2.1. Political context and the populist targeting of associations

The political context of civic associations consists of factors both on the demand and supply sides. The specific context that we study is that of populism. In line with the literature, we define populism to be, at its core about, anti-establishment rhetoric that emphasizes the division between the people and the corrupt elite (Mudde, Reference Mudde2007: 29). For spreading populism, the critical element on the demand side is a part of society that feels left out. Such feelings could result from exogenous shocks, like an economic crisis, increased levels of automatization linked to globalization (Stanig and Colantone, Reference Stanig and Colantone2018; Anelli et al., Reference Anelli, Colantone and Stanig2021) or the influx of migrants (Dancygier et al., Reference Dancygier, Egami, Jamal and Rischke2021). Indeed, those who vote for the populist radical right tend to have lower levels of education, come from lower social classes (Arzheimer and Carter, Reference Arzheimer and Carter2006), are often unemployed (Lubbers et al., Reference Lubbers, Gijsberts and Scheepers2003), and are distrusting of the establishment (Berning and Ziller, Reference Berning and Ziller2017). The spreading of populism and anti-establishment rhetoric is more likely to happen in these communities.

This links to a particular dynamic on the supply side. On the one hand, there is an unresponsive government led by mainstream parties that are unable to address the electoral dissatisfaction (O'Grady and Abou-Chadi, Reference O'Grady and Abou-Chadi2019; Abou-Chadi and Krause, Reference Abou-Chadi and Krause2020). On the other, on the supply side, an opportunistic challenger party uses the emergence of a group with similar grievances and rhetorical innovation (De Vries and Hobolt, Reference De Vries and Hobolt2020) to gain electoral support. In particular, we study a party from the “populist radical right wave” (ibid., 29) of challenger parties that combine populism with a right-wing ideology and nativist elements.

Political parties pursue multiple strategies to gain votes, including online, TV, mail, or newspaper campaigns or door-to-door canvassing. We focus on an understudied strategy of targeting civic associations (see also Greskovits, Reference Greskovits2020). Given the ostracism that many populist radical right parties experience, being accused of aggressive, offensive, and anxiety-fueled rhetoric, they may resort to more covert ways of contact with the community. This also aligns with the idea that challenger parties choose unconventional strategies and channels to spread their message (De Vries and Hobolt, Reference De Vries and Hobolt2020). While the status as a challenger party is the reason for adopting an infiltration strategy, the message spread through civil society networks is inherently populist and anti-elitist. In practice, infiltration can vary: contacts with clubs and associations can be initiated, or existing membership positions can be exploited by party members or supporters without party membership.

2.2. Types of associations and spreading of non-compliance

We contend that both the supply of populist ideas and the nature of targeted associations are crucial for understanding compliance with government measures. When parties with anti-establishment messages target civic associations, their networks become conduits for spreading these views. Many associations serve as meaningful social networks built on longstanding relationships and mutual trust. Within closely connected networks, the high reputational costs of transgression limit opportunistic behavior, fostering greater trust and facilitating the transmission of ideas with minimal scrutiny.

However, not all clubs are equal. Putnam (Reference Putnam2000) famously distinguished between “bridging” and “bonding” social networks. Bridging networks bring together individuals from diverse backgrounds, while bonding networks reinforce existing social divisions. For example, a handball club represents a bridging association with few social or financial barriers to entry, whereas women's shelter associations and elitist male student fraternities exemplify bonding clubs with distinct social structures.

The fact that bridging associations bring together extensive local population makes them even more conducive to spreading norms. Because political views are not salient in bridging clubs, anti-elitist attitudes may be applied covertly without discussion. Instead, bonding clubs with a more homogeneous membership should reinforce members’ existing attitudes, either in line with mainstream or populist parties, as opposed to the spreading of new ideas. Ultimately, the answer to which type of associations will spread populist views will depend on whether specific associations are directly targeted by populist parties or instead aligned with other non-populist political forces.

