1. Introduction
The last two decades have seen an unprecedented rise in violence committed by right-wing extremists. This tendency is demonstrated by high-profile events such as the 2011 attacks in Norway, the 2019 shootings in Halle and Christchurch, as well as systematically collected data showing the spread of such terror attacks in the Western world (Auger, Reference Auger2020; Walters and Chang, Reference Walters and Chang2021). A key question regarding these acts of violence is their relationship to the growing popularity of far-right political parties: to what extent does violence against vulnerable minorities fuel support for far-right parties?
While there is a rich literature exploring the impact of terrorism on public opinion (Berrebi and Klor, Reference Berrebi and Klor2008; Montalvo, Reference Montalvo2011; Legewie, Reference Legewie2013) most studies consider terror attacks committed by minority groups (Godefroidt, Reference Godefroidt2023). In contrast, the literature on the impact of right-wing terror attacks is much sparser, leading to mixed findings. On the one hand, some studies suggest that right-wing terror attacks increase affect for immigrants and lower approval of right-wing positions in their aftermath (Jakobsson and Blom, Reference Jakobsson and Blom2014; Pickard et al., Reference Pickard, Efthyvoulou and Bove2003). This is usually attributed to the “Black Sheep effect”, whereby ingroup members reject deviants to maintain group status (Marques, Yzerbyt and Leyens, Reference Marques, Yzerbyt and Leyens1988) and thus voters reject parties and ideas that they associate with the perpetrators who challenge existing social norms. In contrast, other studies have found the opposite, showing that at least under some circumstances such terror attacks increase support for the far-right, possibly because they raise the salience of issues—such as immigration or law-and-order—owned by these parties (Krause and Matsunaga, Reference Krause and Matsunaga2023; Völker, Reference Völker2024; Sabet, Liebald and Friebel, Reference Sabet, Liebald and Friebel2025).Footnote 1
However, it is important to note that the scope of cases covered by existing studies is quite narrow. They focus almost entirely on Western democracies—like Germany and Norway—and explore the consequences of terror attacks against immigrants. Because the impact of such violent events clearly depends on both the characteristics of the perpetrators and victims (Jacobs and van Spanje, Reference Jacobs and van Spanje2021) and the way the media covers such events (Völker, Reference Völker2024; Sabet, Liebald and Friebel, Reference Sabet, Liebald and Friebel2025), expanding the universe of cases is essential to a more complete picture. The contribution of this research note is to bring new evidence to this question by studying a new case and utilizing a novel methodological approach that can provide more credible causal estimates.
Specifically, we explore the electoral impact of a series of ethnically motivated terrorist attacks against members of the Roma ethnic community by right-wing extremists in Hungary in 2008–2009. The attacks strongly resembled terrorist attacks in other European countries and took place against the backdrop of growing ethnic tensions and the rise of the anti-Roma radical right Jobbik party that increased its support from 2.2% in 2006 to over 16% by 2010. Utilizing an ensemble of approaches to causal inference that combine both quasi-experimental and synthetic control groups with longitudinal data on the settlement-level vote share of Jobbik, we demonstrate that support for the far-right increased by 11–14 percentage points more in the attacked settlements from 2006 to 2010 than it did in the control settlements. Our analysis also provides evidence of the temporal and spatial scope of this effect: the effect of the attacks spilled over to neighboring villages, even quite distant from the location of the attacks, and persisted till the national and EU elections in 2014, but not in the national election of 2018.
2. Research design
2.1. Context
We study the impact of anti-Roma terrorism on far-right political support in Hungary.Footnote 2 The anti-Roma attacks constituted the single most important case of anti-minority violence in the country since World War II. They took place in distant villages across Eastern and Northern Hungary over the course of approximately one year from July 2008 to August 2009, leaving six people dead and many more injured. The terrorists were a group of men who consistently employed the same methods and selected similar targets: they attacked Roma-inhabited houses with arson and gunfire at multiple locations. Details are found in Table OA1. The trial proceedings revealed that the murders were carefully premeditated against civilians with whom the attackers had no previous relations. The motives of the perpetrators were clearly political in that they sought to incite further violence among and against Roma (Miklósi, Reference Miklósi2011; Jászberényi, Reference Jászberényi2016; Tamás, Reference Tamás2019).
