European attitudes to refugees after the Russian invasion of Ukraine

Abstract

Why do attitudes to refugees vary? An original panel is used in five EU states – France, Germany, Hungary, Italy and Poland – to explain European attitudes towards three groups of refugees following the 2022 invasion of Ukraine. It is shown that European attitudes to Ukrainian refugees are determined by predispositions to immigration and perceptions of the war and actors involved, with European identity and contact with refugees being relatively unimportant. These findings are validated with dynamic panel models and attitudes towards the Temporary Protection Directive. A ‘spill-over’ effect is further demonstrated, whereby attitudes to Ukrainian refugees positively affect attitudes to Afghan and Somali refugees, and a declining ‘rally-around-the-flag’ effect over time. These findings contribute to the literature on attitudinal formation, showing the relative malleability of attitudes to refugees as a function of their embeddedness in broader attitudinal patterns (particularly to immigration and geopolitics), changing context (the different stages of the war), and spill-over from views towards other refugee groups.

Keywords:

• Attitudes to refugees
• invasion of Ukraine
• attitudes to immigration
• preferences
• panel data
• threat perceptions

Why do individual attitudes to refugees vary? Are they primarily a function of one’s attitudes to immigration more broadly or do they have distinct contextual determinants? To what extent are attitudes to different types of refugees distinct from each other and how do they interact? To what extent are they formed by non-immigration issues, particularly those related to the cause of the refugee flow? To answer these questions, we examine the case of European attitudes to Ukrainian, as well as Afghan and Somali, refugees following Russia’s 2022 invasion of Ukraine using an original two-wave panel in France, Germany, Hungary, Italy, and Poland. The invasion of Ukraine led to the mass arrival of refugees escaping war in Europe, echoing the smaller in scale though more politicised 2015–2016 ‘migration crisis’. The 7.2 million Ukrainian refugees in Europe as of September 2022, around 4 million of whom are registered for temporary protection, form part of the growing number – around 90 million – of refugees worldwide (UNHCR 2022). Previous refugee flows – notably the arrival of 1.3 million Syrian (and other) refugees and migrants to Europe over the course of 2015–2016 – led to an increased politicisation of immigration in Europe and beyond, affecting party systems, government formation, membership of supranational organisations, and arguably the post-war rights-based consensus (Dennison and Geddes 2018, 2019). Moreover, refugees are highly vulnerable populations, disproportionately likely to suffer both mental and physical health consequences of displacement, trafficking and other forms of exploitation and discrimination in their host countries (Mendola and Pera 2022). Conversely, effective refugee policies have the potential to minimise these risks, uphold a legal- and rights-based migration regime, and offer genuine refuge until the opportunity to return arises.

Understanding attitudes to refugees is of profound substantial importance. First, given the increasing influence of public opinion on migration policy (Böhmelt 2021; Dennison and Geddes 2021a), understanding attitudes to refugees predicts how sustainable various policy programmes are likely to be and why. Furthermore, explaining such attitudes allows us to design interventions to reduce misperceptions and increase support for internationally agreed-upon objectives including ‘safe, orderly, and regular’ migration. Understanding attitudes to refugees is also of profound scientific interest: the extent to which we observe change in such attitudes, how they (and their determinants) vary from attitudes to immigration generally, and how and why attitudes to different refugee groups and crises vary offer evidence in support of various and at times competing theories of attitudinal formation generally.

European attitudes to Ukrainian refugees are an especially useful case to consider these questions for several reasons. The current wave of refugees to Europe comes reasonably soon after another wave of refugees that reached a peak in 2015–2016. Initial evidence suggest that the reception of Ukrainian refugees has been more positive than the reception to previous waves of migrants. This has led to speculation that attitudes to refugees vary according to a long list of proposed factors, including: the geographical, cultural, ethnic, and identity-based proximity of the specific refugee group; levels of understanding and perceptions of the source of the refugee crisis (in this case, the Russian invasion), the gender make-up of the refugees, changing attitudes to immigration generally, and the role of the media and politicians in influencing public opinion.

We hypothesise that – in addition to predispositions to immigration – attitudes to refugees are determined by perceptions of the specific refugee crisis and actors involved and spill-over effects from other attitudes to other refugee groups. Attitudes should also change over time according to these factors. Our original two-wave panel survey includes questions measuring these explanatory variables, as well as questions measuring attitudes towards Ukrainian refugees, the Temporary Protection Directive, and, to allow for comparability, attitudes towards Afghan and Somali refugees. We show that European attitudes to Ukrainian refugees are determined by predispositions to immigration and perceptions of the war, with European identity and contact (except in Poland) being relatively unimportant. We validate these findings with dynamic panel models and attitudes towards the Temporary Protection Directive. We further demonstrate (1) a ‘spill-over’ whereby attitudes to Ukrainian refugees positively affect attitudes to Afghan and Somali refugees, (2) a declining general ‘rally-around-the-flag’ effect over time, and (3) considerable individual variation over time that is explained by shifting attitudes to the war and to immigration. Our findings have ramifications for our understanding of the nature and sources of attitudes to refugees from Ukraine as well as attitudinal formation in general. We thus contribute to theories emphasising that stronger, more salient attitudes (in terms of the war and Ukrainian refugees) affect weaker, less salient ones (i.e. attitudes to Afghan and Somali refugees), as well as theories that emphasise the relative importance of broad predispositions in shaping specific preferences, and how new information can drive attitudinal change (Ajzen 2001; Zaller 1992).

