election Study (CES) dataset (Stephenson et al., 2022) that has an unusually large, nationally representative sample (N = 14,841). With its fine-grained measures, we can incorporate more than 250 covariates in a MI-LASSO model. Our study’s contributions are twofold. The first is methodological: we bring MI-LASSO to the field of voting behavior, and more broadly, research on politics, for it is still an underutilized modeling strategy. As we demonstrate, a MI-LASSO model can uncover several correlates of PRR voting that recent works have not identified. The second contribution is empirical. Our findings suggest that, in line with the current literature, PRR voters in the pandemic were both protest voters and policy-oriented. They were protest voters since anti-establishment attitudes consistently correlate with their vote choice. However, their policy concern was not about immigration, but rather about the pandemic. In the variable selection process, nativist attitudes never emerge as key determinants. Even if we manually add nativist attitudes, their explanatory power on PRR voting is much lower than pandemic-related factors. Also, we uncover that conventional variables like sociodemographics are not key determinants of PRR voting, while the mainstream right party’s ideological placement and defense of hate speech are.

PRR voting during the COVID-19 pandemic

In studies of the individual-level determinants of PRR voting during the pandemic, scholars mainly posit that pandemic-related factors play a key role. This applies regardless of whether they are studying PRR incumbents or PRR in opposition. This emphasis on pandemic-related factors can be linked to the fact that PRR parties embraced vaccination skepticism and politicized pandemic policies (Crulli, 2023; Pirro, 2022; Serra-Silva and Santos, 2022).

Concerning PRR incumbents, research focusing on the United States shows that COVID-19 exposure decreased support for Trump in the 2020 election (Aviña and Sevi, 2021). Likewise, other scholars demonstrate that concerns about public health, knowledge of COVID-19 infections, and health behavior were associated with voting for/against Trump (Gadarian et al., 2021; Miller et al., 2022; Shino and Smith, 2021). Experimental evidence also suggests information on health mismanagement led to a decrease in Trump support among the elderly (Neundorf and Pardos-Prado, 2022).

Regarding PRR in opposition, scholars similarly find PRR voting is associated with pandemic-related factors (Kessenich and Van der Brug, 2022; Pickup et al., 2020). Yet, these works add the notion of anti-establishment to the determinants of PRR voting. For instance, a study in Germany finds that negative attitudes towards government health restrictions are correlated with continued PRR support (Bayerlein and Metten, 2022). Similarly, PRR voters in Austria were much more against the government’s compulsory vaccination program than other mainstream voters (Paul et al., 2021). In a study that is closest to ours, scholars demonstrate that both opposition to health restrictions and populist attitudes were associated with PRR voting in Canada (Medeiros and Gravelle, 2023), but the study highlights that the former is a much stronger factor than the latter.

Following these insights, we expect pandemic-related factors and anti-establishment attitudes will be associated with voting for the PRR opposition party. Moreover, our study explores whether and to what extent other correlates of PRR voting suggested by pre-pandemic research mattered in a pandemic-era election. This is the empirical contribution that we make by using MI-LASSO, as the abovementioned literature only covers a small set of covariates.

Data

Our dataset is the 2021 Canadian Election Study (CES) (Stephenson et al., 2022). We only include respondents who answered the vote choice questions (N = 14,841). The dependent variable is voting for the PPC, coded as 1 if the respondent chose the PPC in the 2021 election (N = 464) and 0 otherwise (abstainers included) (N = 14,377).

The CES includes many variables that have been associated with PRR voting in previous research. These variables include sociodemographics; personality, political attitude, and self-reported behavioral measures; and voting motivations. We include all of them in our MI-LASSO model. The dataset has 252 usable covariates: 34 are continuous, 101 are ordinal, 13 are binary, and 104 are created from nominal indicators.¹ We classify these covariates into the following categories:

pandemic-related attitudes and behaviors;

health status self-assessment;

anti-establishment attitudes: political satisfaction, trust in political and social institutions, and items of populism;

efficacy, political interest, and political knowledge;

social trust;

left/right placement of parties and ideological self-placement;

experience of spoiling votes and reason for choosing the party in the 2021 election;

political information consumption;

political participation;

Big Five personality traits;

cultural attitudes: environmental protection, nativism, freedom of speech and equal rights, personal choice, Quebec issues, indigenous peoples, and gender issues;

identity: elements that constitute personal and Canadian identity;

economic attitudes, economic perceptions, and economic status; and

sociodemographics.