In a political context with populist supply, (bridging) associations can accelerate the spread of anti-establishment views. Through infiltration, populist parties can covertly shift boundaries toward anti-establishment attitudes, entrenching resentment of the government, mainstream parties, intellectual elites, and the privileged rich (Mudde, Reference Mudde2007; Inglehart and Norris, Reference Inglehart and Norris2016).

Previous studies have confirmed the relationship between populism and distrust in scientific elites in the case of climate change denial (Huber et al., Reference Huber, Greussing and Eberl2021) and anti-vaccination movements (Kennedy, Reference Kennedy2019). Such distrust can also result in sub-optimal outcomes in emergencies like Covid-19 (Lasco and Curato, Reference Lasco and Curato2019; Ansell et al., Reference Ansell, Cansunar and Elkjaer2021). Our study focuses on compliance with Covid-19 lockdowns. We argue that mobility during lockdown is a potent behavioral indicator of the spread of anti-establishment attitudes (for individual-level correlates of compliance behavior see Mahdavian et al., Reference Mahdavian, Warren, Evensen and Bouder2022). If people are willing to break lockdown policies at the cost of risking their health, this is a much stronger indicator of anti-establishment attitudes than a survey response.

Taken together, in a political context with populist supply, we expect to see: (1) less compliance in places with more (bridging) clubs. If our proposed mechanism works, we should also find (2) evidence of strategic targeting of (bridging) clubs by radical populist forces and (3) a clear relationship between (bridging) associations and votes for populist parties.

3.1. The political context

In the years leading up to the 2020 Covid pandemic, the German context was conducive to the development of anti-establishment sentiments. Societal divisions and government dissatisfaction heightened following the financial crisis of 2008, the European debt crisis of the 2010s, and the migration crisis in 2015–2016. These events bred resentment toward elites, particularly the CDU and SPD. Consequently, segments of German society became more open to messages from parties that echoed their concerns. However, the government and major parties failed to address these grievances adequately. The CDU's leftward shift before 2017 and Angela Merkel's policies left some former CDU (and SPD) voters feeling neglected.

The Alternative für Deutschland (AfD) is one of the “populist radical right wave” challenger parties emerging in Europe (De Vries and Hobolt, Reference De Vries and Hobolt2020, p. 29–33). It was founded in 2013 as an explicit “alternative” to mainstream parties and has since moved to the right. The party increased the supply of anti-establishment rhetoric and aimed to normalize their views further and attract more voters.

3.2. Mobility restrictions and compliance as civic behavior

Germany in 2020 is a relevant setting for studying civic behavior. Despite uniform social-distancing measures throughout the country, compliance varied. After the first domestic case of SARS-CoV-2 on January 27, 2020 and the accelerating spread of the virus, politicians reacted to reduce infections. Despite Germany's federal system, the national and sub-national state (Länder) governments jointly imposed a “lockdown” on March 16, 2020. This included limits on gatherings in private and public spaces and closures of schools, workplaces, restaurants, etc.Footnote ¹ In addition, the government explicitly asked citizens to limit their mobility, emphasizing that the common good depended on individuals’ collaboration. In a televised statement on March 18, 2020, Chancellor Merkel said:

Lockdown measures were uniformly implemented across Germany, leading to a consistent decrease in mobility after week 11 (see Appendix section 7). These restrictions eased gradually between late April and early May 2020, with the government continuing to advise against unnecessary travel throughout the year. As infection rates rose in the fall, national and regional governments reintroduced restrictions. On November 2nd, a “lockdown light” was imposed, requiring the closure of non-essential shops, restaurants, and cultural venues, along with a call for reduced social interactions. A second, stricter lockdown was enforced on December 16th, imposing tighter limits on private gatherings and urging citizens to stay at home. Mobility restrictions and mask-wearing faced increasing criticism. Initially, AfD politicians advocated for stricter measures until April 2020, before later condemning containment policies. Nonetheless, our research indicates an immediate and sustained decline in compliance from March onward, consistent with our argument that anti-government attitudes predated the pandemic.

