Party Bans: Deterrence or Backlash? Evidence from the Basque Country

This paper investigates the electoral eﬀects of party bans, studying the case of the ban on Batasuna, the political wing of ETA. In an initial electoral term, in 2003, Batasuna was banned from contesting local elections in all Basque municipalities; in a second term, in 2007, it was banned only in a subset of them, and in 2011 it became legal again. Exploiting the ﬁnite and heterogeneous length of the ban across municipalities, I ﬁnd that a longer ban has a negative eﬀect on electoral support for the targeted party. This eﬀect is explained by the extent of the immediate loss in support in treated municipalities under the 2007 ban, observable in this instance because Batasuna called for a null vote. This pattern and further heterogeneous eﬀects are consistent with voters learning from the new electoral scenario rather than with a direct cost attributable to the party no longer having institutional representation. This suggests that party bans may be used by incumbents to reduce their challengers’ strength by triggering an electoral reshuﬄing, with voters learning about new parties and potentially switching allegiances.

with voters learning about new parties and potentially switching allegiances.
Keywords: Party bans; voting behavior; conflict resolution Many democratic countries restrict the participation of political parties whose behavior or viewpoints are deemed to go beyond what is tolerable. In recent decades, over 50 political parties have been banned across Europe, including parties in Spain, Germany, Turkey, the United Kingdom, and Bulgaria. In 2013, the far right extremist Golden Dawn came third in the 2013 Greek election, after EU officials unsuccessfully called for its ban. 1 In Africa, the number of bans on ethnic, religious, and regional parties has increased in sub-Saharan countries since the 1990s. 2 The variation in party ban practices across countries has led to a various studies that discuss the normative and legal arguments surrounding party bans. Capoccia (2013), Fox and Nolte (1995), and Issacharoff (2007) argue that there is an inevitable democratic dilemma in party bans, consisting of a tension between procedural and substantive conceptions of democracy: democratic elections may require, as a precondition to the right of participation, a commitment to the preservation of the democratic process, but at the same time, this limits the scope of democratic deliberation, calling into question the legitimacy of the political process. A common concern in these studies is striking the right balance between the danger of extremist parties gaining excessive power and implementing anti-democratic policies, and the danger that party bans might be used to insulate incumbents from electoral challenge. In this spirit, according to the Guidelines on the Prohibition and Dissolution of Political Parties and Analogous Measures issued by the European Commission for Democracy through Law (Council of Europe -Venice Commission, 1999), party bans are an exceptional measure that requires sufficient evidence that a party is a real threat to the constitutional order or citizens' fundamental rights and freedoms.
Discussions as to the desirability of party bans often include instrumental arguments based on the expected electoral effects of such bans. However, besides a short-run mechanical exclusion, it remains an open empirical question whether party bans increase or reduce support for the targeted parties. According to Capoccia (2005), declaring an extremist party illegal might deprive the extremist movement of organizational resources and social visibility, and prevent extremists from influencing policy in the short-run. However, party bans carry the risk of significant electoral costs. First, such measures might encounter strong principled opposition in public opinion because they contradict democratic principles too starkly; second, they might force the extremists to go underground and adopt other, possibly violent, tactics, which pose a different kind of problem to the incumbents. Bale (2007) compares party bans in Belgium, Spain, and Turkey, and concludes that they are not associated with an increased popularity of the targeted parties. Casal Bértoa and Bourne (2016) compare party bans in Germany, Spain, and Turkey, and conclude that party bans are associated with periods of electoral volatility. Minkenberg (2006) reports that the German fascist Sozialistische Reichspartei, which was banned in 1952 but which re-emerged in 1969, did not manage to clear the 5% electoral threshold, suggesting that the ban reduced party support. The Czech Republic banned the marginal, far-right Workers' Party in 2010, but a successor party later obtained a similar level of electoral support (Downs, 2012). 3 All these qualitative studies suggest that party bans are unlikely to lead to an electoral backlash; nevertheless, it is difficult to know whether these aggregate outcomes are indeed caused by party bans, or if instead party bans are simply more likely to occur in periods when the targeted parties are expected to be weak, or in contexts of political instability. This limits our ability to understand the reasons why party bans might be used in the first place as well as the reasons why party bans may have an effect on electoral outcomes.
This paper contributes to this literature by examining a case study that allows me to construct a counterfactual to disentangle the effect of the party ban from confounding effects. This case study is the ban on Batasuna, a leftistindependentist political party in the Spanish region of the Basque Country that was outlawed because of its connection with the Basque terrorist organization ETA. The ban on Batasuna had a finite and heterogeneous length across municipalities: in the first electoral term, in 2003, the party was banned in all municipalities, in the second, in 2007, it was banned only in about half of them, and in a third electoral term, in 2011, the ban on the party was lifted. Crucially, these two groups of municipalities had reacted very similarly to aggregate popularity shocks before the heterogeneous ban (i.e., they had been following parallel trends in electoral outcomes). This allows me to empirically examine the effects of party bans by comparing the outcomes of two groups of similar municipalities that were exposed to bans of different intensity. Furthermore, the ban on Batasuna is an interesting case study because the party had a high level of pre-ban support, above 20% across municipalities; and because the ban was promoted by the Spanish Parliament against the will of the Basque Parliament.
The main result of this paper is that a longer ban had a negative effect on support for Batasuna. The effect was large and persistent for at least two elections following the lifting the ban. Furthermore, using data on episodes of street violence perpetrated by Batasuna supporters, the results indicate that the extension of the ban for an additional electoral term led to a one-month spike in street violence in treated municipalities, but it had no effects during the following year. The interpretation of the results as a causal effect of the ban relies on the assumption that both groups of municipalities would have followed parallel trends in outcomes in the absence of the heterogeneous length of the ban. An inspection of pre-treatment trends suggests that this is a plausible assumption.