Appendix Part A documents in detail the survey wordings of each construct and variable operationalization. Since the CES uses a split questionnaire design, we also document which variables belonged to the randomized modules.

Estimation method

To examine which variables were the key determinants of PRR voting during the pandemic, we use MI-LASSO logistic regression. LASSO was introduced by Tibshirani (1996) to select variables based on the absolute size of coefficients. More recently, scholars have adopted it to model vote choice with high-dimensional data (Angelucci et al., 2020; Mauerer et al., 2015). The basic idea of LASSO logistic regression is to introduce a penalty term that is subtracted from the log-likelihood (Friedman et al., 2010: pp. 6–7). Unlike standard logistic regression that maximizes the log-likelihood, LASSO logistic regression maximizes the penalized log-likelihood: ${\hat{β}}_{p e n a l i z e d} (λ) = \arg \max l (β) - λ \sum_{j = 1}^{p} | β_{j} |$ where $p = 1, \dots, j$ denotes the individual-specific covariate; each $β_{1} \dots β_{j}$ is a p-dimensional coefficient vector related to the covariates; and $λ$ is the tuning parameter that controls the overall penalty level.

A major advantage of a LASSO estimator over an unpenalized maximum likelihood (ML) estimator is that at different penalty levels of $λ$ , coefficient estimates that are small in magnitude are set at zero. So, even if the number of variables ( $p$ ) is huge and exceeds the sample size ( $n$ ), the model is still identifiable and LASSO can mitigate the issue of multicollinearity (Ahrens et al., 2020). Also, because a LASSO model is more parsimonious, its estimator is more efficient than an unpenalized ML estimator (Mauerer et al., 2015).

However, one disadvantage of LASSO is handling missing data by listwise deletion. Because an individual is deleted if he/she has missing data on any variable, a LASSO model can end up being non-estimable due to insufficient observations. Thus, listwise deletion can pose a challenge to LASSO as the number of variables increases. Moreover, to save costs and reduce respondents’ cognitive burden, large-scale surveys increasingly use split questionnaire design to randomize certain question blocks to subsets of respondents (Axenfeld et al., 2022); in such cases, respondents who do not answer a block of questions are missing completely at random. Because of listwise deletion, LASSO drops these respondents, which considerably reduces the amount of information.

To address this problem, statisticians propose MI-LASSO (Chen and Wang, 2013; Liu et al., 2016). We borrow this method to uncover the determinants of PRR voting in the pandemic era. The procedure is shown in Figure 1. First, we use chained equations to impute five datasets (m = 5). In Appendix Part B, we report the rate of missingness, procedures of imputations, and the diagnostics of all 252 covariates. Also, in the robustness check, we discuss whether the violation of the missing at random (MAR) assumption affects our inference. Second, we randomly split the sample into two—one for variable selection ( ${S a m p l e}_{s e l e c t}$ ) and the other for inference ( ${S a m p l e}_{i n f e r}$ ). Third, we use ${S a m p l e}_{s e l e c t}$ to fit a LASSO model for each imputed dataset.² Regarding the suitable value of $λ$ , we use BIC for each imputed dataset. Each LASSO model results in a set of variables ( $p_{m}$ ) that are strongly associated with our dependent variable (PPC voting). As we shall see, the $p_{m}$ of the five datasets are not fully equivalent, as there is sample-to-sample variability due to the imputation process. Therefore, to ensure consistency, we pick those variables that are present in at least four of the five models. Last, we use ${S a m p l e}_{i n f e r}$ to compute the key determinants’ coefficients and their standard errors according to Rubin’s rules (Rubin, 2004). To evaluate the predictive performance of our model, we follow the literature and use the AUC (Muchlinski et al., 2016).³

Figure 1.

Procedures of the MI-LASSO model.

Results

Variable selection

Table 1 shows the selected determinants for each imputed dataset. Notably, the number of significant determinants is similar in all five columns, ranging from 23 to 27. Crucially, 18 covariates are present at least four times across the five models.

