Flat-hunting experiment suggests men with Arab-sounding names face bias in Germany.
While searching for a flat in Berlin, Hertie School Master of Public Policy student Gabriel Tarriba encountered some misconceptions about his home country Mexico. That got him thinking. How could you examine whether this prejudice was real, and did it actually affect a person’s ability to find a place to live?
Tarriba was also searching for an experiment for Professor Christian Traxler’s Behavioural Economics course in the spring of 2016. Students had to devise an experiment that would test a hypothesis and identify a causal effect – the basis of empirical research in the social sciences.
“The idea was to get some kind of meaningful results,” Tarriba said. “Most people would have said: of course this is true, but how do you prove it? You have to show the evidence.”
In the winter of 2015/1016, Europe was at the height of the refugee crisis, when around a million people from the Middle East, Central Asia and Africa entered Germany seeking asylum. There were many stories in the media that people had a hard time finding a flat. But how could you prove discrimination played a role?
Tarriba and a group of 10 fellow Hertie School students wanted to find out if people with non-German background, in particular from the Middle East, had a lower chance of finding a shared flat. Using a popular flat-search website they – or rather, four fictitious students – sent 1,075 requests for rooms in 11 German cities. The messages were identical except for the names of the senders – one German and one Arab-named female, and one German and one Arab-named male. Jan Müller, Julia Müller, Mohammed Sharif and Fatima Sharif were 23 years old, from Frankfurt, corresponded in flawless German and were all enrolled in master’s programmes.
“We spent eight to ten hours sending messages over several weeks,” Tarriba said. “Then we waited and marked the replies on a spreadsheet. We had similar results in every city.”
Tarriba and his fellow researchers expected a higher response rate for German names than for Arab names and a higher response rate for females than for males. The students were surprised at the clarity of the result: messages signed with an Arabic-sounding name were on average 14 percentage points less likely to receive a positive response compared to a German name, the researchers found. Mohammed was the only recipient with a below-average response rate, at a mere 39%, while Fatima received 54%, Jan 59% and Julia 62%. The average was 53%.
Indeed, after the Hertie School’s 2016 study, data journalists at the German news website Spiegel Online and broadcaster Bayerische Rundfunk performed a similar study on a larger scale, published in 2017, using German and foreign-sounding names – Arab, Turkish, Polish and Italian. Sending out 20,000 mails, they had similar results, discovering that in every fourth case a German-named candidate was invited to look at a flat, a flat-hunter with a foreign name was turned down.
“I used to think that for randomized control trials you needed a lot of funding and a lot of people, but with this kind of experiment you suddenly realize you can do really cool things without a lot of resources,” Tarriba said. “You just find an everyday issue and find a way to design an experiment. In the end we did something quite simple, and got quite a clear result.”
Professor of Economics Christian Traxler teaches the course every year. It is designed to give students practical experience in carrying out evidence-based research.
“Students are in a kind of competition with each other to create a well-designed experiment,” Traxler explained. “Based on the feedback from people working experimentally in government organisations or the academic sector, we then select the best ones to actually carry out.” In the future, the course will also provide students with a basis for writing their master’s thesis.
Students found the results of the first flat-hunting experiment so interesting that in Traxler’s spring 2017 course, a new group took it a step further. This time, they wanted to find out whether it was possible to overcome bias by offering more information about the applicants. The idea was to spark so-called deliberative effects, in which respondents’ reactions might be informed by sympathy or prejudice.
“The idea was to find out how could we mitigate this absurdly high response gap – not with a big policy measure, but rather at the individual level,” Traxler explained.
This time, the students focused only on men and tested six types of emails. As in the 2016 trial, they first sent out an email from Mohammed or Jan expressing interest in a flat. The second set of emails conveyed a bit more information, mentioning that the applicant did volunteer work. The last set augmented this, in hopes of inducing deliberative effects, by mentioning that their volunteer work was with refugees. The students expected that reminding respondents of the difficulties faced by immigrants would increase response rates for the Arab name as compared to the German name. Would these deliberative effects help in overcoming bias?
The results were mixed and somewhat puzzling. The students confirmed the results of the first experiment. Jan had a 36% response rate and Mohammed 20% to the basic emails. Paradoxically, adding social information increased Jan’s response rate to 41%, but Mohammed’s dropped to 14%.
In the mails that mentioned refugee volunteering, the results were not perfectly clear, but it did seem to have a small positive effect, as the response rate for Mohammed rose from 14% back up to 21%, slightly above that of the basic email.
Experiments like this are a way to teach social science students how to test different approaches and to derive causal evidence, a vital key to devising policy solutions. Without proof of causality – knowing for sure what’s at the root of a problem – decisions are made on correlations, which are often misleading. A number of Traxler’s students have gone on from his course to work in the field of behavioural public policy, where they use experimental techniques to evaluate traditional policies or policies using “nudges”.
“Two former Hertie students are working for the UK’s Behavioural Insights Team, and several others are doing impact evaluations, using experimental techniques,” he said.
Among the Hertie School students who participated in the project are Giacomo Bagni, Fabian Bohnenberger, Laurence Hendry, Kevin Müller, Lena Schaffel, Faruk Tuncer and Lukas Wiese.
Christian Traxler, Professor of Economics
Gabriel Tarriba (MPP 2017) will pursue his PhD this autumn with Hertie School Professor of Sociology Michaela Kreyenfeld. He plans to study the integration of second generation children who migrated at an early age to Western countries.