The Influence of the Peer Effect on Farmers’ Agricultural Insurance Decision: Evidence from the Survey Data of the Karst Region in China

Abstract

Low insurance participation rate and low willingness to insure among farmers have always been major problems in the sustainable development of agricultural insurance in China. This paper attempts to examine the peer effect on farmers’ agricultural insurance buying decisions and explore its mechanism. We have established an IVprobit model, using the survey data of 9452 farmers in the karst regions in China. The empirical results show that: (1) Peer effect has a significant influence on farmers’ participation in agricultural insurance. A 10% increase in farmers’ neighbors’ participation in agricultural insurance increases the likelihood of farmers’ participation by 3.25%. (2) Peer effect promotes farmers’ participation by enhancing farmers’ risk perception and insurance cognition. (3) Peer effect is asymmetrical: male and larger-scale farmers have more significant effects on their peers and probably lead the participation in agricultural insurance. The results of the study have the following policy implications: (1) Increasing policy publicity and enforcing policy advocacy would magnify the positive impact of peer effect. (2) Increasing the participation rate of male and larger-scale farmers by policy interventions, which would give full play and a positive demonstration effect of specific groups. (3) Innovating insurance publicity methods to enhance farmers’ risk awareness and insurance cognition. (4) More concentration should be focused on the primary-level governance in rural China. We should advance the stock of rural social capital on all fronts to leverage the role of peer effect within rural areas.

1. Introduction

The inferiority of agriculture has always been a restriction for the development of Chinese agriculture for a long time. In the karst regions, due to the widespread mountains and hills, the rugged terrain, and depleted soils, it is hard to develop agricultural production [1]. In this case, the foundation of agricultural production in the karst regions is relatively weak. This results in the problems of agricultural frangibility and inferiority being more serious. Agricultural insurance, as an important tool to spread the risks of agricultural production, plays an ever-increasingly important role in promoting farm modernization and sustainability, promoting rural industry revitalization, and guaranteeing farmers’ basic income. It is also an essential tool to safeguard agricultural production in the karst regions. Therefore, it is of great necessity to study the development of agricultural insurance in the rural karst regions, to help promote the participation rate of agricultural insurance in rural China.

The Chinese government has always attached great importance to the development of agricultural insurance: In 2022, Central Document No. 1 listed “strengthening financial services for rural revitalization” as an important directive and stressed the importance of “actively developing agricultural insurance and reinsurance”. With the tireless efforts of the government and departments, China’s agricultural insurance has made great progress: According to the data provided by the China Banking and Insurance Regulatory Commission (CBIRC) and China’s Agricultural Insurance Protection Research Report (2021), China’s agricultural insurance premium income in 2020 totaled CNY 81.49 billion, and the premium scale has surpassed the United States. China has become the largest agricultural insurance market worldwide. At the same time, China’s agricultural insurance depth reached 1.05%, and the density reached CNY 460 per person, achieving a year-on-year growth rate of 27.75%. Although the insurance density and insurance depth have achieved rapid growth, the overall level is still far below the average level of the overall insurance industry. The relatively low insurance density and insurance depth also reflect that the number of farmers participating in insurance is insufficient, and their willingness to participate in insurance is still not high. To better solve this problem, the farmers’ willingness to purchase agricultural insurance needs to be further studied. Comprehensively, we should systematically clarify the farmers’ decision-making in purchasing agricultural insurance and its internal influencing mechanism.

Farmers’ decisions are easily influenced by the behaviors of their peers. Especially in the case of inadequate information, farmers’ decisions are more likely to be influenced by the group members. In other words, farmers’ decisions will be affected by the peer effect. As a typical “acquaintance society” in rural China, farmer’s decision-making behavior is more likely to be influenced by each other. Furthermore, some studies have proved that the peer effect can significantly promote individuals to participate in voluntary social programs, such as China’s New Rural Pension Scheme [2], pension plans, social medical insurance, etc. [3]. Meanwhile, the peer effect also plays an important role in many individual decision-making processes, such as risk-taking, fertility issues, financial investment issues, etc. [4,5,6]. However, in the field of studying farmers’ behavior to participate in agricultural insurance, the peer effect still lacks sufficient attention.