Theoretical considerations

What factors explain attitudes to refugees? And what factors are likely to affect Europeans’ policy preferences to Ukrainian refugees, in particular? Although the latter phenomenon is novel, there exist preliminary findings and the far larger and older literature on attitudes to immigration – and, to a lesser extent, refugees – in general which we can use to form hypotheses.

Early studies have already considered attitudes to Ukrainian refugees following the 2022 Russian invasion. De Coninck (2023) argues that greater support amongst Europeans and Americans for admission of Ukrainian than Afghan refugees results from a shared ‘collective conscience’ determined by cultural and political proximity, ethnicity, and the lower perceived threat posed by the migrants, combined with the greater threat posed by Russia compared to the Taliban. Similarly, Moise and Oana (2022; see also Zhou et al. 2022; Kirk 2022; Pettrachin and Abdou 2022; Drazanova 2022) postulate that the greater national-level willingness of Europeans to accept Ukrainian refugees – compared to previous inflows from Africa and Asia – results from: geographical proximity; perceived urgency and moral unambiguity of the war; cultural proximity and significant intergroup contact prior to the war; a common European identity; a shared sense of history and perception of threat vis Russia; the high proportion of women and children amongst Ukrainian refugees (and by contrast the high proportion of young adult males in previous waves); as well as (white) ethnicity and Christian (rather than, especially, Muslim) background.

Whereas these initial analyses focussed on explaining national averages in willingness to admit different types of refugees, we are concerned with explaining individual-level variation in attitudes to Ukrainian and other refugees, the determinants of which are likely to be more expansive. Attitudes to Ukrainian refugees are likely to be determined by one’s broad predisposition to immigration in general. Substantively, admitting refugees is a form of immigration, so attitudes to this specific type fall under the broader umbrella of attitudes to immigration, which have been recognised as largely strong and stable in individuals (Dennison and Geddes 2019; Kustov et al. 2021) making them relatively resilient to external stimuli (Druckman and Leeper 2012; Erikson 2017). On the other hand, Page and Shapiro (1992) characterise attitudes in general as genuine and stable yet rationally responsive to political situations. Moreover, they argue that an increasingly nuanced understanding of a political object – such as immigration – allows distinctive attitudes to sub-types of an object, with distinct attendant beliefs and attitudinal formation processes, to manifest in an individual’s consciousness (Druckman and Lupia 2000: 6). As such, because attitudes to immigration are strong and stable, they are likely to be strong predictors of attitudes to Ukrainian refugees. However, the high profile of the novel event of this refugee inflow may mean that European attitudes to Ukrainian refugees are partially independently formed from broader attitudes to immigration.

Attitudes to the admission of refugees are likely to be determined by one’s perceptions regarding the specific refugee crisis and in particular its causes – in this case, the Russian invasion of Ukraine. Retrospectively, whom one sees as at fault for the outbreak of war is likely to determine perceived levels of ‘deservingness’ and the genuineness of Ukrainian refugee status. Therefore, seeing Russia’s stated grievances (e.g. supposed security concerns regarding NATO expansion) as legitimate or their objectives (e.g. annexation of Ukrainian territory) as tolerable is likely to lower support for the admission of Ukrainian refugees (Bjånesøy 2019; Findor et al. 2021; Lawlor and Paquet 2022; Reeskens and van der Meer 2018). Approval or disapproval of conduct during the war by, on the one hand, Russia and, on the other, Ukraine and its allies is similarly likely to affect one’s beliefs in the ‘deservingness’ and genuineness of the refugees. Prospectively, belief that Russia represents an ongoing common threat beyond Ukraine is also likely to affect such attitudes (Dennison and Geddes 2021b). Such determinants link more broadly to geopolitics, echoing how refugees during the Cold War were ‘often welcomed as a way to discredit the Soviet bloc’ (Braithwaite et al. 2019: 7). This leads us to hypothesise that:

Hypothesis 1: Attitudes to Ukrainian refugees are determined by perceptions of responsibility for the war and on-going security threats posed by conflict participants: Russia, on the one hand, and Ukraine, the US, and NATO, on the other.

A range of other factors related to the specific crisis and refugee group are likely to also have effects, albeit weaker or more distal ones than perceptions of the conflict. These include: (1) sense of a common identity (see Sniderman and Hagendoorn 2007, for effects on attitudes to immigration) – in this case, ‘European’ (see Carl et al. 2019; Dennison et al. 2020, for effects of feeling European; Hooghe and Marks 2005); (2) contact with members of the refugees’ social group (Ukrainians), which may increase empathy and support for admission of members of that group (Allport 1979; Paluck et al. 2019; Clayton et al. 2021; by contrast, fleeting transitory contact has been shown to trigger group threat, e.g. Dinas et al. 2019; Rudolph and Wagner 2022; Gessler et al. 2022); (3) political cues about the crisis, which have been shown to be particularly effective when issues are novel, complex and related to specific episodes, such as the Russian invasion or the ‘deservingness’ of Ukrainian refugees (for cues and attitudes to immigration, see Hellwig and Kweon 2016; Ha et al. 2016; Turper et al. 2015; on cues and ‘deservingness’, see Lawlor and Paquet 2022); and (4) left-right political orientation – although Bansak et al. (2016) show being left-wing leads to positivity to refugees due to humanitarian concerns and weaker anti-Muslim bias, the tendency of right-wing individuals to value security highly may lead to reverse effect, particularly regarding a common aggressor (Dennison et al. 2021). Moreover, the country in which one lives is likely to capture numerous ecological factors including longer-term socialisation and historic effects (Meidert and Rapp 2019).

