Introduction
One of the most alarming trends in contemporary public opinion is rising partisan polarization, that is, the tendency of individuals to evaluate their preferred political party more positively and to evaluate other parties and their followers more negatively. This phenomenon was first detected in established democracies in North America and Europe (Iyengar and Westwood Reference Iyengar and Westwood2015), but more recently has emerged as a global phenomenon affecting countries at varying stages of democratic development (McCoy and Somer Reference McCoy and Somer2019; Reiljan Reference Reiljan2020; Wagner Reference Wagner2021). This rising polarization is thought to contribute to a host of negative consequences for the political system, including the unwillingness of opposing sides to compromise (Hetherington and Rudolph Reference Hetherington and Rudolph2015), an increase in policy gridlock, and an exacerbation of ethnic and religious divides. Perhaps, most dangerously, polarization has been cited as an important source of ‘democratic backsliding’, as political elites seek to weaken democratic institutions or engage in abuses to prevent the opposing side from exercising power (Orhan Reference Orhan2022).
These disturbing effects at the elite level parallel those found for the effects of polarization among individuals. Polarized individuals are less supportive of democratic norms and values, in particular, being willing to tolerate violations of those norms so long as they are committed by co-partisan elites (Kingzette et al. Reference Kingzette, Druckman, Klar, Krupnikov, Levendusky and Ryan2021). This occurs for a variety of reasons, including a ‘win at all costs’ perspective and increased motivated reasoning, leading individuals to attend exclusively to the cues of elite rhetoric on ‘their side’ of the political divide and to then rationalize their anti-democratic behaviour (Krishnarajan Reference Krishnarajan2023; Svolik et al. Reference Svolik, Avramovska, Lutz and Milačić2023). As a result, polarized individuals are more likely to choose party loyalty over democratic values, and thus fail to impede a country’s democratic decline.
How can these anti-democratic effects of partisan polarization among individuals be mitigated? One possibility is through informational campaigns designed to correct individuals’ misperceptions of the degree to which opposing political parties and partisans obstruct political compromise or support anti-democratic behaviour or political violence, and hence lessen the valuation of ‘party’ in the party versus democracy choice (Mernyk et al. Reference Mernyk, Pink, Druckman and Willer2022). While having laudable aims, such interventions, according to much empirical literature, have yielded generally weak, and, in some cases, backlash effects whereby individuals minimize the validity of the corrective information and bolster their pre-existing views (Druckman Reference Druckman2023).
Another possibility lies with democratic civic education, interventions designed to promote political knowledge, engagement, and support for democratic norms and values among ordinary citizens (SE Finkel and Smith Reference Finkel and Smith2011; Hyde et al. Reference Hyde, Lamb and Samet2023). Since the third wave of democratization in the 1990s, civic education programmes in new democracies have proliferated, with most taking the form of in-person events such as workshops, community forums, or lectures (Hyde et al. Reference Hyde, Lamb and Samet2023), with recent efforts utilizing mass media or social media platforms as a cost-effective, scalable means for delivering civic education messages (Ferrali et al. Reference Ferrali, Grossman and Larreguy2023; S Finkel et al. Reference Finkel, Neundorf and Ramírez2024).
Civic education may succeed in mitigating the negative effects of polarization by enhancing the ‘democratic’ side of the party versus democracy choice. Through the development of norms such as tolerance, political equality, and rejection of political violence, individuals may be more likely to uphold democracy by rejecting the anti-democratic behaviours of elites even from their side of the political divide. Civic education messages may also decrease polarization itself by dampening individuals’ antipathies towards opposing parties and their followers. That civic education can enhance democratic values and norms is well established, yet there have been no studies examining its ability to mitigate the anti-democratic effects of partisan polarization. If civic education can succeed in this respect, it will represent another way that these kinds of interventions can help forestall democratic backsliding.Footnote 1
In this paper, we attempt to fill this gap by testing the impact of an online civic education intervention on the choice of democracy over party – that is, in rejecting a hypothetical candidate from one’s preferred party who endorses anti-democratic norms in favour of a democratically-oriented candidate from a lesser-liked party – among individuals of varying levels of pre-existing partisan polarization. We conducted the experiment on over 41,000 subjects recruited with Facebook and Instagram advertisements in thirty-three countries with differing levels of macro-level political polarization and democratic development.Footnote 2 The treatments consist of short videos emphasizing the different benefits of democratic regimes or a placebo video, with the treatments being identical within and across countries to enhance causal inferences and generalizability.
