The Impact of Herders’ Risk Attitudes on Livestock Insurance: Evidence from the Pastoral Areas of Tibetan Plateau

Abstract

In the context of advancing the transformation and upgrading of grassland animal husbandry, encouraging and guiding the widespread adoption of livestock insurance plays an important role in promoting the sustainable development of the livestock industry. This paper explores the impact of herders’ attitudes and perceptions towards climate change risks on their livestock insurance strategies. Firstly, experimental economics methods are employed to measure the risk preferences of herders on the Tibetan Plateau. Secondly, a theoretical model incorporating risk preferences and insurance adoption behavior is constructed. Finally, the effects of herders’ risk preferences on insurance adoption behavior are empirically examined through double-hurdle models, instrumental variable models, and moderating effect models. The results reveal that (1) most herders on the Tibetan Plateau exhibit risk-averse characteristics. (2) The degree of risk preference has a significant negative impact on herders’ insurance adoption behavior, while the risk perception significantly positively influences insurance adoption. The results remain valid even after addressing issues of endogeneity and conducting robustness checks. (3) Livestock income plays a crucial moderating role in the mechanism through which risk attitudes affect insurance adoption behavior. (4) The impact of risk preference on insurance adoption behavior shows regional and income heterogeneity.

1. Introduction

Globally, rangelands cover more area than any other type of land, typically vast and remote with complex and challenging environmental conditions [1]. As global climate change intensifies, rangelands and their inhabitants are increasingly subjected to heightened climate pressures and environmental risk [2]. Extreme weather events triggered by climate change, such as droughts, floods, and extreme temperatures, along with unpredictable precipitation patterns, have significantly affected the productivity of pastoral grasslands [3]. Moreover, climate change is expected to increase the prevalence and frequency of livestock diseases, posing significant threats and risks to both the ecological environment of rangelands and the livelihoods of pastoralists [4,5]. This demands that rangeland communities and policymakers implement stronger adaptation and management strategies to ensure ecological and economic sustainability. In response, the development of effective risk management tools has become crucial for the sustainable development of rangelands and the wellbeing of pastoralists. Countries worldwide have developed various risk management strategies, including livestock insurance, which has proven essential in agricultural and pastoral management. Research conducted across different regions, including Europe [6], the United States [7], South America [8], Kenya [9], Mongolia [10], Ethiopia [11], India [12], and China [13], all highlight the indispensable role of livestock insurance in risk management in pastoral areas. Moreover, the effectiveness and application of livestock insurance varies from one country to another, influenced by geographic and cultural environments, demonstrating its adaptability and diversity [14].

The Qinghai–Tibet Plateau, highly sensitive to global climate change [15,16], acts as an amplifier of global climatic shifts and significantly affects global ecosystems. The plateau’s complex and harsh geographical environment exposes the pastoral areas to increasingly diverse and unpredictable natural risks [17,18]. Additionally, the Qinghai–Tibet Plateau is an important livestock production base in China, with grasslands accounting for over 60% of the region’s total area [19]. Given the escalating global climate changes, pastoralists on the Qinghai–Tibet Plateau are compelled to implement effective strategies to mitigate these emerging risks. Livestock insurance, recognized as an effective risk management tool, is categorized in China as policy-based agricultural insurance. It offers both protective and policy-oriented characteristics, aiming to stabilize economic fluctuations in production and operations by distributing unpredictable risks and providing economic compensation for disaster consequences [20]. Livestock insurance plays an essential role in safeguarding livestock production and management. Currently, over 70% of the premium income from policy-based insurance in China is supported by government subsidies at various levels [21]. The main strategy to increase the adoption of policy insurance is to encourage farmers to participate actively and voluntarily. However, in practice, the adoption rate is not ideal, with a continuous lack of motivation to enroll and a persistent insufficiency in demand [22]. This imbalance between supply and demand is a significant characteristic of China’s agricultural insurance market [23]. The development of agricultural insurance in China has been slow, especially for livestock insurance, which has received less attention, provided insufficient coverage, and even experienced market contraction in some cases [24]. The low adoption rate of agricultural insurance in China can be attributed to several factors. First, the frequent occurrence of natural disasters in Chinese agriculture complicates the insurance claims process, increasing farmers’ concerns about purchasing insurance [25,26] and heightening their reluctance. Second, the existing agricultural insurance products are limited and do not meet the diverse needs of farmers [27]. Third, many farmers exhibit a strong reliance on luck, leading to a weak awareness of the necessity to purchase insurance [28]. Fourth, the presence of positive externalities in the agricultural insurance consumption process, along with moral hazard and adverse selection issues, results in a demand level that is below the socially optimal level [29,30]. Therefore, it is critical to analyze the key factors influencing the adoption of livestock insurance to effectively boost its demand. Understanding these factors is crucial for promoting high-quality development in pastoral areas and addressing the unique needs and challenges faced by these communities.

