Organizational Performance and Discrimination: Are High-Performing Schools Less Likely to Discriminate against Potential Clients?

Abstract Recent studies argue that underperforming public organizations are more prone to stereotyping behavior and discrimination. However, empirical research concerning this subject is limited and focuses only on employment discrimination based on ethnicity. This article is the first to study the relationship between organizational performance and discrimination against clients/costumers. It also takes socioeconomic (SES) discrimination into account. This study is based on a large-scale correspondence experiment in which Swedish school principals were randomly contacted via email by parents interested in enrolling their children; the parents had Arabic- or Swedish-sounding names and were engaged in professions associated with a low or high SES. The results show no clear relationship between client discrimination and student test scores and grades. Hence, the findings of previous studies may not be generalizable to client discrimination and/or other national contexts, and client discrimination may be a result of mechanisms operating in all types of organizations (e.g., discriminatory attitudes).


Introduction
The equal application of laws to all citizens is a vital normative goal of modern public bureaucracies. A growing body of experimental studies has shown that this goal is often not met; public officials discriminate against ethnic minorities all over the Western world (for reviews, see Costa, 2017 andHansen &Tummers, 2020; for two recent studies, see Olsen et al., 2022;Wright et al., 2021). However, the literature also shows important variations in the presence of discrimination across countries and policy areas that are left unexplained. Why are some public organizations and service providers more likely to discriminate than others? Compared to our knowledge on the prevalence of discrimination, much less is known about the contextual factors that can explain differences in discrimination (Gooden, 2015;Guul et al., 2019). This is problematic, given that such knowledge is needed to generate measures that can prevent discrimination.
One organizational factor that has been recently highlighted as possibly being associated with discriminatory behavior is organizational performance. In a recent experimental study, Guul et al. (2019) showed that better-performing Danish schools are less likely to engage in employment discrimination based on ethnicity. This was argued to be a result of managers in underperforming organizations being less open to outgroup applicants and more likely to be under stress or pressure, leading to their unconscious discrimination. This argument aligns with the vast public administration literature concerning organizational performance, which has emphasized the importance of considering past performance as an input for decision-making processes (Nicholson-Crotty et al., 2017;Nielsen, 2014).
Since prior research consists of only one study, more experiments are clearly needed to draw safe conclusions regarding the relationship between organizational performance and discrimination among public organizations. In addition, prior research is too limited in focus as discrimination against clients/costumers and socioeconomic status (SES) discrimination have not been studied. Studying the relationship between organizational performance and client as well as or SES discrimination is inherently important from an equal treatment perspective.
Accordingly, in this article, these relationships are explored in a large field experiment. More specifically, the relationships between school grades/test results and discrimination against parents based on ethnicity and SES in Swedish elementary schools are studied. A total of 3,430 elementary school principals in Sweden were randomly contacted via email by fictional parents interested in placing their children at their schools and who asked them questions about a given school and its admission procedures. The ethnicities (Swedish-and Arabic-sounding names) of the parents and their professions were randomized. The number of emails from the aliases that received responses and the friendliness of the replies were studied. Organizational performance was measured using registry data on test scores and grades. More specifically, I captured the proportion of students in schools who passed the national tests in Swedish and mathematics and the average grades of schools containing ninth graders. Unlike the existing Danish study, which is based only on 71 schools, this study covers almost all principals and a large majority of the schools in an entire country, Sweden, increasing the potential generalizability of the results.
The present study focuses on elementary school principals because schools are important public services that frequently engage with the general public, and the decisions made by principals have potentially large consequences for their clients. Education is arguably the most crucial means of increasing socioeconomic mobility in a society and facilitating the integration of immigrants. In many countries, school systems employ a large group of public employees (Meier et al., 1999). In Sweden, for example, up to 8% (432,000) of all employed individuals work in the education sector (Statistics Sweden, 2021).
The results of this article demonstrate that parents with Arabic-sounding names and those with low-SES professions receive less-friendly emails from Swedish elementary school principals than parents with Swedish-sounding names and high-SES professions. However, no clear relationship between average grades and test scores in Swedish schools and discrimination against clients based on either ethnicity or SES is found. Thus, the results suggest that discrimination may be a result of mechanisms that generate discrimination regardless of the performance level of a school (e.g., discriminatory attitudes or cream skimming as a result of the widespread use of performance metrics; more information concerning this issue is provided in the following section). Accordingly, the present study suggests that the results in the Danish study investigating the relationship between organizational performance and employee discrimination may not be generalizable to client discrimination and/or other national contexts, such as Sweden. Moreover, the Danish findings may possibly not be generalizable to all schools in Denmark.