4.1. Measuring associations and compliance

We begin our analysis by examining the relationship between compliance and different types of associations. We focus on 2020, the first year of the pandemic. To measure compliance with lockdown policies, we rely on mobility data that the German Federal Statistical Office provides based on mobile phone data (see Table 2 in the Appendix). The data capture daily mobility changes compared to the 2019 baseline and measure all entries to and mobility within a jurisdiction. We aggregate the data as weekly means. The mobility data are available at the level of administrative counties (Kreise). The complete sample of 401 jurisdictions comprises rural counties (Landkreise) and urban counties (Kreisfreie Städte). Table 2 in the Appendix shows details on all data sources.

To explore the impact of associations, we utilized data on modern clubs and their density per 1000 inhabitants at the county level from 2008 (Franzen and Botzen, Reference Franzen and Botzen2011). This dataset encompasses all registered non-profit associations in 2008, totaling 553,394 associations across various categories such as nature/environment, culture, social welfare, sports, and more (Franzen and Botzen, Reference Franzen and Botzen2011). Although detailed membership data are unavailable, association density serves as a reliable proxy for membership (Satyanath et al., Reference Satyanath, Voigtländer and Voth2017: 491). We assume these data accurately reflect active club membership leading up to 2020, with club numbers remaining relatively stable over time.

Indeed, stability is evident in the number of sports clubs and members between 2005 and 2017, with minimal fluctuations (Breuer et al., Reference Breuer, Feiler and Rossi2020: 19). Overall association numbers saw slight changes from 2009 to 2019 (Schubert et al., Reference Schubert, Tahmaz and Krimmer2022: 3), which we interpret as negligible. We address concerns regarding club stability and potential bias in our analysis in more detail in the Appendix. While stable club and membership figures do not guarantee unchanged active membership since 2008, there is no systematic evidence suggesting otherwise.

We use principal component analysis to create additive indices to increase conceptual clarity. In section 3.2 of the Appendix, we show the presence of three latent variables corresponding to (1) sport, nature, and culture clubs; (2) social/welfare, political/economic, and special interests clubs; and (3) other free-time (leisure) clubs. This confirms the categorization of clubs in earlier studies on Germany (Franzen and Botzen, Reference Franzen and Botzen2011; Satyanath et al., Reference Satyanath, Voigtländer and Voth2017).

While bonding and bridging may not be perfectly distinguishable theoretically, on an empirical level the latent variables point in the direction of the classification proposed by Putnam (Reference Putnam2000). The first latent variable corresponding to sport, nature, and culture clubs broadly reflects bridging associations. The second latent variable describing social/welfare, political/economic, and interest clubs reflects bonding associations with a distinct socio-political identity. Finally, free-time clubs are their independent latent category. Free-time clubs include organized leisure activities unrelated to another category, such as model railroad, card games, or rabbit breeding. As a robustness check, we rerun the analysis with culture clubs included in the bonding rather than the bridging category in Table A.7: results are highly similar.

4.2. Estimation strategy

To investigate the relationship between bonding and bridging associations and social-distancing behavior, we employ a DiD strategy. We classify counties along the mean density of civic associations as high density and low density (see robustness checks—section 4.5 where we also use a continuous measure for density). We use this classification for three groups of associations: all associations, bridging, and bonding. In the first specification, we pool the weekly observations and compare mobility patterns (a) before and after lockdown and (b) in counties above and below the mean density of clubs. The DiD baseline equation of interest is:

(1)$$\matrix{\Delta Mobility_{i, t} & = \beta_0 + \beta_1Networks_{i} \times Lockdown_{t} + \beta_2Lockdown_{t} + \beta_3Networks_{i}\cr & \quad + {{\bf X}'_{i, t}\beta} + \alpha_i + \gamma_t + \delta_{m, t} + \epsilon_{i, t} }$$

where i defines the county, m the state (Land), and t the week. Mobility _i,t is the dependent variable and evaluates mobility changes in 2020 compared to the previous year. Networks takes the value of one if the county's association density is above the mean and zero otherwise. Networks _i × Lockdown _t is an indicator variable for the interaction between association density, Networks, and an indicator variable Lockdown for the period starting in week 11. Recommended mobility restrictions started in mid-March and continued throughout the year, so for all weeks after week 11, we coded the dummy Lockdown_t as 1. The specification includes county-fixed effects (α _i) for time-invariant heterogeneity at the county level, week-fixed effects (γ _t) for common shocks across counties, and state-week-fixed effects (δ _m,t) to capture structural differences in reaction to lockdown across the different states m. $\epsilon _{i, t}$ is the error term of the model.

Counties with a high density in associations may differ from those with lower density on various characteristics. One could be the degree of urbanization or the urban/rural divide. We run additional models in which we use the vector X′_i,t to define a set of control variables that might impact mobility (Barrios et al., Reference Barrios, Benmelech, Hochberg, Sapienza and Zingales2021). These include turnout as a standard measure for civicness, log total population, GDP per capita, population above the age of 60, gender ratio, percentage of people with a college degree, employment shares in sectors that hardly allow for work-from-home such as hospitality, transport, and manufacturing, a dummy for East Germany, and lagged number of Covid cases, a measure that featured prominently in German media coverage as “case incidence.” For urbanization, we follow Haffert (Reference Haffert2021) and include a variety of indices: population density, percentage of people below the age of 35, percentage of students, and university-to-population ratio. To further alleviate the potential confounding effects of urbanization, we split our sample into two, based on (1) the distinction between free cities (“Kreisfreie Stadt” or “Stadtkreis”) and rural counties (“Kreis” or “Landkreis”) and (2) the presence of universities (either above or below the mean university–population ratio). Results are displayed in Table A.6 of the Appendix.

We aim to estimate the coefficient β ₁ for Networks _i × Lockdown _t. This parameter captures different trends in mobility between counties above and below the mean density of associations after the introduction of the lockdown. Lockdown generally decreased mobility across German territory (see Figure 1). Therefore, positive values of the parameter β ₁ indicate a less severe drop in mobility for counties with denser clubs (all/bridging/bonding) compared with counties where they are relatively scarcer.

Figure 1. DiD and event study for bridging associations: (A) weeks 5–16, (B) weeks 5–16, (C) weeks 1–53, and (D) weeks 1–53.

In the second specification, we consider an event study. The specification takes the following form:

(2)$$\Delta Mobility_{i, t} = \beta_0 + \sum_{\tau \in ( -5, 16) , \tau = -1} \beta_{\tau} Networks_{i} \times {\opf 1} ( t + \tau) + {\bf X'_{i, t}\beta} + \alpha_i + \gamma_t + \delta_{m, t} + \epsilon_{i, t}$$

The event study setup is similar to Equation 1. Still, instead of pooling different weeks in a pre-lockdown versus a post-lockdown period, we interact associations (bridging and bonding above-the-mean value) with an indicator variable for each week (t + τ). We label t the week when lockdown is introduced: week 11. The pre-lockdown period covers the weeks from t − 10 to t − 1, while the post-lockdown period covers the weeks from t to t + 42. As the reference week, we consider the last pre-lockdown week t − 1, week 10 of 2020.

4.3. DiD assumptions

The DiD estimation strategy is based on the parallel trends assumption: the trends in mobility for counties with high and low density of associations must be parallel in the pre-lockdown period. Figure 1 provides a graphical test of this assumption. Panels A and C show the evolution of mobility over time in counties with a higher and lower density of bridging associations. The second test in panels B and D is based on the event study setup. Specifically, we regress mobility at the county level on interactions between Networks and indicator variables for each week. The event study specification only includes county-, week-, and state-fixed effects. Panels A and C report results for the weeks 5–16, around the start of the lockdown. One possible concern is that this pre-lockdown period may not be able to capture secular trends. In panels C and D of Figure 1, we replicate the same analysis focusing on the broader pre-lockdown period starting in week 1 of 2020.

