1. Introduction
Humanitarian admissions account for an increasing proportion of immigrant flows to developed democracies. This trend is particularly pronounced in Europe, where countries have grappled with successive refugee waves following the Syrian civil war and the invasion of Ukraine. Despite its political prominence, the electoral consequences of admitting refugees and asylum seekers remain unclear. Although group threat theory suggests that exposure to refugees will fuel support for far-right parties (Landmann et al., Reference Landmann, Gaschler and Rohmann2019), the empirical relationship is mixed (Cools et al., Reference Cools, Finseraas and Rogeberg2021), with studies documenting positive (Dinas et al., Reference Dinas, Matakos, Xefteris and Hangartner2019; Bratti et al., Reference Bratti, Deiana, Havari, Mazzarella and Meroni2020; Gessler et al., Reference Gessler, Tóth and Wachs2021; Bredtmann, Reference Bredtmann2022; Endrich, Reference Endrich2024), negative (Gamalerio et al., Reference Gamalerio, Luca, Romarri and Viskanic2021; Steinmayr, Reference Steinmayr2021; Vertier et al., Reference Vertier, Viskanic and Gamalerio2023), and null (Hennig, Reference Hennig2021; Schaub et al., Reference Schaub, Gereke and Baldassarri2021; Pettrachin et al., Reference Pettrachin, Gabrielli, Kim, Ludwig-Dehm and Pötzschke2022; Fremerey et al., Reference Fremerey, Hörnig and Schaffner2024) effects.Footnote 1
What explains these divergent results? Previous research shows that (i) exposure duration and (ii) pre-existing ethnic heterogeneity moderate the effect of refugee exposure on voting behavior. Short-term exposure—i.e. in regions through which refugees transit—has been robustly linked to hostility towards immigrants and support for far-right parties (Dinas et al., Reference Dinas, Matakos, Xefteris and Hangartner2019; Hangartner et al., Reference Hangartner, Dinas, Marbach, Matakos and Xefteris2019; Gessler et al., Reference Gessler, Tóth and Wachs2021). In contrast, longer-term exposure may create conditions for positive contact between refugees and citizens.
The ameliorating effects of positive contact may be more likely in regions that previously experienced immigration and are thus less primed to view increases in ethnic diversity as threatening (Newman and Velez, Reference Newman and Velez2014). Consistent with this hypothesis, some studies have documented a shift toward right-wing parties within rural areas unaccustomed to large inflows (Dustmann et al., Reference Dustmann, Vasiljeva and Damm2019). However, the moderating effect of prior diversity remains contested, with other studies finding no evidence for polarization in rural regions (Schaub et al., Reference Schaub, Gereke and Baldassarri2021; Bredtmann, Reference Bredtmann2022).
In this research note, we focus on an additional source of heterogeneity: the spatial distance between voters and refugees. To date, research has typically identified the electoral effects of refugee exposure by comparing administrative units (e.g., municipalities or counties) that host refugees to those that do not. This approach is sound as long as the unit size reflects the geographic level at which voters react to refugee settlement. However, it is plausible that political reactions to refugees may be more localized. The nationalization of media implies many voters have similar exposure to the political rhetoric surrounding refugees (Hopkins, Reference Hopkins2011). In this environment, electorally-relevant responses may not be a function of whether refugees are hosted in an administrative unit, but rather whether one is directly exposed to the potential negative externalities of hosting refugees, such as disrupted access to amenities, negative impacts on real estate valuations (Hennig, Reference Hennig2021), or exposure to protests and elevated police presence (Marbach and Ropers, Reference Marbach and Ropers2018). If the effects of refugee exposure are highly localized, using larger administrative units to operationalize proximity may effectively mask political responses by aggregating voters with different levels of exposure. As a result, aggregated analyses may yield null or even negative effect estimates when a more granular analysis would reveal substantial opposition at the local level.
To estimate how electoral responses vary as a function of spatial distance to refugee housing, we focus on Berlin, which experienced an unprecedented influx of refugees starting in 2015. Our analysis combines multiple data sources to precisely measure the distance between voters and refugee housing. First, we collected the street addresses of all refugee housing facilities. We combine this information with 100x100 meter census grid data collected prior to the refugee influx, as well as precinct-level returns for federal elections.Footnote 2
Our study is not the first to assess the electoral consequences of refugee exposure at the precinct level (Hennig, Reference Hennig2021; Pettrachin et al., Reference Pettrachin, Gabrielli, Kim, Ludwig-Dehm and Pötzschke2022; Endrich, Reference Endrich2024; Fremerey et al., Reference Fremerey, Hörnig and Schaffner2024).Footnote 3 Indeed, two recent studies have also examined Berlin (Hennig, Reference Hennig2021; Pettrachin et al., Reference Pettrachin, Gabrielli, Kim, Ludwig-Dehm and Pötzschke2022), finding null effects.Footnote 4 However, our approach improves upon prior designs in three important respects.