Previous research and theory
Client discrimination among public officials While field experiments in political science are still rather rare, in recent decades, a growing number of such studies have investigated the presence of discrimination against clients among public officials, such as election officials, welfare officers and school principals. Field experiments have become the preferred method in this literature, as they are less likely to generate unrepresentative samples and social desirability bias and allow a more objective assessment of the extent to which discrimination takes place (see, e.g., Pager & Shepherd, 2008;Quillian, 2006). Most studies are correspondence experiments in which emails from fictional persons of varying ethnicities are sent to public officials.
The results from these correspondence experiments show that public officials discriminate against ethnic minorities all over the Western world, though to varying degrees (for a metastudy, see Costa, 2017).
Ethnic discrimination is defined here as unequal treatment based on physical characteristics (e.g., skin color or hair type) or cultural factors (e.g., name or language). In Europe, ethnic discrimination has been primarily found in the quality and friendliness of the officials' responses. In a recent experiment, Olsen et al. (2022) found that parents with Muslim names experienced lower acceptance rates and received more questions when asking school administrators whether it was possible to transfer their child to a school. Hemker and Rink (2017) conducted a similar experiment on German welfare offices and found that clients with Turkish names received significantly lower-quality responses than those of other clients. Similarly, Adman and Jansson (2017) found that clients with Arabic-sounding names received less-friendly responses in their interactions with local Swedish public officials. Research in the US has found even clearer signs of ethnic discrimination, even in terms of response rates (see, e.g., Einstein & Glick, 2017  Clearly, ethnic discrimination against clients occurs among public officials in Western democracies. Thus, a natural next step for the literature is to examine the causes of discrimination and why some officials are more likely to discriminate than others. However, in contrast to our knowledge on the prevalence of discrimination, much less is known about the contextual factors that can explain differences in discrimination (Gooden, 2015;Guul et al., 2019). However, to develop measures and reorganize bureaucracies to minimize discrimination, more research is needed in this regard.