The graphical analysis in Figure 1 suggests that the parallel trends assumption holds in our setting. In panel A, the dotted lines for the linear interpolation in the pre- and post-lockdown periods indicate that the trends for counties above and below the mean bridging clubs are parallel and increasing before the crisis. The event study setup in panel B corroborates the presence of parallel trends.

Concerning the last pre-lockdown week (week 10, t − 1), almost all pre-lockdown coefficients are statistically insignificant and close to 0. The only exception is week 6. These patterns also hold when including trends for additional individual control variables and all control variables simultaneously, as indicated in Figure A.10 of the Appendix.

Panels C and D also confirm the existence of parallel trends when going back to the first week of 2020. We also include a variety of controls in our main specifications as indicated in Table A.4 and interactions between these controls and the time units as in Figure A.10 in the Appendix.

4.5. Robustness and alternative explanations

Our results are robust to sample selection: excluding East Germany or the areas in the north bordering the sea (the results might be driven by changed vacation behavior). The results are presented in Figure A.9 of the Appendix. The conclusions hold when controlling for additional interactions between independent variables and weeks (see section 4.4 of the Appendix) to deal with possible trends in other relevant factors. Finally, the results are not sensitive to the classification of bridging club density; in Table A.11 in the Appendix, we also use mean, median, upper quartile, and continuous measures of associational density.

We also examined whether the results are driven by more people needing to go to work during the lockdown in regions with denser bridging clubs. As already indicated, in Table 1 and Table A.4 in the Appendix, we control for the employment share in services (hospitality and transport), and manufacturing. We also control for local GDP per capita and the percentage of people with a college degree. Neither the occupation variables nor local income is statistically significant, while counties with more college-educated people experienced significantly more mobility reduction.

The results in section 4.7 of the Appendix indicate that higher mobility during the lockdown in areas with denser bridging associations is spread across various places and activities: retail and recreation, transit, and workplaces. Reassuringly, people in areas with denser bridging clubs are less likely to stay home, as indicated by their lower mobility in residential areas. No significant differences exist in mobility to pharmacies and grocery stores (basic necessities) or parks. The null result for parks renders little plausibility to the mechanical effect of bridging clubs; the mobility patterns are not driven by sports club members simply practicing sports in parks.

5.1. Qualitative accounts of infiltration and spreading

After its foundation in 2013, the AfD quickly acknowledged the strategic value of civil organizations for expanding its vote share. According to a leaked strategy paper by the federal board of the AfD, cited in the German press (Rosbach, Reference Rosbach2020), AfD must improve connection between the party and “interest groups, citizens associations, and medium-sized companies” (emphasis added) to “anchor the AfD in the middle of society” (Alternative für Deutschland, 2016: 15). They should keep track of “which members of the AfD work in societies and associations voluntarily. On this basis, a strategy will be developed” (p. 16). Infiltration of associations and clubs should be seen as an “opportunity for the AfD […] to approach their target groups like a ‘fish in water.”’ AfD members “must also be encouraged to participate in clubs or to use their club membership discreetly but consciously for the AfD” (emphasis added) (p. 16). The AfD working “discreetly” and “like a fish in water” suggests that they focus on spreading populist views compatible with the demand side and in a covert way in line with their challenger status.

Besides evidence of infiltration on the supply side (AfD), there are further indications of infiltration from the clubs and associations themselves. Many sports and nature clubs and their umbrella organizations throughout Germany have recognized the problem and developed various forms of training, information, and legal advice against radicalization. A brochure from a regional organization for sports clubs draws attention to infiltration:

To further understand infiltration strategies and their success, we conducted semi-structured interviews with eight representatives of umbrella or peak associations of nature and sports clubs from across Germany. Section 5 of the Appendix details the interview methodology and relevant quotes supporting the proposed mechanisms: infiltration and spreading. While we expect representatives of umbrella organizations to be more open to interviews than representatives of single (infiltrated) clubs, we acknowledge that our interview material is limited insofar as it does not identify a specific case or club where infiltration occurred.

