The Impact of Crop Insurance on Fertilizer Use: Evidence from Grain Producers in China

Abstract

This study aims to examine the influence of crop insurance on the utilization of chemical fertilizers using plot-level data. The dataset utilized in this analysis consists of information obtained from 1039 participants residing in four major grain-producing provinces (Heilongjiang, Zhejiang, Henan, and Sichuan) in China. To address the potential issue of endogeneity, instrumental variables were employed to establish a causal relationship within the empirical model. The findings of this study indicate that crop insurance does not exert a statistically significant impact on overall fertilizer input in China. Nonetheless, the effect varies across different categories of farmers. Specifically, large-scale farmers experience a moderate reduction in fertilizer input as a result of crop insurance, while small-scale farmers do not demonstrate a significant effect. It is essential to strike a balance between risk protection and the potential influence of moral hazard in order to enhance future crop insurance policies.

1. Introduction

The application of fertilizers in China has significantly increased crop yields in the past few decades. However, this has also caused a large number of adverse effects on the agricultural environment, such as soil deterioration [1,2,3] and greenhouse gas emissions [4,5]. Therefore, many policy instruments, aiming to reduce fertilizer use, have been implemented by the Chinese government. For example, in 2015, the Chinese government launched “Action Plans to Achieve Zero Growth of Chemical Pesticides and Fertilizers” by 2020 [6].

This paper examines whether or not purchasing crop insurance can affect the use of fertilizer. China launched crop insurance, and since insurance changes the expected income distribution of farmers, their production behaviors are bound to change accordingly [7]. On the one hand, under the condition that crop insurance can diversify the risk of future income, farmers may be encouraged to take more risky production decisions [8,9]. On the other hand, farmers who purchase crop insurance may reduce the input of production factors because the production losses, induced by natural disasters, can be partly covered by insurance [9,10,11].

The existing studies have focused on the effect of crop insurance on the usage of fertilizer. But the findings still remain mixed. Some studies have found that crop insurance will encourage farmers to increase chemical inputs such as fertilizers and pesticides [8,12,13]. On the other hand, subsequent studies have shown a negative relationship between insurance and chemical input use [14,15,16], especially Weber et al. (2016) [16], who found that the relationship between crop insurance and input factors was not significant.

Therefore, this paper examines the relationship between crop insurance and fertilizer use based on rice and maize production in China. First, the input of fertilizer is the largest factor contributing to the growth of grain output in China. Second, the quantity of fertilizer used is close to 4 times the world average level. The overuse of fertilizers in China has become the main cause of agricultural non-point source pollution [5].

This paper contributes to the existing research in three aspects. First, an analysis of the relationship between crop insurance and fertilizer use in the Chinese context is provided. China has different crop insurance policies and farmer characteristics that determine the effect of crop insurance on fertilizer use. Therefore, it is necessary to further examine the relationship between crop insurance and fertilizer use based on observation in China. Second, we employed the instrument variable method to address the endogeneity problem of crop insurance among farmers arising from reverse causality and omitted variable bias. Third, the heterogeneous effects of crop insurance on fertilizer use were examined, which can provide policy implications adapted to different scales of farms.

We structured the rest of this paper as follows. Section 2 introduces the policy background of crop insurance in China. Section 3 presents the empirical model, data collection, and estimation strategy. Section 4 reports the estimation results and robustness test. Conclusions and policy implications are discussed in Section 5.

2. Theoretical Mechanism and Policy Background

2.1. Theoretical Mechanism

After purchasing crop insurance for farmers with the goal of income maximization, the income risk can be diversified to a certain extent [16], and the probability of disaster loss is reduced, which will inevitably adjust input decisions with risk-management effects, such as fertilizer use. On the one hand, under effective risk protection, farmers may reduce the use of fertilizer in order to save costs. On the other hand, crop insurance can diversify income risks and alleviate liquidity constraints under the influence of moral hazard, which can promote farmers to actively use more fertilizer.