Meanwhile, the political landscape also went through radical realignment. The immediate aftermath of the financial crisis and the ensuing collapse of the governing socialist-liberal coalition government brought not only simmering ethnic tensions but also the rise of the organized radical right (Gyöngyösi and Verner, Reference Gyöngyösi and Verner2020). The radical Jobbik party (“Better” or “More Right” in Hungarian) grew from a marginal organization in 2006 into a sizable parliamentary force, capturing 15% of the national vote in the 2009 European Parliament elections just before the last attacks, and 17% in the 2010 parliamentary elections. Anti-Gypsism was at the center of Jobbik’s political agenda, as the promise of putting an end to “Gypsy crime” and Roma welfare scrounging featured prominently in the party’s platform. It allowed Jobbik to successfully mobilize voters in Hungary’s Northern industrial rustbelt and left-behind rural areas on the Northern Great Plains, the regions with the highest Roma population share (Karácsony and Róna, Reference Karácsony and Róna2011). We provide additional details about anti-Roma prejudice in Hungary as well as the rise of Jobbik compared to general EU-wide trends in the Online Appendix Sections A and B.
2.2. Identification
To estimate the causal effect of the attacks on electoral results, we compare changes in the support for Jobbik in attacked villages to similar but unattacked ones. We do so relying on versions of the following difference-in-differences model:
\begin{equation}
\text{JobbikShare}_{it} = \beta_0 + \beta_1 \text{Attacked}_i + \beta_2 \text{Post}_t + \beta_3 (\text{Attacked}_i \times \text{Post}_t) + \mathbf{X}_{it} \gamma + \varepsilon_{it},
\end{equation}where 
$\text{JobbikShare}_{it}$ is the vote share of Jobbik in settlement 
$i$ at time 
$t$, 
$\text{Attacked}_i$ is a dummy variable indicating whether settlement 
$i$ was attacked, 
$\text{Post}_t$ is a dummy variable equal to 1 if 
$t=2010$ (post-treatment period) and 0 if 
$t=2006$ (pre-treatment period), 
$\text{Attacked}_i \times \text{Post}_t$ is the difference-in-differences interaction term to estimate the average treatment effect on the treated (ATET), and 
$\varepsilon_{it}$ is the error term.
The causal effect of the attacks on subsequent election outcomes is identified by the coefficient 
$\beta_3$ if the parallel trend assumption is met, that is, if Jobbik support in attacked villages would have evolved over a similar path compared to the rest of the country in the absence of the attacks. A key challenge to this assumption is selection: terrorists choose locations with rising frictions between the Roma and the non-Roma population, we expect counterfactual Jobbik support to be also higher in the attacked group than in the control group. This would cause an upward bias in our estimate 
$\hat{\beta}_3$.
To address this challenge, we rely on different control groups that could plausibly be used to estimate counterfactual changes in attacked villages in the absence of the attacks. Specifically, we first rely on a design-based control group which includes settlements where the terrorists were later—during the court phase—revealed to have planned further attacks, but had no opportunity to carry them out due to their capture.Footnote 3 The advantage of this comparison is that they were arguably selected on the same unobservables as the attacked variables. The disadvantage is that there are only four such settlements.
Second, we also create a synthetic control equivalent for each of the 9 towns and villages, and the control group of the regression is the synthetically generated control settlements. We use the set of similar settlements as the donor pool. The synthetic control settlements arguably have the socio-economic fundamentals closest to the attacked settlements. A further advantage of the synthetic control approach is that it also allows us to assess the impact of each terror attack separately. We use these settlement level estimates to provide further evidence of the validity of our assumptions.
Finally, we also present baseline estimates from a broad control group that includes every settlement in Hungary and a restricted control group that includes settlements that had at most 10.000 inhabitants at the time and had a Roma population (the restricted control group also serves as the donor pool for the synthetic control analysis). In these specifications, we also control for a vector of covariates to further mitigate selection. This vector includes the share of the Roma population according to the most recent (2001) census, and indicators of economic distress: the settlement level unemployment rate, the number of crimes reported, and the number of inhabitants with a criminal record.
2.3. Data
We rely on data from several sources. Election data come from the National Election Commission’s website (National Election Commission, 2024). Settlement-level variables come from the TEIR system collected by the Hungarian Central Statistical Office, Hungarian State Treasury, National Tax and Customs Administration, and the Ministry for Home Affairs and their legal successors.(Central Statistical Office (Központi Statisztikai Hivatal, 2018)Footnote 4. The number of citizens identifying as Roma comes from the 2001 Census (Central Statistical Office (Központi Statisztikai Hivatal, 2001).Footnote 5 Finally, we also include the number of times the settlement was mentioned on the far-right news portal kuruc.info’s “gypsy crime” column in 2008 before the first attack took place. We scraped all posts from the far-right kuruc.info news site’s “gypsy crime” column between 2006 and 2018 to gain a settlement-level contemporaneous indicator of tensions between the Roma and the non-Roma.Footnote 6
Table OA3 in the Appendix shows the demographic and economic profile of attacked settlements and different sets of control settlements. We list the 2006 values of all variables that were used to construct the synthetic controls. We show the mean of the variables for the attacked settlements (column 1), the rest of the country (column 3), the donor pool (settlements with at most 10.000 inhabitants, of which at least 1 identified as Roma in the 2001 census, column 5), and the average of those settlements that were eventually used for generating the synthetic control settlements. Figure OA4 in the Appendix shows the geographical location of attacked, planned-attack villages, and similar settlements. These units will also serve as the donor pool for our synthetic control method.
