Introduction
While women account for roughly half of the global refugee population,Footnote 1 the gender distribution can vary significantly across time and space. According to Eurostat, more than 80% of adult Syrians, Afghans, and Pakistanis granted asylum in the EU before 2022 were men. By contrast, during the 2022 Ukraine crisis, 75% of adult Ukrainians under temporary protection in the EU were women.Footnote 2
The gender composition of refugee groups can significantly influence residents’ views and welcoming attitudes, as people tend to be more welcoming toward female migrants and refugees (Bansak, Hainmueller and Hangartner Reference Bansak, Hainmueller and Hangartner2016; De Coninck Reference de Coninck2023; Ward Reference Ward2019) or groups consisting primarily of women (Krawczyk et al. Reference Krawczyk, Blasco, Gajderowicz and Giergiczny2023). At the same time, public perceptions of refugees’ characteristics, including gender composition, are often inaccurate or influenced by misinformation (Lutz and Bitschnau Reference Lutz and Bitschnau2023). For example, French politician Marine Le Pen (MLP) incorrectly claimed that 99% of individuals crossing the Mediterranean were men (Barrera et al. Reference Barrera, Guriev, Henry and Zhuravskaya2020). Similarly, Krawczyk et al. (Reference Krawczyk, Blasco, Gajderowicz and Giergiczny2023) showed that public opinion across six European countries significantly underestimated the share of women among displaced Ukrainians in 2023. These misperceptions can ultimately affect how welcoming societies are toward different refugee populations.
In this paper, we employ a randomized experiment to evaluate the effectiveness of providing information on the gender composition of Ukrainian refugees in influencing Europeans’ propensity to change their beliefs and attitudes. We also vary the source of information on refugee demographics between official national statistics and informal polls, investigating source credibility. The informal polls also varied in accuracy: one reported accurate data (75% women), whereas the other portrayed most refugees as men (55%), contrary to available evidence.Footnote 3
We investigate the effect of information provision about refugee demographics on three key dimensions: (1) perceptions of how refugees impact crime, examining whether individuals believe the situation gets worse; (2) support for refugee family reunification policies within the EU; and (3) perceptions of the economic impact of migrants, which could be seen as positive or negative.
These three dimensions are critical in shaping public opinion about migration. First, crime and safety are widely debated aspects of migration and are closely related to the view of migrants or refugees as predominantly male. Indeed, both crime statistics and public perceptions predominantly link violent crimes with the male gender. Second, the gender composition of the refugee population clearly influences the scope of family reunification policies. Third, migrants’ potential contributions to the economy are often linked to gender composition, with men and women pursuing different occupations and affecting distinct sectors.
Our results demonstrate the effectiveness of information provision in correcting public misconceptions about the proportion of women among displaced Ukrainians in the EU, particularly when delivered via an official source. However, we found no evidence that it affected public attitudes toward refugees, as the distributions of our three outcome variables did not differ significantly from those observed in the control group. These findings suggest that communication about the share of women alone is unlikely to have a substantial effect on individual attitudes toward migrants and, by extension, to facilitate their integration. We need to understand better how widespread misperceptions of gender composition may lead to inefficiency or suboptimal policies. In the discussion section, we elaborate on the implications and generalizability of these findings.
Related literature
Our study contributes to three main strands of the literature: research on gender and preferences toward migrants; work on misperceptions about migration and interventions to correct them; and studies examining the role of misinformation and source credibility in shaping beliefs and attitudes about migration.
