Does the term matter? The labeling effect on the perception of ethnic minorities: The case of The Romani in Serbia

Empirical research on the impact of linguistic labels on social perceptions is scarce, especially in the context of ethnic groups. Across three studies ( N = 1185), we investigated the impact of labels on perceptions of the Romani ethnic group by non-Romani participants in Serbia. In Study 1 ( N = 244), we found some evidence that the Romani elicit more positive perceptions (more sociable and competent) when labeled with the neutral (vs. derogatory) term. In two follow-up studies, we focused on investigating potential mechanisms. In Study 2, we tested whether positive perceptions emerged via perceived higher status, while in Study 3 we focused on the motivation to respond without prejudice and sensitivity to hate speech. Study 2 ( N = 467) replicated the labeling effect showing that Romani were perceived more positively (more sociable and moral) when labeled with the neutral term. However, we found no support for perceived group status as a mechanism. Study 3 ( N = 474) did not corroborate the labeling effect but found the higher external motivation to respond without prejudice (potential mechanism). Meta-analytic effects showed that Romani were perceived as more moral, sociable, and competent when the neutral term was used. These effects were to some extent moderated by ideology as they existed only for right-wing individuals. We conclude that the effect is much smaller than the effects in previous comparable studies.


Introduction
"We are at war, but the only way to win this war is to be as understanding, non-biased and politically correct as possible."PC Principal, South Park As this quote from one of South Park's lead characters satirically demonstrates, political correctness is an important topic in contemporary society; it has changed the public discourse, making normative language more inclusive, symmetrical, non-offensive and non-stigmatizing (for a review, see Maass, Suitner, & Merkel, 2013).Nevertheless, such language remains the topic of heated debates about its impacts on free speech, the beauty of language, or whether it, in fact, has any positive effects (Hughes, 2010).Empirical research on such linguistic effects is heterogeneous: some issues, such as gender-inclusive language, have been studied more extensively (cf.Sczesny, Formanowicz, & Moser, 2016).On the other hand, the effect of offensive language-in which a group is referred to with a derogatory (e.g., "Negro") instead of a neutral term (e.g., "Black")-has been investigated to a lesser extent (Maass et al., 2013).With the present research, we attempted to expand this limited literature by utilizing the fact that similar neutral and derogatory terms exist for the Romani ethnic minority in Serbia.We conducted three studies to answer our central research question-whether the use of neutral or derogatory terms matters for the perception of the Romani (Roma) minority.Additionally, we aimed to discover potential mechanisms through which this effect occurs, and possible ideological moderators of the effect.

The power of (derogatory) labels
The idea that mere labeling may change social perceptions is not without foundation: there is some empirical evidence that using different terms for the same social group might influence prejudice and discrimination.For example, "obese" people elicited more negative attitudes and greater disgust than "fat" people (Vartanian, 2010).Comparable effects were demonstrated for several other social groups: prejudice and punitive behavioral intentions towards immigrants were greater when they were referred to as "illegal aliens", as opposed to "noncitizens" (Rucker, Murphy, & Quintanilla, 2019, also see Pearson, 2010).Similarly, Hall, Phillips, and Townsend (2015) showed that African-Americans were perceived as warmer, more competent, and moral when the term "African--Americans" was used, as opposed to the term "Black" (but see Zilber & Niven, 1995).
However, not all labels have equal valence.Derogatory group labels represent linguistic terms that refer to a social category or its members in an offensive and pejorative manner (Cervone, Augoustinos, & Maass, 2021).Homophobic epithets (e.g., "f*g"), or ethnic or racial slurs (ethnophaulisms; e.g., "n****r") are examples of such terms as they are insulting to the members of a social group they refer to (Carnaghi & Bianchi, 2017;Carnaghi & Maass, 2008).Research on derogatory labels showed how using a neutral term might bring more positive perceptions.For instance, a semantic priming study showed that out-group participants (heterosexuals) had slower reaction times on positive stereotypical traits about homosexuals when primed with a derogatory label, indicating less favorable attitudes ("f*gs" vs. "gays"; Carnaghi & Maass, 2007).In a similar vein, ethnophaulisms might impact social perceptions: overhearing a derogatory label for a black person in the US can lead to a more negative perception of that individual (Greenberg & Pyszczynski, 1985;Kirkland, Greenberg, & Pyszczynski, 1987).Although it seems reasonable to assume that comparable patterns should also emerge in the perception of ethnic groups outside of the US, so far there is not much empirical research on this.We aimed to overcome this gap by investigating how labels affect the perceptions of the Romani ethnic group.
Compared to other ethnic prejudice in social psychological research, there has been neglect of prevalent and unique anti-Roma prejudice ("Anti-Gypsyism"), which is problematic as these prejudices might be at the core of Romani discrimination and marginalization (Kende et al., 2017;2020).Given that using different linguistic labels showed a potential to change perceptions of a group, we believed it would be beneficial to investigate such effects in the social perceptions of the Romani, as a possible step in improving their position.