3. Results
We estimate Equation 1 using the first-difference estimator and present our regression results in Table 1.Footnote 7 Depending on the specification, settlements where attacks took place saw an approximately 9–14 percentage points larger increase in the vote share of Jobbik relative to the control group. The average baseline change in the control group is 15–21 percentage points. Thus, the estimated local impact of the attacks is substantial: the increase in Jobbik support is 53%–70% larger in attacked villages as it is in their respective control groups. Additional analyses—reported in the Online Appendix show that the attacks did not affect turnout; rather, they decreased the combined support of mainstream parties by an equal amount.
The impact of the attacks on Jobbik support in attacked villages
Table 1 Long description
Outcome is the change in extreme-right vote share from 2006 to 2010, measured in percentage points, estimated across four model specifications with different comparison groups and covariates. The attacked indicator is positive in every column: 9.276 in the whole-country sample and 8.779 in the donor-pool sample, both with strong statistical significance. In the planned-attack comparison, the estimate is 13.537 and remains statistically significant, though less precise. In the synthetic-control comparison, the estimate is 10.553 and strongly significant, with no covariates included. Sample sizes vary widely across columns, from 3,124 and 1,965 observations down to 13 and 18, so the smaller-sample estimates should be interpreted with more caution. Model fit is low in the large samples (about 0.01) but higher in the small-sample comparisons (about 0.68 and 0.41). Overall, results consistently indicate larger increases in extreme-right support in attacked settlements, with uncertainty depending on the specification and sample size.
Notes: Standard errors clustered at the micro-region level (175 clusters) in Columns 1 and 2; robust standard errors in Column 3. *: significant at 10%, **: significant at 5%, ***: significant at 1%.
Using data from later elections, we also test the persistence of our results. We regress the vote share of extreme right support across all elections within a 
$\pm$10-year time window, relative to the attacks on time dummies, their interactions with a dummy indicating the attacks, and a set of settlement fixed effects. The coefficients of the interactions capture how the difference in Jobbik support between treated and control settlements evolved over time. We plot these in Figure 1. In Panel A, the control group is the go donor pool; in Panel B, the control groups are the locations of the planned attacks. The results suggest that the effect of the attacks on the surge of Jobbik’s popular support was sustained even in 2014, though it dissipated by 2018.
The impact of the attacks on the far right’s vote share over time.
Figure 1 Long description
Multi-line graph with two panels comparing differences in extreme right vote share. X-axis: Election year (98 to 18). Y-axis: Difference in extreme right share in valid votes (negative 8 percent to 20 percent). Panel A: Attacked 9 villages vs. donor pool. Single line with circular markers and error bars. Key points: near zero from 98 to 06, sharp increase peaking around 9 percent in 10, decline by 14. Panel B: Attacked vs. planned. Single line with circular markers and error bars. Key points: near zero from 98 to 06, sharp increase peaking around 12 percent in 10, decline by 14. Both panels show vertical dashed lines around 09 and 10, indicating reference years. The trend shows a significant rise in vote share differences around 09 to 10, followed by a decrease by 14 to 18.
The same analysis also provides additional evidence that the parallel trend assumption underlying our empirical analysis holds: differences between attacked and control settlements stayed small and insignificant in the period leading to the treatment.Footnote 8 Our estimates based on the synthetic control method also allow the exploration of how treatment effects vary across individual attacks. In the Online Appendix, we show that the magnitude of individual effects is largely consistent with anecdotal evidence on each individual incident. Specifically, the estimated effects are substantially smaller in the cases where the attacks were unsuccessful from the terrorists’ perspective.Footnote 9
So far, our analysis assumed that the local impact of terror was completely concentrated in villages where the attacks took place. This assumption makes our estimates conservative in the sense that if the impact of the terror attacks spilled over to nearby villages, then the control groups used for our estimates are partially contaminated. To explore spatial spillovers directly, in Figure 2, we visualize the relationship between both pre- and post-treatment Jobbik support and the distance from villages where attacks took place (left panel) and where they were planned but not executed (right panel). Exploring these four relationships—i.e., pre vs. post treatment and planned vs. successful attacks—not only allows us to quantify the degree of spillover but also to construct placebo tests.
