Abstract
According to infrahumanization theory, people ascribe outgroups lesser humanity than their ingroup. Infrahumanization has been proposed as a relatively subtle and indirect form of emotional prejudice. We tested this assumption further in an experiment during which participants were asked to take on the perspective of a prejudiced or a nonprejudiced person or received no such instruction before completing a questionnaire that included measures of dehumanization and explicit prejudice. Results showed that whereas the infrahumanization effect was unbiased by faking instructions, participants readily faked their responses on the more direct measure of prejudice. The results support the notion that infrahumanization represents a rather indirect form of discrimination that is less prone to social desirability tendencies than other self-report measures of prejudice. Moreover, our findings point to the importance of emotion valence in the context of infrahumanizing others. Implications of our findings will be disussed.
Previous research has demonstrated that people tend to reserve human essence for their ingroup, thereby infra-humanizing the outgroup in an intergroup setting (for a recent review, see Leyens et al., 2007). According to infrahumanization theory (Leyens et al., 2000, 2001, 2003, 2007), humanity is operationalized in terms of secondary emotions, which are differentially attributed to social groups. Secondary emotions such as hope, love, guilt, and contempt are harder to recognize in outgroupers, are developed later in life and represent one core aspect of humanity (e.g., Demoulin et al., 2004; Leyens et al., 2000, 2001). On the other hand, primary emotions such as happiness, surprise, sadness, and anger characterize both humans and animals.
Because people are largely unaware about the distinction between uniquely human and non-uniquely human emotions, the effect has been characterized as relatively indirect and subtle (Boccato, Capozza, Falvo & Durante, 2008; Boccato, Cortes, Demoulin & Leyens, 2007; Demoulin, Cortes & Leyens, 2009; Leyens et al. 2007). Furthermore, the phenomenon has been coined a form of emotional prejudice.
To examine these propositions further, we conducted an experiment that investigated one variant of infrahumanization, namely endorsing emotions as typical of social groups. First, we tested the subtlety of the phenomenon more directly: In line with existing research on faking measures of attitudes and prejudice (e.g., Dovidio, Kawakami & Beach, 2000; Dovidio et al., 1997, Fazio & Olson, 2003; Fiedler & Blümke, 2005; McDaniel, Beier, Perkins, Goggin, & Frankel, 2009; Olson & Fazio, 2009; Steffens, 2004), we conducted an experiment in which participants were provided with fake-good, fake-bad, or no additional instructions to study whether participants’ responses to direct and indirect forms of prejudice would be affected. The underlying rationale was that infrahumanization, if an indirect phenomenon, should not be biased by specific faking instructions vs. none. This is due to the fact that reactions to indirect measures should be beyond an individual’s conscious control (Steffens, 2004). Accordingly, we would interpret a lack of effects in terms of a lack of awareness of the fact that the typically human, secondary emotions are particularly suitable to discriminate against devalued outgroups. In other words, if people infrahumanize others without being aware of it, they will not know how to turn faking instructions into action.
On the other hand, it seems plausible that people rely on emotion valence instead when having to select emotions as typical of one group, particularly when instructed to make a socially desirable (vs. undesirable) impression.
Second, we examined the notion that infrahumanization would represent a form of prejudice by correlating it with self-reported explicit prejudice. Moreover, we predicted that explicit prejudice should be more prone to faking instruction than infrahumanization. Including explicit prejudice in the present research also served to assure that the lack of effects of the instruction manipulation on infrahumanization would not merely be due to participants’ general unwillingness to follow experimental instructions.
We investigated these open research questions in the German vs. Roma intergroup context 1 because the Roma have been and still remain a target of strong negative prejudice and discrimination throughout history, both in German society and in other countries (see Klicpérova-Baker & Kostál, 2011; Marcu & Chryssochoou, 2005).
The following hypotheses were tested in the experiment: First, we predicted an infrahumanization effect (i.e., a Group × Emotion Type interaction). Accordingly, the German ingroup should be attributed more secondary emotions than the Romany outgroup, whereas no differences between groups were hypothesized regarding primary emotions. Importantly, if infrahumanization were fakeable, we would predict a Group × Emotion Type × Instruction interaction, showing stronger infrahumanization when instructed to “fake bad”, and weaker infrahumanization when instructed to “fake good”.