Social perceptions, prejudices, and stereotypes
According to the Stereotype Content Model (SCM), which is one of the most dominant approaches in social perception, people base their impressions along two dimensionswarmth and competence (Fiske, Cuddy, Glick, & Xu, 2002).Using the SCM, previous research established that Romani were perceived as low in both dimensions, occupying the same space as drug addicts and delinquents in Romania (Stanciu, Cohrs, Hanke, & Gavreliuc, 2017) and Chechens in Russia (Grigoryev, Fiske, & Batkhina, 2019).Recently, however, it has been suggested that the trait of morality is a third fundamental dimension, different from the component of warmth (Goodwin, Piazza, & Rozin, 2014;Goodwin, 2015;Leach, Ellemers, & Barreto, 2007).Given that "moral character" might have a central role in impression formation, we decided to use the three-dimensional model, focusing on the perceptions of morality, warmth (sociability), and competence.
However, focusing on only social impressions would not provide a complete picture.As all these measures are positively worded (e. g., how warm or competent someone is), another, negative component would be missing.Therefore, in the first study, we included valence-based measures (feeling thermometer and social distance in which Romani also score low) and discriminatory attitudes (e.g., not allowing the Romani members to vote).Finally, given the unique stereotypes about Romani, our third study included specific stereotypes (e.g., getting undeserved benefits, being culturally different; Kende et al., 2017), rather than fundamental and more general measures.

Labels might matter, but why? Potential mechanisms
Although the number of studies investigating the labeling effect for different social groups is not negligible, current literature is relatively silent on why different labels would elicit different responses.For example, Rucker et al. (2019) showed that neutral (vs.derogatory) labeling of immigrants leads to less punitive behavioral intentions through less prejudice.Also focusing on immigrants, Wyszynski, Guerra, and Bierwiaczonek (2020) showed that the labels "migrant" and "refugee" (vs."economic migrant") led to less opposition to helping through social perceptions (warmth/competence) (but see Findor, Hruška, Jankovská, & Pobudová, 2021).However, none of these studies investigated why the labels produced different levels of prejudice/social perceptions in the first place.We propose mechanisms that can be grouped in two categories: different prototypes of the target, and/or different contexts (i.e., norms) in responding.

Different prototypes of the target group
One possibility is that different labels provoke different conceptualizations of the target.Carmona, Sindic, Guerra, and Hofhuis (2020) investigated the content underlying six different labels for global identities (e.g., "People all over the world", "Global citizens").They found that these terms which group in two categories-Humanness oriented and Global citizenship oriented-include partially different content so that each refers more to emotional and intellectual traits respectively (also see Spruyt, van der Noll, & Vandenbossche, 2016).The idea that different terms elicit different prototypes resonates well with Hall et al. (2015) suggestion that higher social status estimates of African-Americans (vs.Blacks) led to more positive evaluations.However, they did not formally test this mediation in their study.We agree with their proposition that different terms might have influenced perceived status, which further influenced evaluations of groups.This would mean that the neutral (vs.derogatory) term evokes a concept of a higher (vs.lower) social status target, which in turn brings more positive social perceptions.We tested this possibility in the second study.

Different normative structures
Another reason why different labels would evoke different perceptions could lay in distinct normative structures they elicit.According to the Prejudiced Norm Theory, communicating in a derogatory way (more specifically, using disparagement humor) develops such normative context that allows the expression of prejudices and/or discriminatory attitudes (Ford & Ferguson, 2004).Similarly, the Justification-Suppression model posits norms as one of the potential suppressors of prejudice: their expression will be lower when there are such normative demands (e.g., politically correct environment; Crandall & Eshleman, 2003).This theory draws on the fact that the last few decades brought a non-racist and non-prejudiced normative context, which largely suppressed the overt expression of prejudice (Crandall, Eshleman, & O'Brien, 2002).Following this normative change, Plant and Devine (1998) developed measures that tap into two distinct types of motivation to respond without prejudice: internal and external.The former refers to the reasons one would have regarding her self-perception (e.g., internalized non-prejudiced norms), while the latter concerns the normative pressure from the environment (e.g., concerns with others' opinions if one expressed her prejudices) (also see Klonis, Plant, & Devine, 2005).Applying these insights to the labeling effects, it is possible that some labels, such as derogatory ones, cue normative context in which one is less motivated to respond without prejudice.Accordingly, this individual would report more negative perceptions.In Study 3, we tested this possibility.
Finally, the recently developed Model of expressing hate speech (Bilewicz & Soral, 2020) proposes that mere exposure to hate speech reduces sensitivity to it, which in turn leads to negative evaluations of the outgroup.Indeed, Soral, Bilewicz, and Winiewski (2018) showed that the exposure to hate speech towards Muslims and LGBT individuals decreased sensitivity to it, which subsequently increased prejudice towards these groups.If derogatory group labels are seen as a form of hate speech, then this model implies that exposure to the derogatory label decreases sensitivity to hate speech, which consequently increases self-reported prejudice.This was also tested in Study 3.