Organizational performance and (employee) discrimination
Recently, organizational performance has been identified as an organizational factor possibly associated with discriminatory behavior. In their study of 71 Danish schools, Guul et al. (2019) found that applicants for teacher positions with Middle Eastern-sounding names received higher callback rates (than those with Danish names) from better-performing schools than from underperforming schools. Two causal mechanisms underlying these results are presented.
First, underperforming organizations are arguably less open to new inputs (such as outgroup applicants) by relying more on well-established routines and procedures. This mechanism is supported by studies showing that managers in underperforming organizations are more likely to centralize decisions and rely on proven types of personnel (McDonald & Westphal, 2003;Meier et al., 2015) and to seek advice from ingroup members who confirm current behaviors (McDonald & Westphal, 2003). Second, managers in underperforming organizations are more likely to be under stress or pressure, as their working life may be characterized by a gap between the demands of their jobs from clients/senior managers and the resources/support they have available. This in turn makes them more likely to (unconsciously) attach stereotypical traits or behaviors to ethnic minority applicants. This mechanism is supported by research in social psychology on cognitive load. When under time constraints, individuals tend to unconsciously use mental shortcuts, such as stereotypes, to free up much-needed cognitive resources, whereby they are less likely to notice the effects of stereotypes and prejudices on their behavior. Consistently, laboratory experiments have found that humans may be more likely to be unconsciously influenced by stereotypes and prejudices in situations that are very demanding of cognitive resources (Bodenhausen, 1990;Gilbert & Hixon, 1991;Johnson et al., 2016;Ma et al., 2013).
Although the theoretical mechanisms and results presented by Guul et al. (2019) are convincing, more experiments are clearly needed to draw safe conclusions regarding the relationship between organizational performance and discrimination. In addition, not only is prior research almost nonexistent, but it is also too limited in focus as the relationship between organizational performance and discrimination against clients and SES discrimination have not been studied. Here, client discrimination refers to all types of discrimination against clients/costumers based on demographic and non-demographic characteristics, such as ethnicity and SES. Ethnic discrimination is defined as unequal treatment based on physical characteristics (e.g., skin/hair) or cultural factors (e.g., name or language). SES discrimination is defined as unequal treatment based on an individual's education, occupation and/or income. The occurrence of client discrimination based on ethnicity and SES conflicts with fundamental democratic principles, such as equal treatment under the law. Accordingly, this article contributes to the existing literature by complementing the study by Guul et al. (2019) with a larger study investigating the relationship between organizational performance and both SES and ethnic discrimination against clients.
Why should organizational performance affect client discrimination?
Why should we expect to also find a relationship between organizational performance and client discrimination? First, the stress-and pressure-based mechanism highlighted by Guul et al. (2019) should also apply to client groups against which managers may have prejudices, such as immigrant and low-SES clients. Experiments in social psychology have shown that the relationship between discrimination and stress/cognitive load is not limited to recruitment processes but applies to many different social situations, including interactions between public officials (e.g., police officers/doctors) and their clients (Bodenhausen, 1990;Gilbert & Hixon, 1991;Johnson et al., 2016;Ma et al., 2013). This hypothesis is consistent with Andersen and Guul's (2019) recent study, which demonstrated that Danish teachers were less likely to discriminate against immigrant students when they received extra resources.
In addition, it is possible to identify mechanisms based on cream skimming and rationing. Cream skimming is when officials selectively focus their attention, resources and efforts on certain clients. Rationing is when officials try to limit service availability, attractiveness or expectations for certain client groups (Lipsky, 2010) by, e.g., selectively increasing administrative burdens (cf. Jilke et al., 2018). According to Lipsky, both coping strategies are routinely used by public officials facing insufficient personnel levels and budgetary resources to maximize cost efficiency while increasing their odds of achieving policy goals (Lipsky, 2010). Therefore, attracting self-propelled clients while avoiding more demanding clients should be the most efficient and commonly used method for managers in underperforming organizations to increase the odds of achieving their (measurable) performance goals while decreasing their high workloads. For example, as a result of statistical discrimination, immigrants and students whose parents lack higher education (low SES) may be associated with perceived costs related to language skills and potentially negative grade trajectories. Thus, principals in underperforming schools have strong incentives to attract students with highly educated parents who are not ethnic minorities; such students are likely viewed as less resource-demanding for schools and could automatically help schools raise their average grades through changes in the student composition. Similarly, underperforming recruitment agencies have strong incentives to attract highly educated individuals who belong to the majority population in terms of ethnicity because it may be easier to find jobs for them, thereby improving agency results.

Why would organizational performance not affect client discrimination?
On the other hand, there are several reasons why we should not expect to find a strong relationship between organizational performance and client as well as employment discrimination. First, if discrimination is primarily caused by unconscious or conscious discriminatory attitudes that are not largely affected/activated by the organizational context in which officials operate, then we should not expect to find a relationship between organizational performance and any form of discrimination. For example, discrimination may largely result from a relatively deeply rooted distaste toward outgroup members (taste-based discrimination; Becker, 2010). It may also be a result of widespread prejudices and stereotypes, by which officials (similar to a general population) automatically associate ethnicity and professions with certain negative traits (statistical discrimination; Phelps, 1972), which affects their behavior regardless of the organizational context. Therefore, the variations in discrimination between public service providers, such as schools, should simply reflect the variations in discriminatory attitudes between their employees.
Second, the mechanisms based on stress, cream skimming and rationing described above may operate in all public organizations, regardless of their performance metrics. New public management (NPM) reforms, with their emphasis on competition and measurable performance targets, have changed the conditions for most providers of public services (cf. Bevan & Hood, 2006;Bohte & Meier, 2000;Kroll, 2015). As a result of such reforms, even managers in high-performing organizations may feel pressured to decrease costs and focus on measured tasks, leading to client discrimination. Furthermore, managers in under-performing organizations with free slots may potentially have stronger incentives to not discriminate because they cannot afford to reject any potential clients.
Finally, if the findings regarding employee discrimination in the Danish study were primarily a result of an organizational culture in which underperforming organizations are less open to new inputs, such as outgroup applicants, then these findings may not be generalizable to client discrimination. Recruiting staff entails a much greater risk for management and involves much closer contacts with potential staff members than the contacts needed to attract clients. This could reinforce the effects of potential ingroup/outgroup biases. Furthermore, even underperforming conservative organizations may have strong incentives to try to attract any new clients, regardless of their ethnicity.