Existing research extensively examines the impact of farmers’ risk attitudes on their decision-making, especially in the agricultural sector [31,32]. Firstly, extensive literature evidence around the world suggests that farmers’ risk attitudes play a crucial role in their risk management strategies. This has been observed among farmers in Belgium [33], Australia [34], and Sweden [35]. Secondly, when examining the impact of risk attitudes on farmers’ decision-making, the focus is primarily on agricultural input usage strategies. For instance, studies have explored pesticide usage strategies [36], fertilizer usage strategies [37,38,39], and seed usage strategies [40]. Additionally, research has analyzed crop planting strategies influenced by risk attitudes [41,42,43]. However, a review and analysis of existing research reveal that studies on the impact of risk preferences on insurance decisions primarily focus on life insurance [44,45,46], crop insurance [47,48,49], and motor insurance [50], with a noticeable lack of research on the influence of risk preferences on livestock insurance decisions. It is also noteworthy that the existing literature typically uses risk perception as a moderating variable to analyze the specific mechanisms by which risk preferences influence behavioral decisions [51]. Overall, the existing literature sufficiently demonstrates that risk attitudes have a broad and profound impact on agricultural management and decision-making. However, further research is needed in specific areas, such as livestock insurance, to fill significant gaps and provide policymakers and researchers with a more comprehensive perspective to develop more effective agricultural support policies and risk management strategies.

Through the analysis of the existing literature, we have identified several key limitations in the current body of research. First, most research focuses on crop insurance within the planting industry, with only a few studies addressing livestock insurance. Among these few studies on livestock insurance, there is almost no research on yak and Tibetan sheep insurance on the Qinghai–Tibet Plateau. Second, while it is generally acknowledged that there is insufficient demand for agricultural insurance, most research focuses on objective factors influencing agricultural insurance, with less emphasis on subjective factors. Although a few studies consider farmers’ risk attitudes, the methods used to measure these risk preferences lack economic rigor. Finally, when examining the specific mechanisms through which risk attitudes influence farmers’ decision-making, the existing literature predominantly considers risk perception as the moderating variable, with a notable lack of consideration for other potential moderating variables.

Our study provides several contributions. First, the study represents the first attempt to comprehensively analyze the relationship between the adoption of livestock insurance (specifically for yaks and Tibetan sheep) on the Qinghai–Tibet Plateau and the risk attitudes of local herders, which is significant given the region’s unique and complex climate. Second, it measures the perceptions of climate risk and the risk preferences of herders on the Qinghai–Tibet Plateau from an experimental economics perspective. Finally, in exploring the mechanisms by which risk preferences affect the adoption of insurance by herders, it not only employs risk perception as a moderating variable but also, for the first time, attempts to use livestock income as a moderating variable from both theoretical and empirical perspectives, thereby enriching existing research in the field.

Therefore, this paper aims to explore the impact of herders’ risk preferences on the Tibetan Plateau on their decisions to adopt livestock insurance. Utilizing field survey data from the Tibetan Plateau, this study employs experimental economics methods to rigorously analyze the characteristics of these herders’ risk preferences. A double-hurdle model is employed to assess how herders’ risk attitudes affect the decisions to adopt livestock insurance. An instrumental variable model is used to analyze potential heterogeneity, and a moderating effect model is utilized to examine underlying mechanisms. Additionally, a heterogeneity model is applied to investigate variations in the impact across different demographic groups.

2. Methods and Materials

2.1. Conceptual Framework

Figure 1 illustrates the specific mechanism framework. This study primarily analyzes the specific impact of risk preference and risk perception on livestock insurance from the perspective of subjective factors while controlling for variables at the objective level (including individual, household, and policy levels). First, to address potential endogeneity issues, we employed instrumental variable regression. Second, to explore how risk preference specifically influences the decision to adopt livestock insurance, we used a moderation effect model for mechanism testing, selecting pastoral income as the moderating variable. Finally, considering potential heterogeneity, we conducted a heterogeneity analysis.

2.2. Variable Selection

2.2.1. Dependent Variable

In this paper, the dependent variable captures the herders’ adoption behavior towards livestock insurance. It is assigned a value of 1 if a herder household adopts insurance by insuring their livestock and 0 otherwise. In the survey sample of this study, 77.2% of herder households adopted livestock insurance. It is important to note that the livestock insurance discussed in this study pertains to large livestock, primarily cattle and sheep. The livestock insurance studied in this paper is a form of agricultural policy insurance. The livestock insurance primarily covers the baseline costs, with claim payments including partial costs of breeding stock and breeding expenses. From the questionnaire survey conducted in the Qinghai–Tibet Plateau, it was found that the local premiums for yak and Tibetan sheep are CNY 15 (approximately USD 2.14) and CNY 2.5 (approximately USD 0.36), respectively. In the event of livestock death, the compensation is CNY 3000 (approximately USD 428.57) for yak and CNY 500 (approximately USD 71.43) for sheep. The premium payment structure requires the insured to pay 10%, while a 90% subsidy is provided by the government.

2.2.2. Core Independent Variable

(1) Risk preference. The core independent variable in this study is the herders’ risk attitude, which is quantified using the risk preference index as a proxy. A higher risk preference index indicates that herders are more inclined towards risk-seeking behavior, while a lower risk preference index indicates that herders are more inclined towards risk aversion. In other words, a smaller risk preference index signifies a higher degree of risk aversion. To estimate the herders’ risk preference index accurately, we employed experimental economics methods that combine risk–return theory with expected utility theory in a cohesive measurement framework. This methodology is well-supported by extensive literature [52,53]. The experiment required herders to choose between high-risk and low-risk options. The specific experimental methods are detailed in the Appendix A of this paper. For calculating the risk preference index, this paper referred to the experiment conducted by Holt and Laury (2002) [54], as detailed in Equation (1). Figure 2 presents a kernel density plot of herders’ risk preferences, indicating that the risk preference index of the majority of herders is predominantly clustered within the range of 0.4 to 0.5. In high-return scenarios, the data reveal that most participants exhibited risk-averse characteristics (42.78% of participants were risk-averse, exceeding the 35.55% who were risk-seekers), and 21.67% of participants were risk-neutral.