The results indicate that civic education can promote the choice of democracy over party, with the negative slope of partisan polarization on the likelihood of selecting a democratic candidate from an opposing party being reduced by over 20 per cent among those exposed to the treatment videos. We find strong contextual effects as well, with the positive impacts of civic education being concentrated in autocracies and countries characterized by higher levels of partisan polarization. Further, civic education appears to have a small positive effect on partisan polarization itself, with increases from pre- to post-test being larger among the treated versus placebo groups. While this finding may appear counterintuitive, further analysis provides a more complex view. The effects again are observed primarily in autocratic and more polarized country contexts, where polarization more likely reflects the difference between negative evaluations of anti-democratic parties and the positive evaluations of parties who uphold democratic norms. In these contexts, increases in polarization may be a boon to democratic processes. This interpretation is supported by analysis showing that the treatments had a positive impact on the individual’s overall evaluations of a country’s more democratically-oriented parties. Thus civic education enhances individuals’ tendency to choose candidates who uphold democratic norms but also increases their evaluations of parties that adhere to democratic norms as well.
Research design
Data Collection and Case Selection
Data collection in thirty-three countries took place between May 6 to October 5, 2023.Footnote 3 Of the 10.7 million people who saw our recruitment advertisements, 62,518 people over the age of eighteen were assigned one of the treatments, and nearly 41,000 provided responses on all key variables. Building on lessons from previous research, we used Facebook’s targeting options to create a more balanced sample in terms of age, gender, and education, and incentive-based advertisements to recruit politically less interested respondents (Neundorf and Öztürk Reference Neundorf and Öztürk2022, Reference Neundorf and Öztürk2023; Zhang et al. Reference Zhang, Mildenberger, Howe, Marlon, Rosenthal and Leiserowitz2020). We show the comparisons of key demographics between our sample and the populations of each country in Appendix G. The samples are on average quite representative in terms of age and gender, while still over-representing highly educated people as well as those more interested in politics to a certain extent. In our empirical models, we control for these factors to estimate effects more precisely.
We selected thirty-three countries for our study with various levels of liberal democracy and macro polarization.Footnote 4 To measure levels of liberal democracy, we use the country’s 2022 score on the Varieties of Democracy (V-Dem) Liberal Democracy Index (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Bernhard, Cornell, Fish, Gastaldi, Gjerløw, Glynn, Grahn, Hicken, Kinzelbach, Marquardt, McMann, Mechkova, Neundorf, Paxton, Pemstein, Rydén, Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Uberti, Yt, Wig and Ziblatt2023). The index ranges between 0 and 1, where higher values indicate higher levels of liberal democracy. To measure the countries’ polarization score, we use V-Dem’s 2022 Political Polarization variable (v2cacamps) measuring the degree to which ‘supporters of opposing parties are more likely to interact in a hostile than friendly manner’. Figure 1 plots countries by their Liberal Democracy and Political Polarization scores, showing considerable variation in both.
Liberal Democracy and Macro Polarization
Data: V-Dem (v13).
Experimental treatments
We used simple randomization to assign respondents to one of four study arms consisting of an approximately 3-minute video varying in content.Footnote 5 The content of three videos focused on different beneficial aspects of democracy: 1) institutions, such as independent parliaments and courts that check potential abuses of power; 2) protection of political, social, and economic rights and liberties, and 3) the positive economic and social outputs produced by democratic systems, for example, economic growth, health, and human development. The fourth arm was a placebo video on the advantages of space exploration. In the main analyses presented below, the three treatment arms that promote democracy are pooled and compared to the placebo group, while Appendix J presents results for the individual treatments.
The tone of all videos was informative and upbeat through the use of music, narrative intonation, and the use of positive words. Appendix E shows that randomization produced balanced samples between the pooled treatment and placebo groups on numerous pre-treatment variables.