The impact on fertilizer also depends on whether fertilizer is risk-increasing or risk-decreasing. Quiggin et al. (1993) [17] conclude that fertilizer is generally viewed as a risk-increasing input. But Babcock and Hennessy (1996) [9] found that an input could be both risk-decreasing and risk-increasing at different levels of application. If fertilizer is a risk-increasing input, then farmers may have reason to increase the fertilizer input. The reason is that fertilizer can increase the probability of increasing yield, but the increased risk induced by increasing the use of fertilizer would be shared by the crop insurance. Similarly, if fertilizer is a risk-decreasing input, farmers would choose to decrease fertilizer use.

As stated earlier, mixed evidence exists regarding the relationship between crop insurance and fertilizer use. A similar situation can also be found in papers examining the relationship between crop insurance and other input factors, such as pesticide [18,19,20] and land [21,22]. The reasons for this result may be due to the changes that fertilizer is increasing or decreasing the risk of, which are affected by different countries, farmers’ production levels, insurance policy design, and other factors. Therefore, it is necessary to analyze the impact of agricultural insurance on fertilizer input in the context of China.

2.2. Crop Insurance in China

Crop insurance is an important tool for diversifying natural risks and stabilizing farmers’ income. China officially launched the policy-based crop insurance program in 2007, which subsidizes the purchase of crop insurance for important agricultural products such as wheat, corn, rice, and breeding sows that are related to the national economy and people’s livelihood. With heavy government subsidies, crop insurance in China has seen a rapid expansion. According to the data from the China Insurance Regulatory Commission, the underwriting area of major crops in China reached 2.1 billion mu in 2017, an increase of 1.869 billion mu compared with 2007; the premium reached RMB 47.91 billion, 9 times that of 2007. Crop insurance has become an important part of China’s agricultural support policy.

A prominent characteristic of crop insurance policy in China is the increasing adoption rate and areas covered with low insurance premiums and amounts. A low premium makes the insurance affordable for farmers. However, the low amount associated with low premiums also only covers the cost of agricultural production and cannot cover all the losses of farmers induced by natural disasters.

To avoid the high transaction costs caused by scattered farmers, insurers generally choose to cooperate with villager committees to underwrite insurance, that is, the villager committees serve as an intermediary organization connecting farmers and insurers. In addition, despite the principle of voluntary participation, insurers are still more likely to conduct business in villages where most farmers or plots are willing to participate in insurance.

The maize insurance in Heilongjiang province in 2016 can be taken as an example. In Heilongjiang province, farmers can choose three types of insurance products, the indemnities are RMB 155, 250, and 320 per mu, and the corresponding insurance premiums are RMB 15, 20, and 25 per mu. The government subsidy ratio is 80% for all products, which is to say that farmers only need to pay RMB 3–5, i.e., 20% of the premiums. It can also be found that the upper limit of the insurance indemnity that the insured farmer will receive is still relatively low, generally not exceeding 20% of the final production value of the crop.

When settling claims, insurers calculate the indemnity based on the yield losses caused by disasters. However, in order to curb the moral hazard of farmers, insurers generally set a deductible amount of about 20–30%, which is to say that the insurers only compensate for the 20–30% of the total yield loss, thus increasing the proportion of farmers’ risk-sharing. When the loss caused by natural disasters exceeds 30%, the compensation will be paid according to the following equation: payment = indemnity × [1 − loss ratio] × insured area.

The loss ratio is calculated using the ratio of the standard yield to the actual yield. The standard yield is the average actual county-level production yield in the past 5 years. It will be stated in the insurance policy.

The payment also varies depending on the growing period at the time of the disaster. For example, if the natural shock occurs between the emergence period and the first flowering period, 40% of the indemnity shall be paid at most; if it occurs between the flowering period and the milk period, the compensation shall be 70% of the indemnity. During the final flowering period, 100% of the indemnity will be paid if the occurrence happens.

In addition, the payment is calculated based on the village-level yield, not the individual plot. Insurers often avoid the expense of assessing losses on a per-plot basis. Instead, they typically calculate the total payment based on the average yield loss per village and distribute the funds among insured farmers according to the severity of the disaster. Such a procedure can reduce the operation costs of the insurers but would probably increase the risk basis for the insured farmer and then reduce the probability of obtaining a sound indemnity for the individual farmer.

