To unpack this support, we leverage the data from the conjoint experiment. Figure 2 shows the estimated effects of the asylum seeker attributes on support, pooling across all respondents in the 2022 and the 2016 waves of the survey. These results are based on a forced choice outcome, which denotes whether a profile was preferred or not in a randomly generated pair of profiles. The Supplementary Information provides results for analogous analyses that use a rating outcome (with a scale of 1 to 7 that respondents used to rate profiles individually) and a dichotomized version of the rating variable, and the findings are similar (Supplementary Figs. 11 and 12). Figure 2 reveals several important findings.
Dots with horizontal lines indicate point estimates with cluster-robust 95% confidence intervals from linear (weighted) least-squares regression. The unfilled dots on the zero line denote the reference category for each asylum seeker attribute. The results from 2016 were first presented in ref. 5. Attributes that were only included in 2022 are shown in italics. n = 178,740 profiles evaluated for 2016 and n = 148,460 profiles evaluated for 2022. The underlying regression results are presented in Supplementary Table 9. PTSD, post-traumatic stress disorder.
First, we find that Europeans’ patterns of preferences regarding the desirable characteristics and traits of asylum seekers have remained remarkably stable between the height of the Syrian crisis in 2015–2016 and the height of the Ukrainian crisis in 2022 (Fig. 2). Inconsistencies in asylum testimony, gender, age, previous occupation, special vulnerabilities, reason for migrating, religion and language skills each have similar effects on the probability of being supported for admission in both survey waves. Major inconsistencies have a negative 9.6–10.7 percentage point effect relative to no inconsistencies (|t| = 23.42–28.38; maximum P < 0.00001 for effects in both survey waves based on two-sided t-tests of estimates from linear least-squares regressions; n = 178,740 in 2016; n = 148,460 in 2022). Being male has a negative 5.8–6.0 percentage point effect relative to being female (|t| = 18.72–19.64; maximum P < 0.00001). Being 62 years old has a negative 5.9–6.1 percentage point effect relative to being 21 years old (|t| = 14.39–15.54; maximum P < 0.00001). Having been previously employed has a positive 4.8–14.2 percentage point effect and having been employed in a high-skilled profession (accountant, teacher or doctor) has a positive 7.6–14.2 percentage point effect relative to having been unemployed (|t| = 9.06–25.05; maximum P < 0.00001). Being a victim of torture has a positive 9.8–11.2 percentage point effect relative to having no special vulnerabilities (|t| = 18.95–23.65; maximum P < 0.00001). Being Muslim has a negative 8.8–10.7 percentage point effect relative to being Christian (|t| = 21.39–28.04; maximum P < 0.00001). Having no host country language skills has a negative 11.1–11.7 percentage point effect relative to being fluent (|t| = 27.19–30.08; maximum P < 0.00001). Finally, there is a negative effect of 13.3–18.9 percentage points for asylum seekers who are migrating for economic reasons rather than fleeing persecution or war (with economic reasons set as the reference category, |t| = 25.50–36.26; maximum P < 0.00001).
Furthermore, the differences in the magnitudes of the effects of all of these characteristics (excluding country of origin) between the 2016 and the 2022 survey waves are minimal (Fig. 2). Of all the differences, only 4 out of 21 are statistically significant at P < 0.05 (|t| = 1.96–3.37 for the significant differences; |t| = 0.04–1.70 for the insignificant differences; two-sided t-tests; n = 178,740 in 2016; n = 148,460 in 2022), and the largest difference in absolute value is 0.019 (1.9 percentage points). Further, the two one-sided t-test procedure (TOST equivalence test) with equivalence bounds of −0.03 and 0.03—that is, 3 percentage point differences—yields rejection for all differences at P < 0.05 (t = 2.30–8.71 for one-sided t-tests against lower bound; t = −7.24 to −1.96 for one-sided t-tests against upper bound) (and rejection for all differences at P < 0.0005 for equivalence bounds of −0.04 and 0.04; t = 3.73 to 10.48 for one-sided t-tests against lower bound; t = −9.52 to −3.45 for one-sided t-tests against upper bound). This shows that rather than being sensitive to the repeated crises and protracted trends, the structure of European public attitudes towards asylum seekers is remarkably stable over time. These findings are also similar when we stratify the analysis by voters’ political ideology, age, education and income, or by the host country (Extended Data Figs. 2–4 and Supplementary Figs. 14–17).