Variables
Partisan Polarization
To measure partisan polarization, we use Wagner’s (Reference Wagner2021) ‘weighted spread of scores’ variable measuring variation in respondent’s affect on a 0–10 dislike-like scale towards all political parties in a country receiving greater than 5 per cent of the vote in the previous national elections. The measure is calculated as:
$$Sprea{d_i} = \sqrt {\sum\limits_{p\,=\,1}^p {v_p}{{(lik{e_{ip}} - \overline {lik{e_i}} )}^2}} $$
where
$p$
is the party;
$i$
is the individual;
$lik{e_{ip}}$
is the like-dislike score assigned to each party
$p$
by individual
$i$
;
${v_p}$
is the proportion vote share of each party, with a range from 0 to 1, and
$\overline {lik{e_i}} $
is average like for all parties rated by the respondent, weighted by party size. The variable was measured before and after respondents viewed the treatment or placebo videos.
We note the Wagner measure ignores the social identity component that is key to some conceptualizations of partisan polarization (for example, Bankert Reference Bankert2021). However, it has distinct advantages for assessing polarization cross-nationally, as it can be applied in multiparty systems where individuals may have multiple non-exclusive strong partisan likes and dislikes, and in contexts where explicit positive identification may be weak but negative affect towards some parties may nevertheless be intense. The measure is thus well-suited for purposes of enhancing generalizability.Footnote 6
Defending Democracy
The dependent variable is designed to capture the extent to which a respondent will refrain from supporting a hypothetical candidate from a preferred party who acts in an anti-democratic fashion, compared with a candidate from a less-liked party whose behaviour is consistent with democratic norms. This variable, which we call ‘Democratic Defense’, follows studies that use conjoint experiments to demonstrate people’s willingness to support anti-democratic candidates to prioritize their party affiliations or candidate competence (Frederiksen Reference Frederiksen2022; Graham and Svolik Reference Graham and Svolik2020; Svolik et al. Reference Svolik, Avramovska, Lutz and Milačić2023).Footnote 7
Here we use a single static pair question to measure the likelihood of voting for either of two candidates. One of the candidates (A) is from the respondents’ preferred party, that is, the party with the highest evaluation pre-treatment. This candidate is described as engaging in or endorsing some (randomly assigned) violation of democratic norms regarding media freedom, judicial independence, and electoral competition. The opposing candidate (B), from a lesser-liked party, is described as upholding or endorsing the same norm. Respondents were asked which of the two candidates they would be more likely to vote for, on a 0 (certainly vote for undemocratic, preferred party candidate A) to 10 (certainly vote for democratic, non-preferred party candidate B) scale. The candidate descriptions and the vote likelihood measure were taken from existing conjoint experiments.Footnote 8
Results
Partisan polarization and the effects of civic education
We present in Table 1, and visually in Figure 2, the model predicting the likelihood of ‘Defending Democracy’ by choosing a democratic candidate of an opposing party when a preferred party candidate violates democratic norms. The model contains the interaction between the individual’s pre-treatment level of partisan polarization and exposure to the civic education video treatments.Footnote 9 We estimate the model with country fixed effects and include covariates,Footnote 10 with the results being robust when estimated via pooled OLS with standard errors clustered by country.
Treatment effects on defending democracy
*p
$ \lt $
0.05, **p
$ \lt $
0.01, ***p
$\lt $
0.001.
See Appendix I for the full list of coefficients.
Treatment effects on defending democracy by individual-level partisan polarization
Note: Predicted values are based on Model 1, Table 1.
As can be seen in column 1 and the solid line in Figure 2, the effect of partisan polarization is starkly negative in the placebo group (b=−0.36), so that individuals with a maximum polarization of 5 are 1.8 points higher on the 10-point likelihood scale of choosing the anti-democratic candidate compared with non-polarized individuals. This confirms the negative impact of polarization on democratic commitments. However, among individuals exposed to the civic education videos, the negative slope of pre-treatment polarization on anti-democratic choice flattens considerably to −0.28, a drop of 22 per cent from the control group level (as seen in the dashed line in Figure 2). This translates to a 0.37 increase (on a 10-point scale) in the likelihood of the most polarized individuals in the treatment group choosing a democratic candidate compared to similarly polarized individuals exposed to the placebo video. This is clear evidence that civic education exposure mitigates some of the negative effects of polarization on the choice of party over democracy.Footnote 11
We examine the macro-contextual conditional effects of the treatment-polarization relationship in the next columns in Table 1. We compare countries above and below the mean value (2.84) on V-Dem’s characterization of the extent to which the country is divided into polarized camps, and for countries above and below 0.5 on the 0–1 Liberal Democracy Index, with countries above 0.5 being ‘clear democracies’ and those below 0.5 being, at most, ‘ambivalent’ in terms of democracy versus autocracy, which we label as ‘flawed democracies and autocracies’ (Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Bernhard, Cornell, Fish, Gastaldi, Gjerløw, Glynn, Grahn, Hicken, Kinzelbach, Marquardt, McMann, Mechkova, Neundorf, Paxton, Pemstein, Rydén, Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Uberti, Yt, Wig and Ziblatt2023).
