The Impact of Disasters on Adaptive Collective Action Among Farmers: Evidence from China’s Border Regions

Abstract

Climate change has profoundly impacted human development, with disasters serving as a prominent manifestation of its effects on societies. While disasters impose significant disruptions on agricultural production and rural livelihoods, they may also create a “window of opportunity” for rural development by motivating farmers to enhance their adaptive capacities through social learning and collective action. Existing research on disaster impacts predominantly focuses on individual decision-making under assumptions of full rationality, with limited systematic attention to collective action among rural households. Furthermore, studies examining collective action rarely address how external ecological systems influence such behaviors. To address these gaps, this study employs survey data from 419 households across 80 villages in Guangxi, China, revealing two critical findings: (1) disasters exert a significant positive effect on farmers’ participation in response-focused collective actions but a notable negative impact on their engagement in prevention-focused collective actions; and (2) these relationships are mediated by shifts in farmers’ risk perception levels. Building on these insights, this study proposes strategies to strengthen farmers’ collective action by enhancing risk perception and fostering collaborative governance mechanisms between governments and local communities. These measures aim to improve the effective provision and equitable allocation of disaster-response resources, thereby bolstering rural resilience.

1. Introduction

Climate change has emerged as one of the most pressing challenges facing humanity. Since the mid-20th century, the rate of climate change has accelerated unprecedentedly, leading to a surge in the frequency and intensity of climate-related disasters. These disruptions have compounded systemic risks, posing growing existential threats to human societies [1,2]. According to data from the World Meteorological Organization (WMO), weather-, climate-, and water-related hazards triggered nearly 12,000 disasters globally between 1970 and 2021. Developing nations have borne the brunt of these impacts, with climate shocks and disasters accounting for nine in ten fatalities and 60% of economic losses during this period [3]. Rural regions, deeply intertwined with climate change, face profound disruptions to livelihoods and agricultural systems due to escalating climatic shifts and disasters. The destructive impacts of disasters undermine the viability and sustainability of agricultural production, jeopardizing the livelihoods of millions of smallholder households reliant on agri-food systems. According to the United Nations Food and Agriculture Organization (FAO), disasters have caused an estimated US $3.8 trillion in crop and livestock production losses over the past three decades—equivalent to an annual average loss of US $123 billion, or 5% of global agricultural GDP [4]. Furthermore, disasters frequently devastate homes and undermine the livelihoods of rural households, rendering daily survival precarious. According to World Bank data, disaster-induced displacements reached a record high of 26.39 million people in 2023, reflecting the profound human toll of such events [5].

Confronted with the escalating impacts of disasters, are rural households truly helpless? Recent scholarship grounded in risk management frameworks has shifted the discourse, reconceptualizing disasters not as inevitable natural phenomena but as socially constructed risks shaped by systemic vulnerabilities and adaptive capacities [6]. This perspective argues that societies can adopt proactive measures to avoid or mitigate the adverse effects of disasters while actively learning from such experiences to foster development—a phenomenon conceptualized as the “window of opportunity” through which disasters catalyze advancements in rural governance [7]. Empirical studies demonstrate that adaptive collective action serves as an effective pathway for reducing disaster risks and facilitating post-disaster learning. By collaborating to share information, production tools, capital, and knowledge, rural households can collectively adapt to disaster threats [8]. Internationally, adaptive collective action is typically categorized into two dimensions: response-focused collective actions (RCA), which involve coordinating disaster response plans and maintaining mitigation infrastructure; and prevention-focused collective actions (PCA), which emphasize proactive measures such as constructing resilient infrastructure and adopting protective strategies [9]. While existing research explores adaptive collective action through dual lenses—viewing disasters as both destructive threats and catalysts for developmental opportunities—the mechanisms driving these outcomes remain poorly understood.

Commons governance research primarily focuses on common-pool resources beyond private goods, examining the understanding of collective action dilemmas that arise in commons governance processes and exploring solutions to overcome these dilemmas. Scholarship in commons governance defines collective action through four interrelated dimensions: (1) Groups of Interdependent Individuals [10]: Collective action operates through decision-making units composed of individuals bound by mutual dependencies, with the group itself serving as the primary vehicle for action; (2) Common Interests [11]: Shared objectives motivate collective action, forming the basis for group cohesion and purpose; (3) Collective Decision-Making [12]: Members negotiate strategies to achieve common goals, emphasizing participatory deliberation; and (4) Institutional Arrangements [11]: Formal or informal rules structure how collective action is organized and sustained. In this study, collective action refers to processes in which villagers, as interdependent actors, negotiate responses to shared challenges—such as disaster prevention and mitigation—through institutionalized mechanisms that enable the provision of public goods and advance communal welfare. Consequently, rural collective action capacity is defined as a community’s ability to institutionalize participation, ensuring adherence to agreed-upon rules and sustaining engagement in collective endeavors. Commons governance theory enriches the study of collective action by offering diversified analytical frameworks and systematic research paradigms, providing a nuanced perspective for understanding how rural households engage in adaptive collective action under disaster pressures.

Livelihood systems among farmers in China’s border regions exhibit significant vulnerability [13]. On one hand, villagers in these areas face persistent threats from climate-related disasters such as droughts and floods, resulting in unstable agricultural production environments [14]. On the other hand, the border location exposes them to transboundary livelihood risks, including cross-border disease transmission, illicit trade, and population movements. Compounding these challenges, border regions suffer from relatively limited governance resources, weak emergency management capabilities, and insufficient professional expertise and financial support [15]. Against the dual background of livelihood vulnerability and limited governance resources in border regions, this study addresses the following research questions: (1) How do disasters differentially affect farmers’ participation in RCA and PCA? (2) Does the mediating path of risk perception exhibit contextual heterogeneity? Guided by these questions, we analyze survey data from underdeveloped border regions of China, employing instrumental variable regression and structural equation modeling to explore the pathways through which disasters affect adaptive collective action through shifts in farmers’ risk perceptions. This investigation aims to advance theoretical understanding of adaptive collective action while providing actionable policy recommendations for fostering high-resilience rural communities in developing regions.

This research offers three novel contributions to the literature: (1) We empirically identify disaster shocks as an underexplored driver of collective action within commons governance systems. By examining the role of climate change and other ecological factors, this study expands the scope of ecosystem-driven influences on collective behavior. (2) We propose a conceptual-analytical framework that systematically examines how disasters shape adaptive collective action, supported by empirical analysis of specific mechanisms to bridge theoretical abstraction and actionable insights. (3) By integrating subjective factors (notably risk perception) into the social-ecological system (SES) paradigm, we propose a revised model that enhances the framework’s applicability to crisis contexts.

2. Literature Review

2.1. Adaptive Collective Action in Commons Governance

A central challenge in the commons governance lies in resolving collective action dilemmas, epitomized by the “tragedy of the commons,” where individual rationality diverges from collective rationality. Collective action remains a core concern in public governance, with scholarship extensively exploring how to effectively organize collective action. These efforts have been categorized into first- and second-generation theories of collective action. First-generation theories posit that collective action dilemmas must be resolved through a state-market dichotomy, relying on either hierarchical state intervention or market-driven mechanisms to align individual incentives with collective goals. In contrast, second-generation collective action theory—pioneered by Elinor Ostrom—explores a third pathway for overcoming such dilemmas. Ostrom [11] argues that public goods can be effectively managed through self-governance by resource users themselves, bypassing the limitations of exclusively state- or market-centric approaches.

Existing research has extensively explored factors influencing collective action through three lenses: natural geographic characteristics, socioeconomic attributes, and general institutional rules. Among these, scholars have identified a range of critical variables. Socioeconomic attributes—including social capital, leadership, migration, land transfers, large-scale agricultural operations, formalization of land tenure, and a sense of community belonging—have been widely recognized as pivotal drivers. Similarly, institutional rules such as incentive structures and monitoring mechanisms have been established as universal determinants of successful collective action [16,17,18,19,20,21,22,23].