Analogously, we tested the alternative notion that participants would base their evaluation of both groups on the valence of the emotions rather than their uniquely human content. This would show in a Group × Emotion Valence interaction. That is, we assumed that Roma would be characterized with more negative emotions terms than Germans, whereas Germans would be attributed more positively valenced emotions by German participants than Roma. We hypothesized that this effect could likely depend on Instruction: That is, participants would attribute more positive emotions to the outgroup than the ingroup in an attempt to appear nonprejudiced (i.e., faking good). Similarly, participants would select more negative emotions to characterize the outgroup than the ingroup when asked to appear prejudiced (i.e., faking bad).
Finally, regarding Explicit Prejudice, we hypothesized a main effect of Instruction. This would show that participants readily bias their response according to predefined instructions. To demonstrate that infrahumanization would represent a form of emotional prejudice, we predicted a positive correlation between infrahumanization and explicit prejudice, providing evidence of convergent validity.
Method
Participants and design
Participants were German university students ((N = 225, 110 male, 115 female) with a mean age of 25.26 years (SD = 7.29). They were on average in their 6th semester of study (SD = 3.22). Only 15 participants (7%) were psychology majors. Participants were randomly assigned to one condition, with Instructions (Fake good vs. fake bad vs. control) and Order of Group Rating (Germans first vs. Roma first) as between-subjects factors, and Emotion Type (primary vs. secondary emotions), Group (Germans vs. Roma), and Valence (positive emotions vs. negative emotions) as within-subjects factors. They received confectionary for participation in the experiment.
Materials and procedure
Participants were tested individually in the laboratory using paper-and-pencil questionnaire packages. Initially, participants were told that they would take part in a survey study that involved a brief imaginary scenario.
Instructions
Depending on experimental condition, participants received different instructions that they were to follow during completion of the questionnaires. In both the fake good- and the fake bad-condition, participants were instructed to imagine taking part in “a class on the topic ‘prejudice against minorities in German society.’”. They were told that during one session of the seminar, they had been randomly assigned to a group of individuals who were asked to pretend to be either low (vs. high) in prejudice against the group of Roma. Accordingly, participants were asked to complete the subsequent questionnaires while taking on the perspective of a person who would hold unprejudiced (vs. prejudiced) and tolerant (vs. intolerant) attitudes towards Roma. These instructions were presented on the top of every page of the questionnaire to make sure that participants would comply with them. In the control condition, participants were merely asked to complete the questionnaire package and received no further instructions.
Emotions
The attribution of emotions to Germans relative to Roma was assessed using a list of 20 emotions presented in alphabetical order (see Appendix 1). To evaluate both groups, participants were asked to choose 8–10 features that would be most typically for Germans (vs. Roma). A similar procedure was used by Demoulin, Cortes, Viki, et al. (2009) and Leyens et al. (2001). 2
Explicit prejudice
We used a German adaptation of the 20-item Subtle and Blatant Prejudice Scale (Pettigrew & Meertens, 1995; Zick, 1997) to measure explicit prejudice against Roma. Example items read: “Roma belong to a less talented race, that’s why they’ll never get as far as Germans”, or “How big are the similarities and differences between Roma living in Germany and Germans concerning the values they impart to their children/young adult?”. Seven-point Likert scales were used to collect participants’ responses.
Perceived primary and secondary emotions
To assure that the 20 emotions used in this study would be perceived as uniquely vs. non-uniquely human, participants were asked to rate them at the end of the experiment. That is, they had to indicate the extent to which the respective emotions could 1, be experienced by animals and humans; or 7, by humans only. Importantly, participants were instructed to provide responses that would reflect their honest personal opinion before completing the ratings.
Results
Preliminary analyses
Internal consistencies (Cronbach’s α) were calculated: The Explicit Prejudice index was based on participants’ average responses to the 20 items and was highly reliable (α = .86).
The 10-item index for perceived primary emotions had a low internal consistency, α = .72, whereas high reliability was obtained for the 10-item index for perceived secondary emotions, α = .86.
