Vaccine Skeptics and Vaccine Enthusiasts: What Is the Intergroup Wall Made Of?

We may be inclined to assume that people reluctant to vaccinate either lack credible information or cannot correctly assess the objective state of scientific knowledge and make a logically consequential decision. While this interpretation may be prevalent and seems to be an implicit assumption in many online and offline debates, science communicators frequently criticize it.

Wynne (1991) was one of the first to speak out against this. He called this explanation the “cognitive deficit model” and proposed that instead, we should understand the rejection of scientific knowledge through the lens of the social context, most importantly the context of social identity (Wynne, 1992). What people reject, is not necessarily the content of scientific knowledge itself or the method through which it was obtained. Instead, according to Wynne (1992), people reject the messengers of that knowledge, as they perceive them as outsiders whose interests do not coincide with those of their in-group.

Wynne’s thought was repeated, elaborated and brought into the specific context of vaccine hesitancy by Hornsey and Fielding (2017). In this work, the authors postulate the search for many motivational roots of rejecting scientific knowledge. One such reason is the motivation to maintain and act on one’s social identity. Another source, closely related to social identity, is ideology/value system.

In this vein, the authors argue that the rejection of the HPV vaccine is partly rooted in an aversion to more progressive social norms. Since the HPV vaccine is mainly recommended for adolescent women (preferably before they become sexually active), the decision to vaccinate rests on the shoulders of parents. They must face the realistic prospect that their daughters will soon become sexually active, presumably with multiple partners. Since this could threaten the parents’ preferred values and traditional social order, they become reluctant to vaccinate.

A number of recent empirical studies support the claim that aversion to vaccination is closely linked to broader beliefs and value system that can easily translate into group identity and group affiliation. Much evidence suggests that aversion to vaccination is linked to conservatism and a conspiratorial worldview (Freeman et al., 2020; Hornsey et al., 2020; Stroope et al., 2021). Moreover, analyses of vaccine-related online activity reveal that Russian state-driven disinformation about vaccines specifically exploits existing intergroup (e.g., interracial) tensions. These disinformation activities often aim to radicalize both sides of the conflict (Broniatowski et al., 2018, 2020; Walter et al., 2020).

The question of pro- and anti-vaccine group identity has also been addressed more directly. Maciuszek and colleagues (Maciuszek et al., 2021) studied a targeted sample of individuals who hold pro- or anti-vaccine attitudes and are involved in discussions about vaccines. Representatives of both attitudes manifested a sense of group identity based on their positions on vaccines. Interestingly, pro-vaccine respondents had a stronger group identity and it manifested itself in all measured domains (Importance, Commitment, Superiority and Deference).

Despite many theoretical and empirical suggestions that attitudes toward vaccines may shape (or at least be part of) group identities, there is little research that directly examines how pro- and anti-vaccine people view each other. This issue may be crucial for understanding and mitigating communication barriers between groups.

In the domain of intergroup relations, social psychologists often emphasize the prevalence and importance of various forms of so-called dehumanization (see: Haslam, 2015 for a synthetic overview). Dehumanization occurs when a member of one group (typically in-group) denies the existence of some or all of the prototypical human characteristics of another group (typically out-group).

The occurrence of dehumanization or meta-dehumanization (the feeling of being dehumanized) predicts many negative consequences in intergroup relations (see: Kteily & Landry, 2022 for a recent review).

From available knowledge, it appears that pro- and anti-vaccine individuals can form groups that are prone to polarization and mutual hostility. Mønsted and Lehmann (2022) found that the pattern of online interactions regarding vaccines reveals an “epistemic echo chamber” effect for both pro- and anti-vaccine individuals. These two groups form an internally consistent information environment, supporting their attitudes and rarely interacting with content expressing opposing views.

There are evidence suggesting that vaccine averse individuals may be animalistically dehumanized by vaccine supporters. Animalistic dehumanization (or denial of human uniqueness) is part of the dual dehumanization model proposed by Haslam (2006). It occurs when someone is denied traits that distinguish humans from animals. These traits are related to self-control, high cognitive functions or cultural sophistication.

Rozbroj and colleagues (2022) found that similar qualities are denied to vaccine-skeptics – they tend to be perceived as intellectually inferior, overly emotional and disruptive by vaccine-enthusiasts. In addition, Maciuszek and colleagues (2021) found that pro-vaccine people tend to view anti-vaccine people as lacking in scientific knowledge.

Analogically, pro-vaccine people may experience mechanistic dehumanization from anti-vaccine people. Mechanistic dehumanization (or denial of human nature) is the second part of Haslam’s dual model of dehumanization. It involves denying someone the traits associated with human nature that separate humans from inanimate entities such as robots. These traits refer to virtues such as warmth, empathy or individuality/agency.

Rozbroj and Collegues (Rozbroj et al., 2022) found that people who actively refuse vaccination perceive themselves as rich in virtues that closely resemble the human nature constellation from the dual dehumanization model. They see themselves as courageous, caring (for example, for their children) and independent. Moreover, an analysis of vaccine skeptic’s narratives in online media, revealed that anti-vaccine attitudes are promoted with motives of freedom, individual agency and care (Jamison et al., 2020; Lander & Ragusa, 2021).

Since anti-vaccine individuals see themselves as a minority possessing human nature traits and are additionally surrounded by narratives that support their view, a logical consequence could be the negation of human nature traits in the opposing group—vaccine enthusiasts.

Another type of dehumanization, which can involve the relationship between vaccine skeptics and vaccine enthusiasts, is blatant dehumanization (Kteily et al., 2015). This concept of dehumanization is one of the most comprehensive. It does not specify what human-related properties are being denied. Instead, it invokes the simple notion of being/not being a fully evolved/developed human being.

This type of dehumanization has been shown to be loosely correlated with more subtle forms (such as dual-model dehumanization) and closely linked to general prejudice and hostility toward the dehumanized external group (Kteily & Landry, 2022).

Since blatant dehumanization is a good indicator of general hostility, we predict that vaccine enthusiasts may tend to blatantly dehumanize vaccine skeptics. What’s more, the vaccine-skeptics will feel blatantly dehumanized by the opposing group (meta-dehumanization).

Rozbroj and Collegues (2019) found that pro-vaccine people maintain highly hostile attitudes toward vaccine skeptics. Here are some excerpts from the opinions about vaccine refusers: “A bunch of misinformed, dangerous twits”, “Deluded paranoid narcissists!”, “A selfish group of deliberately ignorant people” (Rozbroj et al., 2019, p. 5988).