Overall, the interviewees acknowledged the presence of populist attitudes and anti-democratic influences in their associations. They come out in the form of racist comments during sports matches or outdoor excursions (racist language is marked as anti-establishment in the German discourse) or in the form of association members promoting extremist or populist views. For instance, one interviewee recalled the case of a youth coach who tried to “spread his ideology regarding ‘no masks and vaccines’ and conspiracy theories” in the sports club's chat group (Representative of Netzwerk Sport & Politik). The interviews reveal that these influences are not limited to AfD members but that other, even more extremist far-right parties and organizations also attempt infiltration.

To combat the spread of such attitudes, most associations developed programs to inform instructors in local clubs about the presence of “infiltrators” to train them to react effectively. Despite these efforts, one sports club representative argued that much remains to be done because “for a large part [of the sports clubs], the problem seems peripheral, and they do not see the urgency.” (Representative of Landessportverband für das Saarland).

One point to note is that AfD's strategy has not always stayed the same. In fact, it changed significantly between 2013 and 2017. Pytlas and Biehler (Reference Pytlas and Biehler2023) discuss how intra-party competition influenced nativist party radicalization. The extremist grouping Der Flügel gained momentum within AfD by portraying itself as “the true party within the party,” moving the AfD further to the right (Pytlas and Biehler, Reference Pytlas and Biehler2023: 6) and permeating mainstream discourse.

Irrespective of the dominating faction with the AfD, the AfD's declared strategy of targeting such associations is unsurprising. The personal and informal relations cultivated in bridging clubs facilitate the spread of norms and attitudes. German sports clubs are an excellent example of horizontally organized clubs that stress equality and community (Putnam, Reference Putnam2000); they actively nurture bonds among their members. According to a report for the National Sports Association (Breuer et al., Reference Breuer, Feiler and Rossi2020), German sports clubs “attach importance to community” (p. 25) and “offer an important space for social exchange” (p. 9), with 93 percent of clubs organizing celebrations and social events.

We argue that these continuous interactions, based on a shared sense of belonging, foster the spread of attitudes and ideology. Although interviewees highlighted that club members are “a mirror of society” and that sports club members may have more right-wing or populist attitudes ex-ante, they also described club structures conducive to spreading such views. One interviewee compared the social interactions in clubs to a family celebration where an aunt's racist comments are met with silence: he argued that many sports club members have known and respected each other for decades and, therefore, avoid discussing controversial topics. The interviewee acknowledged the “danger” that “[extremist] attitudes can gain ground within the club” if no action is taken (Representative of Landessportbund Niedersachsen).

A structural factor that hampers clubs’ decisive action is a legal requirement for clubs to remain neutral politically to keep their non-profit status and related tax benefits (see the brochure by the State Sports Association of Thuringia, 2015). Although this clause does not exclude action against radical behavior, it is often misread. Several interviewees referred to this “misinterpreted neutrality” as potentially conducive to the spread of anti-establishment views. The apolitical appearance of bridging clubs and their focus on the community can prevent dissent with in-group opinion leaders and spread anti-establishment views. As a result of infiltration and spreading, populist and radical right attitudes permeate sports, hunting, nature, and other (bridging) clubs (Rosbach, Reference Rosbach2020).

5.2. Individual-level evidence on the spread of attitudes

To bolster our proposed mechanism and distinguish between socialization within clubs and self-selection, we investigate norms regarding non-compliance at the individual level. We utilize nationally representative data from the European Bank for Reconstruction and Development's Life in Transition Survey. Respondents are categorized as active or passive members of a sport club or non-members. Active members engage in club activities, while passive members are nominal members. To gauge non-compliance with government regulations, we employ a scale where respondents rate their agreement with the statement “People should obey the law without exception” from 1 to 10.