Table 1.

Key determinants of MI-LASSO logistic regression.

Variables	MI dataset 1	MI dataset 2	MI dataset 3	MI dataset 4	MI dataset 5
Pandemic-related variables
Vaccine mandate when traveling by air/rail	X		X	X	X
Vaccine mandate when going to a bar/restaurant			X		X
Vaccine mandate when working in a hospital	X	X	X	X
Liberty threatened by public health recommendations	X	X	X	X	X
Dissatisfied with how provincial government handled COVID outbreak	X	X	X	X	X
Dissatisfied with how local government handled COVID outbreak	X	X		X
Having been vaccinated	X	X	X	X	X
Anti-establishment attitude
Distrust the media	X	X	X	X	X
Distrust the public service					X
Compromise in politics is just selling out on one’s principles	X	X	X	X	X
Most politicians are trustworthy	X			X
Those elected to Parliament soon lose touch with the people	X	X	X		X
Dissatisfied with the way democracy works in Canada	X			X
Dissatisfied with the performance of the federal government under Justin Trudeau		X	X	X
Dissatisfied with the performance of provincial government		X	X		X
Efficacy
The government does not care much about what people like me think				X
Left/right placement
Ideological position of LPC		X		X
Ideological position of CPC	X	X	X	X	X
Ideological position of PPC					X
Experience of spoiling votes and reason for choosing the party in the 2021 election
Reason for choosing the party: Party policies	X	X	X	X	X
Reason for choosing the party: Other	X	X	X	X	X
Spoiling votes			X
Political participation
Used social media to discuss politics or political issues	X	X	X	X	X
Followed any elected officials or candidates for office on social media	X	X	X	X	X
Attended a rally or participated in a protest or demonstration	X	X	X		X
Being a member of a provincial or federal political party					X
Cultural attitude
The federal government should continue the carbon tax to reduce greenhouse gas emissions	X	X	X	X	X
Spending on the environment	X	X		X	X
Do you think that climate change is happening?	X		X
Illegal to say hateful things publicly about racial, ethnic, and religious groups	X	X	X	X	X
The best way to protect women’s interests is to have more women in parliament	X		X	X
Rating of feminists		X	X		X
Abortion			X
Less should be done for women		X		X	X
Economic attitudes, economic perceptions and economic status
More/less should be done to reduce the inequality gap in Canada	X	X		X	X
The government should leave it entirely to the private sector to create jobs			X
People who don’t get ahead should blame themselves, not the system			X
Is income inequality a big problem in Canada?			X		X
No. of observations	7420	7420	7420	7420	7420
No. of covariates	252	252	252	252	252
No. of covariates selected	23	24	27	24	26
No. of cross-validation folds	10	10	10	10	10

Note: X denotes the variables that were selected by the corresponding LASSO model.

To begin, pandemic-related factors are associated with PPC voting. These constructs include attitudes toward vaccine mandates, perception of liberty threat due to public health recommendations, dissatisfaction with how the provincial government handled the coronavirus outbreak, and whether respondents are vaccinated. Also, three constructs of anti-establishment attitudes correlate with PPC voting. These constructs are media distrust, attitudes toward compromise in politics, and beliefs about whether parliamentarians lose touch with the people. Hence, in line with our expectations and existing literature, both pandemic-related factors and anti-establishment attitudes are key determinants of PRR voting during the pandemic.

Next, we move on to the determinants that have not been addressed by existing research. First, in terms of party-related variables, left/right placement of the Conservative Party of Canada (CPC), choosing their preferred party due to party policies, and other reasons⁴ are three key correlates of PPC voting. Second, three items of political participation are associated with PPC voting. They are, respectively, using social media to discuss politics, following politicians on social media, and attending a rally or participating in a protest/demonstration. Third, three cultural items are correlated with PPC voting, including attitudes towards (i) carbon taxes, (ii) environment spending, and (iii) hate speech. Last, one item concerning economic attitudes emerges: attitudes toward the inequality gap.