Institutional context: Swedish elementary schools
This article studies the relationship between organizational performance and discrimination against clients through a large field experiment of Swedish elementary school principals. Sweden is a particularly interesting case in the context of discrimination research. It has a strong reputation for being an immigration-friendly welfare state and has an increasingly ethnically, linguistically and culturally diverse population. Following decades of largescale immigration, today, approximately 19% of the population is foreign-born (Statistics Sweden, 2021). Two percent of the total Swedish population was born in Syria. Sweden has often been described as a European outlier, with public opinion showing particularly positive views of ethnic minorities compared with other countries (World Value Survey, 2014). Accordingly, we may be less likely to find discrimination in Sweden than in other countries. However, more recent studies suggest that Sweden is a fairly representative case of Europe today. Similar to large European countries, such as France, the UK and the Netherlands, a large share of the nativeborn population now views immigrants from the Middle East negatively and considers refugees a threat to security and social welfare (Abdelkader, 2017). Furthermore, experiments have shown that Middle Eastern immigrants are often subject to discrimination in many Swedish societal arenas, including the labor and housing markets, as well as in their contacts with public officials, similar to other European countries (Adman & Jansson, 2017;Ahmed & Hammarstedt, 2008;Bursell, 2014).
The principals contacted in the experiment work at publicly financed public and private schools that operate at the local (municipal) level. Municipalities (local governments) in Sweden have the administrative responsibility of organizing and financing the schools located within their territories, and all municipalities have schools covering years 1-12 (years 1-9 are mandatory). Parents freely choose between all schools in their municipality, which include public schools managed by the municipalities, private schools run by for-profit businesses, and nonprofit private schools run by associations and foundations. All schools are obliged to follow the same laws and adhere to the Swedish curriculum for compulsory schools. As a result of a voucher system, no students have to pay for their education. If a private school is oversubscribed, it can choose students based on their proximity to the school, waiting lists (by the date of application), and/or priority for children whose older siblings are already enrolled in the school. The same criteria apply to public schools, but students are always guaranteed a slot in the public school nearest to their home. Schools are not allowed to discriminate against students in Sweden and most other countries, but frictions in the choice process may allow schools to influence who applies. For example, schools may influence the applicant pool by sending informal messages and information to potential clients. Therefore, the focus is on social interactions between principals and parents interested in placing their child at the school.

Methods
In line with previous studies on discrimination among public officials, this study is based on a correspondence experiment. Each elementary school principal governing a unique school 1 in Sweden was contacted via email by a randomly assigned fictional parent (N ¼ 3,430). The emails were sent in several waves between January 11 and 14, 2020. 2 The email addresses of the school principals were randomly divided into eight parent alias groups that corresponded to the aliases used in the emails (see Table 1). Each parent alias contained a specific combination of traits in terms of ethnicity (Arab-/Swedish-sounding name), SES (dentist/care assistant) and gender (man/woman). An advantage of factorial designs such as the one used here is their efficiency for use in experimental subjects. That is, for a given number of treatments, factorial designs require fewer experimental subjects than alternative experimental designs to maintain a similar level of statistical power (cf. Collins et al., 2009). As all treatment conditions are randomly distributed in the sample in a balanced way, it is possible to estimate the individual independent effects of each treatment in a single model with little loss of efficiency.