A party ban is a repressive measure that prevents a specific political party from contesting elections and obtaining any of the benefits from being in office or holding seats in parliament. Hence, previous studies of the effects of party bans mainly consider two sets of possible mechanisms. A first set of mechanisms concerns voters' reactions to the ban as a policy. These are assumed to be negative because of the democratic contradiction that it might entail, possibly generating sympathies towards the party among voters who are simply opposed to the ban. For instance, Van Spanje and De Vreese (2015) use data from a representative sample of Dutch voters interviewed before and re-interviewed after the unexpected court decision to prosecute the far-right MP Geert Wilders for hate speech, and find that it substantially enhanced his party's appeal. In the context of counter-terrorist policies, a common thread in the literature is that more indiscriminate actions are more likely to result in a backlash (Benmelech et al., 2015;Dragu, 2017;Dragu and Polborn, 2014). Likewise, the empirical literature on repression, which has focused on the effect of state violence, has shown that indiscriminate violence systematically reduces political support for the perpetrator, both at the individual level and overall in the most affected areas (Balcells, 2012;Lupu and Peisakhin, 2017;Rozenas et al., 2017). The negative effect of the ban on support for Batasuna, however, suggests that this set of mechanisms did not play a major role in this case.
A second set of mechanisms concerns the relative disadvantages suffered by the targeted party because of its inability to contest elections and hold seats in city councils. This might be associated with a loss of support for two reasons: the direct material disadvantages of not holding parliamentary representation and its effect on the party's mobilization capacity, and voters' increased information about competing politicians that would not have been elected otherwise. For instance, Dinas et al. (2015) show that small parties that barely cross the electoral threshold to enter parliament substantially improve their electoral performance in the following election, especially in new democracies, with weaker party branding. 4 The second main contribution of this paper is to study the heterogeneous effects of the ban, with the aim of understanding the role of these mechanisms. Because of the negative effect of the 2007 ban, I focus on mechanisms related to the party's inability to contest elections and hold seats in city councils. First, I study how the electoral effects of the 2007 ban unfold over time. This is possible because when the ban did not apply to all municipalities, Batasuna continued to contest the elections in those places where it was banned by calling for a null vote. Because the number of null votes is typically very close to zero, the immediate loss in support for the party under the ban is observable. Because the 2007 ban took place less than three weeks before the elections, if the direct costs of not holding city council seats are the main mechanism, this should not be detectable immediately, but rather emerge gradually over time, during the 2007-2011 legislature. In addition, under the ban, the level of support for the party does not correlate with any resources from holding seats in city councils, because these votes are counted as invalid votes.
On the other hand, if the main mechanism is voter learning about new parties, one would expect a significant correlation between the very immediate effect of the ban on support for Batasuna in 2007 and the effect once the ban was lifted. First, this is because wherever Batasuna voters are more likely to abandon the party under the ban, the composition of the remaining parties in the city council is less likely to be the same, because not all parties gain the same from the ban: I find that those that gain the most are small local parties, with less known platforms. Hence, these voters will be exposed to a more different city council for an additional legislature. Second, because under the 2007 ban, when voting for Batasuna does not translate into seats, the party's lack of potential for policy influence from the outset increases the incentives for instrumental voters to learn about alternative parties before the 2007 elections, and this could lead to an immediate loss in support. These mechanisms would lead to a positive correlation between the immediate and the long-run loss in support due to the ban.
The results indicate that the very immediate loss of electoral support in treated municipalities predicts very well the long-run loss of support following the lifting of the ban, which is consistent with voter learning about new candidates being the main disadvantage of the targeted party due to the ban, rather than the direct cost from being excluded from the city council. This suggests that party bans may be used by incumbents to reduce their challengers' strength by triggering an electoral reshuffling, with voters learning from the new scenario and potentially switching parties.
I also examine in which municipalities Batasuna is more or less likely to lose votes. I find that the vote loss due to the 2007 ban is especially pronounced in municipalities with a higher number of ETA attacks in the past, which is consistent with the idea that more indiscriminate policies are perceived as illegitimate and are more likely to backfire, even if overall this seems to be of secondary importance because of the negative effect of the ban on support for Batasuna. I also find that the effect is larger in municipalities where other nationalist parties and federal parties had similar vote shares in the local elections preceding the 2007 ban. This also suggests that the effect of the ban might depend on the incentives for party switching, and that the existence of alternative organizations in competition with the targeted party might be important, as proposed by Capoccia (2005). In both cases, the immediate and the long-run effects show a similar pattern, indicating again a persistent rather than a gradual pattern in the vote loss. Finally, I also find that the effect of the ban is greater wherever the pre-ban variance of the electoral support for the party is greater (conditional on the pre-ban level of support), which is also consistent with voters' propensity to switch parties playing a role in the effect of the ban. Nevertheless, because these municipalities could differ in other dimensions, the interpretation of these heterogeneous effects is suggestive, and further investigation of these hypotheses and mechanisms is left for future research.

The Ban on Batasuna
The Basque Country is a region in Northern Spain with a multi-dimensional party system. Every party is characterized by a policy position on the leftright spectrum and in relation to its stance regarding regional autonomy, with positions on the latter ranging from full state centralization to Basque independence. Apart from their policy positions, the electorate is similarly divided by cultural identity, which correlates with a range of preferences regarding decentralization. The main parties contesting the Basque elections can be divided into Federal parties (with candidatures everywhere in Spain) and Basque parties (with candidatures only in the Basque Country, typically with a preference for more regional autonomy). Among the latter, Batasuna was the main leftist and independentist party in the region until its ban in 2003. Over the period of analysis, this party adopted different names and electoral brands, including "Herri Batasuna", "Euskal Herritarrok", and "Acción Nacionalista Vasca". For simplicity, I refer to these parties and to the political representation of the "izquierda abertzale" (leftist-independentism) as Batasuna. These organizations had a continued political leadership, were considered by the Basque leftist-independentist organizations (trade unions, youth associations, anti-repressive organizations, social centers) as their electoral wing, and were consistently considered a continuation of each other by the Spanish and European judiciaries. 5 Before its 2003 ban, Batasuna used to obtain between 20 and 25% of the vote in the Basque local elections.
Politics in the Basque Country have been heavily influenced by the existence of Euskadi Ta Askatasuna (ETA), a terrorist organization in favor of the independence of the region. 6 ETA was created in 1958 (during Franco's dictatorship in Spain) and was active until 2011, when it announced a permanent end to its armed activities. 7 Over this period, ETA killed more than 800 people, mostly between the 1970s and the 1980s. Besides its direct victims, ETA's terrorism has had important economic consequences for the region and has also influenced Basque politics. One of the most politically relevant consequences of the existence of ETA has been the ban on Batasuna, which used to represent its ideological space (leftist-independentism) and never rejected ETA's terrorism.
