Source-linked AI summary
The complex relationship between anti-immigrant sentiment and exposure in the Netherlands
Benedikt Meylahn, Tommaso Giommoni, Mike Lees, Alessandro Nai
TL;DR
The paper examines whether immigrant exposure is associated with more or less anti-immigrant sentiment, amid competing contact and threat theories. Using nationwide aggregate data and right-wing voting as a proxy, it finds a national contact-theory-consistent relationship that partially reverses across regions, with wealth shaping responses in highly exposed neighbourhoods.
Problem
Research has not reached consensus on whether greater immigrant exposure improves intergroup relations or increases threat, despite empirical support for both theories.
Method
The study uses nationwide aggregate data, a distance-sensitive exposure measure, and right-wing voting as a proxy for anti-immigrant sentiment across the Netherlands.
Results
Nationally, greater immigrant exposure aligns with contact theory, but the relationship partially breaks down across similarly urban regions; in highly exposed neighbourhoods, WOZ value is the only variable significant at p < 0.01 under all filters.
Takeaways & Limitations
Contact theory captures part, but not all, of the exposure–sentiment relationship, while wealth is central to voting responses in highly exposed neighbourhoods.
Takeaways & Limitations
Aggregated data risk ecological fallacy and may miss individual patterns, while 500-by-500-metre and larger geographic aggregation can obscure smaller-scale dynamics.
Abstract
from arXiv · showhide
We study the relationship between exposure to immigrants and right-wing votes used as a proxy for anti-immigrant sentiment in the Netherlands. Exposure to immigrants is measured using a novel indicator, adapted from the residential segregation literature, applied to grid data available for the entire country at the aggregate level. Using the proxy of election results, we are able to study the entire voting population instead of a limited subset of panel respondents. In our cross-sectional study of the 2023 national election, we find that municipalities and neighbourhoods with greater exposure have less anti-immigrant sentiment. However, looking at the increase in right-wing vote share from the 2021 to the 2023 national elections, we find evidence of an ``overexposure effect'', as this relationship reverses, and the areas that saw significant increases in migrants subsequently experienced an increase in right-wing votes. Furthermore, we see that at high levels of exposure, wealth (and inequality) plays an important role in the voting dynamics, indicating that relative deprivation theory may play a role.
Significance statement
The study examines how immigrant exposure relates to anti-immigrant sentiment in the Netherlands, using right-wing voting as a proxy. Nationally, the relationship follows contact theory, but regional analysis reveals reversals.
- The study uses the full Dutch voting population and right-wing votes as a proxy for anti-immigrant sentiment.
- Nationally, greater exposure to immigrants is associated with less anti-immigrant sentiment, consistent with contact theory.
- At the regional level, the exposure–sentiment relationship can reverse, suggesting contact theory alone does not explain the pattern.
- The study applies a distance-sensitive exposure measure to minority-group exposure and native-group sentiment.
1 Introduction
The introduction frames Dutch anti-immigrant voting as a contested relationship between contact and threat theories, and motivates a nationwide, distance-weighted analysis using election data. The study aims to reconcile apparently conflicting findings by examining both national and regional patterns, including exposure thresholds and local context.
- The 2023 election saw the PVV receive 23% of the vote, despite a reported decline in Dutch anti-immigration sentiment from 2017 to 2020.
- Contact theory predicts improved intergroup relations with exposure, whereas threat theory predicts reduced trust and cooperation as diversity increases.
- Prior reviews find support for both theories and suggest that results may depend on the analytical scale.
- Dutch studies have reported both decreased anti-immigrant sentiment after refugee exposure and increased right-wing voting near visible mosques.
- The paper uses election results to cover the entire Dutch voting population, while recognizing that regional aggregation limits individual-level inference.
- Its distance-weighted exposure measure, adapted from segregation research and applied to 500m by 500m grid data, distinguishes spatial mixing patterns beyond immigrant population shares.
