SOCIAL MOVEMENTS AND ELECTORAL POLITICS

The way social movements affect their political environments has received considerable scholarly attention (e.g., Amenta and Caren Reference Amenta, Caren, Snow, Soule and Kriesi2004; Bosi, Giugni, and Uba Reference Bosi, Giugni and Uba2016; Gamson Reference Gamson1975; Kitschelt Reference Kitschelt1986; Kriesi et al. Reference Kriesi, Koopmans, Duyvendak and Giugni1992) but rarely through the examination of electoral outcomes. Movements have been known to affect polities (Goldstone Reference Goldstone2003; Yashar Reference Yashar1998), policies (Andrews Reference Andrews2004; Fording Reference Fording1997; Weldon Reference Weldon2011), and people (Blee Reference Blee, Bosi, Giugni and Uba2016; Giugni and Grasso Reference Giugni, Grasso, Bosi, Giugni and Uba2016), “yet there is little research on movement influence over elections and the political influence gained through such electoral support” (Amenta et al. Reference Amenta, Caren, Chiarello and Su2010, 297). This is partly because social movements are known to affect outcomes through other institutional channels that are only indirectly related to the electoral arena. This is especially so in the early stages of the policy process (Baumgartner and Mahoney Reference Baumgartner, Mahoney, Meyer, Jenness and Ingram2005; Soule and King Reference Soule and King2006), when movements help set the agenda to which governments and legislatures respond (Hutter and Vliegenthart Reference Hutter and Vliegenthart2018; Walgrave and Vliegenthart Reference Walgrave and Vliegenthart2012; Wouters, Sevenans, and Vliegenthart Reference Wouters, Sevenans and Vliegenthart2021).

The net effect of movements on elections has mostly been studied indirectly, at the party-systemic level, through the analysis of their impact on political parties. In the US, movements have been known to affect electoral outcomes by grounding or pushing political parties to vote-winning or vote-losing programmatic positions (McAdam and Kloos Reference McAdam and Kloos2014; McAdam and Tarrow Reference McAdam, Tarrow, van Stekelenburg, Roggeband and Klandermans2013; Schlozman Reference Schlozman2015). The effect of the antiwar movement on the political revival of the Democratic Party (Heaney and Rojas Reference Heaney and Rojas2015); the way the Tea Party revitalized Republican conservatism (Williamson, Skocpol, and Coggin Reference Williamson, Skocpol and Coggin2011); and the impact of the Women’s Marches (Andrews, Caren, and Browne Reference Andrews, Caren and Browne2018) and Black Lives Matter protests on the anti-Trump opposition (Meyer and Tarrow Reference Meyer and Tarrow2018), have compelled students of American politics to look for systematic patterns of interaction between movements and parties. In different electoral contexts, movements brought about electoral change by transforming themselves to political parties (Kitschelt Reference Kitschelt1986; Peterson Reference Peterson, Bosi, Giugni and Uba2016). The economic crisis gave rise to new political contestants (Hutter, Kriesi, and Lorenzini Reference Hutter, Kriesi, Lorenzini, Snow, Soule, Kriesi and McCammon2018), more willing than their competitors to intermediate interests by resorting to the streets. In some electoral settings, these political contestants became hybrids of movements and parties, squeezing political support for mainstream party competitors. The scholarly interest on “movement parties” (Kitschelt Reference Kitschelt, Katz and Crotty2006) on the left (Della Porta et al. Reference Della Porta, Fernández, Kouki and Mosca2017) and the right (Castelli Gattinara and Pirro Reference Castelli Gattinara and Andrea2019) of the political spectrum has helped narrow the gap in the study of how social movements and political parties interact.

