Intersected Groups and Discriminatory Everyday Behavior
Evidence From a Lost Email Experiment
Abstract
Abstract. De-racialization research suggests that depicting members of ethnic minority groups as gay leads to less stereotypic perceptions of their ethnic group. However, whether the consequences of de-racialization translate into real-world behavior is unclear. In a large “lost letter” field experiment (N = 6,654) where an email was ostensibly sent to the wrong recipient by mistake, we investigate whether the relative impact of signaling gayness (vs. heterosexuality) differs for Arab (minority) versus Swedish (majority) senders. The results show clear evidence of ethnic discrimination where Arab (minority) senders receive fewer replies (prosocial response) than Swedish (majority) senders. However, there is no evidence indicating that Arab senders would receive a lower penalty for revealing gayness. Implications for multiple categorization research are discussed.
A large bulk of research in economics, psychology, and sociology has studied the presence of discrimination in numerous contexts such as the labor market, the rental housing market, and the healthcare sector (for recent reviews, see Bertrand & Duflo, 2017; Neumark, 2018). An outstanding feature of this research is that it has focused on discrimination based on one group category (i.e., ethnicity) while not accounting for the fact that individuals belong to many groups simultaneously (e.g., an Arab gay man). However, recently it has been shown that stigmatized demographic group categories can interact in a nonadditive manner. Rather, one category perceived negatively in isolation can offset the negative impact of another stigmatized category (Pedulla, 2014; Petsko & Bodenhausen, 2019; Strinić et al., 2021). For example, Petsko and Bodenhausen (2019) have found evidence of a de-racialization effect where gay men are perceived as less stereotypic of their racial group (e.g., Black people) than their heterosexual counterparts, and hence are ascribed more positive stereotypes. How and when the combination of multiple stigmatized social categories translates into positive behavior has been scarcely researched (but see Mazziotta et al., 2015, for a noteworthy exception). To address this research gap, we conducted a field experiment in Sweden. Using a modern version of the lost letter technique (Milgram et al., 1965), we sent out thousands of emails that were ostensibly sent to the wrong recipient by mistake. Returning the email would increase the sender’s chances of finding a job. Of primary importance was whether the number of emails returned to Arab minority versus Swedish majority senders would interact with the sexual orientation of the sender.
Discrimination of Ethnic Minorities and Gay People
A large body of research has established that ethnic minorities often receive unequal treatment in various situations. Correspondence studies have documented unequal treatment in the labor and housing markets (Bertrand & Duflo, 2017). In these studies, fictitious applications are sent to employers or landlords, and the number of positive responses (e.g., invitations to job interviews or apartment showings) are recorded. In their meta-analysis of 43 correspondence studies in different OECD countries, Zschirnt and Ruedin (2016) show that ethnic minority applicants have 49% lower odds of being invited to a job interview than equally qualified majority applicants. In Sweden, Carlsson and Rooth (2007) find that job applicants with a Swedish-sounding name have a 50% higher callback rate to a job interview compared to job applicants with an Arab-sounding name.
Correspondence studies have also investigated discrimination against gay men and lesbian women, although fewer studies exist on this issue. A meta-analysis of 18 correspondence studies in different OECD countries shows that openly gay applicants face discrimination in the labor market, especially in low-skilled jobs (Flage, 2019). There is also evidence that gay men and lesbian women are discriminated against in Sweden, although not in all situations (Ahmed et al., 2008, 2013; Ahmed & Hammarstedt, 2009).
Research on Intersecting Ethnicity and Sexual Orientation
While the results of these correspondence studies constitute compelling evidence of discrimination against ethnic minorities and gay individuals in isolation, they seldom consider how ethnicity and sexual orientation may interact to produce discriminatory behavior (but see Mazziotta et al., 2015, discussed below). Although still being a relatively unexplored field, there are more studies on how perceptions and stereotypes are influenced by group intersections (Kulik et al., 2007; Petsko & Bodenhausen, 2020). Interestingly, this research suggests that ethnicity and sexual orientation could interact such that signaling gayness alters people’s perceptions about ethnic minorities. Consistent with their de-racialization hypothesis, Petsko and Bodenhausen (2019) find that Asian, Black, and Hispanic male individuals are perceived as less stereotypic of their ethnic/racial group if portrayed as gay as opposed to heterosexual. For example, they find that Hispanic and Black men are perceived as having more stereotypically “White” traits when portrayed as gay. These individuals are also stereotyped as more feminine, less masculine, and of higher socioeconomic status. Although the de-racialization effect applies to White targets, too, the effect seems smaller than for non-White targets (Petsko & Bodenhausen, 2019). In another study, Strinić et al. (2021) find that recruiters’ stereotype ratings of Arabs become more positive when this group is described as gay (vs. heterosexual), which is consistent with the de-racialization hypothesis.