- The study reports contact-consistent patterns up to a turning point, after which greater exposure is linked to larger increases in right-wing voting, especially in rural and poor regions.
2 Methods
The study measures immigrant exposure with a distance-sensitive indicator adapted to grid data, then aggregates exposure and population-weighted controls to Dutch municipalities and neighbourhoods. Election data and regional classifications support the analysis of right-wing voting across the Netherlands.
- 2.1 Exposure: Exposure Ei weights immigrant residents by distance from cell i across neighbouring layers until a total population of k is reached.The calculation uses immigrant and total cell populations, Euclidean distance, and a distance-sensitivity constant c; the final layer is completed when k is reached.
- 2.1 Exposure: Regional exposure is the mean exposure experienced by Dutch residents, weighted by the number of Dutch persons in each cell.The weighted sum covers all cells within the municipality or neighbourhood, and the resulting value is averaged for Dutch residents in region L.
- 2.1 Exposure: The exposure measure is considered robust to k because added individuals are increasingly distant and make diminishing contributions.Figure 2 compares exposure calculated for k values of 10, 100, 1 000, 5 000, and 10 000; the authors note that aggregated grid data may affect convergence.
- 2.2 Description of control variables: Control variables include population-weighted WOZ home value, address density, and voting-age demographic shares.WOZ value accounts for property-related wealth, address density proxies urbanisation, and age shares focus on residents aged ≥15; S65+ is omitted as the reference category.
- 2.3 Population weighted control variables: Address density and WOZ value are logarithmically transformed and centred after population weighting for regression models.Both variables are positive and skewed, motivating the transformation.
- 2 Methods: The analysis uses 500m-by-500m CBS grid data aggregated to simplified 2023 municipality and neighbourhood boundaries.The study also uses LISS politics-and-values data to classify right-wing anti-immigrant parties and COROP regions or provinces for fixed-effects comparisons.
3 Results
Across the Netherlands, greater immigrant exposure is associated with less right-wing voting, but this relationship varies by spatial scale, urbanisation, and exposure level. The 2021–2023 change analysis indicates a nonlinear pattern in which the association can reverse at high exposure.
- Spatial distribution: Municipalities with the greatest exposure average above 0.2, while the least exposed municipalities approach zero exposure, with concentration around major cities and eastern and southern borders.Exposure is therefore spatially uneven rather than uniformly distributed across the Netherlands.
- Spatial scale: At finer spatial resolution, the municipal inverse relationship becomes heterogeneous: neighbourhood-level patterns in high right-wing-support municipalities no longer show a discernible inverse association.Within highly exposed Amsterdam and Rotterdam areas, the most exposed neighbourhoods generally have lower right-wing vote shares, while low-exposure neighbourhoods show substantial variation.
- National relationship: Greater exposure is associated with less right-wing, anti-immigrant voting at both municipal and neighbourhood scales across model specifications.The negative association remains with and without provincial or COROP regional fixed effects, and survives the reported robustness checks.
- Economic context: Home value is consistently negatively associated with right-wing voting, but its interaction with exposure is inconsistent and nonsignificant, providing no evidence that wealth changes exposure’s relationship with voting.The paper distinguishes the direct association of home value from its unsupported interaction with exposure.
- Urbanisation: Urbanisation moderates exposure’s association with right-wing voting: increased exposure in dense areas has a stronger negative relationship, whereas exposure in low-density areas is associated with more right-wing voting.The exposure-by-address-density interaction is statistically significant across the reported interaction specifications.
- Change from 2021 to 2023: The change in right-wing vote share from 2021 to 2023 has a significant negative linear exposure term and a significant positive quadratic term, indicating a reversal at sufficiently high exposure.Most areas experienced a relatively consistent national increase, while the quadratic specification fit slightly better than the linear specification.
4 Discussion
The discussion identifies contact, threat, and visibility-related mechanisms that may jointly shape the relationship between immigrant exposure and right-wing voting. It also emphasizes the study’s aggregate-election-data scope and limits, including unmeasured perceived and lifetime exposure.