In June 2002, the Spanish Parliament passed a new Law of Political Parties, with the support of more than 90% of its members, with the following aim: ". . . to guarantee the operation of the democratic system and the citizens' essential freedoms, avoiding the possibility that a political party could, in a reiterative and serious fashion, threaten this democratic regime of freedom, justify racism and xenophobia or lend political support to violence and the activities of terrorist groups. . ." 8 5 The Council of Europe refers to Batasuna as Batasuna (a.k.a. Herri Batasuna, a.k.a. Euskal Herritarrok) -Council Common Position 2004/500/CFSP. The Spanish newspaper "La Vanguardia" writes, in an article entitled The different names of Batasuna that "the suspension of activities of Basque Nationalist Action (Acción Nacionalista Vasca) has added the acronym of this party to the long list of names that the Abertzale (Independentist) Left has used to try to have a presence in the institutions since March 23, 2003, when the Supreme Court declared HB, EH, and Batasuna illegal. 6 A number of contributions have studied other questions related to identity, political conflict and terrorism in the Basque Country (Ansolabehere and Puy, 2016;Barros et al., 2009;de la Calle, 2007;de la Calle and Sánchez-Cuenca, 2013;Gardeazabal, 2011;Gil-Alaña and Barros, 2010), and its economic effects (Abadie and Gardeazabal, 2003) 7 During this period, ETA held a number of ceasefires. The last ceasefire started in September 2010. In January 2011, ETA announced that this ceasefire would be permanent and verifiable by international observers. Eventually, on October 2011, ETA announced a definitive cessation of all its armed activities. A few weeks later, the Council of Ministers asked the Supreme Court of Spain to ban Batasuna. In March 2003, following a period of deliberation, the Supreme Court banned Batasuna owing to its links with ETA and its failure to reject terrorism. 9 The ban was not the result of bargaining among the existing political agents in the Basque Country, which mostly opposed it; rather, it was a process led by the Spanish Parliament and subsequently enforced by the Courts. The Law of Political Parties was passed by the Spanish Parliament, with a majority government under the People's Party (PP), the party that had suffered the most victims of ETA terrorism. Indeed, this was a period of increasing international concern about terrorism, especially following the 9/11 terror attacks. 10 The political context surrounding the 2003 ban on Batasuna was a general non-conciliatory stance by the Spanish government and judiciary towards Basque political organizations and the political majorities in the Basque parliament. For instance, in February 2003 the Spanish National Court closed down Egunkaria, the only fully Basque language newspaper in circulation due to allegations of links with ETA, allegations that would be withdrawn seven years later. In June 2003, the Spanish parliament passed a law extending the maximum length of effective prison sentences from 30 to 40 years. A majority of members of the Basque regional parliament voted against the law and accused the Spanish government of basing its policies on a desire for revenge. And in November 2003, the Spanish government passed a new law introducing prison sentences of between three and five years for anyone calling for a referendum on autonomy, in response to the development of a new proposal on autonomy made by the Basque government (a coalition of moderate Basque nationalists and federal leftists).
After its ban, Batasuna adopted different strategies to maintain its electoral presence. In 2003, in the first local election after the ban, they were completely unsuccessful in their efforts. 11 With the aim of participating at the 2007 local election, Batasuna made use of two political parties, the first of these, "Abertzale Sozialisten Batasuna", however was fully outlawed by the Courts. The second, "Eusko Abertzale Ekintza-Acción Nacionalista Vasca" (EAE-ANV), was an old party that had been inactive for decades. In this case, the party was revived by Batasuna very shortly before the elections, so that the Courts and the police had insufficient time to find enough evidence of links to the previously banned organizations to ban the party before the election. After the Public Prosecutor failed to prove that EAE-ANV was the same party as Batasuna, instead of giving up, he decided to denounce the municipalities for which he thought the evidence of a connection with Batasuna was the strongest.
More specifically, the public prosecutor denounced the lists with a relatively large number of former Batasuna candidates (3 or more) and councilors (1 or more), or some other visible connection, such as former regional or provincial Batasuna MPs. 12 The courts, in the main, followed the public prosecutor's denounce and banned most of the denounced electoral lists, and a few nondenounced lists too. This affected around 50% of the municipalities. Although Batasuna denied having any connection to EAE-ANV until the Courts' final verdict was announced, shortly afterwards they endorsed EAE-ANV, revealing that it was part of their strategy to recover their positions in the institutions. 13 Even though Batasuna wanted to conceal the fact that they were behind EAE-ANV and that presenting its former candidates could raise suspicions, it should be borne in mind that the public prosecutor established his criteria ex-post (i.e., after the electoral lists had been presented), and that this was the only instance in which an abertzale organization -among the various organizations that were banned around that period -was partially banned based on local criteria. 14 With the aim of participating at the 2011 local election, Batasuna once more created another party, which was meant to lead a larger candidature named "Bildu", jointly with smaller parties such as EA (centrist independentists) and Alternatiba (a Basque splinter group from IU, federal leftists). For the first time, Batasuna decided to explicitly reject ETA's violence, with the aim of being allowed back into the political arena. Despite this, on the 1st May 2011, Spains' Supreme Court invalidated Bildu's electoral lists and forbade the inscription of the new Batasuna party on the registry of political parties, following police reports of links with ETA. However, a few days later, on May 5th, the Constitutional Court revoked the Supreme Court decision and allowed Bildu to contest the 2011 election to be held on May 22nd, in a close decision: with six judges voting in favor, five against (Garmendia Madariaga, 2011). In short, Batasuna was banned for two local elections (2003 and 2007), although the 2007 ban applied only to a subset of municipalities. Figure 1 shows the level of turnout at every local election in the Spanish region of the Basque Country between 1987 and 2015, the sample being divided between municipalities where Batasuna was banned in 2007 and municipalities where it escaped this ban. 15 Turnout had followed parallel trends across the treated and control municipalities before the ban, and it did not undergo any substantial changes after the lifting of the ban (in 2011 and 2015). Figure 2 reports null votes as a percent of turnout (note that they y-axis scale changes across these first seven figures). Null votes were very close to zero before the ban. In 2003, when Batasuna was banned everywhere and the party called for a null vote, null votes as a percentage of turnout jumped from almost zero to more than 15%, representing around 70% of Batasuna's average pre-ban vote share. In 2007, Batasuna was deemed legal in 98 of the 251 Basque municipalities, and, hence, in those municipalities null votes returned to being close to zero. In the remaining municipalities, null votes remained at around 15%. Note that municipalities escaping the 2007 ban received more null votes in 2003. As the following figure shows, this was to be expected, because they also had more pre-ban support. After taking the pre-ban support difference into account, it seems that, if anything, the municipalities that escaped the 2007 ban dealt slightly worse with the 2003 ban. This outcome is examined in further detail in a later section dedicated to identification threats. conditions, it obtained more support than previously. However, the larger vote shares obtained by Bildu can be attributed to the presence of other political actors in this coalition, such as the centrist and independentist Eusko Alkartasuna (EA), which prior to the ban had contested the elections either alone or in coalition with the conservative nationalists. 