Moving beyond stereotypical perceptions, Pedulla (2014) examined in a survey experiment how race combines with sexual orientation to influence hypothetical salary allocations. The results show that participants set higher starting salaries for White heterosexual men than White gay men and Black heterosexual men. More interestingly, however, Black gay men were given higher starting salaries than Black heterosexual men, supporting the notion that one stigma can offset another. Pedulla (2014) also sheds some light on the mechanism behind this result by showing that the gay male stereotype of being effeminate and weak counteracted the Black male stereotype by lowering perceptions of threat and violence.
Studying real-world behavior, Mazziotta et al. (2015) conducted a field experiment to investigate whether revealing a gay sexual orientation would benefit Turkish (vs. German) couples when inquiring about rental offerings in the German labor market. The results did not support this interaction hypothesis. Turkish couples were discriminated regardless of sexual orientation. Unexpectedly, there was no evidence of discrimination against gay couples, possibly reflecting a positive attitude change toward gay people in Germany. As Mazziotta et al. (2015) concluded, more research is needed to examine how the combination of ethnicity and sexual orientation plays out and influence discriminatory behavior in other real-world settings, and toward other ethnic minorities.
Theoretical Framework
The theoretical argument for expecting sexual orientation to moderate the impact of ethnicity on behavior draws on de-racialization processes and the stereotype content model (SCM; Fiske et al., 2002). De-racialization implies that gay ethnic minorities are often de-racialized in the minds of people because they will be perceived as less typical of their constituent racial or ethnic group (Petsko & Bodenhausen, 2019). In this framework, the most prototypic members of social groups are presumed to be white, male, and heterosexual. Because ethnic prototypes presume heterosexuality, being gay is assumed to reduce ethnic prototypicality. As a result, being perceived as gay could displace, and perhaps even erase, attributes that comprise the group prototype, and gay individuals who belong to the ethnic minority group may be perceived as less stereotypical than heterosexual individuals of the same ethnic minority.
According to Petsko and Bodenhausen (2019), the de-racialization process can involve different mechanisms depending on the group in question. Regarding Arabs as an ethnic group, which is the focus of the current research, we theorize that perceived warmth, one of the two fundamental dimensions of social perception according to the SCM (Fiske et al., 2002), could be the driver of the de-racialization process.1 According to the SCM, people perceive social groups and individuals in terms of how warm (e.g., friendly, trustworthy, and kind) and competent (e.g., intelligent, skillful, and efficient) they are. The reason why we expect warmth to be the driver in the current context is that the Arab minority group is perceived as possessing lower warmth but similar levels of competence relative to gay men (e.g., Fiske et al., 2002).2 Thus, signaling gayness is assumed to infuse warmth related stereotype content and cause the Arab category to be perceived as less stereotypic of their ethnic group. Although research suggests that Arabs are perceived to lack competence too (Agerström et al., 2012; Fiske et al., 2002; Lee & Fiske, 2006; Veit et al., 2021), competence should be of less relevance here since gay people do not seem to differ very much from Arabs in perceived competence. As a result, signaling gayness should not introduce competence content that displaces, conflicts with, or erases the Arab stereotype. A recent study (Strinić et al., 2021) shows that for both male and female social targets, it is Arabs’ perceived warmth in particular that increases when a gay male or lesbian female sexual orientation is also revealed.
Naturally, perceptions and treatment of individuals from the majority group could also be affected if an individual deviates from the prototype. Specifically, gay individuals belonging to the ethnic majority group could be perceived as less stereotypical than heterosexual individuals from the ethnic majority group (Petsko & Bodenhausen, 2019). In this case, however, being gay is presumably negative since it conflicts with the positive stereotypes usually associated with the ethnic majority.
The SCM and its predictions about behaviors from intergroup affect and stereotypes (BIAS) map (Cuddy et al., 2007) posit that the type of stereotype content determines the emotional prejudice directed at the group. The specific emotions elicited (pride, pity, contempt, envy) subsequently predict the type of discriminatory behavior directed at the group in terms of active or passive help or harm (Fiske, 2018). Although these models do not seem to have been developed with intersected social categories in mind, one could expect that the increase in perceived warmth gained by Arab minorities who also signal a gay sexual orientation would entail more positive emotions being felt toward this intersected group. One such emotion would be sympathy, which in turn increases active facilitation behaviors. Helping a stranger by informing him/her that an email has been sent to the wrong recipient to ensure that the stranger does not miss a job opportunity (the behavioral outcome studied in the current research) would be a concrete example of an active facilitation behavior.
The Current Research
In the current study, we investigate the effect of group intersections on actual behavior by focusing on Arabs and gay individuals. The rationale for focusing on Arabs and not on other ethnic minority groups is that Arabs constitute one of the most researched and salient ethnic minority groups in Sweden. During the last decades, Sweden has received a significant influx of immigrants from the Middle East and North Africa, and they constitute a large part of the current immigrant population in Sweden (Aldén & Hammarstedt, 2015; Statistics Sweden, 2019). Importantly, previous research has documented clear evidence of discrimination against this ethnic group in the Swedish labor market, with applicants with Middle Eastern-sounding names receiving fewer invitations to job interviews than applicants with native Swedish sounding names (Agerström et al., 2012; Bursell, 2014; Carlsson & Rooth, 2007). Consistent with the stereotype literature (e.g., Fiske et al., 2002), we use “Arab” rather than Middle Eastern people. Moreover, the more general Arab category is presumed to be more cognitively salient than more specific geographical ones (people from the Middle East and North Africa) and largely activated by our ethnicity signal (e.g., Ali, Mohammed).
