Non-coresident family as a driver of migration change in a crisis: the case of the COVID-19 pandemic

The COVID-19 pandemic decoupled residence from workplace for many, enabling relocation and altering U.S. domestic migration patterns. Observed deviations include increased moves from large cities to smaller, less populated places. The study asks whether non-coresident family, particularly parental ties, helped drive these changes: Did individuals relocate to be closer to kin during the crisis, and how does this relate to city size? The motivation stems from evidence that kin ties are more abundant in smaller cities and become especially salient during crises. The authors focus on parental family as a plausible and measurable pull factor given age- and life-course dependencies of migration and the pandemic’s unique loosening of geographic constraints on employment.

Three strands inform the inquiry. (1) Family ties perspective: Kin offer support, a need for proximity, and unique obligations; face-to-face interactions and practical support require geographic closeness. Parents and adult children exchange assistance far more than other ties; intergenerational informal care is substantial. (2) Crises and kin activation: During COVID-19, formal care closures heightened reliance on kin; studies show increased communication with non-coresident kin, strengthened family bonds, and greater practical help from children and parents; similar activation occurred in prior crises (Gulf War, Hurricane Andrew). (3) Tension with economic goals: Extended family proximity can conflict with economic aspirations; however, COVID-19 uniquely enabled mobility while increasing the salience of kin. Parental ties dominate support flows in Western kinship systems; many adults live within 30 miles of a parent/child. Migration propensity peaks at ages 18–44, overlapping with needs for intergenerational childcare and eldercare, making parental locations especially relevant to crisis-era migration decisions.

Design: The study tests a three-way relation among pandemic-era inter-city migration changes, city population size, and non-coresident family (parental). It combines Spectus mobile GPS county-to-county relocation data aggregated to CBSAs (Jan 2019–Dec 2020) with ACS PUMS microdata (2016–2021), plus ACS/BEA city-level covariates. Data: (1) Spectus provides weekly relocation indices between U.S. counties, with home locations identified via persistent nighttime GPS. The Relocation Index r_hk(t) equals devices in h with new home in k during week t divided by devices in h. Data are aggregated to CBSA-level flows R_ij(t) via county-to-CBSA mapping, using county populations and calibrated device sampling rates against ACS county flows (2016–2019). Period sums R_ij(θ) computed for baseline Apr–Dec 2019 (θ=0) and comparison Apr–Dec 2020 (θ=1). (2) ACS PUMS microdata (IPUMS USA) 2016–2021 provide person and household records (N≈18.7M person-records; N≈8.25M households across six annual samples), including migration in past year, state of birth, residence, and household composition. Additional city covariates from ACS 5-year estimates and BEA employment data. Derived measures: Probability of moving between city-size bins z(P′;P,θ) computed from binned log-populations using migration flows (Eq. 4). Net in-migration ratio per city y_i(t) = (Σ_{j≠i} R_{ij}(t)) / (Σ_{j≠i} R_{ji}(t)) and period analogue y_i(θ) (Eqs. 7–8); change measured by y_i(θ=1)/y_i(θ=0). Investigation 1 (micro-level return-to-home): Three move types from PUMS: Type 1: individual movers who moved into parents’ household (identified via MOMLOC/POPLOC and parent non-mover status); Type 2: individual movers returning to state of birth from another state, excluding Type 1; Type 3: family household units (all members moved) returning to a householder’s state of birth. Annual rates λ_m(t) computed using sampling weights: for m=1,2, weighted share among individual movers; for m=3, weighted share among family household movers (Eqs. 5–6). 2020 weights are experimental due to COVID-19 data issues. Investigation 2 (city-level parental availability and migration): Construct proxy v_i for parental family availability as the share of households likely to be parents of adult children, based on 2019 PUMS: qualifying households include (1) family households where head or spouse is ≥50 years old; or (2) non-family households with head ≥50 and married-spouse absent, separated, divorced, or widowed. Household weights (HHWT) summed within PUMAs and allocated to CBSAs using Geocorr PUMA–CBSA allocation factors; v_i equals allocated qualifying households divided by total weighted households per CBSA. Relate v_i to city log-population, to McLeod et al.’s φ_i (general family availability; via modal regression with KDE), and to migration changes y_i. Investigation 3 (empirical model): Difference-in-differences with continuous treatment v_i to explain ln[y_i(1)/y_i(0)] = β v_i + γ + Σ_a C_{ai} (Eq. 9), where γ is the time fixed effect and the ratio structure provides city fixed effects. Controls reflect relocation drivers: log population size, log population density, log median home value, log median income, log employment level (jobs per person), and share of single-family homes (SFH; level). Robustness across model specifications reported; standard errors provided. Binning choices (e.g., 10 population bins) explored; consistent qualitative patterns for b in 5–10.