16 Figure 4 reports the sum of the votes for all nationalist parties, including Batasuna, EA, the conservative nationalist PNV, EA-PNV coalitions, Aralar, a moderate independentist party, and Null votes (capturing support for Batasuna in 2003 and 2007). The figure suggests that the 2007 ban led to a decline in support for Basque Nationalist parties in treated municipalities, a trend that persists after the lifting of the ban. Figure 5 reports the vote share of Federal Leftist parties (mainly the Socialist Party, United Left, the Green Party, and Podemos). Although prior to the 2007 ban the vote share of these parties was slightly lower in treated municipalities, in 2007 it increased above the level of control municipalities, and it remained at a similar higher level after the lifting of the ban. In contrast, Figure 6 reports the vote share of Federal Right wing parties, without reporting any marked changes over time. This could be because Batasuna is a leftist party and Federal Leftist parties are slightly more sympathetic to Basque Nationalist positions than Right Wing Federal parties, which tend to seek confrontation. Yet, the magnitude of these changes is quite small.  to those where the ban lasted for two elections (2003 and 2007). Specifically, they seem to react in a very similar fashion to aggregate popularity shocks (i.e., they seem to be following parallel trends). Because I observe support for Batasuna before and after the ban in all municipalities, and the length of the ban varies across municipalities, conditional on a parallel trends assumption a difference-in-differences estimate recovers a causal effect of the differential length of the ban on the electoral success of Batasuna. In order to investigate the relationship between the 2007 ban on Batasuna and subsequent electoral outcomes, I therefore estimate the following equation:

Effects on Electoral Support
where Outcome mt is an electoral outcome in municipality m in election t, α m and δ t are municipality and election fixed effects, and I(Post 2007) t × I(Ban in 2007) m is a dummy variable equal to one for the treated municipalities after the 2007 ban. The main outcome of interest is the level of electoral support for Batasuna. In all regressions, 2007 is excluded from the sample, because Batasuna had a different legal status across treated and control municipalities, and, hence, differences in support for Batasuna cannot be directly compared.
In a heterogeneous treatment effects framework, under the assumption that treated and control municipalities would have followed parallel trends in the absence of the 2007 ban, difference-in-differences estimates an average treatment effect on the treated (ATT). This is an interesting treatment effect, because it is the treatment effect for a subset of municipalities with a considerable level of support for Batasuna. Because Batasuna was banned in all municipalities in 2003, this is an intensive margin rather than an extensive margin effect.
It should be stressed that the identification assumption does not require the 2007 ban to be randomly assigned across municipalities, but to depend on time-invariant unobservables. The 2007 ban was heterogeneous because the public prosecutor failed to find enough information before the election to demonstrate that the electoral lists presented by EAE-ANV were actually Batasuna lists. After the electoral lists were presented, they were banned wherever the connection with Batasuna was most evident (i.e., those including high numbers of former Batasuna candidates or councilors). Figure 3 shows that municipalities that escaped the 2007 ban had a higher level of pre-ban support, suggesting that this allowed Batasuna to find clean candidates more easily, which ended up increasing their probability of escaping the ban. Ex-ante, it was not clear how including Batasuna candidates in the list would play a role, because the precise criterion was established ex-post and because this was the first time that an organization had been banned heterogeneously across municipalities. However, given the context of continuous bans, appearing to be as different as possible from Batasuna on paper was a very likely aim when drawing up the electoral lists. Yet, this seems orthogonal to pre-treatment trends. A visual inspection of Figures 1 to 7 shows that treated and control municipalities had followed parallel trends for a long period of time, indicating that they tended to react similarly to aggregate shocks. Hence, if the ability to find clean candidates was related to time invariant unobservables, these are absorbed by municipality fixed effects, and difference-in-differences estimates the causal effect of interest.
However, it is interesting to consider why this assumption might be violated. One possibility is that the aim of the Courts was to ban municipalities where support for Batasuna was growing. In this case, difference-in-differences estimates would be biased towards finding a positive effect of the ban on support for Batasuna. Another possibility is that Batasuna was better able to find clean candidates in those places where its support was increasing (i.e., not only where it was higher). In this case, the estimates would be biased towards finding a negative effect of the ban on support. Although it is not possible to observe electoral support immediately before Batasuna presented its lists, in the section on identification threats I test for the existence of differential trends in support for Batasuna across treated and control municipalities in the regional elections that took place in 2005, just two years before the 2007 ban. Moreover, I exploit the criteria employed by the public prosecutor to denounce EAE-ANV lists, based on the number of former Batasuna candidates and councilors, to provide instrumental variables estimates of the effect of the ban. These estimates rely on a weaker identification assumption but have low statistical power, because they combine triple differences with imperfect compliance and a relatively small sample. The instrumental variables identification assumption and the results, which are qualitatively similar to the difference-in-differences estimates, are discussed in Appendix B.
In the estimation of the baseline difference-in-differences, to compare municipalities in which the party had a more similar local infrastructure before the ban, I group municipalities according to whether Batasuna presented lists in each of the local elections before the ban and include year fixed effects for each of these groups, namely pre-ban presence by election fixed effects. The largest group is the one for which Batasuna presented lists in every election between 1987 and1999. 17 In 2003, I measure support for Batasuna with the share of null votes, and to be consistent across years, the dependent variablesupport for Batasuna -is defined as Batasuna V otes+N ull V otes T otal V otes . Arenas (2016) shows that under the 2003 ban, null votes only increased in municipalities where Batasuna had obtained votes in the past, and that they represented at least two-thirds of the votes that they would have received in a regular election, indicating that this is a good measure of support. Recall that in all regressions, 2007 is excluded from the sample, because support is not comparable because of the different legal status, which is precisely our treatment of interest. Table 1 reports the difference-in-differences estimates of the effect of the 2007 ban on the outcomes in Figures 1 to 7. Table A1 reports descriptive statistics by treatment group and period. As suggested by the figures, the results in the top panel show that a longer ban had no effect on turnout nor on null votes, but that it had a negative effect of around 6pp on Batasuna's vote share, which translates into a negative effect of a similar size on the vote share of nationalist parties overall. The magnitude of the effect is important: it represents around 15% of the average post-ban support for Batasuna, or almost 0.3 standard deviations. This compares to 8.7% of all the nationalists' vote share, or 0.2 standard deviations.
