Differential Treatment of Families with Children: A Correspondence Study of the Moscow Real Estate Market (phase 1)

This is a correspondence study that aims at detecting differential treatment of families with children in the real estate market. We contact subjects through an online platform where landlords post ads showcasing their properties and the conditions for rental. The contact message expresses interest in the property and indicates the composition of the household. Specifically, there will be four treatment arms: single female, childless couple, a couple with a 2 year old son, and a single mother with a 2 year old son. We record response rates and response times, as well as the details of the response message. We test for differential outcomes on average and check for heterogeneous effect based on property value and other pre-treatment covariates.

Phase 1 will assign subjects to four distinct treatments. In correspondence studies, the difference among treatments resides only in the information contained in the message sent to the experimental subjects. Below we present the exact wording of the messages that will be sent to each of the treatment groups:

Phase 1 Treatment Details
Group 1: "Hello! My name is Lyubov, I liked your apartment. Tell me please, when is it possible to see it? I am looking for a place for myself, I will live alone. I am a Russian citizen."
Group 2: "Hello! My name is Lyubov, I liked your apartment. Tell me please, when is it possible to see it? I am looking for an apartment for me and my husband. We don’t have children. We are Russian citizens."
Group 3: "Hello! My name is Lyubov, I liked your apartment. Tell me please, when is it possible to see it? I am looking for an apartment for me, my husband and our 2 year old son. We are Russian citizens."
Group 4: "Hello! My name is Lyubov, I liked your apartment. Tell me please, when is it possible to see it? I am looking for an apartment for me and my 2 year old son. We are Russian citizens."

In summary, the contact messages are fundamentally identical other than the information corresponding to the household composition of the tenant. Treatment group 1 corresponds to a single Russian woman. Treatment group 2 is a Russian couple without children. Treatments 3 and 4 are tenants with a two-year-old son.

Of course, the actual contact message will be in Russian. Please see the pre-analysis document (attached) for the Russian language version.

2026-02-23

2026-05-15

1. Response rate
2. Time to response
3. Positive response (apartment is available for visit without delay and no further questions/tests)

Correspondence study on an online real estate platform.

This is a correspondence study on an online real estate platform. We target the following population of available properties in Moscow (excluding the historical center):

1. 2/3 bedroom apartments
2. 40-70 square meters
3. Available for long-term rental (1 year+)
4. Registration of residents allowed
5. Children allowed
6. Available from owner directly without agent commission

We exclude the city center because it is a niche real estate market with very idiosyncratic characteristics. Requirements 1 through 4 are meant to target properties that would be suitable, within reason, to any of the households described in the different treatment arms.

Requirement 5 deserves separate discussion. The platform gives landlords the option to indicate whether families with children are welcome to apply. Openly declaring a preference against families with children is perfectly legal in Russia. Roughly 26% of the ads in the target population on cian.ru do not welcome applications from families with children. In phase 1 of this study we focus on the 74% of ads that claim to welcome children and test whether discrimination against children is present.

Finally, requirement 6 is meant to avoid real estate agents and focus on the preferences of landlords directly.

The online real estate platform cian.ru provides an excellent setting for the experiment. We are able to search for properties that fit the necessary criteria and sort them by date of publication. We are able to download the full list and apply our random assignment algorithm directly. Finally, the platform allows us to contact landlords directly through a text-only messenger utility available from the advertisement's web page.

One important caveat is that around 20% of ads do not have the messenger utility enabled. In a pilot study, we found no significant relationship between the availability of the messenger and the main property descriptors (number of bedrooms, square meters, asking price, Moscow district). We do not expect this form of attrition to significantly unbalance the treatment groups.

The randomization to treatment will be done by a computer. Specifically, we will stratify the sample by three price categories and randomize into the four treatment arms within strata.

The unit of randomization is the individual property. We will put checks in place to avoid messaging the same landlord more than once over the course of the study.

No

Treatment is administered at the property level. There is no clustering.

Following the power analysis detailed in the attached document, our target sample size for phase 1 is 3500 contacted landlords.

We aim at equal-sized treatment arms. In other words, we expect to have around 3500/4=875 observations per treatment arm.

The minimum detectable effect size is 5 percentage points. See the attached pre-analysis document.