Our first hypothesis (H1) expects the effect of signaling gayness to be more positive for Arabs than for Swedes. The next two hypotheses (H2, H3) concern the direction of this effect. We predict that the effect will be positive for Arabs and negative for Swedes. Our fourth hypothesis (H4) concerns a standard ethnic discrimination effect, whereby Arabs will receive fewer returned emails than Swedes. This ethnic discrimination effect concerns heterosexual senders. Previous field experiments on ethnic discrimination do not reveal the sexual orientation of the fictitious job applicants, and we assume that the applicants are likely to be construed as heterosexual per default. In addition, the differential impact of signaling gayness for Arabs versus Swedes (see H1, H2, and H3) also implies that ethnic discrimination should be more pronounced for heterosexual senders. Stated formally, the four hypotheses are:
There will be a significant ethnicity by sexual orientation interaction such that the relative impact of signaling a gay sexual orientation will differ for Arab and Swedish senders. Signaling gayness will have a more positive impact on reply rates toward Arab (vs. Swedish) senders.
Signaling a gay sexual orientation will increase reply rates toward Arab senders.
Signaling a gay sexual orientation will decrease reply rates toward Swedish senders.
Arab senders will receive lower reply rates than Swedes when heterosexuality is signaled.
The current study contributes to the scarce research literature on the impact of intersected social categories on actual behavior. Unlike the Mazziotta et al. (2015) study, which examines discriminatory rental offerings among property owners, it examines discriminatory behavior in a different context (job-seeking). The studied behavior (whether to return an email) is relatively effortless with minimal consequences for the participants, who, unlike the property owners in the Mazziotta et al. (2015) study, do not risk accusations of formal discrimination. Another contribution is examining how sexual orientation interacts with hitherto unexplored ethnic groups: Arab minority (vs. Swedish majority) groups. Although exploratory, the current study also sheds light on the role of gender when intersecting ethnicity with sexual orientation.
Method
The hypotheses, materials, procedure, and analyses, including a priori power calculations, were preregistered on the Open Science Framework (https://doi.org/10.17605/OSF.IO/XW59B). As Simmons et al. (2011) recommended, we report how we determined our sample size, all data exclusions (if any), all manipulations, and all measures in the study. The study was approved by the Swedish Ethical Review Authority (no: 2019-02650).
Design
The current research was inspired by the lost-letter technique (Milgram et al., 1965). In the seminal study, Milgram et al. were interested in the general public’s attitudes toward political groups and other institutions. Large quantities of ostensibly lost letters (sealed and stamped) were dispersed in various public places. People that came across a letter had the choice of mailing, disregarding, or destroying the letter. The independent variable consisted of the type of organization (e.g., a communist versus medical organization) appearing in the recipient address area of the envelope, and the dependent variable comprised the number of letters that reached its destination, which, in fact, was Milgram’s own address. As suggested in Milgram et al.’s seminal paper, later studies (Ahmed, 2010; Bushman & Bonacci, 2004) have adapted the technique to examine discriminatory behavior toward different racial/ethnic minority groups. The main benefit of the method is that it provides an unobtrusive measure of everyday behavior with participants not knowing they are participating in a study, which avoids the common problem with social desirability.
We implemented an email version of a lost-letter experiment (Stern & Faber, 1997). Specifically, the participants in our study received an email which had ostensibly been sent to the wrong recipient by mistake. The email consisted of a few sentences and aimed to give the impression that it was a follow-up of a previous conversation about a potential job opening. Thus, the participants were faced with a situation in which they could either reply to the sender or ignore the email. Replying to the “lost email” was presumed to indicate helping behavior since this would make the sender aware of the mistake and enable him/her to send the email to the intended recipient.
Whether the recipient replied or not was the outcome in the experiment, and it was coded using a dummy indicator (0/1). Among those who replied, most recipients explicitly informed the sender that the email appeared to have been sent to the wrong email address, while some replies contained follow-up questions, typically to inquire what the email referred to. In the coding of the outcome variable, we did not distinguish between different types of replies, except for automatic replies, which were treated as a no reply.
All participants received the same email with one important exception. Each recipient was randomly assigned to receive an email from a sender belonging to one of the following four group identities: Swedish heterosexual, Swedish gay, Arab heterosexual, and Arab gay individual. Each group identity was assigned with the probability ¼ to a specific email. The group identity of the sender was clearly revealed by the sender’s name and by the email text itself (see below).
Materials
Sender Names
The name of the sender was used to signal the ethnicity of the sender. Therefore, it was essential that the recipients of the “lost emails” clearly associated the sender names with the intended ethnicity and that the names were clearly visible.