Overall, the variable selection process reveals that pandemic-related factors and anti-establishment attitudes are associated with PRR voting. These findings confirm recent works on pandemic-era PRR voting (Aviña and Sevi, 2021; Medeiros and Gravelle, 2023; Miller et al., 2022; Serani, 2023). Importantly, we uncover some novel findings as well. First, the most notable finding is that nativist attitudes never emerge in the variable selection process. This is worth emphasizing because the absence of nativist attitudes raises the question of whether PRR voters’ policy concern has changed, as PRR parties politicized various pandemic policies (Lehmann and Zehnter, 2022; Zanotti and Turnbull-Dugarte, 2022; Zulianello and Guasti, 2023). In the next section, we compare the explanatory power of pandemic-related factors on PRR voting and that of nativist attitudes. Second, we uncover several correlates of PRR voting in the pandemic-era that recent works have not yet clearly identified. These determinants include, for instance, the ideological placement of the mainstream right party and the defense of hate speech. Third, we also note that sociodemographic covariates are not selected in all models.

Inference

We use ${S a m p l e}_{i n f e r}$ to analyze the coefficients’ direction and magnitude. The average AUC of our model is 0.952. As only 18 out of 252 covariates are used (<8%) to fit ${S a m p l e}_{i n f e r}$ , our model’s predictive performance is considered excellent. To facilitate understanding, Figure 2 illustrates the selected variables’ coefficients (for full regression results, see Appendix Table C1.1). Below we discuss the results in detail, but to put them into context, the top-five determinants with the largest coefficients are: (1) policies as a reason to choose the party, (2) liberty is threatened by public health recommendations, (3) ideological placement of CPC, (4) dissatisfaction with how provincial government handled COVID outbreak, and (5) distrust the media.

Figure 2.

Coefficient plot: key determinants of PPC voters. Note: Thick and thin error bars represent 90% and 95% confidence intervals. Full results of the binomial logistic regression are reported in Table C1.1 of the appendix. * denotes a dummy of a nominal variable; the reference is those who do not choose that category.

We first consider pandemic-related factors. PPC voters are more prone to disagree with having vaccine mandates in hospitals (p < 0.1) and in public transport (p < 0.05) than other citizens. Also, they are more likely to feel their liberty is threatened by public health recommendations (p < 0.001) and be dissatisfied with how the provincial government handled the coronavirus outbreak (p < 0.001). Relatedly, PPC voters are less likely to be vaccinated (p < 0.05). These findings indicate that one major concern of PPC voters is pandemic policies, which suggests PRR voters’ concerns reflect the PRR’s politicization of COVID-19 policies. Also noteworthy is that the threatened liberty item is the second strongest determinant.

Second, regarding anti-establishment attitudes, PPC voters are more likely to distrust media than other citizens (p < 0.01). In addition, they are more prone to consider compromise in politics as selling out on one’s principles (p < 0.01). PPC voters are also more likely to think parliamentarians lose touch with the people, but the estimate is insignificant. Nevertheless, these results suggest PPC voters can partially be protest voters, as they have stronger anti-establishment attitudes than other citizens.

Third, for party-related variables, PPC voters are more likely to perceive the mainstream conservative party as more left-wing than other citizens (p < 0.001). This echoes previous findings of contrast effects among PRR voters (Chan, 2022). Regarding the most important reason for choosing a party, PPC voters are more likely to mention party policies than other citizens (p < 0.001). This determinant has the largest magnitude, signifying that PPC voters are partially policy-oriented as well. By delving into the open-ended answers about why people chose the PPC (Table C1.2), we can articulate two main logics: they mention opposing pandemic policies, including vaccine mandates and vaccine passports, and voting for the PPC to protest mainstream parties/party leaders.

Fourth, regarding political participation, PPC voters are more likely to use social media to discuss politics and follow politicians on social media, but these coefficients are insignificant. The only significant estimate is attending a rally or participating in a protest or demonstration (p < 0.01). This can be explained by the fact that PPC voters often protested pandemic restrictions.

Fifth, for cultural attitudes, PPC voters are less prone to support a carbon tax and more likely to support reducing environmental spending than other citizens, but these coefficients are insignificant. PPC voters are significantly more likely to disagree that it should be illegal to say hateful things publicly about racial, ethnic and religious groups (p < 0.01). Last, for economic attitudes, PPC voters are more likely to think more should be done to reduce the inequality gap, but the coefficient is insignificant.

