Perceptions of migration and integration in Germany: The role of regional context and migrants’ origin

Abstract

A substantial body of evidence documents misperceptions among natives regarding the number of migrants and their integration. Using two experimental studies, we show that these misperceptions are context-dependent rather than universal. The first study examines whether the accuracy of natives’ perceptions varies with the spatial scale at which they are elicited. People consistently overestimate the share of migrants, and this misperception does not depend on scale. In contrast, perceptions of migrant integration improve at more localized levels: natives underestimate integration at the national scale but estimates become more accurate at the district and neighborhood levels. The second study shows that perceptions of labor market integration for both natives and migrants are generally inaccurate and vary across different migrant groups. Overall, our findings suggest that researchers studying (mis)perceptions about migrants should carefully consider the spatial context of their data. Misperceptions are more nuanced than previously recognized.

Significance statement.

People overestimate migrant population shares and underestimate migrants’ labor market integration. Such misperceptions fuel opposition to immigration and shape policy debates. Using two survey experiments in Germany, we show that geographic proximity increases the accuracy of one type of perception but not the other. Respondents overestimate the migrant population share by roughly the same margin whether asked about the country, the district, or the neighborhood. Perceptions of migrants’ unemployment, by contrast, become more accurate at the neighborhood level, where they track actual local conditions. This asymmetry is consistent with distinct cognitive processes underlying each judgment: proportion estimates are subject to a domain-general bias that proximity cannot correct, whereas inferences about employment are updated through direct local experience. Accuracy also improves when respondents evaluate specific origin groups rather than migrants in general.

Introduction

Over the past few decades, there has been a surge in migration to Western countries. Research across disciplines has examined the societal implications of the resulting increase in ethnic heterogeneity. Among others, research has explored how ethnic heterogeneity affects social cohesion (1–3), trust (4), support for the welfare state (5–7), and attitudes toward migrants and refugees (8–11).

The results in general remain inconclusive and are characterized as a “cacophony of empirical studies” (3). Nevertheless, in recent years, some important insights have emerged across these research areas which may help to explain the variation in findings. First, individual perceptions of ethnic heterogeneity might be more important predictors of trust, support for the welfare state, or attitudes toward migrants than actual facts (10, 12). Second, research shows the relevance of geographic proximity and supports the assumption that ethnic diversity is more consequential when out-group members are closer and more directly visible (4, 13, 14). Third, research has questioned whether migrants are perceived as one homogeneous group and whether attitudes toward migrants differ conditional on the origin, race, cultural background, level of education, and degree to which they have integrated (15–17).

This article studies how (mis)perceptions about migration vary with the geographical level under consideration and with immigrants’ country of origin. We build on recent literature showing that the presence of immigrants is overestimated and their economic integration is perceived as much worse than it is in reality (see Refs. (10, 18), among others). Correcting misperceptions about the size of the foreign-born population is not enough to change policy views, but adding information on the characteristics of migrants induces more support for immigration (19). Although misperceptions about immigration are extensively documented, little is known about how they are formed. Drawing on recent advances in social cognition, political science, and the study of immigration attitudes (see, e.g. Refs. (20–22), among others), we advance three arguments: first, perceptions of migrant population size and perceptions of migrants’ economic integration are governed by distinct cognitive processes and therefore differ in accuracy; second, geographic proximity improves perceptions of economic integration but does not reduce overestimation of migrant population shares; and third, the accuracy of perceptions varies with migrants’ country of origin.

Using two studies (henceforth study 1 and study 2), we show that the geographical level matters for (mis)perceptions about the labor market integration of migrants but not for estimates of their size (study 1). Additionally, we show that respondents’ perceptions of labor market integration vary depending on migrants’ country of origin (study 2).

In study 1, we rely on data from an online survey conducted with a nonprobability sample in Germany in 2020. We document that the level of regional aggregation is not a universal driver of (mis)perceptions about migrants. We asked respondents for an assessment of the size of immigrant groups and their labor market integration at three different regional levels: the national level, the district level,^a and the neighborhood level. Respondents’ assessments of the migrant population size differ only slightly at the different regional levels. However, respondents severely underestimate the economic integration of migrants in Germany at the national and district levels, while having, on average, very accurate perceptions of migrants’ labor market integration at the neighborhood level. We show that this result is likely driven by a greater responsiveness of perceptions about immigrants’ labor market integration to neighborhood-level conditions. Although respondents typically overestimate migrant unemployment, their perceptions vary systematically with the actual local unemployment rate of immigrants.

