Impact of Human–Elephant Conflict Risk Perception on Farmers’ Land Use Efficiency in Yunnan, China

Abstract

In countries and regions where Asian elephants are distributed, human–elephant conflict has become an important ecological and socio-economic issue. As one of the major habitats of Asian elephants, China faces severe challenges. Based on the theory of planned behavior and the risk perception theory, this study takes the survey data of 449 smallholder farmers in the Asian elephant distribution areas of Pu’er City, Yunnan Province as samples and uses the Tobit model and the mediating effect model to empirically analyze the impact of human–elephant conflict on farmers’ land use efficiency and its mechanism. The results show the following: (1) The human–elephant conflict risk perception has a significant negative impact on farmers’ land use efficiency. A one-unit increase in risk perception decreases land use efficiency by 250.34 CNY/mu. (2) Social networks positively moderate the negative impact of the human–elephant conflict risk perception on farmers’ land use efficiency, further strengthening the negative impact of risk perception. (3) From the perspective of the mechanism, the human–elephant conflict risk perception increases the likelihood of farmers changing their land use behavior. Farmers with high risk perception tend to reduce agricultural capital investment, which in turn leads to a decline in land use efficiency. In view of this, this paper puts forward suggestions in terms of strengthening ecological monitoring and control, increasing support for agricultural production, and guiding rational social network communication, providing theoretical support and practical guidance for alleviating human–elephant conflict and improving farmers’ land resource use efficiency.

1. Introduction

With the changing global ecological environment and the continuous expansion of human activities, conflicts between humans and wildlife have become increasingly prominent [1], emerging as a significant issue in global ecology and sociological research. Human–elephant conflict (HEC), as one of the most representative conflicts, has become more severe in recent years across many countries in Asia and Africa [2,3,4,5,6]. Currently, there are only three major elephant populations globally: the Asian elephant (Elephas maximus) in Asia, the African savanna elephant (Loxodonta africana) in Africa, and the African forest elephant (L. cyclotis) in parts of sub-Saharan Africa [7]. Since the early 20th century, due to habitat loss and fragmentation, elephants have frequently entered human activity areas [8,9,10,11,12,13], leading to crop damage, property loss, and even human casualties [14,15,16]. This conflict not only threatens the survival of elephant populations but also severely impacts the livelihoods of farmers, increasingly endangering the survival of residents [17]. For instance, Sri Lanka has the highest level of HEC in the world [18], with an average of over 300 elephants killed and 90 human deaths annually from 2015 to 2021 [19,20]; in Tanzania, African elephants caused 70 deaths and 16 injuries from 2017 to 2019 [21,22,23]; in India, elephant attacks resulted in 2361 deaths from 2014 to 2019 [24,25]; and in Kenya, approximately 200 people died from HEC between 2010 and 2017 [2,26].

In China, Asian elephants are primarily distributed in three prefectures of Yunnan Province: Xishuangbanna, Pu’er, and Lincang. From 2013 to 2020, HEC resulted in 100 casualties and direct property losses amounting to 210 million yuan, with indirect losses being incalculable [27,28,29]. Around 83% of residents in areas where Asian elephants are active reported crop damage [30]. Additionally, due to the frequent presence of Asian elephants near villages, local residents are often unable to carry out production activities such as tea picking and rubber harvesting in a timely manner [30]. The elephants’ aggressive behavior poses a threat to the personal safety of farmers, causing psychological issues such as fear and anxiety and significantly negatively impacting their quality of life [31]. Although the Chinese government has implemented measures such as establishing nature reserves, restoring habitats, deploying monitoring and early warning systems, and introducing public liability insurance for wildlife damage to mitigate some of the conflicts, the fundamental issue of human–elephant coexistence remains unresolved. Apart from monitoring measures like early warning systems, other initiatives suffer from limited coverage, high investment with low effectiveness, and sustainability issues [32]. Similarly, in Sri Lanka, elephants that are relocated back to protected areas often return to the original conflict zones, leading to further incidents [33]. In India, Asian elephants have adapted over time and no longer fear traditional deterrents such as bonfires and loud noises [34].

Assessing farmers’ risk perceptions and their coping behaviors is of paramount importance, as it can reveal decision-making patterns in the face of uncertainty [35]. Current measurements of HEC primarily focus on the direct negative impacts of elephants on humans, often overlooking the indirect effects on farmers. These measurements rarely address the disruptions to household livelihoods and food security, the psychological health of farmers, or the transaction costs incurred when seeking compensation, thus failing to comprehensively and accurately reflect the extent of HEC [36,37]. Risk perception is a critical factor in evaluating the potential elements of conflict; however, there is often a significant discrepancy between risk perception, actual risk levels, and responses to risk [38,39]. Farmers’ perceptions of the level of human–wildlife conflict tend to be higher than the predicted actual conflict levels, and relying solely on the history of crop raiding by elephants is insufficient to assess the overall level of conflict experienced [40]. Merely quantifying and reducing real risks, and subsequently disseminating information about risk reduction, may have limited effectiveness in preventing conservation conflicts [41]. Therefore, it is essential to also consider risk perception, which is the probability of harm or loss as perceived by individuals facing uncertain dangers, and affective risk perception, which encompasses the fear or concern about potential dangers [42].

The academic community has conducted extensive research on land use efficiency, but most of it has focused on macro and meso scales [43,44,45]. Land is a fundamental resource for the survival and development of farmers, and its utilization efficiency is closely linked to their economic income and quality of life. As the micro-level agents of land use, farmers’ behaviors not only determine the level of land use efficiency but also profoundly influence regional sustainable development [46]. Agriculture is a uniquely high-risk and uncertain industry, requiring farmers to make decisions in a dynamically changing risk environment [47]. After experiencing human–wildlife conflict (HWC), farmers often adopt self-adaptive behaviors [48]. To mitigate the losses caused by HWC to their households, they proactively adjust their lifestyles and production methods [49]. Severe HWCs can lead to soil fertility degradation [50] and farmland abandonment [51,52,53]. For instance, Liu’s survey in Yunnan found that the threat posed by Asian elephants significantly negatively impacts farmers’ production inputs, with farmers reducing their investment in agricultural production factors as the severity of elephant-related damage increases [54]. Similarly, Yan et al.’s study on residents in China’s Giant Panda National Park revealed that HWC experience significantly increases the likelihood of farmers adjusting their land use behaviors, not only raising the probability of altering crop planting structures but also increasing the chance of abandoning farmland. This is primarily driven by heightened farmer perceptions of future HWC risks [55]. Psychology defines perception as the process of information processing [56]. Risk perception refers to an individual’s assessment of the probability and consequences of specific events or their subjective judgment of the nature and severity of potential threats [57]. Farmers’ HEC risk perception largely determines their land use decisions and behaviors, thereby significantly impacting their land use efficiency. Additionally, farmers’ social networks can greatly influence their production behaviors, shaping their learning and decision-making processes [58,59]. Therefore, in-depth research into the impact of HEC on farmers’ land use behaviors and its underlying mechanisms is of great significance for achieving the dual goals of Asian elephant conservation and community harmonious development. This not only helps alleviate current ecological and socio-economic pressures but also provides valuable insights for global biodiversity conservation.

In summary, previous studies have predominantly focused on the current situation of HEC and loss mitigation measures. In terms of the relationship between HEC and land use, they have placed emphasis on the impact of actual crop losses caused by Asian elephants on the behaviors of farmers [55]. However, research has revealed that the HEC risk perception by farmers is higher than the actual losses they incur [40,60]. Previous studies have overlooked the psychological perception of farmers and its influence on land-use behaviors and efficiency. Building on this foundation, this study adopts a sociological and psychological perspective, focusing on the Asian elephant distribution area in Pu’er City, Yunnan Province, with farmers as the research subjects. It aims to explore the impact of HEC risk perception on farmers’ land use efficiency and its underlying mechanisms, providing a basis for mitigating HEC and enhancing farmers’ land use efficiency. The research will specifically address the following questions: (1) What is the level of farmers’ risk perception regarding HEC? (2) How does HEC risk perception affect farmers’ land use efficiency? (3) What is the mechanism through which HEC risk perception influences farmers’ land use efficiency?