- Mechanisms: Greater exposure is consistently associated with less right-wing anti-immigrant voting, aligning with contact theory.The authors interpret meaningful contact as potentially increasing trust in newcomers, while noting that contact opportunities do not always produce warmer out-group attitudes.
- Mechanisms: At high exposure levels, the relationship reverses, with poorer regions showing stronger right-wing voting increases.The discussion links this turning point to possible threat dynamics and economic hardship, while reporting that lower WOZ values correspond to greater right-wing voting.
- Mechanisms: Visible, acute housing arrangements may overshadow opportunities for contact and coincide with sharp local increases in right-wing voting.Brouwhuis and IJmuiden-Zuid had the largest increases; IJmuiden hosted roughly 1,000 asylum seekers on a converted cruise ship.
- Contribution and scope: Using election data broadens coverage beyond survey respondents but requires aggregate measures that risk ecological fallacy and can miss opposing individual responses.The authors frame this as exchanging causal identification for nationwide range and recent-election coverage.
- Policy implications: The authors suggest communicating with locals and creating opportunities for meaningful involvement when housing newcomers.They caution that temporary housing conversions can sharply increase exposure, particularly in regions facing housing shortages and low wealth.
- Limitations: The analysis cannot account for perceived exposure or lifetime exposure, and its 500-by-500-metre aggregation may miss smaller-scale dynamics.The proxy may also capture voting motivations other than anti-immigrant sentiment.
A Non-colinearity of control variables
The regressors show no problematic collinearity, with the strongest correlations remaining interpretable and within reasonable bounds. Address density and WOZ home value are log-transformed and centred for modelling.
- The greatest municipal anti-correlation is -0.73 between S25,45 and S45,65, while the neighbourhood maximum is -0.67 between S45,65 and address density.These relationships are considered sensible because older residents appear less concentrated in high-density urban areas.
- The correlation matrices report regressors for both municipality and neighbourhood scales, including address density, home value, and the share receiving benefits.
- None of the variance inflation factors for model specifications M2 and M6 is high enough to require correction.
- Address density and WOZ home value are log-transformed and centred because both variables are positive and skewed.The transformation is applied after population weighting and illustrated in Figure 10.
B Robustness checks
The robustness-check section illustrates the transformation applied to address density and WOZ home value before modelling.
- Figure 10 depicts the transformation applied to address density and WOZ home value.
B.1 Different value of k
The exposure results are robust to using k values of 100, 1 000, and 5 000. The authors caution that this stability may reflect the aggregated grid structure of the analysis.
- All coefficients retain the same sign across exposure calculations using k ∈{100, 1 000, 5 000}.The exposure coefficients are also identical when rounded to the first digit after the decimal.
- The choice of k does not strongly influence the analysis results, although this may be related to the aggregated, grid-based design.With precise resident locations, convergence of the exposure measure may occur only at higher k values.
B.2 Historic municipality voting as neighbourhood control
The neighbourhood-level relationship between exposure and right-wing voting remains negative and significant after controlling for historical municipality voting. Its magnitude is roughly halved, while the 2021 and 2023 CBS grid data are not directly comparable for some variables.
- The 2021 CBS grid data are not comparable to the 2023 data because heritage reporting groups were redefined without a clear mapping.The categories changed from western/non-western heritage to country of birth inside/outside the European Union.
- The robustness check adds the municipality’s 2021 voting outcome as a control for local ideology and voting behaviour in neighbourhood models.
- Table 6 presents neighbourhood-scale regression results for exposure computed with k ∈{100, 1 000, 5 000}, including insignificant coefficients.
- The negative, significant relationship between exposure and right-wing voting remains after controlling for historical regional voting.This robustness result holds when provincial fixed effects or COROP-region fixed effects are included.
- The exposure relationship roughly halves in magnitude compared with the original models M2, M4, and M6 in Table 1.