In study 2, we examine perceptions of migrant integration. We asked respondents about the unemployment rates of the largest migrant origin-groups in Germany and also that of native Germans. By and large, respondents overestimate unemployment rates for most migrant origin-groups as well as for Germans. However, respondents’ perceptions are not uniformly biased but linked to the actual unemployment rates of migrant groups.

Our results advance our knowledge about how people form their perceptions with respect to outgroups. Perceptions about migrants’ characteristics—but not their size—seem to be influenced by the local exposure of individuals. We extend previous studies that examined perceptions at the national level (10, 18) or the regional level (23, 24) with an analysis of the neighborhood level. We further extend this line of research by investigating perceptions of migrant shares and of unemployment rates, building on recent insights that perceptions of migrants’ characteristics might be more influential for attitudes than perceptions of the size of these groups. Our contribution complements recent work (13, 25), which relates actual regional factors (e.g. local immigrant shares and unemployment rates) to beliefs about and attitudes toward immigrants. Our study adds to this research by directly eliciting and experimentally manipulating geographic reference levels, allowing us to isolate how spatial framing shapes perceptions and evaluations of migrants.

In addition, migrants are not perceived as a homogeneous group and (mis)perceptions about their economic integration are, to some extent, shaped by migrants’ region of origin—a finding which is often neglected when perceptions about migrants are analyzed. Our findings question the common practice of drawing policy conclusions from (mis)perceptions at the national level and for a homogeneous migrant group. Instead, we argue that the geographic scale of measurement and migrants’ region of origin should be taken into account when scholars call for a correction of (mis)perceptions about migrants—but that these factors operate differently depending on what is being estimated. Geographic proximity increases the accuracy of perceptions of migrants’ economic integration but does not reduce overestimation of their population share, suggesting that different types of misperception require different corrective strategies.

Theory

This article contributes to the strand of research examining how people form perceptions about their socio-economic environment. Studies have explored whether people have an accurate understanding of the state of the economy of a country, its unemployment rate (26), inflation (27, 28), or the share of migrants living in a country (14, 29). Knowledge and information about these facts come from personal experience and from the media or opinion leaders (30). If no specific knowledge of these facts is available, personal experience dominates (31). Furthermore, there exists a general tendency to overestimate the share of migrants (10, 18). Along these lines, Gorodzeisky and Semyonov (10) find that the perceived size of the migrant population is correlated with its actual size. In contrast, Herda (32) shows that the overestimation of migrant shares has grown rapidly from 2005 to 2013 in the United States, outpacing the actual modest increase. Previous studies analyzing perceptions about the size of specific ethnic minority populations in the United States find that the average perceptions of respondents are closer to the actual figures for these groups when the geographic scope of the response is given at the county level rather than the national level (23, 24). Moreover, studies have shown that the impact of migration on welfare state attitudes is more salient at the regional level than at the national or cross-national level (20, 33, 34).

Key theoretical arguments to understand attitudes toward migrants refer to contact (35) or to threat and competition (36, 37). While most studies on perceptions about migrants focus on the share of migrants in the population, this might only be an appropriate indicator for contact with migrants. However, it tells us little about the degree of competition with migrants or whether migrants are perceived as a threat. For example, Margalit and Solodoch (38) argue that the stock of migrants and inflow of migrants induce different views among natives, showing that natives are systematically more accepting of the former. In a similar vein, we argue that we should look at the share of migrants and measures for migrant integration, like the unemployment rate, with more nuance. While the share of migrants is primarily a measure of ethnic heterogeneity which leads to more contact, and possibly better intergroup relations (25), the unemployment rate of migrants is an indicator of migrants’ integration. It hence captures aspects of competition for jobs but also the extent to which migrants are perceived as a burden on the welfare system. We study both—perceptions about the share of migrants and perceptions about unemployment rates—and thus complement the existing literature.