$$Risk preference index=\frac{Number of high-risk options are chosen under certain probability}{The total number of options}$$

(1)

(2) Climate risk perception. In this study, we introduce an additional independent variable, risk perception, which we measure through herders’ perceptions of meteorological information. Our survey questionnaire assesses this by evaluating herders’ awareness of natural disasters including droughts, snow calamities, temperature fluctuations, and rainfall patterns, providing a comprehensive measure of their risk perception. The assessment of climate perception is structured into two phases: the first phase reviews herders’ perceptions of climate and weather conditions over the past decade to validate the accuracy of their experiential judgments and common sense. The results from this phase reveal that 99.5% of herders can make accurate and reliable predictions about meteorological disasters, confirming the dependability of their meteorological perceptions. The second phase then shifts focus to their perceptions of upcoming changes in climate conditions, aiming for a more direct assessment of their risk perception. Given the high reliability demonstrated in the first phase, this study uses the meteorological perceptions of herders from the second phase as a proxy for the risk perception variable.

2.2.3. Instrumental Variable

The instrumental variable used in this study is the NDVI (normalized difference vegetation index) of the grassland. This index is a critical parameter for reflecting plant growth and a vital indicator for assessing the quality of grasslands [55]. The grassland quality is essential for livestock farming and forms the basis for herders to counter livelihood and natural risks. (1) Relevance: Grassland quality affects herders’ economic status and capacity to bear risks. Higher-quality grasslands typically result in greater productivity and income, which can influence herders’ risk preferences. This connection satisfies the requirement that instrumental variables must be correlated with the independent variable (risk preferences). (2) Exogeneity: Grassland quality is largely determined by natural conditions and environmental factors, rather than by the herders’ individual choices, making it exogenous to the model. This ensures that the NDVI is unrelated to any unobserved confounding factors that could influence the herders’ risk preferences. Therefore, the NDVI serves as a valid instrumental variable for this study.

2.2.4. Moderating Variable

This paper also considers livestock income as a moderating variable to analyze the potential mechanisms of the baseline regression. In this study, livestock income is defined as the income obtained from livestock production and operations, calculated specifically as the sales revenue plus government subsidies minus the input costs. We chose livestock income as a moderating variable based on the following considerations. (1) Risk sensitivity. Livestock income constitutes a significant portion of herders’ total family income, directly influencing their sensitivity to income fluctuations and potential losses. Higher livestock income may increase herders’ sensitivity to risks as they have more assets at stake. (2) Risk-bearing capacity. The level of livestock income might also reflect herders’ capacity to bear risks. Herders with higher incomes may have more resources to manage risks, thus enhancing their capacity to withstand shocks. Therefore, when exploring the impact of risk preferences on insurance adoption decisions, livestock income is expected to serve as a moderating variable in analyzing the influence mechanism between risk preferences and insurance adoption decisions. Additionally, we also considered risk perception as a moderating variable.

2.3. Data Collection

The data for this study were collected from a micro-survey of herder households conducted in August 2020 in the pastoral areas of Qinghai and Gansu provinces. To address the language barriers of the local herders, specialized translators were hired and trained to ensure the accuracy of the survey questions. The survey employed a stratified random sampling method. Firstly, considering accessibility, Zezhou County, Gangcha County, Gande County, Dari County, Zhiduo County, and Chengduo County were selected in Qinghai Province; in Gansu Province, Sunan County, Subei County, Tianzhu County, and Maqu County were chosen. Secondly, in these ten counties, 2–3 townships per county were randomly sampled. Thirdly, from each selected township, approximately 15 households were randomly chosen. Data were collected through face-to-face interviews with the herder families. It should be emphasized that a total of 362 questionnaires were administered across these counties. However, after thorough verification, two questionnaires were deemed invalid due to missing information and unreasonable answers. Thus, a total of 360 valid questionnaires were obtained.

Table 1. Descriptive statistics for the variables.

Variable	Definition	Mean	S.D.	Min	Max
Dependent Variables
Livestock Insurance	1 if livestock insurance is purchased, 0 otherwise	0.772	0.420	0	1
Insured Quantity	Number of insured livestock in family/thousand sheep units	0.160	0.206	0	1.145
Independent Variables
Risk Preference Index	Higher values indicate a greater preference for risk.	0.491	0.232	0	1
Risk Perception	Perception and expectation of future risks, 1 = risk perceived, 0 = no risk perceived	0.619	0.486	0	1
Individual Characteristics
Age	Age of the head of household/years	50.736	11.421	23	96
Education Level	Years of education received by the head of household/years	2.547	3.792	0	16
Chinese Education	1 = received Chinese education, 0 = did not	0.311	0.464	0	1
Health Level	Health level of the head of household (1–5), 1 = very good, 5 = very poor	2.200	1.273	1	5
Household Characteristics
Laborers	Number of laborers in the family/individuals	4.656	1.918	1	12
Grassland Area	Total grassland area owned by the family/thousand acres	7.634	24.952	0	260
Livestock Quantity	Total number of livestock owned by the family/thousand sheep units	0.254	0.258	0	1.61
Family Income	Annual total income of the herder family/ten thousand yuan	13.570	14.865	0.809	129.231
Loan Situation	Whether there is a loan over 3000 in the family, 1 = yes, 0 = no	0.306	0.461	0	1
Policy Characteristics
Grazing Ban Policy	Participation in grazing ban policy, 1 = yes, 0 = no	0.381	0.486	0	1
Grass–Livestock Balance Policy	Participation in grass–livestock balance policy, 1 = yes, 0 = no	0.528	0.500	0	1
Village Rules and Agreements	Whether there are village rules and agreements, 1 = yes, 0 = no	0.831	0.375	0	1
Other Variables
Livestock Income	Total livestock income of the herder family/ten thousand yuan	5.852	11.004	0	102
NDVI	Normalized difference vegetation index	0.535	0.217	0.118	0.784

