Migration behaviors leaving metropolitan areas: assessing the impacts of health risks and teleworking in the COVID-19 context

The study investigates whether and how health risks (COVID-19 infection rates) and teleworking influenced migration from metropolitan areas (MAs) to local areas in Japan during COVID-19, and how these effects changed over time. Motivated by observed increases in out-migration from large cities worldwide during the pandemic and the expectation that MAs face higher future health risks, the paper seeks empirical evidence on: (1) the impact and temporal dynamics of health risks on LMA migration; (2) the impact, temporal changes, and heterogeneity by occupation and employment status of teleworking on LMA migration. Japan provides a pertinent case with no mandatory lockdowns and high mobility, plus rich panel survey data enabling analysis of behavioral mechanisms under a major health crisis. The research is important for understanding adaptation to health crises and informing policy responses for future events.

Prior work documents increased outflows from metropolitan cores during COVID-19 across multiple countries (e.g., Japan, Spain, UK, Germany, Sweden, Norway, Australia). Explanations often emphasize economic factors (income, home ownership, living costs, job opportunities), social ties (birthplace), and place attributes (rural proximity to cities, second homes). Health risks and teleworking are frequently cited but seldom tested directly. Health risk studies often rely on proxies (density, amenities) rather than direct infection measures; a notable exception (Peng & Dai, 2024) finds infection rates negatively associated with LMA migration but lacks temporal analysis and aligns infection windows imperfectly with migration timing. Teleworking’s role is debated: ICTs may reduce migration via enhanced place attachment and reduced exploratory moves; some studies find no significant effect of teleworking on migration; others suggest telework could facilitate LMA migration during crises due to space needs and flexible work, though direct evidence is limited and often confounded with health risk considerations. Key gaps include: disentangling safety considerations from amenity narratives, measuring health risk and telework at appropriate temporal and individual levels, and exploring temporal dynamics and heterogeneity by occupation/employment status.

Design: Empirical analysis using an unbalanced individual-level panel from the Cabinet Office of Japan’s Surveys on Changes in Attitudes and Behaviors in Daily Life under the Influence of Novel Coronavirus Infection (CABC), waves 3–6. Data: Four survey waves with nationally representative stratified random samples: Wave 3 (Apr 30–May 11, 2021), Wave 4 (Sep 28–Oct 5, 2021), Wave 5 (Jun 1–9, 2022), Wave 6 (Mar 2–11, 2023). Over 10,000 valid responses per wave; more than 60% overlap across adjacent waves; 34% present in all four. Students excluded. Final sample: 37,170 observations from 19,306 individuals. Outcome: Migration (binary) = 1 if respondent reported having migrated from an MA to local areas in the last 6 months; 0 otherwise. Local areas defined as smaller MAs or non-MA areas (relative concept). Key predictors: - COVID: log of newly confirmed COVID-19 cases per 10,000 population in respondent’s residence prefecture over the 6 months preceding each survey (aligned with migration window). Range (log, scaled): 1.1236–7.3465. - Telework: individual teleworking frequency level (ordinal; 0=no telework; categories include occasional to almost-all-time remote), with 22.63% engaging in some telework. Controls: Female, AgeLevel, Married, University, HouseholdIncome, ChildUnder15, prefectural UnemploymentRate; employment status (FormalEmployee [ref], InformalEmployee, Manager, Self-employed, HomeWorker, Unemployed); occupation categories (IT professional, Office worker, Healthcare worker, Manufacturing/Construction/Mechanical technical, Store-based service, Non-store-based service). Models: - Main pooled entity- and time-fixed effects logit model (Model 1) with individual fixed effects and survey-wave fixed effects; odds ratios reported. - Wave-specific fixed-effects logit models (Model 2) estimated separately for each wave to assess temporal changes in COVID and Telework effects. - Moderation analyses: entity- and time-fixed effects logit models testing moderating effects of Telework on relationships between EmploymentStatus or Occupation and Migration. Interaction terms Group × centered Telework (c_Telework). Control variables included throughout. Notes: The COVID and Telework variables reflect conditions in the 6 months before the survey at the outcome location (current residence at survey time). While pre/post changes around the migration decision are not observed, the aligned measurement window supports inference about decision mechanisms.