There are also reasons to expect significant over-time variation. On the one hand, support for refugees may become more negative as the initial support – akin to a rally-around-the-flag effect – wanes due to weakening media interest (and thus stimuli), personal interest, emotional investment, and greater consideration of the costs of supporting refugees (Lorenz-Spreen et al. 2019; for analogy following Covid-19 pandemic outbreak, Altiparmakis et al. 2021; Baekgaard et al. 2020; Bol et al. 2021; Esaiasson et al. 2021; Schraff 2021). Should this happen, we also expect an individual’s attitudes to immigration generally to increasingly affect refugee attitudes as the effects of a novel context and attendant cues wane and individuals revert to their pre-dispositions. In this scenario, attitudes towards Ukraine, NATO, and the US would either become less predictive of support for refugees or would themselves become more negative. On the other hand, support for refugees may become more positive over time via normative socialisation and cueing (see above) into the majority view or as individuals learn more about the conflict and their perceptions of geo-politics change (see Czymara and Schmidt-Catran 2017, for analogous phenomenon in Germany during the 2015 ‘refugee crisis’). In this scenario, we would expect the association between attitudes towards Ukraine, NATO, and the US, and refugee views to become stronger, or the attitudes themselves to become more positive. Conversely, in this scenario, immigration attitudes would become less predictive, or would become more positive over time. This leads us to hypothesise that:

Hypothesis 2: Attitudes to Ukrainian refugees become more (H2a) negative over time due to a reversion to general immigration attitudes or (H2b) positive over time due to cumulative normative socialization, cueing, and information processing.

Finally, we also consider two – again, potentially competing – ‘spill-over’ effects of attitudes to Ukrainian refugees on attitudes to other contemporary refugee groups, such as Afghans and Somalis (see Costa-I-Font and Ljunge 2022, on ‘spill-overs’). On the one hand, this effect may be positive, particularly if attitudes to Ukrainian refugees are ‘stronger’, since: (1) more strongly held attitudes cause more weakly held ones to change to avoid cognitive dissonance (Ajzen 2001), creating cognitive networks in which attitudes are embedded (Dalege et al. 2017); (2) individuals are strongly motivated to control their own internal prejudices and ‘will deliberately seek to control actions, expressions or thoughts that can be deemed to violate these norms’ (see Motivation to Control Prejudice theory; e.g. Schwartz et al. 2021: 1169), which may become apparent in the contrast of support for Ukrainian refugees but opposition to Afghan or Somali ones; and (3) information gained about refugees from one crisis (the invasion of Ukraine) may update beliefs about refugees generally (Page and Shapiro 1992). On the other hand, it may be that support for Ukrainian refugees has a negative spill-over onto other groups as the Ukrainian crisis is deemed of higher relative importance and more deserving of limited resources, as attention to the other crises is diverted away, or as ambivalent or ambiguous feelings towards other refugee groups are highlighted by unambivalent support for Ukrainian refugees (again, see Czymara and Schmidt-Catran 2017, for analogous phenomenon in Germany during the 2015 ‘refugee crisis’; see Boninger et al. 1995, on ordinal limits of humans to deem issues important; see Bansak et al. 2016, on the effects of anti-Muslim prejudice on attitudes to refugees). This leads us to the final set of hypotheses:

Hypothesis 3: Greater positivity to Ukrainian refugees leads attitudes to Afghan and Somali refugees becoming more (H3a) positive over time as individuals avoid dissonance, control their prejudice, and update their information or (H3b) negative over time, as individuals prioritise the needs of Ukrainian refugees in their preferred distribution of resources.¹

Data and context

Our data was collected as part of two surveys conducted in five EU countries (Germany, France, Italy, Poland, and Hungary) in the framework of the SOLID research project.² Interviews were administered in March (hereafter called wave 1) and then in July 2022 (hereafter, wave 2) on national samples obtained using a quota design based on gender, age, macro-area of residence (NUTS-1), and education.

The total sample size, including responses from both waves, is 22,600. However, part of our analysis focuses on the panel in our survey, of which we have 12,676 observations, two responses each from 6338 respondents.³ Table A3 in the online appendix compares our sample for the quota criteria (gender, age, and education) with the European Social Survey sample – the gold standard in European survey research. While there are some minor deviations for some categories (e.g. the oldest age cohorts are somewhat underrepresented in Hungary and Poland, and the youngest cohort is underrepresented in Germany), the overall distributions are quite similar. We also conduct attrition analysis to see whether respondents who stayed in the panel for our second wave differed from those who dropped out. Table A4 in the online appendix shows the results where again we see few dissimilarities, mostly in terms of demographics. Importantly, when we rerun our main analysis on the subgroups of respondents who remained and those who dropped out, we see nearly identical results (Table A4, columns 2 and 3).

The timing of our surveys comes at two crucial moments for the Russian invasion of Ukraine. Wave 1 was carried out between 11 March and 5 April 2022, a period when the beginning of the war dominated media channels across Europe. The share of respondents who viewed the war as the most serious threat to the survival of the European Union was 32.3% in wave 1, compared to 29.0% in wave 2. At the time, European countries had already jointly accepted more than 3.2 million Ukrainian refugees. This figure rose to 5.8 million by July 2022 (UNHCR 2022). Figure 1 shows the number of refugees by country in absolute numbers, as well as relative to the population. We consider this period to be crucial since it captures respondents’ reactions to the war in a period of high salience. For our purposes, it also captures their attitudes towards Ukrainian refugees at a time when their numbers were lower in relative terms and when there was uncertainty regarding how long they would need to be accommodated and how high a burden they might pose for each state. While this does not constitute an absolute baseline for refugee attitudes, it does constitute a different context compared to the second wave. In terms of the war itself, during this period the whole of Ukraine was under attack, and there was great uncertainty over the fate of all civilians, particularly as images surfaced of atrocities in Bucha and Mariupol.