The innovations of this paper lie in the following aspects: (1) This is the first time that risk perception theory has been combined with the theory of planned behavior to analyze the impact mechanism of HEC on land use efficiency. (2) The moderating effect of social networks is introduced to reveal the reinforcing role of group behavior on individual decisions. (3) The mediating effect model is used to verify the key path of agricultural capital input.

2. Theoretical Analysis and Research Hypotheses

2.1. HEC Risk Perception and Farmers’ Land Use Efficiency

HEC refers to the negative impacts arising from interactions between humans and elephants [61], primarily manifested as economic losses suffered by farmers, crop damage, threats to personal safety, and farmers’ psychological perception of potential threats posed by Asian elephants [62,63]. The concept of risk perception originates from the field of psychology, denoting the process through which individuals assess risks of uncertain events based on intuitive judgments and subjective feelings [38]. Farmers’ risk perception specifically refers to their subjective evaluation and cognition of potential adverse effects in agricultural production, such as judgments about risks posed by wildlife conflicts, natural disasters, price fluctuations, and plant diseases to crop production [64].

In terms of risk formation mechanisms, HEC differs from conventional agricultural risks in three key aspects: the duality of threats (intertwined material losses and psychological stress), the predictability of risks (seasonal patterns in elephant activity), and constraints on countermeasures (limited by wildlife conservation policies) [40]. Building on Slovic’s risk perception theory [38], farmers’ cognition of such risks is not merely a cost-benefit calculation but a complex process integrating rational judgment and emotional responses. At the rational level, farmers assess risk probability through historical encounter frequency, damage severity, and prevention efficacy. At the emotional level, they develop a “fear–anger–helplessness” emotional memory, where psychological trauma significantly amplifies risk sensitivity [17,65].

The theory of planned behavior provides a dynamic perspective for understanding farmers’ decision-making. According to Ajzen’s theoretical framework [66,67], risk perception constitutes a core factor influencing farmers’ adaptive decisions, with land use decisions following a “perception–evaluation–action” logical chain [68,69,70]. Lorenzoni et al. further proposed a three-stage theoretical model, hypothesizing that individuals’ behavioral–psychological processes regarding environmental issues involve (1) perceiving environmental changes, (2) assessing the impacts of these changes, and (3) implementing adaptive adjustments [71]. When exposed to risks, individuals’ psychological states become significantly affected, typically manifesting risk-averse behaviors to alleviate anxiety and stress [57]. Varying levels of risk perception lead to distinct adaptation strategies, with farmers exhibiting heightened sensitivity to environmental changes more likely to take proactive measures [72].

Given the inherently risk-averse nature of most agricultural producers [73], farmers often avoid high-return activities accompanied by elevated risks, making defensive production strategies more likely to be triggered under heightened risk perception [74,75]. Furthermore, risk perception theory suggests that farmers make risk-related decisions (acceptance, avoidance, transfer, or control) based on expected benefits and perceived risk levels [76]. The greater the perceived danger, the more actively farmers implement preventive measures to minimize risks [77,78]; conversely, they may neglect risk management when perceived threats are low [79]. To mitigate losses from HEC, farmers tend to adjust agricultural practices [49] and reduce production inputs [54]. Ultimately, farmers’ HEC risk perception fundamentally shapes their land use decisions and behaviors, thereby critically influencing land use efficiency. Based on this, the following hypotheses are proposed in this paper:

Hypothesis 1 (H1).

The HEC risk perception has a significant negative impact on the land use efficiency of farmers.

Hypothesis 2 (H2).

The HEC risk perception reduces land use efficiency by prompting farmers to reduce agricultural capital investment.

2.2. Social Network, HEC Risk Perception, and Farmers’ Land Use Efficiency

In rural risk management, social networks serve dual functions as both information amplification mechanisms and behavioral synchronization catalysts. Social networks are relatively stable systems constructed based on social relationships between individuals. In rural areas, they typically exhibit relationship characteristics such as “kinship”, “geographical proximity”, and “acquaintance”. These networks influence social behavior through the connections among their members [80,81]. Members in social network relationships share resources and tend to have convergent behaviors [82]. Individuals’ social relationships exist in multiple circles, which are relatively independent yet partially integrated [83]. Rural society in China is a relationship-oriented society, where traditional kinship, blood-related, and geographical relationships [84] are interwoven with emerging occupational relationships [85], forming the social networks within Chinese villages. Relationships based on blood ties and geography are more intimate with individuals and have a more significant impact on individual decision-making [86]. According to the theory of social embeddedness, when individuals make decisions, they are influenced not only by the considerations of the “rational economic man” in economics but also by the social networks in which they are embedded. In economically underdeveloped and geographically secluded rural areas, farmers often conform to others due to limited information channels. In daily life, to reduce decision-making risks caused by incomplete information, people tend to learn from others’ decisions or behaviors to reduce uncertainty [87,88]. Farmers form risk perceptions based on their own experiences of encountering Asian elephants and the information obtained from other farmers. When the social network has a significant impact on farmers’ risk perception, risk information will spread rapidly among farmers. Farmers who are spatially close and have close social network connections are more likely to make the same decisions when facing the risks of human–elephant conflict [89], further exacerbating the negative impact of human–elephant conflict risk perception on land use efficiency. Among them, groups related by blood ties and geography are the main subjects of conformity [90]. Although with the acceleration of urbanization, the original social networks in Chinese rural areas are showing a trend of change [91], overall, farmers’ behavioral choices are still closely related to the decisions of their relatives and friends [92].

Hypothesis 3 (H3).

Social networks positively moderate the negative impact of HEC risk perception on the land use efficiency of farmers.

2.3. Theoretical Analytical Framework

Human–elephant conflict, as a unique type of ecological and environmental risk, not only causes direct economic losses and threats to the personal safety of rural households but also affects their land-use decisions through complex psychological mechanisms. Based on the theories of risk perception and planned behavior, this study constructs a theoretical analytical framework of “risk perception–behavioral adjustment–efficiency change”, as shown in Figure 1.

3. Materials and Methods

3.1. Research Area

Asian elephants are mainly distributed in Yunnan Province, China, with a population of around 300, primarily found in the Xishuangbanna Dai Autonomous Prefecture, Pu’er City, and Lincang City. This study selects the Asian elephant activity area in Pu’er City as the research region, which involves the administrative regions of Jinghong City, Menghai County, Mengla County in Xishuangbanna Prefecture, Jiangcheng County, Lancang County, Ning’er County, Simao District in Pu’er City, and Cangyuan County in Lincang City (Figure 2).

Pu’er City, also known as Simao, is situated in the southwestern region of Yunnan Province. It borders Honghe and Yuxi to the east, Xishuangbanna to the south, Lincang to the northwest, Dali and Chuxiong to the north. The city shares borders with Vietnam and Laos to the southeast, and Myanmar to the southwest. With a total area of about 45,385 square kilometers, it is the largest city (prefecture) in Yunnan Province in terms of area. Pu’er City is a grand view garden of national cultures, with multiple indigenous ethnic groups, including the Hani, Yi, and Dai ethnic groups. The permanent population is about 2.37 million people. Pu’er City is influenced by the subtropical monsoon climate, with most areas frost-free throughout the year, no severe cold in winter, and no extreme heat in summer. It is known as the “Green Sea Pearl” and “Natural Oxygen Bar”. Pu’er City is rich in water energy resources, with a forest coverage rate of over 74.59%. It is an important coffee production area and distribution center in China, as well as one of the important birthplaces of Pu’er tea. It is one of the largest tea-producing areas in China and is praised as “embracing gold and bearing treasures”.