In previous studies, many scholars conducted their studies from an objective aspect: Some scholars believed that the main reasons for the agricultural insurance market failures are information asymmetry and high systemic risk [7,8,9]. In that case, the expected utility theory is used to explain the insufficient demand for agricultural insurance, including that the initial wealth accumulation will reduce farmers’ risk aversion consciousness [10]; the larger or more efficient the farm is, the more likely the farmer is to participate in agricultural insurance [11]. However, Chinese scholars attributed the agricultural insurance market failures to the nature of agricultural insurance as a quasi-public good. They believed that the main reason for the insufficient supply of agricultural insurance was inadequate government subsidies, which ultimately resulted in the “double freezes in supply and demand” in the agricultural insurance market [12]. In 2004, China officially established the policy-based agricultural insurance system and achieved nationwide coverage in 2012 [13]. China’s agricultural insurance has made great progress since then, but the overall situation is still not ideal. Since then, many studies have been published on the demand-side research of the agricultural insurance market. Zhang et al. [14] proposed that policy-based agricultural insurance essentially belonged to an agricultural support policy, whose importance lies not in the characteristic itself, but the features compared with other agricultural support tools. It is the insufficient demand that results in market failures. Wu et al. [15] pointed out that one of the keys to the success of agricultural insurance is the demanding analysis. Overall, the studies carried out by Chinese scholars on the problem of insufficient demand can be summarized as follows: (1) From a macro perspective, the suggestions put forward for the reform of the agricultural insurance system consist of the rural revitalization strategy. These suggestions include improving relevant laws and regulations, perfecting management systems, clarifying policy objectives, etc. [16], or refer to the policy frames of agricultural insurance development in other developed countries, such as developing and utilizing the income support function of agricultural insurance [17], gradually improving the guaranteed level of agricultural insurance, implementing differentiated regression proportional subsidy [18], etc. (2) From the micro perspective, studies focused on the influencing factors of farmers’ purchase of agricultural insurance, such as gender, education level, health conditions, and other individual characteristics; meanwhile, household factors such as cultivated land areas and household total income will also significantly affect farmers’ insurance participation decisions [19,20,21]. With the deepening study on farmers’ behavior to participate in agricultural insurance, a growing body of literature has noticed that the previous studies were mostly from the traditional expected utility theory, which did not explain the real insurance participation behavior, and made it hard for the conclusions to be applied to practice. Hence, from the perspective of behavioral economics, some scholars began to distinguish the difference between the farmers’ willingness for insurance participation and their participation behavior. They tried to explore the real law of farmers’ insurance participation behavior [22]. Some scholars also began to study the influence of social capital on farmers’ agricultural insurance purchasing decisions [23], or to explore the dynamic equilibrium of the agricultural insurance market by the evolutionary equilibrium principle [24].

As mentioned above, seldom has empirical literature tested the peer effect on farmers’ decisions to participate in agricultural insurance. Therefore, the main objective of this paper is to quantify the importance of peer effect in household agricultural insurance participation and further explore its internal mechanism. In this paper, we adopt an IVprobit model to identify the peer effect on the farmers’ participation in agricultural insurance using the 2020 China Karst Area Rural Economic Survey (CKRS) data. The CKRS data include Hubei, Guangdong, Guangxi, Sichuan, Guizhou, Hunan, Chongqing, and Yunnan Provinces in China, with 9452 farmers in total. We introduce two mediators of risk perception and insurance cognition, to further explore the main functionary channels of the farmers’ decision-making to buy agricultural insurance. Moreover, we find the peer effect is asymmetrical by dividing the whole sample into different groups.

Compared with the existing literature, our study contributes to the literature by showing that the peer effect can be applied to the study of agricultural insurance decision-making through empirical analysis. Our study finds that the probability of individual farmers participating in agricultural insurance will increase by 3.25% when the farmers’ neighbors’ participation rate increases by 10%. Specifically, we conducted the survey in the karst regions to help increase the agricultural insurance participation rate in the areas with weak agricultural production, and to provide some references for the future development of agricultural insurance in rural China. Meanwhile, the differential distribution of farmers’ participation rate may be caused by the particularity of geographical conditions, which may provide a new probable perspective for regional research on agricultural insurance in the future. The remainder of the paper is organized as follows. Section 2 briefly describes the theoretical framework of this study. Section 3 outlines our materials and the estimation strategy. Section 4 presents our empirical findings, and Section 5 presents further discussions about the peer effect. Section 6 concludes and discusses the policy implications of our findings.