Who is perceiving: the importance of ideology
It has been well-established that ideological variables determine prejudice towards different social groups.For instance, right-wing authoritarianism (RWA) has been shown to predict prejudice towards gays and lesbians (Terrizzi, Shook, & Ventis, 2010), as well as women and immigrants (Zakrisson, 2005).While RWA predicts the prejudice towards "dangerous" groups (e.g., terrorists), social-dominance orientation (SDO) does so for "derogated" ones (e.g., psychiatric patients) (Duckitt & Sibley, 2007).Furthermore, it can be argued that individuals who endorse RWA and SDO ideologies might be more sensitive to normative cues implied by group labels.In line with this, Knoll, Redlawsk, and Sanborn (2011) demonstrated that ideology moderated the labeling effect on the perception of immigrants, so that it only occurred in Republicans.Similarly, RWA was shown to be associated with higher levels of prejudice against groups whose label implies deviance ("homosexuals" as opposed to "gays"; Rios, 2013).However, the moderating role of RWA on this labeling effect was not replicated in a higher-powered study (Crawford, Brandt, Inbar, & Mallinas, 2016).On the other hand, the role of SDO in the labeling effect of social groups has not yet been tested.For these reasons, we included both RWA and SDO in our research.
We did not make any hypotheses about the directions of a possible ideological moderation of the effect: there are also reasons why the labeling effect could as well emerge for those on the ideological left.As Strauts and Blanton (2015) showed, usage of the non-offensive language as a form of political correctness is more prominent among liberal participants.Also, given that individuals low on RWA are more internally motivated to respond without prejudice (Plant & Devine, 1998), the normative pressure for liberal participants may be high.Therefore, a neutral term could remind liberal participants about the norms of political correctness, yielding a labeling effect for this ideological group.

Overview of the present research
Across three studies, we aimed to test the effect of using neutral or derogatory labels on outgroup perception of the Romani ethnic minority in Serbia.In Serbian, there are two coexisting terms for Romani; one of them is considered neutral ("Romi") while the other one is derogatory ("Cigani").Both terms are quite prevalent and widely familiar to Serbian citizens, providing a suitable context for testing the labeling effect.To our knowledge, this is the first attempt to investigate the labeling effect on the perception of ethnic groups outside of the US.We posit that studying this effect in an under-researched population such as Serbian, could provide insights into the generalizability as well as contextual specificity of the previous findings.Moreover, we aimed to advance the literature on the effects of ethnic group labels in two additional ways.As the psychological mechanism(s) behind the labeling effect remains unclear from previous studies, we investigated four possible factors: perceived out-group status, internal and external motivation to respond without prejudice, and sensitivity to hate speech, none of which have been previously tested.Finally, given the well-established link between RWA and SDO, and prejudice towards ethnic groups, we explored the possibility of the labeling effect depending on one's ideology.
In Study 1, we investigated whether the typical member of the Romani minority would be perceived differently with regards to fundamental social perception dimensions (morality, sociability, competence), social distance, the feeling thermometer, and discrimination endorsement depending on the label used.We also tested whether RWA and SDO would moderate the labeling effect.In Study 2, we aimed to replicate the findings from Study 1 and test perceived social status as a potential mechanism of the effect.In Study 3, our goal was to expand our findings by testing the labeling effect on Romani-specific stereotypes, as well as to examine other potential mechanisms (motivations to respond without prejudice and sensitivity to hate speech).All studies were pre-registered at the Open Science Framework (OSF), where the research questions and the data are disclosed.Ethical approval was obtained from the Institutional Review Board of the Department of Psychology at the University of -.