Pu’er City is one of the important habitats for wild Asian elephants. Currently, there are 181 wild Asian elephants active within the city, with their activity areas covering 7 counties (districts), 31 townships, and 138 natural villages, accounting for more than 60% of the Asian elephant population in China. With the increase of the Asian elephant population and the expansion of their activity range, HECs have become increasingly frequent, mainly manifested in crops being trampled, tea trees being damaged, houses being destroyed, and even threatening the personal safety of villagers (Figure 3). According to statistics, in 2021, the direct economic loss caused by Asian elephant incidents in Pu’er City reached 6.631 million yuan, among which the loss of crops was the main part (6.631 million yuan), the loss of livestock was 72,000 yuan, and the compensation for personal casualties exceeded 800,000 yuan [93]. The frequent intrusion of elephant herds not only destroys agricultural production but also threatens the lives of farmers [94]. The combined cost of psychological trauma and economic loss has led to insufficient satisfaction of farmers with Asian elephant protection policies. On the one hand, the activity range of Asian elephants is constantly expanding, and they frequently enter human activity areas, resulting in the destruction of crops, damage to houses, and even threats to the personal safety of villagers, which seriously affects the production and life of local residents. On the other hand, the problem of habitat fragmentation is prominent, and factors such as infrastructure construction further compress the living space of Asian elephants, making their activity areas highly overlap with human activity areas, and intensifying the frequency and intensity of conflicts. In order to alleviate HEC, Pu’er City has taken a series of measures. However, in the actual implementation, there are still many challenges, such as insufficient insurance compensation, high maintenance costs of anti-elephant facilities, and slow progress in habitat restoration. Therefore, how to protect Asian elephants while effectively alleviating HEC and safeguarding the production and life rights and interests of local residents remains an important issue to be solved in Pu’er City.

3.2. Data Sources

Data were collected through face-to-face household surveys carried out by the research group in the Asian elephant activity areas of Pu’er City, Yunnan Province from July to August 2023. Through questionnaires and in-depth interviews, the local HEC and the land-use situation of farmers were understood. The questionnaires for farmers mainly covered the basic information of the respondents and their families, family land and production situations, family economic situations, and HEC situations. The survey team consisted of instructors, postgraduate students, and staff from nature reserves. Before the start of the survey, the questionnaire designers trained the team members on the content of the questionnaire to ensure a consistent understanding of the questionnaire content. The survey objects were the heads of households or family members who were more familiar with the relevant situations. All surveys were completed in the form of one-on-one interviews between the investigators and farmers, and the survey time for each household was 40–60 min. The survey adopted a multi-stage stratified random sampling method. Based on the preliminary research foundation and the typicality of HEC, a total of 4 counties (districts) were selected. Each county (district) randomly selected 2–5 towns, each town randomly selected about 2 administrative villages, each administrative village randomly selected about 2 natural villages, and each natural village randomly selected about 10 farmers. After eliminating the samples with missing key information and outliers, a total of 449 valid questionnaires were obtained, including 192 in Simao District, 113 in Ning’er County, 75 in Jiangcheng County, and 69 in Lancang County (

Table 1. Sample distribution in the survey area.

Research Area	Survey Counties (Districts)	Townships	Sample Size
Pu’er City (449)	Simao District (192)	Yixiang Town	43
		Yunxian Town	42
		Simao Port Town	41
		Liushun Town	35
		Nanping Town	31
	Ning’er County (113)	Mengxian Town	50
		Puyi Town	33
		Mohei Town	30
	Jiangcheng County (75)	Kangping Town	38
		Zhengdong Town	37
	Lancang County (69)	Fazhanhe Town	40
		Sakai Town	29

). The overall effective rate was 87.87%.

3.3. Variable Selection

3.3.1. Dependent Variable

The dependent variable is the land use efficiency of farmers. Land use efficiency is a key indicator for evaluating farmers’ agricultural land management activities, reflecting the proportional relationship between input and output [95]. In this paper, land use efficiency is defined as the output value per unit of land area [96].

3.3.2. Core Independent Variable

The core independent variable is the HEC risk perception. An evaluation index system for the HEC risk perception is designed from three dimensions: perception of damage to life and property, perception of agricultural production losses, and perception of mental health harm (

Table 2. Composition of farmers’ risk perception indicators.

Dimension	Measurement Items	Code	Mean	Standard Deviation
Perceived loss in agricultural production	Losses of food crops caused by Asian elephants	A1	3.454	1.395
	Losses of cash crops caused by Asian elephants	A2	3.450	1.410
	Problems of agricultural disruption caused by Asian elephants	A3	3.194	1.366
Perceived damage to living property	Asian elephants entering houses to eat grains	L1	2.196	1.413
	Vehicle losses caused by Asian elephants	L2	2.020	1.437
	House losses caused by Asian elephants	L3	2.365	1.455
Perceived psychological health harm	Whether to feel nervous and anxious about Asian elephant attacks	P1	4.094	0.916
	Whether to hate and loathe Asian elephant attacks	P2	3.864	1.038
	Degree of fear of Asian elephants	P3	3.546	1.279

). Each dimension includes three measurement items, with values ranging from 1 to 5. Respondents who answer “none” are assigned a value of 1, “not serious” a value of 2, “moderate” a value of 3, “serious” a value of 4, and “very serious” a value of 5. Specifically, agricultural production loss includes crop loss caused by Asian elephants, loss of cash crops, and farming issues. Damage to living property includes Asian elephants entering houses to eat grain, vehicle damage caused by Asian elephants, and house damage. Psychological health harm includes rural households’ tension and anxiety over Asian elephant incidents, their hatred and aversion to Asian elephants, and their level of fear of Asian elephants. The principal component analysis (PCA) is used to extract three common factors, and a composite score reflecting the degree of HEC is computed.

3.3.3. Moderating Variable

The moderating variable in this study is social networks. In 1973, Granovetter [97] proposed the “weak ties theory”, categorizing social networks into strong and weak ties based on interaction frequency, relational closeness, and reciprocity. Strong-tie networks consist of kinship groups with blood ties and geographical proximity, demonstrating high trust and stable interest-based bonds. Weak-tie networks primarily comprise friends connecting heterogeneous groups, featuring less stable interest-based connections and potential information distortion [98]. Drawing on prior research, this paper employs strong/weak-tie networks to operationalize social networks: “The number of relatives and friends from whom one can borrow money” measures strong social networks, while “the frequency of mutual assistance between the family and other villagers” captures weak social networks (

Table 3. Index of social networks.

First-Level Indicator	Second-Level Indicators	Third-Level Indicators	Assignment	Mean	Standard Deviation
Social networks	Strong social network	The number of relatives and friends from whom one can borrow money	Very rarely = 1, Rarely = 2, Moderately = 3, Frequently = 4, Very frequently = 5	3.381	1.236
	Weak social network	The frequency of mutual assistance between the family and other villagers		4.125	0.862

).

3.3.4. Mediating Variable

The mediating variable in this paper is agricultural capital inputs, which refers to all kinds of capital elements such as seeds, seedlings, pesticides, fertilizers, agricultural machinery, and agricultural production-related services that farmers invest per unit area of land during the agricultural production process.

3.3.5. Instrumental Variable

Farmers’ HEC risk perception may in turn be affected by their land use efficiency, with farmers who have lower land use efficiency possibly having a stronger perception of the risk of HEC. Instrumental variables can fully suppress reverse causality and effectively address self-selection bias. Drawing on the research of Zhang [99] and Ma [100], the “average HEC risk perception in the village” (excluding the individual farmer) is selected as the instrumental variable for the endogeneity test. Based on the theory of farmers’ behavioral imitation, the average level of other farmers’ HEC risk perception in the village (excluding the individual farmer) can affect the individual’s risk perception but has no direct relationship with the individual’s land use efficiency. In theory, it meets the relevance and exogeneity conditions of an instrumental variable.