Such opinions are reflected in the accounts of parents who refuse vaccinations. A qualitative study by Wiley and colleagues (2021) found that parents who refuse vaccinations experience labeling, social exclusion and loss of status. In their perspective, these adversities are a consequence of their virtues and best intentions, and as such are unfair and cruel. Such a perspective can provide good ground for a sense of being dehumanized. This feeling may be stronger in people who frequently come into contact with vaccination enthusiasts.

The goal of the study is to empirically verify the predictions concerning mutual dehumanization between vaccine enthusiasts and vaccine skeptics. The theoretical basis for them was presented in previous chapters. The hypotheses are:

H1. Vaccine enthusiasts will animalistically dehumanize vaccine skeptics,

H2. Vaccine skeptics will mechanistically dehumanize vaccine enthusiasts,

H3. Vaccine enthusiasts will blatantly dehumanize vaccine skeptics,

H4a. Vaccine skeptics will experience meta dehumanization (They will believe, they are blatantly dehumanized by pro-vaccine people.)

H4b. In the relationship predicted in hypothesis H4a, the intensity of an online-interactions with vaccine enthusiasts will be a significant covariate—The effect of meta-dehumanization will be stronger in the case of the respondents who have more interaction with pro-vaccine people.

We tested these hypotheses on three separate populations - South African, Polish and American – to establish to what degree postulated effects can be universal vs. local. It is a crucial point of our investigation since most of the empirical works on which we base our predictions, were conducted in North America or Australia. This may pose a serious problem for generalizability, since the political and social discourse around vaccines can be highly specific in different countries - see the reviews of such local contexts in South Africa (Bam, 2021) and Poland (Żuk et al., 2019).

We conducted an online, correlational study, using targeted sampling. The study was pre-registered. Data, methods and reproducible analyses are publicly available through the OSF platform (https://osf.io/67h3w/).

We conducted both confirmatory and exploratory analyses, with key areas of exploration assumed beforehand.

During the data collection process, we were forced to deviate from the pre-registered protocol in two minor ways:

1. We planned to recruit all respondents from a single, multinational online panel - Prolific. This turned out to be impossible because there were not enough Polish participants with negative attitudes towards COVID-19 vaccines available on this panel. To alleviate this problem, we decided to obtain additional Polish participants with negative attitudes through the Polish research panel.

2. Due to resource constraints, the Prolific sample size was slightly smaller than our original target. Out of the planned 400 participants per country, we were able to recruit 383 participants from South Africa and 397 participants from the United States. In contrast, we were able to recruit more Polish participants than originally planned (482). This was due to the additional sourcing from the second panel. We decided to retain this excess data in order to maximize the cost-effectiveness of the research.

We sourced our participants through the Prolific platform, using pre-screen criteria and gender-balancing to obtain the sample of the desired characteristics. We chose Prolific because of its high diversity of participants and advanced tools for customizing the sample characteristics. Additionally, we sourced the participants from local, Polish research panel, using the same pre-screen criteria and gender-balancing.

The first desired characteristics were nationalities and locations. We collected samples from three locations and nationalities:

• Participants located in South Africa and of South-African nationality,

• Participants located in Poland and of Polish nationality,

• Participants located in the USA and of American nationality.

Besides the cultural differences, these three clusters of participants form a pool that is diverse geographically (three continents), socially (a post-colonial society, a post-communist society and a Western democratic society) and economically (respectively 103, 41 and 7 places in the world ranking of GPD per capita at purchasing power parity - World Economic Outlook (October 2022), n.d.).

The second characteristic of our sample which we chose to control was the attitude towards vaccinations. We sourced participants that stated a firm opinion on the Prolific panel pre-screen question: Please describe your attitudes towards the COVID-19 (Coronavirus) vaccines. For each national/regional cluster, we recruited participants who responded either “Against (I feel negatively about the vaccines)” or “For (I feel positively about the vaccines)” in a 50%/50% proportion.

Thirdly, all three regional/national clusters were balanced on the sex criteria with a 50%/50% men and woman proportion (the only available balancing option on the platform, with the default screening question: “What is your sex, as recorded on legal/official documents?”).

Planning our sample size, we chose a test for the hypothesis H2: Vaccine-enthusiasts will blatantly dehumanize vaccine-skeptics as a point of reference. To define a minimal effect size of interest, we examined a recent, publicly available data set containing a measure of blatant dehumanization (Izydorczak et al., 2022). In this data set, a Polish sample of participants responded to a blatant dehumanization item concerning their national in-group and various out-groups.

Following the authors’ advice to consider the dichotomic cut-off point (“full humanity”, “non-full humanity”) as the most essential score, we compared the proportion of “full humanity”/“non-full humanity” scores for the in-group and the “Russians” out-group. Russians were chosen, because out of all outgroups who were negatively perceived in the aforementioned study, they were dehumanized to the least degree.

To detect the effect size, we conducted a mixed-model logistic regression for the binomial distribution with dichotomized blatant dehumanization (full humanity/non-full humanity) as an outcome, group of reference (Poles vs. Russians) as a fixed factor and participant ID as a random factor for intercepts. The probability of Poles being blatantly dehumanized was .56, and the probability of Russians being blatantly dehumanized was .79 (OR = 2.9, 95% CI [2.24, 3.76], p < .001).

Using G*power, ver. 3.1.9.4, we concluded that to detect such effect size with the power 1-β = .95 and α = .05 we need a sample of 184 participants (a-priori analysis for z-test, logistic regression, one-tail, OR = 2.9, Pr(H0) = .56).

Considering this calculation, we need at least 184 vaccine-enthusiast participants for each three national/regional clusters. Analogically, we need the same number of vaccine-skeptical participants.

Taking possible data exclusions into account, we aimed to recruit a 220 x 2 (vaccine-skeptical/vaccine-enthusiastic) x 3 (South-Africa, Poland, USA) = 1320 participants.

We included participants, who:

1. Met any of the convergent location/nationality criteria:

   • Were located in South Africa and had South African nationality,

   • Were located in Poland and had Polish nationality,

   • Were located in the USA and have an American nationality,

2. Met language-comprehension criteria:

   • The USA and South-Africa-based participants had to be fluent English-language users

   • Poland-based participants had to be fluent Polish-language users.,

3. Met our pre-screening criteria regarding attitudes towards COVID-19 vaccines:

   • Respond either “Against (I feel negatively about the vaccines)”, or

   • “For (I feel positively about the vaccines)” to the question: Please describe your attitudes towards the COVID-19 (Coronavirus) vaccines.