Crucially, we argue that these two types of perception are subject to fundamentally different cognitive processes, which explains why geographic proximity should correct misperceptions about migrants’ economic integration but not misperceptions about their population share. Estimating the share of migrants in a population is a proportion judgment. Research in cognitive psychology shows that proportion judgments are prone to a well-documented, domain-general bias: people systematically rescale estimates of proportions toward more central prior expectations, resulting in consistent overestimation of smaller groups and underestimation of larger groups (21, 39). Because migrants are a numerical minority at every geographic level, this bias produces overestimation irrespective of proximity or attitudes toward outgroups. So, although antiminority attitudes are pervasive and might vary with the regional context, they are not the primary source of error in estimates of the size of minority groups (21). Local contact does not correct this error—it may even amplify it, since repeated encounters with outgroup members increase their availability in memory and hence their perceived prevalence (40).

Estimating the unemployment rate of migrants, by contrast, is a behavioral inference: it requires judging what members of a group do in a specific, observable domain. Research in social cognition shows that such inferences are far more amenable to revision through direct experience. When people interact with outgroup members, they retrieve specific remembered exemplars rather than relying on abstract category-level stereotypes (41). Exposure to individuating information updates beliefs about specific outgroup behaviors while leaving estimates of group size largely unaffected (42). Crucially, this updating process depends on direct experience and contact with migrants, e.g. that people observe whether their neighbors, colleagues, and local shopkeepers of migrant background are working. This makes it the kind of attribute that local contact can plausibly correct, unlike population proportions, which are not malleable in this way.

The information environment at each geographic level reinforces this asymmetry. At the national level, perceptions of migrants’ economic integration are shaped predominantly by mass media, which systematically frames immigration through a lens of labor market competition and welfare dependency (43). In the absence of direct personal experience, these nationally circulating narratives go uncontested, sustaining the stereotype of the “unemployed migrant” even when it is empirically inaccurate. At the neighborhood level, however, respondents possess sufficient experiential, exemplar-based knowledge to override this media-constructed image—but only for behavioral attributes like employment status, not for the structurally different task of proportion estimation. Studies of local economic contexts confirm that people have a sense of whether times are good or bad in their immediate environment, but only about a third of respondents are usually able to correctly state the national unemployment rate (see Ref. (31), among others). We expect a parallel pattern for migrants: perceptions of unemployment should be sensitive to local conditions and corrected by proximity, while perceptions of population share should be overestimated at all geographic levels (21).

This line of literature also argues that the extent to which misperceptions occur is shaped by individual characteristics affecting both the ability and willingness to form accurate perceptions. A common finding is that misperceptions are smaller among the highly educated and those interested in politics, whereas anti-immigrant attitudes, right-leaning political ideologies and lower household incomes are associated with greater misperceptions and a higher tendency to overestimate the share of migrants (18, 29, 30, 44–46).

A third aspect we study is whether (mis)perceptions about unemployment rates vary with migrants’ country of origin. Extensive research on attitudes toward migrants documents that natives’ attitudes toward migrants vary conditional on specific migrant characteristics, for example, cultural distance or their legal status (15–17, 47, 48). For example, Schmidt (49) shows that immigration from EU member states does not affect anti-immigrant attitudes, but rising immigration rates from the Global South and inflows of asylum-seekers both lead to stronger opposition to immigrants. Similarly, we would expect that misperceptions vary for different migrant groups and that misperceptions are smaller for migrant groups which are usually perceived as culturally closer and better integrated.

Comparing (mis)perceptions with regard to different migrant groups also provides us with the opportunity to compare differences in misperceptions with regard to the unemployment rate of natives. This will allow us to examine whether the over-estimation of migrants’ unemployment rates is something specific to perceptions of migrants or whether this over-estimation is rather a general bias which affects perceptions of natives in the same way—and has little to do with discriminatory behavior or beliefs that migrants are lazy and unwilling to work.

Survey, regional data, and empirical strategy

Survey: data collection and sample

The data for both studies come from a two-wave nonprobability sample of the population living in Germany, stratified by gender, age, and education. The surveys were conducted in December 2020 and April 2021 and administered by YouGov.

We sampled a total of $N=\mathrm{3,030}$ respondents in wave one. Two-thirds of this sample $N=\mathrm{2,012}$ are a quota sample designed to resemble the population in Germany aged 18 years and older. The study oversampled the foreign-born population for a different study ($N=\mathrm{1,018}$). We restrict our sample to respondents without migration background $(N=\mathrm{1,813})$. Table S1 compares distributions of gender, age, education, and migration background in this general population sample with official statistics. Our sample resembles the target population with regard to gender and age quite well. Low educated respondents and those with migration background are underrepresented in our sample. This is a common observation in general population surveys (50).