We used six Swedish names that were chosen from the list of most common names in Sweden according to Statistics Sweden’s name register (2020). Common Swedish names are distinct, and this strategy for signaling ethnic majority membership has been used successfully in several previous field experiments on discrimination in the labor and housing markets (e.g., Carlsson & Rooth 2007). We used three female and three male Swedish-sounding names.
Before conducting the field experiment, we conducted a name survey (N = 103) using the Qualtrics platform to confirm that the names used to signal ethnicity would be perceived as intended. We included six Arab minority names (three of each gender), for which we wanted to confirm that the participants could correctly classify them with respect to ethnicity (and gender). Each name that was included in the name survey satisfied two criteria. First, the name was relatively common in Sweden according to Statistics Sweden’s name register. Second, it could be clearly identified as a name of Arab origin according to various internet name sources (e.g., https://www.behindthename.com). To make the classification task challenging, we also included 10 (decoy) names of various origins. This resulted in 16 names that were presented to the participants in blocks of 5–6 names appearing on the same page in the survey. The participants’ task was to classify each name into one of the following categories: the Middle East/North Africa, Africa, Asia, South America, Europe/North America/Australia, and “Do not know.” The participants also classified the gender of the name. The results showed that most participants correctly classified all the Arab names and their respective gender. The full set of results of the pilot study is presented in the Appendix, Table A1.
Letter Content and Group Signals
Each participant received an email whose content gave the impression that it was a follow-up email on a previous conversation. The aim was to show that the email had been sent to the wrong recipient by mistake. This was likely to be the recipients’ interpretation of receiving the email, since no previous conversation had taken place between the sender and the receiver. Specifically, the email was framed as a follow-up concerning a previously discussed job opening (see Appendix (translated from Swedish)).
All participants received the same email text except that each email was randomly assigned different signals concerning ethnic group membership and sexual orientation. Ethnic group was signaled through the name of the sender, which was revealed by the sender’s display name and in conjunction with the email closing.3 Sexual orientation was signaled by mentioning either a boyfriend or a girlfriend in the email. Signaling a gay sexual orientation by referring to a partner of the same sex is a strategy that has successfully been used in previous correspondence studies in the labor market (e.g., Baert, 2014).
To draw the receiver’s attention when receiving the email, and possibly increase the likelihood that it would be read and not viewed as junk email, each email saluted the receiver with his or her first name at the very beginning of the message. Most likely, the receivers interpreted this as if the sender had sent the email to a person with the same name by mistake.
Population and Sampling
The study population consisted of all students at a large Swedish university (N = 53,493), but with a few imposed restrictions. First, since the emails were formulated in Swedish, we included only students that were likely to have a Swedish background and, thus, could understand Swedish well. Restricting the population to these students was accomplished by using available information on the names of the students, which the university had provided to us. We only included names that were common in the list of students, which meant that all foreign names, which are less common, were discarded.4 This strategy also ensured that the personal salutation at the beginning of the email was not too unnatural, even though the email appeared to be sent to the wrong recipient. The reason is that for common names occurring frequently in the university’s email address database, it seems likely that an email could erroneously have been sent to a person with the same name as the intended receiver.
The minimum sample size was determined by power calculations (explained in detail in the preregistered research plan). To make power calculations based on realistic assumptions, we conducted a pilot study (N = 200) with the same design as the main experiment. From the pilot study, we concluded that it would be realistic to expect a reply rate of approximately 20% for the baseline group (Swedish heterosexual senders). Next, we had to make assumptions considering expected differences in reply rates among the groups. We assumed that the Arab heterosexual senders would have 10 percentage points lower reply rate than Swedish heterosexual senders based on observed callback rates in previous field experiments of ethnic discrimination in the Swedish labor market (e.g., Carlsson & Rooth 2007). For Swedish gay senders, we assumed that they would have a lower reply rate than Swedish heterosexual senders based on observed callback rates in previous field experiments of discrimination against gay men and lesbian women in the Swedish labor market (Ahmed et al., 2013). We expected that Swedish gay senders would have a reply rate equal to 15%. Based on the presumed de-racialization effect of being gay for Arab senders, we assumed that the gayness signal reduces the ethnic penalty by 5 percentage points. Thus, in the power calculations, we used a reply rate for Arab gay senders equal to 15%. Given these assumptions, our power calculations show that approximately 3,700 (4,900) emails would have to be sent to obtain 80% (90%) statistical power to detect the hypothesized effects.
We viewed the obtained sample sizes from the power calculations as minimum sample sizes, since the calculations relied on assumptions that contain some uncertainty. In practice, as stated in the preregistration, we had resources to collect data for four weeks, which determined the actual sample size. Between May 14 and June 11, 2020, we sent out 6,654 emails to randomly selected students at the university. We received 1,350 replies, making the overall reply rate 20.3%. Data are available on the Open Science Framework (https://osf.io/rycmb/).