Far-right support as a function of distance from realized and planned attacks (invalid votes).
Figure 2 Long description
The left scatter plot shows the far right share in valid votes against the distance to realized attacks in kilometers. The horizontal axis ranges from 0 to 100 kilometers, while the vertical axis represents the percentage of far right share in valid votes, ranging from 0 percent to 60 percent. Data points for the years 2006 and 2010 are plotted, with 2010 data showing a downward trend as distance increases. The right scatter plot displays the far right share in valid votes against the distance to planned attacks in kilometers. The axes are similarly labeled, with the horizontal axis ranging from 0 to 100 kilometers and the vertical axis from 0 percent to 60 percent. Data points for 2006 and 2010 are plotted, with 2010 data showing a slight upward trend. Both plots include a legend indicating the years represented by different symbols.
The key finding is that in 2010, distance from realized attacks became an influential predictor of variation in Jobbik support. The slope of the line is -.11 (s.e.:0.01), meaning that as one gets closer to attacked villages, Jobbik support is expected to rise by one percentage point in every 9 kilometers. On the contrary, the distance from future planned and realized attacks does not explain any variation in Jobbik support in 2006, and proximity to planned attacks does not predict increased Jobbik support in 2010 either. Taken together, this demonstrates that terrorist attacks committed against the Roma had a positive impact on far-right support in Hungary that spilled over regionally.
4. Conclusion
The key contribution of our study is to bring additional evidence to the table in the scholarly debate about the political impact of right-wing terrorism. The lives lost in ethnic violence surely reflect underlying tensions in divided societies, but it would be tempting to think that outbursts of violence against ethnic minorities limit the credibility of parties campaigning on ethno-nationalist platforms. Instead, using our unique research design, we find that right-wing terrorism in fact can strengthen the local support for these parties. These results contradict recent research that challenges the importance of physical proximity to such incidents (Böhmelt, Bove and Nussio, Reference Böhmelt, Bove and Nussio2020; Agerberg and Sohlberg, Reference Agerberg and Sohlberg2021).
We believe that our—relatively understudied setting—deserves attention on its own right, especially given the prevalence of anti-Roma hate crimes in Eastern Europe—including high-profile events such as the 2009 Vítkov arson attack in Czechia, or the 2011 anti-Roma protests in Bulgaria (Amnesty International, 2010; BBC News, 2011). However, several factors might limit the generalizability of our findings to other cases. First, prejudice against the Roma minority in Hungary is quite strong and deep-seated (Simonovits, Kezdi and Kardos, Reference Simonovits, Kezdi and Kardos2018) which might make it easier for the ethnic majority to “relativize the murders by juxtaposing them with criminal acts committed by Roma or by highlighting the putatively legitimate motivations of the perpetrators” (Szombati, Reference Szombati2018, p. xiv). At the same time, compared to some other cases with a history of ethnic violence and weaker institutions (Hager, Krakowski and Schaub, Reference Hager, Krakowski and Schaub2019), our results might understate the overall political impact of right-wing terrorism.
An important limitation of our research design is that it only permits the identification of the local impact of right-wing terrorism. As our identification strategy relies on the comparison of localities with and without an attack taking place, we can only speculate about the broader, nationwide impact of these events. Existing research highlights the central importance of media coverage in shaping how citizens make sense of terror attacks (Shoshani and Slone, Reference Shoshani and Slone2008; Böhmelt, Bove and Nussio, Reference Böhmelt, Bove and Nussio2020; Jacobs and van Spanje, Reference Jacobs and van Spanje2023), including the decision of media outlets to classify an incident as an act of terror (Huff and Kertzer, Reference Huff and Kertzer2018). Thus, it is quite possible that our analysis focusing on local effects understates the total impact of the attacks, especially since the nationwide uptake in the support for Jobbik coincided with the events we study and a corresponding media attention to ethnic tensions.
These lingering questions about both the scope conditions and causal mechanisms explaining our findings call for future research exploring more systematically the role of antecedent conditions on the impact of right-wing terror attacks. As more and more case-based research surfaces, meta-analyses could explore how the link between right-wing terrorism and far-right voting is shaped by inter-group relations, the strength of political institutions, as well as both the supply of and demand for news covering these incidents. In addition, to distinguish between the possible causal mechanisms that mediate the impact of terror attacks, future research could combine observational and experimental evidence to assess the relative importance of temporal and spatial proximity to acts of terror vis-à-vis exposure to various narratives of these events in shaping public attitudes.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2026.10111.
Competing interests
The views expressed in this article are those of the authors and do not necessarily reflect the views of their employers and affiliated institutions.
Open access
