1 Introduction

Climate change has led to more intense and frequent natural hazards and disasters over the past few decades. Extreme weather events, such as heavy rainfall, have triggered floods, landslides, and mudslides in many parts of the world (Yamamura 2015; Cappelli et al. 2021). Notwithstanding significant advancements in early warning systems and post-disaster recovery procedures, both the economic and non-economic ramifications of climatic disasters are on the ascent (Lo 2013; Coronese et al. 2019). Rural populations, particularly those residing in less developed nations, are notably vulnerable to recurrent climatic hazards owing to their high exposure and limited adaptive capacity (Peng et al. 2020; Tan et al. 2020). Literature has underscored that strategies followed by government agencies and characterized by technocratic approaches prove insufficient in addressing the unpredictable and multifaceted nature of such disasters (Wilby and Keenan 2012; Tran and Rodela 2019). Consequently, there is an imperative to advocate for collaborative and participatory approaches to foster effective adaptation measures.

Numerous studies have examined the prerequisites for disaster preparedness, mitigation, and adaptation. One prevalent perspective regards individuals as independent decision makers, with their adaptation strategies influenced by psychological perception (for example, risk perception and self-efficacy), experience, and socioeconomic characteristics (Hoffmann and Muttarak 2017; Peng et al. 2020). Another line of inquiry emphasizes social influence as a critical driver of adaptation. Researchers realize that social networks play an important role in disaster risk reduction because they enable information and resource sharing, mental health support, mutual learning, and collaborative efforts (Adger 2003; Lo 2013; Tan et al. 2020). Several studies have found that when deciding whether to adapt to disaster risks, individuals tend to imitate their neighbors. For instance, Kunreuther and Michel-Kerjan (2009) found that homeowners are motivated to purchase flood insurance if their neighbors are also insured, despite the fact that their risk perception remains unchanged. However, the majority of existing research uses qualitative research or case studies to illustrate the phenomenon, and few empirical studies have investigated how and why peers’ adaptation influences an individual’s adaptation decisions.

Neighborhood or peer effects, as defined by Durlauf and Ioannides (2010), refer to the phenomenon where an individual’s behavior is directly influenced by the behavior of a reference group. In rural communities of China, characterized by low population mobility, neighborhood effects tend to be more pronounced. This heightened influence can be attributed to the strong bonds among residents, often spanning generations within the same village (Loh and Li 2013; Tan et al. 2020). Geological hazards predominantly manifest in mountainous rural regions with relatively fixed geographical locations. A substantial portion of the population is consistently exposed to potential low to moderate-intensity hazards, leading to the gradual development of adaptation strategies in response to persistent threats. Therefore, there is a pressing need to investigate the impact of neighborhood dynamics on long-term adaptation behaviors within China’s mountainous rural communities.

The current study aimed to: (1) Examine whether neighborhood effects affect individuals’ adaptation strategy of geohazards (for example, emergency supply storage and participation in training) in rural China; (2) Explore the underlying mechanisms that drive neighborhood effects in the context of disaster risk adaptation; (3) Propose implications for fostering adaptation through social-based interventions.

This study contributes to existing research in three ways. First, while most previous studies have emphasized the role of social networks and social capital in influencing adaptation behaviors, our research offers insights on neighborhood effects by conducting a comprehensive examination of whether and why individuals’ adaptive measures are influenced by the adaptation decisions of their peers. Confirming the existence of neighborhood effects helps provide important implications for enhancing the efficiency of disaster risk reduction policies. Second, there is a relative scarcity of empirical studies examining the mechanisms of farmers’ adaptation decisions, particularly within the rural Chinese context. This study contributes to the existing literature by focusing on a region characterized by frequent geological hazards and a collective culture in China. By validating two mechanisms of neighborhood effects on farmers’ adaptation strategies, our research enhances the understanding of this aspect within the field. Third, while the spatial econometric model is extensively used for macrodata analysis, microdata research has yielded little empirical evidence. Using data with unique spatial information, this study proposes the application of a spatial econometric model to the field of disaster risk reduction.

This article is structured as follows. Section 2 introduces our research hypothesis. Section 3 encompasses estimation methods, data sources, and variables. Section 4 presents the research findings. Section 5 discusses the results, limitations, and policy recommendations.