3.3.6. Control Variables

In addition to the core variables, to reduce the estimation bias caused by omitted variables, this paper controls variables that have been widely proven to have a significant impact on farmers’ land use efficiency from the aspects of farmers’ individual characteristics, household characteristics, and external environmental characteristics. Farmers’ individual characteristics include gender, education level, health status, and experience as a village cadre. Household characteristics include family size, proportion of labor force, average age of the labor force, part-time employment situation of farmers, and whether to join an agricultural organization. External environmental characteristics include the distance from the family residence to the core area of the county town, whether having received early warning information about wildlife damage, Whether having suffered damage from other wild animals, and land fragmentation degree. The descriptions, value assignments, and descriptive statistics of each variable are shown in

Table 4. Variable descriptive statistics.

Variable Name	Variable Description and Assignment	Mean	Standard Deviation
Dependent variable
Farmers’ land use efficiency	Output value per unit land area (CNY/mu)	1164.974	949.7175
Core independent variable
HEC risk perception	Obtained from factor analysis	0.000	0.579
Moderating variable
Social networks	Take the average of the sum of the above two indicators	3.753	0.808
Mediating variable
Agricultural capital inputs	Total agricultural capital inputs/land area	646.211	905.753
Control variables
Gender	Female = 0, Male = 1	0.679	0.467
Education level	Primary school or below = 1, Junior high school = 2, Senior high school or secondary vocational school = 3, Junior college = 4, Undergraduate or above = 5	1.606	0.725
Health status	Serious illness = 1, Minor illness = 2, Average = 3, Healthy = 4, Very healthy = 5	3.849	1.081
Experience as a village cadre	No = 0, Yes = 1	0.176	0.381
Family size	Number of family members (persons)	4.477	1.556
Proportion of labor force	Number of labor force/family size (%)	0.681	0.238
Average age of the labor force	Years old	43.53	8.203
Part-time employment situation of farmers	Pure farming = 1, Farming and part-time work = 2, Side-line occupations and farming = 3, Non-agricultural occupation = 4	1.425	0.782
Whether to join an agricultural organization	No = 0, Yes = 1	0.163	0.369
Distance from the family address to the core area of the county town	Kilometers, take the logarithm	4.041	0.446
Whether having received early warning information about wildlife damage	No = 0, Yes = 1	0.933	0.25
Whether having suffered damage from other wild animals	No = 0, Yes = 1	0.129	0.336
Land fragmentation degree	Average plot area (mu/plot)	10.412	9.366
Instrumental variable
Village-level average HEC risk perception	The average level of HEC risk perception among other farmers in the same natural village	0	0.368

.

3.4. Research Methods

3.4.1. Principal Component Analysis

Principal component analysis (PCA) describes or explains the whole by using a few components composed of the main parts of the original variables [101]. The goal of the evaluation is to optimize and simplify multivariate data, thereby achieving the effects of dimensionality reduction, simplification, and interpretation of the data [102]. The advantage of this method is that it can eliminate the interrelationships between indicator samples while retaining most of the information of the original variables [103]. Based on this, a small number of representative main indicators, namely the principal components, can be extracted.

Since the indicators of HEC risk perception have consistent units of measurement, standardization is not performed. The specific calculation steps are as follows:

Step 1: Calculate the covariance matrix to determine the relationships and trends among variables. Sometimes variables are highly correlated, resulting in redundant information. Covariance can measure the correlation between two-dimensional random variables.

$$\mathsf{\Sigma }={\displaystyle \frac{{(\mathrm{X}-\overline{\mathrm{X}})}^{\mathrm{T}}(\mathrm{X}-\overline{\mathrm{X}})}{\mathrm{n}-1}}$$

(1)

In Equation (1), n is the number of samples, X is the data matrix, $\overline{\mathrm{X}}$ is the mean matrix of X, ${(\mathrm{X}-\overline{\mathrm{X}})}^{\mathrm{T}}$ is the transpose of $(\mathrm{X}-\overline{\mathrm{X}})$, and Σ is an m*m matrix (where m is the number of variables).

Step 2: Calculate the eigenvalues and eigenvectors. Perform eigenvalue decomposition on the covariance matrix $\mathrm{Z}$:

$$\mathrm{Z}{\mathrm{v}}_{\mathrm{i}}={\mathsf{\lambda }}_{\mathrm{i}}{\mathrm{v}}_{\mathrm{i}}$$

(2)

In Equation (2), ${\mathrm{v}}_{\mathrm{i}}$ is the eigenvector, representing the direction of the principal component; and ${\mathsf{\lambda }}_i$ is the largest eigenvalue, indicating the variance contribution of each principal component.

Step 3: Select the principal components. Based on the magnitude of the eigenvalues, select the top p largest eigenvalues and their corresponding eigenvectors. Typically, principal components are chosen such that their cumulative contribution rate reaches 80–95%. The variance contribution rate is calculated as follows:

$${\mathsf{\alpha }}_{\mathrm{j}}={\displaystyle \frac{{\mathsf{\lambda }}_{\mathrm{j}}}{{\sum }_{\mathrm{i}=1}^{\mathrm{m}}{\mathsf{\lambda }}_{\mathrm{i}}}}$$

(3)

$${\mathrm{a}}_{\mathrm{p}}={\sum }_{\mathrm{j}=1}^{\mathrm{p}}{\mathsf{\alpha }}_{\mathrm{j}}$$

(4)

In Equations (3) and (4), ${\mathsf{\alpha }}_{\mathrm{j}}$ is the variance contribution rate of the j-th principal component, and ${\mathrm{a}}_{\mathrm{p}}$ is the cumulative variance contribution rate of the p principal components.

Step 4: Calculate the principal component scores. Using the selected eigenvectors, compute the coordinates of each sample in the principal component space, which are the principal component scores. These scores are then combined into a composite indicator for easier comparison. The composite evaluation value is derived by weighting and summing the principal components based on their variance contribution rates relative to the cumulative variance contribution rate. The formula is as follows:

$$\mathrm{Y}=\mathrm{X}{\mathrm{v}}_{\mathrm{p}}$$

(5)

$$\mathrm{F}={\sum }_{\mathrm{j}}^{\mathrm{p}}{\displaystyle \frac{{\mathsf{\alpha }}_{\mathrm{j}}}{{\mathrm{a}}_{\mathrm{p}}}}{\mathrm{F}}_{\mathrm{j}}$$

(6)

In Equations (5) and (6), v_p is the matrix composed of the first p eigenvectors, and Y is the principal component score matrix. ${\mathrm{F}}_{\mathrm{j}}$ is the score of the principal component, and F denotes the composite evaluation value.

3.4.2. Tobit Model

Some farmers’ land use efficiency may reach the efficiency boundary of 0 due to severe HEC. Conventional regression methods cannot explain the differences in properties between the limit values and non-limit observations. The Tobit model is a generalized linear model mainly used to handle the modeling problems of left- and right-censored data and perform regression on continuous dependent variables. Therefore, drawing on existing research [104], this paper selects the Tobit model with a restricted dependent variable for estimation. Meanwhile, referring to Wen Zhonglin’s research on the moderating effect [105], an interaction term between the HEC risk perception and the social network is added to the model to examine whether there is a moderating effect of the social network in the process of the HEC risk perception affecting farmers’ land use efficiency. The specific model is set up as follows:

$${\mathrm{Y}}_{\mathrm{i}}=\left\{\begin{array}{c}{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}={\mathsf{\alpha }}_0+{\mathsf{\beta }}_1{\mathrm{C}}_{\mathrm{i}}+\sum {\mathsf{\gamma }}_{\mathrm{j}}{\mathrm{Z}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}},{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}>0\\ 0, {\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}\le 0\end{array}\right.$$

(7)

$${\mathrm{Y}}_{\mathrm{i}}=\left\{\begin{array}{c}{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}={\mathsf{\alpha }}_1+{\mathsf{\beta }}_2{\mathrm{C}}_{\mathrm{i}}+{\mathsf{\beta }}_3{\mathrm{S}}_{\mathrm{i}}+{\mathsf{\beta }}_4{{\mathrm{C}}_{\mathrm{i}}\times \mathrm{S}}_{\mathrm{i}}+\sum {\mathsf{\gamma }}_{\mathrm{j}}{\mathrm{Z}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}},{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}>0\\ 0, {\mathrm{Y}}_{\mathrm{i}}^{\mathrm{*}}\le 0\end{array}\right.$$

(8)

In Equations (7) and (8), ${\mathrm{Y}}_{\mathrm{i}}$ is the land use efficiency [106], ${\mathrm{C}}_{\mathrm{i}}$ is the HEC risk perception, ${\mathrm{Z}}_{\mathrm{i}\mathrm{j}}$ is the control variable, including variables related to the farmer’s individual characteristics, household characteristics, and external environmental characteristics. ${\alpha }_0{, \alpha }_1,{\beta }_1,{\beta }_2,{\beta }_3,{\beta }_4,{\gamma }_j$ are constant terms, and ${\epsilon }_i$ is the random disturbance term.

In terms of testing the influence mechanism, the research by Qiu Junjie et al. [107] shows that agricultural capital input has a positive impact on farmers’ land use efficiency. Therefore, referring to the research by Jiang Ting [108], this paper focuses on identifying the impact of HEC on farmers’ agricultural capital input. Since agricultural capital input is a continuous variable, this paper uses the OLS model for estimation, and the constructed model is as follows:

$${\mathrm{P}}_{\mathrm{i}}={\mathsf{\alpha }}_2+{\mathsf{\beta }}_5{\mathrm{C}}_{\mathrm{i}}+\sum {\mathsf{\gamma }}_{\mathrm{j}}{\mathrm{Z}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(9)

In Equation (9), ${\mathrm{P}}_{\mathrm{i}}$ represents the dependent variable in the mechanism analysis, that is, agricultural capital input. ${\mathsf{\alpha }}_2,{\mathsf{\beta }}_5,{\mathsf{\gamma }}_{\mathrm{j}}$ are constant terms, and ${\mathsf{\epsilon }}_{\mathrm{i}}$ is the random disturbance term. If ${\mathsf{\beta }}_5$ in Equation (3) passes the significance test and is negative, it is considered that the HEC risk perception will lead farmers to reduce agricultural capital input. Combined with the discussions in existing research, it can be proved that the research hypothesis H2 in this paper holds.

3.4.3. Ordinary Least Squares

Since agricultural capital input is a continuous variable, ordinary least squares (OLS) is employed for estimation. Ordinary least squares (OLS) is the method used in basic regression analysis. To estimate the impact of HEC on the behavior of agricultural capital input, the model is set up as follows using OLS:

$${\mathrm{I}}_{\mathrm{i}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{C}}_{\mathrm{i}}+\sum {\mathsf{\gamma }}_{\mathrm{j}}{\mathrm{Z}}_{\mathrm{i}\mathrm{j}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(10)

In Equation (10), ${\mathrm{I}}_{\mathrm{i}}$ represents the dependent variable in the mechanism analysis, namely agricultural capital input. ${\mathrm{C}}_{\mathrm{i}}$ is the HEC risk perception. ${\mathrm{Z}}_{\mathrm{i}\mathrm{j}}$ is the control variable, which includes individual characteristics of the farmer, household characteristics, and external environmental variables. $\mathsf{\alpha }$ and $\mathsf{\gamma }$ are constant terms, and ${\mathsf{\epsilon }}_{\mathrm{i}}$ is the random disturbance term. If ${\mathsf{\alpha }}_1$ in Equation (4) passes the significance test and is negative, it is concluded that HEC inhibits agricultural capital input by farmers.

4. Results

4.1. Measurement of HEC Risk Perception

Before empirically examining the relationship between HEC and the land use efficiency of rural households, this paper first constructs an evaluation system for HEC indicators. It then employs PCA to measure the aforementioned indicators and analyzes the results of the measurement.

Specifically, adhering to the principles of practicality and comprehensiveness in the construction of indicators, this paper designs an HEC indicator system from the perspective of rural households’ risk perception based on the connotation and basic characteristics of HEC. The indicator system is constructed from three dimensions: agricultural production loss, damage to living property, and psychological health harm (Table 4). Each dimension includes three measurement items, with values ranging from 1 to 5. Respondents who answer “none” are assigned a value of 1, “not serious” a value of 2, “moderate” a value of 3, “serious” a value of 4, and “very serious” a value of 5. Specifically, agricultural production loss includes crop loss caused by Asian elephants, loss of cash crops, and farming issues. Damage to living property includes Asian elephants entering houses to eat grain, vehicle damage caused by Asian elephants, and house damage. Psychological health harm includes rural households’ tension and anxiety over Asian elephant incidents, their hatred and aversion to Asian elephants, and their level of fear of Asian elephants.

In the text, the structural validity of the indicators is tested using the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and the Bartlett test of sphericity. The KMO ranges from 0 to 1. The closer it is to 1, the stronger the correlation between variables, and the more suitable it is for principal component analysis. The KMO should be at least greater than 0.5; when the p-value of the Bartlett test of sphericity is ≤0.01, it indicates that principal component analysis is appropriate. As can be seen from Table 2, the KMO for the measure of sampling adequacy is 0.808, and the p-value for Bartlett test of sphericity is 0.000, which shows that the data are suitable for principal component analysis (

Table 5. Results of KMO and the Bartlett test.

Validity Test	Index	Results
KMO	-	0.808
Bartlett test	Chi-square	2601.716
	Degrees of freedom	36
	p-value	0.000

).

Secondly, employing PCA and using STATA 17.0 software, an exploratory factor analysis was conducted on the data, with principal components determined by eigenvalues greater than 1. According to the selection criteria, the first three principal components were extracted, with a cumulative variance contribution rate of 80.29% (

Table 6. Eigenvalues, variance contribution rates, and cumulative variance contribution rates in the principal component model.

Principal Component Factors	Eigenvalues	Variance Contribution Rates	Cumulative Variance Contribution Rates
1	4.3722	0.4858	0.4858
2	1.8025	0.2003	0.6861
3	1.0510	0.1168	0.8029
4	0.5543	0.0616	0.8644
5	0.3450	0.0383	0.9028
6	0.2876	0.0320	0.9347
7	0.2605	0.0289	0.9637
8	0.2147	0.0239	0.9875
9	0.1122	0.0125	1.0000

), indicating that the extracted principal components contain most of the information in the sample data. The three principal components were named Composite Factor I, Composite Factor II, and Composite Factor III, with eigenvalues of 4.3722, 1.8025, and 1.0510, respectively.

Using the varimax orthogonal rotation method, the rotated factor loading matrix was obtained (

Table 7. Orthogonal rotated factor loading matrix.

