Experimental Design
In our correspondence experiment, we utilized the contact form on an online rental platform, which enables users to message the individual who posted a listing. For each listing, we sent two separate messages from two different accounts: one using a Russian-sounding name and the other using a non-Russian-sounding name. The experiment was carried out in two rounds.
The platform provides users with two options for contacting landlords or agents: either by public mobile phone or through an online form. The online form is commonly used for submitting short, clarifying questions about a listing. To make sure that the only difference in applications is the name, the online form was chosen for this experiment.
We used two simple questions for the intervention. Translations of these questions are as follows:
Q1. Hello, I'm interested in your apartment. May I contact you later tonight?
Q2. Good afternoon, I'm interested in your apartment. I would like to ask a clarifying question. When could one move into the apartment? [First name]
The content of the questions is unrelated to ethnic discrimination. Their purpose is simply to elicit a response based on the name of the applicant. The message is a preliminary step before a phone conversation, which precedes a personal visit to the apartment. The online form is rarely used to finalize deals or negotiate terms. The experiment was designed to allow landlords to ignore the potential tenant, disrupting interaction at the very first stage.
When applicants submit their messages through the form, landlords only see the content of the message. Nevertheless, separate accounts with realistic email addresses were created for each identity.
The variation in perceived ethnicity based on names is the key factor in the experiment. Two rounds of the experiment were conducted, with different approaches to name selection.
It is important to note that in Russia, there is no comprehensive dataset of birth names. For the first round, we selected two names: the Russian name Andrei and the Turkic name Arslan, both popular and recognizable in Russia.
In the second round, we adopted a more systematic approach to name selection. We compiled a dataset of popular names in Russia using statistics from the social network vk.com (the most commonly used media in Russia comparable in popularity to Facebook in the U.S. We constructed a ranking of name frequencies for each Russian city.
The most Russian-sounding names were selected using data from Moscow, while the most non-Russian-sounding names were chosen based on data from Makhachkala. According to the 2010 Census, Moscow's population is 90 per cent ethnically Russian, whereas in Makhachkala, only 6.3 per cent of the residents are Russian. The largest ethnic groups from Makhachkala (Avars, Kumyki, Dargins, Lezgins, Laks) face systematic discrimination in Moscow's housing and labor markets. Chechen job seekers had one of the lowest response rates. Given the overlap between names of major ethnic groups in Dagestan and Chechen names, this result applies to many of the most popular names in Makhachkala.
The ten most popular names from Moscow were selected to represent Russian-sounding names, while the ten most common names from Makhachkala were chosen to represent non-Russian-sounding names. These selected names were used in the second round of the study. To avoid overlap, names that ranked highest at the national level, as well as those used in the first round of the experiment, were excluded.
The experiment was conducted in two rounds: June 20-21, 2018, and December 13-14, 2019. The only difference between the rounds was the selection of Russian-sounding and non-Russian-sounding names. The experiment targeted fresh listings no older than one day. To prevent potential bias from contacting the same person through different listings, we excluded all listings with non-unique phone numbers.
We manually sent the first message through the contact form to each listing, randomising Russian-sounding and non-Russian-sounding names. Automating the process was not feasible due to platform restrictions. The next day, we sent the second set of messages with alternative names (Russian-sounding name if there was a non-Russian-sounding name on the first day, and vice versa) from different accounts. A one-day gap was chosen to make control and treatment groups more comparable and minimize the chance that listings would become unavailable by the time of the second message.
Randomization of Russian-sounding and non-Russian-sounding name message order and text content ensured that timing would not bias results.