With regard to primary and secondary emotions attributed to Germans and Roma, sum scores were computed to represent the respective indices. For purposes of correlation analyses, we computed an Infrahumanization index [infrahumanization = (ingroup secondary emotions + 1) / (ingroup secondary emotions + ingroup primary emotions + 1) – (outgroup secondary emotions + 1) / (outgroup secondary emotions + outgroup primary emotions + 1)]. Positive values mean that the relative proportion of typically human emotion ascription is higher for the ingroup than for the outgroup, whereas negative values mean the opposite.
Perceived primary and secondary emotions
As predicted, participants perceived primary emotions as less applicable to humans (M =2.01, SD = 0.79) than secondary emotions (M = 5.66, SD = 1.20), t(224) = 38.46, p < .001.
Test of main hypotheses
A preliminary repeated-measures ANOVA with Group (Germans vs. Roma) and Emotion Type (Primary emotions vs. secondary emotions) as within-subjects factors, and Order of Group Rating as between-subjects factor yielded no significant effects (Fs < 1). Therefore, the factor Order of Group Ratings was excluded from further analyses.
A mixed model ANOVA on Group, Emotion Type, and Emotion Valence as within-subjects variables and Instruction as between-subjects variable revealed a main effect of Group: That is, participants attributed more emotions to Roma (M = 2.10) than to Germans (M = 1.71), F(1, 222) = 38.02, p < .001, η2 = .15. We also obtained a main effect of Instruction, F(2, 222) = 3.21, p < .05, η2 = .03.
Importantly, we had predicted an infrahumanization effect that showed in a significant Group × Emotion Type interaction, F(1, 222) = 142.45, p < .001, η2 = .39. Specifically, participants attributed more secondary emotions to Germans (M = 2.45, SD = 1.63) compared to Roma (M = 1.25, SD = 1.27), t(224) = −8.00, p < .001. Contrary to our predictions, the opposite pattern of results was found for primary emotions, t(224) = 13.58, p < .001. That is, overall, more primary emotions were attributed to Roma (M = 2.95, SD = 1.78) than to Germans (M = 0.96, SD = 1.32).
If the infrahumanization effect were fakeable, we would have obtained a significant Group × Emotion Type × Instruction interaction. However, this was not the case, F(2, 222) = 1.75, p = .18, η2 = .02.
The pattern of means are illustrated in Figure 1.
Attribution of primary and secondary emotions as a function of Group, Emotion Type, and Instruction.
Contrary to previous research by Leyens and colleagues, the infrahumanization effect was substantially moderated by Emotion Valence, F(2, 222) = 35.04, p = .001, η2 =.14. Negative primary emotions were perceived more typical of Roma (M = 1.66, SD = 1.35) than positive ones (M = 1.28, SD = 1.24), t(224) = -3.00, p = .003, whereas positive primary emotions (M = 0.65, SD = 0.95) were perceived as more typical of Germans than negative ones (M = 0.32, SD = 0.67), t(224) = 5.14, p < .001. Additionally, negative secondary emotions were perceived as more typical of Roma (M = 0.83, SD = 0.05) than positive ones (M = 0.72, SD =1.01), t(224) = 2.03, p = .04, whereas positive (M = 1.18, SD = 1.14) and negative secondary emotions (M = 1.27, SD = 1.14) were perceived as equally typical of Germans, t(224) = -.80, p = .43.
Analogously, we examined the role of valence in emotion attribution further: The Group × Emotion Valence interaction, F(2,222) = 4.63, p = .03, η2 =.02 was also inspected more closely using simple effects analyses. Surprisingly, overall, participants attributed an equal amount of positive emotions to both groups (M Germans = 0.92, SD Germans = 0.76 vs. M Roma = 1.00, SD Roma = 0.86), t(224) = -1.25, p = .21. As predicted, however, they attributed more negative emotions to Roma (M Roma = 1.10, SD Roma = 0.76) than to Germans (M = 0.79, SD = 0.69), t(224) = -4.86, p < .001.
Moreover, we obtained a significant Group × Emotion Valence × Instruction interaction, F(2,222) = 28.04, p < .001, η2 =.20. Simple effects analyses showed that when participants were instructed to appear nonprejudiced, they attributed more positively valenced emotions to Roma (M = 1.62, SD = 0.91) than to Germans (M = 0.99, SD = 0.94), t(73) = 5.92, p < .001. There was no difference between both groups regarding the attribution of negative emotions (M Roma = 0.72, SD Roma = 0.66 vs. M Germans = 0.88, SD Germans = 0.72), t(73) = -1.56, p = .12.