Study 1
In Study 1, we investigated whether the labeling effect would occur in the perceptions of the Romani. 2 We tested whether this ethnic group would be perceived differently in terms of morality, sociability, competence, prejudice, and discrimination endorsement when different labels were used (neutral vs. derogatory).We also tested potential moderating effects of RWA and SDO on these effects, to check if these ideological orientations exhibit effects similar to what RWA showed for labels for gay people (Rios, 2013).Research questions were pre-registered via the OSF (link).

Method
We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study.The design of the study was pre-registered via the OSF.Materials, procedure, and exclusion criteria can be retrieved here.

Participants
Based on the power analysis for the global effect in a MANOVA (2 groups, 6 response variables, α = .05,β = .8,f 2 = .0625;Faul, Erdfelder, Lang, & Buchner, 2007), we planned for a sample size of 230.Due to the exclusion criteria, we aimed for around 260 responses.Participants were recruited online via Facebook groups and Reddit.While 516 participants voluntarily started the survey, a total of 268 participants completed the study.Our final sample consisted of 244 participants (101 females, M age = 28.6,SD age = 11.25).Three participants indicated education less than high school, 41 were attending or had completed high school, 101 were students and 99 had completed higher education.The final sample was obtained after excluding 12 participants who failed the attention check question ("This is an attention check.Please answer 'not sure' to this item") and 12 participants who were multivariate outliers (Mahalanobis distance).Although our pre-registered criteria for outliers was Minimal Covariance Determinant 75 (breakdown .25,p = .01;Leys, Klein, Dominicy, & Ley, 2018), this approach identified 42 outliers, which we deemed as too large a proportion of our sample.No data analyses were performed before the participants were excluded.

Materials and procedure
We report all measures that participants completed in the exact order.After agreeing with the consent form, participants filled out demographic information (gender, age, and education).Next, they were presented with the RWA and SDO scales.Means and standard deviations of the scales for each ethnic group and both terms, and scale reliabilities are given in Table 1.
RWA was measured using a 10-item scale from Todosijević (2013).Responses were given on a five-point scale (1 = "completely disagree" to 5 = "completely agree").One of the items was "It would be best for everyone if the proper authorities censored magazines and movies to keep trashy material away from the youth" (α = .85).SDO was measured using six items with the highest factor loadings from Todosijević (2013).Participants used a seven-point scale (1 = "completely disagree" to 7 = "completely agree") to express their 2 In Study 1, we also tested the labeling effect on the perception of Albanians (see Supplement at the OSF).In the main text, we only report the findings for the perceptions of Romani to keep the focus of the paper.
V. Gligorić et al. agreement with items such as "To get ahead in life, it is sometimes necessary to step on others" (α = .81).
Next, participants rated a typical member of a Romani ethnic group on three social perception dimensions (morality, sociability, competence), social distance, the feeling thermometer, and support for discrimination.Each participant was randomly assigned to rate a typical Romani labeled with either the neutral or derogatory label (between-subjects design).Using the derogatory term in the study had to be justified, so we explained that both labels (neutral and derogatory) are used to refer to the group with the derogatory term being the endonym.The introduction of the block was as follows: "Several ethnic minority groups live in Serbia, making up 17 % of the total population.The members of one of these groups are called Romani or Gypsies (endonym of this group).An endonym is a term that the group members use to label their group.In this research, the term Romani [Gypsies] will be used." 3he participants were instructed to think about the "typical member of the group" when answering the questions.The presentation of the following scales, and the items within scales, were randomized.
Social distance.We used a modified social distance scale with a Likert scale instead of binary choices (e.g., Dietrich et al., 2004).This scale consists of five items representing the social relationships to the respondent (neighbors, friends, partners, colleagues, visiting the person's place).Participants expressed their willingness to admit a member of the Romani group on a five-point scale (1 = "not willing at all" to 5 = "definitely willing").
Discrimination endorsement.We used a shortened five-item scale from Sotelo (2002) to measure discrimination toward Romani members.Participants reported their agreement (1 = "fully disagree" to 5 = "fully agree") with items such as "A member of this group should not be allowed to vote".