2.3. Heterogeneity in Farm Size

According to insurance policy regulations, all insured farmers should receive equal compensation when facing the same level of loss. However, the size of the farming operations may influence the probability of receiving full compensation, depending on the negotiating power of the insured farmers and the insurance companies. Firstly, large-scale farmers heavily rely on agricultural income as a significant economic source, making compensation funds crucial for sustaining agricultural production. Secondly, a larger insured area corresponds to higher compensation amounts, increasing the motivation to seek greater compensation while also enhancing bargaining power in negotiations between farmers and insurance companies.

By comparison, small-scale farmers are often at a disadvantage in the process of claiming compensation. Since the agricultural income of small-scale farmers accounts for a very low proportion of the total household income, the land area is small, and the total amount of the insurance compensation funds is very low, and their motivation to fight for insurance compensation is relatively weak, so the actual production guarantee level of its insurance products and the probability of obtaining claims are relatively low.

2.4. Chemical Fertilizer Use in China

China is the largest consumer of fertilizers for agricultural production in the world. The amount of fertilizers used in China was 46.98 million tons in 2018, which accounts for 24.92% of the world’s total fertilizer consumption (FAOSTAT). Considering the adverse environmental and health effects associated with the overuse of fertilizers, the Chinese government has started to reduce fertilizer use through various programs, including promoting biopesticides and organic fertilizers to substitute chemical pesticides and fertilizers and providing technical training of advanced agronomic practices. We can find that fertilizer use in China since 2015 shows a decreasing trend, although it still remains increasing before that (Figure 1).

3. Model and Data

3.1. Model and Variables

This study employed an instrumental variable regression model to estimate the impacts of crop insurance on fertilizer use.

$$y_i=b_0+b_{1}insu{r}_i+b_2{X}_i+e_i$$

In the above formula, the dependent variable represents the fertilizer expenditure per unit area used by farmers. While the data also includes information on the quantity of fertilizer input, the survey results revealed that farmers have a more accurate recollection of the total expenditure of fertilizer used on the plot. Therefore, this study selected the fertilizer input value as the dependent variable for the regression analysis and utilized the fertilizer input quantity for robustness testing.

One limitation of utilizing this variable is the potential influence of regional disparities in fertilizer prices on the quantity of fertilizer input. However, given the wide range of fertilizer types employed by farmers (including nitrogen, phosphorus, potassium, and compound fertilizers), there exists significant variation in the prices of these different types, posing challenges for accurately predicting their overall impact. To address this concern, a standardized price measurement was adopted; nevertheless, farmers often accurately recall the prices associated with various fertilizer types. Nonetheless, considering the relatively negligible fluctuations in fertilizer prices within county-level regions, this study incorporated county-level regional dummy variables to partially mitigate the influence exerted by fertilizer prices.

$insuri$ indicates whether the plot is insured or not. Other control variables, denoted as $X_i$, encompass plot characteristics, family characteristics, and regional dummy variables. Plot characteristics consist of variables such as plot size, plot property rights, plot quality, organic fertilizer application status, and crop type. Family characteristics mainly involve the age of the head of the household, years of education, years of agricultural planting, and the proportion of non-agricultural employment.

3.2. Endogeneity and Estimation Strategy

3.2.1. Endogenity of Insurance Decision

Fertilizer input and crop insurance both serve as crucial mechanisms for farmers to mitigate risks in agricultural production. Fertilizer input acts as a proactive risk-management approach, effectively reducing the likelihood of risk occurrence. Conversely, crop insurance functions as a reactive risk-management strategy by sharing risks and minimizing potential losses, these two factors may exhibit a model due to their interdependent influence on each other. Previous studies have also indicated that although fertilizer input and insurance decisions are made at different stages during a production cycle, they are jointly determined [10,23].

To address the endogeneity issue, instrumental variables need to be selected to accurately identify the causal relationship within the model. While previous studies have employed simultaneous equations to analyze the impact of crop insurance on factor inputs and selected variables such as the yield variation coefficient, total sown area, farmers’ concurrent employment status, and risk preference as instrumental variables for insurance decision-making, there has been a lack of comprehensive discussion and demonstration regarding their selection. Moreover, these instrumental variables exhibit varying degrees of endogeneity and weak instrument strength. Therefore, this paper aimed to incorporate more exogenous instrumental variables into the regression analysis.