First, previous studies at the precinct level do not account for unit instability across elections. In Berlin, the city government redraws a subset of precincts for each election, without maintaining consistent identifiers. Between the 2013 and 2017 federal elections, for instance, 50 percent of in-person precincts and 67 percent of mail-in precincts were altered (see SM, H.2). Panel studies that overlook this issue, such as Hennig (Reference Hennig2021), risk biasing results towards zero due to precinct reshuffling.
Second, studies that acknowledge unit instability typically address the issue by implementing cross-sectional designs (Pettrachin et al., Reference Pettrachin, Gabrielli, Kim, Ludwig-Dehm and Pötzschke2022; Endrich, Reference Endrich2024; Fremerey et al., Reference Fremerey, Hörnig and Schaffner2024). While this approach will return unbiased estimates if refugee housing is as-if randomly assigned, our analysis of census data demonstrates that housing locations are strongly correlated with resident demographics, which can confound the observed relationship between exposure and electoral behavior. For example, the fact that housing was concentrated in districts with a higher share of foreigners could attenuate electoral backlash, since these districts may be less responsive to changes in local diversity.Footnote 5
Finally, since mail-in and in-person precinct boundaries differ, previous studies often exclude mail-in votes entirely (e.g. Endrich, Reference Endrich2024; Fremerey et al., Reference Fremerey, Hörnig and Schaffner2024). However, this coding decision introduces bias if refugee exposure affects the likelihood of voting by mail. In Berlin, where 33.4% voted by mail in 2017, we find that mail-in voting is related to both the treatment (refugee exposure) and the outcome (see Figure H.4 and Table H.2 in the SM), implying that omitting mail-in ballots would create a non-representative sample that does not accurately reflect support for far-right parties.
To address these issues, we harmonized precinct boundaries for in-person and mail-in votes across elections using GIS, creating stable synthetic precincts (see SM, section I). This approach preserves all votes, regardless of mode, while retaining a significant level of disaggregation. Relative to weighting, it also requires minimal assumptions about the spatial distribution of voters. Next, we calculated the pairwise distance between the population-weighted centroid of each synthetic precinct to each refugee housing facility. After evaluating the parallel trends assumption, we estimate difference-in-difference specifications with varying distance cutoffs to obtain causal estimates for the effect of local exposure to refugee housing on far-right voting in the 2017 federal elections.
Our results diverge from recent studies by demonstrating that proximity to refugee housing does, in fact, benefit the far right. In Berlin, support for the AfD increased from 4.9% to 12% in the aftermath of the refugee crisis. Consistent with group threat theory, we observe that this increase was particularly pronounced among voters living close to refugee accommodations. Specifically, we find that AfD support is, on average, 2.6 percentage points higher in precincts immediately surrounding refugee housing facilities compared to more distant precincts. However, since a relatively small share of voters lived close to refugee housing, the aggregate effect of local refugee exposure on the AfD’s support was modest. Although this implies that most of the AfD’s electoral gains between 2013 and 2017 were likely driven by sociotropic concerns, our findings nuance the literature by underscoring that group threat remains active at the local level, albeit at a smaller geographical scale than is detectable within most research designs.
2. Data and empirical strategy
Our main outcome is support for the far-right party Alternative für Deutschland (AfD) in Berlin, across the 2013 and 2017 federal elections. We focus on the 2017 federal election because (1) federal elections are the most salient elections in the German context, and (2) they are proximate to the refugee influx of 2015/2016.
Precinct boundaries vary extensively over time (see SM, Figure H.1). To generate a panel of comparable precincts, we aggregated to the mail-in level and then amalgamated units with changed borders. This approach created 341 synthetic precincts that remain unchanged across federal elections (see SM, section H.2 for more details). To measure local exposure to refugees, we geocoded the street addresses of all housing facilities established since January 2014 (see SM, section H.1) and measured the pairwise distance in meters between each housing facility and the population-weighted centroid of each precinct.
We begin by evaluating whether the placement of refugee housing facilities is correlated with local demographics. To do so, we draw on covariates measured at the 100 
$\times$ 100m grid level from the 2011 census. We regress each covariate on the presence of refugee housing at varying distance cutoffs from the grid centroid. The results, presented in Figure 1, demonstrate that despite the emergency situation, the placement of housing facilities was clearly non-random. Specifically, housing was more likely to be placed in proximity to diverse neighborhoods with a smaller share of German citizens and a higher share of younger residents. The magnitude of imbalance is sizable, at about 0.4 standard deviations. Although this pattern may be a function of the relative availability of vacant buildings, it nevertheless implies that housing facilities were concentrated in particular types of precincts. As a result, a direct cross-sectional comparison of treated to untreated precincts likely yields biased estimates since (1) a higher proportion of residents in affected precincts are unable to cast ballots in federal elections and (2) senior citizens are substantially less likely to support the far-right AfD compared to younger voters (Kobold and Schmiedel, Reference Kobold and Schmiedel2018).