Indicator	Factor Loading			Communality
	F1	F2	F3
A1	0.9163	0.2301	0.0969	0.0980
A2	0.9027	0.2583	0.1027	0.1079
A3	0.8309	0.3058	0.1008	0.2059
L1	0.3040	0.8610	0.1223	0.1514
L2	0.2136	0.8580	0.0869	0.2106
L3	0.3455	0.8184	0.1339	0.1930
P1	0.0870	0.0896	0.9064	0.1628
P2	0.0653	0.1218	0.8325	0.2879
P3	0.1820	0.1272	0.7706	0.3569

). The principal components are linear combinations of the original nine indicators, with the weights of each indicator being the corresponding principal component eigenvectors, which represent the degree of influence of each individual indicator on the principal components. Referring to the study by Lin Haiming et al. [109], and based on the data in the matrix, the following principal component score models were calculated:

$${\mathrm{F}}_1=0.4478{\mathrm{A}}_1+0.4300{\mathrm{A}}_2+0.3725{\mathrm{A}}_3-0.1207{\mathrm{L}}_1-0.1697{\mathrm{L}}_2-0.0832{\mathrm{L}}_3+0.0085{\mathrm{P}}_1-0.0436{\mathrm{P}}_2-0.0618{\mathrm{P}}_3$$

(11)

$${\mathrm{F}}_1=-0.1674{\mathrm{A}}_1-0.1451{\mathrm{A}}_2-0.0901{\mathrm{A}}_3+0.4435{\mathrm{L}}_1+0.4753{\mathrm{L}}_2+0.4010{\mathrm{L}}_3-0.0514{\mathrm{P}}_1-0.0567{\mathrm{P}}_2-0.0216{\mathrm{P}}_3$$

(12)

$${\mathrm{F}}_1=-0.0373{\mathrm{A}}_1-0.0361{\mathrm{A}}_2-0.0364{\mathrm{A}}_3-0.0406{\mathrm{L}}_1-0.0519{\mathrm{L}}_2-0.0336{\mathrm{L}}_3+0.3665{\mathrm{P}}_1+0.4457{\mathrm{P}}_2+0.4067{\mathrm{P}}_3$$

(13)

Based on the scores of the above three factors, calculate the comprehensive index of HEC. The explanatory abilities of the three common factors for the comprehensive index of HEC vary. The weight of each factor is determined according to the proportion of its variance contribution rate to the cumulative variance contribution rate, and a comprehensive index score equation for the degree of HEC can be constructed:

$$\mathrm{F}=0.3663{\mathrm{F}}_1+0.3320{\mathrm{F}}_2+0.3015{\mathrm{F}}_3$$

(14)

From the previous analysis, it can be seen that the comprehensive factor with the greatest impact on the degree of HEC is Factor One, with a contribution rate of 0.4858. The coefficients, in descending order, are A₁, A₂, A₃, L₃, L₁, L₂, P₁, P₂, and P₃. The indicators of losses of food crops and cash crops caused by Asian elephants are the highest, which has severely impacted the agricultural production on which local residents depend for their livelihood. The extensive damage to crops not only directly reduces the economic income of farmers but also affects food security, further intensifying the conflict between humans and elephants and becoming a key factor driving the escalation of the conflict.

The comprehensive factor ranking second in terms of its impact on the degree of HEC is Factor Two, with a contribution rate of 0.2003. The coefficients, in descending order, are L₁, L₂, L₃, A₃, A₂, A₁, P₃, P₂, and P₁. The indicators of vehicle and house losses caused by Asian elephants have the highest impact, indicating that when Asian elephants enter human residential areas, they directly damage vehicles and houses, affecting the daily travel and production activities of farmers. This not only brings direct property losses to residents but also greatly affects the farmers’ daily travel, production activities, and sense of security in their lives.

The comprehensive factor ranking third in terms of its impact on the degree of HEC is Factor Three, with a contribution rate of 0.1168. The coefficients, in descending order, are P₁, P₂, P₃, L₃, L₁, A₂, A₃, A₁, and L₂. The impact of the tension, anxiety, disgust, and resentment emotions of farmers due to the incidents caused by Asian elephants is the highest, indicating that the harm to mental health is also an aspect that cannot be ignored in HEC. The long-term HEC keeps farmers in a state of mental stress, worrying about the safety of themselves and their families as well as property losses. This psychological pressure may affect their normal lives and enthusiasm for production.

4.2. Benchmark Regression Analysis

Before estimating the model, the variance inflation factor (VIF) was used to conduct a multicollinearity test. Results indicate that the VIF values of the independent variables in this model estimation range from 1.03 to 1.47, which are far less than the threshold of 10. Therefore, there is no severe multicollinearity. Then, Stata 16.0 was used to estimate the Tobit model (

Table 8. Estimation results of the Tobit model.

Variable Name	Farmers’ Land Use Efficiency
	(1)		(2)
	Coefficient	Standard Error	Coefficient	Standard Error
HEC risk perception	−252.759 ***	76.498	−240.045 ***	75.882
Social network			136.208 **	53.319
HEC risk perception × social network			−196.643 **	98.905
Gender	136.872	94.139	97.259	94.135
Education level	96.579	64.010	88.339	63.411
Health status	41.251	41.959	333.624	41.736
Experience as a village cadre	127.935	117.701	135.359	116.764
Family size	17.101	33.152	21.402	32.849
Proportion of labor force	144.704	208.078	229.623	207.890
Average age of the labor force	−9.682 *	5.689	−9.900 *	5.632
Part-time employment situation of farmers	−152.891 ***	57.157	−139.220 ***	56.769
Whether to join an agricultural organization	79.272	116.644	75.745	115.553
Distance from the family address to the core area of the county town	13.644	99.878	12.692	99.276
Whether having received early-warning information about wildlife damage	277.260	174.591	287.369 *	173.360
Whether having suffered damage from other wild animals	−130.397	129.801	−140.591	128.494
Land fragmentation degree	−30.854 ***	4.714	−31.152 ***	4.669
Constant	1198.387 **	585.675	1175.915 **	580.372
Log-likelihood	−3598.467		−3593.527
Prob > chi2	0.0000		0.0000
LR chi2	68.26		78.14
Pseudo R2	0.0094		0.0108
Observation	449		449

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% significance levels, respectively.

). According to the regression results in Column (1), the coefficient of the HEC risk perception is significantly negative at the 1% statistical level, indicating that the HEC risk perception can significantly reduce farmers’ land use efficiency. Specifically, for every one-unit increase in the HEC risk perception, the land use efficiency of farmers will decrease by 250.34 yuan per mu. From the perspective of resource allocation, in order to cope with HEC, farmers will divert resources originally intended for agricultural production to the construction of protective facilities and the employment of caregivers. This not only increases implicit costs but also reduces investment in core agricultural production. From the perspective of psychological cognition and behavioral motivation, the risk perception brought about by HEC can easily trigger negative emotions such as anxiety among farmers. These emotions interfere with their rational decision-making process, leading them to prefer conservative strategies. They are neither willing to easily try new technologies nor motivated to increase labor input. Based on this, Hypothesis H1a is verified.

To alleviate the potential collinearity problem in the model after introducing the interaction term, this paper centers the interaction term [110,111,112]. According to the regression results in Column (2), the coefficient of the interaction term between the HEC risk perception and the social network is negative and significant at the 5% statistical level, indicating that social networks play a positive moderating role in the process of HEC affecting farmers’ land use efficiency. Specifically, as the social network strengthens, the negative impact of HEC on farmers’ land use efficiency is further intensified. A possible reason is that a strong social network enables risk information to spread rapidly and be overly magnified. Due to increased fear, farmers make conservative decisions, pursuing risk avoidance rather than maximizing efficiency. The social network strengthens the herd mentality and group pressure, and farmers will imitate others’ land use patterns, further exacerbating the negative impact. Based on this, Hypothesis H2 is verified.

4.3. Robustness Test

Given that there may be sample selection bias and that farmers’ land use efficiency is affected by multiple factors, which may lead to problems such as omitted variables, it is necessary to conduct a robustness check on the above results. First, this paper chooses to replace the measurement method of the core explanatory variable. Specifically, it is measured based on farmers’ ratings of “the severity of Asian elephant damage”, where 1–5 represent “none”, “not serious”, “moderate”, “serious”, and “very serious”, respectively. The estimation results are shown in

Table 9. Robustness test (replacing core explanatory variables).