Taken together, these findings provide important updates to the determinants of PRR voting during the pandemic. First and foremost, the policy dimension of PRR voters’ concern is not about immigration but about pandemic policies. This is confirmed by comparing the explanatory power of pandemic-related factors in determining PRR voting to that of nativist attitudes. We do so by considering the drop in AUC relative to the full model in Figure 2 after regressing PPC voting on these two sets of variables separately.⁵ The drop in AUC for pandemic-related factors is 0.029, while the drop for nativist attitudes is 0.204. This huge difference in AUC decrease signifies that pandemic-related factors are much more important determinants than nativist attitudes.

Second, confirming recent works, anti-establishment attitudes are strong correlates of PPC voting. This suggests that PRR voters during the pandemic are protest voters because they are dissatisfied with mainstream parties and the media. Third, we highlight two determinants of PRR voting during the pandemic that have not been covered in recent works. One is the ideological placement of the mainstream right party. This contrast effect can be linked to the mainstream right party’s support for some public health restrictions and suggests that the PPC’s attraction might come from being more ideologically extreme than the mainstream right-wing party. Another uncovered determinant is the defense of hate speech. This higher propensity to defend hate speech echoes well with the reasons why voters chose PRR parties during the pandemic: they aimed to abrogate all government interventions in the name of freedom.

Robustness checks

We conduct three supplementary analyses. First, we switch ${S a m p l e}_{s e l e c t}$ and ${S a m p l e}_{i n f e r}$ , to see if the key determinants and the magnitude of coefficients are consistent. The key determinants remain similar (Table C2.1), though there are minor deviations. First, for pandemic-related factors, the coefficients of having vaccine mandates in hospitals and public transport are less robust (p = 0.132 and p = 0.250 respectively) and dissatisfaction with how the provincial government handled COVID-19 is only relevant three times in the variable selection process. Still, the results maintain that PPC voters’ concern is pandemic policies. Second, dissatisfaction with the federal government is significant five times in the variable selection process. This further confirms the protesting nature of PPC voters, although attitudes towards compromise in politics only matter three times. Third, for cultural attitudes, the coefficient of spending on the environment becomes significant (p < 0.05), hinting that PPC voters are anti-environmentalists (see Figure C.1 and Table C2.2).

Second, we check whether the violation of the MAR assumption affects our inference. Accordingly, we create an indicator for each predictor: respondents who answer “Don’t know/Prefer not to answer” are coded as 1 and 0 otherwise. Then, we regress these indicators on the PPC dummy separately. Setting α at 0.05, 88% of the predictors pass the test, while 30 of them fail (Table C3.1). We add these 30 failed indicators into the MI-LASSO model. No missing indicators are chosen in the variable selection. That means the missingness of these variables is not strongly associated with PPC voting (Table C3.2).

Last, due to the small number of PPC voters relative to other citizens, our analysis may underestimate the probability of this rare event. Hence, we follow King and Zeng’s advice (2001) in conducting “rare event logistic regression” in the inference stage. Reassuringly, the rarity of PPC voters does not affect our findings (Figure C.2 and Table C.4).

Conclusion

We apply a novel modeling strategy—MI-LASSO logistic regression—to analyze the key determinants of PRR voting in a pandemic-era election. We study the PPC in Canada, as its rapid rise and fall raises a question about the key characteristics of PRR voters during the pandemic. Confirming recent works, we find PRR voters in the pandemic were both protest voters and policy-oriented. They were protest voters since anti-establishment attitudes consistently correlate with their vote choice. They were also policy-oriented, but their policy concern was about pandemic policies instead of immigration, as nativist attitudes never emerge as key determinants. Even if we manually add nativist attitudes to the model, their explanatory power is still much weaker than that of pandemic-related factors. This implies a change in policy dimensions among PRR voters during the pandemic. Moreover, we uncover that the mainstream right party’s ideological placement and the defense of hate speech strongly correlate with PRR voting, which recent works have not yet identified. Also worth mentioning is that sociodemographics are not key determinants. Overall, this research helps update the profile of PRR voters in the pandemic.