presents the descriptive statistics for the variables in this study. The average risk preference index of the herders is 0.491, indicating that most herders exhibit risk-averse characteristics. The mean value of risk perception is 0.619, indicating that herders have a higher sensitivity to natural risks. Specifically, in our sample, herder households account for 69% of the total, while non-herder households (farmers and hunters) make up 21%. Additionally, 12.5% of the sample population primarily relies on herding as their main livelihood. Control variables include individual characteristics, household characteristics, and policy characteristics, aimed at measuring individual differences from various dimensions.

2.4. Theoretical Analysis and Research Hypotheses

This paper defines “risk attitude” as “risk preference” or “risk propensity”, which refers to an individual’s propensity to take or avoid risks. These preferences can influence decision-making in various contexts [56]. Commonly identified types of risk attitudes include risk aversion, risk neutrality, and risk seeking. This paper focuses on the risk aversion characteristics of herders. Risk aversion is a type of risk preference characterized by the tendency to prefer certainty over uncertainty. Individuals who are risk-averse are more likely to avoid risky situations, even if the potential rewards are higher. Several studies have examined the role of risk preferences in decision-making processes. For instance, Pratt (1978) [57] introduced the concept of risk aversion in economic theory, highlighting how individuals’ aversion to risk affects their choices under uncertainty. Similarly, in the context of agricultural insurance, studies have shown that farmers’ risk preferences significantly influence their insurance purchasing decisions [58].

This paper establishes a theoretical model to analyze the impact of risk preferences on herders’ decisions to adopt livestock insurance.

Firstly, this paper assumes that herders are risk-averse. The expected value of herders’ decisions to adopt livestock insurance is represented by the random variable $X$. The random variable $X$ inherently includes the consideration of potential outcomes in different scenarios, specifically whether insured shocks occur or do not occur.

$$E\left(X\right)=a$$

(2)

$$var\left(X\right)={\sigma }^2$$

(3)

where $a$ is a positive number, indicating that the decision-making behavior of herders to adopt livestock insurance has a positive average impact on their income. ${\sigma }^2$ is the variance of the random variable $X$. Specifically, ${\sigma }^2$ represents the degree of uncertainty associated with $X$, capturing the variability or risk in herders’ income when considering livestock insurance. This variance represents the herders’ risk perception when adopting livestock insurance. The larger the ${\sigma }^2$, the stronger the herders’ risk perception.

Secondly, in the following Equation (4), we define the risk premium P when herders purchase livestock insurance. The risk premium P represents the additional amount herders are willing to pay to avoid risk, indicating that the utility of purchasing insurance is equivalent to the utility of losing an uncertain amount P. Therefore, a larger P reflects a higher degree of risk aversion among herders, indicating their willingness to pay more to avoid risk. It is important to note that while an increase in the actual premium usually reduces the willingness to purchase insurance, here, we are discussing the risk premium, not the actual premium. Therefore, the expression for the risk premium is as follows:

$$E\left[U(w+X)\right]=U\left[w+E\left(X\right)-P\right]$$

(4)

In Equation (4), $U(·)$ is the utility function of the herders, and w represents the herders’ income. It is assumed that the utility function $U(w+X)$ is twice continuously differentiable. By expanding $U(w+X)$ in a second-order Taylor series around $\left[w+E\left(X\right)\right]$, we obtain the following:

$$U\left(w+X\right)\approx U\left[w+E\left(X\right)\right]+U^{\prime }·\left[X-E\left(X\right)\right]+0.5U^{″}·{\left[X-E\left(X\right)\right]}^2$$

(5)

In Equation (5), $U^{\prime }=\frac{\mathsf{\partial }U}{\mathsf{\partial }w}$, $U^{″}=\frac{{\mathsf{\partial }}^{2}U}{\mathsf{\partial }w}$ Then, by taking the expected value of Equation (5), we can obtain the following:

$$E\left[U(w+X)\right]\approx E\left\{U\right.\left.\left[w+E\left(X\right)\right]\right\}+U^{\prime }·E\left[X-E\left(X\right)\right]+0.5U^{″}·E\left\{\left[X-\right.\right.\left.{\left.E\left(X\right)\right]}^2\right\}$$

(6)

Since $E\left[X-E\left(X\right)\right]=0$, Equation (4) can be further expressed as following:

$$E\left[U(w+X)\right]\approx E\left\{U\right.\left.\left[w+E\left(X\right)\right]\right\}+0.5U^{″}·{\sigma }^2$$

(7)

Subsequently, by expanding the right-hand side of Equation (4) in a first-order Taylor series around $\left[w+E\left(X\right)\right]$, we obtain the following:

$$U\left[w+E\left(X\right)-P\right]\approx U\left[w+E\left(X\right)\right]-U^{\prime }·P$$

(8)

By combining Equations (7) and (8), we can solve for P to obtain the following:

$$P\approx -0.5\left(\frac{U^{\prime \prime }}{U^{\prime }}\right){\sigma }^2$$

(9)

In Equation (9), let $r\left(w\right)=-\left(\raisebox{1ex}{$U^{\prime \prime }$}\!\left/ \!\raisebox{-1ex}{$U^{\prime }$}\right.\right)$, where $r\left(w\right)$ is the Arrow–Pratt measure of absolute risk aversion, used to measure the degree of the herders’ risk preference. An increase in $r\left(w\right)$ indicates an increase in the herders’ degree of risk aversion, i.e., a lower degree of risk preference.

$$P\approx 0.5r\left(w\right){\sigma }^2$$

(10)

From Equation (10), it is observed that if $r\left(w\right)$ (the degree of risk aversion) and ${\sigma }^2$ (risk perception) increase, then P will also increase. Consequently, the likelihood of herders adopting livestock insurance will also increase.

