Statistical or Taste-Based Discrimination at the Gates of Admission to 5th Grade High School in Germany – Evidence from the Field (Arabic Names)

With regard to the study design, we sent an application via email to all high schools using aliases as parents of prospective 4th-grade graduates—randomized by immigrant background, parental gender, academic performance (with or without a Gymnasium recommendation), and the child’s gender (son or daughter), resulting in a 2×2×2×2 intervention design.

Studies on discrimination against young adults and minors have gained increasing attention in economics (see Beart et al.). Esther Duflo and others have contributed to this field by developing methods to identify discriminatory behavior through randomized controlled trials in naturalistic settings. While evidence of labor market discrimination is abundant, discrimination against minors—especially within school environments—remains less well documented. Teachers play a central role in students’ educational trajectories (Hattie), yet no existing randomized controlled trial has examined the causal effects of student characteristics such as gender or migration background on teachers’ behavior within their natural institutional setting.
To address this gap, we conducted a field experiment testing how admission offices in German secondary schools respond to information requests from parents. Admission decisions are typically made by senior-level teachers, often supported by administrative staff or other teachers. Qualitative evidence indicates that these teachers ultimately compile the definitive list of accepted students who will begin the pathway toward the German Abitur.
The aim of this study is to provide empirical evidence on whether gender, ethnic background, and academic performance influence high school admission in a natural environment. Building on a prior experiment that used German- and Turkish-sounding names—where we observed notable evidence of ethnic discrimination—we now extend the design by using Arabic-sounding names. Given that Arabic names may be perceived as culturally more distant, we test whether discriminatory behavior is at least as strong as, or potentially more pronounced than, the discrimination documented in the Turkish-name experiment.
Using an administrative database of primary contact addresses for admission offices in high schools across the German federal states, we sent emails requesting information, posing as parents of German or Arabic background. As before, we experimentally varied key characteristics, including the child’s gender and whether the child held a high-school recommendation (Gymnasium recommendation) or an equivalent GPA threshold in the junior high school certificate.
The German state school system provides a suitable institutional context to identify potential teacher biases. First, the system allows students to apply for Gymnasium even without an official recommendation, provided their GPA exceeds a threshold comparable to a grade B. This pathway is frequently cited as a justification for the stratified three-tier secondary school system criticized in various OECD and PISA reports. Our design allows us to test whether this merit-based option is applied in a non-discriminatory manner in practice.
Second, admissions are decentralized: each Gymnasium independently selects its incoming students. This institutional structure increases the likelihood that teachers’ individual criteria and perceptions play a role in the selection process. Third, although schools must comply with state regulations prohibiting discrimination based on students’ background, our setting allows us to test whether such legal norms effectively prevent discriminatory behavior in actual decision-making.

2026-01-15

2026-02-15

The primary outcome variable is responsiveness of the addressed high schools. Responsiveness is measured as follows. If we observe a non-automated response to our treatment email, we mark it as “1”. If we do not observe a non-automated response, we mark it as “0”.
(i) We expect that the overall responsiveness towards inquirers without immigrant background is higher compared to high school's responsiveness towards inquirers with immigrant background.
(ii) We expect that high schools within districts that have a higher share of immigrants answer more frequently to inquirers with an immigrant background than schools in districts that have a low share of immigrants.
(iii) We expect that schools are more responsive to male compared to female inquirers.
(iv) We expect that schools are more responsive to inquirers with high GPA (“Gymnasialempfehlung”) compared to inquirers with low GPA (“keine Gymnasialempfehlung”).

(i) There is evidence that German school administrators respond more frequently to inquiries from families without an immigrant background than to inquiries from families with a migration background.
(ii) We expect to observe similar patterns in the present study, as our earlier nationwide field experiment using Turkish-sounding names already documented substantial ethnic discrimination in school admission responses.
(iii) Building on these prior findings and drawing on insights from gender and migration studies, we hypothesize that families with Arabic-sounding names—who may be perceived as culturally more distant—will face equal or potentially higher levels of discriminatory behavior.
(iv) Independently of ethnic background, responses may also be driven by rational-choice considerations: admission offices have incentives to favor high-GPA students over those with weaker academic profiles.