2. The Theoretical Analysis

2.1. The Peer Effect on Farmers’ Agricultural Insurance Decisions: The Direct Effect

It is generally believed that the concept of the peer effect originated from interdisciplinary studies of psychology, sociology, and economics. It refers to the tendency of an individual to behave in accordance with a peer group due to the influence of their peers [25,26]. There is an obvious feature of “acquaintance society” in China. Fei [27] also proposed a “differential order pattern” to explain the individual-centered social network in China’s rural society, which is a network system based on traditional primary social relations such as blood relationships, geography, etc. Studies found that most farmers have not bought agricultural insurance mainly because of not being familiar with agricultural insurance information [28]. However, according to the theory of the peer effect, when facing the case of insufficient information, farmers will search for related information from their peers as much as possible to increase the cognition for agricultural insurance, and reduce the potential risk caused by the information asymmetry [29]. Eventually, their own decisions will converge with those in their peer group. At the same time, some existing literature also proves that peer effect is prevalent in the individual decision-making process and plays a positive role. Therefore, the following hypothesis is proposed in this paper:

Hypothesis 1 (H1).

Peer effect has a positive influence on farmers’ agricultural insurance buying decisions.

2.2. The Peer Effect on Farmers’ Agricultural Insurance Decisions: The Indirect Effect

According to the theory of planned behavior, the decision made by the individual will be affected by personal characteristics, including attitude, perception, and subjective norms [30]. Therefore, to spread the production risks, as the starting point for farmers to participate in insurance, individuals’ risk perception will significantly affect their insurance decisions. In addition, it is generally believed that the higher the risk perception level farmers have, the more likely they are to participate in agricultural insurance [31]. Similarly, farmers may struggle to have a clear understanding of some complex insurance clauses and the operating flows, which could be limited by their education level. All of these may weaken their insured willingness. Thus, if farmers’ agricultural insurance cognition can be strengthened, their misapprehension and even prejudice against agricultural insurance can be significantly reduced. When conventional hedging measures cannot avoid risks, farmers’ demand for agricultural insurance will greatly increase [32]. Moreover, in the theoretical analysis of the peer effect, scholars have divided its mechanism into three categories: The information effect, experience effect, and externality effect [33]. Specifically, the information effect mainly refers to the new information that the individual obtains from the peer group. The experience effect refers to the experience shared by the individuals who have made the decision first in the peer group, generally leading to a “critical mass” in group decision-making. The externality effect can motivate the decision-making of other individuals by causing non-marketable positive or negative effects on their welfare, generated by the individuals who take the lead in decision-making. In summary, our study considers that the peer effect can improve the farmers’ risk perception and insurance cognition through information exchange and experience demonstration, so as to promote their participation in agricultural insurance. Therefore, our study proposes the following hypothesis:

Hypothesis 2 (H2).

The peer effect promotes farmers’ participation in agricultural insurance by enhancing their risk perception and insurance cognition.

3. Materials and Methods

3.1. Data

Our data come from the 2020 China Karst Area Rural Economic Survey (CKRS), which covers 10,111 farmers of karst regions and 8 provinces, including Hubei, Guangdong, Guangxi, Sichuan, Guizhou, Hunan, Chongqing, and Yunnan Provinces. They contain rich information on villages and communities in rural karst regions, allowing researchers to study agricultural and economic development in the karst regions. The karst regions have been limited by geographical conditions that result in the problems of agricultural frangibility and inferiority. Therefore, it is crucial to study the development of agricultural insurance in the karst regions, which can be of great significance to developing local characteristic agriculture industry, guiding the proper distribution of regional agricultural production, and protecting and repairing the eco-environment [1]. Our survey uses a multi-stage sampling approach, which follows the principle of the first level of sampling in the PUS sampling box using the probability-proportional-to-size sampling (PPS) method, sampling 40 counties in total. Then, we use the PPS method to sample villages according to the administrative village serial number generated by the National Bureau of Statistics, there are more than 4 villages selected from each county, and, finally, our survey contains 641 villages. In terms of data cleaning, we eliminated the missing values of the sample data, which are 659 farmer samples in total. Finally, 9452 farmer samples were retained in our analysis.

3.2. Variables

Dependent Variable: Farmer’s agricultural insurance buying decision. Our questionnaire represents a farmer’s agricultural insurance buying decision by asking “Whether you have bought agricultural insurance in the past year”.

Core explanatory variable: The peer effect. In rural China, the village is the basic activity scope of rural residents, and their daily communications with each other mainly occur within the village. Meanwhile, when farmers need to make decisions independently, they often tend to face high risks brought by asymmetric information, environmental uncertainty, etc. Thus, they are more likely to access effective information from their peers to complete decisions, in other words, they make their final decisions consistent with their peers. Therefore, a village can be regarded as a closely related group, and the insurance participation rate of all farmers (excluding the respondent farmer) within a village is the peer effect experienced by each respondent farmer. Therefore, we use the respondent farmer’s neighbors’ participation rate in agricultural insurance as the proxy variable of each farmer’s peer effect.