The instrumental variables for insurance decision-making used in this study primarily included whether crop insurance is implemented at the township level, the proportion of village participation in insurance, and the actual premium rate. Currently, crop insurance subsidies in China are jointly funded by central, provincial, and county finances. However, due to financial constraints in certain counties, the county-level government may be unable to provide supporting subsidies and thus choose not to implement or only partially implement them in select townships. Consequently, the implementation of crop insurance at the township level significantly influences farmers’ purchasing behavior but does not directly impact fertilizer input.

The average village participation rate also serves as an effective instrumental variable. Given that the diffusion of new agricultural technologies heavily relies on neighborhood relatives and friends [24], insurance participation behavior will inevitably influence farmers’ decisions to participate in insurance. However, this behavior does not directly impact farmers’ fertilizer investment choices. Consequently, this study selected the village’s average insurance participation ratio as an instrumental variable.

The premium and insured amount of crop insurance products refer to their respective prices and expected income, which directly affect farmers’ decision-making process regarding purchasing insurance but do not influence fertilizer input levels. Therefore, this paper employed the rate or ratio between premiums and insured amounts as an instrumental variable for analyzing insurance purchasing decisions. Considering that government subsidies significantly contribute to insurance premiums, it is crucial to accurately measure the actual premiums borne by farmers by deducting these subsidies.

3.2.2. Instrumental Variable Estimation

In this study, the decision of farmers to participate in insurance was measured using dummy variables, making the endogenous variables binary. For estimating instrumental variables with binary endogenous explanatory variables, previous studies have employed a two-stage estimation method [25]. This approach involves using Probit or Logit regression in the first stage to obtain the coefficient for insurance decision-making. The fitted values from the first stage are then directly substituted into the second stage for least-squares estimation. However, the conditional expectation function $E[insur|y,X]$ in the first-stage regression is nonlinear, and its residual is only asymptotically uncorrelated with other control variables. Directly substituting it into the second stage would result in a problem known as “forbidden regression”, where efficient estimates of parameter variances cannot be obtained [26].

To address this issue, Agrist et al. (2009) [27] suggested using the fitted value of the nonlinear regression as an instrumental variable and conducting a two-stage least-squares estimation, which yields more accurate and efficient estimation results. Additionally, Smith and Goodwin (1996) [10] utilized the bootstrap method to adjust the standard error, providing a solution to the problem mentioned above.

Theoretically, all three methods mentioned can yield consistent estimators of parameters, but there may be variations in the validity of the estimation results. For the estimation method in this paper, the suggestion of Agrist et al. (2009) [27] was adopted, where the fitted value obtained from the Logit model estimation of the participation decisions was used as an instrumental variable, followed by a two-stage least-squares regression. However, to verify the robustness of the estimated results and the effectiveness of different identification strategies, this paper also employed the other two methods mentioned for regression analysis as additional robustness tests.

3.3. Data Source and Description

In contrast to most previous studies that utilized county-level and farm-level data, this paper utilized survey data at the plot level to enhance control over important variables such as plot quality and production conditions. To ensure an adequate sample size and randomness in the sampling process, a stratified random sampling method was employed to collect relevant data. Before the investigation commenced, four major grain-producing provinces (Heilongjiang, Zhejiang, Henan, and Sichuan) were identified, followed by the random selection of four counties in each province based on grain output. To test the research hypothesis and analyze the impact of crop insurance on fertilizer input for farmers of different scales, the sample needed to include both large-scale and small-scale farmers. Consequently, two townships with a higher number of large-scale households were selected in each county, and four administrative villages were chosen in each township. Then, six large-scale households and ten small-scale farmers were selected from each village, resulting in a total of sixteen farmer samples. To compare the production behaviors of large-scale and small-scale households, small-scale households were selected in proximity to the large-scale households. Consequently, a total of 64 households were sampled per county, resulting in a total of 1039 sample farmers surveyed across the four provinces. This survey was carried out in 2015 by four research teams from the Chinese Academy of Agricultural Sciences, Renmin University of China, China Agricultural University, and Nanjing Agricultural University.

Since this paper focuses on refining the research to the plot level, it was necessary to randomly select specific information regarding the sample farmers’ operating plots. During the specific sampling process, two plots were randomly chosen from all the planting plots of the farmers. To effectively control the impact of land property rights, the largest plot was selected from the farmer’s land and transferred land to better reflect the farmer’s production decisions. However, as some farmers solely owned their land or only had transferred land, they were only able to have one plot selected. Consequently, although there were 1039 sample farmer households, there were ultimately only 1707 plots included in the analysis.