The examination of direct links between protests and elections has been less frequent and has mostly dealt with earlier periods in American politics. Focusing primarily on the local level, a number of studies have sought to trace the effects of social protests on electoral outcomes. For example, the civil rights mobilization in the 1960s and 1970s in Mississippi was shown to have changed Black electoral participation and, over time, local office holding (Andrews Reference Andrews1997). In Southern counties, Ku Klux Klan presence and mobilization had an enduring and positive electoral impact on Republican voting (McVeigh, Cunningham, and Farrell Reference McVeigh, Cunningham and Farrell2014). Evidence of even more direct effects of local protests on elections has come from the analysis of how protest events have affected House results: it has been shown that, depending on the issue, a composite index of protest salience can potentially remove as much as 6 percent from a party’s vote share and reward the other party with 2 percent (Gillion and Soule Reference Gillion and Soule2018). Black protests between 1960 and 1972 have also been shown to have a direct effect on voter results: when nonviolent, they increased Democratic vote share by 1.6–2.5 percent, and when violent, they shifted voter support toward the Republicans (Wasow Reference Wasow2020). Studies of the electoral effects of more recent protest waves are rarer, but the evidence points to the same direction. For example, the size of Tea Party protests has been linked to an increase in Republican vote share (Madestam et al. Reference Madestam, Shoag, Veuger and Yanagizawa-Drott2013).

Beyond the American context, analyses focusing on the direct effects of social protest on electoral outcomes are even less frequent. One exception is a cross-national analysis of protest events in 30 European countries, which shows how protests driven by worsening economic conditions tend to penalize incumbents and mainstream parties, by accentuating economic grievances (Bremer, Hutter, and Kriesi Reference Bremer, Hutter and Kriesi2020). A common thread in studies of European and American contexts is that, depending on the issue area, social protest tends to dampen or boost electoral support for political parties. Although social movements have mostly been left out of analyses of electoral change, a nascent body of literature shows the growing need to link social protest with electoral outcomes (Barrie Reference Barrie2021).

HOW SOCIAL MOVEMENTS AFFECT ELECTORAL POLITICS

Given the scarcity of evidence on the direct effects of social movements on electoral politics, it is not surprising that the exact dynamics linking social protest to voting outcomes remain largely unexplored. There is a tendency to assume that protests affect voting behavior by sending informational cues to voters, but the informational link between protests and voting is primarily drawn from the various characteristics of social protest. Since “not all protests are created equal” (Wouters Reference Wouters2019, 406), the main idea here is that, by varying the main attributes of protest, one can identify features of protests that have a distinct impact on the informational signals sent to voters and, hence, electoral outcomes. The issues over which social movements mobilize protesters are a key feature of any protest. Protests over “liberal” issues have been shown to have a different effect on Democratic voter share than protests over “conservative” issues. The main assumption behind the distinction between liberal- and conservative-leaning protests is that they resonate with voters in a different way (Gillion and Soule Reference Gillion and Soule2018). Other characteristics that can plausibly affect electoral outcomes are numeric strength (Madestam et al. Reference Madestam, Shoag, Veuger and Yanagizawa-Drott2013), unison, diversity, worthiness, and commitment of protesters (Tilly Reference Tilly2004). Once manipulated experimentally, these characteristics are shown to affect how politicians (Wouters and Walgrave Reference Wouters and Walgrave2017) and the public (Wouters Reference Wouters2019) view protests.

Protest tactics have gained growing prominence in efforts to link protest with political outcomes. Taking cues from studies of strikes as well as from analyses of urban riots of the 1960s in the US, earlier scholarship probed the effectiveness of disruptive and violent protests (Giugni Reference Giugni1998). The disruptive nature of some protests has been shown to positively affect policy (Fording Reference Fording1997) and voting (McAdam and Su Reference McAdam and Su2002). For example, the 1992 Los Angeles riots helped shift support in favor of providing public goods to racial minorities. Violent protest increased local support for more spending in public schools, even when controlling for a number of other variables (Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019). Not all the evidence regarding the use of violence in protests points to the same direction, though (Huff and Kruszewska Reference Huff and Kruszewska2016). An elaborate study of Black-led protests between 1960 and 1972 shows that violent protests had a negative effect on the Democratic presidential vote. The key mechanism, here, was the way the media got to frame violent and nonviolent protests and, hence, the way public opinion got to perceive them (Wasow Reference Wasow2020). Individual-level data point to the same direction: extreme protest actions tend to reduce emotional connection, social identification, and overall support for a movement (Feinberg, Willer, and Kovacheff Reference Feinberg, Willer and Kovacheff2020). A study of party identification after the Capitol Insurrection yields similar findings (Eady, Hjorth, and Dinesen Reference Eady, Hjorth and Dinesen2022).