Figure 1. Number of Ukrainian refugees by country, Sept. 2022. Note: these are the number of refugees that were settled in the country at the time. The number of refugees transiting Poland for example, was much larger, about 5 million by July 2022 (UNHCR 2022).

Our wave 2 was conducted between 8 and 28 July 2022. This period was marked by a considerable decrease in media salience of the war as other domestic topics took over, including inflation and energy prices. However, Europe by this point had accepted a much larger number of refugees who had already stayed in their new host countries for up to five months. Part of the uncertainty of the war was lifted as it became clear that the war would likely continue for some time, meaning that refugees would need to be hosted for a longer period. By this point, the costs of hosting refugees also became clearer. The war, by this point, had moved largely into the East and South of the country, with most of the fighting occurring in the Donbas region of Ukraine, with the rest of the country experiencing varying degrees of return to some type of normality. Five months into the war, EU countries had engaged in several debates regarding refugees, amongst other war-related topics, including energy policy, sanctions, and military support for Ukraine. All these debates led to disagreements which may have polarised EU citizens to a greater degree than in March (NYT 2022). These differences in the context of our two waves lead us to expect a diminished rally-around-the-flag effect by the time of the second wave.

Finally, the countries included in our sample also provide for considerable heterogeneity regarding their geopolitical positioning and their role in the refugee crisis. Poland and Hungary share a historical experience of Russian occupation, like Ukraine, which they directly border. We expect respondents in these two countries to be more favourable to Ukrainians given these historical and cultural ties. Regarding refugees, they can be considered frontline states, which had first points of entry. Poland is also the most important destination state since many refugees opt to stay due to linguistic, cultural and geographical proximity (see Figure 1). By contrast, Poland was a bystander and Hungary a transit state during the 2015–2016 refugee crisis, when they both opposed EU refugee solidarity as part of the Visegrad 4 coalition. France, Germany, and Italy, on the other hand, are likely to have lower perceptions of threat as posed by the Russian military and lower cultural proximity to Ukraine. All three countries can be considered destination states for Ukrainian refugees. In the 2015–2016 refugee crisis, France and Germany were destination states, while Italy was a frontline state. These past experiences likely shaped their current positions on refugees and EU-level burden sharing.

National-level differences

Our analysis focuses on two main questions answered by respondents in both surveys. The first question (outcome 1) asks the respondents their level of agreement on an 11-point scale with granting the right to stay to those fleeing the war in Ukraine.⁴ Our first question therefore considers the national level response to refugees, without referring to possible costs or benefits of hosting refugees. We also asked respondents what their level of agreement was to accepting refugees from Afghanistan and Somalia, using the same wording as outcome 1, for purposes of comparison.

Our second question (outcome 2) asks the respondents’ level of agreement on an 11-point scale with the Temporary Protection Directive of the EU, which grants automatic protection to refugees from Ukraine, including a residence permit, and access to employment and social welfare for up to three years. Thus, our second question considers the EU-level of the refugee response and explicitly considers the benefits allotted to refugees and also the costs to taxpayers. The perceptions of these costs may be ameliorated by the fact that refugees are given the right to work and thus contribute to their host states.

Figure 2 shows the average response by country for each wave of our survey to our two outcome variables of interest, including the acceptance of refugees from Afghanistan and Somalia. The figure highlights that across countries and waves, support for Ukrainian refugees is considerably higher than for Afghan or Somali refugees. Given our 11-point scale, the red line at five points indicates whether there is an average agreement in favour of refugees. Only in Italy do respondents show positive attitudes towards Afghan and Somali refugees, in addition to Ukrainians. We only partly explore the reasons for the differences between different types of refugees. For a more extensive discussion, see Moise and Oana (2022). Second, across countries and waves respondents are slightly less in agreement with outcome 2 than outcome 1. We interpret this as implying that respondents also consider the costs of taking in refugees in the Temporary Protection Directive question. The effect of the ability to work might cut both ways. On the one hand giving refugees the ability to work allows them to contribute to state resources. On the other hand, certain respondents might perceive them as a threat on the labour market.

Figure 2. Outcome variables across waves.

Finally, between the two waves, we see a slight waning of support for Ukrainian refugees across the two outcome questions. We take this as a first indication in favour of our hypothesis of an initial rally-around-the-flag effect that weakens over time.

Research design and covariate measurement

We are interested in both the static and dynamic nature of attitudes towards refugees. We therefore employ a variety of models to explain the within- and between-individual variation in attitudes towards refugees. In a first step, we use OLS with country fixed effects to identify the determinants of between-individual levels of support for the first wave of our survey. In a second step, we model the within-individual variation with an individual fixed effects model. In a third step, we model the between-individual change in attitudes across the waves, considering levels and differences of our predictors. Lastly, we conduct two further analyses to see how our predictors of interest change in importance across time: a pooled OLS with individual clustered standard errors interacting wave with our two main predictors, and a multinomial analysis of how individuals change between the waves.