Results
The full database and the R script are available on OSF via this link, where the interested reader can find a correlation table for all scales used.There was no missing data because participants could not move to the next page unless they answered all questions.
Next, to further investigate the labeling effect for the perceptions of the Romani, we examined the moderating effects of RWA and SDO by including them as factors in a MANOVA.RWA was a significant moderator of the labeling effect, Pillai's Trace = .05,F(6, 235) = 2.21, p = .04,η 2 = .05.Specifically, the strongest moderating effect of authoritarianism was registered for discriminating attitudes, F (1, 240) = 10.738,p = .001,η p 2 = .04,so that the labeling effect was registered only for those high in RWA (Fig. 1).The same pattern (marginally significant) was observed for the effect on distance (F = 3.151, p = .08,η p 2 = .01)and competence (F = 2.995, p = .08,η p 2 = .01)such that the effect was observed for participants high in RWA.On the contrary, SDO did not show a moderating effect on dependent variables, Pillai's Trace = .03,F(6, 235) = 1.242, p = .29,η 2 = .03.Note.The thermometer does not have reliability as it is a one-item measure.

Discussion
In Study 1, we investigated the social perceptions of the Romani when this group is labeled using different terms (neutral vs. derogatory).In the neutral condition, the Romani were perceived as more sociable and competent, marginally less distant and the endorsement of discriminating policies was marginally lower.The effect for discrimination endorsement was moderated by RWA (but not SDO) so that it was present only for those high on RWA.In our second study, we wanted to replicate the labeling effect for social perceptions of the Romani and investigate the potential mechanism of the effect.

Study 2
We conducted the second study for several reasons.First, we did not include a question about ethnicity in Study 1, meaning we might have had Romani people in our sample, which could have decreased the validity of the results.Second, we wanted to replicate the findings given that the effects emerged only for some of the social perception variables-sociability, competence, and marginally for distance and discrimination-while it did not emerge for the feeling thermometer and morality evaluations.We excluded measures of discrimination due to floor effects and the feeling thermometer which had large variability.We also wanted to see whether ideology more broadly (as opposed to RWA) shows the same moderation effects.Finally, we wanted to test whether different conceptualizations of the social status of the Romani could mediate the labeling effect.We assumed that when thinking about the Romani referred to with the neutral label, one would be more likely to think of a more respected person with somewhat higher income, education, and occupation as compared to those labeled with the derogatory term.This should, in turn, improve social perceptions.To test this hypothesis, we conducted a high-powered study.Same as the first study, the second study was pre-registered via OSF (link).

Method
We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study.

Participants
Based on the mediation effect with small and large path sizes (α = .14,β = .59;Fritz & MacKinnon, 2007), we aimed at 400 participants.Participants were recruited online via Facebook groups and posts completing the survey voluntarily.While 1276 participants started the survey, 581 participants completed it.We excluded 96 participants who failed the attention check question ("This is an attention check.Please select "Other" and type 1 ′′ ), one participant who indicated Romani ethnicity, and 17 participants who were outliers based on Mahalanobis distance.This left the final sample of 467 (267 females, M age = 35.8,SD age = 10.9)out of which 424 indicated Serbian ethnicity, 10 Hungarian, six Bosniak, one Albanian and 26 indicated other ethnicities.Eight participants indicated education less than high school, 147 had completed high school, 63 were students and 237 had completed higher education.Due to a mistake in programming, we did not record education for 12 participants.There was no other missing data.No data analyses were performed before the participants were excluded.

Materials and procedure
After agreeing with the consent form, participants filled out demographic information (gender, age, and education) and indicated their ideological stance on social and economic issues (two items ranging from 1 = "left-wing" to 9 = "right-wing").Next, they were randomly assigned to one of two conditions.In the first condition, participants indicated their social perceptions of the Romani labeled with the neutral term.In the second one, participants had the same task, but the Romani were labeled with the derogatory term.The participants rated a typical Romani member on the following scales (presented in the randomized order).
Participants filled out the same three-item scales of morality, sociability, and competence as in the first study (Leach et al., 2007).They also completed the social distance scale from Study 1, but with different social relationships (roommate, friend, boyfriend/girlfriend, colleague, spouse).Participants also rated a typical Romani on several indicators of social status: monthly income was estimated using a slider from 0 to 50 000 RSD (converted to euros, the scale ranges from 0 to around 416).Education was estimated using a multiple-choice question with the responses ranging from 1 = "no elementary school" to 7 = "post-graduate degree".
The occupational position was also estimated using a multiple-choice question with nine options ordered from the lowest to the highest in the hierarchy, 1 = "Basic, simple occupation (e.g., hawker, cleaner)" to 9 = "Lawmakers, officials, directorsmanagers (e.g.CEOs)".Social reputation was estimated using a five-point scale (from 1 = "very low" to 5= "very high").
Means and standard deviations of scales for each label, and reliabilities are given in Table 2.

