Discovering the long-term effects of COVID-19 on jobs-housing relocation

The study investigates how the COVID-19 pandemic affected intra-urban jobs-housing relocation behaviours in Beijing, addressing long-term changes in residential moves, workplace moves, and commuting outcomes. The research question centers on whether and how COVID-19 altered relocation patterns relative to pre-pandemic baselines, and whether these changes persist over several years. The rationale is that workplace and residential mobility affect employment distribution, transportation, and urban restructuring. The authors posit that pandemic-driven stressors (infection risk, lockdowns) and new preferences (health, teleworking, accessibility needs) could trigger relocations, leading to measurable shifts in home and work locations and commute times. Hypotheses include: H1a (inward home shifts due to risk-mitigation via shorter trips) vs H1b (outward home shifts to lower-density areas), H2 (decentralization of workplaces), and H3a vs H3b (improvement vs deterioration of jobs-housing balance measured via commute time). The study aims to separate pre/post-pandemic patterns using a long-term event study and large-scale mobile data.

Prior work has documented pandemic-related changes in daily travel (reduced commuting and leisure, growth in essential trips; modal shifts away from public transport and shared modes toward private car use; risk perceptions guiding mode choice). Urban activity intensity dropped in workplaces and increased in residential areas; teleworking and online shopping rose and may remain elevated in the 'new normal'. Job relocation was affected by lockdowns, job losses, sectoral reallocations, telework experiences, and shifting occupational priorities. Residential mobility motives expanded to include income shocks, psychological needs, caregiving, better living conditions under lockdown, and preferences for lower-density 'COVID-proof' environments. Socioeconomic factors shape mobility inequalities: women, older adults, and young people experienced different risk aversion and vulnerabilities; low-income and vulnerable groups faced greater displacement and reemployment challenges. Other strands include inter-city mobility changes, links between mobility and COVID-19 spread, and effects of non-pharmaceutical interventions. Gaps identified: most studies examine job or housing moves and commuting in isolation rather than as interdependent; few long-term comparisons distinguishing pre- and post-pandemic patterns; many rely on surveys with potential bias and smaller samples. A related mobile-data study examined dwell/work intensity trends but not individual-level location changes and commuting ties.

Research design: A citywide event study using 10 months of cellular signalling snapshots across 4.5 years (Apr 2018, Nov 2018, Apr 2019, Nov 2019, Apr 2020, Nov 2020, Apr 2021, Nov 2021, Apr 2022, Nov 2022). Time is divided into nine semi-annual phases (−3, −2, −1 pre-pandemic; 0–5 post-pandemic). For each phase, home and work locations are detected for users within Beijing’s 6th ring road. Relocators are identified as home movers, work movers, or home-and-work movers (including regular commuters, used-to commuters who stopped commuting, and about-to commuters who began commuting). Key outcomes: changes in ring road position of home (Δr^h) and work (Δr^w), housing price (Δp), population density at home (Δdens), accessibility to public transit (Δpt), schools (Δkind, Δpri, Δmid), healthcare (Δmed), green spaces (Δgreen), and commute time (Δt). Research hypotheses draw on stress-response and dissatisfaction models: COVID-19 as a push-pull reallocation shock affecting relocation choices. Empirical models: Event-study-based velocity models estimate cumulative post-pandemic deviation from pre-pandemic average change (six specifications J=0–5), and acceleration models estimate phase-specific deviations across all phases. Controls include sociodemographics (gender, age, affluence index) and pre-relocation jobs-housing characteristics: pre-relocation commute time, ring positions of home and work, housing price, population density, and accessibility metrics. Seasonal fixed effects are included. Income subgroups are defined by affluence index (middle-income <6; high-income ≥6). Data: Cellular signalling data from a major Chinese operator cover ~2 million users per month, with timestamps, triangulated locations, gender, age, and an affluence index (derived from online/offline activity). Spatial datasets include uniform 1.5 km grids within the 6th ring, POI-based accessibility indicators (0/1 presence of transit stations, schools, medical facilities), green space area shares, housing prices (Lianjia resale prices, Dec 2021), and population density (calibrated with Nov 2019 signalling data and the 2020 Census). Commute times between grid centroids are estimated via Baidu RouteMatrix API (driving, shortest route) for evening peak (17:00–19:00) in Mar 2023. Policy stringency is captured by the Oxford COVID-19 Government Response Tracker; infection counts sourced from Beijing Municipal Health Commission. Sampling and processing: For each phase, continuous core users aged 25–54 present in both start and end months are filtered; home/work detection uses stay-time within working (9:00–17:00 workdays) and resting (21:00–5:00) windows; locations within 1.5 km are considered the same; only users with home/work within the 6th ring are retained. Regular commuters, used-to, and about-to commuters are identified and classified by relocation type. Affluence index missingness is imputed via a Naive Bayes model using gender, age, monthly fee, housing price, and ring positions. Coordinates are mapped to grid IDs and joined to grid attributes and OD commute-time matrices. Average phase sample size is 363,546; pooled final sample includes 3,271,918 users. Analyses also apply seasonal fixed effects and robustness checks (models without season FE, weighted models by age-gender census margins, and Poisson regressions).