Mediator: Farmers’ risk perception and insurance cognition. Our questionnaire used the Likert scale to measure farmers’ insurance cognition. Specifically, we asked respondents “How much do you know about agricultural insurance?” and we set 5 levels of answers from “Very well” to “Don’t know”. For the measurement of farmers’ risk perception, we divided it into subjective risk perception and objective risk perception, referring to the literature. By inquiring “Have you suffered any natural disasters in the past five years?”, we measured farmers’ objective risk perception; by inquiring “Do you rely on agricultural insurance to resist risk?”, we measured farmers’ subjective risk perception. Ultimately, we calculated the arithmetical average of subjective and objective risk perception to represent the risk perception level of each farmer’s household.

Instrumental variable: Farmer’s neighbors’ average political status. On the one hand, there may be some influences that cannot be observed in the study. On the other hand, the interplay between the farmers in villages may cause omitted variables and bidirectional causality problems. Therefore, we measured farmer’s neighbors’ participation rates using their average political status, referring to the literature [34]. An individual’s political status is known as a reasonable predictor of uptake of agricultural insurance decisions. Party members will be exposed to more policy publicity and may have a higher insurance cognition, and our subsequent study will prove this as well. Simultaneously, the farmer’s neighbors’ political status is an individual characteristic that can be treated as exogenous because it will not directly influence the respondent farmer’s decision to participate in agricultural insurance.

Control variables: Based on research needs and the existing literature [19,20,21], we selected control variables of household head, household, and village. Household head variables include age, gender, political status, ethnicity, education level, health conditions, and subsidy policies cognition. Household variables include types of farmer households, household monthly average expenditure, and household total cultivated land area. Village variables include village development level. In addition, our study also controls provincial dummy variables.

Table 1. Summary statistics.

Variable	Description	Mean	Std. Error	Max.	Min.
Dependent variable
Agricultural insurance buying decision	Whether farmer has bought agricultural insurance in the past year	0.077	0.266	1	0
Explanatory variables
Core explanatory variable
Peer effect	Farmer’s neighbors’ average participation rate of agricultural insurance	0.076	0.126	1	0
Mediator variables
Farmers’ insurance cognition	Don’t know = 1; Understand a little = 2; Commonly understand = 3; Familiar = 4; Very well = 5	1.522	0.904	5	1
Farmers’ risk perception	Objective risk perception: Whether having suffered any natural disasters in the past five years: Yes = 1; No = 0	0.241	0.427	1	0
	Subjective risk perception: Whether relying on agricultural insurance to resist risk: Yes = 1; No = 0	0.174	0.380	1	0
Household head variables
Age	Actual age	53.977	13.4391	88	15
Gender	Male = 1; Female = 0	0.729	0.444	1	0
Educational level	Years of schooling	7.122	2.917	18	0
Health conditions	Unable to function independently = 1; Chronic diseases = 2; Subhealth = 3; Good = 4; Very good = 5	3.779	1.025	5	1
Political status	Party member or not: Yes = 1; No = 0	0.051	0.219	1	0
Ethnicity	Han = 1; Ethnic minorities = 0	0.775	0.418	1	0
Subsidy policies cognition	Familiar with government subsidy policies: Yes = 1; No = 0	0.090	0.287	1	0
Household variables
Types of farmer household	Pure farmers = 1; Others = 0	0.711	0.453	1	0
Household monthly average expenditure	Actual household monthly average expenditure, logarithm (unit: CNY)	7.360	0.834	9.903	3.689
Household total cultivated land area	Actual household total cultivated land area (unit: acres)	3.488	7.945	265	0
Village variable
Village development level	Whether the village has cement roads: Yes = 1; No = 0	0.983	0.130	1	0