Results
The full database and the R script are available on OSF (under Study 2 menu) where the correlations between used scales can be found.
Given that MANOVAs give a conservative estimate, we proceeded with the preregistered ANOVAs.These showed that the Romani were perceived as more sociable (F(1, 465) = 7.260, p < .01,η 2 = .02)and moral (F(1, 465) = 6.301, p = .01,η 2 = .01)when the neutral term was used.They were not rated as significantly more competent (F(1, 465) = 1.833, p = .18,η 2 = .004)and less distant (F(1, 465) = 2.042, p = .15,η 2 = .004)when the neutral term was used.Next, we tested whether different labeling led to different status perceptions.MANOVA did not show the labeling effect, Pillai's Trace = .01,F(4, 462) = .682,p = .60,η 2 = .006,suggesting that there was no effect on the status estimation.Therefore, status estimation could not have mediated the effect on social perceptions, so we did not proceed with the mediation analyses.In sum, using the term for the Romani led to slightly more positive attitudes, but did not affect the perception of their status (Table 2).

Ideology as a moderator
We investigated the moderating role of ideology on the labeling effect by including ideology for social as well as economic issues as factors in a MANOVA.While ideological stances on social issues did not moderate the labeling effect on dependent variables (Pillai's Trace = .008,F(4, 460) = 0.997, p = .41,η 2 = .009),economic ideology marginally moderated the labeling effect (Pillai's Trace = .019,F(4, 460) = 2.298, p = .058,η 2 = .02).Separate ANOVA showed that economic ideology moderated only the labeling effect on the perception of competence (F(1, 463) = 8.597, p < .01,η p 2 = .02).As Fig. 2 shows, the neutral term increased the perception of competence, but only among those with left-wing views on economic issues.

Discussion
Although we partly replicated the findings that the neutral term brings more positive perceptions of the Romani, the moderating effect of ideology showed unexpected patterns.Namely, we found that the neutral term increased the perception of Romani's competence only for those on the economic left, which is in contrast to the findings from the first study where the effect was found only for those high in RWA.We also did not find the support that perceived status is the mechanism underlying the effect, since the participants estimated the status of the Romani as relatively low regardless of the term used.To resolve these inconsistencies about the role of ideology and test other potential mechanisms, we conducted a third study.

Study 3
The first two studies indicated that the neutral label for the Romani might lead to more positive perceptions of this group in terms of general judgements about group members' warmth and competence.To further test the scope of the labeling effect on the perception of the Romani, in Study 3 we tested this effect for specific anti-Romani attitudes which include specific blatant stereotypes related to criminality, receiving undeserved benefits, and being culturally diverse from the rest of the population (Kende et al., 2017).Additionally, we continued our exploration of potential mechanisms of the effect.To this end, we tested internal and external motivation to respond without prejudice, as well as sensitivity to hate speech.Finally, given that the first two studies pointed to somewhat opposing moderating effects of ideological variables, we included all measures of ideology used across the first two studies.As for previous studies, Study 3 was also pre-registered via the OSF (link).

Method
We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study.

Participants
Based on the mediation effect with small and large path sizes (α = .14,β = .59;Fritz & MacKinnon, 2007), we aimed at 400 participants.Participants were recruited in the same way as in Study 2. Out of 1190 participants who started the survey, 623 completed it.We excluded 119 participants who failed either of two attention checks ("This is an attention check question.Please answer agree/agree to some extent to this question."),already participated in a similar study (10), indicated Romani ethnicity (2), or were

Materials and procedure
The design of the study was the same as in Study 2. After demographic information, participants indicated their ideological stance on social and economic issues (Study 2) and filled out the RWA (α = .82)scale from Study 1.4 Next, participants were randomly assigned to one of two conditions (neutral vs. derogatory label), which asked about the perceptions of the Romani.The participants reported their view of a typical member of this group on the following scales (presented in the randomized order).
Participants filled out the same three-item scales of morality, sociability, and competence from the previous two studies.They also filled out three measures of Roma-specific stereotypes indicating their agreement on a seven-point Likert scale, 1 = "completely disagree" to 7 = "completely agree" (Kende et al., 2017).These three scales were Blatant Stereotyping (e.g., "Roma people tend to make more criminal acts than other people"), Undeserved Benefits ("I think that Roma people in this country are given preferential treatment in certain aspects"), and Cultural difference ("Music and dancing is something Roma children already learn in the womb").Next, using a five-point Likert scale (1 = "strongly disagree" to 5 = "strongly agree") they filled out the five-item measures of internal ("Because of my personal values, I believe that using stereotypes about Roma people is wrong") and external motivation ("I try to hide any negative thoughts about the Romani in order to avoid negative reactions from others") to respond without prejudice (Plant & Devine, 1998).Finally, to test whether the terms are seen as indeed neutral and derogatory, we asked participants in both conditions to indicate how offensive neutral and derogatory terms are (1 = "not at all offensive" to 5 = "very offensive").The question about the offensiveness of the derogatory term was a measure of the sensitivity to hate speech which we tested as a mediator.Means and standard deviations (by label), and reliabilities of the scales are given in Table 3.