Treatments
As ethnic treatments, Swedish-and Arabic-sounding names were chosen. As mentioned above, immigrants from the Middle East constitute a large ethnic minority in Sweden that is often discriminated against. For the Arabic parents, the male name Mahmoud and the female name Fatimah were chosen, both with the surname Hassan. As shown in Table 9 in Supplementary Appendix D, both names are rather common in Sweden. The chosen names were intended to clearly signal ethnicity, as Arabic names are rarely used by ethnic Swedes, and to my knowledge couples from Arabic countries rarely choose traditionally Swedish names (or surnames) for their children.
For the Swedish parents, the names Kevin and Melissa with the surname Andersson were chosen because they were associated with SES levels similar to the Arabic names according to a pretest that was conducted among upper secondary school principals (see Supplementary Appendix D, Table 8). Names usually carry not only an ethnic association but also a certain SES association, and names belonging to ethnic minorities are often associated with a low SES (Carnes & Holbein, 2019;Elchardus & Siongers, 2011). Therefore, to prevent the possibility that the results would be driven by SES associations with the chosen names, it was important that all names, regardless of ethnic marker, were associated with the same SES levels. All four names are currently commonly used in Sweden, and the mean age of the names varies between 15 (Melissa) and 35 (Mahmoud). To facilitate an analysis of the relationship between organizational performance and SES discrimination, half of the emails were randomly sent from parents with a highly skilled profession (dentist) and the other half from parents with a low-skill profession (care assistant). 3 According to the emails sent, both parents shared the same profession. The chosen professions were intended to clearly signal SES. In contrast to care assistants, who require only a high school education, dentists require an extensive university education and a license to practice. The average wage of dentists (47,400 SEK/month) is almost twice that of care assistants (24,800 SEK; Statistics Sweden, 2020). These professions are common among immigrants and native-born individuals, but men are underrepresented among care assistants (only 10%; Socialstyrelsen, 2017).

Content of the emails and coding scheme
The letter sent to the principals is shown in Figure 1. The emails were written as if they were sent by someone who was considering moving to the municipality to prevent the somewhat unusual names from arousing suspicion among principals in small municipalities. Regarding the questions asked, it was considered important to not use overly standard or easily answered questions, but due to ethical considerations, presumably timeconsuming questions were avoided. These considerations were balanced by including the following three relatively simple but important questions in the emails: Does your school have a special profile? How do you register? Are there open slots available at your school?
The coding scheme used to analyze the responses from the principals contained a large number of variables (see the pre-analysis plan), whereas Where we move depends on how the job opportunities are (my husband and I are both dentists/nursing assistants), but it is of course also important that we find a good school for our son.
Thank you for taking the time to reply (preferably via email if possible).
Sincerely, Name Surname the three most important variables were used in this article (see Table 2). Two variables measured formal aspects, i.e., whether any reply was received within two weeks and how many of the three questions were answered. Replies were registered from anyone working at the school or the municipality of the school, except for autoreplies and noninformative emails from principals who left their position, which were excluded from the sample. The third variable was an index that measured the friendliness of the emails (0-5). 4 More specifically, it captures whether the replier invites future contact (1p), welcomes the sender to the municipality/school (1p), invites the sender to visit the school (1p), provides additional information not directly related to the three questions (1p) and/or uses the name of the sender (1p).
To illustrate the coding procedures for the three variables, two complete email answers are presented and coded in Supplementary Appendix A. Nonresponses were coded as zero for all variables to avoid conditioning on a posttreatment variable and introducing selection bias (cf. Coppock 2019). Two research assistants independently coded the emails following instructions and a coding form. After coding all responses independently, the assistants were instructed to pay special attention to emails in which their coding differed to reconcile the final coding. For the simplest variables (reply and the variables forming the friendliness index), a third assistant reviewed differing cases. The names of the fictitious emailers were removed before coding started.
Of the original 3,430 emails sent, 3,394 units of analysis were ultimately included in the dataset. Thirty-one emails bounced back and were excluded from the analysis. Five emails were also excluded because the principal replied that she or he left the position, and the email was not forwarded further. In the few cases in which several responses were received from a principal or a school, the overall "best" response was chosen, i.e., the response that scored the highest on the three variables combined. No signs of spillover or disclosure of the experiments were found in the email responses.