PROTESTING AGAINST THE FAR RIGHT

Social protest against far right actors provides a new opportunity to examine how the interaction of social movements and political parties affects electoral politics. Across different European settings, there are, by now, many examples of social movements mobilizing against the far right. In recent years, dozens of thousands of Germans have taken to the streets to protest against the electorally ascendant Alternative for Germany, holding “Never again” and “Nazis out” banners. The brief attempt of the governing Christian Democrats to cooperate in one region with the far right party sparked protests in major cities, with protesters chanting antifascist slogans. Historical alarm bells have also gone off in Germany against the street-level mobilization of the electorally insignificant National Democratic Party of Germany. Its events have, at times, pushed to the streets thousands of anti-far right protesters. Even gatherings of a few dozen NPD members have attracted sizable crowds across various German localities. Although Germany is a particularly important case for protests against the far right, such protests constitute a broader contemporary phenomenon. For example, the 2019 Sardines movement that sprung up in Bologna and spread throughout Italy managed to gather large crowds to counter the politics of the Lega Nord. The initial aim of the founders was to gather as many people as possible in major public squares to protest the anti-immigrant appeals of the Lega (Hamdaoui Reference Hamdaoui2022).

The visible societal mobilization against the far right has been largely overlooked in the sizable literature on the electoral ascendance of the far right (e.g., Golder Reference Golder2016). This literature has treated these organizations solely as parties, downplaying the movement-type elements in their repertoire. Recent scholarly emphasis on “movement parties” (Kitschelt Reference Kitschelt, Katz and Crotty2006; Della Porta et al. Reference Della Porta, Fernández, Kouki and Mosca2017; Pirro and Castelli Gattinara Reference Pirro and Castelli Gattinara2018; Borbáth and Hutter Reference Borbáth and Hutter2021; Castelli Gattinara, Froio, and Pirro Reference Castelli Gattinara, Froio and Pirro2022) has helped draw attention to this movement-type, street-level mobilization of the far right and has generated a need to systematically examine the responses against it. Such responses are not only political and institutional (Art Reference Art2007; Capoccia Reference Capoccia2013; Müller Reference Müller2016) but also societal (Müller Reference Müller2022). Protests against the far right have started gaining scholarly attention (Castelli Gattinara, Froio, and Pirro Reference Castelli Gattinara, Froio and Pirro2022; Vüllers and Hellmeier Reference Vüllers and Hellmeier2022; Zeller Reference Zeller2021; Reference Zeller2022), but analyses of how these protests affect electoral outcomes remain rare (Colombo et al. Reference Colombo, Ferrara, Vassou, Bernardi and Dinas2021).

PROTEST AGAINST THE FAR RIGHT IN GREECE

To analyze the electoral effects of anti-far right protests, we analyze subnational evidence (Pepinsky Reference Pepinsky2019), by leveraging original data from Greece. Like most European countries, and even before the onset of the economic crisis of the 2010s, Greece witnessed the electoral surge and parliamentary breakthrough of the far right. But unlike most other countries, the country experienced the ascendance of a clearly antidemocratic variant of the far right (Pappas Reference Pappas2016). Greek protests against the far right primarily targeted the GD. Founded in the early 1980s as a National Socialist magazine, GD became a political party in the 1990s and contested all national elections between 2009 and 2019. Mostly known as a violent neo-Nazi gang, the party gained notoriety in the 1990s and 2000s for its assaults on migrants and leftists. In the 2009 parliamentary election, the party received 0.3%. In the 2010s, its extreme ideas and street-level mobilization in migrant-rich communities started gaining electoral traction, and in the May and June 2012 national parliamentary elections, GD managed to enter the national legislature, winning about 7 percent of the vote. By the January and September 2015 parliamentary elections and amid unprecedented refugee inflows (Dinas et al. Reference Dinas, Matakos, Xefteris and Hangartner2019), GD became the third biggest political party, sustaining its share of the national vote at around 7%, despite the criminal prosecution of its top leadership. In the 2019 parliamentary election, GD narrowly missed the 3% electoral threshold and failed to win parliamentary representation. In 2020, its leader, former members of parliament, and dozens of cadres were convicted for running or being members of a criminal organization and many have been serving substantial prison sentences.