B.3 Robustness against definition of right-wing
The right-wing proxy is narrowed to PVV votes as a robustness check, and the regression results remain robust under this alternative definition.
- PVV votes alone provide an alternative proxy for anti-immigration sentiment because PVV consistently receives more votes than the other right-wing parties.
- The regression results are robust when anti-immigration sentiment is measured using only PVV votes.
B.4 Removing Ter Apel
Excluding Ter Apel and its surrounding regions preserves the main exposure–voting pattern and slightly strengthens the neighbourhood-level negative exposure coefficient. Additional analyses examine nonlinear exposure and population-size differences.
- B.4 Removing Ter Apel: The robustness analysis excludes both the entire province of Groningen and the municipality of Westerwolde before re-estimating the main specification.
- B.4 Removing Ter Apel: The main results remain robust after excluding Groningen and Westerwolde, including Ter Apel.The excluded areas comprise 112 neighbourhoods in Groningen and 8 in Westerwolde.
- B.4 Removing Ter Apel: The neighbourhood-level exposure coefficient becomes more negative after removing Ter Apel from the models.The adjusted R-squared also improves slightly, suggesting Ter Apel contradicts the broader negative exposure–right-wing voting pattern.
- Nonlinear exposure: The quadratic exposure specification tests whether the exposure relationship weakens or reverses at high exposure levels.The quadratic exposure term is significant in all specifications, but its negative sign does not support a weakening or reversal.
- Population size: Population-size analyses assess whether exposure effects vary across measurement units, motivated by prior evidence that contact and threat patterns differ by unit size.The models use province or COROP fixed effects depending on the specification.
- Exposure measurement: The paper attributes possible differences from earlier findings partly to its more spatially sensitive exposure measure.Conventional measures capture group shares within units but not the spatial distribution of the group across scales.
C.3 The role of unemployment
Benefit-recipient prevalence is positively associated with right-wing voting, while controlling for it strengthens rather than attenuates the negative exposure–voting relationship. Its interaction with exposure is strongly negative.
- Measurement: Benefit counts are transformed into the share of the working-age population receiving unemployment, assistance, or disability benefits.The denominator is the population aged 15–65.
- C.3 The role of unemployment: Benefit-recipient prevalence is positively associated with right-wing voting when included as a control.
- C.3 The role of unemployment: Including benefit prevalence strengthens the exposure–right-wing voting relationship rather than weakening it.
- Model specification: The analysis estimates benefit prevalence both as a control and as an interaction variable with exposure.The corresponding coefficients are reported in models M15 and M16.
- C.3 The role of unemployment: The exposure-by-benefit interaction is strongly negative in model M16.This contrasts with the expectation that immigrant-related resource competition would intensify threat among economically disadvantaged groups.
D.2 Responses to selected questions per voter group
Survey responses differ consistently across Dutch voter groups in their views on immigration, with PVV, FvD, and JA21 voters more opposed than central and left-wing voters. Agreement with one anti-immigration statement also rises slightly over time.
- D.2 Responses to selected questions per voter group: PVV, FvD, and JA21 voters consistently express more anti-immigration opinions than CDA and VVD voters.GL, PvdA, later GL-PvdA, and D66 voters are more favorable toward immigration.
- D.2 Responses to selected questions per voter group: Responses to the statement that there are too many people of foreign origin or descent clearly separate left, centre, and right party clusters.
- D.2 Responses to selected questions per voter group: Agreement with that statement increases slightly among left-wing parties and in the total sample over time.Figure 13 reports means and 95% confidence intervals for responses from 2016–2025.
- Proxy validation: The authors use the alignment between survey attitudes and voting outcomes from 2021–2023 to justify right-wing vote share as a proxy for anti-immigration sentiment.
- Data: The analysis combines LISS survey data with aggregate election and demographic data from publicly available sources.The data and code are available through the listed repositories and websites.