Consequently, this paper proposes the following hypothesis:

H1. 

The higher the degree of risk aversion among herders, the greater the likelihood of adopting livestock insurance.

H2. 

Herders with a higher perception of risk are more likely to adopt livestock insurance.

2.5. Econometric Model

2.5.1. Baseline Model

To examine the relationship between risk preference and livestock insurance adoption, this study employs the double-hurdle model, which involves a two-step decision-making process: the first step is the “participation decision”, and the second step is the “quantity decision”. The specific settings of the model are as follows [59]:

In the first stage, we used a Probit model to estimate the participation decision.

$${Y_i}^{\ast }={\beta }_0+{\beta }_1{Risk}_i+{\beta }_2{Perception}_i+{\beta }_3{X}_i+{\epsilon }_i$$

(11)

where ${Y_i}^{\ast }$ is the decision variable, where $Y_i=1$ indicates the decision to adopt insurance and $Y_i=0$ indicates non-participation in the decision. ${Risk}_i$ represents the herder’s risk preference index. ${Perception}_i$ denotes the herder’s risk perception. $X_i$ includes a series of control variables. ${\beta }_1$ represents the direct effect of risk preference on the livestock insurance adoption decision. ${\beta }_2$ represents the direct effect of risk perception on the livestock insurance adoption decision. ${\epsilon }_i$ is the random error term.

In the second stage, we used the ordinary least squares (OLS) model to estimate the degree of participation.

$$Z_i={\delta }_0+{\delta }_1{Risk}_i+{\delta }_2{Perception}_i+{\delta }_3{X}_i+u_i$$

(12)

where $Z_i$ is the dependent variable, representing the degree of participation in the decision-making process. In this study, it specifically refers to the number of livestock insured. ${\delta }_1$ indicates the impact of risk preference on the quantity of livestock insured, and ${\delta }_2$ indicates the impact of risk perception on the quantity of livestock insured. ${\delta }_3$ is the regression coefficient for the control variables. $X_i$ includes a series of control variables. $u_i$ is the random error term.

2.5.2. Instrumental Variable Model

The focus of this study is to examine the impact of herders’ risk preferences on their decisions to adopt livestock insurance, which may be susceptible to endogeneity issues. For instance, risk preferences may influence the behavior of adopting insurance, while the experience of adopting insurance, in turn, may affect the risk preferences of herders. This endogeneity issue, caused by reverse causality, is addressed in this paper using an instrumental variable model. The specific settings of the model are as follows [60]:

The first stage is as follows:

$${Risk}_i={\theta }_0+{\theta }_1{IV}_i+{\theta }_2{Perception}_i{+\theta }_3{X}_i+{\vartheta }_i$$

(13)

The second stage is as follows:

$${Insurance}_i=w_0+w_1{\widehat{Risk}}_i+w_2{Perception}_i+w_3{X}_i+{ϵ}_i$$

(14)

In Equation (13), ${IV}_i$ represents the instrumental variable, which, in this paper, refers to the NDVI index. In Equation (14), ${Insurance}_i$ denotes livestock insurance adoption behavior. ${\widehat{Risk}}_i$ is the predicted value of risk preference ${Risk}_i$ obtained from the first stage. ${\theta }_1$, ${\theta }_2$, ${\theta }_3$, $w_1$, $w_2$, and $w_3$ are parameters to be estimated. ${\theta }_0$ and $w_0$ represent the intercept terms. ${\vartheta }_i$ and ${ϵ}_i$ are random disturbance terms.

2.5.3. Moderating Effect Model

To examine the specific mechanism through which risk preferences affect livestock insurance adoption behavior, first, we constructed a moderating effects model using livestock income as the moderating variable. The specific settings of the model are as follows [61]:

$${Y_i}^{\ast }={\gamma }_0+{\gamma }_1{Risk}_i+{\gamma }_2{M}_i+{\gamma }_3{Risk}_i\times M_i+{\gamma }_4{X}_i+{\tau }_i$$

(15)

where ${Y_i}^{\ast }$ is the dependent variable, representing the adoption behavior of livestock insurance. $M_i$ is the moderating variable, which is the logarithm of livestock income in this paper. The interaction term ${Risk}_i\times M_i$ is used to estimate the moderating effect. ${\gamma }_1$ represents the direct impact of risk preference on the decision to adopt livestock insurance. ${\gamma }_2$ represents the direct impact of livestock income on the adoption decision, and ${\gamma }_3$ is a key parameter of interest in this paper, indicating how livestock income affects the relationship between risk preference and the adoption of livestock insurance. $X_i$ represents control variables. ${\tau }_i$ denotes random disturbance terms.

3. Results

3.1. Baseline Regression

This study examines the impact of risk attitudes on herders’ decisions to adopt livestock insurance, with the regression results presented in

Table 2. The impact of risk preference on livestock insurance decisions.