In studies of disaster resilience, existing research typically categorizes collective action into two dimensions: response-focused collective action (RCA) and prevention-focused collective action (PCA). These dimensions, respectively, analyze how communities adapt to disaster impacts through coordinated efforts. RCA encompasses cooperative measures implemented during or immediately preceding a disaster to mitigate its immediate harm. For instance, households may collaboratively develop community response plans, conduct emergency skills training, maintain specialized infrastructure, and stockpile critical supplies. These actions aim to minimize casualties and economic losses by enhancing preparedness for imminent or ongoing disasters. In contrast, PCA emphasizes proactive strategies enacted prior to disaster occurrence. Examples include cooperative initiatives to strengthen protective infrastructure (e.g., flood barriers), adopt risk-reduction land management practices, and organize disaster risk education programs. By addressing systemic vulnerabilities and fostering long-term preparedness, PCA seeks to reduce both the likelihood and potential severity of disasters before they occur. RCA refers to short-term responsive actions aimed at mitigating the adverse impacts of a disaster when its occurrence is inevitable, whereas PCA entails long-term preventive actions designed to reduce the likelihood of disaster occurrence in the absence of any immediate signs.

In border regions, collective action for disaster resilience demonstrates distinctive regional characteristics. On one hand, historically formed transnational social networks significantly influence villagers’ collective action [24]. These established cross-border relationships facilitate mutual assistance through information and material exchange during disasters, though such interactions may simultaneously pose governance challenges at the national security level [25]. On the other hand, as areas predominantly inhabited by ethnic minorities, border regions exhibit unique cultural influences on collective action. Traditional organizational structures particular to certain ethnic groups—such as the “village commune” system among Dai communities and the “village elder” institution in Miao communities—provide established channels for organizing collective initiatives [26].

2.2. Impact of Disaster

2.2.1. Impact of Disaster on Rural Areas from Social Risk Perspective

Disasters have long been a focal point of interdisciplinary research. In the social sciences, scholarly perspectives have shifted from conceptualizing disasters as exogenous natural phenomena—passively experienced events with intrinsic destructive power—to reconceptualizing them as sociopolitical risks that can be systematically mitigated through proactive governance measures. Contemporary studies emphasize disaster risk reduction (DRR) strategies including land-use optimization planning [27], multistakeholder governance frameworks [28], and adaptive resilience mechanisms to minimize systemic impacts [29].

Alongside this paradigm shift, a significant body of research—particularly in low-income rural regions with high disaster exposure—demonstrates causal linkages between disasters and poverty traps. Scholars show that disasters amplify rural vulnerabilities by disrupting agricultural production, reducing household incomes, and causing human casualties [30,31,32]. Under the disaster–poverty cycle theory, scholars argue that rural areas—due to ecological fragility and economic marginalization—are trapped in a “disaster → poverty → intensified disaster → deepened poverty” cycle, positioning disasters as a key driver of poverty [33,34,35].

This perception stems from rural areas being stereotyped as economically backward, environmentally fragile, and knowledge-deficient, leading them to be framed in research as passive subjects requiring external intervention. Such framing overlooks rural communities’ agency and underestimates their adaptive capacities to address disasters [36,37]. Recent studies, however, reveal that rural communities possess inherent resilience [38]. Strengthening this resilience can foster farmers’ collective action capabilities and enhance community cohesion, enabling villages to act as proactive agents in mitigating external shocks through bottom-up approaches [39].

2.2.2. A Study of Disaster and Farmer Behaviour

Scholars have systematically examined farmer behavioral responses to disasters, which manifest in three key domains: participation in disaster prevention infrastructure such as water-saving irrigation systems, reinforced fencing, and pest control facilities [40,41]; adoption of soil conservation practices such as straw mulching, no-till farming, and deep loosening techniques [42,43]; and willingness to utilize risk management tools such as agricultural insurance [44]. However, disaster-induced resource scarcity frequently constrains individual farmers’ capacity to achieve comprehensive risk mitigation. This creates a pressing need for innovative resource allocation mechanisms, driving communities to redefine resource distribution rules and establish new social orders. Such contexts often catalyze collective action—defined as coordinated behaviors emerging spontaneously to address shared challenges [29,45]. Whether through cooperative disaster prevention infrastructure or conservation tillage practices, these collective actions facilitate risk-sharing within communities and rebuild social norms fractured by disasters [46].

2.3. Studies on Risk Perception

2.3.1. The Concept of Risk Perception

In studies on farmers’ willingness to adopt agricultural insurance in the context of disasters, risk perception is defined as farmers’ subjective assessment of the probability of disaster risks based on their personal experiences [47]. As a subjective variable, risk perception arises primarily from farmers’ assessments of external environmental stimuli, shaped by their lived experiences, intuitive feelings, and direct observations [48]. Research on its influencing factors highlights that quantitative risk evaluation involves assessing both the likelihood of occurrence and the severity of impacts. Farmers may adjust their risk perception by evaluating the recurrence probability and potential harm of past agricultural disasters or climate-related risks they face [49]. Consequently, risk perception is shaped by factors such as disaster frequency, shifting patterns of hazards, anticipated climatic conditions, and awareness of the importance of agricultural insurance [50]. A higher level of risk perception indicates that farmers perceive either a greater likelihood of risk occurrence or more severe negative impacts on agricultural production [51]. Some studies further explore specific determinants of risk perception, incorporating objective factors like geographic location, disaster characteristics (such as magnitude and scope), and socioeconomic and demographic attributes. Subjective factors, meanwhile, are categorized as either direct or indirect personal experiences, suggesting that farmers’ risk perception is jointly shaped by rational and emotional factors [52]. This implies that risk perception is not solely influenced by individual subjective factors but also varies across different objective disaster scenarios [53].

2.3.2. Studies of the Relationship Between Risk Perception and Farmer Behaviour

Current research on the relationship between risk perception and collective action remains limited. Nevertheless, extant studies on risk perception’s influence on individual farmer behavior offer critical insights. Existing literature predominantly frames farmer decision-making through rational choice theory. First, scholars posit that risk emerges from the interaction between hazards and vulnerability, shaping behavioral drivers. Reducing vulnerability lowers risk exposure, prompting farmers to adopt risk-coping strategies such as insurance purchases to offset potential losses [54]. Second, perceived disaster severity heightens anticipation of adverse outcomes, inducing anxiety that motivates protective behaviors to alleviate distress [55].

The aforementioned research suggests that farmers’ participation in collective action under disaster conditions primarily aims to mitigate negative risks. However, some studies argue that, contrary to assessments based on full rationality, farmers display bounded rationality in collective decision-making, with most individuals relying on intuitive risk judgments [56]. Specifically, when constrained by factors such as limited access to information, knowledge gaps, and cognitive constraints, farmers often resort to risk analysis approaches, engaging in relatively rational and scientifically grounded deliberation [57]. In disaster-risk contexts, farmers process external risk information—through subjective feelings and intuitive judgments—by systematically recording, filtering, and transforming it into actionable knowledge and stored memories. This cognitive process informs their subjective evaluation of the severity of external shocks and guides decisions regarding whether to avoid, adapt to, or accept risks. Such evaluations ultimately shape their behavioral strategies for coping with disasters.

2.4. Research Commentary

While existing studies have extensively examined the interplay between collective action and farmers’ risk perception in disaster contexts, three critical gaps remain understudied First, while determinants of collective action in commons governance have been thoroughly investigated [58,59], few studies have empirically quantified climate change’s impact on collective action dynamics. Research on adaptive collective action could provide critical theoretical-methodological linkages to address this gap. Furthermore, existing studies on border regions mostly focus on cross-border disaster emergency coordination, while there is insufficient exploration of micro-level collective action among farming households. Second, although prior work recognizes disasters’ broad effects on adaptive collective action, the mechanisms underlying disaster-induced cooperation lack empirical validation. Third, discussions of farmers’ disaster-response behaviors remain narrowly focused on individual-level decision-making, predominantly analyzed through risk-benefit frameworks. However, the mediating role of psychological factors—specifically risk perception—in linking disasters to adaptive collective action remains underexamined.

3. Theoretical Analyses and Research Hypotheses

3.1. Theoretical Analyses

3.1.1. Framework of Analysis

Adaptive collective action includes RCA and PCA. Disasters influence adaptive collective action by affecting social-ecological systems. On the one hand, disasters influence adaptive collective action through their impacts on the social system. For example, disasters alter factors such as risk perception and social capital, thereby influence adaptive collective action [8,60]. On the other hand, disasters affect adaptive collective action by disrupting the ecological systems. For example, disasters may destroy critical resources such as crops and land, thereby further impact adaptive collective action [61,62,63]. To capture the complex mechanisms through which disasters influence adaptive collective action through social-ecological systems, this study employs the social-ecological system (SES) analytical framework proposed by Ostrom [64] for address multifaceted contextual analysis. The SES framework bridges the gap between isolated disciplinary approaches to studying social–ecological systems, offering a common language for interdisciplinary research and facilitating more precise descriptions and systematic diagnoses of these systems. The SES framework is a multi-tiered analytical structure comprising four core subsystems: the Resource System (RS), Resource Units (RU), Governance System (GS), and Actors (A). These subsystems collectively shape the action arena, driving behavioral outcomes among participating actors. Simultaneously, these outcomes are influenced by two external subsystems: the Related Ecosystem (ECO) and the broader Social, Economic, and Political Contexts (S) [65].

Traditional SES frameworks primarily focus on objective attributes within the actor subsystem, such as the number of actors, their socioeconomic characteristics, historical resource-use patterns, or geographic relationships to resources [66]. These variables reflect the role of material conditions in shaping collective action. incorporating farmers’ subjective attributes—particularly risk perception—as determinants of collective action in disaster contexts. Consequently, we expand the actor subsystem in the conventional SES framework by incorporating subjective variables, thereby enhancing its capacity to explain the formation of collective action. Figure 1 presents the revised SES framework, which integrates subjective factors into the actor subsystem for a comprehensive analysis.

3.1.2. Framework for Analysing Rural Social-Ecological Systems Under the Disaster

A critical feature of the SES framework is its modularity, allowing researchers to analyze it horizontally or vertically based on the research question [17]. As shown in

Table 1. Disaster-Specific Social–Ecological Systems (H-SES) Framework.

Level 1 Variables	Level 2–Level 4 Variables
Related Ecosystem (ECO)	ECO1—Climatic condition
	ECO1-a Disaster ECO1-a-1 Frequency of Disasters ECO1-a-2 Degree of Impact of the Disaster
Actor (A)	A2-Socio-economic Attributes of Actors (objective attributes)
	A10- Subjective Attributes of Actors (subjective attributes)
	A10-a Risk Perception
	A10-a-1 Understanding Weather Forecast
	A10-a-2 Sensing Climate Change
	A10-a-3 Sensing Climate Impacts
	A10-a-4 Fear of Disaster
Action Scenario: Interaction (I) → Outcome (O)	I1: Level Of Resource Harvesting    I2: Information Sharing I3: Consultation Process           I4: Conflict Situation I5: Investment Activity          I6: Lobbying Activity I7: Self-Organised Activity       I8: Network Activity I9: Oversight Activity          I10: Evaluation Activity
	O1—Social Performance Measurement
	O1-A Adaptive Collective Action
	O1-a-1 RCA
	O1-a-2 PCA
Social, Economic and Political Context (S); Resource System (RS); Resource Unit (RU); Governance System (GS)	Relevant Control Variables

, this study extends the SES framework through a fourth-tier decomposition under disaster conditions, resulting in the Disaster-Specific SES (H-SES) framework. First, disasters (ECO1-a) are incorporated as a contextual factor under the second-tier variable “climatic conditions” (ECO1) within the ecosystem (ECO) subsystem, reflecting their role as climate-related ecosystem variables. Following Nohrstedt et al. [67], the third-tier disaster variable (ECO1-a) is further divided into fourth-tier variables: disaster frequency (ECO1-a-1) and disaster impact severity (ECO1-a-2), allowing detailed quantification of disaster effects. Second, disasters significantly shape farmers’ willingness to engage in collective action [8]. Following Manomita et al. [9], adaptive collective action (O1-a) is categorized into RCA (O1-a-1) and PCA (O1-a-2), which are linked to the second-tier variable “social performance metrics” (O1) within the outcomes (O) subsystem. Finally, in addition to direct disaster impacts on adaptive collective action, disasters also indirectly affect such action indirectly through farmers’ risk perception (A10-a). Risk perception is nested under the second-tier variable “actors’ subjective attributes” (A10) within the actor (A) subsystem. As risk perception is a latent construct that requires measurement through observable indicators, this study adopts Yamagata et al. [68]’s approach, breaking down the third-tier variable “risk perception” (A10-a) into fourth-tier measurable proxies: awareness of weather forecasts (A10-a-1), perception of climate change (A10-a-2), perception of climatic impacts (A10-a-3), and fear of disasters (A10-a-4).

The Disaster-Specific Social-Ecological System (H-SES) framework developed in this study is illustrated in Figure 2. Notably, the remaining subsystems within the H-SES framework—socioeconomic–political settings (S), resource system (RS), resource units (RU), and governance system (GS)—are treated as control variables. This approach enables the study to focus on the relationship between disasters and farmers’ adaptive collective action by controlling for subsystem characteristics, thus mitigating their potential confounding effects.

3.2. Theoretical Discussions and Research Hypotheses

First, regarding the overarching relationship between disasters and adaptive collective action: On one hand, disasters influence farmers’ RCA. Specifically, RCA is manifested through farmers’ capacity to assist each other during emergencies, strengthening their coping abilities and reducing reliance on external resources. Examples include collaboratively developing community response plans, conducting first-aid training, maintaining dedicated facilities, and stockpiling essential supplies [9]. Under disaster conditions, crises—characterized by high stress, urgency, uncertainty, and conflict-proneness—disrupt institutional frameworks and resource allocation systems, especially when critical resources are depleted [69]. To accelerate post-disaster recovery and restore social order, farmers engage in RCA to mitigate disaster-induced individual burdens.

On the other hand, disasters shape farmers’ PCA. PCA involves addressing root causes of disaster risks through collective mitigation of environmental stressors, such as implementing controlled burning in fire-prone areas or constructing dams in flood-prone zones [70]. However, unlike RCA, farmers may neglect PCA if they have not yet experienced direct disaster impacts, leading to reluctance to invest upfront costs in pre-disaster collective efforts [71]. Moreover, when disasters occur, they cause severe disruptions to livelihoods and production systems [72]. Frequent disasters can exacerbate economic burdens, leaving farmers without sufficient labor or financial resources to engage in PCA initiatives like building flood barriers or adopting resilient crop varieties [73]. Repeated exposure to disasters may lead to desensitization, particularly toward high-magnitude events perceived as inevitable, which can prompt farmers to revert to traditional agricultural practices and reduce PCA engagement [71]. Based on this analysis, we propose the following hypotheses:

H1: 

Disasters have a significant positive effect on RCA of farmers.

H2: 

Disasters have a significant negative effect on PCA of farmers.

Second, focusing on risk perception’s mediating role: On one hand, disasters intensify farmers’ risk perception. Owing to disasters’ unpredictability and abruptness, farmers in threatened areas experience heightened anxiety and fear, culminating in elevated risk perception [74,75]. Risk perception reflects farmers’ subjective judgments based on direct or indirect experiences—specifically, their evaluation of disasters through personal encounters or acquired information. Past disasters and climate-related risks reshape farmers’ risk perception, as severe impacts predict heightened anticipation of future risks [76]. For instance, when a disaster occurs, communities face emergencies that require rapid decision-making, resource allocation, and organized rescue efforts. This may stimulate leadership among rural households, enabling them to acutely perceive disaster-related risks and heighten their risk awareness in response to the disaster [77]. Secondly, disasters encourage villagers to share information and resources to overcome the crisis, thereby enhancing social capital. Through this process of information exchange, households gain more disaster experience, risk knowledge, and early-warning information, allowing them to develop a more comprehensive understanding of potential threats [78]. Furthermore, shared experiences in disaster response strengthen villagers’ sense of belonging to the community, motivating them to more proactively assess disaster risks and linking personal safety to the collective fate of the village, thereby enhancing risk perception [79]. Fourth, disasters may damage critical resources, making villagers more dependent on the allocation of scarce supplies. This increased resource dependency raises their awareness of the fragility and scarcity of these resources [80]. Finally, economic heterogeneity among households becomes more pronounced in the aftermath of a disaster. By observing the varying recovery capacities of different groups, households gain a deeper understanding of their own economic vulnerabilities. This perception leads them to pay greater attention to the economic impacts of disasters, thereby strengthening risk perception [81]. On the other hand, risk perception directly influences farmers’ collective action. As a subjective appraisal of future risks, variations in risk perception drive divergent coping strategies [82]. Under bounded rationality, farmers prioritize intuitive judgments in selecting risk responses [83]. In disaster contexts, farmers with heightened risk perception—anticipating severe consequences—are more likely to adopt risk-averse measures [84]. Among these strategies, engagement in RCA or PCA emerges as a critical mechanism to mitigate perceived risks [85,86]. Thus, we propose the following hypotheses:

H3: 

Disasters positively influence farmers’ risk perception, which in turn contributes to farmers’ RCA.

H4: 

Disasters positively influence farmers’ risk perception, which in turn contributes to farmers’ PCA.

4. Research Design

4.1. Study Area

The Guangxi Zhuang Autonomous Region, situated in southern coastal China and bordering Southeast Asia, is one of China’s five ethnic minority autonomous regions [87]. It has a resident population of 50.13 million [88] and a total land area of 237,600 square kilometers, with a population density of approximately 211 persons per square kilometer [89]. In 2024, the per capita disposable income of residents reached 31,125 yuan, with an urban-rural income ratio of 2.16 [88]. Guangxi was selected as the study area due to its distinctive advantages for examining the relationship between disasters and collective action.

First, Guangxi is characterized by a high frequency and diversity of natural disasters. Geologically, the region features diverse topography, with plateaus encircling its western and northern areas, while mountains, hills, basins, and plains interlace across the south and east. Moreover, situated within active fault zones such as the Bama-Bobai Fault Zone, Guangxi frequently experiences geological disasters including landslides, mudflows, and land subsidence. Climatically, as a typical subtropical monsoon region with extensive coastal areas, Guangxi exhibits hot and rainy summers, unstable weather patterns, and susceptibility to typhoons, resulting in frequent extreme floods and droughts, as well as numerous plant diseases and pest infestations. Additionally, Guangxi contributes over 60% of China’s total sugar cane production, more than 50% of its silkworm cocoons, over 40% of its timber, and above 12% of its orchard fruits [90]. Consequently, any disaster occurrence significantly impacts agricultural production and rural livelihoods in the region.

Secondly, as a border region of China, Guangxi exposes its villagers to transboundary livelihood risks. Bordered by the Beibu Gulf to the south with a coastline of 1595 km and sharing a 696 km land border with the Socialist Republic of Vietnam to the southwest, Guangxi’s unique geographical and socio-natural conditions render local livelihood systems highly vulnerable [89]. Villagers must not only frequently confront recurring and intensifying climate-related disasters such as typhoons, floods, and seasonal droughts, but are also significantly affected by transboundary livelihood risks characteristic of the China–Vietnam border area. These include fluctuations in border trade policies, instability in cross-border labor markets, and transboundary spread of animal and plant epidemics. Consequently, the context in which disasters influence collective action in this region is considerably more complex than in other areas.

Furthermore, as the birthplace of China’s first villagers’ self-governance organization, Guangxi possesses a long-standing tradition of collective action development. Simultaneously, being one of China’s five ethnic minority autonomous regions, it maintains deeply rooted traditions of ethnic settlement and self-governance. Given its frequent disaster exposure, historical continuity in collective action development, and complex socioeconomic context, Guangxi presents an ideal research setting for examining the intricate relationships between disasters and collective action.

4.2. Data Sources

The data used in this study were obtained from the “Hundred Villages and Thousand Households” survey conducted in border areas by the School of Public Administration at G University in 2023. The survey utilized a two-level questionnaire design (village- and household-level) to investigate village and household characteristics, agricultural water management, land use, infrastructure, digital technology adoption, disaster impacts, social capital, and environmental governance during the three-year period preceding the survey. For this study, selected questions from the original survey were reorganized and refined, resulting in a restructured questionnaire comprising four sections: (1) Sample characteristics: Demographic attributes of farmers, including gender, age, education level, political affiliation, marital status, and health status; (2) Disaster conditions; (3) Risk perception measurement; and (4) Collective action participation.

From March to June 2023, the research team consulted with the director of Tsinghua University’s China Rural Governance Survey Project, who is responsible for the “Thousand Villages and Ten Thousand Households” survey program in rural China, and obtained constructive feedback on the preliminary questionnaire draft. Additionally, the team sought advice from professors from Renmin University of China and Peking University who have led field survey teams in household interviews. These consultations helped identify potential challenges in questionnaire implementation, and the professors provided revision suggestions based on their practical field research experience. On this basis, a preliminary survey questionnaire was developed. From July to August 2023, the research team conducted a pre-survey in Babu District of Hezhou, Guangxi Province, China, to test the validity and feasibility of the questionnaire and refine it based on field conditions. At the end of August 2023, the research team recruited 79 questionnaire surveyors at the School of Public Policy and Management of Guangxi University; in early September, a further 10 graduate students and doctoral students with extensive fieldwork experience were invited to train the surveyors. During the training, villages and households were sampled based on socioeconomic development levels across Guangxi. Fourteen counties were randomly selected from Guangxi’s 14 cities, with 6–8 villages randomly chosen per county and 7–11 households per village. By November 2023, the research team completed the survey work of this project and finally achieved 649 questionnaires from 80 villages in 10 Guangxi cities. Questionnaires with incomplete data, illogical responses, or patterned answers were excluded, resulting in 419 valid responses. Figure 3 presents the research sample distribution of this paper.

4.3. Variable Selection

4.3.1. Implicit Variable

This paper investigates the effect of disasters on farmers’ adaptive collective action, defining the dependent variable as farmers’ participation in adaptive collective action. Based on Manomita et al. [9], adaptive collective action is classified into RCA and PCA. On the one hand, for farmers’ RCA, this study employs Exploratory Factor Analysis (EFA) to selects four variables “I am happy to participate in discussions about disasters in the village”, “My family will do disaster-coping work (e.g., clearing silt, repairing roads, etc.), “If the village organisation requests it, would you be willing to participate in the maintenance of irrigation canals” and “If called by the village organisation, are you willing to participate in road maintenance”. EFA indicates that all four variables load on a single dimension (KMO = 0.708, sig < 0.01), specific measures are shown in Table 2. From a theoretical perspective, RCA aims to mitigate the negative impacts of disasters. Disaster events typically cause damage to both livelihood infrastructure (such as roads and houses) and productive infrastructure (including irrigation channels and dams). The restoration of such infrastructure requires substantial human and material resources, which can only be effectively mobilized through collective action. Furthermore, disaster response involves the integration and allocation of extensive resources, including food and medical supplies. Given the common-pool resource characteristics of disaster relief resources, effective disaster response and resource allocation generally require villagers to develop cooperative solutions through collective action. The implementation details of such collective actions—including operational methods, responsible parties, and timing—require consensus-building through villager discussions that integrate community preferences. Therefore, the four indicators selected in this study effectively capture the essential dimensions of RCA.

On the other hand, for the farmers’ PCA, three items were selected via EFA, “If the village organisation requests it, are you willing to reduce pesticide use”, “If the village organisation requests it, are you willing to carry out chemical fertilizer reduction” and “If the village organisation requests it, are you willing to carry out straw resource utilization?”. The EFA confirms that all three items load on a single dimension (KMO = 0.648, sig < 0.01), with measures detailed in Table 2. From a theoretical perspective, PCA aims to mitigate the likelihood of disaster occurrence. Reducing pesticide and fertilizer usage can improve soil structure and ecosystem health, thereby decreasing the probability of disasters such as soil erosion, floods, and droughts. The utilization of straw resources not only reduces fire hazards but also enhances soil resilience to disasters. However, implementing these measures typically requires substantial resource investments, including technical guidance, financial support, and equipment. Furthermore, the effectiveness of these preventive measures depends on widespread adoption among all villagers, as isolated efforts cannot sufficiently improve environmental conditions or reduce disaster risks. Therefore, both the promotion and implementation of such disaster prevention initiatives require organization and coordination at the community level, which can only be achieved through collective action. Consequently, the three indicators selected in this study effectively capture the core dimensions of PCA.

4.3.2. Core Independent Variables

This study draws on the research of Nohrstedt et al. [67] and the contextual conditions of rural households in Guangxi. We used EFA, to select three survey items measuring disaster occurrence and severity: “How frequently have disasters (e.g., typhoons, floods, droughts, pests, diseases, locust plagues) affected your household”; “To what extent did disasters caused significant damage to your housing, property, gardens, and other assets in recent years”; and “How severely have disasters harmed your household’s agricultural production in recent years.” The EFA results indicated that these three variables loaded on a single dimension (KMO = 0.607, sig. < 0.01). Measurement details are provided in Table 2.

To ensure the objectivity of this metric, we have adopted the following approach. Regarding the source of the disaster variable, the frequency of disasters reported by villagers was not based on subjective perceptions but rather on objective records of actual disaster occurrences. During data collection, investigators worked with villagers to chronologically document specific disaster events they had personally experienced. For instance, when villagers encountered disasters such as floods, typhoons, or droughts within the past year, investigators systematically assisted them in recalling and counting each event, rather than relying on generalized impressions regarding disaster frequency.

Furthermore, to ensure the accuracy of the disaster data reported by villagers, we implemented a verification procedure during the survey. While collecting villager recollections on disaster frequency, we cross-referenced their reports with official disaster records maintained by village cadres. As community administrators, village cadres maintain detailed documentation and possess comprehensive knowledge of local disaster occurrences. This cross-verification process ensured general consistency between villager reports and village-level records, thereby enhancing data reliability. Where discrepancies were identified, we conducted further verification with both villagers and village cadres to reconcile the data. Therefore, we contend that the data collection methodology employed in this study ensures the objectivity of the disaster frequency metric.

4.3.3. Intermediary Variable

The mediating variable in this study is farmers’ risk perception. Drawing on the work of Yamagata et al. [68], four survey items were selected through EFA to measure this construct: “Our household regularly seeks weather forecasts (weather conditions) through various channels”; “I perceive that the climate in our village has changed in recent years”; “I believe climate change has impacted my household’s daily life”; and “I feel apprehensive when disasters occur.” The EFA results confirmed that these four variables loaded on a single dimension (KMO = 0.664, sig. < 0.01). Full measurement specifications are provided in Table 2.

4.3.4. Control Variable

This study draws on the research of Su et al. [15] and Su et al. [21] to select other variables that may influence farmers’ collective action as control variables. These include household characteristics such as gender, age, education level, and political status. Controlling for these variables, the analysis aims to isolate the impact of disasters on farmers’ collective action and examine the mediating role of risk perception. Specific measurement indicators are provided in Table 2.

Table 2. Variables and descriptive statistics.

Variable	Notation	Description	Mean	Std.	Min	Max
Implicit Variable
RCA (RCA)	RCA1	Participate in disaster discussions (1 = disagree completely − 5 = strongly agree)	3.97	0.93	1	5
	RCA2	Response work (1 = disagree completely − 5 = strongly agree)	3.97	0.93	1	5
	RCA3	Repair activities (1 = disagree completely − 5 = strongly agree)	4.32	0.79	1	5
	RCA4	Maintenance of hydraulic irrigation canals (1 = disagree completely − 5 = strongly agree)	4.30	0.82	1	5
	RCA5	Road maintenance (1 = not at all − 5 = fully willing)	4.43	0.72	2	5
PCA (PCA)	PCA1	Resource utilisation of straw (1 = not at all − 5 = fully willing)	4.02	1.03	1	5
	PCA2	Pesticide reduction (1 = not at all − 5 = fully willing)	3.85	1.09	1	5
	PCA3	Fertiliser reduction (1 = not at all − 5 = fully willing)	3.74	1.16	1	5
Independent Variable
Disaster (DISASTER)	D1	Frequency of occurrence of disaster (1 = less than 1/year; 2 = 1 or 2/year; 3 = 3 or 4/year; 4 = 5 or 6/year; 5 = more than 6/year)	1.82	0.87	1	5
	D2	Disaster is harmful to the house, etc. (1 = Disagree completely − 5 = Strongly agree)	2.70	1.25	1	5
	D3	Disasters are harmful to agricultural production, etc. (1 = Disagree completely − 5 = Strongly agree)	3.23	1.33	1	5
Intermediary Variable
Risk Perception (RP)	RP1	Understanding weather forecasts (1 = disagree completely − 5 = strongly agree)	4.27	0.87	1	5
	RP2	The climate has changed (1 = disagree completely − 5 = strongly agree)	3.80	1.10	1	5
	RP3	Impact of climate change on daily life (1 = disagree completely − 5 = strongly agree)	3.41	1.18	1	5
	RP4	Frightened by the occurrence of a DISASTER (1 = Disagree completely − 5 = Strongly agree)	3.53	1.15	1	5
Control Variable
	GEN	Gender (1 = male; 0 = female)	0.67	0.47	0	1
	AGE	Age (years)	49.14	12.27	20	82
	EDU	Educational attainment (1 = not attending school; 2 = primary school; 3 = junior high school; 4 = high school; 5 = secondary/vocational/technical school; 6 = junior college; 7 = university; 8 = postgraduate and above)	3.58	1.44	1	8
	PA	Political profile (1 = Communist and reservist; 0 = other)	0.34	0.48	0	1
	MS	Marital status (1 = married; 0 = other)	0.86	0.34	0	1
	CON	Health status (1 = ill; 2 = good; 3 = healthy)	2.60	0.61	1	3

Table 2. Variables and descriptive statistics.