Conversely, when participants were instructed to present themselves as prejudiced, they selected more positive emotions for Germans (M = 0.97, SD = 0.76) than for Roma (M = 0.49, SD = 0.63), t(73) = -4.04, p < .001, whereas they attributed more negative emotions (M = 1.39, SD = 0.81) to Roma than to Germans (M = 0.78, SD = 0.77), t(73) = 5.23, p < .001.
Table 1 depicts additional significant interaction effects. No other significant findings emerged.
Additional interaction effects
Explicit prejudice
To test whether Instructions and Order of Group Rating would have an effect on more explicit self-reported prejudice against Roma, ANOVA and subsequent t-tests were conducted. Order of Group Rating did not affect reports of Explicit Prejudice, F < 1. Importantly, we obtained the hypothesized main effect of Instruction on Explicit Prejudice, F(2, 217) = 198.27, p < .001, η2 = .65.
Specifically, t-tests revealed that the fake-good instructions reduced self-reported Explicit Prejudice (M fake-good = 2.71, SD fake-good = 0.77 vs. M control = 4.21, SD control = 1.19, t(149) = 9.76, p < .001) relative to the control group without additional instructions. Furthermore, as predicted, fake-bad instructions increased prejudicial responses (M fake-bad = 5.81, SDfake-bad = 0.77 vs. M control = 4.21, SD control = 1.19, t(149) = 9.03, p < .001). The mean difference in Explicit Prejudice between participants in the fake-good vs. the fake-bad condition was also significant, t(144) = 24.27, p < .001.
Figure 2 illustrates that participants clearly followed experimental instructions and biased their responses on the explicit prejudice measure accordingly.
Average responses to the Prejudice index as a function of Instructions.
Correlation analyses
Correlation analyses were conducted on data of the control group without additional instructions to examine the relationship between the Infrahumanization index and Explicit Prejudice. As predicted, the correlation was positive, r(77) = .63, p < .001, providing empirical evidence for the fact that the infrahumanization measure used in our study constitutes a valid measure of emotional prejudice.
Discussion
Previous research on infrahumanization has characterized the phenomenon as a mild and subtle form of emotional prejudice (e.g., Leyens, 2007). We argued, however, that if the effect were indeed so subtle in nature, infrahumanization should not be biased by explicit additional faking instructions. If it were prone to faking, however, German participants should infrahumanize Roma outgroupers more when faking bad, whereas we hypothesized that they would infrahumanize Roma less when instructed to fake good. Furthermore, we predicted that—on the contrary—questionnaire measures that commonly serve to assess outgroup prejudice more directly would be more strongly affected by specific instructions to respond to them as if one were a prejudiced or an unprejudiced person. Moreover, we reasoned that if infrahumanization were a form of emotional prejudice, both measures should be positively correlated.
Besides investigating the interplay between infrahumanization and explicit prejudice and the measures’ immunity to conscious faking, we were also interested in examining the role of emotion valence in infrahumization. This was based on the following assumptions: We agued that instead of distinguishing between uniquely human and non-uniquely human emotions, participants might take into account the valence of the emotions when asked to select specific emotions that they deem typical of the ingroup vs. the outgroup. Thus, we predicted that participants would select more negative emotions to characterize the outgroup, whereas positive ones would be reserved for the ingroup.
Furthermore, when instructed to appear nonprejudiced, this should result in the selection of predominantly positive emotions for the outgroup relative to the ingroup, whereas more negative emotions might be chosen to characterize the outgroup vs. the ingroup when instructed to make a prejudiced impression.
To our knowledge, the present research is the first to test the subtlety of the infrahumanization effect more directly, and likewise, it is the first to show the direct link of emotion attribution to explicit measures of anti-Roma prejudice. Importantly, it is among the few works that critically address the role of valence in infrahumanization.