The results indicate that the positive effects of civic education on the polarization-democratic choice relationship are more pronounced in more polarized (Model 3) and less democratic macro contexts (Model 5) contexts. These are precisely those contexts where civic education is most needed, given the greater negative effect of polarization on anti-democratic choice among the control group compared to their counterparts in less polarized (Model 2) and more democratic (Model 4) contexts, and given that these contexts are more likely to have viable anti-democratic parties.Footnote 12 The results are consistent with previous civic education research showing greater impact at the micro-level among individuals with greater ‘need’ (SE Finkel and Smith Reference Finkel and Smith2011; Neundorf, Öztürk et al. Reference Neundorf and Öztürk2023).
These conclusions are reinforced when examining the effects on a country-by-country basis, as shown in Appendix K. With caution in interpreting the effects given the smaller sample sizes, the table shows clearly that the effects of civic education on dampening the polarization-defending democracy relationship are strongly related to levels of democracy and partisan polarization. Among the countries with the largest interaction effects (column 5), the top thirteen are in the flawed democracy and autocratic group, and eight of those are polarized. Moreover, there is a strong negative relationship between ‘need’, as evidenced by the magnitude of the slope coefficient for polarization (column 3), and the magnitude of the interaction – the correlation between the two coefficients among the thirty-three countries is −0.71. This is powerful evidence that civic education has greater impacts in precisely the contexts where the deleterious effects of polarization are most pronounced.
The effects of civic education on change in partisan polarization and democratic party affect
Does civic education exposure affect the level of partisan polarization? In Table 2 we present treatment effects on changes between pre- and post-treatment levels of polarization for the entire sample and then for individuals in different democratic and polarized macro contexts, controlling for pre-treatment covariates. For the entire sample (Model 1) there is a slight positive treatment effect on changes in polarization. Further analyses suggest that the effects are concentrated almost exclusively in the more polarized and flawed democratic and autocratic contexts (Models 3 and 5). These effects are consistent with those in Table 1, whereby civic education has the greatest impact in the places where it is most needed, where more viable anti-democratic parties are likely to be present in the political system. A positive interpretation of this pattern is that individuals exposed to civic education in these contexts come to value democratic parties more and value parties that violate democratic principles less, leading to increases in what might be called ‘positive polarization’ in the face of parties that challenge democratic norms.
Treatment effects on defending democracy
*p
$ \lt $
0.05, **p
$ \lt $
0.01, ***p
$ \lt $
0.001.
See Appendix I for the full list of coefficients
We explore this further in the final two models in Table 2 by estimating the effect of civic education on the individual’s average change in affect toward parties that endorse democratic norms, and those parties that do not. To classify parties, we use the ordinal V-Dem Pluralism variable (v2paplur) from the 2018 V-Party data, treating democratic parties as either ‘committed’ or ‘fully committed’ to electoral and political rights, and undemocratic parties as ‘not committed’ or ‘not at all committed’. The results show that the average affect toward democratic parties (Model 6) significantly increases as a result of the treatments, while there is no change towards anti-democratic parties (Model 7). These results are consistent with a process whereby the positive effect on partisan polarization is due to increases in affect towards parties upholding democratic norms, leading to greater distances between evaluations of democratic and anti-democratic parties.