While club members may differ from non-members, we assume no significant distinctions between passive and active members apart from exposure to fellow club members. If socialization within clubs is influential, active members, exposed to club dynamics, should exhibit more lenient attitudes toward rule adherence compared to passive members. Conversely, if self-selection prevails, differences between passive members of bridging clubs and non-members should emerge, with passive members showing a higher tendency to disregard rules. To address potential differences in attitudes toward obeying the law, we control for age, income, education, and state-fixed effects. Control variables’ results are omitted to streamline the presentation.

Results in Figure 2 support both hypotheses and add a temporal dimension. While in 2010, few differences existed between sports club members and non-members, disparities became more pronounced in 2016, aligning with our expectations. Sports club members, both active and passive, were notably more inclined to justify breaking the law. Furthermore, the data support the socialization hypothesis: in 2016, active sports club members were significantly more prone to endorsing law-breaking compared to passive members. This suggests that active participation in club activities may shape attitudes, possibly through exposure to club dynamics. Additionally, some evidence backs the self-selection hypothesis; in 2016, passive members were more inclined than non-members to justify breaking the law.

Figure 2. Sports club membership and agreement with breaking the law. Notes: Standardized effect of sports club membership on agreement with the question that “There are times when people have good reasons to break the law.” Control variables include age, income, education, and state-fixed effects.

Overall, the individual-level results support our proposed mechanism's socialization effect but cannot rule out self-selection. Bridging clubs act as meeting points for people who already question government rules and places where government questioning becomes even more widespread, which is in line with the interview results. The different findings for 2010 and 2016 support our argument that the decisive infiltration and spreading of anti-establishment attitudes occurred between the foundation of the AfD in 2013 and the onset of the Covid pandemic in 2020.

5.3. County-level evidence on the spread of political views

We explore another implication of our theory to substantiate the proposed mechanism. Suppose anti-establishment views were diffused through bridging associations. In that case, we should not only see that people in counties with more associations complied less with the lockdown but also that they voted more for the main populist party, AfD. Here, we investigate the relationship between the density of bridging associations and votes for AfD in the national elections of 2013 (right after the AfD was founded) and 2017 and the change between 2013 and 2017 in a bivariate setting.

Panel A in Figure 3 confirms that counties with denser bridging clubs saw sharper rises in AfD vote share. Panels B and C reveal a nearly flat relationship between bridging club density and AfD vote share in 2013, contrasting with a much steeper correlation in 2017. This suggests a significant diffusion role for bridging clubs preceding the 2017 election, aligning with the AfD's infiltration strategy. However, we acknowledge the possibility of the AfD's increasing radicalization and changes in messaging influencing these trends. Encouragingly, disaggregating different bridging clubs (sport, culture, and nature) in panels D, E, and F shows similar effects on AfD vote increases. In contrast, the negative relationship in panel G implies that bonding clubs may not spread these political views and could even protect communities from them.

Figure 3. Associations and AfD infiltration: (A) bridging and change in AfD votes, (B) bridging and AfD votes, 2017, (C) bridging and AfD votes, 2013, (D) sport clubs, (E) culture clubs, (F) nature clubs, and (G) bonding clubs.

We also investigate the change in votes for AfD in the DiD framework. Figure A.14 in the Appendix shows the interaction between AfD (continuous measure) and indicator variables for each week in 2020 over mobility in an event-study format. The positive results suggest that bridging associations and the change in votes for AfD are almost interchangeable, indicating a close link between bridging clubs, change in AfD votes, and compliance.

To further illustrate the relative importance of the change in votes for AfD for non-compliance with lockdown, we perform a mediation analysis, following the procedure proposed by Tingley et al. (Reference Tingley, Yamamoto, Hirose, Keele and Imai2014). The results presented in section 6 of the Appendix indicate that between 17 and 25 percent of the total effect of bridging clubs on mobility is mediated by the change in AfD votes.