Sending Emails
The study design meant that emails with similar content were repeatedly sent to different email recipients at the university. This raises the question of whether our emails, in some cases, were recognized as spam or blocked by the receiving email server at the university. However, the IT department at the university monitored our emails and confirmed that the emails were delivered to the recipients as intended. In sending the emails, we took several actions that may have contributed to the successful delivery. First, we did not send out all emails at the same point in time but over approximately 1 month. This should have lowered the risk that some of the emails were recognized as spam. Second, we registered four different email addresses for each sender (one university, two Gmail, and one Outlook email address), resulting in 48 email addresses in total. Using this large amount of different email addresses ensured that the number of emails sent from each email address on a given day was low, which should have further reduced the risk that some emails were recognized as spam or blocked. Third, we sent most emails from the registered university email addresses (i.e., within the receiving server’s domain), which should have been less likely to be considered spam.
Results
Main Analysis
We analyzed the effect of group memberships using the linear probability model.5 We used the reply indicator as the dependent variable and dummies for Swedish, Arab, and their interaction as the independent variables. The four hypotheses were examined by conducting tests of the estimated parameters of this regression model.
Figure 1 shows the main results graphically (the results of the formal hypothesis testing are reported in connection with Table 1). The dots in the figure show the mean reply rate for Swedish heterosexual, Swedish gay, Arab heterosexual, and Arab gay stimuli, and the lines show the 95% confidence intervals. A visual inspection of the confidence intervals suggests the absence of ethnicity by sexual orientation interaction whereby the relative impact of signaling gayness would be different for Arab and Swedish senders (H1). Furthermore, signaling gayness does not seem to increase helping behavior toward Arab senders (H2), nor does it decrease helping behavior toward Swedish senders (H3). Finally, in relation to H4, the confidence intervals for the Swedish heterosexual senders are highly separated from those of the Arab heterosexual senders. However, Arab senders seem to be helped less across the board, regardless of sexual orientation.
We now proceed to the formal tests of our hypotheses. Table 1 shows the point estimates and the standard errors of the results from the estimated linear probability models. The first column shows the results with no other covariates than the indicators for Gay, Arab, and Gay × Arab. In this regression, the constant shows the average reply rate for Swedish heterosexual senders, while the coefficients for Gay, Arab, and their interaction show the difference in reply rate compared to Swedish heterosexual senders. This means that the reply rate for Swedish gay senders is the sum of the constant (Swedish heterosexual; .2188) and the parameter for Gay (.0102), the reply rate for Arab heterosexual is the sum of the constant (Swedish heterosexual; .2188), and the parameter for Arab (.0102), and, finally, the reply rate for Arab gay is the sum of the constant (Swedish heterosexual; .2188), the parameter for Gay (.0102), the parameter for Arab (.0102), and the parameter for the interaction effect Arab × Gay (−.0146).
Hypothesis 1 expects that the relative impact of signaling a gay sexual orientation will differ for Arab and Swedish individuals, such that signaling gayness will have a more positive impact on helping behavior toward Arabs. This implies that the parameter of the interaction term Arab × Gay should be positive. However, we find no evidence supporting H1 (p = .457). Hypothesis 2 (signaling gayness will increase helping behavior toward Arabs) implies that the sum of the parameters for Gay and Arab × Gay is greater than zero, but there is no support for this hypothesis either (p = .740). We also find no support for H3, which expects that stating signaling gayness will decrease helping behavior toward Swedes (p = .480). Regarding H4 (that Arabs will be helped less than Swedes when heterosexuality is signaled), the Arab parameter is negative and statistically significant (p = .012). However, Arab senders are also helped less than Swedish senders when gayness is signaled.
In our field experiment, the receivers were randomly assigned to receive an email from one of the four groups of senders using randomization tools built into the statistical software Stata. Following the preregistered research plan, we conduct a test of random assignment (balance check) by also presenting results with covariates for the gender of the sender, the gender of the receiver, and the department of the receiver (which is all the information about the receiver we have access to). These results are presented in column 2 of Table 1. The fact that the results are almost identical to those in column 1 suggests that the random assignment of the group belonging to the sender has worked as intended.
Exploratory Analyses
We have explored the possible gender effects of both the sender and the receiver. For brevity and conciseness, we have included the analyses (and rationale for conducting them) in the Electronic Supplementary Material, ESM 1.
Discussion
We contribute to prior research on group intersections and discriminatory behavior by investigating whether sexual orientation intersects with an Arab (vs. Swedish) ethnicity to impact unobtrusively measured prosocial behavior in a job-seeking context. Our first hypothesis that the effect of signaling gayness would be more positive for Arabs than for Swedes was not supported. The second and third hypotheses that signaling gayness would increase reply rates for Arabs, but decrease reply rates for Swedish senders, were not supported either. Regarding our fourth hypothesis (standard ethnic discrimination effect), we find evidence of ethnic discrimination for our heterosexual senders. We labeled it the “standard” discrimination effect since the fictitious job applicants in previous field experiments on ethnic hiring discrimination are most likely presumed to be heterosexual. Hence, it was assumed that previous field experiments largely reflect ethnic discrimination against heterosexual job applicants. Interestingly, however, we also observe ethnic discrimination when the sender is gay. Although H4 is silent with respect to gay senders, the highly similar amounts of ethnic discrimination affecting gay and heterosexual senders are inconsistent with our intersectional main hypothesis.