All inclusion criteria were implemented through internal pre-screening offered by Prolific or through filters within the questionnaire (in the case of participants sourced from the local Polish panel). While Prolific platform allows researchers to filter respondents invited to participate based on their responses to an internal demographic questionnaire, Polish local panel did not offer this option. For than reasons, screeners had to be added manually, using survey engine.

Additionally, we planned to exclude participants who would meet at least one of the following conditions:

1. Fail the bot-detection check (“I am not a robot” re-CAPTCHA test),

2. Indicate a different answer in a pre-screening question: Please describe your attitudes towards the COVID-19 (Coronavirus) vaccines., which we will incorporate as a screener-check in our questionnaire,

3. Finish the survey extremely fast (less than one second per item),

4. Fail both of the attention check questions

We used Qualtrics to design an online survey. The survey consisted of 5 blocks: 1) Vaccine attitudes block, 2) blatant dehumanization, 3) dual model dehumanization, 4) human/animal words-based dehumanization, and 5) in-group and intergroup communication and attitudes. Block 1 was displayed as the first, block 5 as the last one. The order of display for blocks 2-4 was randomized. In multi-item blocks, the order of items was randomized as well.

Demographic data were delivered by Prolific and local, Polish research panel. Collected information were age, sex, fluent languages, ethnicity, country of birth, country of residence, nationality, native language, student status, and employment status.

In block 3 we included the first attention check question:

How typical do you think the listed trait is for the Pacific Ocean:

It contains water: (Answers 1- not at all typical, 2- rather untypical, 3 - rather typical, 4- completely typical). Answers 1 and 2 will be considered failed attention check.

In block 5 we included the second attention check question:

When asked about your favorite color please indicate “blue”. This is an attention check.

Name your favorite color: (Answers: blue, red, yellow, green).

In the vaccine attitudes block, participants were asked two questions. The first question is a direct reiteration of a pre-screening question. If the participant provided an answer which is inconsistent with the pre-screened data, they were automatically filtered out of the sample.

1. Please describe your attitudes towards the COVID-19 (Coronavirus) vaccines. With four possible answers: Against (I feel negatively about the vaccines), For (I feel positively about the vaccines), Neutral (I don’t have strong opinions either way) and Prefer not to say.

The second question was analogical to the first one, but instead of asking about the COVID-19 vaccine, the question concerned attitudes about vaccines in general:

1. Please describe your attitudes towards vaccines in general. With four possible answers: Against (I feel negatively about the vaccines), For (I feel positively about the vaccines), Neutral (I don’t have strong opinions either way) and Prefer not to say.

Blatant dehumanization was measured by the tool developed by Kteily et al. (2015). Participants read instructions (in Polish or English):

“Some people think that people can vary in how human-like they seem. According to this view, some people seem highly evolved, whereas others seem no different than lower animals. Using the sliders below, indicate how evolved you consider the group of people to be.”

The question was asked regarding two groups of people: “people who feel positive about vaccines and are eager to get vaccinated” and “people who feel negative about vaccines and are uneager to get vaccinated”.

Participants rated their answers on a slider scale ranging from ‘0’ to ‘100’ (full humanity). The slider did not contain any numerical labels, the slider dot became visible upon a click on a slider scale, and the currently indicated numerical value was displayed under the slider dot. Above the slider scale, an illustration of 5 silhouettes which symbolize the evolution of the human species from quadrupedal animal to anatomically modern human was placed.

See the illustration below as an example.

Meta-dehumanization was assessed by the additional question: How do you imagine a typical [person with attitude opposite to the participant] would evaluate someone who is [attitude convergent with participants’] on this scale?

Dual model dehumanization was assessed through Polish and English sets of 20 traits of which 10 pertained to the aspect of human uniqueness (traits which separate humans from animals) and 10 to the aspect of human nature (traits that separate humans from robots) - See Haslam (2006) for a detailed review of the theoretical concept behind the method. Two subsets contained an equal number (5 vs 5) of socially desirable and undesirable traits.

Both sets had been recently validated in the unpublished research in which the corresponding author was collaborating. In the validating study, the human uniqueness scale proved to be highly reliable in both Polish (Cronbach’s ⍺ = 0.89) and English sets (Cronbach’s ⍺ = 0.84). The same can be said about the human nature scale (Cronbach’s ⍺ = 0.90 for Polish set, Cronbach’s ⍺ = 0.81 for English set)

See supplementary materials for complete lists in both languages.

Respondents were asked three questions with respect to the listed traits:

1. How typical do you think each listed trait would be for the people who feel positive about vaccines and are eager to get vaccinated?

2. How typical do you think each listed trait would be for the people who feel negative about vaccines and are uneager to get vaccinated?

3. How do you imagine an average [person with an attitude opposite to the participant] would assess the typicality of these traits among those with [attitude convergent with participants’]?

Respondents were presented with a slider scale ranging from 0 to 100, labeled as ‘0 - not at all typical’, ‘50 - somewhat typical’ and ‘100 - very much typical’.

By averaging the answers for the traits in the respective sets, we calculated human uniqueness, human nature, desirable traits and undesirable traits scales.

Direct dehumanization (Animal/human-related words) is a method of measuring dehumanization originally developed by Viki and colleagues (2006). In this method, participants are asked to indicate the extent to which they think certain words can be used to describe a given group or individual. The set of words contains two subsets—the words considered to be appropriate for describing animals and unfitting for humans, and the words considered to be appropriate for describing humans and unfitting for animals. Respondents were presented with a set of 8 words, four of which were animal-related and four human-related.

The human/animal word list we used was recently validated in the unpublished study in which the corresponding author was collaborating. The human-words list had high internal consistency in both English (Cronbach’s ⍺ = 0.83) and Polish (Cronbach’s ⍺ = 0.88) versions. The animal-related word list had a moderate internal consistency in the English set (Cronbach’s ⍺ = 0.73) and very high in the Polish set (Cronbach’s ⍺ = 0.90)

Full sets for Polish and English languages can be found in supplementary materials.

Participants were asked three questions about the word list:

1. Please indicate the extent to which you think each of the following words can be used to describe people who feel positive about vaccines and are eager to get vaccinated.

2. Please indicate the extent to which you think each of the following words can be used to describe people who feel negative about vaccines and are uneager to get vaccinated.

3. How do you imagine an average [person with an attitude opposite to the participant] would assess the extent to which the following words can be used to describe people who [attitude convergent with participants’]?

Participants provided answers on a 0-100 slider scale, where 0 = not at all and 100 = very much.