Variable	Notation	Description	Mean	Std.	Min	Max
Implicit Variable
RCA (RCA)	RCA1	Participate in disaster discussions (1 = disagree completely − 5 = strongly agree)	3.97	0.93	1	5
	RCA2	Response work (1 = disagree completely − 5 = strongly agree)	3.97	0.93	1	5
	RCA3	Repair activities (1 = disagree completely − 5 = strongly agree)	4.32	0.79	1	5
	RCA4	Maintenance of hydraulic irrigation canals (1 = disagree completely − 5 = strongly agree)	4.30	0.82	1	5
	RCA5	Road maintenance (1 = not at all − 5 = fully willing)	4.43	0.72	2	5
PCA (PCA)	PCA1	Resource utilisation of straw (1 = not at all − 5 = fully willing)	4.02	1.03	1	5
	PCA2	Pesticide reduction (1 = not at all − 5 = fully willing)	3.85	1.09	1	5
	PCA3	Fertiliser reduction (1 = not at all − 5 = fully willing)	3.74	1.16	1	5
Independent Variable
Disaster (DISASTER)	D1	Frequency of occurrence of disaster (1 = less than 1/year; 2 = 1 or 2/year; 3 = 3 or 4/year; 4 = 5 or 6/year; 5 = more than 6/year)	1.82	0.87	1	5
	D2	Disaster is harmful to the house, etc. (1 = Disagree completely − 5 = Strongly agree)	2.70	1.25	1	5
	D3	Disasters are harmful to agricultural production, etc. (1 = Disagree completely − 5 = Strongly agree)	3.23	1.33	1	5
Intermediary Variable
Risk Perception (RP)	RP1	Understanding weather forecasts (1 = disagree completely − 5 = strongly agree)	4.27	0.87	1	5
	RP2	The climate has changed (1 = disagree completely − 5 = strongly agree)	3.80	1.10	1	5
	RP3	Impact of climate change on daily life (1 = disagree completely − 5 = strongly agree)	3.41	1.18	1	5
	RP4	Frightened by the occurrence of a DISASTER (1 = Disagree completely − 5 = Strongly agree)	3.53	1.15	1	5
Control Variable
	GEN	Gender (1 = male; 0 = female)	0.67	0.47	0	1
	AGE	Age (years)	49.14	12.27	20	82
	EDU	Educational attainment (1 = not attending school; 2 = primary school; 3 = junior high school; 4 = high school; 5 = secondary/vocational/technical school; 6 = junior college; 7 = university; 8 = postgraduate and above)	3.58	1.44	1	8
	PA	Political profile (1 = Communist and reservist; 0 = other)	0.34	0.48	0	1
	MS	Marital status (1 = married; 0 = other)	0.86	0.34	0	1
	CON	Health status (1 = ill; 2 = good; 3 = healthy)	2.60	0.61	1	3

4.4. Models and Methods

Consistent with the study’s hypotheses, disasters may positively influence farmers’ willingness to engage in RCA while negatively influencing their willingness to participate in PCA. Additionally, disasters may indirectly shape farmers’ adaptive collective action by influencing their risk perception. Critically, the key constructs examined in this study—disasters, risk perception, and adaptive collective action—are latent variables requiring measurement through observed indicators. To estimate the relationships among these latent variables, a structural equation model (SEM) was constructed (Figure 4). The model’s fit was assessed using Mplus 8.3, and a bias correct percentile bootstrap CI was applied to examine the mediation effect’s significance. This methodology enables testing the mediating role of risk perception in disasters’ impact on adaptive collective action.

4.5. Data Processing and Analysis

As most variables in this study are latent constructs measured by observed indicators, the reliability and validity of the questionnaire design were assessed prior to empirical analysis to verify its rationality and effectiveness. Following common practice in the literature, Cronbach’s alpha coefficients were computed using SPSS 27.

Table 3. Results of questionnaire reliability analysis.

Latent Variable	Observed Variable	Standardized Regression Weights	Cronbach’s α	CR	AVE
RCA	RCA1	0.696	0.745	0.8041	0.4524
	RCA2	0.670
	RCA3	0.757
	RCA4	0.603
	RCA5	0.626
PCA	PCA1	0.701	0.838	0.8817	0.7159
	PCA2	0.905
	PCA3	0.915
DISASTER	D1	0.678	0.658	0.7937	0.563
	D2	0.779
	D3	0.789
RP	RP1	0.592	0.69	0.797	0.5009
	RP2	0.812
	RP3	0.796
	RP4	0.600

Note: In this paper, CR > 0.6, AVE > 0.4 [91], and Cronbach’s α > 0.6 [16] were used as criteria for judging the validity of the results.

presents reliability analysis results for independent, dependent, and mediating variables. As summarized in Table 3, the reliability coefficients were 0.745 (RCA), 0.838 (PCA) 0.658 (disaster), and 0.69 (risk perception). All coefficients exceeded the threshold of 0.6, demonstrating adequate scale reliability.

Following confirmation of adequate satisfactory reliability, further validity tests were conducted. First, factor analysis was performed using SPSS 27 to derive factor loadings for each item within their respective principal components. Convergent and discriminant validity were then evaluated through the average variance extracted (AVE) and composite reliability (CR). The validity analysis results for the independent, dependent, and mediating variables are summarized in Table 3. As indicated in Table 3, the AVE values for all scales ranged between 0.45 and 0.72, while CR values fell between 0.79 and 0.88. These results confirm strong convergent validity and composite reliability, demonstrating that the observed variables adequately reflect the latent constructs. This supports the credibility of the quantitative findings in this study.

5. Estimated Results

5.1. Total Effects Test

Since both the independent and dependent variables in this study are latent constructs that cannot be directly observed, continuous variables were extracted from both constructs using principal component analysis. Given that the derived dependent variable was continuous, ordinary least squares (OLS) regression models were employed to test Hypotheses H1 and H2.

Table 4. Total effect of disaster affecting farmers’ willingness to collective action.

Variable	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
	RCA	PCA	RCA	PCA	RCA	PCA
Disaster	4.738 **	−0.101 **	6.716 ***	−0.100 **	21.001 **	−0.546 ***
	(2.086)	(0.049)	(2.127)	(0.048)	(6.569)	(0.143)
GEN			−1.263	−0.167	−2.043	−0.168
			(4.600)	(0.107)	(4.790)	(0.116)
AGE			−0.061	−0.010 **	0.078	−0.014 **
			(0.212)	(0.005)	(0.228)	(0.005)
EDU			1.996	0.041	3.119	0.014
			(1.800)	(0.042)	(5.075)	(0.046)
PA			5.048	0.247 **	3.656	0.284 **
			(4.851)	(0.111)	(5.075)	(0.121)
MS			13.708 **	0.024	6.930	0.254
			(6.274)	(0.145)	(7.148)	(0.172)
CON			−2.083	0.283 ***	−2.619	0.286 ***
			(3.459)	(0.079)	(3.601)	(0.086)
AREA			Controlled	Controlled	Controlled	Controlled
p	0.024	0.038	0.000	0.000	0.000	0.000

Note: ***, ** indicate significant at the 1 per cent, 5 per cent statistical levels, respectively.

presents the regression results for the total effects. Models 1 and 2 estimate the impact of disasters on farmers’ willingness to participate in RCA and PCA without including control variables.

The results showed that the effect of disaster on RCA showed a positive coefficient which was statistically significant (b = 2.086, p < 0.05), indicating that disaster exerted a significant positive effect on RCA. In the effect of disaster on PCA, the coefficient was negative and the result was significant (b = 0.049, p < 0.05), indicating that disaster had a significant negative effect on PCA. Models 3 and 4 examined the effect of disaster on farmers’ willingness to participate in RCA and PCA after adding control variables, and the results show that the disaster coefficient is positive and the result is significant (b = 2.127, p < 0.001) in the effect of disaster on RCA, still presenting that disaster has a significant positive effect on RCA; and in the effect of disaster on PCA, the coefficient of disaster was negative and the result was significant (b = 0.048, p < 0.001), still presenting a significant negative effect of disaster on PCA.

The estimates from the conventional multiple regression may be subject to endogeneity issues. First, reverse causality could arise; for instance, villages with stronger collective action capacity may implement more effective disaster prevention and response measures [92]. Second, omitted variable bias is a concern, as factors such as land fragmentation and labor migration have been shown to significantly influence collective action [17,20], yet these were not included in the model specification in the OLS regression. To address these concerns, this study employs instrumental variables (IVs) to mitigate the impact of endogeneity on the estimates.

