Do Nursing Homes Provide Better Care When Inspections Are Harder to Predict?
Gandhi, Olenski, and Shi (2026) ask whether predictable nursing home inspections weaken incentives to provide consistently high-quality care. Using federal inspection records, daily staffing data, patient assessments, and Medicare mortality records, they find that nursing homes reduce effort when an inspection is unlikely and increase effort as the next inspection becomes more likely. These changes matter for patients: a one-standard-deviation increase in facility effort is associated with a 5.4% reduction in mortality. Their model estimates that the current inspection system saves about 886 lives each year. Simply making inspection timing unpredictable, without conducting more inspections, could increase the number of lives saved by about 10%.
Why This Article Was Selected for The Policy Scientist
The broader importance of this research lies in the design of regulatory monitoring: inspections are used across health care, workplace safety, environmental regulation, food safety, and other settings where organizations may alter behavior when oversight becomes predictable. The question is especially timely as regulators face staffing and budget constraints that limit opportunities simply to conduct more inspections. Gandhi, Olenski, and Shi have developed substantial research agendas in health-care and nursing-home economics, and this study extends that work by examining inspection timing rather than inspection frequency alone. The administrative data are unusually rich, combining inspection histories, daily staffing records, patient assessments, and mortality outcomes. The fixed-effects analyses, robustness tests, and structural modeling provide substantial statistical leverage, although the observational design does not establish causality as cleanly as a randomized inspection schedule would. Future randomized or quasi-experimental research could strengthen causal inference and test whether the findings generalize to other regulatory systems and jurisdictions.
Full Citation and Link to Article
Gandhi, A., Olenski, A., & Shi, M. (forthcoming). Predictably unpredictable inspections. American Economic Review. American Economic Association.
Central Research Question
The study asks how the predictability of regulatory inspections affects the behavior of regulated organizations and, ultimately, the people whose welfare depends on that behavior. The empirical setting is U.S. nursing homes, which receive nominally unannounced inspections approximately once a year. Because inspections occur at fairly regular intervals, however, facilities can infer when inspection risk is low and when the next inspection is becoming increasingly likely. The central question is whether this predictable inspection cycle causes nursing homes to vary their effort over time and whether those changes affect patient health. A second question follows directly: could regulators improve outcomes by making inspection timing less predictable without increasing the number of inspections?
The study also examines a tradeoff inherent in inspection design. Inspections provide incentives because organizations want to avoid poor inspection outcomes, but they also provide regulators with information about underlying quality. More unpredictable inspections may generate stronger and more persistent incentives, yet irregular timing may leave regulators with somewhat less current information about facilities that go longer between inspections. The analysis therefore evaluates both the incentive and informational consequences of changing inspection schedules.
Previous Literature
The article builds on a broad literature showing that inspections and monitoring can alter organizational behavior. Previous research has studied inspection frequency, including Olken (2007), Levine et al. (2012), Blundell et al. (2020), Shi (2024), and Lin et al. (2025). Other work examines how inspections are targeted, including Duflo et al. (2018) and Johnson et al. (2023), while Jin and Leslie (2003), Dranove and Jin (2010), and Johnson et al. (2023) study the importance of information disclosure and monitoring salience. The distinguishing contribution here is the focus on predictability itself rather than simply how often inspections occur or which organizations receive them.
The study is also connected to theoretical work by Varas, Marinovic, and Skrzypacz (2020) and Ball and Knoepfle (2023), which identifies a fundamental tradeoff between unpredictable monitoring and the informational value of regularly timed inspections. Previous empirical work generally could not test this tradeoff because compliance is typically observed only when an inspection actually occurs. Zou (2021) is an important exception, using satellite pollution data to observe behavior between environmental inspections. The nursing-home setting allows the authors to extend this literature because staffing and patient outcomes can be measured continuously between inspections.
The article also contributes to the nursing-home economics literature. Prior studies have examined quality of care, staffing, ownership, reimbursement, access, and public quality ratings. Examples include Grabowski et al. (2008), Friedrich and Hackmann (2021), Einav et al. (2022), Grabowski et al. (2023), and Konetzka et al. (2021). More recent research by Chen and Dillender (2025) and Lin et al. (2025) begins to examine nursing-home inspections directly, but the timing of inspections over the entire inspection cycle has received substantially less empirical attention.
Data
The study combines several unusually detailed administrative datasets. Nursing-home inspection records come primarily from Centers for Medicare & Medicaid Services Survey Summary files covering 2006–2019. These records identify inspection dates, inspection types, and deficiencies cited by inspectors. The authors supplement these data with additional CMS records and Freedom of Information Act data on inspection start dates and follow-up revisits. Federal rules require nursing homes to receive an unannounced inspection at least every 15 months and roughly once annually on average. In practice, 74 percent of inspections occur between 40 and 60 weeks after the previous inspection, creating substantial predictability in inspection risk.