Since we restrict our sample to respondents without migration background, in a sensitivity check, we include those, i.e. born outside of Germany or with at least one parent who was not born in Germany, and the results remain virtually unchanged (Tables S3 and S5). In wave two, we were able to reinterview 1,550 out of 1,813 respondents which corresponds to a recall rate for the second wave of 85%. Table S2 presents summary statistics of wave 1 of our main sample, and Table S4 presents summary statistics of wave 2.

Study 1

In study 1, we use both waves of our online survey. In the first wave, we asked respondents about their perceptions of the share of migrants. In particular, we asked the following: “Per 100 people living [in Germany/in your district/in your neighborhood] how many do you think were born outside of Germany?.” Respondents could answer this question by choosing either one of six brackets (“0–5%,” “6–10%”, …, “46–50%” or “more than 50%”). This question was followed by another asking respondents to assess how accurate their estimate was: “How confident are you in your estimate of how many people in Germany were born outside of Germany?” with a five-point scale.

We randomly assigned respondents to three different groups. Each of the groups was asked about the share of migrants at a different geographical level, i.e. the national, the district, or neighborhood level. In doing so, we ensured that each respondent is treated only with one geographic unit. This procedure allows us to compare potential biases in perceptions across different geographical levels between individuals. By adopting a between-individuals instead of a within-individuals design, i.e. asking the same individuals several times about their perceptions of migrants at different geographical levels, we circumvent a potential learning mechanism that could lead to updated beliefs.

In the second wave, we repeated this procedure but instead asked about migrants’ labor market integration, i.e. about the unemployment rate of the foreign-born population, again at the three different geographic levels. We asked: “Per 100 people that were not born in Germany and that live [in Germany/in your district/in your neighborhood] how many are unemployed?.” Respondents could give their answers using the same six brackets as above and also received the same geographic reference as in wave 1.

In addition, we asked respondents for their ZIP code of residence. This information is key to being able to match the official numbers of migrant group sizes and their labor market integration levels to the answers of the respondents.

Study 2

For study 2, we rely on the question about respondents’ beliefs on the unemployment rates of migrants from different countries of origin. More specifically, we asked: “How many out of 100 people of working age [who live in Germany and have no migration background/who live in Germany/who live in Germany and were born in Turkey/who live in Germany and were born in Poland/who live in Germany and were born in Italy/who live in Germany and were born in Syria] do you think are unemployed?” with the same 5-percentage interval possible responses described above. Although the mean responses for the first two categories were fairly similar, we recoded the “who live in Germany” condition as missing and excluded it from the analysis to avoid ambiguity in the reference category. The analysis therefore focuses on clearly defined origin groups and comparisons involving individuals with a migration background. We randomized the country of origin so that each respondent only had to estimate the unemployment rate of one specific migrant group. Again, this between-individual design has the major advantage that beliefs are not updated and estimates are not anchored to each other.

The main question used in study 2 was asked in the first wave of our survey. The summary statistics of the respective sample are presented in Table S6.

We did not provide incentives for correct answers to these questions. Previous research suggests that incentives can increase respondent effort and improve accuracy. However, in a self-administered survey, it is difficult to determine whether higher accuracy reflects greater cognitive engagement or simply increased use of external information sources (e.g. online searches). Moreover, our primary interest lies in the randomly assigned experimental treatments. We do not expect incentives to have heterogeneous effects across experimental conditions, and thus their omission should not bias the estimated treatment effects.

Regional data

We use three data sources to check the respondents’ perceptions against actual data. For the neighborhood level, we use the RWI-GEO-GRID provided by the FDZ Ruhr at the RWI (51). This dataset contains socioeconomic information at the $1\times 1$ km raster cell level, which we then map to the ZIP code level. From this dataset, we use the variable “share of foreigners,” which is defined based on an onomastic procedure. Nonetheless, this dataset only contains information on the overall unemployment rate rather than disaggregated by nationality. Therefore, we use data from the German Federal Statistical Office (Destatis) at the municipality-level to compare the perceptions about foreign unemployment at the neighborhood level.

To benchmark perceptions of migrant shares at the district and national level, we aggregate the ZIP code level data to the corresponding levels. For the foreign unemployment rate, we use data from the German Federal Employment Agency for the year 2020 by regional level (districts and municipalities) and nationality (52).