Our results emphasize the following issues: Generally, participants were able to distinguish uniquely human emotions from non-uniquely human emotions when reporting the perceived humanity of the various emotions used in this research. That is, when asked about their perception of primary vs. secondary emotions, participants interpreted primary emotions as the ones that can be experienced by humans and animals, whereas secondary emotions were perceived as reserved for humans only. Even though participants clearly interpreted the emotions in terms of their uniquely vs. non-uniquely human content, they apparently failed to take the human content of emotions into account when under faking instructions. Instead, we found that participants infrahumanized the Roma in all conditions by attributing lesser secondary emotions and more primary emotions to them compared to the German ingroup. We did not predict that participants would attribute a greater amount of primary emotions to Roma than to Germans. However, we interpret this choice of more animalistic emotions as typical of Roma in terms of emotional prejudice as well. Generally, this form of infrahumanization occurred independent of experimental faking instructions. Apparently, participants were unable to turn the faking instructions into action.
Equally important, our results emphasize the important role of valence-based emotion selection as a phenomenon that might occur parallel to infrahumanization. That means, contrary to previous findings by Leyens and colleagues (2000, 2003, 2007), participants apparently do select emotions to describe their ingroup vs. an outgroup based on their valence, rather than relying on the perceived humanity of the emotions only. First, as a form of outgroup derogation, participants described the outgroup using more negative emotion terms than for the ingroup. However, positive emotions were attributed to both groups to the same extent.
In addition, results showed that overall, participants saw both negative primary and secondary emotions as more typical of the outgroup. On the other hand, positive, relative to negative primary emotions were perceived as more typical of the ingroup. With regard to secondary emotion attribution to Germans, valence had no effect. Thus, overall, the choice of characteristic emotions for the ingroup relative to an outgroup did not occur independent of valence.
Moreover, we obtained a significant three-way interaction of Group × Emotion Valence × Instruction: When instructed to appear nonprejudiced, participants attributed more positive emotions to Roma than to Germans. We did not find a difference between both groups in the attribution of negative emotions, though. On the other hand, when instructed to fake bad, participants selected more positive emotions as typical of Germans vs. Roma, whereas they attributed more negative emotions to Roma than to Germans. Thus, emotion valence was used as a cue to turn faking instructions into action.
Turning faking instruction into action was not a problem when participants responded to scales assessing more explicit forms of self-reported outgroup prejudice. As predicted, explicit prejudice was clearly influenced by experimental instructions: That is, when instructed to present themselves favorably, participants reported lower levels of explicit Prejudice than in the Control condition without any additional instruction. However, when participants were instructed explicitly to present themselves as intolerant and prejudiced individuals instead, they reported higher levels of anti-Roma prejudice. Because our findings show that the Prejudice index, which was based on Subtle and Blatant Prejudice subscales, was susceptible to faking instructions in both directions, these results document that the lack of instruction effects on infrahumanization cannot be explained away by participants’ potential general unwillingness to follow experimental instructions.
Clearly, additional research is needed to replicate and extend our findings. This could be done in other intergroup contexts (e.g. Germans vs. Turks). Furthermore, additional measures of infrahumanization and dehumanization, e.g., uniquely human and human nature traits as proposed by Haslam (2006) should be used. Importantly, future research should also include indirect measures of prejudice, for instance, the Semantic Misattribution Procedure (Imhoff, 2010) to gather more support for the claim that the infrahumanization paradigm represents a valid measure of indirect emotional prejudice against social outgroups.
The present findings demonstrate the need to more critically address and test the issue of emotion valence in infrahumanization research. Our data show that infrahumanization is a form of emotional prejudice and that it is subtle, because the distinction between typically human and non-uniquely human emotions seemed not to be obvious to our participants. However, our data show that the effect depends on emotion valence. Whereas participants apparently did not base their emotion selection on the uniquely human content of the emotions, it seemed far easier for them to use the positive or negative connotation of the emotions as a cue and subsequent basis for their choice. Further research is needed to address this issue and certainly, more research should focus on how to counteract and reduce infrahumanization and other forms of dehumanization. However, in line with a quote from Bob Feller, a famous baseball pitcher (see Bargh, 1999, p. 378), reducing infrahumanization might be a difficult endeavour because “You can’t hit what you can’t see”.
Footnotes
Appendix 1
List of primary and secondary emotions administered in this experiment in English language: