The study, led by Tao Liu of Xi’an University of Architecture & Technology Huaqing College together with colleagues at Xi’an University of Architecture and Technology, surveyed households across three flood-prone counties: Gaoxian, Jiajiang, and Yuechi. These are not abstract case studies. Jiajiang County endured the largest flood since hydrological records began in 1955 during the “8.18” flood of 2020, which affected more than 118,000 residents. Yuechi County suffered rainstorm floods in July 2014 that touched all 44 of its towns and affected 460,000 people. Gaoxian saw flooding in 2019 that damaged over 6,100 hectares of crops. In short, the households in this sample live with flood risk as a recurring fact of life, not a hypothetical one.

The researchers structured their inquiry around a well-established framework in livelihood research. Livelihood resilience — a household’s ability to absorb disturbances and keep functioning — was decomposed into three capacities: buffering ability, meaning the land, labor, assets, and income a family can draw on when livelihood is disrupted; self-organization ability, meaning the capacity to mobilize collective action, community resources, and reciprocal relationships; and learning ability, meaning the knowledge and skills that widen a household’s options when conditions change. On the risk side, the team measured four dimensions: health, environmental, financial, and social risk, each built from multiple survey indicators.

The data came from face-to-face interviews conducted in July 2021 across nine towns and twenty-seven villages, using a stratified, multi-stage random sampling design. Respondents averaged 58.5 years of age with roughly 6.7 years of schooling, and 70 percent reported that weather had damaged their crops. The researchers weighted indicators using the coefficient of variation method, an objective statistical approach that assigns greater weight to indicators showing more variation across households, and then modeled the relationship between risk scores and resilience scores using fractional logit regression with county fixed effects. Because the dependent variables were proportions bounded between zero and one, this model class avoids the distortions that ordinary linear regression can introduce at the boundaries.

The headline result is stark. Social risk was negatively associated with all three component capacities and with overall livelihood resilience, with statistical significance at the 1 percent or 5 percent level across the adjusted models. In practical terms, a one-standard-deviation increase in the social risk index was associated with measurable declines in buffering, self-organization, learning, and overall resilience. The authors are careful to stress that these are cross-sectional associations, not causal effects — the data capture a single moment in time, so the direction of the relationship cannot be pinned down. But the consistency of the pattern across every outcome the team tested is difficult to dismiss.

Health risk told a more partial story. It was negatively associated with self-organization capacity at the 10 percent significance level and with overall livelihood resilience at the 5 percent level, but showed no statistically reliable association with buffering or learning capacity. This aligns with a well-documented mechanism: serious illness drains labor capacity, generates medical expenses, and shifts caregiving burdens onto healthy family members, undermining the social and organizational dimensions of a household’s livelihood without necessarily depleting its physical assets. The authors interpret this as limited, outcome-specific support for their hypothesis rather than a general relationship.

Perhaps the most counterintuitive finding concerns money. Financial risk appeared positively associated with overall livelihood resilience in the baseline model — a result that seems, on its face, to suggest that financial strain somehow strengthens households. The robustness checks dismantled that interpretation. When the team recalculated the financial risk index excluding loan-amount indicators, the positive association vanished. The explanation lies in what loan amounts actually measure: a household carrying substantial formal and informal borrowing may be under debt pressure, or it may simply have good access to credit because lenders trust it. Those two channels push resilience in opposite directions, and the index could not separate them. The authors explicitly warn against reading the result as encouragement for more borrowing.

Environmental risk — the category covering floods, extreme weather, pollution, soil erosion, and crop pests — showed no statistically significant association with any resilience outcome. This null result is striking given the study area’s flood history, and the authors resist over-explaining it. They note that possible contributors, such as survey timing, limited variation in reported exposure, or the breadth of the perception-based environmental index, were not directly tested. They also point to disaster research showing that household welfare often recovers over time as assets are rebuilt and assistance arrives, which can flatten the apparent cross-sectional relationship between hazard exposure and resilience. The finding should be read as a null association within this sample and measurement framework, not as evidence that environmental hazards do not matter.

Methodologically, the study took unusual care with inference. Standard errors were clustered at the village level across the twenty-seven villages, and significance was assessed using wild cluster bootstrap procedures with 1,000 replications — a technique designed for settings where the number of clusters is too small for conventional clustered standard errors to be trustworthy. The team also ran robustness checks replacing the fractional logit with ordinary least squares, switching to equal weighting for the indices, and excluding potentially overlapping indicators. Social and health risk results held up across these variations; the financial result did not. The authors also acknowledged a conceptual wrinkle: some social-risk indicators sit close to the self-organization measures, which is precisely why they frame their policy suggestions as priorities for further local assessment rather than prescriptions.

For policymakers in flood-exposed rural regions, the practical takeaway is a reordering of priorities. The strongest signal points toward strengthening social support structures — identifying households with weak family or community ties, limited access to formal assistance, or low trust in local channels, and reviewing whether existing aid is actually known, reachable, and available when floods strike. The weaker health signal supports a cautious review of village health services and the needs of families facing serious illness. The financial findings argue for recording loan use, repayment difficulty, and default risk separately before proposing credit policies. And the environmental null result, paradoxically, is a call for more evidence rather than less action: local hazard exposure and program performance need direct study before subsidies or preparedness programs can be justified by this data. What the study ultimately demonstrates is that livelihood risk is not one number. Treating it as a single aggregate score, the authors conclude, conceals patterns that matter — and in the villages of Sichuan, the pattern that matters most is social.

Subject of Research: Associations between multidimensional livelihood risks and livelihood resilience among rural households in Sichuan Province, China

Article Title: Livelihood risks and resilience of rural households: evidence from Sichuan Province, China

Article References: Liu, T., Zhao, T., Tian, J., Zhao, S., Zeng, T., Xu, L., & Liu, W. (2026). Livelihood risks and resilience of rural households: evidence from Sichuan Province, China. Frontiers in Sustainable Food Systems, 10, Article 1934397. https://doi.org/10.3389/fsufs.2026.1934397

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1934397

Keywords: livelihood resilience, livelihood risk, rural households, Sichuan Province, floods, social capital, fractional logit, China, household survey, disaster resilience, sustainable food systems, buffering capacity