2 Research Hypothesis

According to Manski (1993, 2000), three effects are constructed to explain why an individual’s choice is typically similar to those of their peers: neighborhood effects (or endogenous effects/peer effects), contextual effects (or exogenous effects), and correlated effects. Neighborhood effects occur when an individual’s behavior is directly influenced by the behavior of a neighboring group. Contextual effects refer to the influence of neighbors’ characteristics on one’s behavior. Correlated effects describe how a group’s shared natural and social environments may lead to consistent decision making within the group (see Fig. 1). Only neighborhood effects can generate a social multiplier effect,Footnote 1 which describes a phenomenon that a small exogenous shock experienced by a targeted group has the potential for a greater influence and a larger aggregate level than intended (Glaeser et al. 2003). For instance, governmental agencies provide disaster prevention and mitigation training to specific rural households. This training directly enhances the adaptive capabilities of the participating households. As these trained households become better prepared, they naturally engage with and exert influence on neighboring households that have not undergone the training. This interplay subsequently results in an indirect improvement in the disaster preparedness of the entire community. Consequently, the presence of this social multiplier effect significantly magnifies the impact of the policy, thereby enhancing the overall resilience of the entire community in the face of disasters.

Fig. 1
figure 1

Reasons underlying the phenomenon whereby members of the same group behave similarly

Full size image

Neighborhood effects have been found in a range of behaviors, such as pro-environmental behavior and new technology adoption (Adjognon and Liverpool-Tasie 2015; Zheng et al. 2021). These studies have shown that individuals are susceptible to the influence of others in their intentions and behaviors. Previous research on the role of neighborhood effects in disaster risk reduction has been limited. A relevant study by Tan et al. (2021) employed the linear-in-means method to identify peer effects in disaster-induced relocation intention. Kunreuther and Michel-Kerjan (2009) have also observed that residents tend to imitate their neighbors’ disaster risk reduction strategies, such as insurance purchases. However, in the context of frequent and recurrent geohazards in rural mountainous regions of China, there remains a lack of evidence regarding the existence of neighborhood effects among households rooted in a collectivist culture when it comes to their disaster risk adaptation actions. Based on these studies, the following hypothesis is proposed:

H1

Neighborhood effects exist in farmers’ adaptation strategy of geohazards.

In addition, researchers have suggested some mechanisms underlying neighborhood effects, such as complementarities, comparisons, convergence, and social learning, which may have varying policy implications (Bursztyn et al. 2014; Tan et al. 2021). According to these studies and our field investigation, there are two possible mechanisms by which the neighbors’ adaptation actions influence an individual’s decision. First, social interaction shapes group opinions about how things should be done (social norms). Second, social interaction facilitates learning processes and knowledge exchange (social learning). They are specifically elaborated as follows.

Social norms entail widely accepted behavioral standards within a community, encompassing notions of “appropriate conduct” and “typical behavioral responses” in given situations. These norms function as soft constraints, diverging from more rigid legal regulations. Individuals within a social group conform to these norms in pursuit of social validation while avoiding potential social sanctions (Abrahamse and Steg 2013; Bergquist et al. 2019). Consequently, groups adhering to shared social norms tend to display congruent beliefs and behaviors. Although social norms serve as potent drivers of specific behaviors, individuals often underestimate their influence due to a lack of awareness (Bergquist et al. 2019). China’s collectivist culture, rooted in a strong emphasis on social connections and peer evaluation, amplifies the impact of social norms. Members who embody qualities of sociability, cooperation, and conformity tend to garner more favorable regard than those who deviate from these norms. Rural populations, in particular, more closely follow the principle of “the rule of the mean” and are highly susceptible to influence from reference groups (Lo 2013). Consequently, when a norm, such as “adapting to disaster,” is established within a group, individuals are motivated to emulate the predominant choice.