Gender and attitudes toward migrants
Most studies find that people tend to prefer female over male refugees and economic migrants (Bansak et al. Reference Bansak, Hainmueller and Hangartner2016; Denney and Green Reference Denney and Green2021; Findor et al. Reference Findor, Hruška, Jankovská and Pobudová2022; Hainmueller and Hopkins Reference Hainmueller and Hopkins2015; Ward and Denney Reference Ward and Denney2022), with one notable exception (Hager and Veit Reference Hager and Veit2019). Consistent with these findings, Mascini and van Bochove (Reference Mascini and van Bochove2009) show that male asylum applicants have lower success rates than women. One explanation for these patterns is the fear of physical or cultural threats attributed to men (Ji et al. Reference Ji, Tybur and van Vugt2021; Ward Reference Ward2019), particularly in contexts where male migrants are perceived as more prone to violence. In addition, mate competition concerns among native men may also contribute to negative attitudes toward an inflow of male refugees or migrants, as evidenced by the association between hate crimes and perceived difficulties in local mating markets (Dancygier et al. Reference Dancygier, Egami, Jamal and Rischke2022)
At the same time, women may be perceived as more deserving, especially in humanitarian crises. De Coninck and Matthijs (Reference de Coninck and Matthijs2020) apply Van Oorschot’s (Reference van Oorschot2000) deservingness framework to migrants, showing that refugees displaced by war are perceived as more deserving than other migrants, particularly those perceived as physically weaker, such as women (Glick and Fiske Reference Glick, Fiske and Zann2001). However, despite these insights, relatively little research has examined how these gendered preferences interact with public misperceptions about the gender composition of refugees.
Misperceptions about migration and the difficulty in correcting them
Beliefs about migrants are often incorrect. Individuals tend to overestimate migrants’ numbers, poverty, cultural distance, and welfare dependence (Lutz and Bitschnau Reference Lutz and Bitschnau2023) and to view their impact on the economy and the labor market as more negative than experts do (Wu Reference Wu2021). Individuals also tend to overestimate migrants’ criminality and to blame them for worsening crime rates (Ceobanu Reference Ceobanu2011; Semyonov et al. Reference Semyonov, Raijman and Gorodzeisky2008). The media seem to reinforce these misperceptions, which are more common among supporters of right-wing parties (Arendt Reference Arendt2010; Blinder and Jeannet Reference Blinder and Jeannet2017).
Correcting such misperceptions is challenging (Jørgensen and Osmundsen Reference Jørgensen and Osmundsen2022; Lergetporer, Piopiunik and Simon Reference Lergetporer, Piopiunik and Simon2018; Rolfe et al. Reference Rolfe, Ahlstrom-Vij, Hudson-Sharp and Runge2018), partly because they seem to reflect a more general limited ability to handle numbers (Landy et al. Reference Landy, Guay and Marghetis2018). Although some studies find corrections effective (Swire et al. Reference Swire, Berinsky, Lewandowsky and Ecker2017; Swire-Thompson et al. Reference Swire-Thompson, Ecker, Lewandowsky and Berinsky2020), individuals may ignore new information (Rolfe et al. Reference Rolfe, Ahlstrom-Vij, Hudson-Sharp and Runge2018) or reinterpret it to fit their original views (Jørgensen and Osmundsen Reference Jørgensen and Osmundsen2022). Strikingly, inaccurate information may persist and continue to influence people’s thinking even after misinformation is corrected and the new information is accepted as accurate, a phenomenon called the continued influence effect (Chan et al. Reference Chan, Jones, Jamieson and Albarracín2017; Ecker, Lewandowsky and Cook Reference Ecker, Lewandowsky, Cook, Schmid, Fazio, Brashier, Kendeou, Vraga and Amazeen2022; Johnson and Seifert Reference Johnson and Seifert1994; Walter and Murphy Reference Walter and Murphy2018; Walter and Tukachinsky Reference Walter and Tukachinsky2020).
However, there is limited direct evidence of public beliefs about the gender composition of refugees or in identifying interventions that could effectively correct these misperceptions, a gap our study addresses.
Source credibility and impact on migration beliefs and attitudes
The credibility of information sources is also crucial, as people tend to update their beliefs based on perceived trustworthiness and expertise (Ecker et al. Reference Ecker, Lewandowsky, Cook, Schmid, Fazio, Brashier, Kendeou, Vraga and Amazeen2022; Pornpitakpan Reference Pornpitakpan2004; Vraga and Bode Reference Vraga and Bode2018). Differences in source credibility can effectively influence the spread of misinformation (Prike et al. Reference Prike, Butler and Ecker2024). However, the role of official statistics in shaping public perceptions of refugees remains poorly understood.
Much research has examined the credibility of the source providing correcting information (Ecker et al. Reference Ecker, Lewandowsky, Cook, Schmid, Fazio, Brashier, Kendeou, Vraga and Amazeen2022), including perceived trustworthiness and expertise (Ecker et al. Reference Ecker, Lewandowsky, Cook, Schmid, Fazio, Brashier, Kendeou, Vraga and Amazeen2022; Guillory and Geraci Reference Guillory and Geraci2013; Nadarevic et al. Reference Nadarevic, Reber, Helmecke and Köse2020; Pornpitakpan Reference Pornpitakpan2004; Vraga and Bode Reference Vraga and Bode2018). However, most of these studies rely on correcting experimentally induced beliefs, which are more open to correction.