Table 1. Variables description.

	Variables	Unit	Mean	SD	Min	Max
Dependent Variables	Log fertilizer expenditure	RMB/mu	4.855	0.622	0	6.89
	Log fertilizer quantity	kg/mu	1.73	0.598	0	2.53
Independent Variables	Insured or not	1 = yes 0 = no	0.364	0.481	0	1
	Plot quality	1 = good 2 = Medium 3 = Bad	1.624	0.642	1	3
	Rented in	1 = yes 0 = no	0.425	0.494	0	1
	Manure	1 = yes 0 = no	0.320	0.467	0	1
	Plot size	Mu	18.49	75.39	0.1	1750
	Farming years	Years	31.87	13.70	0	67
	Crop type	1 = Corn 0 = rice	0.533	0.499	0	1
	Gender	1 = Male 0 = Female	0.965	0.183	0	1
	Education	years	6.791	3.071	0	16
	Rate of non-farm labor	%	0.411	0.337	0	1
	Large farmer or not	1 = year 0 = no	0.439	0.497	0	1
Instrumental Variables	Village participation rate	%	48.67	39.38	0	100
	Implemented in town or not	1 = yes 0 = no	0.847	0.361	0	1
	Premium rate	%	0.015	0.027	0	0.333

presents the descriptive statistics of the main variables examined in this paper.

To visually examine the relationship between crop insurance and fertilizer use, we categorized the sample plots into insured and uninsured groups to observe the differences in fertilizer input among the different sample groups. As shown in Figure 2, on average, the farmers in the insurance group used less fertilizer than those observed for both the large-scale and small-scale farmers; however, the difference was more significant for large-scale household farmers. These preliminary findings suggest that crop insurance may have varying impacts on fertilizer use among different types of farmers; however, further analysis is needed to control for other variables.

4. Estimation Results

4.1. The Impact of Crop Insurance on Fertilizer Use

The first column in

Table 2. The impact of crop insurance on fertilizer use.

	OLS	IV
		(1)	(2)	(3)	(4)	(5)
Insured or not	0.0344	−0.392 ***	−0.141 *	−0.125	−0.193	−0.172
	(0.0316)	(0.0799)	(0.0799)	(0.0786)	(0.138)	(0.117)
Plot quality
Medium	0.0540 *		0.0538 *	0.0688 **	0.0434	0.0514
	(0.0319)		(0.0308)	(0.0318)	(0.0311)	(0.0321)
Bad	0.0858 *		0.0896 *	0.0986 *	0.0783	0.0823
	(0.0507)		(0.0511)	(0.0512)	(0.0507)	(0.0510)
Property right	−0.00392		−0.0132	−0.0120	−0.00524	−0.00727
	(0.0307)		(0.0306)	(0.0318)	(0.0293)	(0.0309)
Manure	0.0347		−0.0111	−0.00395	0.0264	0.0274
	(0.0345)		(0.0332)	(0.0330)	(0.0353)	(0.0352)
Plot size	−0.0004		0.0006	0.0003	0.00003	0.00004
	(0.00021)		(0.00021)	(0.0002)	(0.00022)	(0.00021)
Crop type	0.298 ***		0.262 ***	0.282 ***	0.314 ***	0.313 ***
	(0.0549)		(0.0330)	(0.0359)	(0.0558)	(0.0566)
Farming years	0.000343			0.000350		−0.000113
	(0.00112)			(0.00109)		(0.00116)
Gender	0.165			0.232 *		0.144
	(0.114)			(0.123)		(0.116)
Education	−0.00721			−0.00244		−0.00750 *
	(0.00448)			(0.00482)		(0.00455)
Large farmer or not	−0.00616			0.0153		0.0193
	(0.0287)			(0.0305)		(0.0306)
Rate of non-farm labor	0.115 **			0.156 ***		0.0997 **
	(0.0452)			(0.0483)		(0.0447)
County dummy	Yes	No	No	No	Yes	Yes
Constant	4.367 ***	4.998 ***	4.744 ***	4.427 ***	4.548 ***	4.401 ***
	(0.156)	(0.0313)	(0.0483)	(0.150)	(0.0969)	(0.159)
Observations	1707	1707	1707	1707	1707	1707
R2	0.150	0.048	0.051	0.065	0.121	0.130
Prob > chi2	0.000	0.000	0.000	0.000	0.000	0.000
F value in first stage	-	407.45	63.81	45.93	44.77	42.12
Over-identification test	-	0.11	0.67	0.75	0.62	0.69