The result on turnout is important for the interpretation of these results. One possible concern about interpreting the results as a negative effect of the ban on support for Batasuna's positions is the party's move towards a more moderate stance in relation to ETA after the lifting of the ban. For instance, Gould and Klor (2010) show that in Israel terrorism makes voters more likely to vote for the right, but at the same time it shifts their preferences to the left. In this setting, an analogous outcome would be people no longer voting for Batasuna because they find them too moderate. Although without individual-level data I cannot completely rule this possibility out, given that Batasuna remains the most extreme platform, the result that a longer ban does not affect turnout, combined with the parties' vote share patterns, suggests that this is unlikely.
The bottom panel of Table 1 reports a small positive effect on the vote for the Federal left, a slightly smaller negative effect on the vote for the Federal 17 Groups are defined for all possible combinations of presenting electoral lists between 1987 and 1999. For instance, municipalities with no lists until 1991 but with lists in 1995 and 1999 are grouped together, as are municipalities with lists only in 1991, municipalities present in all elections, municipalities present in all elections but 1995, and so on. right, and an increase in votes for local parties, as suggested by the previous figures.

Identification Threats
The interpretation of the previous results as causal effects relies on the common trends assumption. This assumption is fundamentally untestable, but pretreatment differences can nonetheless be informative about its plausibility. To formally test for pre-treatment differences in trends, Table A2 reports placebo regressions, that include a lead treatment indicator which is never significant and mostly close to zero, indicating that the estimates in Table 1 are not driven by differential pre-ban trends, as suggested by the figures.
In the case of null votes under the 2003 ban, Table A8 further tests for differences in null votes between treatment and control municipalities. The estimates in the first column, accounting for municipality and election fixed effects, indicate that in 2003 null votes increased slightly less in treated municipalities, as suggested by Figure 2. Note however that municipalities in which Batasuna was banned in 2007 also presented a lower level of support for Batasuna before the ban, and, hence, one would also expect fewer null votes in those municipalities in 2003. In fact, the difference in null votes in 2003 is smaller than the pre-ban difference in support for Batasuna between treated and control municipalities, which means that treated municipalities had lost less support in 2003, compared to municipalities escaping the ban in 2007.
Column 2 reports estimates that include pre-ban presence by year fixed effects, as in the baseline specifications throughout this paper, with the aim of comparing municipalities with more similar pre-ban strengths. After conditioning on these, municipalities with a longer ban received slightly more null votes in 2003, but again this has to be compared with the pre-ban difference in support after conditioning on pre-ban presence by election fixed effects. This is done in Column 3, which reports difference-in-differences estimates on support for Batasuna (defined as Batasuna Votes + Null Votes) for the full sample, including a lead treatment indicator to test for differences under the 2003 ban. The results in Columns 2 and 3 indicate that the difference between treated and control municipalities under the 2003 ban is small, which is consistent with the common trends assumption.
In another test of pre-treatment trends, Column 2 of Table A5 examines the effect of the 2007 ban in local elections on support for Batasuna in the Basque regional elections. In addition to considerations of identity (Basque elections are seen as national elections for Basque Nationalists), in Basque Regional elections Batasuna can obtain a significant share of the seats and wield influence. 18 Regional elections present a very similar party system to that of the local elections, and also a proportional electoral rule, with seats being allocated at the province level. 19 At the Basque regional level, Batasuna always had the same legal status across municipalities, and it was banned in 2009, when it called for a null vote, as it did in the local elections during the ban. The results show that in the 2005 regional election, just two years before the 2007 ban, treated and control municipalities were still performing similarly. However, Columns 1 and 2 in Table A5 report that, after 2007, Batasuna lost support (around 1.2pp) in treated municipalities in the regional elections.
18 This is in contrast with the Spanish General elections and the European elections. 19 The electoral districts are the three provinces making up the Basque Autonomous Community: Araba/Álava, Bizkaia/Vizcaya, and Gipuzkoa/Guipúzcoa, and the electoral rule is d'Hondt within each district, with a 5% threshold. Regional elections took place in 1986, 1990, 1994, 1998, 2001, 2005, 2009, and 2012. Another identification threat would be the existence of compound treatment effects. For instance, the changes in Batasuna's platform after the end of the ban could have a differential appeal across treatment and control municipalities. As a robustness check, Table A10 includes estimates that allow for heterogeneous post-ban effects, depending on a number of covariates that could plausibly induce heterogeneous effects. Across columns, the estimates allow for heterogeneity in post-ban election effects by municipality size, (log) terror attacks in the 10 years before the ban, (log) street violence attacks in the year before the ban; by pre-ban local electoral outcomes, including Batasuna's vote shares, the sum of Basque nationalists, and the sum of federal parties; and by pairwise interactions between each of these six variables. The results indicate that the estimates are robust to these heterogeneous time effects. 20 Another source of violation of the common trends assumption concerns the definition of the outcome of support for Batasuna. As explained in the section on the ban on Batasuna, after the ban, Batasuna re-emerged as Bildu, a new coalition led by Batasuna, but in conjunction with smaller parties that enjoyed little electoral success before the ban, such as EA (social democrat Basque nationalists), and Alternatiba (a splinter group from IU, federal leftists). This could be problematic if there had also been a differential increase in support for some of Batasuna's new partners at the same time of the ban. The effect on the Basque nationalist vote share overall, which includes almost all of these partners (and also the large Christian-democrat Basque nationalists (PNV)), suggests that this is not a concern. Nonetheless, I also present entropy balancing matching estimates (Hainmueller, 2012) to further deal with this. Entropy balancing calibrates municipality weights so that the re-weighted treatment and control group satisfy a set of pre-specified balance conditions. More specifically, I use entropy balance weights so that the pre-treatment support for each political party is on average the same across treatment and control municipalities. With these weights, any differential increase in support for any political party would have the same effect in treated and control municipalities, on average. 21 The entropy balancing estimates in Table A9 are very similar to those in 21 More precisely, I balance the level of support for a number of political parties in each of the four pre-treatment elections (2003, 1999, . . ., 1987). These are the vote shares of Batasuna; the vote shares of the Basque Nationalist Party (PNV, Basque Christian Democrats), the vote shares of Eusko Alkartasuna (EA, Basque Social-Democrats, which becomes part of Bildu); the vote shares of EA-PNV coalitions, that take place in some municipalities in some pre-treatment elections, the vote shares of IU (Federal leftist, a splinter of which becomes part of Bildu); the vote shares of Aralar, a small leftist independentist moderate party that joined Bildu (which pre-treatment only contested the 2003 local election); the vote shares of the PP (the Federal conservatives), and the vote shares of the PSOE (the Federal social democrats); and also dummies for Batasuna being present in each of the pre-ban elections.