We measure predisposition to immigration with three questions with wording taken from the European Social Survey, which ask respondents the extent to which they favour immigration of the same ethnic group, of a different ethnic group, or from poorer countries outside Europe, respectively. These three items form a strong factor,⁵ which we use in our analysis. We measure attitudes towards the refugee crisis (and the war from which the crisis originates) with a variety of questions. Trust towards Ukraine, the United States, and NATO, respectively, and approval of the actions of these three actors form a coherent factor. We measure trust towards Russia separately. Another set of variables measure identity and proximity to the refugee group. We operationalise the individuals’ primary identity (borrowing a measure from the Eurobarometer) as either European or national.⁶ For the left-right orientation we split responses on an 11-point scale into four categories. Left and right-wing are defined as scoring below and above five on the 11-point scale, respectively. Those scoring 5 were labelled as non-ideologically affiliated while non-respondents were coded separately. We also measure levels of threat perception towards people who are close to the respondents. We treat these perceptions as a proxy for personally knowing Ukrainians or other people directly affected by the war. We rescaled all explanatory factors to the 0–1 range to make their effects directly comparable. We also provide interpretations in terms of standard deviation changes below and in the online appendix. Tables A1 and A2 in the online appendix provide descriptive statistics for all variables of interest.

Results

Static analysis

Figure 3 shows the results of our static wave 1 analysis, using country fixed effects (country fixed effects not shown, see Table A6 in online appendix). Coefficients are shown for our two outcomes of interest, as well as for support for Afghan and Somali refugees.

Figure 3. Static analysis – first wave, 5 countries: OLS coefficients. Note: country fixed effects not shown. N = ∼8000. See Table A6 in the online appendix for further details.

We see that attitudes towards Ukrainian refugees are primarily associated with underlying immigration attitudes and attitudes towards the crisis (i.e. the war). These effects are substantial: going from the minimum to the maximum general support of immigration increases the support for Ukrainian refugees by around 35 percentage points.⁷ The association with attitudes towards the war is even stronger. Trust in Russia has a negative association of more limited proportions once accounting for attitudes to Ukraine/US/NATO: its maximum effect decreases support for Ukrainian refugees by slightly more than 10 percentage points. The association of immigration attitudes with the second outcome (the Temporary Protection Directive: TPD) is more limited, though still substantial. As expected, taking the costs, as well as possible benefits, of refugees into account reduces the effect of the overall immigration attitude. European identity and contact (perceived threat to close ones) have much weaker or insignificant effects. The effect of contact on the TPD is positive, while its effect on support of Ukrainian refugees is insignificant. The effect of contact is in fact driven by respondents from Poland (see Table A8 in the online appendix). This supports our use of this variable as a proxy for contact since respondents from Poland are more likely to have acquaintances or family in or from Ukraine. Overall, the differences between our main dependent variable and the TPD are substantively small. Both variables attest to the importance of war and immigration attitudes. Differences between the two variables may be due to different elements that are specified in more detail in the TPD: the EU level, employment, benefits, and limited time horizon. We thus find support for H1.

We note several interesting dynamics when comparing support for Ukrainian refugees to support for Afghan or Somali refugees. Immigration attitudes have a stronger effect for the latter two. This may be a result of the fact that refugees from these countries are culturally perceived to be more different from Europeans than Ukrainian refugees are, that they are more likely to be men, and they are more likely to be perceived as economic migrants than refugees. By contrast, support for Ukrainian refugees is strongly impacted by other factors, such as views of the war, while the factor capturing attitudes towards Ukraine, the US, and NATO is much less relevant for support for Afghans and Somalis and trust in Russia is a trivial factor.

The most important additional factor that distinguishes between support for the different groups is one’s general left-right orientation, i.e., political ideology. We note that compared to right-wing respondents (our baseline category), left-wing voters are much more likely to support Afghan and Somali refugees, whereas reporting no strong ideological views or no views at all has a smaller effect in the same direction. By contrast, being left-wing does not increase support for Ukrainian refugees compared to being right-wing.

When analysing the second wave of our survey we notice very similar results. To test for the differential effect of our main explanatory variables of interest, attitudes to the war and attitudes to immigration, we pool together observations from the first and second waves and run an interaction between our wave dummy and the two independent variables of interest. Table A7 in the online appendix shows the results for attitudes to Ukrainian and Afghan refugees.⁸ We first note the negative coefficient for our wave dummy, confirming what we see in Figure 2, that attitudes are becoming more negative over time. We see that for Ukrainian refugees both interactions are significant and positive. Thus, both attitudes towards migration and towards the war become more salient as the war advances. As the descriptive statistics in Tables A1 and A2 show, both immigration attitudes and war attitudes become more negative in the second wave. Together, they partly account for the overall small negative shift of attitudes towards refugees. We thus find initial support for H2a over H2b. We further explore these dynamics in the next section. We note also that these interactions are not significant for Afghan refugees.

Dynamic analysis

We have so far explored the static, between-individual variation in our sample. We now turn to the dynamic part of our analysis and start with within-individual variation. We observe a considerable change in the outcome variables between the two waves. We show this in several ways. First, we have divided the outcome variables into three categories by summarising the 0–4 categories into a ‘contra’ position and the 6–10 categories into a ‘pro’ position.⁹ We kept the 5 category as a separate ‘undecided’ position. Considering a respondent’s positions over the two waves, we arrive at seven types of respondents: stable pro, stable contra and stable undecided types, as well as four types who changed sides, either from pro or undecided to contra or from contra or undecided to pro. We expect significant stability between the two waves, given that they are only four months apart and the widely reported stability of attitudes to immigration (Kustov et al. 2021). However, we find sizable change, as is shown in Table 1. Roughly a fifth of the respondents changed sides on Ukrainian refugees and on the TPD, with change from pro to contra prevailing. Roughly sixty percent of those who changed sides turned against Ukrainian refugees/the TPD, and forty percent changed in favour of them/the TPD. Attitudes towards Afghan and Somali refugees changed in a similar way. As Table A5 in the online appendix demonstrates, the firmer the respondents’ attitudes are to begin with, the more the probability to change sides decreases. But all respondents have a non-zero probability to change. Moreover, at all levels of intensity, the probability to change is higher from the pro- to the contra-side. This pattern of changing attitudes provides yet another indication of overall decreasing support for refugees (H2a).