To assess the extent to which a participant is engaged in debate with people expressing opposite attitudes towards vaccination, we asked the following questions:

1. Are you engaged in an online discussion with people who [attitude opposite to the participants] on the subject of vaccinations?

2. Are you engaged in real-life (offline) discussion with people who [attitude opposite to the participants] on the subject of vaccinations?

To assess the extent to which a participant is engaged in discourse with a group that shares their beliefs, we asked the following question:

1. Do you participate in online conversations with people who [attitude consistent with that of participants] about vaccination?

2. Do you participate in real-life (offline) conversations with people who [attitude consistent with that of participants] about vaccination?

Participants indicated their answers on a 0-100 slider scale, where 0 = never , 50 = from time to time and 100 = on daily basis.

As a measure of emotional attitude towards in-group and out-group, participants were asked to rate their feeling towards vaccine-skeptics and vaccine-enthusiasts in the form of feeling-thermometer slider scale:

1. ’How warm (favorable) or cold (unfavorable) do you feel towards the following groups?:

   • Vaccine enthusiasts,

   • Vaccine skeptics,

Answers were given on a 101-point scale slider-scale, where 0 = very unfavorable, 50 = neutral and 100 = very favorable.

We conducted our analysis in Jamovi software, ver. 2.3.18. All confirmatory analyses were frequentist.

As the first step, we tested whether vaccine skeptics and vaccine-enthusiasts differentiate their emotional attitudes towards vaccine skeptics and vaccine-enthusiasts. .

In confirmatory analysis, all hypotheses were tested on three sub-samples simultaneously, including “country” as a random factor/cluster variable in the mixed-model analysis. Intracluster Correlation Coefficient (ICC) for “country” were interpreted as a degree to which the obtained results are universal/country-specific.

To test the first hypothesis H1) Vaccine-enthusiasts will animalistically dehumanize vaccine-skeptics, we used a mixed-model linear regression (fixed slopes, random intercepts) with dehumanization index as a dependent variable, group of reference as fixed effect factor, and “respondent ID” along with “country” as a random cluster variable.

Following the theoretical and empirical critique by Enock and colleagues (2021), we acknowledged that dual-model dehumanization might be confounded with a general positive bias towards in-group to a large extent.

To disentangle this possible confusion, we tested this hypothesis in two ways. The first way was the classic one: the dependent variable was a combined index of animalistic dehumanization. In the second variant, we tested two components/sub-scales of animalistic dehumanization separately: desirable human uniqueness and undesirable human uniqueness. This hypothesis was tested on a group of vaccine-enthusiasts.

The second hypothesis, H2) Vaccine-skeptics will mechanistically dehumanize vaccine-enthusiasts, was tested in the same way as H1, but instead of vaccine enthusiasts, it was tested on the vaccine-skeptics group and instead of an animalistic dehumanization, the tested variables were: combined index of mechanistic dehumanization, desirable human nature and undesirable human nature.

To test the third hypothesis H3) Vaccine-enthusiasts will blatantly dehumanize vaccine-skeptics, we conducted a mixed-model logistic regression. We dichotomized the Ascent of Humans score (‘100’ score was be coded as “fully human”, scores < 100 were coded as “partially human”). This dichotomized score was the dependent variable. The reference group (vaccine-enthusiasts vs vaccine-skeptics) was a fixed factor while respondent ID and “country” were random factors for intercepts.

Fourth hypothesis H4a) Vaccine-skeptics will experience meta-dehumanization (They will believe, they are blatantly dehumanized by pro-vaccine people) was tested analogically to H2 - we conducted mixed-model logistic regression with dichotomized Ascent of Humans score as the dependent variable, a point of reference (self-evaluation of vaccine skeptics vs assumed evaluation of vaccine skeptics by vaccine enthusiasts) as a fixed factor and responded ID and “country” as a random factors for intercepts.

To test the H4b) - In the relationship predicted in hypothesis H4a, the intensity of an online-interactions with vaccine enthusiasts will be a significant covariate - The effect of meta-dehumanization will be stronger in the case of the respondents who have more interaction with pro-vaccine people, we conducted the same analysis as for the H4a with one additional element: the intensity of the interactions with vaccine enthusiasts was assigned as a covariate.

All tested relationships, as well as the distribution of the variables, were visually analyzed with additional statistical analyses.

Furthermore, In the exploratory section, we tested the occurrence of blatant dehumanization, dual model dehumanization and meta dehumanization in all remaining out-group/in-group combinations which were not investigated in the confirmatory section. Moreover, we investigated the occurrence of direct dehumanization (Viki et al., 2006).

Below we present the test of pre-registered hypotheses along with the post-hoc exploratory analyses. All data, scripts for data-wrangling and visualizations and reproducible statistical analyses (in .omv format) can be found in the OSF repository (https://osf.io/67h3w/).

For analyses we used Jamovi, ver. 2.3.18.0 (jamovi project, 2022), supported by visualization and data processing in R programming language ver. 4.1.3 (R Core Team, 2022) with tidyverse (Wickham et al., 2019) and ggstatsplot (Patil, 2021) packages.

The final sample consisted of 1262 participants (630 women, 632 men, M_Age = 34.5, SD_Age = 13.3, Min_Age = 18, Max_Age = 82). Questions about COVID-19 vaccine attitude and attention checks were used as automated screeners, so no data exclusion has been made based on these criteria. All participants who passed the pre-screening criteria passed the bot-detection test as well. No data were excluded based on the completion time criterium. The detailed demographic characteristics of the sample are presented in the table below.

In order to explore the characteristics of the participants and test the validity of our assumptions, we decided to investigate whether attitudes toward vaccines could serve as a group-forming factor and induce in-group favoritism. We considered two indicators of intergroup division: intergroup bias as measured by the feelings thermometer and the intensity of online and offline communication with individuals who have similar and opposing attitudes toward the COVID-19 vaccine.

For the feelings thermometer we found evidence of strong mutual biases between people with different attitudes toward the vaccine, although biases were stronger among vaccine enthusiasts. We conducted a 2×2×3 within-between-subjects ANOVA with the reference group (in-group, out-group) as a within-subjects factor, COVID-19 vaccine attitudes (positive, negative) as a between-subjects factor, and country of residence (Poland, South Africa, USA) as a between-subjects factor. The feelings thermometer score (ranging from 0 to 100) was the dependent variable.