Our results mirror those of the Mazziotta et al. (2015) study conducted on the German housing market, which found evidence of ethnic discrimination but not discrimination based on sexual orientation or an interaction effect. The fact that the two studies probed for discrimination in different contexts, intersected sexual orientation with different ethnic minority groups, and employed different manipulations of sexual orientation, strengthens our (and Mazziotta et al.’s) findings. As for similarity, the two studies were conducted in two geographically close European countries (Germany and Sweden) with similar cultures. This could have contributed to the converging findings.
What can explain our findings? Let us first consider the absence of a discrimination effect toward gay individuals. Although the overall picture from previous studies on discrimination against gay people in Sweden is that discrimination exists (Ahmed et al., 2008, 2013; Ahmed & Hammarstedt 2009), it should be noted that the most recent study was conducted in 2013. This might be important since attitudes toward gay people appear to have changed rapidly (in a positive direction) over time in Sweden, and this seems to be an ongoing process (OECD 2019). Furthermore, recent surveys show that Sweden scores substantially higher on attitudes toward sexual minorities than the OECD average, which is much more positive than non-OECD countries. An illustrating example of the attitudes in Sweden is that 77% of Swedish respondents report being comfortable with having a transgender or transsexual person in the highest elected political position, as a co-worker, or as a daughter-or son-in-law. If we add to this picture that young and highly educated individuals, i.e., the population we study, have the most positive attitudes toward sexual minorities, the absence of a negative effect of signaling gayness on helping behavior may not be surprising after all. Returning to the study by Mazziotta et al. (2015), they also argue that their null effect of sexual orientation on rental discrimination can be explained by an attitude change where attitudes toward gay people have become more positive over the years. Attitudes toward gay people may be changing in other European countries as well. Whether such an attitude change is accompanied by less discrimination in different contexts, such as hiring, is an important topic for future research.
Turning to the significant negative main effect on helping behavior toward the Arab group, this is likely to be explained by negative attitudes and stereotypes about this group. This result aligns with a previous Swedish lost letter study (Ahmed, 2010) and field experiments showing discrimination toward Arab job applicants in the Swedish labor market (e.g., Carlsson & Rooth, 2007). Interestingly, annual attitude surveys, such as the attitude barometer, suggest that Swedish people still hold strong negative attitudes toward individuals with a Middle Eastern background (Ahmadi & Palm, 2018). Unlike attitudes toward gay individuals, ethnic attitudes appear to be more resistant to change.
The absence of a gay by ethnicity interaction effect can probably best be understood in the light of the absence of the main effect of sexual orientation and the significant ethnicity main effect. De-racialization is based on the idea that ethnic minorities are de-racialized in the minds of people if they are also gay. This argument relies on the presumption that groups are strongly associated with prototypes. De-racialization occurs when the prototypes of one group – an ethnic minority – clash with the prototypes of another group – a sexual minority. The fact that the population we study, as explained above, is relatively positively inclined toward gay individuals together with the absence of a discrimination effect against gays in our study speaks against the idea of prototypical thinking regarding gay people. If there are no strong prototypes associated with the category gay (e.g., male gays being effeminate and weak), the mechanism (infusion of warmth) operating in a de-racialization process appears less probable.
To situate our results in the broader literature on multiple group membership, one could claim that our data are more in line with the ethnic-prominence hypothesis, which claims that ethnicity is a strong signal that will drive discriminatory behavior more than other group signals (Levin et al., 2002). According to the ethnic-prominence hypothesis, ethnicity may be a more salient category in many contexts, reflecting a numerical minority in a society unlike other categories (e.g., gender). Another reasoning behind ethnicity driving discrimination is that it is often connected to perceptions of conflict and antipathy.
We did also, albeit in an explorative manner, investigate if the gender of the sender and the receiver matter for the results (see ESM 1). The results were very similar for female and male receivers. For senders, we did find some evidence of gender differences. Compared with male senders, we found that for female senders (i) the reply rate for Swedish heterosexuals is higher, (ii) the reply rate for Swedish gays is lower, and (iii) the reply rate for Arab heterosexuals is lower (while there is no indication of a difference for Arab gays). Although these differences are small, they align with Strinić et al.’s (2021) findings showing that Swedish women receive a penalty for signaling gayness in the form of reduced perceived warmth, whereas Arab women’s perceived warmth is boosted by signaling gayness. This different impact of signaling gayness for Arab versus Swedish women might partially explain the observed gender differences pertaining to the sender of the email. Although it is difficult to pinpoint the mechanisms, these results remind us that gender could be an important factor to consider when investigating discrimination against minority groups from an intersectional perspective.