The electoral traction of the GD in the early 2010s along with its growing street-level presence across dozens of Greek localities set the stage for unprecedented societal mobilization against it. To confront the ideological and spatial threat (Andrews and Seguin Reference Andrews and Seguin2015) posed by the electoral ascendance of the GD, the self-proclaimed Greek antifascist movement brought together hundreds of organizations, groups, and networks, which ranged from trade unions, human rights organizations, and migrant associations to numerous radical leftist and autonomous groupuscules. A few of these were national organizations, but the vast majority were local groups. Antifascist groups organized thousands of events like protest marches, public speeches, and neighborhood assemblies. Across many Greek municipalities, protesters held banners in the streets, handed out leaflets demanding the imprisonment of “the neo-Nazi gang,” and organized festivals and concerts to galvanize support against the GD. Some events involved thousands, and others, only a handful of protesters. The vast majority of these events—like the anti-racist festivals organized on the International Day for the Elimination of Racial Discrimination every March, or demonstrations held every September to commemorate the 2013 murder of an antifascist musician—contested far right claims without involving direct encounters with the GD.

A significant segment of the antifascist movement tactically chose to directly challenge the organizational expansion of the party in various localities, generating the tango pattern described earlier. At the time when the GD was busy setting up local branches across Greece, organizing food rationing in public squares, and holding military-style marches and torch-bearing gatherings at major monuments, antifascist groups staged hundreds of highly localized counter-protests. Anti-far right protests were particularly prevalent in urban, migrant-rich localities where the GD sought to acquire grassroots presence and challenge the local dominance of leftist groups. The sequence of the tango pattern was that the GD would announce an open event in a particular setting and the antifascists would make a call for a counter-event. For example, every June the GD would hold an annual gathering at the Alexander the Great Statue in Thessaloniki, to celebrate his birthday. Antifascists would typically organize a counter-protest at the same time, in a nearby location.

DATA AND METHODS

To examine the electoral effects of protests against the far right, this study relies on the analysis of protest events (Fisher et al. Reference Fisher, Andrews, Caren, Chenoweth, Heaney, Leung and Perkins2019; Hutter Reference Hutter and Porta2014b; Koopmans and Rucht Reference Koopmans and Rucht2002). The analysis here relies on 4,479 protest events organized against the Greek far right between the parliamentary elections held on October 4, 2009, and the parliamentary elections held on July 7, 2019 (Ellinas and Lamprianou Reference Ellinas and Lamprianou2023). To examine how earlier anti-far right protests might affect electoral outcomes, the dataset also includes 266 protest events against the Greek far right organized between June 2006 and the elections held on October 4, 2009, bringing the total sample size to 4,745 protest events (Figure 1). During the period between October 2009 and July 2019, the GD contested all six national parliamentary elections. The dataset includes all events posted by users on the Athens Indymedia in the category “anti-fascism, anti-racism and anti-nationalism.” Founded in 2001, Athens Indymedia is part of the global Indymedia network, set up after the late 1990s, to provide grassroots coverage of anti-globalization protests (Fenton Reference Fenton2020).

Figure 1.

The Distribution of Protest Events per Year

Notes: The first bar aggregates 2006 and 2007 events for purposes of visibility. The 2019 data only represent protest events for half year (up to the elections of July 2019).

As an aggregator of information from multiple sources, Indymedia is a unique data source because it provides a comprehensive record of events organized across Greece, like demonstrations, marches, pickets, canvassing, assemblies, festivals, and concerts. The events were identified, assessed, and added to the dataset after meticulous piloting and coding by two coders that lasted several years (Appendix I in the Supplementary Material). The ideological slant of the medium and its reliance on grassroots rather than professional journalistic coverage of protests imposes a notable limitation on the compilation of anti-far right events. The multiplicity of sources reporting these events limits the availability, veracity, and consistency of measures of protest size, a key component for analyzing protest outcomes (Biggs Reference Biggs2018).