Measuring organizational performance
As argued by previous studies, academic achievement is one of the most important measures of organizational performance among schools (Andersen & Serritzlew, 2006;Guul et al., 2019;Nielsen, 2014). Therefore, organizational performance in this article is measured through test scores and grades. As Table 3 and Table 10 in Supplementary Appendix F demonstrate, no statistically significant differences were found in grades and test scores between the treatment groups in terms of SES and ethnicity, which suggests that the randomization procedure was successful. The test scores and grades were collected from administrative registries and are listed according to the year before the collection of experimental data (2019).
The main measure for organizational performance here is labeled as 'test scores' and captures the proportion of students who have passed the national tests in Swedish and mathematics in grade 3, 6 or 9 (0-1). The tests completed by the youngest cohort in the schools are used. For example, for schools covering grades 6 to 9, national tests from grade 6 are used. The results remain the same in terms of statistical significance after controlling for the cohorts attending the school (see Supplementary Appendix G, Table  13) and when comparing the results of elementary schools and secondary schools (see Supplementary Appendix K, Tables 24 and 25). The test-score data are available for almost all the schools in the dataset (2,996/ 3,394 ¼ 88.27% of the schools). Thus, the test-score sample appears to be very representative of the full experimental dataset comparing school and municipality characteristics (see Table 11 in Supplementary Appendix F). This is arguably my best measure of organizational performance as the tests are standardized and are designed at the national level, and teachers are provided guidelines for grading the tests, which could reduce grade inflation. Finally, I include an additional measure labeled 'grade scores' that captures the average grades in all subjects for the students graduating from elementary school (grade 9). It is bounded between 0 and 340 points, where the latter corresponds to the highest possible grade in all subjects. Grade data are available for all middle schools (grades 7-9, 1,109/3,394 ¼ 32.67% of the schools). Grades 7-9 in Sweden are comparable to lower secondary education in the United States. In contrast to the test score sample, the grade sample is not completely representative of the full experimental dataset; there is an overrepresentation of private schools, large schools and municipalities in the sample (see Table 11 in Supplementary Appendix F). Therefore, the results are also presented, after controlling for these and additional characteristics, in Supplementary Appendix H. The results are similar when these controls are not included.

Limitations
While it is arguably difficult to imagine a better feasible research design to capture the relationship between organizational performance and discrimination, 5 a clear limitation of the study is that the organizational performance measures are not experimentally manipulated. If there are variables not included in the statistical models that correlate with both the dependent variables and the independent variables (omitted variable bias), the models will not capture the causal effect of organizational performance on discrimination. The robustness tests shown in Supplementary Appendices G and H, Tables 12-20 illustrate that the results hold after controlling for a large number of potentially important covariates and potential confounders: student composition, number of students, the ethnic composition of the school management, school ownership, age cohorts and so on. The models that control for student composition are particularly important, as they do not suggest that the study suffers from reverse causality (cf. Guul et al., 2019). 6 However, since it is not possible to completely exclude the possibility that the results are affected by omitted variable bias, the results can only capture whether there is a relationship/correlation between organizational performance and discrimination.

Research ethics
To reduce social desirability bias and capture real-life discrimination, the research subjects were not aware that they were a part of a field experiment. Discrimination is potentially sensitive, and the results would not have been accurate if the elementary school principals were informed and asked to participate beforehand. However, I strived to decrease the potential negative effects of the experiment as much as possible. First, the principals were anonymized, and only aggregate-level tendencies are shown rather than specific answers. The emails and the discrimination effects are presented in a way that prevents the identification of the municipalities from which they were sent. Second, I minimized the time the principals spent on emails by keeping the questions simple. Some principals asked questions in their replies, but these questions were not answered to prevent them from working additional hours. To the best of my knowledge, no similar experiments involved the research subjects, and no future experiments by the author will involve them (without seeking consent) such that future encounters between principals and citizens will not be affected and to avoid using more of their working time. The data of how much time the principals spent replying to our emails in the experiment are provided in Supplementary Appendix B. The design was approved by the Swedish ethical review board.

Results
The presence of client discrimination in swedish schools Examination of the general presence of discrimination in the different samples (see Table 4) revealed no negative discrimination effects on the response rate (Models 1-3). The same is true to some degree for questions answered; however, a negative statistically significant discrimination effect on Arabic aliases among middle schools (model 6) was identified. In regard to the friendliness index, we observe clear discrimination effects on our Arabic and low-SES aliases (models 7-9) in all of the samples. The responses to these aliases were rated as less friendly on average (a decrease of approximately 10-24% in friendliness points from the baseline of 1.376 to 1.466 friendliness points). 7 These results align with previous experiments on public officials in Europe, which have found discrimination mainly in terms of the quality and friendliness of officials' responses and not in terms of their response rate.