Variable Name	Farmers’ Land Use Efficiency
	(1)		(2)
	Coefficient	Standard Error	Coefficient	Standard Error
HEC risk perception	−110.973 ***	34.245	−86.674 **	34.820
Social network			121.090 **	53.560
HEC risk perception × social network			−104.572 **	45.863
Constant	1666.337 ***	591.887	1274.352 **	578.790
Control variables	Controlled		Controlled
Log-likelihood	−3598.672		−3593.755
Prob > chi2	0.0000		0.0000
LR chi2	67.85		77.69
Pseudo R2	0.0093		0.0107
Observation	449		449

Note: **, and *** indicate significance at the 5%, and 1% significance levels, respectively.

. From the results in Column (1) and Column (2), it can be seen that the coefficient of the HEC risk perception is significantly negative at the 1% statistical level, and the interaction term is significantly negative at the 5% statistical level, which proves the robustness of the previous estimation results.

Secondly, the robustness test is conducted by narrowing the sample interval (

Table 10. Robustness test (narrowing the sample interval).

Variable Name	Farmers’ Land Use Efficiency
	(1)		(2)
	Coefficient	Standard Error	Coefficient	Standard Error
HEC risk perception	−182.946 **	88.569	−179.661 **	87.894
Social network			86.798	64.354
HEC risk perception × Social network			−259.980 **	122.444
Constant			1436.893 **	662.535
Control variables	Controlled		Controlled
Log-likelihood	−2576.259		−2573.380
Prob > chi2	0.0000		0.0000
LR chi2	63.50		69.25
Pseudo R2	0.0122		0.0133
Observation	311		311

Note: ** indicate significance at the 5% significance levels.

). Most of the farmers in the survey area are full-time farmers. Among the sample farmers, there are 320 full-time farmers, accounting for 64.93%. Therefore, this paper selects the samples of full-time farmers for Tobit regression. Results indicate that the HEC risk perception is significantly negative at the 5% statistical level, and the coefficient of the interaction term is also significantly negative at the 5% statistical level, which is consistent with the results of the benchmark regression.

Furthermore, the OLS model is used to re-estimate the impact of the HEC risk perception on farmers’ land use efficiency (

Table 11. Robustness test (OLS).

Variable Name	Farmers‘ Land Use Efficiency
	(1)		(2)
	Coefficient	Standard Error	Coefficient	Standard Error
HEC risk perception	−243.487 ***	75.776	−232.648 ***	75.338
Social network			131.898 **	52.908
HEC risk perception × Social network			−172.430 *	97.839
Constant	1280.567 **	579.714	1260.080 **	575.799
Control variables	Controlled		Controlled
Prob > F	0.0000		0.0000
R-squared	0.1446		0.1619
Observation	449		449

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% significance levels, respectively.

). Results indicate that, compared with the Tobit model, there are no substantial changes in the coefficients and significance levels of the HEC risk perception and the interaction term. This indicates that the HEC risk perception leads to a decrease in farmers’ land use efficiency, and the social network strengthens its negative impact. The regression results are consistent with those of the benchmark regression, which proves that the empirical results of this paper are robust and further validates the research hypotheses of this paper.

4.4. Endogeneity Test

The main factors leading to endogeneity problems can be summarized into the following three aspects: (1) There is a reciprocal relationship between the HEC risk perception and the control variables. The HEC risk perception is often influenced by factors such as farmers’ personal characteristics, family characteristics, and the external environment. (2) There may be a reverse causal relationship between the HEC risk perception and farmers’ land use efficiency. In other words, the higher the land use efficiency of farmers, the lower their HEC risk perception may be. (3) Although this paper has made relatively comprehensive considerations in model construction and variable selection, there may still be cases of omitted variables and measurement errors, which in turn lead to estimation biases.

This study refers to the existing literature [113] and uses the instrumental variable method to correct potential endogeneity problems. Selecting an instrumental variable needs to follow two key conditions: first, the instrumental variable should be highly correlated with the endogenous explanatory variable in the model; second, the instrumental variable should be uncorrelated with the random error term in the model. Based on these two conditions for instrumental variable selection, this paper selects the “average HEC risk perception in the village” as the instrumental variable. Based on the theory of farmers’ behavioral imitation, the average level of the HEC risk perception among other farmers in the village (excluding the individual farmer) can affect the individual’s risk perception but has no direct relationship with the individual’s land use efficiency. In theory, it meets the relevance and exogeneity conditions of an instrumental variable.

Drawing on relevant research [114,115], this paper intends to use the 2SLS and IV-Tobit models to address the endogeneity problem in model estimation. The interaction term between the HEC risk perception and the social network is ignored in the model, and the results are shown in

Table 12. Endogeneity test.

Variable Name	2SLS				IV-Tobtit
	The First Stage		The Second Stage		The First Stage		The Second Stage
	Coefficient	Standard Error	Coefficient	Standard Error	Coefficient	Standard Error	Coefficient	Standard Error
HEC risk perception			−326.368 **	142.561			−317.528 **	146.677
Instrumental variable	0.850 ***	0.066			0.851 ***	0.065
Constant	0.333	0.320	1230.286 **	575.472	0.333	0.315	1158.159 *	591.235
Control variables	Controlled		Controlled		Controlled		Controlled
LM			122.96 ***
Wald F			163.67 ***
Observation	449		449		449		449

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% significance levels, respectively.

. To test the problem of under-identification, this paper calculates the estimated value of the LM statistic to be 122.96, with a p-value of 0.000, indicating that there is no under-identification problem. Secondly, for the test of the weak instrumental variable problem, this paper calculates the Cragg–Donald Wald F statistic. Its estimated value is 163.67, with a p-value of 0.000, which is far higher than the critical value for rejecting the hypothesis of weak instrumental variables at the 10% statistical level, strongly rejecting the null hypothesis of “instrumental variable redundancy”. Therefore, the instrumental variable selected in this paper has a strong explanatory power for the HEC risk perception among sample farmers, and the selection of the instrumental variable is reasonable.

In the 2SLS model, the first stage is to perform an ordinary least-squares regression on the HEC risk perception. The result is significant at the 1% statistical level, and the coefficient of the instrumental variable is positive. In the second stage of the model, results indicate that for every one-unit increase in the HEC risk perception, the land use efficiency of farmers decreases by 317.53 yuan per mu, and the absolute value of this coefficient is larger than the relative coefficient estimated in Table 12. This indicates that if the endogeneity of the HEC risk perception is not considered, its impact on farmers’ land use efficiency may be underestimated.

The IV-Tobit model takes into account both the sample selection bias and the endogeneity problem. The estimation results are basically consistent with those of other models, indicating that the model has strong stability. From the estimation results, the HEC risk perception significantly reduces farmers’ land use efficiency, which is consistent with the conclusion drawn from the benchmark regression.

4.5. Analysis of the Mechanism of Action

According to Jiang Ting’s research [108], in the test of the mediation effect in economics, the step-by-step method is not suitable, and the two-step method should be adopted instead. According to the two-step method, on the basis that the HEC risk perception has an impact on agricultural capital input, if it is theoretically valid that agricultural capital input affects land use efficiency, it means that agricultural capital input plays a positive mediating role between the HEC risk perception and land use efficiency. That is, there is an action path of “HEC risk perception → agricultural capital input → farmers’ land use efficiency”.

The results in

Table 13. Mechanism analysis.

Variable Name	Agricultural Capital Input
	Coefficient	Standard Error
HEC risk perception	−193.321 **	75.543
Constant	561.397	577.932
Control variables	Controlled
Prob > F	0.0083
R-squared	0.0653
Observation	449

Note: ** indicate significance at the 5% significance levels.

show that in terms of the intensity of agricultural production input, the coefficient of the HEC risk perception on agricultural capital input is −193.32, and it is significant at the 5% statistical level. This indicates that for every one-unit increase in the HEC risk perception, agricultural capital input will correspondingly decrease by 193.32 yuan per mu. A possible reason is that when farmers make agricultural production decisions, they expect that even if they increase input, they may not get corresponding returns due to elephant damage. Therefore, as the HEC risk perception increases, they will rationally reduce the intensity of input in agricultural production to avoid potential losses, which in turn leads to a decline in their land use efficiency. Based on this, Hypothesis H1b is verified.