A party ban is a repressive measure that prevents a specific political party from contesting elections and obtaining any of the benefits from being in office or holding seats in parliament. Hence, previous studies of the effects of party bans have primarily considered two sets of possible mechanisms. A first set of mechanisms concerns citizens' reactions to the ban as a policy. These are assumed to be negative because of the democratic contradiction that it might entail, possibly generating sympathies towards the party among people who are simply opposed to the ban, in line with existing evidence (for instance, Van Spanje and De Vreese (2015)). Given the overall negative effect of the ban on support for Batasuna, these mechanisms do not seem to be playing a central role in this case. A second set of mechanisms concerns the relative disadvantages faced by the targeted party, given its inability to contest elections and hold parliamentary (city council) representation. This might be associated with a loss of support because of (1) the direct material disadvantages of not holding parliamentary representation and its effect on the party's mobilization capacity, and (2) voters' increased information about competing politicians and parties that may not have been competitive or elected otherwise.
It is difficult to disentangle empirically these mechanisms associated with the inability to contest elections and hold city council seats, which have not been formalized in previous qualitative party ban studies. A particular feature of the ban on Batasuna is that when it did not apply to all municipalities, the party continued to contest elections in those places where it was banned by calling for a null vote. Hence, it is possible to observe the immediate loss in support for the party under the ban, because null votes are otherwise typically very close to zero. This is interesting because an understanding of the timing of the effect of the ban can be informative. Here, given the sudden nature of the 2007 ban, any difference in the party's mobilization capacity and the material benefits of holding city council seats should emerge gradually over time. In addition, differently from a regular election, under the 2007 ban, the support for Batasuna within treated municipalities does not correlate with resources from parliamentary representation during the 2007-2011 legislature, because those votes do not translate into seats. Hence, we would not expect a correlation between the very immediate effect of the ban on support for Batasuna in 2007 (with Batasuna calling for a null vote in treated municipalities) and the effect once the ban was lifted, if the main effect of the 2007 ban is a resource effect.
On the other hand, if the main mechanism involves voters learning about new parties, one would expect a significant correlation between the very immediate effect of the ban on support for Batasuna in 2007 (when Batasuna called for a null vote in treated municipalities) and the effect once the ban was lifted. There are two explanations for this: first, that when more Batasuna voters abandon the party under the ban, the composition of the remaining parties on the city council is less likely to be the same: recall, those gaining the most from the ban are small local parties, with less-known platforms, as shown in Figure 7 and Table 1. And second, that because under the ban voting for Batasuna does not translate into seats, instrumental voters have incentives to learn about new parties before the 2007 elections, which could result in an immediate vote loss that persists on the lifting of the ban. These mechanisms would lead to a positive correlation between the immediate and the long-run losses in support resulting from the ban.   similar, suggesting that those municipalities where voters immediately switched allegiance to another party were also those municipalities that lost support on the lifting of the ban. Column 1 in Table 2 reports estimates of the effect of the ban that control for an interaction of the average pre-ban vote share of Batasuna and a post-2007 dummy. This does not affect the results, indicating that the average effect of the ban is orthogonal to the average pre-existing strength of the party. Column 2 in Table 2 further controls for an interaction of the 2007 vote share (driven by votes for Batasuna in those municipalities where it was legal, and by null votes in those where it was banned) and a post-2007 dummy. Note that this is a bad control (i.e., given that it is in itself an outcome of the ban), and, hence, our interest in this specification is descriptive rather than causal.
The results indicate that those municipalities where Batasuna immediately lost support were also those municipalities where Batasuna lost support on the lifting of the ban. This is interesting because the immediate loss in support occurred before any difference in the direct disadvantages from not being on the city council could make themselves manifest as a result of the 2007 ban, which is inconsistent with this factor being very important for the estimated effect. 22 Rather, this pattern is more consistent with voter learning about new candidates. 22 An effect through candidates' incentives to join Batasuna also seems less consistent with the data because one would also expect this to emerge progressively over time. In addition, the ban also has an effect on support for Batasuna in regional elections, where treated and control municipalities vote for the same candidatures, as shown in Table A5, and also in regional elections the immediate vote loss predicts the effect of the ban. I also examine in which municipalities Batasuna was more or less likely to lose votes, studying heterogeneous effects of the ban across municipalities depending on their level of pre-ban violence and on voters' incentives to abandon Batasuna under the ban. The literature on counter-terrorism has emphasized that a backlash is more likely with measures that inflict damage to a broader population than that of the terrorist group, which might perceive the measure as illegitimate (Benmelech et al., 2015;Bueno de Mesquita and Dickson, 2007). In this case, one might expect voters in municipalities that have not witnessed many episodes of terror to consider the ban as being less legitimate, and, hence, to maintain their vote for Batasuna. A second source of heterogeneity concerns voter incentives when votes for Batasuna are counted solely as null votes. In this case, one might expect voters to stick more firmly to their vote when such a decision is less consequential, for instance when it does not make it easier for antagonist parties to gain power. Table A3 reports descriptive statistics showing that in those municipalities in which Batasuna was banned in 2007, the number of terror attacks in the previous decade had been 13% lower than in the control group, that the margin between nationalist and federal parties in 2003 was a little narrower, and that the population size was smaller. Table 3 reports estimates that investigate these questions by including a number of interaction terms. Column 1 reports estimates on the immediate vote loss (i.e., support for Batasuna under the 2007 ban), while Column 2 provides estimates on the vote loss once the ban had been lifted (i.e., excluding 2007). The interacted variables are demeaned so that the main term of the effect of the ban can be interpreted at their mean values. Note that the regression also controls for interactions with a post-treatment dummy. First, the results indicate that in those municipalities where ETA had been more active in the decade before the 2007 ban, the loss in support was more pronounced, which is consistent with the idea that more indiscriminate policies are perceived as less legitimate. 