A second major takeaway from the results in Fig. 2 relates to the role of country of origin. There is a statistically significant positive Ukraine effect in 2022, compared with a much smaller effect in 2016, suggesting that there have been some minor changes to European preferences in the face of the war in Ukraine. If forced to choose, Europeans in 2022 have a 5.5 percentage point higher probability of choosing a Ukrainian versus a non-Ukrainian asylum seeker (|t| = 11.71; P < 0.00001; two-sided t-test; n = 148,460), whereas in 2016 there is only a 0.9 percentage point effect (|t| = 2.13; P < 0.05; two-sided t-test; n = 178,740), where these percentage point effects are the estimates if all the other origins are pooled as the reference category. The 4.6 percentage point difference between these two is statistically significant (|t| = 7.18; P < 0.00001; two-sided t-test; n = 178,740 in 2016; n = 148,460 in 2022), and a TOST equivalence test of the 0.9 percentage point effect in 2016 with equivalence bounds of −3 and 3 percentage points rejects at P < 0.00001 (t = 9.03 for one-sided t-test against lower bound; t = −4.76 for one-sided t-test against upper bound). Nonetheless, the importance of Ukrainian origin is limited compared with the collective effects of the other attributes in 2022. For instance, in contrast to the 5.5 percentage point Ukraine effect, Europeans in 2022 expressed an 8.8 percentage point preference for Christians versus Muslims (|t| = 21.39; P < 0.00001; two-sided t-test; n = 148,460), a 7.6–14.2 percentage point preference for skilled professionals (accountants, teachers and doctors) relative to unemployed individuals (|t| = 13.70–24.88; maximum P < 0.00001) and a 6.0 percentage point preference for females versus males (|t| = 18.72; P < 0.00001).
In sum, European preferences with respect to the desirable features of asylum seekers have changed little across the Syrian and the Ukrainian humanitarian crises. It is in fact this resilience in preferences (rather than a sudden reaction to the state of affairs) that appears to be the primary driver of current support for Ukrainian refugees in Europe. Indeed, large-scale surveys carried out by the United Nations High Commissioner For Refugees (UNHCR) show that compared with refugees from other countries of origin, Ukrainian refugees are predominantly female, younger and Christian, they are more highly educated, and they are more likely to have worked in middle- and high-skilled occupations15. In light of this, our results suggest that the predominant source of support for Ukrainian asylum seekers in 2022 is the strong and longstanding preferences for specific traits that Ukrainian asylum seekers happen to possess. In other words, the main driver underlying the sympathy for Ukrainians is not a new attitudinal phenomenon, but rather a consequence of the socio-demographic composition and displacement profile of Ukrainian refugees.
As a more minor influence, we also find evidence for European solidarity as a new source of support for asylum seekers who are Ukrainian per se. Indeed, the small positive Ukraine effect that we see in the 2022 conjoint results is 2.7 percentage points higher among Europeans who possess higher sentiments of European solidarity compared with Europeans with lower sentiments (specifically, a 7.5 percentage point effect versus a 4.7 percentage point effect, the 2.7 percentage point difference of which is statistically significant at P < 0.01; |t| = 2.61; two-sided t-test; n = 148,460). Supplementary Information, ‘Additional analyses’ and Supplementary Fig. 18 provide further details, including a supplementary analysis that suggests that this effect may indeed be causally moderated27 by (rather than simply correlated with) European solidarity attitudes (Supplementary Fig. 19)—a result that resonates with the conscience collective discussed in ref. 16. Furthermore, we find that the Ukraine effect is specific to Ukraine during the invasion: we did not observe an effect of this magnitude in 2016 as described above, and it is also not driven by a general preference for all European asylum seekers, as we find no differential treatment of asylum seekers from Kosovo relative to those from Syria, Afghanistan, Eritrea, Iraq and Pakistan in 2022. For this latter comparison, we set the reference category to Kosovo and estimate the effects for Syria, Afghanistan, Eritrea, Iraq and Pakistan—none of the effects are statistically significant at P < 0.05 (|t| = 0.34–1.20; two-sided t-tests; n = 148,460), and TOST equivalence tests with equivalence bounds of −0.03 and 0.03 (3 percentage point bounds) yield rejection for all effects at P < 0.0001 (t = 3.74–5.43 for one-sided t-tests against lower bound; t = −6.18 to −4.46 for one-sided t-tests against upper bound).