Research also highlights that belief updating may not translate into attitude changes or policy support. On the one hand, Grigorieff et al. (Reference Grigorieff, Roth and Ubfal2020) or Haaland and Roth (Reference Haaland and Roth2020) observed that correcting the characterization of migrants led to more positive attitudes toward migrants. Likewise, Japanese respondents of Facchini et al. (Reference Facchini, Margalit and Nakata2022) were more supportive of migrants when exposed to information on social problems that migration could help address. Similarly, Allen et al. (Reference Allen, Ahlstrom-Vij, Rolfe and Runge2023) observed that providing evidence on the (modestly positive) effect of migrants from the EU on the British economy changed attitudes and policy preferences of their UK participants. On the other hand, several studies have shown that even when self-reported beliefs change in response to correcting information, attitudes and subsequent judgments may remain unchanged (Hopkins et al. Reference Hopkins, Sides and Citrin2019; Lawrence and Sides Reference Lawrence and Sides2014; Lergetporer et al. Reference Lergetporer, Piopiunik and Simon2018). Consistent with these findings, Kustov, Laaker and Reller (Reference Kustov, Laaker and Reller2021) argue that immigration preferences are remarkably stable over time and robust to economic and political shocks, suggesting a limited effectiveness of information interventions.
In this context, Barrera et al. (Reference Barrera, Guriev, Henry and Zhuravskaya2020) examined the effects of misinformation on the gender composition of migrants and its consequences for political behavior. The study focused on the support for the (extreme-right) candidate MLP in France. One group of participants was exposed to MLP’s three widely incorrect claims about migrants, one of which was that 99% of illegal migrants coming to Europe were men; another group was exposed to the correct official statistics; another group was exposed to both; and a control group was exposed to none. While exposure to facts effectively corrected beliefs, it did not reduce participants’ willingness to vote for MLP or agree with her views on migration. Interestingly, compared with the control group, participants exposed only to facts were also more supportive of MLP. Mentioning neutral facts or reminding people of migration may have made the topic more salient, apparently sufficient to improve ratings of politicians who take an extreme stance on it. Barrera et al. (Reference Barrera, Guriev, Henry and Zhuravskaya2020) underscore the complexity of translating corrected beliefs into attitudinal or behavioral change and highlight the need for interventions tailored to gender-specific misperceptions, an issue our study aims to address.
Further research has explored the limits of such information interventions. Narrative interventions, such as portraying migrants as hard-working people, as in Alesina et al. (Reference Alesina, Miano and Stantcheva2023), could have a greater impact on attitudes toward migrants than information alone. Adida et al. (Reference Adida, Lo, Platas, Prather and Williamson2025) further show complementarities between perspective-getting interventions and information provision in shaping individual attitudes on refugee inclusion, underscoring the importance of bundling information provision with other intervention strategies.
Experimental design
We conducted an online survey experiment with a sample representative of the general adult population across six European countries (Czechia, France, Germany, Hungary, Poland, and Romania). A total of 13,008 respondents completed the survey. Among other elements (see Appendix A for the English master version), we implemented a belief-correction experiment, which is the focus of the current paper, the rest being discussed in companion papers. Note that these earlier modules are orthogonal to the belief-correction experiment, thus cannot explain treatment effects (if any).
All study participants were asked to estimate the demographic characteristics of displaced Ukrainians, including the proportion of women. Specifically, we asked them to guess “How many of the adult Ukrainians who have arrived in [the host country]” were women, were native speakers of Ukrainian, were native speakers of Russian, could communicate in English, could communicate in the language of the host country, and had higher education.
After a short buffer of other questions (see Appendix A), we reminded respondents of their estimate of the share of women among displaced Ukrainians (which, in this report, we will refer to as the First Guess). We then randomized at the individual level, assigning respondents to one of the four treatments.
