|
Field
Last Published
|
Before
November 21, 2023 11:26 AM
|
After
November 21, 2023 04:53 PM
|
|
Field
Intervention (Public)
|
Before
We collect survey data on people’s beliefs about the extent of labor market discrimination against women with a Turkish background in Germany using an online panel. We first investigate to what extent beliefs respond to information from different sources. In a second step, we examine to what extent these beliefs drive support for anti-discrimination policies.
The experiment hast two waves separated by two weeks. In the first wave, we elicit prior beliefs from all respondents about the extent of discrimination against female applicants with a Turkish background in the German labor market. We also elicit post-treatment beliefs about hiring discrimination using qualitative scales as well as beliefs about discrimination in another market, namely, the carpooling market. This enables us to check if there are short-term belief updates based on the information provision. Furthermore, we ask questions about support for anti-discrimination policies. In the second wave, we ask again the same questions about beliefs about hiring discrimination in the labor market and the policy support questions.
In the first wave, we elicit prior beliefs distinguishing between beliefs attributed to the Turkish background and beliefs attributed to presumed religious affiliation. We present respondents with the same profiles used in the correspondence study by Weichselbaumer (2020), which we use as a benchmark for objectively measured callback discrimination. The Turkish background is signaled through the name of the applicant (Meryem Öztürk instead of Sandra Bauer) while all other characteristics that are part of the job application are held constant (including qualifications, age, German nationality, and photograph).
We present respondents with hypothetical job applicants who are identical in all dimensions except either their name or their profile picture (see main part in the appendix). We inform them about the callback rate for the reference scenario of a female applicant aged 27 named Sandra Bauer (candidate A), which is considered a German-sounding name. Respondents are asked to estimate the callback rate for another candidate for whom a Turkish background is signaled through the name (Meryem Öztürk instead of Sandra Bauer, candidate B) while all other characteristics are held constant (including qualifications, age, German nationality, and photograph). Using the same procedure, we isolate the specific role of religion by assessing beliefs about the callback rate for a third candidate with the same Turkish-sounding name (Meryem Öztürk) with photographs showing the same candidate with headscarf (candidate C). She is again compared to Sandra Bauer, i.e., the same candidate with a German-sounding name and without headscarf.
After eliciting the prior beliefs, respondents are allocated to one of three groups using computer-assisted randomization. The first two subsets of respondents (the treatment groups) receive information about the results from the correspondence study by Weichselbaumer (2020). In her study, the author tested for labor market discrimination by randomly varying (i) whether names on fictitious resumes were Turkish- or German-sounding or (ii) whether the photograph of candidates with a Turkish-sounding name were with or without headscarf. We present the first treatment group with information on the study results using a summary we wrote ourselves. We include a link to a YouTube video where the author presents the study results herself. We refer to this treatment as the academic treatment. We offer the second treatment group an excerpt from a newspaper article reporting on the same study results. We refer to this treatment as the newspaper treatment. The remaining respondents constitute the control group and do not receive any information from the research article. Instead, they receive a general article about gender differences in the labor market.
Following the information treatment, we measure post-treatment belief updates using explicit questions on beliefs about discrimination as well as results from a different correspondence study and assess preferences for anti-discrimination policies.
To reduce concerns regarding anchoring, we elicit post-treatment beliefs in the first wave in another market, namely, carpooling. This question uses a smaller range of possible answers (0-10 instead of 0-100) and success rates are much higher than in the job application study. These differences should make anchoring more difficult for respondents. Two weeks after the first wave, we measure again beliefs about hiring discrimination in the labor market and we elicit our respondents’ views regarding anti-discrimination policies. This second wave allows us to verify the stability of belief updates in response to the information provision in the first wave and to mitigate concerns about pure experimenter-demand effects.
|
After
The intervention is an information treatment. Respondents in treatment groups receive information about the results from a correspondence study conducted in Germany. The study finds evidence of hiring discrimination against women with a migration background signaled by their name.
|
|
Field
Primary Outcomes (Explanation)
|
Before
|
After
See pre-analysis plan.
|
|
Field
Experimental Design (Public)
|
Before
We collect survey data on people’s beliefs about the extent of labor market discrimination against women with a Turkish background in Germany and examine to what extent these beliefs drive support for antidiscrimination (i.e, affirmative action) policies.
The experiment proceeds in two waves. In the first wave of the experiment, we elicit prior beliefs from all subjects about the extent of discrimination against female applicants with a Turkish background in the German labor market. We also elicit post beliefs in the first wave in another market, namely, the carpooling market, to check if there are short-term belief updates based on the information provision. Furthermore, we include questions about support for anti-discrimination policies. In the second wave, we ask again the same questions about beliefs about hiring discrimination in the labor market and their policy support questions.
In the first wave, we elicit prior beliefs distinguishing between beliefs attributed to the Turkish background and beliefs attributed to presumed religious affiliation. We present respondents with the same profiles used in the correspondence study by Weichselbaumer (2020), which we use as a benchmark for objectively measured callback discrimination. The Turkish background is signaled through the name of the applicant while all other characteristics that are part of the job application are held constant (including qualifications, age, German nationality, and photograph).
We present respondents with hypothetical job applicants who are identical in all dimensions except either their name or their profile picture. We inform them about the callback rate for the reference scenario of a female applicant aged 27 with a German-sounding name. Respondents are asked to estimate the callback rate for another candidate for whom a Turkish background is signaled through the name while all other characteristics are held constant. Using the same procedure, we isolate the specific role of religion by assessing beliefs about the callback rate for a third candidate with the same Turkish-sounding name with photographs showing the same candidate with headscarf (candidate C). She is again compared to the same candidate with a German-sounding name and without headscarf.