Note: brackets in the table are robust standard errors, and *, **, and *** indicate significance levels of 10%, 5%, and 1%, respectively.

presents the OLS estimation results of the basic regression equation. The coefficient of insurance decision-making indicates that, according to the OLS estimation, the influence of insurance decision-making on fertilizer input is positive but not statistically significant. To address potential endogeneity issues, the instrumental variable method was employed, as shown in columns two to six of Table 2. Models (1) to (5) present the regression results with instrumental variables, incorporating control variables such as plot characteristics, family characteristics, and regions in a stepwise manner. The p-values of the over-identification test for all five models exceeded 0.1, hence failing to reject the null hypothesis that the instrumental variables are exogenous. Additionally, the smallest F statistic for the first-stage joint significance test in the five models was 42.12, all exceeding the threshold of 10, indicating the absence of weak instrumental variable problems. Therefore, the instrumental variable regression method effectively addressed the endogeneity concerns and enabled a more accurate estimation of the impact of insurance decision-making on fertilizer input among farmers. In the results of the instrumental variable regression, model (1) did not include any control variables, and the coefficient for insurance decision-making was, significantly, −0.392 at the 1% significance level. Upon adding land characteristics to model (2), the coefficient remained negative but was not statistically significant. This suggests that after accounting for family and regional characteristics, the decision to participate in insurance does not significantly affect fertilizer input. It also implies that the impact of insurance decision-making on fertilizer input varies across different types of families, supporting the hypothesis of this study. This indicates that the impact mechanism of insurance decision-making on fertilizer input differs among different types of farmers. Thus, this study further analyzed the impact of crop insurance on the fertilizer input of farmers of different scales. Model (5) represents the estimated results after including all the control variables, revealing a coefficient of −0.172 for insurance decision-making, which remained statistically insignificant. Overall, these findings suggest that the current crop insurance mechanism in China does not have a significant impact on farmers’ fertilizer input. The land property variables in the regression results were not statistically significant, likely because property rights stability primarily influences long-term investments closely related to land fertility, such as organic fertilizers [28], rather than short-term inputs like fertilizers on the plot.

4.2. The Impact of Crop Insurance on Fertilizer Use among Farms of Different Scales

To analyze the impact of crop insurance on fertilizer input among farmers of different scales, this study conducted separate regression analyses for large-scale and small-scale farmers. It is important to note that the criteria for categorizing farmers into different scales vary across regions and are typically based on the scale being at least three times higher than the local average. These large-scale households receive subsidies. To accurately capture the influence of crop insurance on fertilizer input for farmers of different scales, this study defined large-scale households as those with a scale of more than three times the local average land management scale, while the rest were considered small-scale farmers.

Table 3. The impact on farmers among farms of different scales.

Dependent Variables	Large-Scale Farmers	Small-Scale Farmers
Insured or not	−0.279 *	−0.102
	(0.147)	(0.283)
Plot quality
Medium	0.0182	0.0720
	(0.0447)	(0.0493)
Bad	0.152 *	0.0180
	(0.0826)	(0.0716)
Property right	0.0166	−0.0185
	(0.0435)	(0.0450)
Manure	0.00001	−0.0023
	(0.00024)	(0.0019)
Plot size	0.184 ***	0.405 ***
	(0.0699)	(0.0833)
Crop type	−0.110 *	0.132 ***
	(0.0604)	(0.0425)
Farming years	−0.00115	−0.000357
	(0.00169)	(0.00167)
Gender	0.0829	0.283 *
	(0.199)	(0.151)
Education	−0.000265	−0.0195 ***
	(0.00681)	(0.00669)
Large farmer or not	−0.00421	0.175 ***
	(0.0652)	(0.0625)
Rate of non-farm labor	—	—
County dummy	Yes	Yes
Constant	4.553 ***	4.366 ***
	(0.284)	(0.189)
Observations	751	956
R2	0.133	0.173
Prob > chi2	0.000	0.000