presents the summary statistics of key variables. The agricultural insurance participation rate of the full sample is 7.7%, and there are obvious distinctions among provinces (Figure 1): The average participation rate of farmers in the western region, including Guangxi, Sichuan, Guizhou, Chongqing, and Yunnan Provinces, is the highest (10%). The average participation rate of farmers in the central region including the Hubei and Hunan provinces is 2.59%, which is between the eastern and western regions. Meanwhile, the participation rate of farmers in the eastern region is the lowest (0.6%). This distribution is basically consistent with the previous study that showed that the insurance participation rate in central and western regions is higher than the rate in the eastern region, and the participation rate in each province is also close to the calculation results in the previous study. Both of them show that the sample in our study is highly representative [35]. Although it is slightly different from the three step decline distributions of participation rates from the central to the eastern region, our research takes karst regions as samples that may result in certain particularities in geographical conditions and agricultural productions, which provides a new perspective for the regional study of agricultural insurance to some extent. In addition, Figure 2 shows the distribution of the average insurance participation rate in each village, ranging from 0 to 10%, which basically coincides with the participation rate estimated by [35]. The average participation rate in each village is relatively concentrated, and there are only three out of 641 villages where the participation rate is more than 50%. We consider that the probable impact of strong government restrictions or intensive promotions is not yet evident and, furthermore, our study will explore those problems in the endogenetic treatment and robustness test.

3.3. Research Methods

As whether farmers have bought agricultural insurance in the past year is a typical binary selection problem, the probit model is selected in our study to analyze the peer effect on farmers’ agricultural insurance buying decisions. The model is set as Equation (1):

$$P_r\left(D=1|x_i\right)=\varphi \left({\beta }_0+{\beta }_{1}peerrati{o}_i+{\beta }_{2}control\right) ,$$

(1)

where “D” is the dummy variable for farmers’ agricultural insurance decisions. It takes the value “1” if the farmers have bought agricultural insurance in the past year, and “0” otherwise. The $\varphi$ function is a standard cumulative normal distribution function; $peerrati{o}_i$ indicates each farmer’s peer effect; ${\beta }_1$ indicates the regression coefficient of the core explanatory variable ($peerrati{o}_i$); ${\beta }_2$ represents the regression coefficient of each control variable; while ${\beta }_0$ is a constant term.

There may be omitted variables and bidirectional causality problems in the model specification and the hypotheses of this study. Specifically, there may be some decisive factors within villages, such as the interactions between the villagers, that were not observed in this study; or there may be some omitted variables that come from externality, such as the strong government restrictions or intensive promotions. All of these problems will make the stochastic disturbance term related to the endogenous explanatory variable that a consistent estimation cannot be obtained using the probit model alone. Therefore, we introduce an instrument variable to overcome endogeneity problems by using the IVprobit model. In particular, the ordinary least square (OLS) estimate is performed for the endogenous explanatory variable ($peerrati{o}_i$) with instrumental variable ($aco{m}_i$) and control variables in the first-stage regression. The fitted value of peer effect ($peerrati{o}_i^{*}$) is obtained, as Equation (2) shows:

$$peerrati{o}_i^{*}={\gamma }_0+{\gamma }_{1}aco{m}_i+{\gamma }_{2}control+{\mu }_1+{\eta }_i,$$

(2)

where ${\mu }_1$ is a region dummy variable; ${\eta }_i$ is a classical error term. In second-stage regression, Equation (3) is estimated using probit to obtain a consistent coefficient of the peer effect (${\lambda }_1$).

$$P_r\left(D=1|x_i\right)=\varphi \left({\beta }_0+{\lambda }_{1}peerrati{o}_i^{*}+{\beta }_{2}control\right)$$

(3)

In mechanism analysis, we use mediating effect and group regression. In mediating effect analysis, IVprobit is a nested non-linear probability model, uncontrolled and controlled coefficients can differ not only because of confounding but also a rescaling of the model. The difference in coefficients arises whenever the mediator variable has an independent effect on the dependent variable. To solve this problem, we refer to the KHB method proposed by Kohler, Karlson, and Holm to decompose the total effect into direct effect and indirect effect, which makes the results conform more precisely to the real coefficient difference and makes the mediating analysis more intuitive [36].

4. Results

We use Stata 16.0 for statistical analysis and

Table 2. Baseline effect of peer effect on farmers’ agricultural insurance decision.

