|
Field
Experimental Design (Public)
|
Before
We conduct a pre/post randomized correspondence experiment using publicly available rental listings. Eligible listings are assigned to one of 12 treatment cells defined by applicant race (Black- vs White-coded name), Housing Choice Voucher status (disclosed HCV receipt, disclosed non-receipt, or not disclosed), and applicant quality (high vs low). Race and quality are assigned with equal probability within each HCV condition; HCV=Yes and HCV=No each receive 32% of assignments and HCV=Not Disclosed receives 36%. We compare landlord responses and listing content before and after the October 1, 2026 policy changes.
Treatment effects will be estimated using regression models that compare outcomes across randomized treatment groups and policy periods, including interaction terms corresponding to the study’s primary hypotheses. Specifications include pre-specified covariates and fixed effects for listing and inquiry characteristics, with standard errors clustered at the landlord level. We also estimate higher-order interactions to test whether post-policy changes differ by race, HCV status, and applicant quality.
We describe these features in greater detail in our Pre-Analysis Plan.
|
After
We conduct a pre/post randomized correspondence experiment using publicly available rental listings. In each stage of measurement (pre and post Oct 1) we will identify eligible listings using pre-specified criteria, and we will attempt to send two inquiries to each eligible listing.
For the first inquiry, each eligible listing will be randomly assigned to one of 12 treatment cells defined by applicant race (Black- vs White-coded name), Housing Choice Voucher status (disclosed HCV receipt, disclosed non-receipt, or not disclosed), and applicant quality (high vs low). Race and quality are assigned with equal probability within each HCV condition; HCV=Yes and HCV=No each receive 32% of assignments and HCV=Not Disclosed receives 36%.
Race, voucher status, and quality for the second inquiry will be assigned conditional on assignments made for the first inquiry. We use pre-specified rules, as defined in our Pre-Analysis Plan, to generate second inquiry assignments.
We compare landlord responses and listing content before and after the October 1, 2026 policy changes.
Treatment effects will be estimated using regression models that compare outcomes across randomized treatment groups and policy periods, including interaction terms corresponding to the study’s primary hypotheses. Specifications include pre-specified covariates and fixed effects for listing and inquiry characteristics, with standard errors clustered at the landlord level. We also estimate higher-order interactions to test whether post-policy changes differ by race, HCV status, and applicant quality.
We describe these features in greater detail in our Pre-Analysis Plan.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
Eligible inquiries will be randomized across 12 Race × HCV Status × Applicant Quality treatment cells. HCV=Yes and HCV=No will each receive 32% of the sample, while HCV=Not Disclosed will receive 36%. Within each HCV condition, inquiries will be assigned 50/50 to Black- versus White-coded applicant names and 50/50 to high- versus low-quality inquiries.
For a planned total sample of 15,000 inquiries, the target cell sizes are therefore:
HCV=Yes, White, Low Quality: 1,200
HCV=Yes, White, High Quality: 1,200
HCV=Yes, Black, Low Quality: 1,200
HCV=Yes, Black, High Quality: 1,200
HCV=No, White, Low Quality: 1,200
HCV=No, White, High Quality: 1,200
HCV=No, Black, Low Quality: 1,200
HCV=No, Black, High Quality: 1,200
HCV=Not Disclosed, White, Low Quality: 1,350
HCV=Not Disclosed, White, High Quality: 1,350
HCV=Not Disclosed, Black, Low Quality: 1,350
HCV=Not Disclosed, Black, High Quality: 1,350
|
After
Each sampled rental listing receives two inquiries. Treatment is assigned at the inquiry level using a blocked 2 (race) × 3 (HCV disclosure status) × 2 (applicant quality) design, with the two assignments within a listing jointly randomized subject to the prespecified pairing rules. Across the full set of inquiries, the design targets 50% Black-coded and 50% White-coded names, 50% high-quality and 50% low-quality inquiries, and approximately 25% HCV-Yes, 25% HCV-No, and 50% HCV-not-disclosed inquiries.
For a planned total sample of 15,000 inquiries, the target cell sizes are therefore:
HCV=Yes, White, Low Quality: 1,200
HCV=Yes, White, High Quality: 1,200
HCV=Yes, Black, Low Quality: 1,200
HCV=Yes, Black, High Quality: 1,200
HCV=No, White, Low Quality: 1,200
HCV=No, White, High Quality: 1,200
HCV=No, Black, Low Quality: 1,200
HCV=No, Black, High Quality: 1,200
HCV=Not Disclosed, White, Low Quality: 1,350
HCV=Not Disclosed, White, High Quality: 1,350
HCV=Not Disclosed, Black, Low Quality: 1,350
HCV=Not Disclosed, Black, High Quality: 1,350
|