There are some limitations that only future works can address. First, our research focuses on a single country in which the PRR rose rapidly during the pandemic. Hence, more studies are needed to test whether our findings are generalizable. Nevertheless, the key determinants of PRR voting we uncovered should be relevant to other countries because, like other democratic regimes, the Canadian mainstream parties were almost in consensus about COVID-19 restrictions while the PPC was the sole challenger (Merkley et al., 2020). Second, our research is descriptive and does not offer causal explanations: the determinants we uncover may be endogenous to PRR voting and the direction of the causal arrow is unknown. Thus, we cannot ascertain what caused citizens to vote for PRR during the pandemic.

Still, understanding the profile of PRR voters during the pandemic offers guidance for future research on their political behavior. Returning back to our Canadian case, the PPC’s support merely reaches 2% in a public poll at our time of writing, which resembles the 1.6% popular vote in the 2019 pre-pandemic election. This low support compared to the pandemic era can be linked to two of our findings. One is the importance of pandemic-related factors: as the pandemic becomes less relevant, the PPC will have a hard time consolidating its voter base. Another is the ideological placement of the mainstream right party. Because the Conservative party elected a new populist leader—Pierre Poilievre—and shifted rightward, previous PPC voters may return to supporting the mainstream right party, as research suggests the mainstream right party’s positioning can tame PRR support (Ryan, 2016). Therefore, our analysis hints that the PRR in Canada might fade back into being a minor player if it only appeals to anti-establishment sentiment.

Supplemental Material

Supplemental Material - Using MI-LASSO to study populist radical right voting in times of pandemic

Supplemental Material for Using MI-LASSO to study populist radical right voting in times of pandemic by Ka Ming Chan and Laura B Stephenson in Research & Politics

Footnotes

Acknowledgments

An earlier version of this research was presented at the 2022 Toronto Political Behavior Workshop and the 2023 Consortium on Electoral Democracy Forum.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research,authorship,and/or publication of this article: This research project is funded by the Social Sciences and Humanities Research Council of Canada (895-2019-1022).

ORCID iD

Ka Ming Chan

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Ahrens

Hansen

Schaffer

(2020) lassopack: model selection and prediction with regularized regression in Stata. STATA Journal 20: 176–235.

Angelucci

De Sio

Paparo

(2020) Europe matters … upon closer investigation. Italian Political Science Review 50: 334–349.

Aviña

Sevi

(2021) Did exposure to COVID-19 affect vote choice in the 2020 presidential election? Research and Politics 8: 1–4.

Axenfeld

Bruch

Wolf

(2022) General-purpose imputation of planned missing data in social surveys: different strategies and their effect on correlations. Statistics Surveys 16: 182–209.

Bayerlein

Metten

(2022) The impact of COVID-19 on the support for the German AfD: jumping the populist ship or staying the course? Politische Vierteljahresschrift 63: 405–440.

Budd

(2021) Maple-glazed populism: political opportunity structures and right-wing populist ideology in Canada. Journal of Canadian Studies 55: 152–176.

Chan

(2022) The making of radical-right voters: persuasion and contrast effects in a dynamic political context. Political Psychology 43: 1043–1060. DOI: 10.1111/pops.12801

Chen

Wang

(2013) Variable selection for multiply-imputed data with application to dioxin exposure study. Statistics in Medicine 32: 3646–3659.

Cohen

(2020) Between strategy and protest: how policy demand, political dissatisfaction and strategic incentives matter for far-right voting. Political Science Research and Methods 8: 662–676.

10.

Crulli

(2023) Unmet expectations? The impact of the Coronavirus crisis on radical right populism in North-Western and Southern Europe. European Journal of Cultural and Political Sociology 10: 379–401.

11.

Erl

(2021) The people and the nation: the “Thick” and the “thin” of right-wing populism in Canada. Social Science Quarterly 102: 107–124.

12.

Friedman

Hastie

Tibshirani

(2010) Regularization paths for generalized linear models via coordinate descent. Journal of Statistical Software 33: 1–22.

13.

Gadarian

Goodman

Pepinsky

(2021) Partisanship, health behavior, and policy attitudes in the early stages of the COVID-19 pandemic. PLoS One 16: e0249596. DOI: 10.1371/journal.pone.0249596.

14.

Hastie

Tibshirani

Wainwright

(2015) Statistical Learning with Sparsity. New York: Chapman and Hall/CRC.

15.