Social learning, on the other hand, involves individuals observing choices made by reference groups and interpreting them as possessing greater “private signals/values,” reinforcing their own inclination toward those choices (Conley and Udry 2010). For example, Bursztyn et al. (2014) found that an individual may infer that the consumption decisions made by others are of higher quality, thereby revising their own beliefs regarding the quality of a given product upon learning about others’ revealed preferences. In situations marked by intricate decision making, individuals lacking comprehensive information are inclined to learn from others since it substantially reduces the cost of acquiring information for a specific action (Tan et al. 2021). According to Zhang and Maroulis (2021), geographical congruence is a prerequisite for social learning, as similar disaster risks enhance the comprehension and anticipation of potential threats. Rural residents living in China’s vast mountainous areas often have insufficient private information concerning mountain disasters and adaptation strategies. However, they have the freedom to communicate with or observe their neighbors (Tan et al. 2021). Individuals can adapt their expectations and preferences based on information revealed by neighbors, which may be acquired through direct peer communication or observation of their decisions (Moretti 2011). In our study, a farmer may actively update his/her beliefs (that is, learn from others) and make choices aligned with those of the neighbors. Therefore, the following hypotheses are proposed:

H1a

Social norm is a mechanism underlying the neighborhood effects in adaptation.

H1b

Social learning is a mechanism underlying the neighborhood effects in adaptation.

3 Methodology

In this section, we present the methods, including the spatial econometric model, data collection, and variable selection.

3.1 Spatial Econometric Model

As it is challenging to distinguish between neighborhood effects, contextual effects, and correlated effects, the spatial Durbin model (SDM) is considered an appropriate approach because it includes both the spatial lags of the outcome and the neighbors’ and farmers’ private characteristics (Ajilore 2015; Läpple and Kelley 2015; Yang and Sharp 2017; Zheng et al. 2021). The equation for the spatial Durbin probit model is as follows:

$$y=\lambda Wy+X\beta +WX\theta +\alpha {\iota }_{n}+\varepsilon$$
(1)
$$\upvarepsilon \sim {\text{N}}(0,{\sigma }_{\varepsilon }{I}_{n})$$

where the dependent variable \(y\) is a binary variable. If farmer i adopts the adaptation behavior, then \({y}_{i}=1\); if farmer i does not take any action against disasters, then \({y}_{i}=0\). \(Wy\) is the spatial lag of the dependent variable, reflecting the spatial dependence of the adaptation choices among farmers. The spatial autoregressive parameter λ measures the strength of spatial dependence and captures the coefficient of neighborhood effects. Similarly, \(WX\) is the spatial lag of the independent variable, which captures the weighted average characteristics of neighboring farmers. The k × 1 vector θ captures the contextual effects. X is a matrix of independent variables. W is an n × n spatial weight matrix, which reflects the spatial relationships among individuals (the diagonal element \({w}_{ii}=0\)). \({\iota }_{n}\) is a constant term vector.

According to previous research (Li et al. 2013; Ling et al. 2018), our study assumed that individuals residing within a shared village are “neighbors.” Typically, a village encompasses descendants from diverse families, implying shared kinship or deep-rooted geographic connections among its inhabitants. Each village maintains its grassroots governance structure, conducts regular village meetings to disseminate official policies, administers public affairs, and allocates resources. Consequently, individuals dwelling within a village are highly likely to be familiar with one another and susceptible to the influence of their fellow villagers (Tan et al. 2021). Additionally, these studies posit that designating residents cohabiting within the same village as neighbors can effectively mitigate issues associated with self-selection bias. China’s household registration system (hukou) imposes limitations on the migration of residents into rural areas; and the establishment of rural settlements frequently spans multiple generations, with even newly constructed residential addresses primarily planned within designated homestead areas. In essence, the spatial arrangement of neighborhoods in mountainous rural regions is predominantly externally determined, and residents possess limited options in selecting their neighbors. In the benchmark model, individual interactions are represented by an inverse distance weights matrix, where \({ w}_{ij}={d}_{ij}^{-1}\)(the individual i, j in the same village, i \(\ne\) j) and \({w}_{ij}=0\) (the individual i, j in different villages), indicating that individuals are more influenced by their “nearer neighbors.” Geographic distance \({d}_{ij}\) is the distance calculated based on each household’s longitude and latitude. According to LeSage and Pace (2009), we estimated the spatial probit model using the Bayesian Markov chain Monte Carlo (MCMC) method.

Because the reported coefficients in spatial probit models are not equal to the partial derivatives, LeSage and Pace (2009) proposed a method to interpret the marginal effects of an explanatory variable. These marginal effects can be decomposed into direct and indirect effects.Footnote 2 The direct effect assesses the influence of a change in an explanatory variable on \({y}_{i}\). Indirect effect, also known as spatial spillover, describes the cumulative effect of a change in neighboring explanatory variables on the outcome (\({y}_{i}\)) of farmer i (LeSage and Pace 2009; LeSage et al. 2011; Läpple and Kelley 2015). Total effects are equal to the sum of direct and indirect effects.