Note: the brackets in the table are robust standard errors, and * and *** indicate significance levels of 10% and 1%, respectively.

represents the regression equations for large-scale and small-scale farmers. The results indicate that for large-scale farmers, the coefficient of crop insurance on fertilizer input was −0.279, which was statistically significant at the 10% level. This suggests that, compared to uninsured farmers, insured large-scale farmers exhibited a 27.9% decrease in fertilizer input. Conversely, the coefficient for small-scale farmers was −0.102, but it was not statistically significant. These findings confirm the research hypothesis of this study. The mechanism at play is two-fold. Firstly, large-scale farmers rely more heavily on agricultural income as a proportion of their total household income. Consequently, they have a stronger awareness of risk management and adopt diverse risk-management strategies, including production inputs and crop insurance. Secondly, despite relatively low levels of insurance coverage, large-scale households possess a larger land area, allowing them to access greater total insurance compensation, which can mitigate the impact of adverse events on production. Furthermore, when negotiating claims with insurance companies, large-scale households possess greater bargaining power, enabling them to secure higher insurance compensation. Therefore, participating large-scale households exhibit a certain degree of moral hazard, leading to a reduction in fertilizer input. In contrast, small-scale farmers have a significantly smaller proportion of agricultural income contributing to their total household income. Consequently, their total insurance compensation is also modest, exerting minimal impact on households and production. Consequently, small-scale farmers lack the motivation and bargaining power to seek higher insurance compensation, resulting in weaker moral hazard. Therefore, their participation in insurance does not affect fertilizer investment. The analysis reveals that, under the same terms and operating mechanisms of crop insurance, both large-scale and small-scale farmers optimize their input choices based on their specific factors and resource endowments, resulting in different responses to fertilizer input behavior.

4.3. Robustness Checks

4.3.1. Different Estimation Methods

Different treatment methods have been proposed in previous studies for binary endogenous explanatory variables. This paper aimed to test the robustness of the regression results under different treatment methods by using the two-stage estimation method and the bootstrap-adjusted standard error two-stage estimation method.

Table 4. Robustness checks result.

		All	Large-Scale Farmers	Small-Scale Farmers
		(1)	(2)	(3)
Two-stage estimation	Insured or not	−0.155	−0.282 *	−0.105
		(0.124)	(0.151)	(0.292)
	Observations	1707	751	956
	R2	0.150	0.160	0.186
	Prob > F	0.000	0.000	0.000
Bootstrap-S.E.-adjusted two-stage estimation	Insured or not	−0.155	−0.282 *	−0.105
		(0.125)	(0.147)	(0.290)
	Observations	1707	751	956
	R2	0.150	0.160	0.186
	Prob > F	0.000	0.000	0.000
Fertilizer quantity as the dependent variable	Insured or not	−0.120	−0.314 **	0.0504
		(0.107)	(0.158)	(0.266)
	Observations	1707	751	956
	R2	0.138	0.133	0.204
	Prob > F	0.000	0.000	0.000
Controls	Plot	Yes	Yes	Yes
	Farm	Yes	Yes	Yes
	County dummy	Yes	Yes	Yes

Note: * and ** indicate significance levels of 10% and 5%, respectively. Due to space constraints, regression results for other control variables are not reported here.

presents the regression results using the two-stage method and bootstrap-adjusted standard error two-stage method in the first two rows. Comparing the estimation results of the two treatment methods, we found that the coefficient and significance level of insurance participation were the same under both methods, with only a small change in the standard error. This result implies that insurance participation behavior has no significant effect on fertilizer input among farmers. However, for large-scale households, the coefficient of insurance participation was −0.282, and both coefficients were significant at the 10% significance level, suggesting that insurance participation behavior can, to some extent, reduce fertilizer input for large-scale households. On the other hand, for small-scale farmers, the coefficient of insurance participation was −0.105 and was not significant, indicating that insurance participation behavior has no significant impact on the fertilizer input of small-scale farmers. Comparing these results with the previous estimation results, we observed no significant changes in the coefficient size and significance level of these two processing methods, thus obtaining consistent research conclusions. This indicates that the regression results in this paper are robust.