- Health risk (COVID-19 infection rate): In pooled FE-logit (Model 1), higher prefectural infection rates are significantly associated with lower odds of LMA migration (OR=0.6344, p<0.001). Interpreted per e-fold increase in infection rate, odds of having migrated fall by about 37%. - Temporal dynamics (Model 2 by wave): COVID effect is significantly negative during waves 3–5 and reverses in wave 6: • Wave 3: OR=0.5743***; Wave 4: OR=0.4126***; Wave 5: OR=0.4601***; Wave 6: OR=4.8157***. Thus, during emergency/soon-after periods, lower infection risk attracted LMA migration; 17 months after the last emergency, higher-infection prefectures saw higher odds, consistent with shifting preferences as perceived threat fell. - Teleworking main effect: Not significant overall in pooled model (OR=1.0485, n.s.), indicating no long-term average effect on LMA migration when controlling for infection risk. By wave, telework ORs >1 but only wave 4 is marginally/significantly positive (OR=1.1504*), suggesting a short-lived facilitative effect in the later crisis stage. - Employment status effects (Table 5): Relative to reference, formal and informal employees are less likely to engage in LMA migration (multiple models ORs ~0.63–0.73, p<0.001), whereas self-employed, home workers, and unemployed are more likely (e.g., pooled models: Self-employed OR up to 2.90*** in Table 3; Unemployed OR=2.17***; HomeWorker OR=2.90***). Interpreted as opportunity-cost differences. - Telework moderation by employment status: Telework strengthens staying for formal employees (Group×c_Telework OR=0.8436*, Model 4a: each higher level of telework reduces odds of LMA migration by ~16% for formal employees) and strengthens leaving for self-employed (Group×c_Telework OR=1.3887***, Model 4d: each higher telework level raises odds by ~39%). - Occupations: Office workers and manufacturing/construction/mechanical technical professionals are less likely to migrate (significant negative Group effects). Telework does not significantly moderate occupation–migration relationships (all interactions n.s.). No significant main effect detected for IT professionals. - Descriptive migration rates: Share reporting LMA migration in prior 6 months increased across waves: 2.93% (wave 3), 3.61% (wave 4), 4.20% (wave 5), 4.65% (wave 6); overall 3.85% (1,430 of 37,170 observations). - Controls: Higher prefectural unemployment rate associated with lower LMA migration overall (OR=0.5725***), becoming more negative post-emergency (wave 6 OR=0.6031***). Females, higher age, and higher household income less likely to migrate; university education positively associated in most waves.

The findings directly address the research questions. First, health risks significantly shaped LMA migration: residents were more likely to leave MAs for lower-risk prefectures during and soon after public health emergencies, confirming safety-driven (health-risk-aversion) motives. This effect persisted several months after emergency measures ended, consistent with uncertainty and “deferred decisions,” and later reversed as perceived COVID-19 threat declined. Second, teleworking, measured at the individual level and net of infection risk, had no robust overall effect on LMA migration, countering narratives that telework alone drove urban outflows. Its effect appears time-sensitive—becoming positive only in the later crisis stage—and heterogeneous across employment statuses. Telework amplified the trade-off between economic opportunity costs and safety: for formal employees, higher telework levels were associated with staying in MAs (likely due to job retention and opportunity costs), while for self-employed individuals, telework increased the likelihood of moving to local areas (lower opportunity costs and flexibility). Occupation-specific analyses found limited roles for telework in moderating migration. Collectively, the results emphasize that in major health crises, safety concerns initially dominate migration decisions; as threat subsides, economic and amenity considerations regain prominence. The study’s alignment of infection-rate measurement with the migration window strengthens causal interpretation compared to prior work, offering actionable insights for crisis planning and regional policy.

This study contributes to the literature on health-crisis-led migration by: (1) providing robust panel-based evidence that lower infection risks attracted LMA migration in Japan during COVID-19, using infection rates aligned with the migration decision window; (2) showing that teleworking does not generally promote LMA migration when controlling for health risks, but exhibits time-sensitive and employment-status-specific effects—discouraging LMA moves among formal employees and encouraging them among the self-employed; (3) documenting temporal dynamics, including persistent negative infection-risk effects beyond emergency periods and a later reversal, consistent with deferred decision-making; and (4) highlighting the role of opportunity costs: self-employed, home workers, and unemployed were more likely to leave MAs, while employees were less likely. Policy implications include targeting self-employed migrants and entrepreneurship during crises, and investing in telework infrastructure as a long-term strategy for local areas. Future research should integrate measures of living costs (e.g., housing prices) and refine dynamic data to observe pre/post conditions around migration decisions, explore sector-specific mechanisms (e.g., IT professions), and examine lag structures of health risk and telework effects across different crises and institutional contexts.

The study does not control for living costs (e.g., housing prices) due to the lack of monthly, prefecture-level housing price data aligned with survey waves, leaving open the possibility that lower costs outside MAs also influenced LMA migration. Additionally, while key variables (infection rates, telework) are aligned to the 6-month migration window, the data do not observe pre- and post-decision changes at the individual level, limiting direct identification of within-person shifts around the move. The precise lag structure of the negative infection-rate effect cannot be established due to data constraints. Finally, occupation coverage is partial and some subgroup estimates (e.g., IT professionals) may warrant larger samples for more precise inference.