- Flows from large to small cities increased after COVID-19 onset. Movers from large cities were more likely to relocate to small cities and less likely to move to other large cities in 2020 relative to 2019. - Quantitatively, cities with population <500,000 saw an excess influx of nearly 52,000 movers from cities >500,000 relative to 2019 during Apr–Dec 2020. Net in-migration to these smaller cities rose by 80% (from 60,000 to 100,000), with excess inflow from large cities accounting for ~95% of this increase. The 10 largest cities’ net out-migration roughly doubled (−82,000 vs −45,600). - Micro-level return-to-home dynamics: - Type 1 (individuals moving in with parents): λ1 spiked in 2020, then returned to pre-pandemic levels in 2021. Demographics: younger (median age 25), lower mean income ($25,266), about 60% employed; income and education among movers higher in 2020–2021 than prior years. - Type 2 (individuals returning to native state without joining parents): λ2 decreased in 2020 but increased in 2021; movers older (median age 29) with higher mean income; income and education also higher in 2020–2021. - Type 3 (family households returning to native state): λ3 increased in 2020 and more strongly in 2021. 2021 family movers skewed toward slightly older householders and teens as eldest children, hinting at eldercare relevance. - Parental availability proxy v_i validation and patterns: - v_i monotonically increases with McLeod et al.’s φ_i (general family availability) via modal regression (KDE), supporting v_i as a kin-availability proxy. - v_i declines with city population size (higher parental availability in smaller cities). - Cities in higher v_i quintiles experienced proportionally larger increases in net inflow per outflow y_i(t) after April 2020 than a year prior; bottom quintile did not. - Binned analysis shows log[y_i(1)/y_i(0)] increases with v_i; for higher v_i, the ratio exceeds zero, indicating larger inflow per outflow during the pandemic. - Empirical model: Across specifications, β on v_i is positive and statistically significant (e.g., Model (4): β=0.631, SE=0.066, p<0.01). Interpretation: a city with v_i 10 percentage points higher than another saw a 6.5% larger positive change in inflow per outflow after the pandemic began. Controls behave as expected: higher population and higher median home values associate with smaller net influx changes; greater share of single-family homes associates positively.

Findings indicate that kin ties—proxied by parental family availability—contributed to the pandemic-era shift toward smaller cities. This mechanism complements established drivers such as population density and housing costs but adds a critical social dimension: proximity to extended family influenced destination choices when the crisis heightened support needs and remote work loosened geographic constraints. The results reconcile the increased out-migration from large cities with the spatial distribution of kin (more abundant in smaller places), and align with theory on trade-offs between economic aspirations and family proximity. Implications span migration modeling (e.g., incorporating kin/place-of-nativity into gravity-type models), urban vulnerability (heterogeneous kin distribution as a latent factor in population realignments), and potential epidemiological consequences in receiving locales where intensified family face-to-face contact may have affected later COVID-19 outcomes. Policies that bolster institutional support for individuals lacking nearby kin could mitigate asymmetric impacts on cities during future disruptions.

The study provides multi-source empirical evidence that elevated migration to smaller U.S. cities during COVID-19 was partly driven by moves toward non-coresident family, especially parents, intersecting with the uneven spatial distribution of kin. Contributions include: documenting strengthened return-to-home patterns at the micro level; constructing a city-level proxy for parental availability that correlates with known family availability measures and declines with city size; and demonstrating, via a DiD framework, that higher parental availability predicts larger positive changes in net inflow per outflow during the pandemic, even after controlling for key urban attributes. Future research should develop richer, privacy-preserving data linking extended kin networks to geography, extend analysis beyond parental ties to broader kinship, and examine long-run impacts on labor, housing, and industry location. Integrating kinship into migration models could improve forecasting under crisis conditions.

Key limitations include reliance on proxy measures for family location due to the absence of large-scale, geocoded data on non-coresident kin. The parental availability proxy v_i is inferred from age and marital status and allocated from PUMA to CBSA, introducing measurement error. PUMS geography constraints (no precise parental addresses) and use of state of birth as a proxy for family ties may misclassify some moves. 2020 ACS PUMS employs experimental weights due to COVID-19 data collection issues. Spectus mobile data, while timely, requires calibration and is not public; device sampling biases may remain. The analysis centers on parental ties and may understate effects from other extended kin. Despite multiple levels of analysis and controls, unobserved confounders may persist.