One-fourth of the participants were informed that:
-
• “Statistics available for six EU countries, including [country], indicate that women represent about 75% of adult Ukrainian refugees there.” [Official High treatment or Offi-High for short]
Another group was informed that:
-
• “One group of people we surveyed in six EU countries, including [country], believe that women represent about 75% of adult Ukrainian refugees there.” [(Previous) Survey High treatment or Survey High for short]
Yet another group received the following message:
-
• “One group of people we surveyed in six EU countries, including [country], believe that women represent about 45% of adult Ukrainian refugees there.” [(Previous) Survey Low treatment or Survey Low for short]
The remaining respondents formed the control group as they received no additional information. Participants were unaware of existence of other treatments.
The statement “One group of people…” was chosen so that both statements, while plausibly shaping beliefs in different ways, were simultaneously factual: in the first wave of our study, conducted in the Autumn of 2022, we have indeed already elicited beliefs about the share of women among displaced Ukrainians. They turned out to be quite dispersed, with two groups of respondents, the first judging it to be 40–50%, the second judging it to be 65–80%, representing over 60% of the sample (with another prominent group of 8.5% of respondents stating all displaced Ukrainians were female, and the rest of the sample being spread across other values). The design was approved by the Research Ethics Board of the Joint Research Centre.
All respondents were asked whether they wanted to reconsider their previous answer: “In view of this information, do you want to change or confirm your previous answer? How many of the adult Ukrainian refugees who arrived in [country] over the last year do you think are women?” The introductory words “In view of this information” were dropped in the Control Treatment. The response was registered as the Second Guess.
We measured the impact of this intervention on three questions. First, respondents were asked whether they believed that their country’s crime problems were being made worse or better by the presence of displaced Ukrainians. Then, they were asked whether relatives of displaced Ukrainians registered in their country should be permitted to travel there (family reunification). Finally, they were asked whether migration was generally beneficial or detrimental to their country’s economy. All three were answered on a 0–10 scale.
Hypotheses
We hypothesize that it is possible to “correct” the perception of refugee characteristics, when it is initially further off the mark. We further hypothesize that such a belief shift may affect the perception of refugees, notably the extent to which they are perceived as a threat.
We have thus the following hypotheses concerning the belief correction:Footnote 4
Hypothesis 1: We expect belief updates to be rare in the Control treatment (probability of updating not greater than 5%).
Hypothesis 2: We expect the probability of updating the First Guess and the expected update to be higher, the larger the difference between the First Guess and the provided information.
Hypothesis 3: We expect an effect of the intervention on the probability of revising the belief in the Official High treatment to be larger than the effect in the Previous Survey High treatment, due to the source of information having the approval of the statistical office and, thus, being more reliable.
Hypothesis 4: Conditional on initial beliefs, we expect an effect of the belief-correction intervention on the probability of updating to vary depending on the level of education, self-reported social trust, attitudes toward Ukrainian refugees, and views about migration.
Hypothesis 5: We expect an effect of the belief-correction intervention on participants’ perceived link between refugees and crime. Compared to the Control treatment, we expect a more negative assessment of the effect of Ukrainian refugees on the crime situation in the Previous Survey Low treatment and a more positive assessment in the Previous Survey High and Official High treatments. The larger the difference between the First Guess and the information provided in the treatment, the stronger the effect.
Hypothesis 6: We expect the effect on crime perceptions to vary by gender, age, education, and other key demographics.
Note that it is only one of the first dependent variables for which an effect is clearly predicted because of the common association between crime and the male gender. For the other two, it is an open question whether an effect shows up.
Sample and procedures
The master English-language version was developed, programmed, tested, and professionally translated into national languages of the six EU countries in which the study was administered: Czechia, France, Germany, Hungary, Poland, and Romania. The final sample comprised 13,008 respondents. Distribution of key demographic variables is shown in Table F1 of Appendix F, whereas Table F2 shows it was balanced across treatments.
Participants were recruited by Valdani Vicari & Associati through proprietary and third-party panels associated with the global marketing company Cint. The main fieldwork started on March 8, 2023, and lasted for about two weeks.
The anonymized preregistration document is attached, along with the information about the correspondence between the manuscript and the preregistration document.