Correlations between organizational performance and client discrimination
The main results regarding the relationship between organizational performance and client discrimination can be found in Tables 5 and 6. 8  Table 5 shows whether principals in higher performing schools are less likely to discriminate against parents by using an interaction variable between the proportion of students (0-1) that passed the national tests and the Arab and low-SES treatments. The coefficients are generally small and negative, and none are statistically significant below the 10% level, indicating that there is no strong relationship between school performance and discrimination. The results are similar after controlling for variables that could potentially affect the test scores and discrimination, such as student body compositions (Supplementary Appendix G, Tables 14 and 15), the composition as well as the size of the school management/the teaching team (Supplementary Appendix G,  Table 16), private/for-profit schools and the cohorts at the schools (Supplementary Appendix G, Table 13), and when using a logistic regression (Supplementary Appendix G, Table 12). Hence, the robustness analyses show similar results. An alternative measure of organizational performance is found in Table 6. It shows whether principals in middle schools with higher average grades are less likely to discriminate based on ethnicity and SES. Once again, we find small interaction coefficients that are not statistically significant at the 10% level, and the results are similar after controlling for a large number of school level variables that could potentially affect the test scores and discrimination (see Tables 18-20 in Supplementary Appendix H) and when using a logistic regression (Supplementary Appendix H, Table 17). No statistically significant two-way interaction effects were found (see Supplementary Appendix I: Table 21: Arab X Low-SES X test scores, Table 22: Arab X Woman X test scores).