Following Wooldridge [93], valid IVs must satisfy two criteria: (1) strong relevance to the endogenous regressor, and (2) exclusion restriction (i.e., uncorrelated with the error term conditional on covariates). Drawing on Su et al. [21,94], this study selects “Household location in Hechi City (HECHI)” as the IV for disasters in the RCA model and “Household location in Hezhou City (HEZHOU)” as the IV for disasters in the PCA model. The rationale for these IVs is twofold: On the one hand, in 2023, Hechi City experienced frequent disasters such as wind-hail [95], droughts [96], and floods [97], while Hezhou City faced wind-hail disasters [98] and flash floods [99], causing substantial economic losses and livelihood disruptions. Households in these cities are thus systematically exposed to higher disaster risks, ensuring a strong correlation between the IVs and the core independent variable. On the other hand, in China, residential location is predominantly determined by historical path dependence and institutional constraints (e.g., ancestral migration patterns) rather than individual choices [100]. The decision to reside in Hechi or Hezhou was predominantly determined by historical circumstances and family lineage, minimizing its correlation with contemporary individual-level variables.

Additionally, to further address potential endogeneity, this study constructs an instrumental variable (IV) defined as the interaction of disaster exposure and residence in Fangchenggang City (Disaster × Fangchenggang, DFCG) based on the methodology of Rao and Zhang [101]. This interaction term serves as the IV for disaster exposure. The disaster exposure component in DFCG ensures relevance by directly correlating with the endogenous regressor. Meanwhile, the locational factor—Household location in Fangchenggang City—is largely unrelated to individual decision-making by farmers, thereby ensuring that DFCG remains uncorrelated with the regression’s error term.

Models 5 and 6 presented the estimation results after employing instrumental variables. Weak instrument identification tests revealed that Model 5 yielded a Cragg-Donald Wald F-statistic of 25.902, with a Sargan overidentification test p = 0.230 (> 0.1). Similarly, Model 6 produced a Cragg-Donald Wald F-statistic of 30.591 and a Sargan overidentification test p = 0.852 (> 0.1). These results indicate that the selected IVs are strongly correlated with the core independent variable (no weak instrument problem) and the tests fail to reject the null hypothesis that the IVs are uncorrelated with the error term. Thus, the instrumental variables employed in this study are validated as reasonable and effective. The estimation results of Model 5 and Model 6 showed that for RCA, the disaster coefficient was significantly positive (b = 6.569, p < 0.05), indicating that disaster had a significant positive effect on RCA; for PCA, the disaster coefficient was significantly negative (b = 0.143, p < 0.001), indicating that disaster had a significant negative effect on PCA. Hypotheses H1 and H2 were empirically supported.

5.2. Mediating Effects and Structural Equations

Building on the significant impact of disasters on farmers’ willingness to engage in collective action, this study employs a mediation effect model to elucidate the underlying mechanisms. We constructed a structural equation model (SEM) to estimate the mediation effects, and after iterative refinement, the final model achieved the following fit indices: χ²/df = 2.15, CFI = 0.913, RMSEA = 0.052, and AGFI = 0.901. According to Wang et al. [15], these indices meet the criteria for excellent model fit, indicating that the SEM specification is theoretically and statistically valid. To test Hypotheses H3 and H4, the bias correct percentile bootstrap CI was implemented using Mplus 8.3 statistical software. Given that the mediating variable (risk perception) is also a latent construct, all observed variables measuring the mediator were extracted into a single factor through principal component analysis, resulting in a continuous variable for the analysis. The mediation test results are presented in

Table 5. Estimated results of complete standardization (STDYX) of mediating effect.

Path	Coefficient	Bootstrap Bias-Correct 95% Confidence Interval
		LL	UL
Disaster → Risk Perception	0.223	0.134	0.306
Risk Perception → RCA	0.207	0.126	0.285
Risk Perception → PCA	0.107	0.013	0.191
Disaster → RCA	0.072	0.005	0.154
Disaster → PCA	−0.124	−0.212	−0.034
Disaster → Risk Perception → RCA	0.046	0.025	0.074
Disaster → Risk Perception → PCA	0.024	0.005	0.052

Note: bootstrap = 5000.

.

As shown in Table 5, first, disaster exposure exhibited a significant positive association with farmers’ risk perception (b = 0.223, CI: 0.134–0.306), indicating that higher disaster frequency or severity strengthen farmers’ risk perception. When disasters occur more frequently or with greater severity, farmers proactively engage in monitoring meteorological patterns, develop objective assessments of disaster impacts, and are more likely to exhibit heightened disaster-related anxiety. Conversely, reduced disaster exposure diminishes household-level disruptions, leading farmers to pay less attention to disasters and exhibit weaker fear responses. Second, the degree of farmers’ risk perception is significantly positively correlated with their willingness to participate in RCA (b = 0.207, CI: 0.126–0.285) and shows a marginally significant positive correlation with PCA (b = 0.107, CI: 0.013–0.191). This suggests that farmers with stronger risk perception have clearer cognition and judgment regarding disasters, thereby displaying greater willingness to engage in both RCA and PCA. In contrast, weaker risk perception corresponds to more vague disaster awareness; farmers may fail to accurately recognize disaster severity and thus opt out of participating in RCA or PCA as coping strategies.

Table 5 also presents the mediation effects test results. First, the path coefficient for “disaster → risk perception → RCA” is 0.046, which is significant at the 95% confidence level. This indicates that disasters significantly enhance farmers’ willingness to engage in RCA by intensifying their risk perception, thus supporting Hypothesis H5. Second, the path coefficient for “disaster → risk perception → PCA” is 0.024, significant at the 95% confidence level, demonstrating that disasters also positively influence farmers’ PCA participation through elevated risk perception. Hypothesis H6 is thereby supported. Figure 5 visualizes the model’s goodness-of-fit indices, The calculated results indicate that the proportion of the mediating effect of risk perception is 39.066% for RCA and 23.861% for PCA. This variation in the proportion of the total effect accounted for by risk perception suggests that disasters may influence villagers’ RCA and PCA through additional factors beyond risk perception. For instance, existing studies have substantiated that leadership, social capital, sense of community belonging, resource dependence, and economic heterogeneity among villagers may serve as significant mediating variables affecting collective action [16,102]. These potential mediating mechanisms are expected to be systematically examined in subsequent investigations by the research team.

6. Discussion

6.1. The Role of Subjective Factors in Collective Action

As shown in Figure 5, disasters exert a significant negative impact on farmers’ participation in PCA, with a coefficient of −0.128 after controlling the mediating effect of risk perception. However, after accounting the mediating effect, the total impact coefficient of disasters on PCA decreases to 0.024. This indicates that the introduction of risk perception as a mediating variable effectively mitigates the negative influence of disasters on PCA. These findings highlight the critical role of subjective factors, exemplified by risk perception, in analyzing collective action dynamics. Existing research has demonstrated that losses incurred by individuals during natural disasters extend beyond economic dimensions to include emotional, socio-spatial, and religious aspects [103]. Consequently, motivations for participating in collective action may arise not only from economic compensation needs but also from emotional trauma, yet current policies often overlook these non-economic factors. Our findings, the results visualized in Figure 5 further reveal that omitting subjective factors from the analysis may lead to an overestimation of the negative effects of climate-related variables (e.g., disasters) on collective action while overlooking the potential contributions of subjective mechanisms. The positive mediating role of risk perception in the relationship between disasters and collective action, as demonstrated in Figure 5, carries an important implication: although disasters exert a significant negative influence on PCA, this impact is not insurmountable. Therefore, emphasizing the role of subjective factors in collective action aimed at mitigating climate change effects can enable future research to more comprehensively and objectively assess the impacts of climate-related variables (e.g., disasters) on social order and societal challenges. Concurrently, these findings provide theoretical support for leveraging human agency in practical efforts to address disaster-induced challenges.