Table 1. Changing support of refugees and of the temporary protection directive: percentages.

	Ukrainian refugees	Afghan refugees	Somali refugees	TPD
Con-con	12.2	32.9	33.2	14.7
Pro to con	7.2	5.3	5.3	6.8
Undecided to con	4.4	6.8	6.9	4.4
Undecided	9.2	11.1	11.8	8.3
Undecided to pro	5.4	5.1	5.3	5.4
Con to pro	2.2	4.4	2.7	2.7
Pro-pro	59.4	34.5	33.4	57.8
Total	100.0	100.0	100.0	100.0
n	6338	6338	6338	6338

Second, we run an individual fixed effects analysis on the continuous outcome variables. Figure 4 plots the results, while Table A9 in the online appendix shows the details. We include only factors that could have plausibly changed during the four-month period between our two surveys. The results largely corroborate our static analysis and lend further causal weight to our main findings. Confirming the static results, changes in threat perception and identity are not relevant, while changes in immigration attitudes and perceptions of the war are associated with changes in support for refugees. We note a much lower effect of immigration attitudes and trust in Russia on temporary protection compared to outcome 1 of simply admitting Ukrainians.

Figure 4. Panel fixed effects: GLS regression coefficients. Note: see Table A9 in the online appendix for details on coefficients and model fit. N = ∼10,000. Groups = 2.

We also note that growing distrust of Russia increases support for Ukrainian but not for Afghan or Somali refugees. Trust in Ukraine/US/NATO has a stronger effect for Ukrainian refugees while also enhancing support for Afghan and Somali ones. As discussed in the theory section, it is likely to result from a ‘spill-over’ effect whereby factors that make individuals more welcoming to Ukrainian refugees might make them more welcoming to other types of refugees as well.

To further test for this possibility, we rerun the models for Somali and Afghan refugees and include the change in support for Ukrainian refugees as an independent variable. Figure 5 shows the coefficient plot for this effect. We cannot explore the mechanisms for this effect further with our data, but we posit several reasons for why this spill-over effect might be plausible. The first is a general empathy mechanism. If European respondents did not previously identify with the struggles of Afghan and Somali refugees, the experience of empathising with Ukrainian refugees may have spilled over into empathy for other human beings fleeing conflict. A second mechanism may simply be a cognitive dissonance mechanism. Respondents, during the four months of the war, might have felt a tension between their views on Ukrainian and other refugees, and resolved to increase their support for other refugees. The fact that the media has repeatedly highlighted the hypocrisy of supporting Ukrainian versus other refugees lends further support to this hypothesis.

Figure 5. Effect of increase in support for Ukrainian refugees on other refugees: GLS regression coefficients. Note: see Table A11 in the online appendix for details on coefficients and model fit. N = ∼10,000. Groups = 2.

While we cannot distinguish between these mechanisms, we can rule out certain possibilities. First, reverse causality is not likely to be an issue here given that there were no obvious international events that could have changed attitudes towards Afghans and Somalis apart from the war in Ukraine and the ensuing refugee crisis. It is thus extremely unlikely that changes in attitudes towards these groups influenced attitudes towards Ukrainians.

Second, we are also able to robustly rule out ordering effects. Our questions on the three refugee groups are randomised, meaning that our average effects control for ordering. However, we also conduct an analysis to see whether the ordering of the questions affects the outcome. Table A11 in the online appendix shows that there is no ordering effect in the fixed effects analysis. Changing the order in which one saw the three questions from wave 1 to wave 2 has no impact on whether one changes their attitudes to Afghan or Somali refugees, and the coefficient for the spill-over effect remains unchanged. For further robustness, we rerun the analysis on the subgroup who received the Afghan or Somali question first and find nearly identical results (Table A11, columns 3 and 4). Third, we do not expect social desirability¹⁰ to play a role here. While social desirability might explain why an individual might rate the acceptance of Afghan or Somali refugees higher in a single survey, it cannot account for the within-individual effect across the surveys.¹¹ We thus find robust support for H3a.

The third way that we model change is a dynamic analysis, combining between and within-individual variation. In this final model we collapse our dataset and take the difference of the outcome variables between the second and the first wave of our survey. Independent variables are taken both as differences (D) between the waves and as the absolute levels (L) from the first wave. We also include the lagged dependent variable as a predictor.

Figure 6 shows the coefficient plot for this final model. The findings largely confirm our results from previous models. The effects of both the absolute levels as well as the differences between waves follow a similar pattern. Attitudes and changes of attitudes towards refugees are determined primarily by levels and changes of immigration attitudes, and by perceptions of the war. Other factors are either statistically insignificant or substantively trivial.

Figure 6. Dynamic analysis of difference: OLS regression coefficients. Note: Country fixed effects not shown. N = ∼4400. See Table A12 in the online appendix.