3.2 Data Collection

China is a mountainous country, with approximately 70% of its territory covered by mountains, hills, and plateaus. More than one third of its total population lives in mountainous areas, and 74 million of them are directly threatened by geohazards (Cui 2014). In recent decades, extreme weather events (extreme rainfall, strong gusts, and flooding), as external triggers of geohazards, are exacerbated by global climate change.

The data used were collected from a questionnaire survey conducted in Chongqing Municipality between June and July 2018. Chongqing is a disaster-prone region in China. According to data released by the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences,Footnote 3 94% of its terrain is mountainous or hilly, and 15,647 hidden geohazards have been identified. In recent years, the increase and concentration of precipitation in Chongqing have exacerbated the frequency of geological disasters. The questionnaire included basic information about rural residents, psychological factors, social networks, willingness and behavior related to disaster risk reduction, among others. The questionnaire items were chosen after several rounds of trial research and expert consultation. We also placed test questions to determine the quality of the questionnaires. In particular, we recorded each respondent’s location (latitude and longitude). The stratified random sampling method was used to select the samples: based on the ranking of GDP, disaster distribution, and the feasibility of the survey, four counties were selected; based on population size and economic status, three to five townships were selected within each county. Next, one to three sample villages were selected from the rural areas of each township based on the distribution of hidden hazards (Fig. 2). In each village, 10–20 households were randomly selected for the survey. Ten well-trained interviewers conducted semistructured face-to-face interviews, with each interview taking an average of 50 minutes. In the end, 516 valid questionnaires were collected, for a response rate of 100%. The questionnaires show acceptable reliability (Cronbach’s alpha greater than 0.7).

Fig. 2
figure 2

Location of the surveyed village

Full size image

Figure 3 presents the characteristics of the survey participants. Of the respondents, 55.2% are male. The respondents are relatively old, with only 5.8% of them under 40 years of age, and nearly half aged between 56 and 70. The level of education is low: 72.2% of the respondents have only primary education (no more than six schooling years), and only 2.7% have a bachelor’s degree or above, with an average of 5.22 years of schooling. Nearly 40% of the respondents have an average annual household income of less than 30,000 yuan, while 17.1% earn more than 80,000 yuan per year (including earnings from family members who labor in urban areas). The majority of the respondents (87.4%) stated that they have no additional property besides their local residence, and only 12.6% of the respondents own additional properties. Approximately 70% of the surveyed households have two or fewer laborers (number of family members aged 16–64, excluding students and those with poor health conditions). The sampling results are generally representative of rural China. According to the National Bureau of Statistics, the proportion of permanent residents aged 60 or older in China’s rural areas reached 23.81% in 2020, while the proportion of those aged 65 or higher reached 17.72%Footnote 4; 91.8 % of farmers have not attended a high school (< 10 years of schooling) as of 2016.Footnote 5 Rapid urbanization drives a large number of youthful and educated rural laborers to the city in search of employment. Additionally, due to restrictions in the household registration system and high housing prices, the old rural residents who used to work in cities return to their rural hometowns, thereby exacerbating the aging of the rural population.

Fig. 3
figure 3

Sociodemographic characteristics of the respondents (N = 516)

Full size image

3.3 Variables

The dependent variable in this study is binary, assuming a value of 1 for the sample that prepared disaster emergency items or participated in disaster mitigation training and drills, and 0 for the sample that did not adopt any adaptation strategies. Prepared for emergency encompasses the proactive actions undertaken by households to assemble and maintain essential supplies and resources, including provisions such as food, water, medical supplies, emergency shelter materials, and other necessary provisions. These measures are intended for use during or immediately after a disaster, aiming to ensure readiness and enhance resilience. Participation in disaster mitigation training and drills refers to the active involvement of individuals in structured training programs that cover various aspects of disaster risk adaptation, such as hazard identification, evacuation procedures, emergency response protocols, first aid techniques, and other relevant skills and knowledge. Additionally, it is crucial to acknowledge that local government authorities play a significant role in disaster prevention and preparedness within these communities. Many of the hazard adaptation actions undertaken by residents rely on government initiatives or assistance. For instance, participation in disaster mitigation training and drills requires government organization and facilitation, while residents have the right to decide whether to participate. This highlights the collaborative effort between authorities and the community in preparing for and responding to potential hazards.