Despite this limitation, this unique data source helps deal with other known biases of event catalogs. Unlike major national newspapers, which are noted to have a selection bias in their coverage of protest events (Earl et al. Reference Earl, Martin, McCarthy and Soule2004; Walgrave and Vliegenthart Reference Walgrave and Vliegenthart2012), Indymedia posts offer coverage for a much wider range of events. Moreover, Indymedia provides coverage of micro-events in areas away from central or media locations, thereby limiting some of the methodological biases evident in event catalogs compiled from mainstream newspapers (Fisher et al. Reference Fisher, Andrews, Caren, Chenoweth, Heaney, Leung and Perkins2019). Because of the wide geographical range of its coverage, Indymedia allows the compilation of events held simultaneously in multiple locations, avoiding the propensity to treat multi-sited protests as a single record in event catalogs (Biggs Reference Biggs2018, 366). Furthermore, given the emphasis of this study on how protest affects electoral outcomes, Indymedia helps moderate a plausible endogeneity problem—the drop in national news coverage of protest due to the diversion of attention to the national election and limited availability of “news holes” (Oliver and Maney Reference Oliver and Maney2000). Finally, Indymedia provides grassroots coverage of highly localized micro-events that are usually off the radar of major national media. These events are usually posted by various anti-far right collectivities or by individual Indymedia users. The aggregation of these micro-events permits an exploration of electoral outcomes that does not rely on the agenda-setting role of high-profile, media-covered protests (Hutter and Vliegenthart Reference Hutter and Vliegenthart2018; Walgrave and Vliegenthart Reference Walgrave and Vliegenthart2012).

The six parliamentary elections between October 2009 and July 2019 form five intermediate periods of protest (Figure 2). For the period leading to the elections of 2012, protest activity remained at relatively low levels, but there was a significant increase in the last few months before the elections. After peaking in 2013, anti-far right protests lost some steam in the years leading to the January and September 2015 elections. However, there was a gradual but steady increase in the number of events leading to the 2019 elections. Violence was a rather infrequent phenomenon in anti-far right protest events. Out of 4,745 events, around 6% (N = 285) had some degree of violence from anti-far right organizers or protesters.

Figure 2.

The Six Parliamentary Elections Contested by the GD, Define Five Intermediate Periods of Protest

Notes: The solid line graphs between elections illustrate the frequency of protest events per month. The pie chart in the right top inset shows the proportion of municipalities with no protests, moderate protest, and frequent protest events, for the whole period from October 2009 to July 2019.

Time-wise, the length of the in-between election periods is very uneven and ranges from 42 to 1386 days. The shortest period of 42 days corresponds to the repeat election of June 2012, due to the undecisive outcome of the May 2012 election. In 2015 there was a repeat election in September 2015, only eight months after the previous election.

The main independent variable (IV) of the study is the count of protest events per municipality, per period. As the pie chart of the right top inset of Figure 2 shows, almost half of the municipalities had no protest events for the whole period from October 2009 to July 2019. The distribution of counts of events per municipality, per period, is extremely skewed, with an excessive number of zeroes and a number of extreme outliers, so it is not appropriate to use in parametric models such as regressions (Appendix II in the Supplementary Material). For this reason, we recoded the protest events per period as a trichotomous IV where category “No protest events” represents municipalities with no events at all, category “Moderate protest” represents municipalities with at most one event per year, and category “Frequent protest” represents municipalities with more than one event per year (see Appendix III in the Supplementary Material for more details). As the pie charts of Figure 2 show, the majority of municipalities had no protest events in any given period, but the proportion of municipalities with moderate and frequent protest increased significantly for the longer periods. The exact distribution of the trichotomous variable per period is presented in Appendix III, Table A in the Supplementary Material.

The main dependent variable (DV) is the electoral results of GD, measured as the percentage of votes received by the GD per municipality. In total, there are 325 municipalities clustered in 74 regional units (see Appendix IV in the Supplementary Material). The correlations between the electoral results of the GD for adjacent elections range from 0.514 to 0.917 (see Appendix V in the Supplementary Material). As illustrated in Figure 2, the electoral results of October 2009 are used as the baseline measure for the elections of May 2012 (r = 0.514, p < 0.001). The correlation between the electoral results of May and June 2012 is very high (r = 0.917, p < 0.001), and the period between the two elections is extremely short (42 days). We hence use the electoral results of October 2009 as a baseline for the elections of June 2012. Subsequently, all previous electoral results are used as baseline for the next election.