Results
Aggregating across all countries (see Table B1 for country-specific values), the mean of the First Guess regarding the share of women among displaced Ukrainians was 59.22% (with a standard deviation of 2.43). This is almost ideally between the 45% and 75% values featured in our belief correction treatments, as well as far away from the “floor” and “ceiling,” thus setting the stage for identifying possible treatment effects, even in subgroups deviating somewhat from the global mean. We first looked at how respondents adjusted this initial guess. Across all treatments, nearly one in four responders (27%) updated their beliefs upward, indicating a higher proportion of women; one in six (16%) revised their beliefs downward, and most respondents (57%) maintained their initial guess.
Figure 1 shows the distribution of a three-value variable (“Update Direction”) representing the direction of changes (upward, zero, and downward) across treatments. There was a significant difference in the Update Direction across the three treatment groups (χ 2(6) = 1167.9, p < .001). Specifically, in the Official High treatment, 43% of respondents increased their estimate of the share of women, while 34% did so in the Survey High treatment. By contrast, in the Survey Low treatment, only 17% of respondents increased their estimate while a notable share of 26% individuals decreased it.
Propensity to update beliefs about the share of women for each treatment. Update Direction takes the value “upward” when the difference between Second Guess and First Guess is positive; “downward” if negative; and “zero” when there is no difference. P-value for the association between treatment assignment and Update Direction is also reported.
In the control group, most respondents (71%) did not update their reported beliefs, while 17% increased it and 13% decreased it. Unexpectedly, the share of participants in the control group who changed their beliefs (29%) considerably exceeded what we postulated in Hypothesis 1 (below 5%). Thus, merely prompting a reflection on the First Guess was a powerful mechanism to induce people to adjust their initial estimates.
A pairwise test of the difference in the Update Direction identified a statistically significant difference between any treatment pairs (p < 0.001 in Mann–Whitney tests for all pairs). They remain statistically significant even after a Bonferroni correction, thus providing strong evidence of treatment effects.
More specifically, compared to the control group, informing participants that 75% of adult refugees were women boosted the likelihood of increasing their estimates (“upward” change) by over 20 percentage points, a substantial and highly significant impact. The source of information played a major role in this effect: there was a 9-percentage-point difference in the likelihood of increasing the estimates between the Official High and Survey High treatments, supporting Hypothesis 3 about the effect of the credibility of sources (official vs non-official) on the effectiveness of the intervention.
Similarly, informing participants that 45% of refugees were women, according to a survey, increased the probability of reducing their estimates (“downward” change) by about 13 percentage points compared to the control group. Therefore, people were inclined to respond to such information provision, even if this involves incorrect information from unofficial sources.
Appendix B shows that analogous results hold when we analyze the size of the change, not merely the direction.
Figure 2 illustrates the joint distribution of the First and Second Guesses across the treatment groups for the subsets of respondents who modified their estimates. The distributions peaked along the reference lines corresponding to the share of women provided in the given treatment (representing participants who updated their beliefs completely). Indeed, under Official High and, to a lesser extent, under Survey High, observations concentrate on the 75% horizontal line. In the Survey Low group, it is the 45% line instead. Observations between the 45-degree line (no update) and the reference line are also common, representing participants who only partly updated their beliefs in the direction dictated by the treatment. In the control group, by contrast, the (much less numerous) updated guesses were not concentrated on any specific value. These findings are corroborated by the analysis presented in Table B2 and in Appendix D.
Scatterplot illustrating the relationship between the first and second guesses across the different treatment groups, with a contour plot depicting the joint density distribution. Each panel includes only respondents who updated their initial beliefs (the number of such is displayed below each panel).
Treatment effects on attitudes and policy preferences
We have established that the belief-correction intervention had a systematic and significant effect on the respondents’ propensity to update their estimates of the share of women among displaced Ukrainians. Now, we test whether this intervention impacted their attitude toward the presence of refugees, including their concerns about increased crime or adverse economic consequences and the support for family reunification policies. Figure 3 illustrates the distributions of the responses to the questions capturing these dimensions for each country.
Distribution collapsed across all treatments. Crime = Are [country]’s crime problems made worse or better by the presence of refugees from Ukraine? (0 = Crime made worse; 10 = Crime made better); Economy = Regarding immigration in general, would you say it is good or bad for [country’s] economy that people come to live here from other countries? (0 = very bad; 10 = very good); Family reunif. = Do you agree or disagree that family members (spouse, children, immediate relatives) of recognized Ukrainian refugees should be free to travel to [country] for family reunification? (0 = Strongly disagree; 10 = strongly agree).