For the control variables, referring to existing studies, we control a series of psychological variables that may influence adaptation behaviors, including risk perception, self-efficacy, and response efficacy (Lindell et al. 2016; Peng et al. 2019; van Valkengoed and Steg 2019). Based on Slovic (1987) and Yang et al. (2020), risk perception is measured using three statements: “You believe that a disaster will occur in the future,” “If a disaster occurs, your family’s house, land, and life may be affected,” and “When you think of natural hazards and disasters like mudslides and landslides, you feel afraid and scared.” The scale assumes that 1 corresponds to “completely disagree” and that 5 corresponds to “completely agree.” The risk perception level is determined by averaging the responses to the three queries, with a higher value indicating a greater risk perception. We asked “Do you believe that disasters can be managed through certain behaviors?” (0 = do not know, 1 = cannot be, 2 = partially, 3 = can be) to evaluate respondents’ confidence in adaptation strategies (that is, response efficacy). This study used the statement “Although the occurrence of disasters cannot be prevented, you can take some preventive measures to reduce the loss” to measure self-efficacy, with 1 representing low self-efficacy and 5 representing high self-efficacy (Mertens et al. 2018).

In addition, control variables related to social interactions are included. “Public access” refers to disaster prevention and mitigation services or facilities provided by the public sector (for example, warning signs, drills, and prevention cards) with yes/no responses. Collective resources reflect the collective capital of a village. Referring to Tan et al. (2020), we chose four items: “The village leaders are capable,” “The community has the resources/capacity to solve the problems,” “The community has joined forces with organizations/institutions outside the village to help solve the problems,” and “The community has implemented some disaster preparedness policies/programs to respond to future disasters.” Neighborhood trust is the degree to which a respondent has faith in their neighbors. The interpersonal connection reveals the emotional bond between farmers and surrounding residents. If farmers have positive relationships with their neighbors, they may communicate and share more information daily (Tan et al. 2021). Place dependence focuses on the connection between a farmer and his/her area of residence, which is measured by the statements: “You are proud to live in this village,” “You feel that you cannot leave this village and its people,” and “You like this village more than any other place” (Williams and Vaske 2003; Walker and Ryan 2008; Song et al. 2019; Peng et al. 2020). All variables are measured on a 5-point Likert scale, with a higher score indicating a higher level of neighborhood trust, collective resources, interpersonal relationships, and place dependence.

Furthermore, this study controled for some demographic characteristics. The selected variables are described in Table 1.

Table 1 Descriptions of the selected variables
Full size table

4 Results

In this section, we present the main empirical findings, including the results of the benchmark model, marginal effects, and robustness checks. Two potential mechanisms, as explained in Sect. 2, are also examined in this section.

4.1 Benchmark Model

This study used LeSage and Page’s (2009) spatial regression to estimate the coefficients of neighborhood effects. The estimates are presented under model 1 of Table 2. The McFadden pseudo-R2 demonstrates the strong goodness-of-fit of our spatial Durbin probit model. The spatial lag coefficient (λ = 0.389, t = 5.218) is significantly positive at the 1% level, indicating the existence of neighborhood effects; that is, an individual’s adaptation decision is influenced by the adaptation decisions made by others in the village. Besides, the individual’s characteristics, such as age, skills mastery, risk perception, self-efficacy, response efficacy, interpersonal relationships, collective resources, neighborhood trust, and place dependence have significant effects on the adaptation to hazards. Finally, the results in the second column of Table 2 indicate that the spatial lag terms of the response efficacy and interpersonal relationships are statistically significant, indicating that an individual’s adaptation decision is influenced by the characteristics of their neighbors. Therefore, contextual effects exist.