Results
The full database and the R script are available on OSF via this link (under Study 3 menu) where the correlations between used Note.Estimated status measures do not have reliability as they are one-item measures.scales can be found.As in previous studies, there was no missing data.Firstly, we wanted to test whether the derogatory term is indeed perceived as such, compared to the neutral one.A paired-sample t-test showed that the derogatory term (M = 2.24) was perceived as more offensive than the neutral (M = 1.21), t(473) = 16.614,p < .001,Cohen's d = .763.

Meta-analytic effects of the labeling
While the first two studies indicated that using neutral labels for the Romani might improve some aspects of social perception, we did not fully replicate this in Study 3, where none of the effects was significant.As the effects in the first two studies were small, it can be expected that the effect will not be detected in each empirical study even though it exists (Lakens & Etz, 2017).Therefore, as the same conditions and measures of morality, sociability, and competence were used, we decided to conduct three meta-analyses (one per dependent measure).All three fixed-effects meta-analyses showed significant effect of labeling: Romani were perceived as more moral (Cohen's d = .135,CI 95 [0.021, 0.250], p = .02),sociable (Cohen's d = .193,CI 95 [0.078,0.307],p < .001),and competent (Cohen's d = .128,CI 95 [0.014, 0.242], p = .027)when the neutral (vs.derogatory) label was used.This shows that there indeed is a small, but significant effect of different labels on the perceptions of the Romani.

General discussion
In the present research, we investigated the labeling effect (the use of neutral vs. derogatory terms) on the social perceptions of the Romani in Serbia.In Study 1, we found that the Romani were perceived more positively (on dimensions of sociability, competence, and Note.Perceived offensiveness of the derogatory term is a one-item measure and thus does not have reliability.
marginally on distance and discrimination endorsement) when the neutral term (as opposed to the derogatory one) was used.In the second study, we replicated the findings for the perception of the Romani (on dimensions of sociability and morality) but did not find the support for perceived status as a mediator of the relationship.In the third study, we obtained the same pattern of the results as in the first two studies but failed to attain statistical significance.However, we found the labeling effect on the external motivation to respond without prejudice, suggesting this could be a mediator of the effect found in the first two studies.Importantly, meta-analytic effects from all three studies showed that the neutral term for the Romani increased the perceptions of morality, sociability, and competence.These results provide some support towards the notion that using neutral labels can, to some extent, improve the perception of an ethnic minority (in this casethe Romani in Serbia).Although small, these effects were obtained from one-time, brief manipulations, whereas it would be beneficial to explore the potential cumulative effect of repeated exposure to neutral vs. derogatory labels.In any case, if there is the prospect of even a slight improvement in the perception of a group as marginalized as the Romani from something so simple as using the neutral label (to which our results point), it might be wise to use this insight.Yet, we did not investigate actual behavior, which would also be beneficial.This could be one of the avenues for future research.
Notably, the effect sizes we found are lower than those found in previous comparable studies that investigated linguistic effects on social perception (e.g., Rucker et al., 2019;Hall et al., 2015).This illustrates that the labeling effect is not uniform: likely, it is dependent on the perceived and the perceiving group, and the exact terms used.Future studies should aim to discover the specific groups and contexts for which the labeling effect exists.

Potential mechanisms
In Studies 2 and 3, we aimed to investigate why the term used for the Romani might matter for social perception.In Study 2, we found no support for the notion that different labels influenced the estimated social status of the Romani.This contradicts the finding in the U.S. where "African-Americans" were estimated to have higher status than "Blacks" (Hall et al., 2015).Therefore, the labeling effect probably did not occur via the conceptualization of different prototypes (at least regarding the status).In the third study, we tested three more mediators.As mentioned earlier, another possible reason for the labeling effect is that different labels elicit different norms, so that using the derogatory one allows for more honest responses (i.e., less motivation to respond without prejudice).Although we could not test this directly as we did not detect the labeling effect for the main dependent variables in Study 3, we found a higher external (but not internal) motivation to respond without prejudice when the neutral term is used.This would suggest that the neutral term reminded participants that there is social pressure to answer without prejudice, which resulted in more positive perceptions.Such findings would be in line with the Prejudiced Norm Theory (Ford & Ferguson, 2004) and the Justification-Suppression model (Crandall & Eshleman, 2003) which indicate that norms can affect the expression of prejudice (or social perceptions in our case).Future studies should test this possibility.
Finally, the Model of expressing hate speech (Bilewicz & Soral, 2020) suggested that exposure to hate speech (derogatory labels in our case) decreases sensitivity to it, thus leading to more expressions of prejudice.However, in our study, participants exposed to the derogatory label did not perceive that label as less offensive (i.e., they were not less sensitive to it).This is in contrast with the predictions of the model; one of the reasons we did not find the differences might be that the derogatory label we used does not represent hate speech.Another option is that the exposure to the derogatory label was not long enough to affect sensitivity.In any case, these boundary conditions of the model should be taken into account.