Figure 1. Demographic balance: proximity to refugee housing.
Because housing facilities are not as-if randomly assigned, we cannot identify the causal effect on voting behavior using cross-sectional data. To address this issue, we implement a panel design and estimate difference-in-differences specifications of the following form:
\begin{align*}
\Delta Y_{i,d} = \alpha_d + \tau T_i^{k} + \epsilon_{i,d},
\end{align*} where 
$\Delta Y_{i,d}$ is the change in AfD vote share between 2013 and 2017 for unit 
$i$ nested in city district (Bezirk) 
$d$. Because the placement of housing facilities is coordinated between the city and district governments (see SM, section A), we include district fixed effects in all models. We thus compare over-time changes in AfD vote share, as a function of distance from housing, for precincts within the same district. To examine the treatment effect of refugee exposure, we estimate a series of regression models where we define the binary treatment indicator 
$T_i^{k}$ using varying distance calipers 
$k$. For a distance caliper of 150 meters, for example, 
$T_i^{150} = 1$ for all electoral units that experienced the opening of at least one housing facility within 150 meters of their respective population-weighted centroid between the 2013 and 2017 elections.
We vary 
$k$ between 150 and 1000 meters to examine how the estimated treatment effect 
$\hat{\tau}$ varies as a function of how granular we measure local exposure to refugee housing. This approach is also robust to the possibility of a non-linear relationship.Footnote 6
In contrast to previous cross-sectional work, our empirical strategy does not require as-if random placement of refugee housing to obtain causal estimates. Instead, the key identification assumption in our difference-in-differences design is parallel trends in voting behavior between treated and control precincts located within the same city district (Bezirk) in the absence of local exposure. The best evidence in support of this assumption would be to demonstrate that our treatment variable does not predict changes in the outcome variable (AfD support) prior to the refugee influx, i.e., between the 2009 and 2013 federal elections. However, because the AfD first ran in 2013, we cannot implement this test in our setting. Instead, we conduct two alternative tests to provide supporting evidence for our identification assumptions.
First, we show that electoral outcomes closely associated with support for the AfD evolved in parallel in treated and control districts between 2009 and 2013. Specifically, we focus on support for (i) the center-right CDU, (ii) the combined vote share of the liberal FDP (the party from which the AfD gained the largest number of votes in 2013, see Tagesschau (2013)) and the AfD, (iii) the far-right parties National Democratic Party (NPD) and the German Republican Party (REP)Footnote 7, and (iv) turnout. We present these results in SM Figure D.1. Within narrow distance bandwidths, we do not find statistically significant differences in pre-treatment trends for any of these outcome variables. We interpret these findings as partial evidence in support of parallel trends. Second, we conduct a spatial randomization test by randomly placing counterfactual housing units within the city boundaries and recalculating pairwise distances (SM, Figure G.1). The test strongly suggests that our findings are unlikely to be driven by random chance (
$p \lt 0.04$).
As with all temporal designs that rely on aggregated voting data, the specification assumes that the population of voters within precincts is comparable across time periods. While no censuses were conducted during the period we examine, we draw on data collected by the Berlin city government to assess within-city migration (SM, Figure B.1). If widespread, such migration would bias our estimates downwards. We find that migration rates were secularly declining and did not significantly change during the refugee crisis. Similarly, we observe no statistically significant difference in the number of eligible voters within treated districts (SM, Table G.5), as well as no mediating effect of the precinct’s prior political orientation. In other words, we do not find evidence that selective outmigration in response to the refugee influx affected electoral trends in our setting.
3. Results
The main results, shown in Figure 2, indicate that local exposure to refugee housing substantially increases electoral support for the far-right AfD. However, these effects only materialize within a narrow radius around refugee housing facilities. We find that the placement of housing within a distance of 150m from a precinct’s population-weighted centroid increases AfD support by about 2.6 percentage points (95% CI [0.05, 5.1]). This is a substantively large effect given the overall performance of the AfD in the 2017 elections (12%), representing an increase of approximately half of a standard deviation. While our aggregate data does not allow us to precisely trace vote switching at the individual level, point estimates (SM, Figure F.1) suggest that the gains for AfD are predominantly at the expense of the CDU/CSU, as well as Die Linke, a populist left-wing party.Footnote 8 Notably, however, this effect fades rapidly. At a distance of 250m, the placement of a refugee housing facility has no observable effect on far-right vote share.
Figure 2. Effect of refugee exposure on AfD voting.
