Who and which regions are at high risk of returning to poverty during the COVID-19 pandemic?

Crises such as climate-related disasters, conflicts, and pandemics cause differential socioeconomic impacts with low-income groups suffering most. COVID-19 disrupted economic and social systems globally, threatening to push many recently lifted-out-of-poverty households back below poverty lines through direct health impacts and indirect income losses due to mobility restrictions and job disruption. Hubei Province, China—epicenter in early 2020 and a key region for poverty reduction—offers a critical context to examine which households and regions are most at risk of returning to poverty during lockdown-induced income shocks. This study focuses on households lifted out of poverty by end-2019, aiming to identify who is at high risk, where they are located, and why, across scenarios reflecting 1-, 2-, and 3-month lockdowns that primarily affect migrant workers’ wage income.

The study situates its analysis within research on COVID-19’s socioeconomic effects, highlighting forecasts of increased global poverty and mechanisms linking non-pharmaceutical interventions (NPIs) to income loss among vulnerable groups (e.g., migrant and informal workers). It reviews evidence on lockdowns’ impacts on employment, mobility, and livelihoods, and on heterogeneous vulnerability during recessions, which often identifies remote rural areas as more vulnerable. Contrasting with prior work, the study finds proximity to railway stations—typically associated with development—as a vulnerability factor during the early COVID-19 lockdown, due to higher reliance on migrant wage income. The paper also references work on policy responses (e.g., transport facilitation for workers, enterprise support), and on the continuing roles of NPIs and vaccination in mitigating pandemic impacts.

• Study design and data: Two field surveys (January and September 2020) in 10 counties across eastern, central, and western Hubei Province covering 100 towns. Initial dataset of 91,125 registered poor households (end-2019) with total income and income structure; after cleaning (missing/outliers), final analytical sample of 78,931 households. Household characteristics from county poverty alleviation offices; detailed non-income characteristics available for Yingshan County (n=17,972).
• Poverty threshold and return-to-poverty definition: Annual household income < 5000 CNY (Hubei standard). Households lifted out of poverty by end-2019 were assessed for risk of returning below this line under lockdown scenarios.
• Lockdown scenarios: 1-, 2-, and 3-month inability of migrant workers to return to work, implying proportional losses of wage income. For each household with W total wage income over n working months in 2019, wage income under scenario i is w_i = W*(n−i)/n for i=1,2,3. Household’s total income recalculated; if <5000 CNY, marked as returned to poverty. Town-level risk R computed as P/M (households returning to poverty divided by total poor households).
• Household-level risk factors: 13 indicators: income diversity (Shannon-Weiner index), proportion of wage income, migrant working hours, family size, head’s gender, fruit crop area, path type to house, grain-to-green area, woodland area, irrigation area, distance to village main road, power supply for production (binary), supported by leading enterprises (binary).
• Statistical analysis (household level): Two logistic regression models. Model 1 (all counties): income characteristics (proportion of wage income, income diversity). Model 2 (Yingshan County): 11 additional household characteristics. Outcome: returned to poverty (yes/no). Reported odds ratios with 95% CIs (Supplementary Table S1; Fig. 3a,b).
• Town-level risk factors and importance: Potential regional variables drawn from remote sensing, internet, and statistics (e.g., distances to transport hubs, land-use per capita, sectoral employment proxies). Variable importance assessed via LMG metric in multiple linear regression (relaimpo R package), with PMVD as robustness (Supplementary Tables S4–S6). Distance to nearest railway station emerged as most important under all scenarios.
• Spatial risk mapping (town level): Generalized Additive Model (GAM) with beta family link for proportion outcomes. Steps: (A) variable selection with variance inflation factor to remove multicollinearity (selected variables in Table S3); (B) choose beta family link; (C) fit full model then remove variables with p>0.01; (D) evaluate residuals and select model by adjusted R² maximization and AIC minimization (Tables S7–S9). Final GAM applied to all towns to predict risk under each scenario (Fig. 4b).

• Overall risk increase: Under the 3-month lockdown scenario, the percentage of households at risk of returning to poverty rose from 5.6% to 22% across the sample of 78,931 households.
• Household-level determinants:
  • Proportion of wage income strongly and consistently associated with higher risk across all scenarios.
  • Income diversification and working hours protective: a one-unit increase in income diversity (Shannon index) and in working hours decreased the likelihood of returning to poverty by 15% and 32%, respectively.
  • Support and infrastructure protective: households supported by leading enterprises and those with power supply had 41% and 22% lower likelihood of returning to poverty, respectively.
  • High-risk profile: single income source, shorter working hours, larger family size, no support by leading enterprises, and no power supply.
• Spatial/regional findings:
  • Towns at high risk (≥2% of households returning to poverty) doubled from 27.3% to 46.9% under a 3-month lockdown.
  • Transport accessibility dominant: Distance to nearest railway station was the most important factor (LMG), with a negative linear relationship to risk under all scenarios. On average, each 10 km decrease in distance to the nearest railway station increased the share of households returning to poverty by about 2%.
  • Spatial pattern: High-risk towns clustered within railway station buffer zones; low-risk towns tended to be far from stations. Some near-station areas with more balanced development showed lower vulnerability (quadrant LN in Fig. 5).
• Quantified station-distance effects: An average decrease of 10–50 km in distance to the nearest railway station increased risk from 1.8% to 9%, emphasizing transport-proximity-related vulnerability during lockdowns.

The study directly addresses who and where is at high risk of returning to poverty during early COVID-19 lockdowns by linking income dependence, employment disruption, and spatial accessibility to transport. Findings highlight that households reliant on wage income, with limited diversification and fewer working hours, were most vulnerable to mobility restrictions, while support from leading enterprises and access to power reduced risk. Regionally, towns closer to railway stations—areas typically benefiting from development—were paradoxically more vulnerable during lockdown due to higher dependence on migrant wage income disrupted by travel bans. These insights inform rapid, targeted policies in early stages of shocks: prioritize support to households with high wage-income dependence, facilitate safe resumption of migrant work (e.g., chartered transport), strengthen village-level industrial support, and improve resilience through diversification and infrastructure. The results also align with evolving pandemic management, where combined NPIs and vaccination shape recovery and labor mobility; premature relaxation could sustain poverty risks by delaying enterprise and transport recovery. Overall, the work refines understanding of spatial-economic vulnerability under pandemic controls and guides targeted interventions to mitigate poverty reversals.

This study quantifies the risk of returning to poverty among previously lifted-out-of-poverty households in Hubei during COVID-19 lockdowns, identifying key household and regional risk factors. Under a 3-month lockdown, risk rose markedly (5.6% to 22%), with wage-income dependence, low diversification, and proximity to railway stations driving vulnerability. The integration of logistic regression, LMG-based importance analysis, and GAM-based spatial prediction provides actionable maps and determinants for targeted policy. Future research should: (1) analyze later pandemic stages as vaccination and policies evolve; (2) broaden samples beyond registered poor to include marginalized and newly poor groups; (3) assess multidimensional poverty impacts (housing, education, healthcare); and (4) examine compound risks from concurrent shocks (e.g., natural disasters, conflicts).

• Timeframe limited to a 3-month lockdown; longer-term and post-lockdown effects were not assessed.
• Focus on households previously lifted out of poverty; newly poor households not captured by the pre-pandemic poverty line were not included.
• Primary emphasis on income loss pathways; other poverty dimensions (e.g., housing, education, healthcare access) and broader environmental/economic impacts of COVID-19 control measures were not fully incorporated.