4.3.2. Fertilizer Quantity as an Independent Variable

The fertilizer application value was utilized to quantify the level of fertilizer input by farmers. To test the robustness of these results, this study also employed the fertilizer quantity to perform a regression analysis and to determine if similar conclusions could be derived. The regression analysis using fertilizer quantity as the dependent variable is presented in the third row of Table 4. It is evident that, for all the samples, the coefficient of insurance participation was negative but insignificant, suggesting that the decision to participate in insurance has no substantial impact on farmers’ fertilizer input. However, in the regression equation for large-scale households, the coefficient of insurance participation was −0.314 and significant at a 5% significance level. This implies that, compared to uninsured farmers, insured farmers experience a 31.4% decrease in fertilizer input. Although the coefficient in the regression equation for small-scale farmers was positive, it was not significant, suggesting that insurance participation does not have a significant effect. The fertilizer input by small-scale farmers also has no significant impact. The results align with the regression analysis of the fertilizer input value, as there were no substantial changes in the regression coefficients or significance. This paper’s findings exhibit good robustness.

5. Conclusions and Discussion

Taking fertilizer use as an example, this study examined the impact of crop insurance on farmers’ input behavior and its underlying mechanism. Unlike previous studies, the instrumental variable method was employed to assess the influence of crop insurance on fertilizer input by farmers of different scales using plot-level survey data. The findings demonstrate that the impact of crop insurance on input behavior is contingent upon the risk nature of inputs and the level of farmers’ moral hazard. Specifically, the analysis shows that while crop insurance does not have a significant overall impact on fertilizer input, its effect varies across different farmer types. Notably, it moderately reduces the fertilizer input of large-scale households but does not significantly affect small-scale farmers. This discrepancy can be attributed to the current operational mechanism of crop insurance in the country, which leads to strong moral hazard among large-scale households and provides them with ample motivation to reduce the risk-reducing input factor of fertilizers.

The results of this study can also account for the disparities among existing studies. Firstly, different inputs have distinct risk properties. The existing research examines inputs such as pesticides, fertilizers, and new technologies, each of which carries its risk level, thereby yielding disparate findings. Secondly, the operational mechanism of crop insurance varies across different regions, directly influencing the degree of farmers’ moral hazard. In China, the current policy framework adheres to the principle of “low security and wide coverage” and incorporates indemnity or deductible clauses to mitigate moral hazard. Consequently, crop insurance in China generally has an insignificant impact on fertilizer investment at the farmer level. However, in practice, large-scale households possess greater bargaining power in seeking compensation, exhibit relatively higher moral hazard, and have stronger incentives to reduce fertilizer investment. This discrepancy arises due to the differences in the operational mechanisms of foreign crop insurance, resulting in incongruent research results.

The conclusions of this paper hold significant policy implications. Firstly, crop insurance can effectively reduce the fertilizer investment of large-scale households, aligning with China’s green agriculture development strategy and contributing positively to the reduction in fertilizer usage and control of non-point source pollution. However, reducing the level of fertilizer investment through crop insurance may potentially have adverse impacts on agricultural production. Since agricultural production relies on various inputs, the effect of crop insurance may differ depending on the risk properties of these inputs. Therefore, it is imperative to conduct further comprehensive research to explore the relationship between crop insurance and other inputs in the future.

Secondly, the findings of this study offer valuable insights into the development of crop insurance in China. Currently, crop insurance offers relatively low levels of risk protection. While it successfully curbs farmers’ moral hazard, its risk-sharing function remains limited. Hence, there is a need to focus on developing crop insurance products with higher levels of risk protection that can truly support and benefit farmers. Particularly, the responsiveness of large-scale households to crop insurance production and their heightened risk-management requirements emphasize the importance of formulating insurance products specifically tailored to this group. Such a development is crucial for enhancing crop insurance coverage in the future. As the level of risk protection improves, the moral hazard effect of farmers may also intensify, potentially influencing agricultural production input and output. Balancing the impact of crop insurance on ensuring farmers’ income and national food security and promoting agroecological sustainability is, therefore, a vital issue that demands immediate attention in refining crop insurance policies and the development of related products.