We found the main effect of the group of reference – F(1, 1255) = 1624.50, p < .001, η_p² = 0.56. The feeling towards the in-group was more positive (on average 38.86 higher on a 0-100 scale). We also found the interaction effect between a group of reference and attitude: F(1, 1255) = 141.14, p < .001, η_p² = 0.1. The analysis of simple main effects revealed that in-group/out-group difference in the feeling thermometer was statistically significant for both vaccine enthusiasts and vaccine skeptics, but greater for vaccine enthusiasts (Mean difference 50.31 vs. 27.4). The interaction between the group of reference, attitude, and country was also significant - F(2, 1255) = 21.73, p < .001, η_p² = 0.03. The visual analyses revealed that in the case of RPA, the prejudice towards out-groups is similarly strong among vaccine-skeptics and vaccine-enthusiasts while in the case of Poland, the prejudice among vaccine-enthusiasts surpasses those among vaccine-skeptics the most.

Analogical 2×2×3 between-within subject ANOVAs was conducted for the “online contact intensity” and “offline contact intensity” dependent variables. It turned out that in the case of both types of contacts, participants interacted with members of an in-group more frequently. This effect was stronger in the case of vaccine-enthusiasts.

In the case of online contact, we found a significant main effect of the group of reference - F(1, 1248) = 189.13, p < .001, η_p² = 0.13. The frequency of online interactions with in-group contact was on average 10.97 higher. (the scale ranged from 0 - never, 100 – on daily basis). We found a significant interaction effect between a group of reference and attitude towards the COVID-19 vaccine - F(1, 1248) = 28.14, p < .001, η_p² = 0.02. The analysis of simple main effects revealed that the difference in the online contact between in-group and out-group is significant for both vaccine-enthusiasts and vaccine-skeptics, but larger in the case of vaccine-enthusiasts (mean difference 15.2 vs 6.74).

In the case of an offline contact, we identified the same effects. The frequency of contacts was higher for in-group interactions (on average 14.91 points) - F(1, 1253) = 266.13, p < .001, η_p² = 0.18. The interaction between the group of reference and attitude was also significant - F(1, 1248) = 39.81, p < .001, η_p² = 0.03. The main effect (higher intensity of contacts with in-group) was significant for both vaccine-enthusiasts and vaccine-skeptics but stronger in the case of vaccine-enthusiasts (mean difference – 19.96 vs 8.83).

Besides examining, whether attitudes towards COVID-19 vaccine creates intergroup rifts, we tested to what degree attitudes towards this particular vaccine are associated with general attitudes towards vaccines. It turned out that these two attitudes are highly convergent – V = 0.67, chi² (3) = 570.94, p <.001.

To test the first hypothesis (H1): Vaccine-enthusiasts will animalistically dehumanize vaccine-skeptics, we estimated (REML method) a mixed linear regression model with respondent ID and country as a random factor for intercepts and a group of reference (in-group vs. out-group with in-group coded as “0”) as a fixed factor. The hypotheses were tested with three dependent variables separately: animalistic dehumanization (full human-uniqueness index), desirable human-uniqueness traits, and undesirable human-uniqueness traits. It is worth noticing that evaluating out-group members lower on a full human-uniqueness index can be interpreted as evidence for animalistic dehumanization of the out-group. Evaluating the out-group higher on negative human-uniqueness and lower on positive human-uniqueness traits is evidence for negative bias (prejudice) towards the outgroup.

In the case of the full index and desirable traits, we found evidence for animalistic dehumanization in the predicted direction. In the case of undesirable traits, we found evidence for the opposite effect – these traits were ascribed more to the out-group than the in-group.

Moreover, we found that the ICC (intracluster correlation coefficients) for the “Country” variable was very small (ranging from <.001 to .06), indicating that the degree of dehumanization did not vary significantly between the three populations (RPA, USA, and Poland). See the detailed results in the tables (Table 2 and Table 3) below:

Summing up, the hypothesis was partially confirmed. It is worth noticing that the extent of overall animalistic dehumanization is smaller than a sheer positive bias towards in-group. Members of the outgroup have been evaluated on average 4.31 lower on “human uniqueness” (0-100 scale), 35.4 points lower on desirable uniquely human traits, and 27.8 points higher on undesirable uniquely human traits.

To test the second hypothesis (H2): Vaccine skeptics will mechanistically dehumanize vaccine enthusiasts, we estimated (REML method) a linear mixed-model with respondent ID and country as a random factor for intercept and group of reference (in-group vs. out-group, with in-group coded as “0”) as a fixed factor. Analogically to the first hypothesis, we tested H2 with three separate dependent variables: with general mechanistic dehumanization index (human-nature index), with desirable human-nature traits, and with undesirable human-nature traits.

The prediction was not supported. In the case of general mechanistic dehumanization, we found an effect in the opposite direction: vaccine skeptics tended to ascribe human-nature traits more to vaccine enthusiasts than themselves. The same can be said about undesirable human-nature traits. Only in the case of desirable human-nature traits, the relationship was in the predicted direction – Vaccine skeptics ascribed more of these traits to themselves than to vaccine enthusiasts.

When it comes to the difference between the three populations, we found evidence in favor of the universality – ICC for “Country” factor was low, ranging from .02 to .13. See detailed results in the tables (Table 4, Table 5) below.

Summing up, we found no evidence for the mechanistic dehumanization of vaccine enthusiasts by vaccine skeptics, but we found evidence for negative bias. Vaccine skeptics estimated positive aspects of human nature as more prevalent among them (on average 5.91 points higher on a 0-100 scale) and negative aspects of human nature as less prevalent (on average 10.41 lower on a 0-100 scale).

To complement the results obtained in the confirmatory analyses, we decided to explore the patterns of animalistic dehumanization of vaccine enthusiasts by vaccine skeptics and to present a visual analysis of all the patterns for three investigated populations separately.

It turned out that there are no indications of general, animalistic dehumanization of vaccine enthusiasts by vaccine skeptics. On the contrary, in linear mixed model analyses (random factor for intercepts: ID, country, fixed factor: a group of reference, with in-group coded as “0”), we found that vaccine skeptics estimated the human-uniqueness index to be 1.78 points higher (0-100 scale) for an outgroup (vaccine enthusiasts) than for themselves - b = 1.78, 95% CI [0.83; 2.73], t (2.03, 678) = 3.67, p <.001.

Despite the lack of evidence for general animalistic dehumanization, we found a pattern suggesting positive bias towards in-group: positive aspects of human nature were ascribed more eagerly to the vaccine-skeptics - b = -11.01, 95% CI [-13.14; -8.89], t(672,672) = -10.16, p <.001), while negative aspects of human nature less eagerly (b = 14.57, 95%CI [12.44; 16.71], t (2.03,672) = 13.39, p <.001.