	(1)	(2)
Peer effect	3.182 ***	0.325 ***
	(0.148)	(0.015)
Age	−0.006 ***	−0.001 ***
	(0.002)	(0.000)
Gender	0.092 *	0.009 *
	(0.056)	(0.006)
Political status	0.161 *	0.016 *
	(0.092)	(0.009)
Education level	0.010	0.001
	(0.009)	(0.001)
Types of farmer household	0.195 ***	0.020 ***
	(0.057)	(0.006)
Ethnicity	−0.056	−0.006
	(0.062)	(0.006)
Health condition	−0.016	−0.002
	(0.029)	(0.002)
Subsidy policies cognition	0.897 ***	0.091 ***
	(0.048)	(0.005)
Household monthly average expenditure	0.059 **	0.006 **
	(0.029)	(0.003)
Household total cultivated land area	0.009 ***	0.001 ***
	(0.002)	(0.000)
Village development level	0.100	0.010
	(0.214)	(0.022)
Control of regions	Yes	Yes
Constant	−2.346 ***
	(0.363)
LR chi2(19)	1487.91
Number of obs.	9452
pseudo R-squared	0.2908

Note: Reported values are estimated regression coefficients (column 1) and marginal effects (column 2) using probit model with standard errors clustered in brackets; *** p < 0.01, ** p < 0.05, * p < 0.1.

presents our baseline results using the probit model. The LR chi2(19) of the model is 1487.91, and the pseudo R-squared is 0.2908, which shows good goodness of fit for further study. Column (1) shows the variables’ coefficients of regression, and column (2) shows the marginal effects. The result shows that the peer effect has a significant influence on farmers’ participation in agricultural insurance: A 10% increase in farmers’ neighbors’ participation in agricultural insurance increases the likelihood of farmers’ participation by 3.25%. Hypothesis H1 is verified.

Moreover, from the perspective of individual characteristics. The coefficient of farmers’ age is negative and statistically significant at 1%. With a farmer’s age increasing by 1 year, the probability of participating in agricultural insurance will decrease by 0.1%. As farmers get older, they are more inclined to use traditional risk avoidance tools, and their understanding of policies could be weaker, so their willingness to buy insurance is also lower. Meanwhile, the coefficient of the farmers’ gender variable is positive and statistically significant at 10%, which means that male farmers are more likely to buy agricultural insurance. Compared with female farmers, the probability of male farmers participating in agricultural insurance will increase by 0.9%. Male farmers usually play the role of breadwinners and decision-makers in the household, so they are exposed to more new information about agricultural production. Therefore, male farmers are more likely to buy agricultural insurance to spread the agricultural production risks. The coefficients for the farmers’ political status are also statistically significant and positive, and the probability of party members participating in agricultural insurance will increase by 1.6% compared to non-party members. The most likely explanation is that party members are more likely to learn the relevant policies for the promotion and publicity of agricultural insurance, so they will have a deeper cognition and understanding of policies and insurance clauses. So, they are more likely to participate in agricultural insurance. The same argument can be used for the result of farmers’ subsidy policies cognition, which indicates that if a farmer’s insurance cognition increases by one point, the probability of buying agricultural insurance will increase by 9.1%.

Additionally, for farmers’ households, the coefficient of farmers’ household types is positive and statistically significant at 1%. The probability of pure farmers participating in agricultural insurance will increase by 2% compared with the part-time farmers. Pure farmers spend more time on agricultural production and may have a deeper understanding of relevant information, that is, they are more inclined to participate in agricultural insurance. At the statistical level of 1%, household cultivated land area significantly promotes farmers’ probability of buying agricultural insurance. With the expansion of farmers’ planting scale, natural risks will have a greater and greater impact on agricultural production. To spread risks and reduce probable losses, farmers with a larger planting scale could be more willing to purchase agricultural insurance at that time. In the meantime, as another indicator related to production risk, household monthly average expenditure significantly promotes farmers’ probability of buying agricultural insurance at the statistical level of 5%, which can reflect the cash flow of farmers’ households in production and living to some extent. With the household monthly average expenditure increasing by 1%, the probability of a farmer participating in agricultural insurance will increase by 0.6%. It may indicate that a farmer household with larger cash flows may face more uncertain risks in production, thus, they are more likely to spread the uncertain risks with agricultural insurance.

5. Discussion

Table 3. Effect of peer effect on farmers’ agricultural insurance decision (IVprobit results).

Dependent Variable: Agricultural Insurance Buying Decision	(1)	(2)
Peer effect		2.869 **
		(1.453)
Farmers’ neighbors’ political status	0.243 ***
	(0.019)
Controls	Yes	Yes
F-stat, instrument	122.79
Number of observations	9452
Weak instrument variable test
AR statistic	3.79 **
Wald statistic	3.90 **

Note: Reported values are estimated regression coefficients using the IVprobit model with standard errors clustered in brackets; Results for control variables in the IV probit model are not reported, those results are available upon request; *** p < 0.01, ** p < 0.05.

reports the IVprobit estimation results when the farmers’ neighbors’ average political status is used as the instrument variable. As shown in Table 2, the farmers’ neighbors’ political status significantly promotes their probability of buying agricultural insurance. Meanwhile, the neighbors’ political status, as a personal characteristic, will not directly influence the agricultural insurance buying decision of the respondent farmer.