Table 2 Results of the spatial Durbin probit model
Full size table

4.2 Marginal Effects

The estimates \(\beta\), \(\theta\) from the spatial Durbin probit model are not straightforwardly interpretable. To reveal the marginal effects of each explanatory factor on geohazard adaptation, the marginal effects were decomposed into direct and indirect effects (model 2 in Table 2). The estimates of direct effects capture the effects of farmers’ characteristics on their adaptation decisions. The results of the direct (marginal) effects are shown in column 4 of Table 2. Individuals who have a younger household head, family members with non-agricultural skills, high risk perception, strong self-efficacy, strong response efficacy, good interpersonal relationships, high collective resources, high trust in neighbors, and high place dependence are more likely to adopt adaptation behaviors. The skills mastery variable is significantly positive at the 1% level, suggesting that the presence of non-farming family members increases the likelihood of adaptation by 13.2%. This finding is consistent with that of Tan et al. (2021), who found that households were more likely to migrate if they had skilled members working in non-agricultural occupations. Regarding the psychological factors, risk perception, self-efficacy, and response efficacy all have positive and statistically significant direct effects. Each unit increase in risk perception, self-efficacy, and response efficacy raises the adaptation probability by 9.5%, 8.6%, and 7.4%, respectively. These results have been widely confirmed by previous studies (Trainor et al. 2015; Lindell et al. 2016). Interpersonal relationships, collective resources, place dependence, and neighborhood trust all exhibit a significant positive effect. There is a sizeable (12.2%) direct impact of interpersonal connections on adaptation behavior.

The estimates of the indirect effects capture the impact of one’s neighboring farmers’ characteristics on an individual’s adaptation decision (column 5 of Table 3). Neighborhoods with skilled family members and high risk perception, self-efficacy, response efficacy, interpersonal relationships, and trust in the community contribute to an individual’s adaptation to hazards. In particular, the presence of a family member with non-agricultural skills in a neighboring household would encourage a farmer to adopt adaptation measures. One possible explanation is that the competent neighbor in rural communities obtains information from a greater variety of channels, which could increase one’s adaptation behaviors through communication. Farmers’ adaptation is also significantly influenced by other farmers’ interpersonal relationships and neighborhood trust, implying that a harmonious social atmosphere will increase the adaptability of rural settlements to disasters. This finding is in line with previous studies that highlighted the importance of social capital on community disaster resilience (Peng et al. 2020; Tan et al. 2020).

Table 3 Results of the mechanism analysis
Full size table

The total effects are the sum of direct and indirect effects and represent the overall impact of a change in a particular explanatory variable on the probability of adaptation. Households with other non-farming skills are 20.7% more likely to adopt hazard adaptation strategies (13.2% from direct effects and 7.5% from indirect effects) than households that rely solely on agriculture. The total effect of interpersonal relationships on the dependent variable is 19.1%, of which 12.2% arises from direct effects and 6.9% from indirect effects. Other social interaction and psychological perception variables also exhibit significant total effects (see model 2, column 6).

4.3 Robustness Checks

Four additional robustness tests were conducted to determine the reliability of the empirical results of the spatial econometric model.

First, we changed the spatial Durbin probit model to the spatial auto-regressive (SAR) probit model. The findings indicate that the spatial lag λ is still significantly positive, which validates the presence of neighborhood effects in farmers’ adaptation behavior. Then, this study adopted different spatial weight matrices:

  1. (1)

    Spatial adjacency weight matrix \({w}_{ij}=\left\{\begin{array}{c}{({n}_{r}-1)}^{-1}, i, j\,in\,the\,same\,village, i\ne j \\ 0, others\end{array}\right.\): The matrix element \({w}_{ij}\) takes the value of \(1/({n}_{r}-1)\) (\({n}_{r}\) is the number of residents in the r-th village) when farmer i and farmer j are in the same village, and 0 when they are in different villages. It is important to note that, by construction, W is row-stochastic, meaning that it contains non-negative elements, and each row sums to 1.

  2. (2)

    Economic distance weight matrix \({w}_{ij}=\left\{\begin{array}{c}{\left|{e}_{ij}+1\right|}^{-1}, i, j\,in\,the\,same\,village, i\ne j\\ 0, others\end{array}\right.\): In this matrix, the element \({w}_{ij}\) equals the inverse of the absolute value of the annual household income difference between farmer i and farmer j when they both reside in the same village. For all other cases, \({w}_{ij}\) is set to 0. \({e}_{ij}\) is the annual household income difference between household i and household j; To ensure non-zero denominators, we employed a denominator addition of 1. It assumes that farmers tend to interact more closely with individuals who share similar income levels. Therefore, individuals are more influenced by groups with comparable income levels.