Overall, the patterns of mutual animalistic dehumanization display four trends:

1) Animalistic dehumanization towards vaccine skeptics by vaccine enthusiasts is identifiable while being absent or reversed in the attitude of vaccine skeptics towards vaccine enthusiasts,

2) Positive in-group bias (ascribing more positive and less negative traits to the in-group members) is observed among both vaccine enthusiasts and vaccine skeptics, but is stronger in the case of the former,

3) In comparison to vaccine skeptics, vaccine enthusiasts are more eager to attribute uniquely-human traits to themselves and more inclined to view themselves in a favorable way (by attributing many positive uniquely-human traits and few negative uniquely-human traits),

4) Results are largely similar across all three investigated populations.

As the final, exploratory analysis concerning animalistic dehumanization, we decided to test the compensatory dehumanization hypothesis, recently proposed by Bustillos et al. (Bustillos et al., 2023). Authors of this theoretical review suggest that self-dehumanization in one domain (e.g. mechanistic) might lead to compensatory out-group dehumanization on the complementary domain (e.g. animalistic). They argue that when one feel worse than out-group in one aspect (e.g. less cultured), it might lead them to evaluate out-group as worse in other aspect (e.g. less empathetic).

To test this prediction, we investigated the partial correlation between scores for in-group animalistic dehumanization and out-group mechanistic dehumanization while controlling for the base-rate (in-group mechanistic and out-group animalistic). We computed the analogical partial correlation between in-group mechanistic dehumanization and out-group animalistic dehumanization. We did find support for the compensatory dehumanization hypothesis among both vaccine-enthusiasts and vaccine-skeptics. For vaccine enthusiasts the effect of mechanistic(ingroup)-animalistic(outgroup) compensation was r(589) = .17, p < .001, for vaccine-skeptics it was r(673) = .10, p = .007. The animalistic(ingroup)-mechanistic(outgroup) compensation effect was r(589) = .21, p < .001 for vaccine-enthusiasts and r(673) = .12, p = .002 for vaccine-skeptics.

Our prediction that vaccine-skeptics will mechanistically dehumanize vaccine-enthusiasts have not been confirmed, but we decided to test this type of dehumanization in the opposite direction. Linear mixed-model analyses (random factor for intercepts: ID, country, fixed factor: group of reference, with in-group coded as “0”) suggest that vaccine enthusiasts mechanistically dehumanize vaccine-enthusiasts – Vaccine enthusiasts estimated the average prevalence of human-nature traits to be 6.28 lower among vaccine-skeptics - b = -6.28, 95% CI [-7.23; -5.34], t(588,588) = -12.99, p <.001.

Apart from mechanistic dehumanization, we also found evidence suggesting positive bias towards the in-group. Desirable human-nature traits were estimated to be 40.27 less prevalent among vaccine-skeptics than vaccine-enthusiasts - b = -40.27, 95% CI [-42.46; -38.07], t(2,588) = -35.97, p <.001. Undesirable human-nature traits were estimated to be 27.70 more prevalent among vaccine-skeptics than vaccine-enthusiasts - b = 27.70, 95% CI [25.70; 29.70], t(2,588) = -27.13, p <.001.

Summing up, the processes of mechanistic dehumanization are less symmetrical and less universal but similar to the ones concerning animalistic dehumanization. We can sum up the patterns in four points:

1) While vaccine enthusiasts mechanistically dehumanize vaccine skeptics, there is no evidence for the opposite process,

2) Positive in-group bias (ascribing more positive and less negative traits to the in-group members) is observed among both vaccine enthusiasts and vaccine skeptics, but is stronger in the case of the former,

3) In comparison to vaccine-skeptics, vaccine-enthusiasts are more eager to attribute human-nature traits to themselves and more inclined to view themselves in a favorable way (by attributing many positive human-nature traits and few negative human-nature traits),

4) Investigated populations displayed effects of similar direction and general pattern, but the magnitude differed. Patterns of positive in-group bias were the most pronounced in the USA and the least pronounced in Poland.

To test the third hypothesis (H3): Vaccine enthusiasts will blatantly dehumanize vaccine skeptics, we estimated a logistic regression mixed model with respondent ID and country as a random factor for intercept and group of reference (in-group vs. out-group) as a fixed factor. The dependent variable was a dichotomized blatant dehumanization (0: non-full humanity, 1: full humanity).

The hypothesis was confirmed. There was a significant difference between the probability of ascribing full humanness to the members of the in-group (vaccine enthusiasts) and out-group (vaccine skeptics). The probability of ascribing the full humanity to the in-group was 89%, for the out-group it was 20 %: χ2 = 70.64, p <.001. The out-group/in-group odds-ratio (P(1)/P(0)) was exp(B) = 0.03, 95% CI [0.01, 0.07], p < .001.

Once again, the effects proved to be similar among investigated categories (RPA, USA, Poland): The ICC for country < .01.

Testing the pre-registered hypotheses, we confirmed our prediction that vaccine enthusiasts blatantly dehumanize vaccine skeptics. We decided to explore the opposite direction – the dehumanization of vaccine enthusiasts by vaccine skeptics. Moreover, we present a visual analysis of mutual blatant dehumanization in three investigated populations.

It turned out that vaccine skeptics tend to blatantly dehumanize vaccine enthusiasts. Logistic regression mixed model with respondent ID and country as a random factor for intercept and group of reference (in-group vs. out-group) as a fixed factor revealed a significant effect of a group of reference on dichotomized Ascent of Humans score (χ2 = 61.19, p <.001).

Vaccine skeptics attributed full humanness to 63% of their in-group members and 32% of the out-group members. The out-group/in-group odds-ratio (P(1)/P(0)) was exp(B) = 0.28, 95% CI [0.20, 0.38], p < .001.

When it comes to mutual blatant dehumanization among all investigated populations, we identified two striking patterns:

1) Blatant dehumanization (difference in the attribution of full humanness between in-group and out-group) is easily identified in all populations among both vaccine-skeptics and vaccine enthusiasts,

2) Vaccine enthusiasts and vaccine skeptics differ in how they attribute humanness to in-group and out-group – Vaccine enthusiasts humanize themselves more (attribute full humanness more often) than vaccine-skeptics. For this reason, the dehumanization of vaccine skeptics by vaccine enthusiasts is stronger than the opposite process.