6.2. Impact of Relevant Ecosystem Factors Represented by Disaster on Collective Action

The adaptive collective action explored in this study refers to farmers’ adaptive behaviors under disasters, which constitutes a form of collective action in commons governance. Consequently, established analytical frameworks for studying collective action in commons governance can be applied to interpret our findings. Existing research has extensively examined collective action within the Social–Ecological Systems (SES) framework. Specifically, scholars have investigated relationships between collective action and key SES components, including resource systems and resource units [104], governance systems [105], actors [16], and socioeconomic–political contexts [17]. However, few studies have systematically analyzed the relationship between related ecosystems and collective action. Against this research backdrop, our study investigates the pathways through which disasters influence collective action. This approach not only identifies disasters as a novel explanatory variable in collective action research but also establishes a conceptual linkage between ecosystems and collective action, thereby extending the applicability of classical institutional analysis frameworks. These contributions significantly enrich the theoretical landscape of collective action in commons governance.

6.3. Why Do Disasters Promote RCA but Inhibit PCA?

Our findings indicate that disasters exert a positive influence on RCA but a negative impact on PCA. To explain this discrepancy, Ao et al. [70], basing on the Theory of Planned Behavior (TPB), found that individuals only take action when the negative impacts of a disaster reach a certain threshold. The TPB posits that human behavior is not arbitrary but is directly driven by “behavioral intention.” This intention, in turn, is shaped by three key factors: attitude toward the behavior, subjective norms, and perceived behavioral control [106].

On the one hand, behavioral attitude refers to an individual’s evaluation of the favorability or unfavorability of performing a specific behavior. Individuals are more likely to adopt a behavior when they perceive its outcomes positively. RCA can directly reduce losses after a disaster occurs, demonstrating higher cost-effectiveness. Consequently, people are more inclined to pool resources to address disasters, enabling RCA to secure greater financial, human, and material support. In contrast, PCA requires substantial resource investments before a disaster occurs and demands long-term commitments to demonstrate tangible effects, resulting in relatively lower cost-effectiveness. Therefore, individuals are reluctant to incur significant costs for PCA when disaster risks are not imminent, making it challenging for such initiatives to obtain sufficient resource support.

On the other hand, subjective norms refer to the perceived social pressure individuals experience when deciding whether to perform a particular behavior. Given the suddenness and uncertainty of disasters, failure to promptly contain their impacts can lead to rapidly escalating threats to survival and expanding socioeconomic consequences. To mitigate the direct losses caused by disaster-induced pressures, people tend to rapidly engage in RCA such as rescue operations, resource allocation, and population evacuation. Existing research indicates that individuals only take action when disaster-induced pressures reach a certain threshold [70]. Therefore, when disaster occurrence is inevitable, people experience heightened fear and significantly enhanced risk perception, making them more inclined to act either immediately before or after a disaster strikes. Conversely, when disaster occurrence remains uncertain, the relatively lower perceived pressure results in weaker risk awareness, making it difficult for PCA to gain sufficient support. Consequently, there is a prevailing tendency to avoid undertaking long-term preventive collective actions.

Furthermore, the empirical findings of this study regarding the differential impacts of RCA and PCA on collective action provide a plausible explanation for the persistent controversies in existing literature concerning the effects of disasters on collective action. Specifically, some scholars argue that disasters exert a positive influence on collective action. For instance, disasters can foster the formation of collective action, and communities with prior disaster experience are more likely to self-organize to mitigate disaster risks, thereby encouraging villagers to engage in preparedness and mitigation activities before disasters occur [8,9,107]. Conversely, other researchers have found that disasters negatively affect collective action. For example, disasters may trigger social unrest and erode social cohesion and cooperation among community members [108]. The empirical results of this study demonstrate that disasters generally exert a positive overall impact on RCA but a negative overall impact on PCA. Therefore, the divergent conclusions in existing studies can likely be attributed to the insufficient differentiation between types of collective action, particularly the distinction between RCA and PCA.

7. Conclusions, Insights and Shortcoming

7.1. Conclusions

This study integrates commons governance theory to theoretically analyze farmers’ behavioral decisions, risk perception, and collective action, supported by empirical evidence from a survey of 649 households across 80 villages in Guangxi, a border region of China. The findings yield three key conclusions: First, regarding total effects, disasters exert a significant positive influence on farmers’ willingness to engage in RCA in border regions. Higher disaster severity correlates with stronger RCA participation intentions. Conversely, disasters demonstrate a significant negative impact on farmers’ willingness to adopt PCA in border regions. Elevated disaster severity reduces PCA participation, likely due to farmers’ reluctance to bear perceived costs, risks, and knowledge gaps associated with PCA-driven disaster responses.

Second, from the mediating effect of risk perception. People are not only affected by objective factors when evaluating a risk, but also by subjective factors, and people will adopt different coping behaviours in different risk perception situations. Specifically, on the one hand, the experience of disaster increases the degree of risk perception of farmers in border regions, which in turn increases the willingness of farmers to participate in RCA. On the other hand, the experience of disaster increases the degree of risk perception of farmers in border regions, but further decreases the willingness of farmers to participate in PCA. It can be seen that subjective factors represented by risk perception actually still have an impact on farmers’ adaptive collective action, reflecting the important role of human subjective factors in analysing issues such as disaster shocks and collective action.

7.2. Policy Implications

Based on the empirical findings and field observations, this study proposes the following policy recommendations: First, intensify disaster risk communication to elevate farmers’ risk perception. Farmers’ risk perception levels depend on their ability to collect and interpret external risk signals, with varying risk awareness leading to divergent collective action choices. Therefore, government agencies should not only enhance public awareness through expert-led rural workshops and science dissemination programs, but also implement activities such as disaster simulation drills and visual exhibitions of disaster cases. These initiatives help villagers viscerally appreciate the potential impacts of disasters through realistic scenarios and case studies. Such approaches can foster a comprehensive and clear understanding of disaster risks in agricultural production and daily life, enable a thorough grasp of disaster response measures and their implementation mechanisms, and ultimately strengthen risk perception to mitigate the societal disruptions caused by disasters. Second, establish bottom-up participatory mechanisms to foster government–community collaboration. While both governmental and civil entities play distinct roles in disaster prevention and mitigation, top-down governance models often fail to prioritize farmers’ needs, resulting in inefficient delivery of public goods for disaster resilience. To address this, a “bottom-up” demand-responsive mechanism should be established, grounded in the fundamental needs of villagers. This can be achieved through regular revisions of village-level disaster emergency management plans and ensuring substantial villager participation via relevant working meetings. Furthermore, it is essential to fully mobilize and integrate nongovernmental resources, including technical expertise, human capital, and specialized skills. For instance, local human resources can be effectively utilized through existing village institutions such as accumulated points system and labor contribution mechanisms, while emergency management funds can be allocated to compensate villagers for their labor inputs. Such approaches can harness the collective power of disaster response, enhance the role of collective action, and thereby improve the efficiency of disaster prevention and mitigation resource allocation while addressing distribution challenges.

7.3. Shortcoming

This study has the following two limitations: First, this study has yet to establish a theoretical framework for the nonlinear relationship between disasters and collective action. Future research could integrate more systematic theories, such as the adaptive cycle theory, with the Social–Ecological Systems (SES) framework to advance targeted investigations into the nonlinear dynamics shaping disaster–collective action interactions. Second, the research sample is limited to Guangxi, a border region of China, which may not fully represent the realities of rural households in all areas. Therefore, future studies could expand the geographical scope to enhance the sample’s representativeness.