Limitations
We studied a specific population consisting of young and highly educated individuals, which we know have relatively positive attitudes toward minorities, especially gay individuals. We cannot rule out that the results would be different for other populations. In addition, we studied helping behavior in a situation in which minimal effort was required to help and the consequences associated with helping (or not helping) were relatively small. Possibly, unequal treatment could be more pronounced in situations requiring more effort and/or in situations where the consequences of helping (or not helping) are larger. Future research might want to focus on different populations, situations, and potential moderators that might influence the results. For example, lost email studies may be combined with stereotype measures (both implicit and explicit ones) administered to the same participants later (see Agerström & Rooth, 2011). Also, the lost email technique may be employed in organizations whose employees may be more diverse regarding political orientation compared with university student samples that tend to have relatively liberal political attitudes (Zell & Bernstein, 2014). Additionally, although we have taken measures to bypass junk email filters, we cannot confirm that all emails were seen by the participants. This could have introduced noise into our experiment and made it more difficult to discover effects. Relatedly, pertaining to the strength of the sexual orientation manipulation, the receiver had to open the email to be exposed to the signal of sexual orientation. This has probably rendered it particularly difficult to detect a discrimination effect based on sexual orientation. However, we believe that the current sexual orientation signal appearing in the email body as the sender mentions their partner when referring to a previous conversion about a job opportunity is highly realistic, mirroring real-world settings. After all, it would be odd to reveal one’s sexual orientation in other places, such as in the subject field of the email. Increasing the salience of the sender’s sexual orientation would most likely result in lower ecological validity. Achieving strong manipulations of sexual orientation that remain naturalistic is a challenging task for researchers designing field experiments.
Conclusion
The current research contributes to the scant research literature examining the potential positive effects of combining one stigmatized social category with another stigmatized social category in the context of real-world behavior. We find no support for our main hypothesis that Arab senders would be replied to more frequently when signaling gayness compared with native Swedish senders when sending an email to the wrong recipient. These results suggest that de-racialization processes may not have been operating. One explanation could be that the studied population does not associate gay people with clear prototypical attributes. Our results are more consistent with the ethnic-prominence hypothesis that claims that ethnicity is a strong signal that will drive discriminatory behavior more than other group signals.
Future research studying the intersection of ethnicity and sexual orientation using a de-racialization framework should study other populations, preferably those with less positive stereotypes of gay people. It would also be interesting to see if the results would change in situations where helping requires more effort and the consequences associated with helping are larger.
Electronic Supplementary Material
The electronic supplementary material is available with the online version of the article at https://doi.org/10.1027/1864-9335/a000464
1Note that Petsko and Bodenhausen (2019) propose that changes in perceived socioeconomic status is the major mechanism through which de-racialization occurs for Black versus White social targets. Specifically, they propose that gay men are perceived to be higher in SES than heterosexual men. Because Black people are typically associated with lower SES, the increase in SES produced by signaling that one is gay leads Black men to become stereotypically “Whitened” with fewer race-typical traits being assigned to them.
2This applies to the general category of gay men. Numerous subtypes of gay men have been identified that differ in perceived warmth and competence (Clausell & Fiske, 2005).
3The following names were used in the study: 1) Arab males: Amir Hosseini, Ali Mahmoodi, Mohammed Aboud; 2) Arab females: Jamila Hosseini, Fateme Mahmoodi, Laila Aboud; 3) Swedish males: Simon Johansson, Anton Andersson, Alexander Karlsson; 4) Swedish females: Emma Johansson, Julia Andersson, Amanda Karlsson.
4We excluded students whose last name was unique and the first name had a frequency of less than 20 among all students at the university.
5Since the explanatory variables consist of mutually exclusive and exhaustive categories, the fitted probabilities are simply the average outcomes in the cells, and these cannot fall outside the unit interval (Wooldridge, 2002). This is analogous to the simple group mean comparisons and using t-tests. Robust standard errors were estimated.
References
(2012). Warm and competent Hassan = cold and incompetent Eric: A harsh equation of real-life hiring discrimination. Basic and Applied Social Psychology, 34(4), 359–366. https://doi.org/1.1080/01973533.2012.693438
(2011). The role of automatic obesity stereotypes in real hiring discrimination. Journal of Applied Psychology, 96(4), 790–805. https://doi.org/10.1037/a0021594
(2018). Mångfaldsbarometern
[Diversity barometer] . Högskolan i Gävle.(2010). Muslim discrimination: Evidence from two lost‐letter experiments. Journal of Applied Social Psychology, 40(4), 888–898. https://doi.org/1.1111/j.1559-1816.201.00602.x
(2008). Are lesbians discriminated against in the rental housing market? Evidence from a correspondence testing experiment. Journal of Housing Economics, 17(3), 234–238. https://doi.org/1.1016/j.jhe.2008.06.003
(2013). Are gay men and lesbians discriminated against in the hiring process? Southern Economic Journal, 79(3), 565–585. https://doi.org/1.4284/0038-4038-2011.317
(2009). Detecting discrimination against homosexuals: Evidence from a field experiment on the internet. Economica, 76(303), 588–597. https://doi.org/1.1111/j.1468-0335.2008.00692.x
(2015). Boende med konsekvens: En ESO-rapport om etnisk bostadssegregation och arbetsmarknad
[Housing with consequences: An ESO-report on ethnic segregation on the housing and labor market] . Finansdepartementet, Regeringskansliet.(2014). Career lesbians. Getting hired for not having kids? Industrial Relations Journal, 45(6), 543–561. https://doi.org/1.1111/irj.12078
(2017).