Direct dehumanization (Viki et al., 2006), also known as human/animal-related words, is another method of investigating more literal forms of dehumanization. We did not pose any hypotheses related to this measurement. Instead, we conducted an exploratory analysis in order to estimate its prevalence, magnitude, direction, and universality.

Linear mixed model analyses (random factor for intercepts: ID, country, fixed factor: a group of reference) revealed that vaccine-skeptics estimated the animal-related words to be a more fitting description of vaccine enthusiasts than vaccine skeptics. The difference was 8.66 (0-100 scale scale), and it was statistically significant - b = 8.66, 95% CI [6.81; 10.52], t(2.01, 672) = 9.17, p <.001. This effect was less universal than other types of investigated dehumanization – ICC for the “Country” cluster equaled 0.2.

We tested the analogical model for human-related words. It turned out that vaccine skeptics estimated these words as more fitting to themselves than to their opposition. The difference was 5.86 (0-100 scale), and it was statistically significant - b = -5.86, 95% CI [-7.64; -4.08], t(2.01, 672) = -6.42, p <.001. ICC for “Country” cluster equaled 0.25.

Concerning vaccine enthusiasts, linear mixed model analyses (random factor for intercepts: ID, country, fixed factor: a group of reference) revealed that they found animal-related words more adequate description of an out-group (vaccine skeptics) than themselves. The difference was 6.79 (0-100 scale), and it was statistically significant - b = 6.79, 95% CI [5.09; 8.50], t(2, 588) = 7.81, p <.001. ICC for “Country” cluster equaled 0.08.

When it comes to human-related words, vaccine enthusiasts found them more adequate as a self-description than the description of vaccine skeptics. The difference was 15.04 (0-100 scale), and it was statistically significant - b = -15.04, 95% CI [-17.02; -13.06], t(2, 588) = - 14.9, p <.001). ICC for “Country” cluster equaled 0.04.

To test the fourth hypothesis (H4a): Vaccine skeptics will experience meta-dehumanization (They will believe, they are blatantly dehumanized by pro-vaccine people.), we estimated a logistic regression mixed model with respondent ID and country as a random factor for intercept and group of reference (in-group vs. meta) as a fixed factor. The dependent variable was a dichotomized blatant dehumanization (0: non-full humanity, 1: full humanity).

The hypothesis was confirmed. The probability of ascribing full humanity to an in-group (vaccine skeptics) was 63%, while the estimated probability of ascribing full humanity to them by vaccine enthusiasts was 7%. The difference was statistically significant: χ2 = 99.01, p <.001. The meta/in-group odds-ratio (P(1)/P(0)) was exp(B) = 0.05, 95% CI [0.03, 0.09], p < .001.

The last prediction (H4b) (In the relationship predicted in hypothesis H4a, the intensity of online interactions with vaccine enthusiasts will be a significant covariate) was tested analogously to H4a, with one addition - the intensity of online and offline communication with members of an out-group were added to the model as covariates. We found no evidence for the presence of covariate effects (online interactions - b < 0.00, p = .83, offline interaction - b < 0.00, p = .46).

In conclusion, we found evidence of strong meta-dehumanization experienced by vaccine skeptics, but there is no evidence that this effect is related to the level of social interaction between vaccine skeptics and vaccine enthusiasts.

Visual analyses revealed that all types of meta-dehumanization were experienced by both vaccine skeptics and vaccine enthusiasts. This was true regardless of country. For all dehumanization types (animalistic, mechanistic, blatant, and direct) and their subtypes, we found that in-group members predicted out-group members’ less favorable perceptions (see Figures 2-7, 8-9).

We decided to formally test and compare meta-dehumanization as measured by the Ascent of Humans scale and animal/human-related words (direct dehumanization). In the case of the Ascent of Humans scale, analysis for H4a revealed that vaccine skeptics experience meta-dehumanization. Post-hoc analyses revealed that vaccine enthusiasts experience this type of meta-dehumanization even more. We tested a logistic regression mixed model (random factor for intercept: ID, country; fixed factor: reference group [in-group vs. meta]; dependent variable: Ascent of Humans for the vaccine enthusiasts). It turned out that the probability of attributing full humanity to in-group members (vaccine enthusiasts) was > 0.999. The probability of in-group members believing that out-group members would attribute full humanity to them was < 0.001. The difference was statistically significant - χ2 = 2.80 x 10⁹, p <.001.

In the case of human- and animal-related words, we tested linear regression mixed models with ID and country as random factors for the intercepts, reference group (in-group vs. meta) as a fixed factor, and the respective dehumanization scale as the dependent variable.

It turned out that vaccine skeptics rated themselves as more human than they predicted out-group members to see them. For animal-related words, the difference in ratings was 17.98 (on a 0-100 scale) - b = 17.98, 95% CI [15.63; 20.34], t(2.07, 672) = 12.69, p = .005. For human-related words, the difference in ratings was 17.53 - b = - 17.53, 95% CI [-19.84; -15.22], t (2.02, 672) = -14.89, p <.001.

Vaccine enthusiasts experienced an even higher degree of meta-dehumanization. For animal-related words, the difference in ratings was 21.52 (on a 0-100 scale) - b = 21.52, 95% CI [18.94; 24.10], t(2, 588) = 16.32, p <.001. For human-related words, the difference in ratings was 23.10 - b = - 23.10, 95% CI [-25.31; -20.88], t(2, 588) = -20.44, p <.001.

As a final analysis regarding meta-dehumanization, we tested the extent to which meta-dehumanization (feeling dehumanized) is accompanied by dehumanization of the very out-group that is supposedly dehumanizing the in-group. To test this, we computed partial correlations between meta and out-group scores while controlling for in-group scores. We found that for each measure of dehumanization (blatant dehumanization, mechanistic dehumanization, animalistic dehumanization, human-related words, and animal-related words), there was a significant correlation (p < .001 in each case). The strongest relationship was found for mechanistic dehumanization - r(1262) = .46, p < .001, the weakest for animal-related words - r(1262) = .27, p < .001.

We examined a diverse, multicultural, and multi-ethnic sample of COVID-19 vaccine enthusiastic, and skeptical individuals. Mutual dehumanization of vaccine skeptics and enthusiasts proved to be strong and universal.

We found ample evidence that vaccine enthusiasts dehumanize vaccine skeptics in all three types of dehumanization studied (dual-model dehumanization, blatant dehumanization, and direct dehumanization). Vaccine skeptics dehumanized vaccine enthusiasts on all scales except one – the human-nature subscale of dual-model dehumanization. The existence of mutual dehumanization and its magnitude was largely independent of the nationality and country of residence of the participants.