Discussion and conclusion
Recently, it has been argued that a relationship exists between organizational performance and discriminatory behavior among public officials. However, research on this issue is almost nonexistent, consisting of only one study concerning Danish schools . Furthermore, this study focused only on employee discrimination based on ethnicity and did not study the relationship between organizational performance and discrimination against clients/costumers or SES discrimination. Accordingly, this study filled this knowledge gap through a large field experiment on Swedish elementary schools. In line with previous experiments on discrimination among public officials, the results show clear signs of differential treatments of clients. That is, parents with Arabic-sounding names and low-SES professions received less-friendly emails from Swedish elementary school principals than parents with Swedish-sounding names and high-SES professions. However, surprisingly, and in contrast with the previous Danish study, the results do not indicate that higher-performing schools in terms of grades and test scores are less likely to discriminate based on either ethnicity or SES. Why do the results differ from the Danish study? One possibility is that the disparities are simply a result of differences in sample sizes. While my study covered almost all school principals and a large majority of the elementary schools in Sweden (2,996/4,822 ¼ 62% of all schools), the Danish study only analyzed 71 schools (71/2,237 ¼ 3.2%; data from 2015; Statista 2021). Hence, it is possible that the results from the Danish study are not generalizable to all schools in Denmark. Significant effects in small-n studies may be larger by chance (Loken & Gelman, 2017).
Another possibility is that organizational performance has stronger effects on employee discrimination than on client discrimination, and the latter may be affected by other mechanisms. For example, the effects on employee discrimination may be a result of an organizational culture in which underperforming organizations are less open to new inputs (such as outgroup applicants). Here, recruiting staff entails a much greater risk for management and involves much closer contacts with potential staff members than the contacts needed to attract clients, thereby amplifying ingroup/outgroup biases. Hence, this mechanism may only be relevant for employee discrimination and not for client discrimination.
Finally, it is possible that the differences are a result of disparities between Sweden and Denmark. Even though Sweden has more in common with Denmark than most other countries, there are clear differences in their school systems and existing discriminatory attitudes that could potentially affect the relationship between organizational performance and discrimination. Survey and registry data suggest that neither of the stress and ingroup/outgroup mechanisms highlighted in the Danish study may be active in the Swedish case, explaining the null findings. No strong correlations were found between organizational performance and stress/time resources as captured by a survey of the principals included in the field experiment. In the case of intergroup contacts, a negative correlation was found between the proportion of teachers/school managers born in the middle east and student performance, which contradicts the proposed ingroup/outgroup mechanisms (see Table 23 in Supplementary Appendix J). In addition, there may exist differences in the prevalence of discriminatory attitudes between the two countries, potentially making it easier to find discrimination effects in the Danish case. In the 2017 World Value Survey, 9.5% of Danes identified immigrants as a group that they would not like to have as neighbors, in contrast to only 3% of Swedes (World Value Survey 2020).
Given the size of the present study and the lack of any clear signs of an overrepresentation of discrimination among low-or high-performing schools, the most reasonable interpretation is that a strong relationship between organizational performance and discrimination may possibly not exist, at least in email interactions between school officials and clients in Northern Europe. One possible implication of this is that discrimination may be caused by relatively rigid conscious or unconscious discriminatory attitudes that are not typically generated/activated by the organizational contexts in which public officials operate. Hence, measures to prevent discriminatory behaviors should focus on discriminatory attitudes rather than organizational changes at workplaces. However, to draw safe conclusions regarding these relationships and create appropriate policy proposals, more large-scale discrimination experiments in various contexts are clearly needed. In particular, we need more large-N studies in which organizational performance is experimentally manipulated or measured over time to increase the possibilities of capturing causal effects on discrimination.
Notes 1. Therefore, no schools are shared between principals, and the study is very close to being representative of all elementary school principals in Sweden. For more information regarding how our list of principal email addresses was generated, see Supplementary Appendix C. 2. The study followed the preregistered pre-analysis plan (https://osf.io/3bju7?view_only=3ed1f21c95844935b379396f8e77588f). This pre-analysis plan covered a large project that included numerous hypotheses for testing in several articles. It is logical to publish several articles based on the data given the large number of principal working hours spent on answering our emails (cf. Supplementary Appendix B, Table 7). The articles share some dependent variables and experimental data but differ in the use of registry/survey data (independent variables) and theories/ hypotheses being tested. I have published one article testing H2 and H5 (Taghizadeh, 2022a), one article testing H9 (Taghizadeh, 2021) and one article testing H6 (Taghizadeh, 2022b). When I find statistically significant effects, the critical p-values are adjusted for multiple hypothesis testing. The hypotheses tested in this article do not exist in the pre-analysis plan. However, the article closely follows the pre-analysis plan in terms of the design of the experiment and the dependent and independent variables used, which are all listed in the plan. 3. The alternative would be to instead capture SES by varying the formality of the language used. The advantage of using treatments based on professions is that the principals are provided clear information regarding the SES of the senders and do not need to base their SES evaluations on guesses regarding the language used. In line with this, previous research has found much stronger treatment effects of information regarding professions compared to treatments based on the formality of the language used in the letters (cf. Adman & Taghizadeh, 2020;Hemker & Rink, 2017). 4. A principal component analysis supported treating the items as one dimension. Using a principal component analysis with varimax rotation, the eigenvalue of the only dimension that survived the Kaiser criterion was 1.8. This dimension explained 36 percent of the total variance. The factor loadings varied between 0.38 and 0.52. 5. Ideally, we would like to assign schools to different levels of performance and subsequently monitor whether they engage in client discrimination. However, conducting such an experiment is not possible in the real world. The alternative would be to conduct a survey experiment and present a hypothetical situation to the principals where they are assigned different levels of performance. However, the disadvantage of this kind of survey experiment is that the hypothetical situation may not be generalizable to real-life scenarios, where the principals may suffer from stress and may have real material incentives to cream-skim. Furthermore, surveys may suffer from unrepresentative samples and social desirability as topics such as discrimination are inherently sensitive (the same is true for interviews). 6. The arguably most important causal pathway through which discrimination should affect school performance is by changing the student composition on a school. In other words, schools that often discriminate against Arab students and students with a low socioeconomic status should have fewer such students in their schools, which in turn could raise their average grades and test results. However, the results remain largely the same even when controlling for student composition (see Tables 14-15 and  Table 19). 7. In addition, we observe statistically significant interaction effects between SES and ethnicity and between ethnicity and gender, suggesting that SES discrimination (Table  21, models 1, 3, and 5) and gender-based discrimination (Table 22, models 1, 3, and 5) could be more prevalent against ethnic minority aliases. 8. The statistical power levels for the most important models (where we detected statistically significant discrimination effects in Table 4) were above 90 (power !0.90: models 5-6 in Table 5 and model 6 in Table 6, power !0.70: model 6 in Table 6. The power levels were calculated using the InteractionPowerR package in R).

Ethics statement
The design was approved by the Swedish ethical review board (see 2017/234 and 2018/371).