At the first stage of our statistical modeling, linear regression models were used to explain the electoral results of the GD per municipality, for each national election beginning May 2012. The main IV is the trichotomous variable of anti-far right protest described above. For each of the models, 2011 census data are used to control for a number of theoretically significant variables. We used immigration as a percentage of the total population to control for the propensity of GD to perform better in municipalities with higher immigrant concentration (Arzheimer Reference Arzheimer2009; Hangartner et al. Reference Hangartner, Dinas, Marbach, Matakos and Xefteris2019). In accordance with studies measuring the electoral effects of protest events (Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019; Gillion and Soule Reference Gillion and Soule2018; Wasow Reference Wasow2020), we also controlled for population size. To avoid numerical problems in the estimation of the model, we use the natural logarithm rather than the absolute size of the population of municipalities which ranges from 766 to circa 664,000. We use the average age of the population for each municipality to control for the propensity of far right parties in general (e.g., Stockemer, Lentz, and Mayer Reference Stockemer, Lentz and Mayer2018) and the GD in particular to perform better among younger voters. We use the proportion of unemployed people per municipality to control for economic grievances (e.g., Arzheimer Reference Arzheimer2009; Sipma and Lubbers Reference Sipma and Lubbers2020) and a dichotomous variable as an indicator of whether GD has a registered branch in each municipality. To control for the possible spillover effects of geography (Hainmueller and Kern Reference Hainmueller and Kern2008), we model municipalities as nested in regional units using generalized linear mixed-effects models.

For all models, we investigated both the assumptions and the model fit (Gelman and Hill Reference Gelman and Hill2006). Finally, we investigated additional alternative models, by replicating our analysis using additional controls and other theoretically interesting variables to confirm the robustness of our findings.

At the second stage of the analysis, we investigated more refined models, to test the key protest dynamics of tango and timing, which we expect drive the electoral results of the GD. To investigate the tango effect, we counted the number of counter-mobilization events organized in each municipality as a response to a GD event during any given period of time in-between elections (see Appendix III, Table B in the Supplementary Material). A tango event is an event organized in the same municipality as a GD-organized event, explicitly as a reaction to that GD-organized event. As mentioned earlier, such events would typically include a counter-protest, -march or -gathering taking place at around the same time and place where the local GD branch would organize an event. The GD event is also recorded in our dataset, when an explicit reference to it is made in the Indymedia post mentioning the anti-far right event. In other words, when the anti-far right event is explicitly organized as a response to the far right event.

There were 59 tango events in 31 municipalities for the period between October 2009 and May 2012. For the period of May 2012 to June 2012, there were only seven tango events, one per municipality, and we hence excluded the June 2012 election from the analysis. There were 178 tango events in 77 municipalities during the period June 2012 to January 2015. For the period of January 2015 to September 2015, there were only 14 tango events in nine municipalities, so we excluded the elections of September 2015 from further statistical analysis. There were 85 tango events in 38 municipalities during the period September 2015 to July 2019. Due to the small number of tango events per municipality per period, we operationalized new dichotomous IV, which had the value of “1” or “0” depending on whether there were tango events organized in a municipality in a given period or not. For each of the three periods remaining in the analysis, the dichotomous IV was used as the main IV in linear regression models.

To investigate the expected effect of timing, we split each period in-between elections in two parts: The six months preceding an election is defined as the “proximate” part of the period, and the rest of the time going backwards to the previous election is the “distant” part. For example, for the elections of May 2012, the period between the previous election of October 4, 2009, to November 6, 2011, is the distant period and the six-month period between November 6, 2011, to the elections of May 6, 2012, is the proximate period. By default, the two shortest in-between election periods are too short for the six-month cutoff between proximate and distant protests and are hence excluded for the timing models, as was the case for the tango models.

For the distant and proximate parts of the three remaining periods, we counted the number of protest events (Appendix III, Table C in the Supplementary Material). For the period October 2009 to May 2012, there were 0.4 events per day for the distant period and 0.9 events per day for the proximate period. For the period June 2012 to January 2015, there were 1.9 events per day for the distant period and 0.7 events per day for the proximate period. For the period of September 2015 to July 2019, there were 1.5 events per day for the distant period and 1.8 events per day for the proximate period.

