Differential Treatment of Families with Children: A Correspondence Study of the Moscow Real Estate Market (Phase 2)

We send a message to landlords whose listings state that children are not accepted. The message expresses interest in the property and states that the applicant has a two-year-old child. Acknowledging the restriction, the applicant asks the landlord to make an exception.

Listings are randomly assigned to one of four versions of this message. The first arm offers no financial incentive (control). The remaining three offer alternative financial incentives. The specific incentives are described in the pre-analysis plan and will be made public when data collection closes.

2026-09-22

2027-02-28

1. Response rate
2. Positive response (robustness)

The response rate is binary: 1 if we receive any response within 7 days, 0 otherwise. Positive response is a robustness recoding of the primary outcome. We set it to zero for any response that does not agree to a viewing or indicate that the apartment is available. Non-committal replies and any stated reason not to show the apartment—including that the property is unavailable or that children are not accepted—are coded as zero. Non-responders remain zero.

Time to response (latency)

Time to response (in minutes) is the timestamp difference (registered in the platform) between the time the message is sent to the landlord and the landlord's response time.

We conduct a correspondence study on an online real estate platform. The target population is listings that explicitly state that children are not accepted.

We randomly assign listings to one of four arms: a control arm in which the applicant asks for an exception with no financial incentive, and three treatment arms in which the applicant offers a different financial incentive alongside the exception request. The specific incentive amounts are specified in the pre-analysis plan and will be made public when data collection closes.

Randomization is at the listing level. Listings are stratified by asking price and divided into blocks of four, with one listing per arm per block. Because new listings appear continuously on the platform, sampling proceeds in rounds; a round whose length is not a multiple of four may leave leftover listings, which are assigned to arms uniformly at random without replacement and pooled with the blocked listings for analysis.
All messages follow a common template and differ only in the incentive clause. The applicant is a married woman with a generic Russian name, seeking housing for herself, her husband, and their two-year-old son. All applicants are described as Russian citizens, and all landlords are contacted through the platform's messaging system.

The target sample is 2,750 contacted listings. Data collection is expected to take between 16 and 18 weeks.

Not available

Randomization will be done by a computer after block-stratifying by asking price

The randomization unit is the individual listing

No

We sample at most one listing per landlord: once a landlord has been contacted, any subsequent listings posted by that landlord are dropped from the sampling frame. Responses are therefore independent across listings in the sample.

2750 listings

The treatment arms are equal weighted, hence around 688 per treatment

The MDE is a 5pp increase in the response rate at 80% power and 5% significance (two-sided), corresponding to approximately 6% of the baseline response rate. Larger effects are detectable with power approaching 100%.