Data on fears of increased crime are particularly informative on the hypothesized mechanism of preference against male refugees. However, as shown in the figure, almost half of respondents chose the neutral point 5 on the 0–10 scale in each country, thus considering the crime situation not being much affected by the arrival of refugees. Nonetheless, a substantial 30% minority said that crime would be worse, and 24% that it would be better, with Czechia being notably more fearful of increased crime (i.e., a thicker left tail).
Additionally, respondents tended to look favorably at family reunification policies, and the impact of migrants on their country’s economy, with the responses to the question about the impact of immigration on economy and support for reunification shifted to the right. In Romania, the distribution of support for family reunification peaked at maximum (i.e., 10), whereas the midpoint (i.e., 5) was the modal response in other countries.
Because of the distribution of the original dependent variables, we recoded them into variables taking just three values. For example, instead of using a continuous measure for perceptions of the crime situation, we categorized them as perceiving the crime situation as being worse, the same, or better, respectively. As can be seen in Table 1 reporting ordered logit regressions, in violation of Hypothesis 5, we identified no treatment effects, regardless of the set of controls.
The impact of treatment on perception of crime, controlling for FG as a continuous variable
Table 1. Long description
A table with four columns and multiple rows, showing the impact of treatment on perception of crime, controlling for FG as a continuous variable. The table includes variables such as crime direction treatment effects, age, male, size of place of residence dummies, country dummies, education level dummies, education X treatment dummies, demeaned trust, demeaned trust X treatment dummies, demeaned support for temporary protection, demeaned support for temporary protection X treatment dummies, and cut values. Each row lists the values for these variables across four different models. The table also includes significance indicators for the values.
Note: *p<.10, **p<.05, ***p<.01
The same is true when we control for the First Guess as a type rather than a continuous variable, see table E7 in Appendix E. Likewise, for either approach (FG as a continuous variables vs. as a type), there is no treatment effect in regressions taking perception of economic situation (Tables E3 and E4) or preference for family reunification (Tables E5 and E6). Overall, there are no systematic effects of treatment, with only a very small part of the dependent variables’ variation being explained, mostly by variables reflecting trust in others and support for temporary protection of displaced people from Ukraine.
Discussion
A large literature shows that people are generally more welcoming toward female than male migrants or refugees. Yet, public perceptions of refugees’ characteristics are often inaccurate. In the context of Ukrainians displaced by war, where the majority are women, misperceptions of the gender composition of displaced people may negatively affect attitudes toward refugees and support of inclusionary policies. This raises the question of whether awareness-raising interventions can correct perceptions and increase acceptance of such policies.
Our study shows that EU residents tend to underestimate the share of women among displaced Ukrainians hosted in the EU, with a difference of 15 percentage points compared to official statistics, which is consistent across the six EU countries in our sample.
While refugees’ gender composition varies between countries and fluctuates over time, this gap remains notable. For example, it is greater than the 10% drop in the share of women among newly approved applicants for temporary protection from its highest peak at the beginning of the war in Ukraine (first quarter of 2022) and a year later (the first quarter of 2023), according to Eurostat.
Given the extent of these misperceptions, our results underscore a possible solution. We find that providing accurate information is an effective way to align people’s beliefs of the gender composition of refugees with correct figures. Moreover, our findings help identify groups, especially women and people with low initial estimates, for whom this intervention is particularly effective. Still, the post-intervention estimates of the share of women remained significantly below official statistics.
We also show that the source of such information affects its effectiveness. Providing figures from national statistics is more effective than information from non-official sources, such as pools of people’s opinions. This finding underscores the great potential of information awareness campaigns based on official statistics, adding to the literature that studies source credibility as a mediator factor.
Conversely, interventions offering inaccurate information from unofficial sources are far less persuasive. While somewhat reassuring, this evidence still highlights the dangers of misinformation regarding migrants’ characteristics, especially for some segments of the population.