The results in Tables 1 and 2, moreover, are complementary, as they show in different ways how civic education leads individuals to reward parties that are more democratically oriented. In the static-pair analysis (Table 1), civic education results in a lower likelihood of voting for one’s preferred party if it hypothetically engages in undemocratic behaviour. In the party affect analysis (Table 2), civic education leads individuals to evaluate more positively parties that are objectively (according to V-Dem) more democratic, while not increasing support for anti-democratic parties. We interpret all these findings as positive impacts of civic education.Footnote 13
Conclusion
We tested the effect of an online civic education intervention conducted on over 41,000 individuals in thirty-three countries that was designed to promote the ‘democracy’ side of the ‘party over democracy’ choice by emphasizing the benefits of democratic versus autocratic regimes. The results are striking: exposure to civic education messages significantly dampens the negative effect of partisan polarization on anti-democratic co-partisan candidate choice, with the negative effect of polarization on democratic choice being reduced in the treatment group by approximately 22 per cent relative to the placebo group.
Further analysis shows that the effects are concentrated in countries where the negative impact of polarization was most acute, for example, in a range of flawed democratic and autocratic regimes, and in countries characterized by more intense partisan conflict. Civic education also has a small positive effect on polarization itself, with additional analyses showing that this is the result of increased evaluations of parties that uphold democratic norms, resulting in greater differences in affect between democratic and anti-democratic parties.
The paper joins the growing literature on civic education’s favourable effects in various contexts on supportive democratic attitudes (SE Finkel and Smith Reference Finkel and Smith2011; Hyde et al. Reference Hyde, Lamb and Samet2023), showing for the first time the impact of civic education on polarized partisans’ rejection of anti-democratic practices by candidates of a preferred party. That the messages were imparted in a single 3-minute video bodes well for the potential of online civic education, given its low cost and potential for wide reach via social media platforms (Ferrali et al. Reference Ferrali, Grossman and Larreguy2023). If future research can show longer-lasting effects of the video treatments (see [reference removed for anonymity] for one such effort), this would provide even greater support for online civic education in promoting positive democratic outcomes.
The results here suggest several additional avenues for future work. Our sample contains countries with widely varying levels of democracy and polarization among English, Spanish, and Turkish-speaking countries. But it is unclear whether the results can be generalized to the global population of democracies and autocracies, particularly if the relationship between macro levels of polarization and democracy seen in our sample conditioned the individual-level effects that we found. Moreover, the use of Facebook samples tends to some extent to over-represent highly educated and politically interested individuals. We note that these online samples are likely to be more representative of the general population than typical face-to-face civic education interventions (SE Finkel and Smith Reference Finkel and Smith2011); further, in our analyses, we find no significant differences in treatment effects for individuals with different levels of education or political interest. Still, additional research should be conducted with more diverse samples to improve our understanding of the generalizability of these results.
More theoretically, we need a greater understanding not only of the causes and negative effects of partisan polarization on democratic attitudes but also of how and in what contexts polarization may be beneficial for upholding democracy in the face of anti-democratic challenges (Harteveld et al. Reference Harteveld, Mendoza and Rooduijn2022). Further, we need to know how messages can be crafted to call attention to anti-democratic parties and their behaviour so that the positive effects seen here on evaluations of democratic parties can be matched by negative evaluations of parties who fail to uphold democratic norms. As such parties become more common in contemporary autocracies and backsliding democracies, this challenge will take on greater urgency.
Supplementary Material
The supplementary material for this article can be found at https://doi.org/10.1017/S000712342500016X
Data Availability Statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/B0OKER.
Acknowledgements
The authors are grateful for the helpful comments from Lucas Perello and Oana I. Armeanu at the 2024 Midwest Political Science Association Meetings. We thank Mert Moral and the participants at the workshop, ‘Promoting Democratic Citizenship Online: Challenges and New Directions in Civic Education’, Istanbul, Türkiye September 2024, and Bret Barrowman and the participants at the workshop, ‘The Case for Democracy’, IRI, Washington, DC December 2024.
Funding Statement
The research of this project was generously funded by a European Research Council Consolidator Grant, ‘Democracy under Threat: How Education can Save it’ (DEMED) (Grant number: 865305).
Competing Interests
None.
Ethical Standards
The study received ethical approval (number: 400210195) from the Ethics Committee of the College of Social Sciences at the University of Glasgow on May 17, 2022; see Appendix B for further discussion.
Open access
