The Experiment

Fictitious persons

In order to find out whether people looking for apartments are discriminated because of their name and associated migration background, we have constructed the following fictitious persons:

Name	Age	Origin	Sex
Hanna Berg	27	German	female
Nina Weiss	26	German	female
Stephan Braun	26	German	male
Daniel Buschle	27	German	male
Aylin Demirci	27	Turkish	female
Hamit Yilmaz	26	Turkish	male
Milena Adamowicz	27	Polish	female
Mikolaj Janowski	26	Polish	male
Maryam Abedini	26	Arabic	female
Ismail Hamed	27	Arabic	male
Vittoria di Lauro	26	Italian	female
Stefano Loguercio	27	Italian	male
Carsten Meier	42	German	male
Lovis Kuhn	25	German	male

Turkish, Italians and Polish are the largest groups of immigrants in Germany. In addition, persons of Arab origin were chosen as they are supposed to be particularly exposed in times of migration crisis. Otherwise, all our fictitious persons had more or less identical characteristics. They are 26 to 27 years old and first-time employees in marketing - a large professional field for young academics.

In order to record, how positive or negative the feedback can be besides , we have created two more profiles. On the one hand: Carsten Meier, a doctor in his early 40s, single, with an exemplary cover letter - the ideal tenant. On the other hand Lovis Kuhn, long-term student. His cover letter is informal, casually written.

Selection of apartments

The housing supply in Germany is extremely diverse. We put our focus on situations happening thousands of times day to day. An average tenant is looking for an ordinary apartment. He or she is not looking for a penthouse or a car-free housing project, but simply an affordable apartment with 30 to 60 m2 of living space. As demanded by the steadily growing number of single households.

In order to have a broad selection of suitable apartments, we have scraped the two largest online platforms, ImmobilienScout24 and Immowelt. In June and September 2016, we responded twice a day to new housing advertisements in the cities of Berlin, Dresden, Dortmund, Frankfurt, Hamburg, Cologne, Leipzig, Magdeburg, Munich and Nuremberg.

In order to take account of the varying price levels on the housing markets in different cities, we used the 2014 IMX rental price index. In Dresden, for example, we have chosen apartments with a monthly rent between 150 and 496 euros per month, in more expensive Munich apartments in the range of 312 to 937 euros.

Requests

In the mornings and evenings we have selected up to ten of the newly exposed apartments per city and automatically filled in the respective contact forms on the renting portals. We used ten different cover letters that are similar in length and wording, but at the same time different enough to not arouse suspicion among the providers. For the above-described candidates with extreme profiles, we each used a particularly good and a particularly bad cover letter.

(Due to linguistic subtleties, we did not translate these original German texts)

The advertiser of the chosen apartments were contacted with one German and two foreign profiles of the same gender each. The requests were sent every 30 minutes. One out of ten flats was additionally contacted during the same period by the two candidates with extreme profiles.

Categorizing

Following the pattern just described, we sent a total of 20,728 applications. We then received 8377 mails from homeowners, administrators and brokers. We have stored these answers in a database and assigned them to the respective request and person in a first step. The assignment often succeeded automatically on behalf of the unique ID in the apartment advertisement or the address of the apartment. When assigning by hand, the time of the answer, the title of the apartment advertisement, the price or the name of the contact person were helpful for the allocation.

Subsequently, suspected housing frauds and spam were sorted out in conspicuous, recurring passages of text and the remaining emails were categorized by hand. In doing so, we have rated a reply email as positive that either includes an invitation to a viewing appointment or specifically offers such a prospect. All other responses, as well as non-responses, were considered negative.

Discrimination rate

Discrimination is most evident when our fictitious persons have received different answers when applying for the same flat. Every apartment successfully requested by a German and a non-German person can be classified in the following cross tabulation:

	Positive foreign	Negative foreign
Positive German	pp	pn
Negative German	np	nn

In addition to a preference of German applicants (pn), there are also cases in which only the person with a foreign name receives a positive answer (np). This can be a deliberate decision of the landlord. Or simply related to the fact that the German applicant wrote later and the landlord has not even read the request. In order not to overestimate the actual extent of discrimination against persons with a foreign name, these cases are also considered in the calculation of the discrimination rate:

discr = (pn - np) / (pp + pn + np)

So we relate cases of unequal treatment to the number of apartments whose landlords responded positively to at least one of our two requests. Cases in which none of our persons has received a commitment (nn), we do not take into account when determining the rate of discrimination. They rather show how tense the housing market is.

We back up the discrimination rate with a logistic regression to control factors such as the order in which cover letters were sent. The model confirms the results from the descriptive calculation. It also helps us to determine whether differences in the extent of discrimination against different subgroups (e.g. women and men) are statistically significant. This corresponds to a study by the empirical social science researcher Prof. Auspurg from the LMU Munich.

Share of positive feedback

In addition to the extent of discrimination, we also look at how often the considered groups received positive responses in total. And relate that to the total number of requests:

p = positive_answers / requests

The proportion of positive feedback (p) helps us classify the importance of individual factors when looking for an appartment. This shows that the prospects for positive feedback also vary with the gender of the applicant, the city and the organisational form of the provider (commercial or private).

There are different types of landlords. While some offer mass visits, others invite only a few candidates to a local meeting. Therefore, we also calculate a logistic regression with a random factor for each apartment. This model accounts for the possibly different response behaviors of different landlords.

Odds ratio

A regression model also forms the basis for calculating the odds ratio between Germans and applicants with foreign names. The chance of a positive feedback results from the proportion of positive feedback (p) as follows:

chance = p / (1 - p)

The odds ratio (r) between foreign and German applicants is thus:

r = chance_foreign / chance_German

It shows how difficult it is for people with foreign names to find an apartment compared to German applicants. According to this logic, the chances of success in the respective cities result from a logistic regression, in which an interaction factor between the city and the ascribed migration background of an applicant is explicitly modeled.

A detailed description of the analyses and the associated program code is linked here (German). You can also find the complete dataset in the corresponding repository on GitHub.

If you have any questions about our research, please contact us via e-mail or on Twitter (BR Data, SPIEGEL ONLINE).