Sebastian Olascoaga

City of Boston · MIT

Published Work·Housing & public health

Quasi-experiment · 2021

When states ended their eviction bans, what happened to COVID-19?

Sebastian Sandoval-Olascoaga · Atheendar S. Venkataramani · Mariana C. Arcaya

Published · JAMA Network Open

Twelve weeks after a state let its eviction moratorium expire, the risk of a COVID-19 diagnosis was 83% higher than in states that kept the moratorium in place. The hazard was more than twice as high for people with preexisting health conditions, those living in low-income neighborhoods, and those in rent-burdened communities. Restricting evictions, in those months, also restricted COVID-19 transmission.

The same moratorium lift produces four hazard trajectories, stratified by comorbidity.

Hazard ratio of a first COVID-19 diagnosis after a state lifted its eviction moratorium, stratified by individuals' Charlson Comorbidity Index (a standard health-vulnerability score). Each curve compares lift states to non-lift states over 27 weeks centered on the expiration. Higher baseline morbidity tracks with steeper post-expiration hazard curves.

Small multiples event study showing four hazard-ratio panels by Charlson Comorbidity Index strata, with the post-expiration curve rising more steeply at higher comorbidity scores HR 4.5 3.5 2.5 1.5 1.0 0.5 CCI = 0 Generally healthy HR 1.55 −15 0 +12 CCI = 1 One condition HR 1.93 −15 0 +12 CCI = 2 Two conditions HR 2.44 −15 0 +12 CCI ≥ 3 The medically vulnerable HR 2.36 at week 12 −15 0 +12 Weeks since the state lifted its eviction moratorium · vertical dashed line marks week 0

Source: Sandoval-Olascoaga, Venkataramani & Arcaya, JAMA Network Open (2021) · OptumLabs Data Warehouse · Cox proportional-hazards difference-in-differences, n=509,694 individuals · 95% confidence interval shown · Pre-trends joint test p=0.98 · Each panel uses the same log-scale Y axis (HR 0.5 to 4.5) and same week window

509,694

Individuals in the OptumLabs cohort across 43 states + DC

HR 1.83

83% higher hazard of COVID-19 diagnosis at week 12 after expiration

HR 2.31

In high-rent-burden ZIPs. More than 2× the hazard of non-lift states

Why it matters

COVID-19 transmission depended on who could afford to stay home.

By mid-2020, 43 states and DC had issued eviction moratoria, a housing intervention with no obvious health purpose at the time. Then 26 of them let those moratoria expire while the pandemic was still active. That divergence in policy across otherwise similar states and weeks is the kind of natural experiment policy researchers usually only get to imagine.

If lifting the moratorium drove evictions, and evictions drove movement, doubling-up, and crowding, transmission would follow. We test whether that signal appears in administrative health data on diagnosed cases, not in models or projections.

How we did it

The first individual-level study of eviction policy and COVID-19.

We built an individual-week panel of 509,694 commercially insured and Medicare Advantage beneficiaries from the OptumLabs Data Warehouse, all living in the 43 states (plus DC) that had implemented a moratorium between March and September 2020. The 26 that lifted form the treatment arm. The 18 that maintained form the control. The cohort balances 254,847 first COVID-19 diagnoses (ICD-10 U07.1) with the same number of matched non-diagnosed individuals.

The model is a Cox proportional-hazards difference-in-differences with time-dependent covariates, in an event-time framework. The week of the state's moratorium lift sets time zero. We control for concurrent non-pharmaceutical interventions (mask mandates, stay-at-home orders, business reopening, school closures), county and state case and test counts, sex, age, insurance type, industry, prior ICD-10 Z-codes flagging socioeconomic distress, and ZIP-code poverty. Pre-trends pass the parallel-trends test cleanly (χ²₁₄ = 5.35, p = .98). A 2% random-sample sensitivity check confirms the balanced design isn't driving the result.

What we found

The effect emerged five weeks after the lift and grew thereafter.

Timing. The effect was invisible at weeks 1 to 4. By week 5, lift-state residents had a hazard ratio of 1.39 (95% CI, 1.11 to 1.76; p = .004). By week 12, HR 1.83 (95% CI, 1.36 to 2.46; p < .001). An 83% increase relative to states that kept moratoria. The dynamic shape matters. This looks like compounding transmission, not a one-time shock.

The equity gradient. The strongest effects concentrated where they were already hardest to absorb. Among people with three or more comorbidities (Charlson ≥3), HR 2.36 (1.67 to 3.36). In high-rent-burden ZIPs, HR 2.31 (1.64 to 3.26). In high-poverty ZIPs, HR 2.14 (1.51 to 3.05). Only the healthiest stratum (CCI = 0) plateaued after week 4.

The mechanism, revealed by a null. Was the higher risk because those individuals got evicted? No. We saw no association between moratorium expiration and individuals' own ZIP-code changes, a proxy for being personally displaced. The mechanism was community-level spillover. Renewed evictions in a community changed transmission dynamics for everyone living in it.

Caveats

What we couldn't see.

The OptumLabs cohort excludes Medicaid and uninsured Americans. Those are the populations facing the highest eviction and COVID-19 risk. By construction this biases our estimates toward the null. The published hazard ratios are a lower bound on the true effect. We also lack data on beneficiary race and ethnicity, so the paper cannot speak directly to racial disparities, a meaningful gap given the racialized geography of eviction in the United States.

In the press

Cite this paper

Sandoval-Olascoaga, S., Venkataramani, A. S., & Arcaya, M. C. (2021). Eviction moratoria expiration and COVID-19 infection risk across strata of health and socioeconomic status in the United States. JAMA Network Open, 4(8), e2129041. doi:10.1001/jamanetworkopen.2021.29041

Related workHousing · Public health

2024 · Paper

When buyers see a home's flood risk, where do they go instead?

NBER w33119 · R&R, American Economic Review

2025 · Paper

What happens to property values in flood-exposed neighborhoods?

Frontiers in Environmental Economics

2023 · Paper

Blocks exposed to frequent flooding flip from +5.76% growth to −1.42% loss.

Nature Communications