5. Discussion

5.1. Findings

This study, grounded in risk perception theory and the theory of planned behavior, empirically analyzes the impact of HEC risk perception on farmers’ land use efficiency and its underlying mechanisms. Results indicate that HEC risk perception significantly reduces farmers’ land use efficiency, primarily by inhibiting agricultural capital investment. This suggests that HEC not only causes direct economic losses to farmers but also exerts a profound negative impact on land use efficiency by altering farmers’ production behaviors and decision-making patterns.

The results of the baseline regression reveal a negative correlation between HEC risk perception and farmers’ land use efficiency. Specifically, for every unit increase in risk perception, farmers’ land use efficiency decreases by 250.34 yuan per mu. Farmers with higher risk perception are more inclined to take measures to mitigate both internal and external threats [116,117]. From the perspective of resource allocation, to cope with HEC, farmers tend to divert resources originally intended for agricultural production to building protective facilities and hiring guards, which not only increases hidden costs but also reduces investment in core agricultural activities. Analyzing from the angle of risk preference and behavioral motivation, most agricultural producers are inherently risk-averse [73]. Risk-averse farmers are less willing to engage in activities and investments that, while potentially offering higher expected outcomes, also carry the risk of failure. Even if the returns are higher, they tend to avoid risky situations [74]. Risk leads farmers to be less willing to undertake activities and investments that could yield higher expected results but also pose a risk of failure [75]. Consequently, farmers lean toward conservative strategies, hesitating to adopt new technologies and reducing their enthusiasm for labor input.

The study further explores the moderating effect of social networks. Results indicate that social networks play a positive moderating role in the impact of HEC risk perception on farmers’ land use efficiency. Specifically, as social networks strengthen, the negative impact of HEC on farmers’ land use efficiency is further exacerbated. Social networks facilitate the rapid dissemination and excessive amplification of risk information, leading farmers to make conservative decisions driven by heightened fear, prioritizing risk avoidance over efficiency maximization. Information dissemination and group behavior within social networks significantly influence individual decision-making [85]. Social networks reinforce herd mentality and group pressure, and to reduce decision-making risks caused by incomplete information, farmers tend to imitate others’ land use decisions [92], further intensifying the negative impacts. This finding aligns with the results of Kong [86] and Xu [87].

Finally, the study reveals the mediating mechanism through which HEC risk perception affects land use efficiency by influencing farmers’ agricultural capital investment. Specifically, for every unit increase in HEC risk perception, agricultural capital investment decreases by 193.32 yuan per mu. This result is largely consistent with the findings of Liu [54], who noted that the threat posed by Asian elephants significantly negatively impacts farmers’ production inputs. As the severity of elephant-related damage increases, farmers reduce their investment in agricultural production factors, thereby affecting land use efficiency [107]. This suggests that when making agricultural production decisions, farmers anticipate that even increased investments may not yield corresponding returns due to elephant damage. Consequently, as their perception of HEC risk rises, they rationally reduce the intensity of their agricultural investments to avoid potential losses, leading to a decline in land use efficiency.

5.2. Policy Recommendations

Based on the above analysis, this paper puts forward the following suggestions to provide references for alleviating HEC and improving the efficiency of land resource utilization.

First, strengthen ecological monitoring and control, and enhance the monitoring capacity of Asian elephant activities. Utilize technological means such as drones and infrared cameras to build a precise monitoring system to keep track of elephant herd dynamics in real time. Timely release early-warning information to farmers to reduce the risk of HEC and minimize farmers’ losses. At the same time, optimize the planning of Asian elephant habitats and build ecological corridors and food source bases to guide elephant herds away from farmers’ production and living areas, thus alleviating the conflict over living space between humans and elephants.

Second, strengthen support for agricultural production. The government and relevant departments should increase agricultural subsidies to reduce farmers’ production costs, compensate for losses caused by elephant damage, and boost farmers’ confidence. Improve the insurance mechanism for wildlife damage to ensure that farmers can receive timely and sufficient compensation for crop losses and facility damage caused by Asian elephant attacks. This can help disperse farmers’ risks, encourage reasonable investment, and safeguard the land’s production capacity. Meanwhile, organize agricultural experts to provide customized technical guidance and training for farmers, promote anti-elephant-damage agricultural technologies and varieties, enhance agricultural production efficiency and risk-resistance ability, and optimize land use.

Lastly, guide rational social network communication. Strengthen publicity and education for farmers, popularize ecological knowledge about Asian elephants and coping strategies for HEC, improve their scientific understanding, and reduce panic caused by information asymmetry. Encourage farmers to share objective and accurate information and effective experiences on social networks, guide rational discussions and scientific decision-making, avoid the spread of negative information and blind conformity, create a positive community atmosphere for dealing with HEC, and contribute to the improvement of land use efficiency and sustainable agricultural development

5.3. Limitations and Prospects

This study also has some limitations, and future research can further improve upon these aspects. (1) Subjectivity in variable measurement. For instance, the perception of HEC risk relies on farmers’ subjective evaluation. Despite multi-dimensional measurement and reliability and validity tests, it is still difficult to completely eliminate subjective bias. Future research could explore more objective measurement methods or combine other data sources for verification. (2) Lack of research on dynamic changes. The situation of HEC is dynamic. The current study, based on cross-sectional data, cannot capture the dynamic evolution process, which weakens the predictability of the conclusions. Future research could use dynamic research methods to track the change process. (3) Limitations of sample data. The research data were collected from 449 households in the Asian elephant distribution areas of Pu’er City, Yunnan Province. The scope of the sample is relatively limited, which may affect the universality and generalizability of the research findings. Future studies may consider expanding the research domain to cover more regions and different types of farmers and establishing a long-term panel database to enhance the credibility and universality of the research results. (4) The exploration of the mediating mechanism is not comprehensive enough. This paper only takes agricultural capital investment as the mediating variable, which may overlook other potentially important mediating factors. For example, the labor input of farmers, the adoption of land use technology, and the participation in agricultural insurance may also play a mediating role between the perception of HEC risk and land use efficiency. Future research should further expand the scope of mediating variables and take into account more possible mediating factors to more comprehensively reveal the impact mechanism of HEC risk perception on farmers’ land use efficiency.

6. Conclusions

This study, grounded in risk perception theory and the theory of planned behavior, empirically analyzes the impact of HEC risk perception on farmers’ land use efficiency and its underlying mechanisms. The findings reveal the following: (1) HEC risk perception has a significant negative impact on farmers’ land use efficiency. For every unit increase in risk perception, land use efficiency decreases by 250.34 yuan per mu. (2) Social networks play a positive moderating role in the process where HEC risk perception negatively affects farmers’ land use efficiency, further intensifying the negative impacts of risk perception. (3) In terms of mechanisms, HEC risk perception increases the likelihood of farmers altering their land use behaviors. Farmers with higher risk perception tend to decrease agricultural capital inputs, thereby reducing land use efficiency.

This research enriches the application of risk perception theory and the theory of planned behavior in the context of HEC, uncovering the deep-seated mechanisms through which risk perception influences farmers’ land use behaviors. By considering agricultural capital investment as a mediating variable and incorporating the moderating effect of social networks, the study provides a new perspective on farmers’ decision-making processes under external risks, expanding the application of relevant theories in natural resource management and land use. The conclusions offer significant practical guidance for formulating HEC management policies and enhancing farmers’ land use efficiency.