  3. (3)

    Social status distance weight matrix \({w}_{ij}=\left\{\begin{array}{c}{\left|{s}_{ij}+1\right|}^{-1},i, j\,in\,the\,same\,village, i\ne j\\ 0, others\end{array}\right.\): To determine the elements of this matrix, we included a question in our questionnaire asking respondents to self-assess their “social status” using a 5-point Likert scale, where 1 represents “very low” and 5 represents “very high.” In this matrix, the element \({w}_{ij}\) is calculated as the inverse of the absolute value of the self-assessed social status when both farmer i and farmer j are residents of the same village. In all other cases, \({w}_{ij}\) is set to 0. \({s}_{ij}\) is the self-assessed social status difference between household i and household j; To ensure non-zero denominators, we employed a denominator addition of 1. It assumes that farmers interact more closely with people of similar social status. All the results show that the spatial lag variables are significantly positive, implying that neighborhood effects exist in adaptation behavior.Footnote 6

4.4 Mechanism Analyses

4.4.1 Social Norms

The more a farmer wants to conform to other village residents, the more he/she is influenced by social norms. This study introduced a variable “interpersonal importance” into the benchmark model, which measures how much a farmer agrees with the statement “It is important for you to keep in pace with other residents in the village” (1–5: strongly disagree–strongly agree). Table 3 shows that the spatial lag coefficient of the spatial Durbin model remains significant at the 1% level. The spatial decomposition results indicate that the “interpersonal importance” variable has a significant positive impact on adaptation behavior. A one-unit increase in interpersonal importance raises the probability of engaging in adaptation behavior by 14.5%. The more farmers value conformity with others, the more they are constrained by social norms.

4.4.2 Social Learning

As argued by Tan et al. (2021) and Bursztyn et al. (2014), people with higher social status have a higher informational advantage when making decisions, and individuals with less information are more likely to learn from those with more information. In mountainous rural areas where information is insufficient, it is expected that low social status families will learn from high social status families, while high social status families will be less influenced by low social status groups. To investigate the social learning mechanism, the samples were separated into two groups based on their self-evaluated social status: high-status group (N = 252) and low-status group (N = 263). For farmers in the high social status group, our study focused on analyzing the extent to which they are influenced by their low social status neighbors in making adaptation decisions. To facilitate this investigation, we made specific adjustments to the weight matrix of the spatial Durbin model. These adjustments involve setting \({w}_{ij}={d}_{ij}^{-1}\) when both farmer i and farmer j reside in the same village, as well as farmer j belonging to the low social status group. Besides, we set the spatial weight values between high social status group farmers within the same village to zero, that is, \({w}_{ij}=0\), thereby excluding considerations of the impact of high social status neighbors on farmer i (high social status). Likewise, for farmer i in the low social status group, our focus shifted to assessing the impact of their high social status group neighbors. Specifically, we set \({w}_{ij}={d}_{ij}^{-1}\) when both farmer i and farmer j are situated in the same village, and farmer j belongs to the high social status group. For all other cases, \({w}_{ij}\) were set to 0.

The neighborhood effect coefficients in panel B of Table 3 show that the spatial lag coefficient for the high social status group is 0.129, which significantly differs from that in the low social status group of 0.302. This finding suggests that residents with higher social status are less influenced by the adaptation strategies of their lower-status neighbors compared to the extent to which lower social status residents are influenced by their higher social status neighbors. In essence, the decision making of farmers with informational advantages tends to be relatively independent, while individuals at an informational disadvantage are more inclined to learn adaptation choices from neighbors with greater access to information.

5 Discussion, Limitations, and Policy Implications

The results of the spatial econometric analysis confirm the existence of neighborhood effects in geohazard adaptation. This finding is consistent with that of Esham and Garforth (2013), who also discovered that farmers learn about climate change adaptation measures primarily from observing neighboring farmers. Moreover, social learning and social norms have been identified as two mechanisms that underlie the neighborhood effects in farmers’ geohazard adaptation.