The role of ideology in the labeling effect
We also found evidence that ideology moderates the labeling effect, such that it only emerges among right-wing individuals.While Studies 1 and 3 gave similar patterns of results (the neutral term brought about more positive evaluations only for the individuals on the ideological right), there was no such effect in Study 2. On the contrary, in Study 2, the effect on one dependent variable (competence) emerged only for left-wing economic ideologues.However, since this was the case in only one statistical analysis in one of the studies, while the effect was found among right-wing individuals 11 times (three times in Study 1 and eight in Study 3), the overall pattern strongly suggests that ideology moderates the effect so that it only emerges for individuals on the right, which is in line with other studies (e.g., Knoll et al., 2011).Yet, the question of why this would be the case remains open: one possibility is that the derogatory term implies deviance to which high-RWA individuals are more sensitive (Rios, 2013).Another possibility is that it occurs because people high on RWA are more sensitive to norms (Altemeyer, 1981), adhering to them more when the neutral term is used.Finally, it is possible that the effect occurs only when there already is a high level of negative attitudes in the first place, which would be the case for those high on RWA.This would be in line with the Prejudiced Norm Theory which posits that offensive communication in the form of disparagement humor brings more tolerance to discrimination only for those high in prejudice (Ford & Ferguson, 2004).Future research should investigate these possibilities.

Limitations and conclusion
Our study is not without limitations: we did not record ethnicity in the first study.However, as Studies 2 and 3 showed, there was only a small number of the Romanis in our sample (two out of 468 in Study 2 and four out of 476 in Study 3), suggesting that this probably did not influence the findings of the first study.Next, we used several scales that might have been too short (e.g., one-item measure of the feeling thermometer) and unreliable (e.g., economic conservatism).Using longer and more reliable scales should be done to quantify the effect more reliably.Nevertheless, our research contributes to the literature in several important ways.First, we investigated the usage of different terms on the perceptions of ethnic groups (and especially Romani), an area understudied in empirical research.Second, this effect might not be uniform for all individuals-their ideological stances should also be considered.These boundaries of the effect ought to be taken into account in the development of much-needed theoretical frameworks for the effects of language on group perception.

Fig. 1 .
Fig.1.RWA as the moderator of the labeling effect for discrimination attitudes towards Romani.High-RWA participants were susceptible to the labeling effect on discrimination policies.A similar pattern emerged for social distance and the perception of competence.

V
.Gligorić et al.   outliers based on Mahalanobis distance (18).This left the final sample of 474 (278 females, M age = 34.7,SD age = 11.1)out of which 399 indicated Serbian ethnicity, 10 Hungarian, 12 Bosniak, two Albanian, and 51 indicated other ethnicities.Six participants indicated completed elementary school, 162 completed high school, 95 were students, and 211 had completed higher education.No data analyses were performed before the participants were excluded.

Fig. 2 .
Fig. 2. Moderating role of ideology of the labeling effect on perceived competence.The labeling effect emerged only among those who had left-wing views on the economy.

Fig. 3 .
Fig. 3. Violin plots of the labeling effect on different scales of the perception of Romani for right-wing individuals (using three different cut-off variables).Right-wing individuals tend to judge the Romani more positively when the neutral (vs.derogatory) term is used.

Table 1
Reliabilities, means, and standard deviations (in brackets)for scores on dependent variables for the perceptions of the Romani by the two terms.

Table 2
Reliabilities, means and standard deviations (in brackets) for scores on dependent variables (top) and estimated status (bottom) for two labels.

Table 3
Means (standard deviations in brackets) for scores on dependent variables (top) and potential mediators (bottom) for two labels.Reliabilities (Cronbach's α) are given in brackets next to the scale name.