Results

Study 1

In this section, we document the relevance of accounting for regional level when evaluating (mis)perceptions about migrant numbers and their labor market integration. Furthermore, we show which factors are associated with larger misperceptions and how the results vary by respondents’ characteristics.

A key claim in our article is that it is important to distinguish between perceptions about migrant population sizes and perceptions about migrants’ (labor market) integration. We argue that perceptions of the migrant unemployment rate should be more accurate when referring to more proximate and closer regional entities, whereas estimates of the size of the migrant group should not be affected. We experimentally varied this proximity by asking for respondents’ perceptions about the number of migrants and the share of unemployed migrants at the country, district, and neighborhood levels. Figure 1 shows the perceived share of migrants (Fig. 1A) and the perceived unemployment rates of migrants (Fig. 1B) for three different regional levels. The triangles in both panels denote the actual numbers.

Figure 1.

Average actual and perceived share of migrants (A) and their unemployment rates (B) by geographical level. The figure shows the average perceived share of migrants and migrants’ unemployment rates by regional level, and the corresponding 95% CI (in blue). The triangles show the actual migrant shares and migrants’ unemployment rates at each regional level. To illustrate the distribution of perceptions, plus symbols indicate the median and light bars depict the interquartile range. Source: Own survey, Destatis, Federal Employment Agency.

Our results are consistent with previous results in the literature and show that respondents overestimate migrant numbers to a large extent at all regional levels. For example, the share of foreign-born individuals in Germany at the time of the survey was 12.2%, while the average estimate in our sample is 23.6%. These misperceptions are slightly smaller at the neighborhood than at the district level ($\beta =-3.8$, $P=0.001$), but overall perceptions of the migrant share do not vary with the geographical level. When considering the unemployment rates of migrants, the picture looks different. Figure 1B shows that misperceptions, on average, decline with the level of regional aggregation. At the national level, respondents overestimate migrants’ unemployment rate by almost ten percentage points. At the district level, this misperception is reduced to a little less than six percentage points ($P<0.01$), while the gap is close to zero at the neighborhood level ($P=0.91$). The gap between perceptions and the actual number is largest at the national level, smaller at the district level ($\beta =-3.7$, $P=0.030$), and further reduced at the neighborhood level ($\beta =-9.54$, $P<0.001$).^b

We further investigate the distribution of respondents’ perceptions of the migrant share and migrants’ unemployment rates in comparison to the actual values of these statistics at the district level (A and C) and the neighborhood level (B and D) in Fig. 2. The green line represents the 45^∘ gradient. If respondents overestimate, observations are recorded above the gradient, while for underestimations observations are below the gradient.

Figure 2.

Actual and perceived share of migrants and their unemployment rates at district and neighborhood levels. Each panel shows binned scatter plots of actual vs. perceived migrant shares and unemployment rates. The longer line represents a 45^∘ gradient. Values above this gradient indicate overestimation ($\text{ perception}>\text{actual}$). The shorter line is a linear fit. The slope is 0.68 in (A) and 1.09 in (B); the difference between slopes (0.41) is statistically significant ($P=0.015$). For C and D), the corresponding slopes are 0.004 and 0.46; their difference (0.45) is statistically significant ($P=0.020$). Source: Own survey, Destatis, Federal Employment Agency.

Figure 2A and B depicts the results for the migrant shares at the district and the neighborhood level, respectively. Generally, migrant shares are always overestimated, but perceptions at the district level are less responsive to actual variation in migrant shares than at the neighborhood level (the slope is 0.68 in Fig. 2A and 1.09 in Fig. 2B; the difference between slopes (0.41) is statistically significant ($P=0.015$). Put differently, perceptions seem to vary with the actual migrant share, but respondents always believe that the migrant share is larger than it actually is. This bias tends to be less pronounced for districts with higher migrant shares. At the neighborhood level, perceptions and the real migrant share seem to have the same slope. Hence, despite a general propensity for overestimation, respondents hold perceptions that are closely correlated with the actual share of migrants at the neighborhood level.

The distribution of estimates of the migrant unemployment rate in Fig. 2C and D appears to be very different from the distribution of estimates of migrant shares. Perceived migrant unemployment is almost perfectly flat at the district level, i.e. respondents’ perceptions about the local migrant unemployment rate do not vary with the actual unemployment rates of migrants. This result suggests that although misperceptions about migrants’ labor market integration are, on average, less biased at the district level, respondents do not seem to have a good feeling for the actual integration of immigrants in their own district compared to other districts.