23 Second, the results indicate that wherever the margin in the most recent local election between Basque Nationalist and Federal parties had been close (i.e., less than 10pp), the vote loss was larger, which is consistent with the idea that pivotality considerations might be relevant. It is interesting to note that the immediate and the long-run heterogeneous effects of the ban are similar. This is consistent with the previous finding that the immediate vote loss due to the 2007 ban is a good predictor of the vote loss on the lifting of the ban. Again, this pattern is more consistent with voters learning about competing parties, which could be proportional to the immediate loss (either because of voter learning before the 2007 election, or during the electoral term), and less consistent with the direct benefits from holding seats in the city council, which one would expect to emerge progressively over time. Hence, this suggests that party bans may be used by incumbents to reduce their

Street Violence
In addition to ETA's terrorist attacks, which killed more than 800 people, its supporters perpetrated thousands of low-intensity violent attacks while the group remained active. These attacks, known as Kale Borroka (or street fights), involved primarily the burning of public buses, private vehicles and trash containers, and incendiary bomb (Molotov cocktail) attacks on court buildings, banks, political party headquarters and police stations, typically perpetrated at night or following demonstrations. de la Calle (2007) studied their determinants between 1996 and 2000, and described it as a persistent type of non-organized violence (that is, no permanent organizations claimed responsibility for the attacks), which tended to occur in the more polarized municipalities. All political parties except Batasuna condemned the violent squads involved in the Kale Borroka, forming part of which was typically the first step towards becoming an ETA militant. In 2000, Spanish-based insurance companies paid out almost 9 million euros because of the acts of violence of the previous three years (de la Calle, 2007). In 2002, the courts held Batasuna civilly responsible for a number of these violent street attacks, and condemned them to pay part of their costs, arguing that the finances of Batasuna and the Kale Borroka were jointly determined. 24 Given that under Spanish law, such attacks are considered acts of terrorism, in what follows, I use the terms street violence and street terrorism interchangeably. Between 2006 and 2008, more than 350 acts of street terrorism took place in the Basque Country, and in earlier periods they had been even more frequent. The effect of the party ban on this type of violence is interesting for two reasons. First, after accounting for municipality and time fixed effects, it is a measure of the intensive margin of support for the targeted movement by violent means. Second, it is informative of the reaction of a terrorist organization to a measure of political repression. Batasuna and its supporters combined both political and violent mobilizations to maximize the threat of revolution, subject to a certain budget constraint. 25 A party ban might be seen as a tax on political mobilization, and given the central role played by Batasuna in the movement, one would expect it to have large income effectsreducing overall the strength of the movement. These income effects would go in the opposite direction to that of the substitution effect arising from the relative decrease in the price of violent mobilization with respect to the price of political mobilization due to the ban. However, if the ban were to make the aggrieved population more willing to support the targeted movement, increasing their ideological motivation, this could also lead to an income effect of opposite sign, a motivation effect. Bueno de Mesquita (2005) fully models the strategic interaction between a terrorist organization, a government, and a population of potential terrorist volunteers, where counterterrorism reduces the ability of terrorists to carry out attacks, by weakening their infrastructure, while at the same time it diminishes their economic opportunities, by reducing their opportunity costs, albeit while increasing their ideological motivation. 26 I estimate the effect of the May 2007 ban on street violence using detailed data at the municipality level for street terrorism episodes perpetrated between April 2006 and June 2008, as reported in Buesa (2008), based on media reports. Table A6 reports descriptive statistics by treatment group and period. 27 Figure 9 displays the average monthly number of street terrorism episodes, by treatment group. Figure 10 shows the same data but averaged over quarters 27 I include all the episodes in Buesa (2008) involving petrol bombs, damage and destruction with blunt objects, or heavier attacks. Demonstrations, or attacks with eggs or paints are excluded from the sample. The data in de la Calle (2007)   Standard errors clustered at the municipality level in parentheses. The dependent variable is a dummy equal to one if there is one or more episode of street terrorism in a month.
rather than months. The figures show that before the ban was extended in May 2007 the trends in street terrorism ran parallel, and that the extension of the ban triggered a peak in street terrorism in the very short run. However, after this spike, the ban led to a slight decline in violence in treated municipalities, especially compared to pre-ban differences, the overall effect being ambiguous. 28 Table 4 reports difference-in-differences estimates. In the top panel, the dependent variable is the monthly number of street terrorism episodes in a municipality; in the bottom panel, a dummy for whether there was a street terrorist episode in a given municipality in a given month. All regressions include municipality fixed effects and period (month × year) fixed effects by pre-ban presence of Batasuna. The regression results in Column 1 of both panels show that wherever the ban was extended, the probability of an attack in the following month increased by almost 10pp, with 0.2 more attacks being carried out, on average, in those municipalities. With around 150 treated municipalities, this implies a spike of around 30 attacks in the month after the ban. However, if I consider the full post May 2007 period, the overall effect is close to zero. The results in Column 3, which exclude May 2007 from the sample, show that this is because after the very short-run spike, the extension of the ban slightly decreased the monthly probability of a street terrorist attack in treated municipalities in the remaining months by 1.3pp, although this is not precisely estimated.
The results are consistent with a positive motivation effect of the ban that vanishes quickly, in line with the findings of Jaeger et al. (2012) of a fleeting radicalization of the Palestinian population after Palestinian fatalities. Table  A7 reports the results of a placebo test, indicating that shortly before May 2007 treated and control municipalities were performing similarly, as suggested by Figures 9 and 10.