For each period, we created a new variable where we classified each municipality in one of four categories: (a) had no events in both distant and proximate parts of the period, (b) had at least one event in both proximate and distant parts of the period, (c) had at least one event but only in the distant part of the period, and (d) had at least one event but only in the proximate part of the period. For each of the three periods remaining in the analysis, the categorical IV was used as the main IV in linear regression models.

We considered the possibility that confounding effects might introduce bias in our results. To investigate the possibility of endogeneity between protest events and other theoretically significant explanatory variables, we used a two-stage (instrumental variable) model (Sovey and Green Reference Sovey and Green2011). Our instrumental variable of choice is the number of students studying at various university departments and programs in each municipality. It is a variable that takes into account the historically and empirically grounded expectation that “university campuses are frequently hotbeds for student mobilization” (Zhang and Zhao Reference Zhang, Zhao, Snow, Soule, Kriesi and McCammon2019, 101; see also McAdam Reference McAdam1986; Sewell Reference Sewell, Aminzade, Goldstone, McAdam, Perry, Sewell, Tarrow and Tilley2001). A fundamental requirement for such a two-stage model is that the instrumental variable must be independent of the DV. We consider this requirement to be satisfied because student numbers in various municipalities are independent of the electoral results of the GD and, more importantly, because university students usually vote in the municipality of their permanent residence, not that of their university department or program.

TANGO AND TIMING: THE DYNAMICS OF PROTEST AGAINST THE FAR RIGHT

Moving from our basic to more refined models, we first examine whether the electoral outcomes of the GD are affected by a tango pattern, an action–reaction relationship involving events organized by the GD and counter-protests organized in the same municipality in response to these GD events. We used the same mixed-effects models as in Table 1, the only difference being that we replaced the ordinal IV of protest events with the dichotomous version of the tango variable as explained earlier (also see Appendix III, Table B in the Supplementary Material for descriptive statistics). Figure 3 compares the frequency of tango events to the frequency of all protest events per period. It is interesting that the frequency of tango events drops dramatically after 2013, although the frequency of protest events rebounds before the 2019 elections.

Figure 3.

The Distribution of Tango Protest Events per Year

Notes: The first bar aggregates 2006 and 2007 events for purposes of visibility. The bars represent anti-far right events and the solid line represents tango events. The 2019 data only represent protest events for half year (up to the election of July 2019).

The results (Table 2) suggest that, for the elections of May 2012, municipalities with at least one tango event had much lower electoral outcomes for the GD compared to municipalities with no tango events. With a national yield of around 7% in the elections of May 2012, the impact of having tango events corresponds to a reduction of around 13% of the electoral power of GD (circa 0.9 percentage point out of 7). The coefficients of tango events were not significant for the January 2015 and July 2019 models, although they had the same sign as that of May 2012.

Table 2.

“Tango” Linear Models Explaining Electoral Gains of the GD (Regions as Random Effects; DV Is the Electoral Results; “Tango” Effects as a Dichotomous Variable)

Second, we examined whether the timing of protest events, more specifically, their temporal proximity to the parliamentary election, has an effect on the electoral outcome of the GD. We used the same mixed-effects models as in Table 1, the only difference being that we replaced the ordinal IV of protest events with the categorical variable of distant/proximate events and, as explained before, we dropped two elections with very short in-between periods. Some descriptive statistics are presented in Appendix III, Table C in the Supplementary Material. Figure 4 presents the relative frequency of distant and proximate events. For example, for the period October 2009 to May 2012, there were 0.9 events per day for the last six months before the May elections, but only 0.4 events per day for the previous part of the period.Footnote ³

Figure 4.

Timing of Protest Events per Period

The results (see Table 3) for May 2012 suggest that organizing at least one proximate protest event against the far right, compared to organizing no events at all, corresponds to a smaller electoral outcome for the GD, similar in magnitude to that identified by previous models which was around one-sixth of the 7% of the national vote. Municipalities with protest events in the distant period only, did not have a smaller electoral outcome compared to those which had no events at all. Municipalities which had both distant and proximate events also yielded significantly lower electoral outcomes for the GD. The results were partly statistically significant for the January 2015 model as the coefficient of having both distant and proximate events was statistically significant.

Table 3.

“Timing” Linear Models Explaining Electoral Gains of the GD (Regions as Random Effects; DV Is the Electoral Results; “Timing” Effects as a Categorical Variable with Four Categories)