After the prior-belief elicitation, a random subset of subjects (the treatment group) receives information about the true results from the correspondence study by Weichselbaumer (2020). The remaining subjects constitute the control group and do not receive any information from the research article. Instead, they receive a general article about gender differences in the labor market.
Subsequently, we measure post-treatment belief updates using explicit questions on beliefs about discrimination as well as results from a different correspondence study and assess preferences for antidiscrimination (affirmative action) policies.
To reduce concerns regarding anchoring, we elicit post-treatment beliefs in the first wave in another market, namely, carpooling. This question uses a smaller range and success rates are much higher than in the job application study. These differences should make anchoring more difficult for respondents. In the second wave of the experiment, two weeks after the first wave, we measure again beliefs about hiring discrimination in the labor market and we elicit our subjects’ views regarding antidiscrimination policies. This second wave allows us to verify the stability of belief updates in response to the information provision in the first wave and to mitigate concerns about pure experimenter-demand effects. In addition, it ensures that anchoring, if any, is mitigated.
|
After
We collect survey data on people’s beliefs about the extent of labor market discrimination against women with a Turkish background in Germany using an online panel. We first investigate to what extent beliefs respond to information from different sources. In a second step, we examine to what extent these beliefs drive support for anti-discrimination policies.
The experiment hast two waves separated by two weeks. In the first wave, we elicit prior beliefs from all respondents about the extent of discrimination against female applicants with a Turkish background in the German labor market. We also elicit post-treatment beliefs about hiring discrimination using qualitative scales as well as beliefs about discrimination in another market, namely, the carpooling market. This enables us to check if there are short-term belief updates based on the information provision. Furthermore, we ask questions about support for anti-discrimination policies. In the second wave, we ask again the same questions about beliefs about hiring discrimination in the labor market and the policy support questions.
In the first wave, we elicit prior beliefs distinguishing between beliefs attributed to the Turkish background and beliefs attributed to presumed religious affiliation. We present respondents with the same profiles used in the correspondence study by Weichselbaumer (2020), which we use as a benchmark for objectively measured callback discrimination. The Turkish background is signaled through the name of the applicant (Meryem Öztürk instead of Sandra Bauer) while all other characteristics that are part of the job application are held constant (including qualifications, age, German nationality, and photograph).
We present respondents with hypothetical job applicants who are identical in all dimensions except either their name or their profile picture (see main part in the appendix). We inform them about the callback rate for the reference scenario of a female applicant aged 27 named Sandra Bauer (candidate A), which is considered a German-sounding name. Respondents are asked to estimate the callback rate for another candidate for whom a Turkish background is signaled through the name (Meryem Öztürk instead of Sandra Bauer, candidate B) while all other characteristics are held constant (including qualifications, age, German nationality, and photograph). Using the same procedure, we isolate the specific role of religion by assessing beliefs about the callback rate for a third candidate with the same Turkish-sounding name (Meryem Öztürk) with photographs showing the same candidate with headscarf (candidate C). She is again compared to Sandra Bauer, i.e., the same candidate with a German-sounding name and without headscarf.
After eliciting the prior beliefs, respondents are allocated to one of three groups using computer-assisted randomization. The first two subsets of respondents (the treatment groups) receive information about the results from the correspondence study by Weichselbaumer (2020). In her study, the author tested for labor market discrimination by randomly varying (i) whether names on fictitious resumes were Turkish- or German-sounding or (ii) whether the photograph of candidates with a Turkish-sounding name were with or without headscarf. We present the first treatment group with information on the study results using a summary we wrote ourselves. We include a link to a YouTube video where the author presents the study results herself. We refer to this treatment as the academic treatment. We offer the second treatment group an excerpt from a newspaper article reporting on the same study results. We refer to this treatment as the newspaper treatment. The remaining respondents constitute the control group and do not receive any information from the research article. Instead, they receive a general article about gender differences in the labor market.
Following the information treatment, we measure post-treatment belief updates using explicit questions on beliefs about discrimination as well as results from a different correspondence study and assess preferences for anti-discrimination policies.
To reduce concerns regarding anchoring, we elicit post-treatment beliefs in the first wave in another market, namely, carpooling. This question uses a smaller range of possible answers (0-10 instead of 0-100) and success rates are much higher than in the job application study. These differences should make anchoring more difficult for respondents. Two weeks after the first wave, we measure again beliefs about hiring discrimination in the labor market and we elicit our respondents’ views regarding anti-discrimination policies. This second wave allows us to verify the stability of belief updates in response to the information provision in the first wave and to mitigate concerns about pure experimenter-demand effects.
|
|
Field
Planned Number of Clusters
|
Before
3.300 respondents from an online panel.
|
After
Wave 1: 3.500 individual respondents from an online panel.
Wave 2: 2450 individual respondents after 70% attrition
|
|
Field
Planned Number of Observations
|
Before
3.300 respondents from an online panel.
|
After
Wave 1: 3.500 respondents from an online panel.
Wave 2: 2450 respondents after 70% attrition
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
1.100 respondents from an online panel.
|
After
1.166 individual respondents per treatment arm in the first wave.
816 individual respondents per treatment arm in the second wave after 70% attrition
|
|
Field
Power calculation: Minimum Detectable Effect Size for Main Outcomes
|
Before
|
After
The minimum detectable effect size for the main outcomes is 0.15 of a standard deviation with 80% power.
|
|
Field
Secondary Outcomes (End Points)
|
Before
|
After
See pre-analysis plan.
|
|
Field
Secondary Outcomes (Explanation)
|
Before
|
After
See pre-analysis plan.
|