General Equilibrium Effects and External Validity
One limitation of this case study is the possible existence of general equilibrium effects. The first of these, common to most public policy evaluations, concerns the difference between the effect of the implementation of the ban and the effect of a credible announcement of a reaction function by the government that could eventually lead to a ban (i.e., the equilibrium effect). For instance, Dragu (2011Dragu ( , 2016 shows how reducing privacy or constraining free speech in response to terrorist attacks could increase terrorism in the first place, by changing the security agencies' stakes for terrorism prevention or by possibly increasing the terrorist organization's initial stake for a successful terrorist attack. Hence, one avenue for future research is identifying an empirical setting with potential for capturing how the existence of the ban might affect, in the future, the probability of observing a ban, or terrorism, in the first place, through its effects on terrorism or political behavior. A second type of general equilibrium effect, more specific to this setting, concerns the existence of externalities across municipalities, which could violate the Stable Unit Treatment Value Assumption (SUTVA) (Rubin, 2005) that is required for a causal interpretation of the estimates in this paper. In the ideal experiment, party bans would be randomly assigned across independent party systems. However, this is not feasible, but without random assignment, it is difficult to find plausible control groups in independent party systems. This paper overcomes this problem by exploiting the differential intensity of a given ban within a party system, across municipalities that are comparable and which provide a plausible control group. The cost of this approximation is the possibility of spillovers between municipalities. For instance, the effect on party mobilization capacity and resources that would lead the ban to reduce political support could be attenuated by the presence of Batasuna in some municipalities in 2007, which would bias the results towards finding no effect of the ban, because the party could be mobilizing and spending resources from control to treated municipalities. Nevertheless, given the large estimate of the effect of the 2007 ban, this is unlikely to be a concern. On the other hand, the population in municipalities where Batasuna was legal in 2007 could still react negatively to the ban as a policy because Batasuna was banned in some other municipalities, which could increase support for Batasuna in control municipalities. This would bias the results instead towards finding effects of the ban that are overly negative. While this cannot be fully ruled out, the immediate effect of the 2007 ban clearly seems driven by a decrease in support in treated municipalities (i.e., by null votes in treated municipalities being substantially lower than Batasuna votes in pre-ban elections) rather than by an increase in support for Batasuna in control municipalities. This immediate drop is the main predictor of the long-run effect, which suggests that this explanation is not very likely.
Finally, the focus of this paper on the Basque conflict raises concerns regarding the external validity of the results. A first external validity concern is that the ban of Batasuna was a temporary ban. Nonetheless, it turns out that this is no exception, and often the targeted political movement manages to return, following periods of variable length, under a similar guise. 29 For instance, the Sinn Féin was banned between 1956 and 1974 in Northern Ireland, and the Turkish Islamist Welfare Party was succeeded by the Islamist Virtue Party shortly after its ban in 1998. In Israel, the far right Kach Kahanist party, which was banned in 1994, was succeeded by Otzma Yehudit in 2012, and tolerated by the courts. The Communist Party of Germany (KPD) was banned in 1956 and was reconstituted in 1968 as the German Communist Party (PKD).
A second external validity concern is that the 2007 ban on Batasuna is informative of the intensive margin effect of an additional or longer ban rather than of the extensive margin, and Batasuna is a party with a fairly solid support base. Table A11 reports heterogeneous effects of the ban by the average level and variance of pre-ban support. Note that the interacted variables are demeaned so that the coefficient of the main term can be interpreted as the effect at the mean. While the heterogeneous effect by average pre-ban support is imprecisely estimated and the effect at the mean is very close to the baseline estimate, the heterogeneous effect by the variance of the pre-ban support suggests that the loyalty of the support base, or the voters' eagerness to switch parties, is significantly related to the effect of the ban. 30 However, this case study cannot identify possible non-linearities concerning ban length and electoral support.
Given the inevitable trade-off between identification via case studies and external validity, future research focused on other conflicts should be able to shed more light on this question and its associated mechanisms, because empirical studies of conflict inherently take place in unstable and changing institutional environments, making the results more contingent on a particular context. 31 Nevertheless, it should be borne in mind that, overall, the estimates in this paper are consistent with the qualitative study by Bale (2007), who following an analysis of the bans on Batasuna, the Islamist Welfare Party in Turkey, and the far-right Vlaams Blok in Belgium, concludes that the pragmatic predictions that party bans routinely occasion -namely perversity, futility and jeopardy, as in Hirschman (1991) -are not borne out in practice.

Conclusions
Party bans are common in many countries, but disentangling their effects from the political circumstances leading to the proscription is not easy. In this paper, I have attempted to circumvent this problem by empirically evaluating the electoral effects of the ban on Batasuna, the political wing of ETA, exploiting the finite and heterogeneous nature of the ban across Basque municipalities. The main result of this paper is that a longer ban had a negative effect on support for Batasuna, the effect being large and persistent for at least two elections following the lifting of the ban. Furthermore, using data on episodes of street violence perpetrated by Batasuna supporters, the results indicate that the extension of the ban for an additional electoral term led to a one month spike in violence in treated municipalities, but it had no effect during the following year. These results are interesting because while party bans deprive the extremist movement of organizational resources and social visibility, they might encounter strong principled opposition in public opinion because of the contradiction with democratic principles. The interpretation of the results as a causal effect of the ban relies on the assumption that both groups of municipalities would have followed parallel trends in electoral outcomes in the 30 Nonetheless, for any value of the average pre-ban support and its variance in the sample, the predicted effect of the ban is negative.
31 For instance, in this case, the context was a general non-conciliatory stance by the Spanish government and judiciary. Hence, a limitation of this paper is that it is difficult to know whether the effect of a ban would change with a different mixture of repressive and non-repressive government policies. absence of the heterogeneous length of the ban. An inspection of pre-treatment trends suggests that this is a plausible assumption.
Although it is difficult to disentangle the specific mechanisms that led to the loss of support for the party, a number of heterogeneity analyses point towards the importance of voter learning about parties that otherwise would have not been competitive or elected. This is because the immediate effect of the ban on support for Batasuna, which is measurable because Batasuna contested the elections under the ban by calling for a null vote, is a very good predictor of the loss in support following the lifting of the ban. A greater immediate loss in support is associated with a greater change in the make-up of the city council, and instrumental voters have an incentive to learn about new parties under the ban, which would correlate both with the immediate and with the long-term loss in electoral support. If instead the effect of the ban were due to the material advantages of being on the city council, one would not expect the effect to emerge immediately but rather more progressively. The loss in support is also more marked in municipalities where voters are more prone to switching parties, which is also consistent with this interpretation. Hence, overall the results suggest that party bans may be used by incumbents to reduce their challengers' strength by triggering an electoral reshuffling, with voters learning from the new scenario and potentially switching parties. Further examination of these mechanisms is left for future research.