Social norms, in contrast to policies and laws, are a “soft” normative constraint. Within the context of Chinese rural culture, individuals tend to conform their thoughts and actions to those of other group members (Lo 2013). When the majority of a community adopts adaptation strategies, participation becomes the norm in the small society. Thus, non-participating families become a minority within the group and are likely to be labeled as misfits, isolated, and socially deficient (Lo 2013; Tan et al. 2021). A potential negative evaluation puts them at risk of damaging their social image or even losing aid. Therefore, farmers’ behavioral decisions that imitate those around them may be influenced by social norms. This conclusion is also consistent with Blume et al.’s (2015) theoretical derivation, which states that, in order to achieve the Bayes-Nash equilibrium, social pressure compels individuals to minimize the disparity between their behavior and the group’s average level.

Due to the uncertainty of disaster risks, farmers face high learning costs when making decisions regarding geohazard adaptation. Therefore, farmers who lack information are likely to optimize their decisions by obtaining information from other neighbors. Individuals are constrained by their bounded rationality, but communication and interaction with others can enhance disaster awareness and adaptation behavior. When farmers observe their neighbors (especially households with higher social status or information) adopting adaptation behaviors, they assume that these authoritative village residents have more private information regarding the efficacy of adaptation, and thus adopt the same measures as other farmers. Much like what several interviewees mentioned, “I noticed that our village’s teacher had also stocked up on emergency supplies. He is well-educated and knowledgeable, so I figured these items must be effective to some extent. That is why I decided to buy them as well.” This learning mechanism has also been demonstrated in disaster-induced migration studies (Tan et al. 2021).

Admittedly, this study has certain limitations. First, our study primarily focused on the neighborhood effects of whether to adopt two fundamental geohazard adaptation strategies, with limited discussion on the impact of neighborhood effects on the degree of participation and many other types of adaptation strategies. Future research should strive to include a broader range of adaptation strategies and consider the intensity of adaptation for a more comprehensive analysis. Second, addressing endogeneity challenges in spatial probit models remains a complex task, and this study has not yet tackled potential endogeneity issues. Although our research has undergone a series of robustness tests, further investigation could greatly benefit from addressing the endogeneity problem to enhance the study’s credibility. Additionally, while our survey areas encompass the most extensive and representative geological disaster-prone areas in China, future research would greatly benefit from collecting data from other regions. Diversifying the geographic scope of the study could provide valuable insights into regional variations in geohazard adaptation strategies. Lastly, our study did not differentiate between villages based on factors such as population size and proximity to urban areas. These distinctions could yield valuable insights into variations in the influence of social learning versus social norms on adaptation in different village settings. Future research endeavors should explore these nuances to provide a more holistic understanding of the subject matter.

The findings of this study suggest that incorporating a perspective of neighborhood effects can enhance the effectiveness of interventions aimed at promoting geohazard adaptation, given that policies affect not only target groups but also non-target groups. First, it is difficult for a policy or initiative to simultaneously cover all groups, as the implementation of a single policy in rural China necessitates door-to-door advocacy and grassroots officials’ mobilization. Therefore, the government could initially encourage participation from groups that are more receptive to adaptation, such as rural cadres, village elites, and farmers with higher levels of education. Through the process of learning among rural residents, more groups that have not yet comprehended relevant policies can increase their participation awareness. Second, the publicity used to encourage geohazard adaptation can convey what others are doing and/or what others disapprove of. Moreover, residents who adapt to disasters deserve recognition (for example, through ceremonies in rural communities). Notably, the governments should ensure that the information related to disasters and adaptation is both obvious and accurate; otherwise, the neighborhood effects could result in irrational group decision making.

6 Conclusion

This study investigated neighborhood effects on adaptation in geohazard-prone rural areas of China. Leveraging the survey data from 516 participants with unique spatial information, we employed the spatial Durbin model to capture neighborhood and contextual effects. Four robustness tests were conducted by modifying the benchmark model and spatial weight matrices. The results underscore the pivotal role of individuals’ neighbors in shaping their adaptation decisions in mountainous rural areas of China. The adaptation choices and characteristics of neighbors wield a significant influence over rural residents’ adaptation behaviors. Social learning and social norms are the mechanisms underlying neighborhood effects: On the one hand, traditional Chinese values of collectivism and conformity increase group pressures on farmers. Social interactions through communication or observation compel farmers to shape their beliefs about what is socially acceptable (social norms). On the other hand, when confronted with uncertain and complex disaster conditions, social interactions between neighbors provide farmers with more private information, encouraging them to adopt others’ adaptation behaviors (social learning).