Home relocation (n=480,611 home relocators): An immediate outward shift occurred at the outbreak (Phase 0 Δr^h≈0.22), likely reflecting temporary moves to suburban/second homes. Thereafter (Phases 1–5), overall Δr^h settled to a mild outward trend (~0.05), near pre-pandemic levels. Income stratification reveals structural change: middle-income relocators showed a strong pre-pandemic suburbanization that significantly slowed post-pandemic (post-pandemic deviation M_Δr ≈ −0.11), indicating inward relative shifts and greater pursuit of accessibility (notably increases in housing price Δp, middle school Δmid, and healthcare Δmed access; p<0.001). High-income relocators had negligible suburbanization pre-pandemic but developed a mild post-pandemic suburbanization (M_Δr ≈ +0.06), with associated decreases in accessibility metrics due to outward moves. Thus, H1a holds for middle-income (inward relative shift), and H1b holds for high-income (outward relative shift). Work relocation (n=1,080,683 work relocators): A sharp initial outward scatter in Phase 0 (Δr^w=0.42, p<0.001), followed by partial reversion around Phase 2 and then renewed outward movement. Pre-pandemic Δr^w≈0.08 indicated mild decentralization; as of Phase 5, Δr^w≈0.15 remains significantly above pre-pandemic (p<0.001). Patterns are similar for middle- and high-income groups. H2 holds: employment decentralization accelerated and persisted. Jobs-housing relation (n=1,387,418 relocators): Commute time changes indicate improved jobs-housing balance. Pre-pandemic mean Δt≈−27 s/half-year (slight inverse separation). Post-pandemic, Δt reached about −424 s (≈7 min) by Phase 5 (p<0.001), implying an acceleration of inverse separation by ≈397 s per half-year. H3a holds across groups, with pronounced long-term effects among middle-income home movers and high-income work movers. A back-of-envelope estimate suggests that, absent the pandemic, commuting-related CO2 emissions would have been ≈3.03×10^4 t higher over Nov 2019–Nov 2022 (assuming 20% relocators of the working population per half-year and ~150 g CO2 per vehicle-minute).

Findings support that COVID-19 induced lasting changes in intra-urban relocation and commuting. Middle-income relocators shifted towards more central or accessible areas, likely to reduce infection risk during daily activities and improve access to schools and healthcare; high-income relocators increasingly pursued lower-density suburban living. This compositional change may preserve inner-city vitality by retaining or attracting younger, middle-income families while older/wealthier households move outward. Employment locations decentralized more strongly and persistently than residences, consistent with telework expansion and labor market reallocation, weakening the monocentric structure. Most importantly, combined home and work shifts improved jobs-housing balance, reducing commute times and potentially urban congestion and emissions. These outcomes directly address the research question by demonstrating measurable, persistent post-pandemic deviations from pre-pandemic relocation and commuting patterns, with policy-relevant implications: supporting emerging subcenters and suburban services, enhancing central-area accessibility and walkability for families, and leveraging improved jobs-housing balance to reduce transport emissions.

Using a multi-year event study with large-scale mobile signalling data, the study shows that COVID-19 produced durable shifts in jobs-housing dynamics: slowed suburbanization among middle-income households, increased suburbanization among high-income households, accelerated decentralization of employment, and improved overall jobs-housing balance (shorter commutes). These changes may make cities more vibrant at the center, more polycentric in employment, and greener due to reduced commute times. Planners should capitalize on these trends to guide structural improvements and sustainability—strengthening services and walkability in central areas, planning for infrastructure in emerging subcenters, and aligning land use-transport policies with new commuting patterns. Future research should extend analyses to low-income and temporary populations, assess generalizability beyond Beijing, incorporate actual modal shifts into commute metrics, and disentangle pandemic effects from concurrent policies and macro shocks.

- Potential confounders: Beijing’s limited-competition housing policy (2018–2019) likely drove middle-/low-income households to near suburbs until ~2021, possibly overstating the pandemic’s role in slowing middle-income suburbanization; the Russia–Ukraine conflict (from Feb 2022) further strained labor markets, possibly overstating work relocation effects. - Sample representativeness: Mobile users with socioeconomic data skew younger, wealthier, and more digitally active; strict filters excluded lower-SES and temporary populations. Even with age-gender weighting, the sample remains wealthier, so results mainly reflect middle-to-high-income workers. - Generalizability: Beijing’s unusually strict and prolonged controls (Jan 2020–Dec 2022) may limit applicability to other cities/countries with different policy regimes. - Causal mechanisms: Motives are inferred from observed characteristics and context rather than directly measured. - Commute time metric: Estimated for private car only; ignores modal changes during the period, potentially biasing Δt for non-car users.