Kessenich

van der Brug

(2022) New parties in a crowded electoral space: the (in)stability of radical right voters in The Netherlands. Acta Politica. DOI: 10.1057/s41269-022-00269-0

16.

King

Zeng

(2001) Logistic regression in rare events data. Political Analysis 9: 137–163.

17.

Kurl

Korzinski

(2023) Carbon pricing troubles drive a widening gap between leading CPC and lagging Liberals in vote intention. Retrieved from https://angusreid.org/carbon-pricing-vote-intention/

18.

Lehmann

Zehnter

(2022) The self-proclaimed defender of freedom: the AfD and the pandemic. Government and Opposition. DOI: 10.1017/gov.2022.5.

19.

Liu

Wang

Feng

, et al. (2016) Variable selection and prediction with incomplete high-dimensional data. Annals of Applied Statistics 10: 418–450.

20.

Mauerer

Pößnecker

Thurner

, et al. (2015) Modeling electoral choices in multiparty systems with high-dimensional data. Journal of Choice Modelling 16: 23–42.

21.

Medeiros

Gravelle

(2023) Pandemic populism: explaining support for the People’s Party of Canada in the 2021 Federal Election. Canadian Journal of Political Science/Revue Canadienne de Science Politique 56: 413–434.

22.

Merkley

Bridgman

Loewen

, et al. (2020) A rare moment of cross-partisan consensus: elite and public response to the COVID-19 pandemic in Canada. Canadian Journal of Political Science 53: 311–318.

23.

Miller

Woods

Kalmbach

(2022) The impact of the Covid-19 pandemic in a polarized political system: lessons from the 2020 election. Electoral Studies 80: 102548.

24.

Muchlinski

Siroky

, et al. (2016) Comparing random forest with logistic regression for predicting class-imbalanced civil war onset data. Political Analysis 24: 87–103.

25.

Neundorf

Pardos-Prado

(2022) The impact of COVID-19 on Trump’s electoral demise: the role of economic and democratic accountability. Perspectives on Politics 20: 170–186.

26.

Paul

Eberl

J-M

Partheymüller

(2021) Policy-relevant attitudes toward COVID-19 vaccination: associations with demography, health risk, and social and political factors. Frontiers in Public Health 9: 671896. DOI: 10.3389/fpubh.2021.671896

27.

Pickup

Stecula

van der Linden

(2020) Novel coronavirus, old Partisanship: COVID-19 attitudes and behaviours in the United States and Canada. Canadian Journal of Political Science/Revue Canadienne de Science Politique 53: 357–364.

28.

Pirro

ALP

(2022) Performing (During) the coronavirus crisis: the Italian populist radical right between national opposition and subnational government. Government and Opposition. DOI: 10.1017/gov.2022.28.

29.

Rubin

(2004) Multiple Imputation for Nonresponse in Surveys. New York: Wiley.

30.

Ryan

(2016) Does Canadian multiculturalism survive through state repression? Nationalism and Ethnic Politics 22: 342–350.

31.

Serani

(2023) The Covid pandemic enters the ballot box: the impact of conspiracy theories on Italians’ voting behaviour during the COVID-19 crisis. Italian Political Science Review/Rivista Italiana Di Scienza Politica 53: 50–67.

32.

Serra-Silva

Santos

(2022) The 2021 Portuguese Presidential Elections under extraordinary circumstances: covid-19 and the rise of the radical right in Portugal. Mediterranean Politics 1–11.

33.

Shino

Smith

(2021) Pandemic politics: COVID-19, health concerns, and vote choice in the 2020 General Election. Journal of Elections, Public Opinion, and Parties 31: 191–205.

34.

Stephenson

Harell

Rubenson

, et al. (2022) 2021 Canadian Election Study (CES). Harvard Dataverse

35.

Tibshirani

(1996) Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B 58: 267–288.

36.

Zanotti

Turnbull-Dugarte

(2022) Surviving but not thriving: VOX and Spain in times of COVID-19. Government and Opposition. DOI: 10.1017/gov.2022.7.

37.

Zulianello

Guasti

(2023) The demand and supply of pandemic populism: a global overview. Government and Opposition DOI: 10.1017/gov.2023.35.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

1.75 MB

0.00 MB