Variables	First Stage			Second Stage
	(1)	(2)	(3)	(4)
	Marginal Effects	Marginal Effects	Marginal Effects	Marginal Effects
Risk preference	−0.2120 ***		−0.2320 ***	−0.0759 **
	(0.0804)		(0.0780)	(0.0297)
Risk Perception		0.0845 **	0.0964 ***	0.0184
		(0.0369)	(0.0368)	(0.0121)
Individual Characteristics
Age	−0.0027	−0.0027	−0.0024	−0.0007
	(0.0016)	(0.0017)	(0.0016)	(0.0005)
Education Level	−0.0247 ***	−0.0239 ***	−0.0244 ***	−0.0098 ***
	(0.0088)	(0.0086)	(0.0082)	(0.0037)
Chinese Education	0.0423	0.0395	0.0397	0.0717 ***
	(0.0691)	(0.0684)	(0.0652)	(0.0253)
Health Level	−0.0206	−0.0210	−0.0207	0.0043
	(0.0144)	(0.0138)	(0.0138)	(0.0053)
Household Characteristics
Laborers	0.0137	0.0149	0.0147	−0.0009
	(0.0092)	(0.0094)	(0.0091)	(0.0031)
Grassland Area	−0.0017 *	−0.0018 **	−0.0015 *	−0.0008
	(0.0010)	(0.0008)	(0.0009)	(0.0006)
Livestock Quantity	1.1275 ***	1.1405 ***	1.1227 ***	0.6538 ***
	(0.1364)	(0.1423)	(0.1352)	(0.0578)
Family Income	−0.0946 ***	−0.1058 ***	−0.0839 ***	0.0070
	(0.0277)	(0.0281)	(0.0279)	(0.0094)
Loan Situation	0.1414 ***	0.1437 ***	0.1481 ***	0.0148
	(0.0436)	(0.0449)	(0.0433)	(0.0124)
Policy Characteristics
Grazing Ban Policy	−0.0017	−0.0076	0.0020	−0.0046
	(0.0392)	(0.0402)	(0.0384)	(0.0124)
Grass–Livestock Balance Policy	−0.0455	−0.0518	−0.0609 *	0.0160
	(0.0366)	(0.0373)	(0.0366)	(0.0118)
Village Rules and Agreements	0.0614	0.0472	0.0354	0.0068
	(0.0455)	(0.0476)	(0.0477)	(0.0155)
Observations	356	356	356	356
Wald chi2				23.30
Prob > chi2				0.0000
Pseudo R2				0.7051

Note: Parentheses contain heteroskedasticity-robust standard errors; * P < 0.1, ** P < 0.05, *** P < 0.01.

. Columns (1) to (3) estimate the results of the first stage of the baseline model. Column (1) reports that herders’ risk preferences have a significant negative impact on their decisions to adopt livestock insurance at the 1% statistical significance level. The marginal effects indicate that for every 0.1 increase in the risk preference index, the probability of adopting livestock insurance decreases by 2.12%. This suggests that herders with higher levels of risk aversion are more likely to adopt livestock insurance, thus supporting hypothesis H1. Column (2) reports that herders’ risk perception has a significant positive impact on their behavior of adopting livestock insurance at a 5% statistical significance level. The marginal effects show that the presence of risk perception increases the probability of adopting livestock insurance by 8.45%. This implies that herders with stronger risk perceptions are more likely to adopt livestock insurance, supporting hypothesis H2. Column (3) reports the effects of risk preference on herders’ insurance decisions while considering risk perception. The marginal effects demonstrate that with risk perception present, for every 0.1 increase in the risk preference index, the probability of adopting livestock insurance decreases by 2.32%. Column (4) reports the results of the second stage of the baseline model. These results indicate that the greater their risk aversion, the greater the number of livestock they insure. The regression results reveal that for every 0.1 decrease in the risk preference index, the number of insured livestock increases by 0.759 thousand sheep units.

Regarding control variables, at the individual level, the lower the education level of the herder, the higher the probability of adopting livestock insurance. This may be because herders with lower education levels have weaker risk forecasting abilities, and thus lower resistance to risks, leading them to adopt more protective measures against potential losses. At the household level, households with larger grassland areas and higher incomes are less likely to adopt livestock insurance. This is because larger grassland areas provide more fodder, offering sufficient resources to mitigate vulnerabilities in livelihood and risks associated with livestock breeding. Higher income also provides more capital to withstand potential risks. Additionally, families with a greater total number of livestock and those with loans are more likely to adopt livestock insurance. This is because the greater the number of livestock, the higher the risk they face, necessitating more livestock insurance to mitigate potential risks. Families with loans have more financial capital available to adopt livestock insurance.

3.2. Robustness Analysis

To ensure the reliability of the baseline regression results, this study implemented two robustness checks: (1) model substitution, which involved replacing the model used in the baseline regression with the Logit model, and (2) sample reduction, using the Winsorization method to trim the tails of the data distribution. The results in

Table 3. Robustness test results.

Variables	(1) Logit Model			(2) Sample Reduction
	(1)	(2)	(3)	(1)	(2)	(3)
Risk Preference	−0.2223 **		−0.2340 ***	−1.1560 ***		−1.2879 ***
	(0.0859)		(0.0783)	(0.4126)		(0.4130)
Risk Perception		0.0892 **	0.0992 ***		0.3640 *	0.4528 **
		(0.0391)	(0.0379)		(0.1904)	(0.1997)
Controls	YES	YES	YES	YES	YES	YES
Observations	356	356	356	356	356	356
Wald Chi2	76.22	68.17	74.87	88.89	86.00	97.62
Prob>F	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
R2	0.3425	0.3365	0.3604	0.3719	0.3606	0.3852

Note: Parentheses contain heteroskedasticity-robust standard errors; * P < 0.1, ** P < 0.05, *** P < 0.01. Due to space limitations, only the estimation results for the core independent variable are reported. The estimation results for the control variables are also largely consistent with the baseline regression results.

indicate that the coefficients and the direction of the effect for the core explanatory variables remain consistent with those of the baseline regression, with minor variations in the coefficients. These results confirm the robustness and reliability of the baseline regression, further supporting hypotheses H1 and H2 of this paper.