Facility effort is measured primarily using the Payroll-Based Journal, which contains daily staffing records for CMS-certified nursing homes during 2017–2019. These payroll-derived data provide hours worked by different categories of direct-care employees and are considered more reliable than self-reported staffing measures. The researchers combine staffing with additional care inputs, including the use of antipsychotic medications and physical restraints, to construct a composite index of facility effort. Total staffing, weekend staffing, full-time staffing, registered-nurse staffing, and lower antipsychotic use contribute most strongly to the index.
Patient information comes from Medicare administrative records. The Minimum Data Set provides detailed clinical assessments of nursing-home residents, while the Master Beneficiary Summary File supplies demographics, chronic conditions, enrollment information, and mortality. Medicare Provider Analysis and Review records provide information on prior hospitalizations. Together, these sources permit the authors to observe both nursing-home inputs and subsequent patient health, including mortality, at unusually high frequency. The breadth of the administrative data is important because the analysis depends on observing what happens during the long periods between inspections rather than only on inspection dates themselves.
Methods
The empirical analysis begins by examining how outcomes evolve week by week after the previous inspection. The principal specifications include nursing-home fixed effects, which account for stable differences across facilities, and calendar-week fixed effects, which absorb common changes over time. Patient-level models additionally control for predicted survival based on health characteristics measured before nursing-home admission. Standard errors are clustered at the facility level. The key coefficients therefore compare a facility with itself at different points in its inspection cycle rather than simply comparing different facilities with one another.
The researchers construct the effort index using factor analysis, treating observable staffing and treatment inputs as indicators of an underlying level of facility effort. They then estimate how this index changes as the probability of inspection rises. The empirical results show a nonlinear relationship: effort responds strongly when inspection risk begins rising but does not increase proportionately with each additional increase in risk. This concavity becomes important for the counterfactual analysis because a moderate inspection probability spread across the entire year can generate more total effort than concentrating most inspection risk in a shorter period.
The reduced-form analysis is supplemented with a dynamic structural model. Nursing homes are modeled as choosing costly effort in anticipation of future inspections, balancing current staffing and other costs against the benefits of performing well if inspected. The model also treats inspections as signals about underlying facility quality and allows that information to become less current as time passes. The estimated model is then used to simulate alternative inspection regimes, including more frequent inspections, completely unpredictable inspections, and more predictable schedules.
Findings/Size Effects
The central empirical finding is that nursing homes systematically vary their effort according to inspection risk. Immediately after an inspection, effort initially remains elevated because facilities may receive follow-up visits. Once that revisit period ends and the probability of another routine inspection becomes very low, effort falls. Beginning at roughly 35 weeks after the prior inspection, the probability of inspection rises sharply, and nursing homes correspondingly increase effort as the next inspection approaches.
Patient outcomes move in parallel with these changes. Mortality is higher during periods when inspection risk and facility effort are low and declines as facilities increase effort later in the cycle. The estimated relationship implies that a one-standard-deviation increase in nursing-home effort reduces patient mortality by approximately 5.4 percent. The authors also find related patterns in measures of patient functioning and pressure ulcers, providing evidence that the mortality relationship is not the only health outcome associated with the inspection cycle.
The structural model puts these effects into more tangible terms. Under the current inspection regime, the effort induced by the threat of inspections is estimated to save approximately 886 lives each year, equivalent to about 57 lives per 1,000 inspections. The existing inspection system also reduces regulators’ uncertainty about underlying nursing-home quality by approximately 56 percent.
Increasing inspection frequency by 25 percent is estimated to save approximately 25 percent more lives, but doing so would require substantially greater inspection resources. In contrast, keeping the number of inspections unchanged while making their timing unpredictable is estimated to increase the number of lives saved by about 10 percent. That mortality benefit is approximately equivalent to increasing inspection frequency by 10 percent under the existing scheduling system.
Unpredictability does impose an informational cost. Regulators would possess approximately 2.7 percent less information about facility quality in an average week because some facilities would go longer between inspections. However, the model also indicates that inspection frequency and unpredictability reinforce one another: increasing inspection frequency is approximately 9 percent more effective at saving lives when inspections are unpredictable. At the opposite extreme, perfectly predictable inspections would reduce the number of lives saved by 12.9 percent while increasing regulatory information by less than 1 percent.
Conclusion
The study concludes that the timing of regulatory oversight can be as important as its frequency. Nursing homes appear to respond strategically to predictable inspection schedules by reducing effort when inspection risk is remote and increasing it as an inspection becomes more likely. Because patient survival changes with these effort patterns, predictable inspection cycles have consequences beyond administrative compliance.
The larger contribution is to show that inspection policy involves more than deciding how many inspections to conduct. The distribution of inspection risk over time changes organizational incentives. A more unpredictable schedule can maintain a meaningful probability of inspection throughout the year and thereby sustain higher average effort without increasing the inspection budget. The model indicates that this approach could produce health gains comparable to a moderate increase in inspection frequency, although regulators would sacrifice a small amount of information about current facility quality. The results therefore identify a quantitatively important tradeoff between the incentive and information functions of inspections and provide a framework that may also be relevant to regulatory monitoring outside the nursing-home sector.
Comments