Apart from responsiveness we are interested in the tonality and the text length of the written answers to the students. We expect that the tonality also varies across the treatment dimensions. This will be an additional variable we will use once we have a sufficient number of observations. In addition, we are interested in categorizing the answers between helpful answers, rejections and other categories that exemplify good, bad or even ugly answer types.
(i) We expect that migrant inquires receive non-helpful answers more often compared to non-migration inquires.
(ii) We expect that that migrant inquires receive neutral or even negative answers more often compared to non-migrant inquires.
(iii) We expect that non-migrant inquirers receive shorter answers compared to migrant inquires because respondents are spending more time to write non-helpful or negative answers compared to positive answers.

All effects are based on rational behavior or cognitive dissonances:
(i) The respondent shows discrimination not by neglect but by choosing a lower effort level.
(ii) The respondent shows discrimination not by neglect but by choosing a lower effort level quality.
(iii) We believe that humans who are intentionally discriminate are aware of this action and justify their behavior by writing longer answers.

This experiment aims to find effects of taste-based discrimination within the German high school system in the responsiveness toward students with and without an immigrant background and with either very good or intermediate grades.
Therefore, we designed an experiment with four binary treatment dimensions on the side of the inquirers who ask high school admission offices about information on how to apply for high school.
We take all high schools of the 16 states in the Federal Republic of Germany into account.
The first treatment dimension is a potential immigrant background of the family. The inquirer’s name signals, from a German point of view, whether the family has an immigrant background or not.
The second dimension is the gender of the inquirer, again signaled by a female or male first name.
The third dimension varies the academic performance of the child, indicating whether there is a Gymnasium recommendation (high GPA) or not (intermediate GPA).
The fourth dimension concerns the gender of the child; the email explicitly states whether the inquiry refers to a son or a daughter.
Each school is contacted via email by one of the sixteen profiles formed by all convex combinations of these four dimensions. We will then observe whether a high school responds to the request and, if so, how much time is required.

Inquiries are sent via email by four fictional parent personas. First and last names are chosen such that they strongly signal whether the family has an immigrant background (e.g., Maryam Mahmoud Al-Numan) or not (e.g., Sabrina Hartmann) according to common naming patterns in Germany. The first names also convey the gender of the inquirer, allowing us to vary parental gender.
In addition, the inquiries explicitly state whether the child in question is a son or a daughter, providing a separate signal for the child’s gender as an additional treatment dimension. The academic performance of the child—high or intermediate—is transmitted through the wording in the inquiry by indicating the presence or absence of a Gymnasium recommendation.
The text of the inquiry makes clear that the parents are writing on behalf of a 4th-grade student transitioning from primary school. Thus, the target of all inquiries is a minor, a population of particular interest within social science research.
During our experiment, we generate two samples of data. First, we create an anonymized dataset in which each contacted school receives a randomly generated ID, and in which we document outcomes such as whether the school responded. Second, we maintain a password-protected control dataset that contains administrative information on the schools contacted. The latter is stored separately and securely, while the primary dataset is fully anonymized to ensure that individual schools or administrators cannot be identified. This setup guarantees that data protection requirements are fully met.

Randomization is done in office by a computer.

State level

Yes

15 clusters with respect to the randomization unit (state level).
16 clusters with respect to treatment clusters (four binary dimensions) within each randomization unit.

Around 3000+, which is the number of High Schools in the 16 States of the Federal Republic of Germany.

Approximately 350+ observations per treatment arm.
High GPA (Gymnasium Recommendation)
1. Immigrant background + Male parent + High GPA + Son
2. Immigrant background + Male parent + High GPA + Daughter
3. Immigrant background + Female parent + High GPA + Son
4. Immigrant background + Female parent + High GPA + Daughter
5. No immigrant background + Male parent + High GPA + Son
6. No immigrant background + Male parent + High GPA + Daughter
7. No immigrant background + Female parent + High GPA + Son
8. No immigrant background + Female parent + High GPA + Daughter
Low GPA (No Gymnasium Recommendation)
9. Immigrant background + Male parent + Low GPA + Son
10. Immigrant background + Male parent + Low GPA + Daughter
11. Immigrant background + Female parent + Low GPA + Son
12. Immigrant background + Female parent + Low GPA + Daughter
13. No immigrant background + Male parent + Low GPA + Son
14. No immigrant background + Male parent + Low GPA + Daughter
15. No immigrant background + Female parent + Low GPA + Son
16. No immigrant background + Female parent + Low GPA + Daughter