3.3. Endogeneity

Considering the potential for endogeneity issues in the study, a 2SLS (two-stage least squares) instrumental variable analysis was conducted, examining both the first and second stages of the baseline regression model, with results presented in

Table 4. Instrumental variable regression.

Variable	Insurance Adoption	Number Insured
Two-Stage Least Squares Regression (2SLS)
Risk Preference	−0.2452 **	−0.1690 **
	(−0.2452)	(0.0659)
Risk Perception	0.1295 ***	0.0420 **
	(0.0440)	(0.0202)
Control Variables	YES	YES
Observations	356	356
Wald chi2	52.97	52.97
Prob > chi2	0.0000	0.0000
R-squared	0.1308	0.2087
Root MSE	0.3892	0.1774
First-Stage IV Regression
NDVI	0.7854 ***	0.7854 ***
	(0.0431)	(0.0431)
Control Variables	YES	YES
Observations	356	356
F	45.62	45.62
Prob > chi2	0.0000	0.0000
R-squared	0.5552	0.5552
Adj R-squared	0.5382	0.5382
Root MSE	0.1557	0.1557

Note: Parentheses contain heteroskedasticity-robust standard errors; * P < 0.1, ** P < 0.05, *** P < 0.01.

. The main results of the instrumental variable in both the first and second stages are reported. Using NDVI as an instrumental variable, the first-stage regression results show an F-value greater than the critical value of 10 and a significant correlation between the instrumental variable and risk preferences, thus ruling out the issue of a weak instrumental variable. The second-stage regression results demonstrate that risk preference still significantly negatively impacts livestock insurance adoption behavior, while risk perception significantly positively affects the livestock insurance adoption decision. This further validates that higher levels of risk aversion and higher risk perception increase the likelihood of livestock insurance adoption. The absolute values of the coefficients of the core independent variables increase, suggesting that the baseline regression estimates might have some biases caused by endogeneity. The 2SLS regression, by addressing these issues, may provide an estimation closer to the true effects for the study.

3.4. Mechanism Analysis

Table 5. Moderating effect of livestock income.

Variable	Coef.
Risk Preference	−14.3575 **
	(5.9158)
Livestock Income	−0.7570 **
	(0.3174)
Risk Preference × Livestock Income	1.1742 **
	(0.5378)
Control Variables	YES
Observations	203
Wald chi2	51.10
Prob > chi2	0.0000
Pseudo R2	0.3815
Log pseudolikelihood	−52.5846

Note: Parentheses contain heteroskedasticity-robust standard errors; * P < 0.1, ** P < 0.05, *** P < 0.01.

presents the regression results with livestock income as a moderating variable. The coefficient of the interaction term is our primary interest. The results in the table show that, firstly, the impact of risk preference on the decision to adopt livestock insurance is significantly negative, indicating that herders more willing to take risks are less likely to adopt livestock insurance, which aligns with the baseline regression results. Secondly, livestock income has a significant negative effect on the decision to adopt livestock insurance. This may be because herders with higher incomes have more financial buffers or alternative risk management strategies available and thus may not need insurance to manage risks as much. Lastly, the coefficient of the interaction term is significantly positive, suggesting that the level of livestock income positively moderates the relationship between risk preference and the decision to adopt livestock insurance. Overall, these results suggest that an increase in herders’ livestock income enhances the impact of risk preference on the adoption of livestock insurance.

3.5. Heterogeneity Analysis

Considering that the environmental conditions and lifestyle variations across different regions may reflect herders’ production and management styles, it is possible that the understanding of risk strategies could vary among herders from different areas. Therefore, this paper conducts subgroup regression analyses at both the regional and income levels to explore the heterogeneity in the impact of risk preferences on insurance adoption decisions, thereby providing insights for region-specific policies.

Table 6. Heterogeneity analysis.

Variables	Regional Heterogeneity		Income Heterogeneity
	Gansu	Qinghai	High Income	Low Income
	(1)	(2)	(3)	(4)
Risk Preference	−0.2637 **	−0.1022	−0.2794 ***	−0.2070 *
	(0.1337)	(0.0898)	(0.0975)	(0.1151)
Risk Perception	0.1454 **	0.0126	0.1332 ***	0.0452
	(0.0590)	(0.0283)	(0.0475)	(0.0465)
Controls	YES	YES	YES	YES
Observations	141	215	179	177

Note: Parentheses contain heteroskedasticity-robust standard errors; * P < 0.1, ** P < 0.05, *** P < 0.01.

, Columns (1) and (2) report the results of the regional heterogeneity analysis. The influence of risk attitudes on livestock insurance adoption is more pronounced among herders in the Gansu region compared to those in Qinghai. Specifically, in Gansu, risk attitudes significantly negatively affect the adoption of livestock insurance, while risk perception has a significantly positive impact on insurance adoption. In contrast, the insurance decision-making of herders in Qinghai is not notably influenced by risk factors. To explore the reasons behind these differences, this paper conducts a detailed mean analysis of herders’ characteristics in both Qinghai and Gansu, as presented in

Table 7. Grouping characteristics of variables.