We highlight the fact that both sets of substantively important attitudes have larger coefficients for the difference compared to the level. This confirms that attitudes towards refugees are in flux and not set in stone. It is therefore plausible that further changes in the dynamic of the war will shift these attitudes. We further investigate these over-time dynamics using our trichotomized variables to see if those who become more positive towards refugees are different from those who become more negative. Tables A13 and A14 in the online appendix show the results.¹² We see that both the level and the change in pro-immigration attitudes and attitudes towards the war have stronger effects for those becoming more anti-refugee than for those becoming more pro-refugee. This is yet another reason why attitudes get more negative over time (H2a).

Discussion

How do our generalisable theoretical framework and findings solve the specific puzzle of why Europe accepted seven times the number of refugees that generated a deep political crisis in 2015–2016, without obvious, immediate political repercussions? We explored the differences in European preferences towards the acceptance of refugees from Ukraine and those from Afghanistan and Somalia. We find that three factors, from those we explore, explain the difference between these groups.

First, in the background of refugee attitudes stand immigration predispositions. Those with more favourable predispositions to immigration are more likely to accept refugees. We note that these attitudes are not fully deterministic. They are more predictive of support for Afghan and Somali refugees than for Ukrainian ones. The effect of these attitudes also changes over time, becoming stronger following an initial crisis or high-profile event.

Second, and most importantly, the context of the war explains much of this puzzle. Europeans feel involved in this conflict. European governments are arming Ukraine and sanctioning Russia. Individuals who support Ukraine, the US, and NATO, and distrust Russia, are therefore more likely to support Ukrainian refugees as well. Europeans who follow the war are more likely to be taking cues from European elites, which for the most part stand united for Ukraine. Europeans might also feel more moral responsibility for refugees from a neighbouring country that they feel was unjustly attacked by a belligerent neighbour. Simply put, the extraordinary event of having a war on their doorstep fundamentally shifted Europeans’ perspectives on refugees fleeing that war. Notably, when these attitudes towards the war shift, so do attitudes to refugees.

Thirdly, we demonstrate a robust spill-over effect, whereby attitudes to Ukrainian refugees shift attitudes to other types of refugees (Afghan and Somali). Although our data do not allow us to probe the mechanisms for such an effect, we offer several plausible explanations, including cognitive dissonance, greater awareness, and empathy spill-over. Whether these effects are long-lasting remains to be seen. Future research should seek to further identify and explain this mechanism.

Our findings also show a small but important role of political identity and conflict. The Russian war in Ukraine was universally condemned by European elites across the political spectrum. While Europeans disagree about sanctions, arms deliveries, and energy, support for refugees is widely accepted. We therefore do not see the left-right divide that is present when looking at refugees from other countries. This is likely due to proximate culture and identity that, when more distant, are typical immigration concerns of those on the right and interact with a broader personal value of the right: security (Dennison et al. 2021). Right-wing respondents do not feel the same threat to their culture and identity from Ukrainian refugees as they do towards refugees from Muslim countries. Across all countries, however, right-wing individuals remain strongly anti-immigration in general. This creates a tension in the right-wing, including the populist radical right, across European countries, with implications for the ability of Russia to use these parties as a means of creating opposition towards the mainstream policy of supporting Ukraine.

Our panel survey and dynamic analysis highlighted the importance of changes in attitudes on refugees. At the start of the war, Europeans were incredibly supportive of refugees. However, this occurred in a context of uncertainty over the duration and intensity of the war. The shock of the war likely outweighed considerations of the costs that countries must bear. This rally-around-the-flag effect meant that individuals were more likely to support solidaristic policies despite costs. Should the war continue and the costs to European countries increase, both those coming from refugees as well as from energy prices and sanctions, opposition to European support for Ukraine could increase, in which case attitudes to Ukrainian refugees would shift.

For the moment Europeans remain strongly supportive of Ukrainian refugees. This support is slightly lower five months into the war than at the start. Furthermore, when asked whether they would support refugees for an indefinite period, this support further drops (Figure 7¹³).

Figure 7. Attitudes to long-term support of Ukrainian refugees by country: average scale values.

Public opinion surrounding support for refugees is relevant for policymaking at several levels. First, public backlash to refugees can constrain elites when they consider accepting refugees. Far right parties can ride the wave of dissent to threaten the political order. The 2015–2016 refugee crisis showed how one-seventh the number of refugees seen now could disrupt EU and national politics. The much higher acceptance of refugees in this crisis and the low polarisation on the issue of refugees is a reason for why policymakers were and continue to be able to have pro-Ukrainian refugee policies. It is also the reason why far-right parties have not so far been able to capitalise on this crisis.

Second, given the sheer number of refugees, at least in some destination countries and at least at the peak of the crisis, governments were not able to process them logistically. They therefore partly relied on individuals to host refugees directly in their homes. Figure 8 shows the share of individuals in each country who had (by their own reporting) hosted refugees by July 2022 or who were willing to do so unconditionally, conditional on receiving state aid, or were unwilling due to lack of space in their homes or unwilling in general.

Figure 8. Willingness to host refugees.

Conclusion

In this article we offered and tested a novel theoretical framework to explain attitudes to refugees and, in doing so, sought to explore the reasons why this influx of refugees has received a distinct reaction in terms of public opinion to the 2015–2016 ‘refugee crisis’. Refugees have been coming into the European Union faster and in much larger numbers than in 2015–2016, yet without causing the social and political disruptions seen then. We argue that a key part of the explanation for this puzzle lies in the specificities of the crises which drive different dynamics of public opinion. Simply put, Europeans are much more accepting of Ukrainian refugees than they were of Syrian refugees and than they currently are of refugees from Afghanistan or Somalia. This in turn is likely an important reason for why elites have managed to stay united in their strong support for refugees. Initial and continued elite unity may also be responsible for unity seen in public opinion.