The first-stage IVprobit result is reported in column (1), the farmers’ neighbors’ political status has positive and statistically significant coefficients; the first-stage F-statistic is above 10, which indicates that the instrument is statistically strong [37]. The second-stage IVprobit result is presented in column (2), the coefficient of the peer effect is also positive and statistically significant. Hence, from this table, we can conclude that the peer effect increases the likelihood that a farmer participates in agricultural insurance. Furthermore, the coefficient of the peer effect using IVprobit is less than the coefficient estimated in Table 1, which indicates that the probit model will overestimate the positive impact of the peer effect, and the estimations of the IVprobit model are closer to the real situation. In addition, in the case of this model, when the number of endogenous explanatory variables is equal to the number of instrumental variables, the overidentified test is not necessary according to the existing study [38].

Table 4. Robustness test of IVprobit model.

	(1)	(2)	(3)
Dependent Variable	Agricultural Insurance Decision	Insurance Payment	Low Participation Rate Sample
Peer effect	0.327 ***	0.282 ***	0.116 ***
	(0.015)	(0.014)	(0.022)
Controls	Yes	Yes	Yes
Number of obs.	9452	9452	3453
Pseudo R-squared	0.290	0.307	0.243

Note: Reported values are marginal effects using the IVprobit model with standard errors clustered in brackets; results for control variables are not reported, those results are available upon request; *** p < 0.01.

reports the robustness test results of the IVprobit model by replacing dependent variables and sample groups. Firstly, we use the logarithm of the payment for agricultural insurance as the dependent variable to repeat the above estimation. Column (2) shows the results that are consistent with the baseline estimation: The peer effect still has a significant influence on the farmers’ agricultural insurance buying decision. In addition, there are probable differences between local governments’ invisible implications, and some specific villages may be affected by promotions and publicity, which results in a higher overall participation rate to interfere with the test of the peer effect. Therefore, our study excludes the village with an insurance participation rate that is higher than the average level of the whole sample. The results are shown in column (3): The marginal effect of the peer effect is slightly reduced, but the peer effect still significantly promotes farmers’ agricultural insurance decisions, which also indicates that the empirical model in this paper is robust.

As the empirical analysis shows, the peer effect has a significant positive impact on farmers’ purchase of agricultural insurance. We introduce the proxy variables for risk perception and insurance cognition to further explore the mechanism. According to the above inference in this paper, if the peer effect promotes the farmers’ insurance decisions by influencing their risk perception and insurance cognition, the influence of peer effect in the groups with higher risk perception or insurance cognition should be greater. Hence, we divide the whole sample into two groups by the threshold of the average level of risk perception and insurance cognition, respectively, to test the peer effect in each group. The results are shown in

Table 5. Peer effect in different groups.

	Dependent Variable: Agricultural Insurance Buying Decision
	Higher Risk Perception	Lower Risk Perception	Higher Insurance Cognition	Lower Insurance Cognition
Insurance participation rate	0.473 ***	0.273 ***	0.752 ***	0.110 ***
	(0.036)	(0.016)	(0.074)	(0.011)
Number of obs.	2281	7171	1313	8139

Note: Reported values are marginal effects using the IVprobit model with standard errors clustered in brackets; results for control variables are not reported, those results are available upon request; *** p < 0.01.

. The farmers with higher risk perception and insurance cognition are significantly affected by the peer effect, suggesting that risk perception and insurance cognition may be the channels of peer effect in farmers’ agricultural insurance decision-making. Moreover, the KHB method is used to test the mechanism and further quantify the indirect effect.

Table 6. The decompositions of the peer effect.

	Coefficient	Proportion	Std. Error	z	p > |z|
Total effect	0.470	100.00%	0.023	20.43	0.000
Direct effect	0.299	63.62%	0.023	13.13	0.000
Indirect effect	0.171	36.38%	0.102	1.68	0.093

shows the results of the KHB method. The total effect of the peer effect is 0.470 and the direct effect is 0.299, both of which are statistically significant at 1%. The indirect effect is 0.171, composed of risk perception and insurance cognition, which is statistically significant at 10% and accounts for 36.38% of the total effect. Hence, the group regression and KHB decompositions indicate that the peer effect could promote farmers’ purchase of agricultural insurance directly and indirectly through promoting risk perception and insurance cognition. Therefore, there are no grounds for rejecting Hypothesis H2.