Region		Investment in Enclosed Feeding Areas	Distance from Highways
Gansu Province	Mean	0.7292	5.4760
	Std	0.4459	13.6180
Qinghai Province	Mean	0.3009	6.5255
	Std	0.4597	12.5986
Whole Sample	Mean	0.4722	6.0978
	Std	0.4999	13.0145

. Herders in Gansu invest more in enclosed feeding areas than those in Qinghai, reflecting a cautious approach to risks during livestock rearing. Additionally, transportation is more convenient in Gansu, with grasslands closer to highways, which facilitates better information exchange. As a result, herders in Gansu have more channels to receive early warning information about risks, making their insurance-adopting decisions more responsive to risk preferences.

Columns (3) and (4) of Table 6 report the responses of herders from different income groups to livestock insurance strategies based on their risk attitudes. These groups are delineated by the median per capita family income. For herders in the higher income group, their decision-making on adopting insurance is more significantly influenced by risk preferences. Specifically, the wealthier herders, who are risk-averse, are more likely to adopt livestock insurance as a precaution to mitigate potential risks to their production and livelihood when they perceive risks. The possible reason is that families with higher per capita income possess greater financial power to afford insurance and have better capabilities to proactively manage potential risks, thereby preventing severe financial losses.

4. Discussion

This paper aims to analyze the impact of herders’ risk preferences on their decision-making to adopt livestock insurance in the Tibetan Plateau area, as well as the underlying mechanisms. Based on data from field surveys, we employ a double-hurdle model and an instrumental variable model to analyze the specific impact effects and potential endogeneity issues. Furthermore, we use a moderating effects model to explore the potential mechanisms involved. This comprehensive approach allows us to gain deeper insights into how risk preferences influence insurance adoption among herders in this region.

An interesting result that emerged from our study is that livestock income also plays a moderating role in how risk preferences influence herders’ decisions to adopt insurance. Specifically, the higher the livestock income, the less likely herders with a higher degree of risk preference are to choose insurance. This indicates that risk-averse herders are more likely to choose livestock insurance if their livestock income is higher. One possible reason is that higher-income herders have a greater financial capacity to cope with potential losses, making them more inclined to purchase insurance to protect their assets when facing risks. Additionally, high-income herders may place greater importance on long-term financial stability and therefore choose insurance to mitigate unpredictable risks, ensuring their economic situation remains stable.

Finally, our study also further emphasizes that the impact of risk preferences on herders’ decisions to adopt livestock insurance is heterogeneous. In regions with greater pastoral investment and more developed transportation infrastructure, particularly in high-income pastoral areas, the influence of risk preferences on insurance decisions is more pronounced. A possible reason for this is that areas with developed transportation facilities facilitate easier communication, providing more channels for obtaining risk warning information. Additionally, regions with higher per capita incomes have greater adopting power for insurance, so these groups may be more likely to purchase livestock insurance to a greater extent to mitigate risks. Similar findings are also evident in the research conducted by Guiso and Paiella (2008) [62]. This refined analysis highlights the significant variability in how different factors influence insurance adoption across various contexts within the pastoral community.

While this study makes significant contributions to the existing literature, it also has its limitations. Firstly, the data used in this research are cross-sectional and thus do not adequately capture the dynamic effects of herders’ insurance adoption decisions. Addressing this will be crucial in our future research endeavors. Secondly, our analysis of livestock insurance did not consider herders’ satisfaction with their insurance policies, which could provide deeper insights into their purchasing behavior. Lastly, while this study focuses on livestock insurance adoption in China, the methods and models used have universal applicability and can be employed in studies of other insurance decisions globally.

5. Conclusions

This paper analyzes the impact of herders’ risk preferences on their livestock insurance adoption decisions from the perspective of subjective factors on the Tibetan Plateau. The results indicate that herders in the Tibetan Plateau regions of China generally exhibit risk-averse tendencies, with those indicating higher degrees of risk aversion being more inclined to adopt livestock insurance. The influence of herders’ risk preferences on their insurance adoption decisions is more pronounced when risk perception is present. Additionally, among groups with better livestock income, the impact of herders’ risk preferences on their decisions to adopt livestock insurance is also greater. Lastly, the findings of this study suggest that the influence of herders’ risk preferences on livestock insurance adoption decisions exhibits both regional and income heterogeneity.

The study reveals significant policy implications. Firstly, early warning and information dissemination mechanisms for natural risks should be strengthened to ensure that herders can fully understand and mitigate potential risks. Secondly, a comprehensive livestock insurance system should be established, and the benefits of insurance should be actively promoted to overcome traditional herders’ cognitive barriers and increase the adoption rate of insurance. Lastly, regionally adapted insurance policies that take into account the different habits and needs of herders should be implemented, thereby providing differentiated insurance policies tailored to their specific circumstances.

There are several limitations to our study. Firstly, our sample dataset is cross-sectional, which restricts our ability to observe dynamic impacts over time. To address this limitation, we plan to conduct follow-up research in the Qinghai–Tibet Plateau. This will enable us to perform more comprehensive dynamic analyses and gain deeper insights into the temporal aspects of herders’ livestock insurance adoption behavior. Additionally, to fill the research gap and enrich our study, we intend to conduct similar research in the Inner Mongolia Plateau. This will allow us to better understand the heterogeneity and common characteristics across different regions, thereby providing a more comprehensive understanding of the factors influencing herders’ decisions.