Another important contribution is to empirically assess how changes in beliefs about the gender composition of refugees affect attitudes toward refugees and support for inclusionary policies. However, we found no evidence of an impact of the belief intervention on relevant individual attitudes. Specifically, we found no statistically significant effect of treatments on the perception of migrants’ impact on crime and economic situation or the willingness to support family reunification for displaced people already in the country. This corresponds to findings, based on non-experimental survey methods, of temporal stability of immigration attitudes, despite shocks, including the refugee crisis (Kustov et al. Reference Kustov, Laaker and Reller2021; Lancaster Reference Lancaster2022), and to experimental findings of low impact on belief correction on logically related preferences and attitudes (Swire et al. Reference Swire, Berinsky, Lewandowsky and Ecker2017; Swire-Thompson et al. Reference Swire-Thompson, Ecker, Lewandowsky and Berinsky2020).
This result is remarkable because, as our analysis shows, there is a strong association between the initial guess of gender composition of the population of refugees and attitudes or support for policy, suggesting that beliefs of gender composition are closely linked with individual attitudes. Yet, we found no significant effect on these attitudes. Importantly, the large sample size of our study implies that the lack of statistical significance cannot be attributed to low statistical power.
One interpretation of this pattern is that participants only claim they update their beliefs while they do not really change their internal beliefs (a “good subject effect,” see Haaland et al. Reference Haaland, Roth and Wohlfart2023, McCambridge et al. Reference McCambridge, De Bruin and Witton2012, Zizzo Reference Zizzo2010). However, given that the experiment was run online, not in person, that participants reacted differently to different treatments, and that they often only updated partially, we do not think it is likely to have been a very important factor. Likewise, it seems rather unlikely that the participants missed the strong statistical link between male gender and crime (de Norohna Reference de Noronha2015; Lauritsen et al. Reference Lauritsen, Heimer and Lynch2009).
The most likely explanation of the lack of significant effects on attitudes is that the newly acquired information about the share of women is not readily used when forming subsequent judgments. This is consistent with the continued influence effect, which refers to the persistence of false information and its impact on subsequent judgments despite having been corrected. Some research suggests that new information does not simply replace the old; instead, both views can be available and may be activated, depending on the cues provided (Kendeou et al. Reference Kendeou, Walsh, Smith and O’Brien2014; Shtulman and Valcarcel Reference Shtulman and Valcarcel2012). It results from a selective retrieval of false information due to the availability and level of activation even when corrective information has been encoded in memory. When information is first encoded and subsequent information corrects or discredits this initial information, the initial information remains in memory. Correct and incorrect information then compete for activation, and even discredited information can be retrieved and can influence attitudes and judgments (Ecker, Lewandowsky and Tang Reference Ecker, Lewandowsky and Tang2010; Kendeou et al. Reference Kendeou, Walsh, Smith and O’Brien2014).
Finally, what does this finding say about changing attitudes by correcting information about other demographics? While our interventions focused on the gender composition of a group of migrants, a dimension with particularly strong association with public attitudes, we would expect weaker effects for other variables such as age or education level.
Likewise, it is possible that simply telling people the official numbers (or others’ guesses) make little difference: 75% is just an abstract number; seeing it may fail to profoundly change participants’ gut feelings about a given population. It cannot be excluded that actions making the demographic composition salient, such as family events integrating locals and refugees, could potentially improve the overall perception of displaced population (a mechanism related to but distinct from the contact hypothesis, Allport Reference Allport1954). Another important distinction here is one made by Thorson and Abdelaaty (Reference Thorson and Abdelaaty2022). They find that providing information about a policy makes more difference than providing information about its outcomes. In the current project, we informed about the outcomes (gender composition); informing about policies (e.g. the fact that, in general, working-age men are not allowed to leave Ukraine) could have a greater effect. These important considerations do not invalidate the key conclusion arising from our exercise: that communication concerning demographic composition of migrants is likely to correct misperception but is unlikely to yield a large impact on public attitudes or support for policy.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/XPS.2026.10037.
Data availability statement
The data and code to replicate all analyses presented in this article are available at the Journal of Experimental Political Science Dataverse within the Harvard Dataverse Network at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/HYGRWD (Blasco, Krawczyk and Nohlen Reference Blasco, Krawczyk and Nohlen2026).
Funding statement
This study was funded by the Joint Research Centre of the European Commission.
Competing interests
The authors declare none.
Ethics statement
The study presented in this paper was approved by the Research Ethics Board of the Joint Research Centre, number 30506_CCBI.
Open access