When considering local unemployment rates for migrants in Fig. 2D, respondents’ estimates seem to be distributed fairly close to the 45^∘ gradient. The slope is positive (0.46 and different from the slope at the district level by 0.45, $P=0.02$)) but crosses the 45^∘ line at an unemployment rate of about 15%. This result means that respondents living in neighborhoods with comparatively low migrant unemployment rates tend to believe that migrants’ unemployment is somewhat higher than it is in reality. In contrast, respondents living in neighborhoods with higher migrant unemployment rates tend to underestimate the unemployment rate. This pattern adds important nuance to the finding that neighborhood-level estimates appear relatively accurate on average, as illustrated in Fig. 1B. The accuracy on average masks offsetting misperceptions, with roughly similar shares of respondents overestimating and underestimating migrants’ unemployment rates across different local contexts.

We provide a formal examination of the relevance of the geographical unit to respondents’ (mis)perceptions relying on a linear regression (Table S7). We regress the gap in perceptions (i.e. perceived share minus the actual share) on respondents’ treatment assignment (indicators by geographical level), their individual characteristics, and regional covariates. The results corroborate the descriptive evidence found in Fig. 1: misperceptions of the migrant share do not vary with the regional context in a meaningful and robust way.

Furthermore, Table S7 reveals that men ($\beta =-4.7$, $P<0.001$) and the highly educated ($\beta =-2.51$, $P=0.004$) have more accurate perceptions. This is also true for voters of immigrant-friendly parties like The Left ($\beta =-6.28$, $P=0.005$) and The Greens ($\beta =-6.2$, $P=0.004$). Conversely, voters of the main anti-immigration party (the AfD) experience a greater degree of mismatch between perceptions and actual figures ($\beta =8.0$, $P<0.002$). Regional characteristics do not seem to exert too much influence on this gap in perceptions. Only the share of migrants at the district level is significantly different from zero and negatively associated with the gap ($P=0.009$). Put differently, respondents living in districts with a greater share of immigrants have fewer misperceptions. This is also documented in the slope in Fig. 2A.

Correspondingly, Table S8 shows the coefficients for the same regressions for the gap between the perceived and actual migrant unemployment rates. We see that the gap observed at the national level is reduced when respondents are asked to evaluate migrants’ labor market integration at the district ($-3.5$ percentage points ($P=0.005$)) or neighborhood level ($-10.3$ percentage points ($P<0.001$) (model 3). This result is robust to the inclusion of additional covariates (model 5).^c

Again, we observe larger misperceptions for women ($\beta =2.7$, $P=0.006$), but there are no differences with respect to education ($\beta =-1$, $P=0.37$). Instead, respondents over the age of 60 have more biased perceptions ($\beta =3.9$, $P=0.008$). Concerning vote intentions, the same pattern emerges as for the gap in migrant shares.

Overall, the regression results corroborate our assumption that the regional context does not affect all perceptions similarly. Misperceptions about the share of migrants do not vary with the regional context, whereas misperceptions of the unemployment rate become smaller with geographical proximity and even disappear on a neighborhood level.

Nevertheless, we see that greater misperceptions are associated with being female and vary with voting for immigrant-friendly (the Left and the Greens) vis-à-vis anti-immigrant parties (the AfD). In Fig. 3, we examine whether asking respondents to evaluate migrant numbers and integration affects those subgroups differently. In both panels, we show the subgroup average assessments for the three different geographical levels. Subgroups vary by gender, education, and political self-description.

Figure 3.

Misperceptions in migrant shares (A) and foreign unemployment rate (B) by individual characteristics. The figure shows the raw average misperceptions in migrant shares and unemployment rates and their corresponding 95% CI, by individual characteristics and level of response requested in our main sample.

Figure 3A shows perceptions of the migrant population share by individual characteristics and geographic level. Although CI largely overlap, there is some indication that misperceptions are smaller at the neighborhood level (green diamonds), particularly among female, noncollege-educated, and right-wing respondents—precisely the groups that exhibit the largest misperceptions overall. So, even for proportion judgments, where the domain-general bias should not be eliminated by proximity, respondents with the most inflated priors may see a modest attenuation when prompted to consider their immediate surroundings.