These results are consistent across a series of robustness checks. First, rather than using a binary treatment indicator, we use a continuous measure that counts the number of housing facilities, and obtain similar findings (SM, Figure E.1). Second, to address the possibility that the results are being driven by atypical precincts, we weight the main specification by population and land area, respectively (SM, Table G.2). We find that either approach strengthens the findings. Likewise, when we decompose the effect between former East and West Berlin, we find similar patterns in each region (SM, Table G.1). Finally, while our difference-in-differences approach should reduce concerns about spatial autocorrelation in levels, we show that our results are robust to using Conley standard errors (SM, Table G.6).
Precincts may have experienced differential electoral trajectories between 2013 and 2017 not because of the placement of refugee housing but because of differences in sociodemographic structure. Specifically, the level-differences in covariates shown in Figure 1 may interact with the increased appeal of the AfD after 2013 and its programmatic emphasis on immigration. To address this concern, we estimate a series of additional specifications that allow for varying electoral trends as a function of covariates. We present the results in SM, Tables G.3 and G.4. We find that the estimates remain substantively significant across a range of specifications. This suggests that the differential increase in AfD support in treated precincts is not reducible to covariate imbalance between treated and control precincts.
4. Discussion
Following the Syrian refugee protection crisis, far-right parties in Europe experienced a significant surge in popularity. Research suggests that this support can be largely attributed to increased national-level concerns about immigration (Halikiopoulou and Vlandas, Reference Halikiopoulou and Vlandas2020; Van der Brug and Harteveld, Reference der Brug, Wouter and Harteveld2021; Gessler and Hunger, Reference Gessler and Hunger2022). Our results complement this body of research by demonstrating that direct, localized exposure to refugees nevertheless played a complementary role. Examining Berlin, where support for the AfD surged after the refugee crisis, we find that precincts located in close proximity to refugee housing had elevated support for the far-right Alternative for Germany (AfD). This backlash was not ephemeral, but rather persisted across the two years separating the start of the refugee crisis from the 2017 federal elections.
These findings have several implications. Evaluating a series of mixed and null results, recent literature has converged on the consensus that citizens did not meaningfully alter their voting behavior in response to local exposure to the refugee crisis. The absence of significant backlash appears to contradict group threat theory, which anticipates that a rapid increase in the number of arrivals would provoke localized backlash. Although a null result is encouraging in that it suggests that the public can successfully accommodate and adapt to demographic change, our findings suggest that this conclusion may be overly optimistic: voters do shift towards the far right in response to local refugee arrivals. However, we find that the geographic scope of electoral backlash is quite narrow, implying limited aggregate effects. Our results thus nuance prior findings: while voters do react negatively, as expected by group threat theory, they do so at a smaller scale than is detectable within most studies.
Although the data do not allow us to concretely identify the mechanism behind increased local support for the AfD, the effects are plausibly driven by neighborhood-level impacts. Although Berlin’s public services may have been affected by the refugee influx, we find no effects when aggregating to city districts, which control funding and service delivery. Rather, we observe a shift toward the far right only among voters living in close proximity to refugee housing. Since selective outmigration was uncommon (SM, Table G.5), we propose two complementary mechanisms. First, the temporary disruption caused by refugee housing may have had a negative impact on residents’ perceived safety, privacy, or community cohesion, leading them to differentially support the AfD as a form of protest voting. Second, since affected precincts also experienced a decline in electoral turnout (SM, Figure F.1), it is possible that the presence of refugees temporarily reduced levels of trust or political engagement, consistent with other research on diverse residential settings (Bellettini et al., Reference Bellettini, Ceroni and Monfardini2016; Förster, Reference Förster2018). In turn, increased rates of abstention among non-AfD voters may have contributed to the comparative success of the far-right.
Beyond demonstrating a localized effect of refugee presence on far-right vote share, our analysis also provides a methodological contribution by showing that the placement of housing facilities in Berlin was demonstrably non-random. The imbalances in demographic characteristics we observe suggest that researchers should be cautious in assuming that local exposure to refugees constitutes a natural experiment, even within emergency settings. Accordingly, researchers studying the effects of refugee exposure should either rely on an exogenous source of variation (Dinas et al., Reference Dinas, Matakos, Xefteris and Hangartner2019; Hangartner et al., Reference Hangartner, Dinas, Marbach, Matakos and Xefteris2019; Bratti et al., Reference Bratti, Deiana, Havari, Mazzarella and Meroni2020) or leverage panel data to control for differences between units.
Finally, our findings imply policy trade-offs: strategically locating refugee housing in less populated areas may further reduce localized xenophobic backlash, yet potentially hinder refugee integration (Hilbig and Riaz, Reference Hilbig and Riaz2022). While previous research has typically examined the effects of dispersal policies on either refugees or citizens, future research should adopt a more holistic approach to examine potential trade-offs across both domains.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.10061. To obtain replication material for this article, https://doi.org/10.7910/DVN/SKXJDN
Open access