Studies have shown that when individuals are making decisions under the influence of their peers, there may be “leaders” in the group that influence the decision of the whole group [39]. It is of great significance for us to clarify the possible leaders in the purchase decision of agricultural insurance. By mobilizing the participation enthusiasm of leaders and giving full play to the significant demonstration effect of leaders, can we better carry out policy publicity and promotions of agricultural insurance in the future.

It has been proved that male farmers are more likely to buy agricultural insurance in Table 1. Therefore, our study further divides the whole sample into a male group and female group to test the influence of the full-sample peer effect, and we also test the influence of the peer effect within the groups on their insurance participation, respectively. The results are shown in

Table 7. Heterogeneity analysis.

	Dependent Variable: Agricultural Insurance Buying Decision
a. Group by gender	Full sample	Male sample	Female sample
Females’ participation rate within the same village	0.282 ***		0.155 ***
	(0.029)		(0.025)
Males’ participation rate within the same village	0.350 ***	0.313 ***
	(0.018)	(0.017)
Number of obs.	9452	6878	2273
b. Group by cultivation scale	Full sample	Larger-scale cultivated sample	Smaller-scale cultivated sample
Larger-scale farmers’ participation rate	0.227 ***		0.226 ***
	(0.016)		(0.017)
Smaller-scale farmers’ participation rate	0.462 ***	0.302 ***
	(0.031)	(0.022)
Number of obs.	9452	2818	6542

Note: Reported values are marginal effects using the probit model with standard errors clustered in brackets; results for control variables are not reported, those results are available upon request; *** p < 0.01.

: Both the peer effects in the whole sample and the subsample have a greater influence on the male sample’s insurance buying decision. In other words, the male farmers participating in agricultural insurance may have a demonstration effect. Similarly, the larger the areas of cultivated land are, the greater the risk that farmers may have to bear which results in the increasing demand for agricultural insurance. Therefore, the whole sample is divided into larger-scale and smaller-scale farmers with an average cultivated land area of 3.49 acres as the boundary. The results are the same as the groups divided by gender, as shown in Table 7. The peer effect in the whole sample and subsample plays a more important role among the larger-scale farmers. In summary, male farmers and larger-scale farmers have more significant effects on their peers and probably lead the participation in agricultural insurance.

6. Conclusions

In this paper, we examine the peer effect on farmers’ agricultural insurance decisions, based on the survey data of 9452 farmers in the karst regions of Hubei, Guangdong, Guangxi, Sichuan, Guizhou, Hunan, Chongqing, and Yunnan Provinces. We find that the peer effect has a significant positive influence on farmers’ participation in agricultural insurance within a village, and a 10% increase in farmers’ neighbors’ participation in agricultural insurance increases the likelihood of farmers’ own participation by 3.25%. We also find that the peer effect promotes farmers’ agricultural insurance participation by enhancing farmers’ risk perception and insurance cognition. Meanwhile, we find the heterogeneous effects that exist in the peer effect: Male farmers and larger-scale farmers have larger effects on their peers and probably lead the farmers’ participation in agricultural insurance.

As a related finding to the existence of opinion leaders, our evidence suggests: (1) Increasing policy publicity and enforcing policy advocacy, which would magnify the positive impact of the peer effect, such as building a digital information-sharing platform so that the farmers can access more comprehensive technical and policy information. (2) Increasing the participation rate of specific groups by policy interventions, as male and larger-scale farmers have larger effects on their peers and probably lead the agricultural insurance decisions, we could perfect the subsidy policy to encourage moderate scale management of agriculture to give full play to the positive demonstration effect of specific groups. (3) Innovating insurance publicity methods to enhance farmers’ risk awareness in agriculture production and improving farmers’ insurance cognition. On the one hand, we could conduct more relative educational lectures. On the other hand, it would be more efficient if we simplify the publicity materials, avoiding just a single jumble of insurance product details, as we can add more simple cases and process diagrams to explain the operation process of agricultural insurance. (4) Focusing more on primary-level governance and construction in rural China. As the basic administrative unit in China, village communities occupy most of the production and living activities in time and space for residents. We should make full use of village resources, strengthen elementary education, and organize more group activities to advance the stock of rural social capital on all fronts, increasing the sense of identity among farmers and the sense of belonging, to leverage the role of the peer effect within rural areas.