We find that immigration predispositions are much more strongly predictive of support for refugees from outside of Europe, even though they still retain strong explanatory power for Ukrainian refugees as well. The context of the war – and attitudes to it – makes the difference for this crisis. Individuals who support Ukraine, and the efforts of the West to back it, and who distrust Russia, are much more likely to support Ukrainian refugees. Secondly, distinctly from support for other refugees, left-right self-placement does not strongly predict whether someone supports Ukrainian refugees. At both the elite and the individual level support for Ukrainian refugees is a valence issue, supported broadly.

Our dynamic analysis shows that attitudes to Ukrainian refugees in this crisis are to some degree malleable. As individuals change their perceptions of the war, of Russia, and around immigration, they also change their views on Ukrainian refugees. The war is constantly evolving and the events and media narratives in turn partially drive people’s perceptions. The length of the war and the ensuing costs to host countries also play a role. We notice a small decrease in the overall support for refugees (explained by a decrease in support for Ukraine and the West, and more negative immigration attitudes) already five months into the war. This trend will likely continue but will in turn depend on the specific dynamics of how the war evolves.

Finally, we also note a spill-over effect of support for other types of refugees. In our dynamic panel models, we see that individuals who became more supportive of Ukrainian refugees during the first four months of the war were also more likely to become more supportive of other types of refugees. The mechanisms for such a spill-over effect remain to be explored in further research. Further research can also explore the more long-term dynamics of refugee acceptance, as well as the acceptance of refugees from different contexts.

This article contributes to the literature on attitude formation to refugees, highlighting their specific determinants beyond simple predispositions to immigration broadly. Our original panel survey allowed us to see the dynamics of attitude formation and change, underscoring their stability over time, but also showing their malleability according to embeddedness in broader attitudinal patterns (in this case, geopolitical and to a lesser extent identitarian), spill-overs to ensure cognitive consonance (when confronted with similar questions about distinct groups), and changing contexts (as the initial effects of novel and emotive events wear off). Our analysis is not without limitations. Even with our panel data, we cannot fully rule out endogeneity. We also do not have comparable questions for attitudes to the conflicts in Afghanistan and Somalia, although we expect them to be of lower salience generally and therefore of lower impact on refugee attitudes. Future work can investigate the likely mechanisms behind the spill-over effect that we observe and also more minutely investigate the effect of the ethnic and gender composition of refugees.

Acknowledgements

We would like to thank the anonymous reviewers for their detailed feedback which helped greatly to strengthen and develop the key findings of the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Replication data and code are available online at https://doi.org/10.7910/DVN/KTQNOW.

Additional information

Funding

This work was supported by European Research Council [Grant Number ERC-2018-SyGSH], in the scope of the project SOLID – Policy Crisis and Crisis Politics, Sovereignty, Solidarity and Identity in the EU post 2008.

Notes on contributors

Alexandru D. Moise

Alexandru D. Moise is a postdoctoral research fellow at the Department of Social and Political Science at the European University Institute, where he works as part of the SOLID-ERC project.

James Dennison

James Dennison is a Researcher at the University of East Anglia and a part-time Professor at the Migration Policy Centre of the European University Institute, Florence.

Hanspeter Kriesi

Hanspeter Kriesi is a part-time professor for Comparative Politics at the European University Institute in Florence. Previously, he held the Stein-Rokkan Chair at this university as well as positions at the universities of Amsterdam, Geneva and Zurich. His wide-ranging research interests include the study of various aspects of democracy, political communication, political mobilisation and opinion formation.

Notes

1 Figure B3 in the online appendix shows the directed acyclic graph (DAG) that summarizes our theoretical assumptions for the identification of the hypothesized effects (Truchlewski et al. 2023).

2 The survey was conducted via CAWI methodology using the YouGov proprietary panel in all countries to recruit participants.

3 Our recontact rates range from 56% in Hungary to 69% in Germany, with 62% in France, 64% in Italy, and 59% in Poland. Table A4 in the online appendix shows the results of an attrition analysis.

4 Section C in the online appendix contains full question wording for all dependent variables and major independent variables.

5 Details on factors can be found in online appendix section C.

6 See exact question wording in online appendix section C.

7 This is equivalent to a 1 standard deviation change in pro-immigration leading to a 0.33 standard deviation increase in support for Ukrainian refugees. To see all effects standardized, see Table A16 in the online appendix.

8 We cluster standard errors at the individual level in order to obtain correct standard errors.

9 For a continuous measure of change across the two waves, see Figure B2 in the online appendix which plots the density of average change from wave 1 to 2 across our 5 countries.

10 Figure B1 in the online appendix shows a comparison with a pre-war question on the acceptance of refugees fleeing war. While the questions are not directly comparable, the fact that respondents in 2022 report considerably lower support for Afghans than the pre-war question, suggests to us that respondents feel comfortable expressing their preferences.

11 Our original survey covered a broad range of topics, including political attitudes, attitudes towards energy and environmental policy, the rising cost of living, and foreign policy, among others. Respondents were therefore not primed to consider refugee policy in particular.

12 For robustness, we conduct an analysis on a dichotomized version of the variables taking into account solely whether there is positive or negative change, without considering if an individual crossed the 0.5 mark. Results are substantively similar. See table A15 in the online appendix.

13 Note, this question, together with the question on hosting refugees, were only asked in the second wave of our study and thus were not used in the analysis.