Figure 3B shows the corresponding results for perceptions of migrants’ unemployment rates. Here, the geographic gradient is far more pronounced: misperceptions shrink substantially at the neighborhood level for every subgroup and are virtually absent on average for most of them. The reduction is largest among those groups that exhibit the greatest overestimation at the national level—noncollege-educated, and right-wing respondents—suggesting that local experiential knowledge and contact might be most consequential precisely where media-driven stereotypes would otherwise dominate.

Taken together, the two panels underscore the asymmetry at the core of our argument. For proportion estimates of migrant population shares, geographic proximity yields at best a modest reduction in bias. For behavioral inferences about employment status, proximity produces a sharp correction—and this correction is strongest among precisely those respondents whose national-level perceptions are most distorted.

Study 2

Based on recent contributions to the literature of attitudes toward migrant populations (47, 48), we argue that perceptions about migrants should also vary by migrants’ characteristics. Hence, in a second experiment, we asked respondents to estimate the unemployment rates for different migrant groups and for Germans. This figure shows the average perceived unemployment rates (with 95% CI) and compares them to the actual unemployment rates.

As expected, respondents typically overestimate unemployment rates. Perceived unemployment is fairly similar for those of a German, Italian, or Polish background. The accuracy of the perceptions mostly varies by the actual unemployment rate, which is a little larger for Italians than for Polish or German people. Due to the fact that Germans have the lowest unemployment rate, misperceptions for this group are the largest compared to the gap for the Turkish unemployement rate ($\beta =-6.2$, $P<0.001$) and the Polish unemployment rate ($\beta =-3.4$, $P=0.009$).^d

Respondents perceive unemployment rates for Turkish and Syrians higher than for Italians, Poles, and Germans. This finding is also in line with the actual unemployment rates for Turkish and Syrians. This difference reflects actual unemployment rates, which are well above the unemployment rates of the other three groups. Gaps between actual and perceived unemployment rate are highest for German unemployment rates and are significantly smaller for Polish ($\beta =-3.4$, $P=0.009$), Turkish ($\beta =-6.2$, $P<0.001$), Italian ($\beta =-7.9$, $P<0.001$), and Syrian ($\beta =-15.1$, $P<0.001$) unemployment rates and hence do not increase with cultural distance. This suggests that the process of the formation of these perceptions are not racialized or driven by antimuslim bias (47). Notably, people do hold quite accurate perceptions about Syrians’ labor market integration and Syrians are the only group for which people underestimate the actual unemployment rate.

When we inspect the heterogeneities by characteristics of the respondents in Fig. S4, we do not detect any notable differences by gender, education, or political attitudes.

In sum, Fig. 4 documents that respondents indeed have more accurate perceptions of migrants’ labor market integration if they receive more information about the type of migrant. Priming the respondents with the country of origin of a migrant group seems to anchor perceptions in a way that decreases respondents’ tendency to overestimate unemployment rates.

Figure 4.

Average perceived and actual unemployment rates of migrants’ country of origin. The figure shows the average perceived unemployment rates by country of origin, and the corresponding 95% CI. The triangles show the actual unemployment rates by country of origin. To illustrate the distribution of perceptions, plus symbols indicate the median and light bars depict the interquartile range.

Conclusion

This article documents that (mis)perceptions about migration are not monolithic: their severity depends on what is being estimated, at what geographic scale, and about whom. Through two studies based on an online survey in Germany, we advance the existing literature on immigration perceptions by showing that the cognitive processes underlying different types of perception respond differently to geographic proximity.

Our central finding is the difference between perceptions of migrant population shares and perceptions of migrants’ labor market integration. Study 1 reveals that respondents overestimate the share of migrants at all geographic levels—national, district, and neighborhood—by roughly the same margin. Geographic proximity does not meaningfully reduce this bias, consistent with our theoretical argument that proportion judgments are subject to a domain-general cognitive bias that local experience cannot correct (21, 40). Perceptions of migrants’ unemployment rates, by contrast, become substantially more accurate at the neighborhood level. We argue that this is because employment status is updated through direct, everyday contact with migrants in one’s local environment (41, 53). At the national level, where such experiential knowledge is unavailable, media-driven stereotypes of migrants as economically unintegrated go uncontested (43), producing large overestimates of migrant unemployment.