Besides the dehumanization, we investigated mutual prejudice and the echo-chamber effect - the tendency to communicate with members of an in-group and avoid contact with members of an outgroup. In both these domains, we found conclusive evidence of strong, inter-group hostility and avoidance. Both vaccine skeptics and vaccine enthusiasts maintain more online and offline communication with people who shares their views. Both of these group holds significantly warmer feelings towards members of their in-group.

Taken together, these results support our assumption that attitudes towards vaccines can be the source of group identity and a driving force for negative, inter-group processes. This finding can be well understood within the theoretical framework of “opinion-based groups” proposed by McGarty (McGarty et al., 2009). Within this framework, agreement on a particularly controversial and important issue can provide a solid foundation for group identity and drive engagement in collective action better than broader identities such as gender or political affiliation.

We were struck by the lack of a clear, distinct set of views that vaccine skeptics and enthusiasts hold about one another. We predicted that the denial of human-uniqueness would be a specific element of the attitude of vaccine enthusiasts toward vaccine skeptics (H1). It turned out that this particular type of dehumanization was relatively weak and present in the views of both vaccine skeptics and enthusiasts. We also predicted that vaccine skeptics would deny the human-nature of vaccine enthusiasts (H2). This prediction has been disproved. On the contrary, this kind of dehumanization appeared to be maintained by vaccine enthusiasts against vaccine skeptics. We also predicted that vaccine enthusiasts would overtly dehumanize vaccine skeptics (H3). This prediction was confirmed, but we also found evidence of the same kind of dehumanization in the opposite direction.

In summary - the specific themes identified in narratives from and about vaccine skeptics and enthusiasts (identified by: Jamison et al., 2020; Lander & Ragusa, 2021; Rozbroj et al., 2019, 2022) did not translate into characteristic forms of dehumanization. Instead, we found that:

1) More extreme and direct forms of mutual dehumanization (Ascent of humans, animal/human-related words) were much more prevalent than subtle forms (dual-model dehumanization),

2) In all but one case (mechanistic dehumanization), dehumanization was mutual,

3) In all cases, vaccine enthusiasts humanized themselves more than vaccine skeptics,

4) In all but one case (animal-related words), vaccine enthusiasts dehumanized vaccine skeptics more than vaccine skeptics dehumanized vaccine enthusiasts.

We predicted (H4a) that vaccine-skeptics will experience meta-dehumanization (within the Ascent of Humans scale), which turned out to be true. However, contrary to our predictions, meta-dehumanization was not moderated by the extent of inter-group contact (H4b). Moreover, we found that experiencing meta-dehumanization is universal – it applies to both vaccine-skeptics and vaccine enthusiasts across all investigated types of dehumanization.

Summing up – in the case of COVID-19 vaccine, the wall of intergroup division between skeptics and enthusiasts is tall and solid and is being built from both sides. The wall seems to be made out of the general hostility and dislike rather than elaborated and specific stereotypes – the sheer amount of out-group derogation/hostility expressed in measures such as “feeling thermometer” (d = 1, 95%CI [0.93, 1.07]) , “Ascent of humans” (d = 0.6, 95%CI [0.54, 0.66]), “animal-related words” (d = - 0.34, 95%CI [- 0.40, -0.28]) and “human-related words” (d = 0.42, 95%CI [0.36, 0.47]) surpasses the subtle forms of dehumanization captured by dual-model dehumanization scale (human-uniqueness - d = 0.09, 95%CI [0.03, 0.14], human-nature - d = 0.14, 95%CI [0.08, 0.20]). To put these results in context, the magnitude of the expressed prejudice in “feeling thermometer” and “Ascent of humans” scale was comparable to the extent in which Polish in-group expressed prejudice towards one of the most derogated minority outgroup (Roma) - “feeling-thermometer” - d = 1.03, “Ascent of humans” - d = 0.53 (Izydorczak et al., 2022, database: https://doi.org/10.17605/OSF.IO/C5K8Q ).

From the perspective of theory-buliding in the field of dehumanization and inter-group attitudes, our results bear significance in at least two domains. First domain is the recent approach of investigating dehumanization with consideration for three, intertwined phenomena: self-dehumanization, other-dehumanization and meta-dehumanization (Bustillos et al., 2023). Authors of the mentioned theoretical paper propose that 1) feeling dehumanized might lead to reciprocal dehumanization in the same domain and 2) self-dehumanization in one domain might lead to compensatory dehumanization of others in the opposite domain. We found empirical support for both hypotheses

The second theoretical domain in which our results weight in is the opinion-based attitudes framework. Our results support the notion, that taking a firm stance on an issue of vaccines might be accompanied by similar processes that are expected in the case of inter-group conflict (strong mutual dehumanization, contact-avoidance and emotional prejudice). Following this line of reasoning, it would be worthwhile to further explore to what extent opinions on vaccines translate into self-aware social identity and readiness for participating in collective action.

This pattern of results sheds light on what may have been one of the reasons limiting the effectiveness of pro-vaccine interventions. There are numerous accounts of ineffective (or even counter-effective) attempts to influence attitudes toward vaccination, despite the use of well-established techniques (for example Doliński et al., 2022; Sadaf et al., 2013). This should not come as a surprise, given that the target group (vaccine-averse individuals or vaccine-skeptics) may view the pro-vaccine message as a message from a hostile group that disparages them and essentially tries to pull them over to their side. On the other hand, pro-vaccine people (who are obviously behind pro-vaccine campaigns) may find it difficult to develop a message untainted by their oversimplifications and negative stereotypes about their target group.

The study was funded by The National Centre for Research and Development in Poland (award: Gospostrateg-II/0007/2020-00) and by Registered Report Funding Partnership (RRFP) between the Society for the Improvement of Psychological Sciences (SIPS) and Collabra: Psychology.

Contributed to conception and design: KI, DD.

Contributed to acquisition of data: KI.

Contributed to analysis and interpretation of data: KI, DD.

Drafted and/or revised the article: KI. DD.

Approved the submitted version for publication: KI. DD.

Contributed to the funding acquisition: KI, DD.

Authors declare no competing interests regarding presented work.

All data, reproducible files for data analyses, scripts for data visualization and experimental materials are publicly accessible via Open Science Framework - https://osf.io/67h3w/.

The study design was pre-registered in accordance with the protocol approved in Stage 1 Registered Report. The preregistration is accessible via Open Science Framework - https://osf.io/vkghr.

Study was approved by Ethical Board of Psychology Department of the SWPS University of Social Sciences and Humanities.