Measuring selective admissions in nursing homes

Using on-line contact forms on nursing home websites, we submitted fictitious inquiries asking about availability to approximately 3,000 nursing homes, experimentally varying the patient’s payer (Private, Medicaid, or unstated), medical condition (focusing on Alzheimer's vs other conditions) and perceived race (Black or white). Inquiries come from daughters inquiring about placement for their elderly mother. Names and email addresses signal race, while payer and medical condition are signaled in the message content.

2025-03-13

2025-08-28

Availability, defined as a response that did not indicate a lack of availability (in other words, either a positive or neutral response, which we term non-negative availability).

Any response; number of responses received within 30 days (a proxy for nursing home's eagerness to reach the patient); positive sentiment (tone).

We conduct an audit study wherein we send fictitious inquiries to nursing homes and measure outcomes based on the responses we receive. We use a 2x3x4 factorial design that includes race (Black or white); payer status (Medicaid, private pay, or unspecified); and four different vignettes for different medical conditions (focusing on Alzheimer's vs other conditions).

Each inquiry to a nursing home involves a daughter seeking admission to a nursing home from the community on behalf of her elderly mother. The inquiry includes a vignette describing the mother’s (randomly varied) condition and payer status; it also includes mother and daughter’s (randomly varied), racially concordant names and ask if a bed is available. To preserve power and limit the treatment space, we included only inquiries from and on behalf of women, removing the possibility of gender effects in the study. We focus on women because the majority of nursing home residents are women, and because adult daughters tend to take on a greater proportion of elder care responsibilities (Grigoryeva 2017).

We attempted to send inquiries to nursing homes that accept patients from the community and that have online contact forms (our mode of sending inquiries). We formed this target sample using the Centers for Medicare & Medicaid Services June 2024 Provider Information nursing home dataset, a publicly available list of all active, certified nursing facilities in the United States.

Not available

We assigned each facility in our sample to receive two inquiries, to be delivered over two study implementation rounds (as well as a small pilot). Round one inquiries were constructed using independent, simple random assignment of race (Black or white assigned with equal probability), payer (Medicaid, private, unspecified assigned with equal probability), and medical vignette (one of 16 possibilities, with four each corresponding to the four core vignettes, again with equal probability). Within race, we randomly assigned one of eight possible name pairs with equal probability. Round two inquiries were constructed using random assignments conditional on not repeating the same payer, race, or vignette treatment used for that facility in round one.

Random assignment was done at the facility-inquiry level, conditional on prior assignments for that facility

No

3000 nursing home facilities

6000 observations at the inquiry level

All treatments randomized using simple random assignment in equal proportions.