An important nuance in these unemployment results deserves emphasis. The average accuracy of neighborhood-level perceptions masks systematic variation. Respondents in neighborhoods with comparatively low migrant unemployment tend to overestimate it, while those in neighborhoods with higher migrant unemployment tend to underestimate it. One plausible interpretation of this pattern, consistent with the exemplar-based account we propose (41), is that perceptions track actual local conditions but imperfectly, with estimates regressing toward a common prior about migrant unemployment. If respondents form their impressions from a limited and potentially unrepresentative set of everyday encounters, their inferences would be drawn toward a central expectation—overshooting where actual rates are low and undershooting where they are high (54). Yet, this would call for more systematic empirical validation since our nonprobability sample does not allow us to rule out compositional differences between neighborhoods that might drive part of this pattern. A more rigorous test would require a probability-based sample with sufficient coverage across a representative set of neighborhoods, combined with controls for other local characteristics—such as the socioeconomic composition or the visibility of migrant economic activity—that may independently shape perceptions

Second, study 2 shows that perceptions about migrants’ economic integration vary with migrants’ country of origin. Respondents overestimate unemployment rates for most migrant groups as well as for Germans, but the degree of overestimation is linked to actual unemployment rates rather than to cultural distance or anti-immigrant sentiment. Notably, the gap between perceived and actual unemployment is largest for Germans—who have the lowest actual rate—and smallest for Syrians, for whom respondents slightly underestimate unemployment. Providing respondents with information about a specific origin group appears to anchor perceptions closer to reality, suggesting that the common practice of asking about “migrants” as a homogeneous category inflates measured misperceptions.

Our study challenges the prevailing assumption that perceptions about migration are primarily formed at the national level and that they are uniformly biased. The significant reduction in misperceptions of migrants’ labor market integration at the neighborhood level—particularly among groups traditionally prone to the largest misperceptions, such as low-educated individuals and those on the political right—suggests that localized experiences can mitigate misconceptions where the attribute in question is directly observable. Combined with the finding that origin-specific information improves accuracy, our results indicate that more nuanced and localized information about migrants’ characteristics may be more effective in correcting biased perceptions than general national-level statistics (19). At the same time, our findings suggest that correcting overestimates of migrant population shares requires different strategies, since these misperceptions are rooted in cognitive biases that are not responsive to local context.

Both surveys were conducted during the COVID-19 pandemic, which may limit the external validity of our findings. This unprecedented crisis could have influenced perceptions of migrants through several channels, including heightened economic and health risks, increased perceptions of threat, border closures, and shifts in labor demand across occupations and migrant groups. While existing empirical research suggests that pandemic-related changes in political attitudes were largely temporary and had mostly returned to prepandemic levels by late 2020 and early 2021 (55), we cannot fully rule out that the pandemic context affected respondents’ evaluations. Nevertheless, given our focus on randomly assigned experimental treatments, we expect our main experimental findings to be relatively robust to these contextual influences.

In conclusion, this article advances our understanding of how perceptions about migration are formed by demonstrating that different types of perception are governed by distinct cognitive processes with different susceptibilities to correction. For scholars, our findings emphasize the need to distinguish between perceptions of group size and perceptions of group characteristics, and to account for the geographic scale at which perceptions are elicited. For policymakers, the results caution against drawing conclusions from national-level misperceptions alone: the widely documented overestimation of migrant population shares appears resistant to local experience, whereas misperceptions about economic integration are substantially correctable through proximity and group-specific information.

Supplementary material

Supplementary material is available at PNAS Nexus online.

Competing interest

The authors declare no competing interests.

Funding

This work is supported in part by funds from the German Ministry of Labour and Social Affairs (BMAS) via the Forschungsnetzwerk interdisziplinäre Sozialpolitikforschung (FIS).

Author contributions

E.N. and K.S.: conceptualization, validation. E.N., P.B., M.B., K.G.T., M.L., and K.S.: formal analysis, methodology, writing—original draft. E.N., K.G.T., M.L., and K.S.: supervision. E.N., M.L., and K.S.: funding acquisition, project administration. E.N., P.B., K.G.T., M.L., and K.S.: investigation, writing—review and editing. P.B., K.G.T., M.L., and K.S.: visualization.

Data availability

The data underlying this article are available in OSF at https://osf.io/g569c/overview and can be accessed with https://doi.org/10.17605/OSF.IO/G569C.