Field experiments on discrimination . In A. V. BanerjeeE. DufloEds., Handbook of economic field experiments (Vol. 1, pp. 309–393). Cambridge University Press. https://doi.org/1.1016/bs.hefe.2016.08.004(2014). The multiple burdens of foreign-named men – evidence from a field experiment on gendered ethnic hiring discrimination in Sweden. European Sociological Review, 30(3), 399–409. https://doi.org/1.1093/esr/jcu047
(2004). You’ve got mail: Using e-mail to examine the effect of prejudiced attitudes on discrimination against Arabs. Journal of Experimental Social Psychology, 40(6), 753–759. https://doi.org/1.1016/j.jesp.2004.02.001
(2007). Evidence of ethnic discrimination in the Swedish labor market using experimental data. Labour Economics, 14(4), 716–729. https://doi.org/10.1016/j.labeco.2007.05.001
(2005). When do subgroup parts add up to the stereotypic whole? Mixed stereotype content for gay male subgroups explains overall ratings. Social Cognition, 23(2), 161–181. https://doi.org/10.1521/soco.23.2.161.65626
(2007). The BIAS map: Behaviors from intergroup affect and stereotypes. Journal of Personality and Social Psychology, 92(4), 631–648. https://doi.org/10.1037/0022-3514.92.4.631
(2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology, 82(6), 878–902. https://doi.org/10.1037/0022-3514.82.6.878
(2018). Stereotype content: Warmth and competence endure. Current Directions in Psychological Science, 27(2), 67–73. https://doi.org/10.1177/0963721417738825
(2019). Discrimination against gays and lesbians in hiring decisions: A meta-analysis. International Journal of Manpower, 41(6), 671–691. https://doi.org/10.1108/IJM-08-2018-0239
(2007). The multiple-category problem: Category activation and inhibition in the hiring process. Academy of Management Review, 32(2), 529–548. https://doi.org/1.5465/amr.2007.24351855
(2006). Not an outgroup, not yet an ingroup: Immigrants in the stereotype content model. International Journal of Intercultural Relations, 30(6), 751–768. https://doi.org/10.1016/j.ijintrel.2006.06.005
(2002). Perceived discrimination in the context of multiple group memberships. Psychological Science, 13(6), 557–560. https://doi.org/10.1111/1467-9280.00498
(2015). The effects of multiple stigmas on discrimination in the German housing market. Social Psychology, 46(6), 325–334. https://doi.org/10.1027/1864-9335/a000249
(1965). The lost-letter technique: A tool of social research. Public Opinion Quarterly, 29, 437–438.
(2018). Experimental research on labor market discrimination. Journal of Economic Literature, 56(3), 799–866. https://doi.org/1.1257/jel.20161309
. (2019). Society at a glance 2019: OECD social indicators. OECD Publishing. https://doi.org/10.1787/soc_glance-2019-en
(2014). The positive consequences of negative stereotypes: Race, sexual orientation, and the job application process. Social Psychology Quarterly, 77(1), 75–94. https://doi.org/1.1177/0190272513506229
(2019). Racial stereotyping of gay men: Can a minority sexual orientation erase race? Journal of Experimental Social Psychology, 83, 37–54. https://doi.org/1.1016/j.jesp.2019.03.002
(2020). Multifarious person perception: How social perceivers manage the complexity of intersectional targets. Social and Personality Psychology Compass, 14(2), Article e12518. https://doi.org/1.1111/spc3.12518
(2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366. https://doi.org/10.1177/0956797611417632
. (2019, February 21). Invandring till Sverige. https://www.scb.se/hitta-statistik/sverige-i-siffror/manniskorna-i-sverige/invandring-till-sverige/
. (2020, August 4). Name Search. https://www.scb.se/en/finding-statistics/sverige-i-siffror/namesearch/
(1997). The lost e-mail method: Milgram’s lost-letter technique in the age of the Internet. Behavior Research Methods, Instruments, and Computers, 29(2), 260–263. https://doi.org/1.3758/BF03204823
(2021). Multiple group membership: Warmth and competence perceptions in the workplace. Journal of Business and Psychology, 36, 903–920. https://doi.org/10.1007/s10869-020-09713-4
(2021). The “Big Two” in hiring discrimination: Evidence from a cross-national field experiment. Personality and Social Psychology Bulletin. Advance online publication. https://doi.org/10.1177/0146167220982900
(2002). Econometric analysis of cross section and panel data. MIT Press.
(2014). You may think you’re right … young adults are more liberal than they realize. Social Psychological and Personality Science, 5(3), 326–333. https://doi.org/10.1177/1948550613492825
(2016). Ethnic discrimination in hiring decisions: A meta-analysis of correspondence tests 1990–2015. Journal of Ethnic and Migration Studies, 42(7), 1115–1134. https://doi.org/1.1080/1369183X.2015.113327
Appendix
