The Impact of Government Subsidies on Income Inequality Among Farm Households in China: Evidence from CFPS Panel Data

Abstract

Effectively measuring the income-generating effects of government subsidies is an important issue in assessing agricultural policies and implementing precision poverty alleviation. This study utilizes panel data from the China Family Panel Studies (CFPS) to screen a sample of 13,583 individual residents. We empirically analyze the impact of government subsidies on income inequality among farm households using fixed effects models, quantile regressions, and mediated effects models. Our study found that the following results: (1) The mechanism of subsidy action combines the functions of increasing income and regulating income distribution. Marginal benefits to low-income groups are more obvious. (2) Increasing productive agricultural inputs can reduce the income gap of farmers, which is more significant than the direct subsidy of funds. (3) The incentive effect of subsidies is strongest in the Northeast and weakest in the East. Compared to middle-aged and older farmers, the utility level of the subsidy is stronger in the youth group.

1. Introduction

Addressing China’s urban–rural development inequality and narrowing the income gap between residents has long been a concern of policymakers. The report of the twentieth session of the Communist Party of China (CPC) likened the concept of “common prosperity” to the concept of “making the cake bigger and better, and cutting it well” [1]. On this basis, the Third Plenary Session of the 20th CPC Central Committee further pointed out that “a system should be formed to effectively increase the income of the low-income group, steadily expand the size of the middle-income group, and reasonably regulate excessively high incomes” [2]. In order to serve national strategic objectives, the intervention of government subsidies as a coercive tool is needed. Resource redistribution and industrial support can, to a certain extent, interrupt the intergenerational transmission of poverty and promote the vertical mobility of social capital [3,4]. Therefore, systematic subsidy intervention is driven by the pressure of economic transformation and the need for social equity [5].

The reality that cannot be ignored is that, compared with urban residents, rural residents are in a disadvantaged position in the areas of income, expenditure, and public services, and the imbalance in income distribution within rural areas even exceeds that within cities and between urban and rural areas [6,7]. In 2023, the per capita disposable income of the low-income group of rural residents was RMB 5264, while that of the high-income group was RMB 50,136, a ratio of 1:9.52, compared with 1:6.33 for urban residents. The amount of per capita consumption expenditure was similarly disparate, with a ratio of 1:1.82 between rural and urban residents. The data also showed that China’s rural Gini coefficient had reached 0.467 in 2022. The wide income gap, on the one hand, exacerbates the process of rural impoverishment and creates bottlenecks for economic development by suppressing effective demand and capital investment [8]. Meanwhile, it has weakened the degree of participation in collective action in villages, leading to an increase in social instability [9]. Therefore, how to address income inequality within farming households has become a real problem that needs to be solved urgently, posing a major challenge to China’s high-quality development.

Many scholars have discussed in depth the factors affecting income inequality among farm households, with regional disparities in levels of development being the most significant influence [10,11,12]. For farmers, the available land area directly affects the income from agricultural output. Part of the region has a better natural endowment base, people engaged in non-farm activities tend to receive higher remuneration [13,14]. The transfer of agricultural land has led to an increase in the Gini coefficient of the farmers’ income, and overall has widened the income gap between farm households [15]. However, scholars differ on the income situation of transfer-in and transfer-out households. Li et al. found that land transfers increased the income for both transfer-in and transfer-out households, but that the income-enhancing effect was greater for higher-income groups than for lower-income groups [16]. Liu et al., on the other hand, argued that land transfers have a negative effect on the income distribution of transfer-out households, while the opposite is true for transfer-in households [15]. From a human capital perspective, increasing the size of the labor force can directly contribute to the growth of household economic returns [17]. Meanwhile, the optimization of human capital can both improve the efficiency of land use and help to expand non-farm employment channels [10]. Inhibiting the process of demographic aging and synergistically promoting the accumulation of educational capital, investment in health and the optimization of household factor endowments can effectively regulate the pattern of income distribution disparities [18,19,20,21]. Wage income and business income are another critical factor explaining the income gap among rural residents [22,23]. Non-farm part-time activities generate income-boosting Matthew effects through differences in factor allocation, and their elasticity coefficients show a significant advantage in the group of high-resource-endowed farm households [24,25]. The factor return premium brought about by ICT favors high-income as well as high-knowledge groups [26]. The ability of highly educated farming groups to rely on digital technology to access information has been strengthened, inducing a polarization of income distribution [27,28]. However, some scholars believe that the information equality and the gig economy brought about by digital technology access has reduced the level of income inequality through the restructuring of the labor market [29,30].

We find that previous research on the impact of farm household income inequality often focuses on physical capital, human capital, and regional geography, while research on government subsidies focuses more on their ability to increase farmers’ incomes and to change growers’ business decisions. Fewer scholars have paid attention to the impact of government subsidies on farm household income inequality. Therefore, the purpose of this paper is to explore whether government subsidies can effectively reduce the internal income inequality of farm households, and the mechanism of its influence. The rest of the article is organized as follows: Section 2 formulates the research hypotheses and develops the framework structure. Section 3 outlines the research methodology and data used in this article. Section 4 provides the results of the baseline regression, heterogeneity analysis, robustness test, and mechanism test. Section 5 discusses the study and summarizes the conclusions.

2. Theoretical Framework and Research Hypotheses

2.1. Government Subsidy and Farm Household Income

Government subsidies, as transfer income for farm households, have a structural moderating effect on intra-group income disparities [11]. The magnitude of this adjustment is closely related to individual resource endowments and the level of regional development [31]. In terms of income effects, subsidies, as exogenous transfers, directly enhance the disposable funds and consumption capacity of low-income residents [32]. The study by Yin and Guo showed that China’s precision poverty alleviation policy increased per capita consumption of poor households by 4.37%, and that subsistence consumption and developmental consumption increased by 5.76% and 13.12%, respectively [33]. Heterogeneity in the marginal propensity to consume brings about a deeper equalizing effect [34]. Poor households convert 82% of subsidized funds into immediate consumption (necessities such as food, medical care, etc.), while rich households consume only 35%. This gradient difference in consumption elasticity gives the policy a pro-poor livelihood improvement effect [35]. In terms of production effects, the government subsidy indirectly promotes farmers’ incentives to grow food, further enlarging productive income [36]. It further enriches their wallets through indirect ways, such as improving the dilemma of farmers’ credit constraints and promoting production [37]. Motivated by the policy, farmers may choose to expand the size of their land and increase labor inputs to improve their operating income [38]. Such long-term mechanisms to activate production can sink human capital, help increase the school enrolment of farmers’ children, and mitigate intergenerational inequalities [39].

Accordingly, Hypothesis H1 is proposed:

H1: 

Government subsidies can increase the income of farming households.

2.2. Positive and Negative Effects of Government Subsidy

Many scholars have discussed the ability of subsidy policies to reduce the incidence of inequality [40,41,42]. However, with the depth of the research, it has been argued that there may be a paradox in the utility of government subsidies [43]. For example, the stimulus effect of the policy is very limited, and it has little effect on regulating the distribution of farmers’ income, even widening the gap between rich and poor [44]. The discussion of negative effects focuses mainly on the creation of dependency and elite capture. Government subsidies’ gratuitous universality makes beneficiaries psychologically and physiologically dependent, and the process of passive waiting can easily lead to a poverty trap [45,46]. The significant reduction in people’s willingness to work means that the subsidies become, to a certain extent, the invisible income of the insured families, which in effect creates a situation of continuous dependence [47,48]. Similarly, many scholars have pointed out that government subsidies may be subject to elite capture [49,50]. For instance, agricultural subsidies are more often targeted at the economically more affluent large agricultural households [51]. After receiving policy funding support, they will aim to maximize their own interests and use special poverty alleviation funds and policy concessions to create large profits. This process is difficult to generate a pro-poor effect, and the trickle-down effect enjoyed by ordinary farming households is very restricted [42]. Therefore, it has been argued that the gap between large and small farmers still exists, and that elite capture has instead become a form of deprivation for low-income groups [52].

In recent years, government subsidies have made an enormous contribution to China’s fight against poverty [53,54]. On the one hand, government subsidies provide direct assistance to low-income people; while on the other hand, developmental government subsidies cultivate farmers’ own income-generating capacity and promote the continuous improvement of long-term income-generating capacity. Therefore, considering China’s development situation, the government subsidies defined in this paper are divided into two levels. The first is safeguard policy subsidies, such as low-income insurance, special hardship assistance and relief payments. The second is agricultural development subsidies, such as direct food subsidies and agricultural machinery subsidies.

Accordingly, Hypothesis H2 is proposed:

H2: 

Government subsidies have a dampening effect on farm household income inequality and can reduce the farm household income gap.

2.3. Intermediation of Agricultural Inputs

The existing studies generally agree that subsidies have a positive impact on the allocation of agricultural production factors [55,56,57]. The empirical analysis by Zhong et al. indicated that special agricultural support (subsidies for the purchase of food and agricultural machinery) can significantly increase the economic returns of the land operators [32]. Such policy instruments are universal in nature and effectively regulate the structure of internal income distribution in rural areas through the equalization of distribution mechanisms, creating production incentives at the micro level [58,59]. Beneficiary farmers usually convert subsidy funds into factor inputs, such as land improvement, labor optimization, and the upgrading of agricultural machinery and equipment [60]. This kind of factor reorganization is reflected in the linear expansion of the crop sowing area, which is associated with a substantial increase in production efficiency per unit of labor time [61]. Some studies have further revealed the dual impact of agricultural capital deepening the pattern of income distribution [62]. Wang et al. concluded that agricultural productivity enhancement creates labor force transfer [63]. Rural households are able to break through the land factor bondage and realize Pareto improvements in the non-farm employment market. However, it was also pointed out that this moderating effect is significantly heterogeneous [64]. Some groups of rural households convert financial transfers into earmarked assets, creating incremental returns to scale through mechanization as a substitute for labor [65]. Instead, the trend towards the polarization of income distribution was exacerbated.

Accordingly, Hypothesis H3 is proposed:

H3: 

Agricultural inputs play a partial mediating role in the effect of government subsidies on farm household income inequality.

In summary, the research framework of this paper is shown below (Figure 1):

3. Materials and Methods

3.1. Data Sources

This paper utilizes data from the China Family Panel Studies (CFPS). Given that the information on the receipt of government subsidies, household income, and personal characteristics of farm households nationwide is recorded in detail in the CFPS, we use the panel data for a total of three years, 2014, 2016, and 2018 for empirical analysis. Before data processing, to ensure data consistency, we matched the household economic module with the adult module in the database through household coding. For sample selection, “your current household status” was strictly defined as “agricultural household”. Missing data and data with serious outliers were excluded. Finally, a total of 13,583 individual samples were obtained.

3.2. Variable Descriptive Statistics

3.2.1. Independent Variable

We set the independent variable as government subsidy, which is reflected in the database as the “total amount of government subsidy (yuan/year)” and logarithmically treated as a continuous variable lnx0. In the CFPS questionnaire, the amount of government subsidy is the equivalent amount of various types of subsidies in cash or in kind received by the respondent from the government in the past year. It includes low-income insurance, subsidies for returning farmland to forests, agricultural subsidies (including direct food subsidies, agricultural machinery subsidies, etc.), and subsidies for five-guaranteed households.

3.2.2. Dependent Variable

The dependent variables of this paper are set as the net income of farm household and income inequality. The variable lny1 is generated by taking the logarithmic treatment of the net income of farm household. It consists of five major categories of itemized income summing up wage income, business income, property income, transfer income, and other income. Given that the Kakwani index can overcome the shortcomings of the Gini coefficient that does not satisfy the summation and decomposability, and has good properties in income fitting, so it can reflect the income inequality problem more accurately. Therefore, the income inequality of farm households refers to the study of Yang and Deng [66] and Deng et al. [6], and the Kakwani relative deprivation index is used to measure income inequality to generate the incD1 variable.

3.2.3. Control Variable

In the choice of control variables we consider the three aspects of human capital characteristics, family social characteristics, and geographical factor characteristics. First, considering the impact of age, education, health, and household size on income distribution, the age, edu, hea, and size variables were defined for analysis. The age variable is calculated from the birth year of the household head. Educational attainment is measured by the “highest level of education”. The physical condition variable was measured by “how healthy do you think you are”. Household size is measured by the number of people in the household. Next, the social characteristics of the household were analyzed by selecting the situation of agricultural land transfer, land expropriation, migrant labor, and the degree of importance attached to the Internet. In the case of agricultural land transfer, the zc variable is generated according to “whether the land is rented out to others”, on the one hand, and the zr variable is generated according to “whether your family rents out land to others”, on the other hand. In the case of expropriation, the zd variable is generated by “whether or not you have experienced land expropriation”. The dg1 variable was generated based on the “number of migrant workers” for the migrant labor situation of the farm household. Then, the net variable is generated to measure the degree of digitization of the household based on the importance of the Internet as a channel of information for farmers. Finally, given that regional differences are an important factor contributing to income disparities among farm households [13,67], we select geographic regions to measure inter-regional differences and generate the area variable.

Descriptive statistics for the variables are shown in

Table 1. Statistical description of variables.

Variables	Definition of Variables	Full Sample		2014		2016		2018
		Mean	SD	Mean	SD	Mean	SD	Mean	SD
lnx0	Government subsidy amount	4.288	3.225	4.710	2.723	4.358	3.211	3.986	3.460
lny1	Farm household net income	10.643	0.986	10.280	1.205	10.610	0.891	10.880	0.859
incD1	Income inequality	0.488	0.233	0.462	0.271	0.522	0.209	0.472	0.226
age	Household head’s age	35.950	13.460	34.550	13.070	35.430	13.500	37.220	13.510
edu	Educational level is distributed in increasing order from 1–8 No schooling, Elementary school, Middle school, High school, Secondary school, Vocational education, University, Masters and above	2.655	1.615	2.275	1.069	2.249	1.143	3.246	2.014
hea	Health level is distributed in increasing order from 1–5	3.273	1.187	3.374	1.189	3.254	1.178	3.234	1.192
size	Household size	4.963	1.981	5.038	1.820	5.111	1.927	4.784	2.099
zd	Land expropriation have = 1, otherwise = 0	0.069	0.254	0.068	0.252	0.064	0.244	0.075	0.264
zr	Land transfer in have = 1, otherwise = 0	0.174	0.379	0.191	0.393	0.200	0.400	0.141	0.348
zc	Land transfer out have = 1, otherwise = 0	0.142	0.349	0.070	0.255	0.088	0.284	0.233	0.423
dg1	Number of migrant laborers	1.028	1.070	1.063	1.104	1.089	1.083	0.951	1.033
net	Internet importance in increasing order of distribution from 1–5	2.856	1.594	2.337	1.510	2.755	1.579	3.220	1.561
area	Region Western = 1, Northeastern = 2, Central = 3, Eastern = 4	2.410	1.230	2.521	1.205	2.339	1.227	2.414	1.241

below:

3.3. Model

3.3.1. Fixed Effects Model

Given that we used panel data for three years, 2014, 2016, and 2018, a Hausman test reveals p < 0.05, constructing a fixed effects model as follows:

$$Y_{it}=\mathsf{\alpha }+{\beta }_0{X}_{it}+{\beta }_i{Z}_{it}+\gamma +{\mu }_t+{\epsilon }_i$$

(1)

In Equation (1) above, $Y_{it}$ denotes net household income as well as income inequality, $X_{it}$ represents government subsidies, $Z_{it}$ indicates a series of control variables affecting individual characteristics of income, social characteristics of the household, and characteristics of geographic factors, $\alpha$ is a constant term, ${\beta }_0$ and ${\beta }_i$ stand for the parameter to be estimated, $\gamma$ is an intercept term for individual heterogeneity that does not vary over time, ${\mu }_t$ stands for a time fixed effect, and ${\epsilon }_i$ is a random perturbation term.

3.3.2. Quantile Regression Model

Considering that mean regression describes the average effect of the explanatory variables on the explained variables, it is susceptible to being perturbed by extreme values that are too large or too small, and may even lead to biased results due to the omission of variables that cannot be directly measured. Hence, we refer to the studies of Koenker and Bassett [68], Lv and Hu [69], Li et al. [18], Wan and Wang [70] to analyze the impacts of policy subsidies on farm households of different income classes. The quantile regression model established is as follows:

$$Q_t\left(\mathrm{Y}\right|\mathrm{X})=\alpha \mathrm{A}+{\beta }_0{X}_{it}+{\beta }_i{Z}_{it}+{\epsilon }_i$$

(2)

In Equation (2) above, $Q_t\left(\mathrm{Y}\right|\mathrm{X})$ represents the net household income of the sample farmers at quartile t, $X_{it}$ indicates the government subsidy variable, $Z_{it}$ is a series of control variables affecting net household income, ${\epsilon }_i$ is the estimation error, and α and $\beta$ are the parameters to be estimated. Specifically, we categorize farm household incomes into five different income classes: low-income group, low-middle income group, middle income group, middle-high income group, and high-income group. A total of five quartiles, 0.1, 0.25, 0.5, 0.75, and 0.9, are taken to analyze the impact of government subsidies on the income levels of different farm households.

3.3.3. Mediation Effects Model

In order to further explore whether government subsidies, in addition to the direct income-raising effect, also have the role of indirectly regulating income distribution, we use the Sobel test to analyze the mediating effect of agricultural inputs between government subsidies and farm household income inequality. First, test the total effect, c, of government subsidies on farm household income inequality in Equation (3) and observe whether the impact coefficient passes the significance test. Second, test the indirect effects, a and b, arising from the mediating variable of agricultural inputs separately. Finally, test the direct effect, c′, of government subsidies on farm household income inequality according to Equation (5). If the coefficient, c′, is significant, then the mediation effect is “partially” significant. The mediation effect model is constructed as follows:

$${incD}_{it}=\mathsf{\alpha }+c{X}_{it}+{\beta }_i{Z}_{it}+\gamma +{\mu }_t+{\epsilon }_i$$

(3)

$${Tr}_{it}=\alpha +a{X}_{it}+{\beta }_i{Z}_{it}+\gamma +{\mu }_t+{\epsilon }_i$$

(4)

$${incD}_{it}=\alpha +c^{\prime }{X}_{it}+b{Tr}_{it}+{\beta }_i{Z}_{it}+\gamma +{\mu }_t+{\epsilon }_i$$

(5)

where ${incD}_{it}$ represents the income inequality status of farm households, $X_{it}$ is the government subsidy status, ${Tr}_{it}$ is the mediating variable agricultural inputs, $Z_{it}$ is the control variable, ${\epsilon }_i$ shows a random perturbation term, $\gamma$ stands for the intercept term of individual heterogeneity that does not vary over time, and ${\mu }_t$ stands for a time fixed effect. In particular, the agricultural input ${Tr}_{it}$ uses the amount of money spent by the farm household in the past 12 months on cultivation, forest land, and related to farming production, treated in logarithmic terms.

4. Results

4.1. Baseline Result

The results of the benchmark regression in

Table 2. Baseline regression results.

Variable	(1)	(2)
	Net Household Income	Income Inequality Among Farm Households
lnx0	0.019 ***	−0.004 ***
	(5.84)	(−5.42)
age	0.001	0.000
	(0.05)	(0.07)
edu	0.017 **	−0.007 ***
	(2.04)	(−3.39)
hea	−0.004	−0.001
	(−0.41)	(−0.25)
size	0.085 ***	−0.023 ***
	(9.93)	(−11.06)
zd	0.258 ***	−0.073 ***
	(6.94)	(−8.06)
zr	0.038	−0.010
	(1.43)	(−1.58)
zc	0.009	−0.006
	(0.34)	(−1.01)
dg1	0.179 ***	−0.043 ***
	(18.43)	(−19.17)
net	0.006	−0.002
	(0.79)	(−1.11)
area	0.149 ***	−0.031 ***
	(3.28)	(−2.81)
_cons	9.067 ***	0.754 ***
	(18.35)	(6.40)
Year dummy variable	Controlled
N	13,583	13,583
r2	0.215	0.117
F	128.3	79.98

Note: *** p < 0.01; ** p < 0.05.

showed that the government subsidy facilitates the improvement of net household income and, at the same time, reduces income inequality among farmers. The data indicated that both coefficients are significant at the 1% level. Hypotheses 1 and 2 are preliminarily tested. The mechanism of subsidy action is revealed to have both income-raising and redistributive functions.

In terms of control variables, it is basically consistent with theoretical expectations. Specifically, it is manifested as follows: (1) The effect of educational attainment on income is positive and passes the test of significance at the 5% level. It indicates that the literacy level of residents is positively correlated with household income. The regression result of educational attainment on income inequality is negative and significant at the 1% level. It represents that, for every 1-year increase in the average educational time limit of residents, the income Gini coefficient can be reduced by 0.7 percentage points. (2) Family size and the number of migrant workers are both significant at the 1% level. This indicates that the increase in family size and the number of migrant workers enriches human capital and promotes the rise of total family income. Labor mobility increases the possibility of low and middle-income people to break the original solidified class, and realizes the reduction of the internal income gap in rural areas. (3) Land expropriation has a significant positive effect on net income. Its coefficient reaches 0.258 and passes the 1% significance level test. It shows that the land expropriation policy implemented by the government is financially beneficial to farmers. Compensation also obviously alleviates the income distribution pattern. (4) Geographic location and digitization have a contributing effect on income growth, but exacerbate the degree of internal inequality. The former is significant at the 1% level, while the latter is statistically insignificant.

In addition, the heteroskedasticity test and autocorrelation test showed the Breusch–Pagan test p-value > 0.05 and the Durbin–Watson statistic of 1.9, indicating that the model residuals meet the assumption of homoskedasticity, and there is no significant autocorrelation. The VIF of each variable is less than 10, suggesting that there is no multicollinearity in the selection of control variables (not listed due to space constraints, available from the authors upon request).

4.2. Quantile Regression

Table 3. Quantile regression results.

Variable	FE	QR_10	QR_25	QR_50	QR_75	QR_90
lnx0	0.019 ***	0.014 ***	0.005	0.002	−0.002	−0.005
	(0.003)	(0.005)	(0.003)	(0.002)	(0.002)	(0.003)
age	0.001	−0.002	0.000	0.003 ***	0.004 ***	0.005 ***
	(0.016)	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)
edu	0.017 *	0.115 ***	0.103 ***	0.087 ***	0.086 ***	0.100 ***
	(0.009)	(0.013)	(0.009)	(0.006)	(0.006)	(0.009)
hea	−0.004	0.029 **	0.024 **	0.028 ***	0.018 ***	0.007
	(0.009)	(0.014)	(0.009)	(0.007)	(0.007)	(0.009)
size	0.085 ***	0.059 ***	0.074 ***	0.073 ***	0.084 ***	0.094 ***
	(0.008)	(0.008)	(0.006)	(0.004)	(0.004)	(0.006)
zd	0.258 ***	0.050	0.178 ***	0.264 ***	0.334 ***	0.458 ***
	(0.034)	(0.060)	(0.042)	(0.029)	(0.029)	(0.041)
zr	0.038	0.087 **	0.088 ***	0.097 ***	0.064 ***	0.051 *
	(0.027)	(0.041)	(0.028)	(0.019)	(0.020)	(0.028)
zc	0.009	0.099 **	0.112 ***	0.119 ***	0.147 ***	0.166 ***
	(0.027)	(0.046)	(0.032)	(0.022)	(0.022)	(0.031)
dg1	0.179 ***	0.329 ***	0.251 ***	0.155 ***	0.078 ***	0.025 **
	(0.009)	(0.015)	(0.010)	(0.007)	(0.007)	(0.010)
net	0.006	0.084 ***	0.063 ***	0.073 ***	0.058 ***	0.044 ***
	(0.008)	(0.012)	(0.009)	(0.006)	(0.006)	(0.008)
area	0.149 ***	0.095 ***	0.085 ***	0.080 ***	0.079 ***	0.063 ***
	(0.048)	(0.013)	(0.009)	(0.006)	(0.006)	(0.009)
Year dummy variable	Controlled
_cons	9.067 ***	7.436 ***	8.441 ***	9.113 ***	9.687 ***	10.150 ***
	(0.555)	(0.116)	(0.080)	(0.055)	(0.056)	(0.078)

Note: *** p < 0.01; ** p < 0.05; * p < 0.1.

displays that, of the five quartiles, only the low-income group satisfies positive significance at the 1 percent level. At the lower end (10% quantile), the lower quartile (25%), and the median (50%) positions of the income conditional distribution, the subsidy exhibits a stable boosting characteristic. However, the marginal effect of the subsidy on household contribution is negative for the upper-middle-income and high-income groups. The low-income group is most affected by the subsidy. As the quartile rises, the extent of the effect tends to diminish (coefficients gradually decrease).

In addition, the control variables exhibit marked between-group differences in the income-stratified regressions, specifically as follows: (1) The age variable shows a significant negative effect at the bottom of the income distribution (10% quantile). Whereas it shows a positive effect at the 1% significance level in the middle and upper income ranges. (2) Regression coefficients are positive (significant at the 1% level) across all quartiles for the educational attainment, family size, out-of-home labor, importance of the Internet, and regional variables. At the 90% quantile, migrant labor has the smallest effect on net household income. (3) The regression coefficients of healthiness are positive for all quartiles. It is not significant only at the 90% quantile. (4) Except for the 10% quantile, the remaining four quantiles are positively significant at the 1% level on the land acquisition variable. All groups are positive for both land transfer in and land transfer out. Only the absolute values of the coefficients differed between the groups.

Although the coefficients in the quantile regressions are able to represent, to some extent, the difference in marginal contributions by income class, however, it cannot be determined whether this difference is statistically significant. We therefore conducted a quantile difference test (

Table 4. Quartile difference test results.

Variable	Q10–Q50	Q10–Q75	Q10–Q90	Q20–Q75	Q25–Q90
lnx0	−0.011 ***	−0.015 ***	−0.019 ***	−0.006 **	−0.010 **
	(0.004)	(0.004)	(0.005)	(0.003)	(0.004)
age	0.005 ***	0.006 ***	0.007 ***	0.004 ***	0.005 ***
	(0.001)	(0.001)	(0.002)	(0.001)	(0.001)
edu	−0.028 **	−0.029 **	−0.015	−0.017 *	−0.003
	(0.012)	(0.013)	(0.015)	(0.009)	(0.011)
hea	−0.002	−0.012	−0.022	−0.007	−0.017
	(0.012)	(0.011)	(0.015)	(0.010)	(0.011)
size	0.014 **	0.025 ***	0.036 ***	0.010 **	0.020 ***
	(0.007)	(0.008)	(0.009)	(0.005)	(0.007)
zd	0.214 ***	0.284 ***	0.409 ***	0.155 ***	0.280 ***
	(0.060)	(0.068)	(0.080)	(0.044)	(0.076)
zr	0.009	−0.024	−0.036	−0.024	−0.036
	(0.037)	(0.040)	(0.043)	(0.025)	(0.035)
zc	0.020	0.047	0.067	0.034	0.054 *
	(0.035)	(0.039)	(0.047)	(0.025)	(0.032)
dg1	−0.174 ***	−0.251 ***	−0.304 ***	−0.174 ***	−0.226 ***
	(0.013)	(0.012)	(0.013)	(0.010)	(0.012)
net	−0.011	−0.027 **	−0.040 ***	−0.005	−0.018
	(0.011)	(0.013)	(0.015)	(0.008)	(0.011)
area	−0.015	−0.016	−0.031 **	−0.006	−0.022 *
	(0.012)	(0.015)	(0.014)	(0.009)	(0.012)
Year dummy variable	Controlled
_cons	1.676 ***	2.250 ***	2.714 ***	1.246 ***	1.709 ***
	(0.099)	(0.107)	(0.132)	(0.096)	(0.118)

Note: *** p < 0.01; ** p < 0.05; * p < 0.1.

). The results show that all five columns of coefficients are negative in direction. The coefficients in the first three columns are significant at the 1% level and the coefficients in the last two columns are significant at the 5% level. The absolute value of the coefficient in column three is the largest, indicating that the subsidy suppresses the extreme variance most strongly.

4.3. Mediation Effects Test

Table 5. Mediation effects test results.

Variable	(1)	(2)	(3)
	incD1	lntr	incD1
lntr			−0.004 ***
			(0.00)
lnx0	−0.004 ***	0.146 ***	−0.0001 ***
	(−5.42)	(0.006)	(0.00)
Control variables	Controlled	Controlled	Controlled
Adj R2	0.117	0.268	0.207
F-value	79.98	451.86	295.86
Sobel	0.000 ***
Goodman-1	0.000 ***
Goodman-2	0.000 ***

Note: *** p < 0.01.

presents the results of the mediation effect analysis. The data show that the effect of government subsidies on both farm household income inequality and agricultural inputs is significant at the 1% level (Column (1) and Column (2)). Column (3) indicates that the coefficients of subsidies and agricultural inputs on economic disparities are significant at the 1% level. This illustrates that the effect of subsidies on farmers’ income structure is partially mediated through the mediating effect of agricultural inputs. The Sobel, Goodman-1, and Goodman-2 tests support the existence of this mediating effect. Accordingly, hypothesis 3 is confirmed.

4.4. Heterogeneity Discussion

4.4.1. Government Subsidy Type Heterogeneity

Since the CFPS data does not disclose the specific categories of government grants for each year from 2014 to 2018, here we simply use the available cross-sectional data for 2014 to test for heterogeneity. The results are shown in

Table 6. Heterogeneity results across government subsidy types.

Variable	(1)	(2)
	Agricultural Development Subsidy	Safeguard Policy Subsidy
lnx0	−0.024 ***	−0.014
	(0.006)	(0.014)
age	−0.001 **	−0.003 **
	(0.001)	(0.002)
edu	−0.039 ***	−0.094 ***
	(0.007)	(0.020)
hea	−0.013 ***	−0.008
	(0.005)	(0.013)
size	−0.011 ***	−0.046 ***
	(0.003)	(0.009)
zd	−0.095 ***	−0.218 ***
	(0.021)	(0.061)
zr	−0.014	0.054
	(0.014)	(0.049)
zc	−0.026	−0.068
	(0.021)	(0.058)
dg1	−0.072 ***	−0.077 ***
	(0.005)	(0.017)
area	−0.032 ***	0.003
	(0.005)	(0.013)
net	−0.016 ***	−0.019
	(0.004)	(0.013)
_cons	1.038 ***	1.324 ***
	(0.052)	(0.148)
r2	0.206	0.358
r2_a	0.201	0.324

Note: *** p < 0.01; ** p < 0.05.

. It can be seen that the regression coefficient of agricultural subsidies on income inequality is −0.024, and is significant at the 1% level. The regression coefficient of safeguard policy subsidies on farmers’ income inequality is −0.014, and fails the significance test.

4.4.2. Regional Heterogeneity

The results in

Table 7. Regional heterogeneity results.

Variable	(1)	(2)	(3)	(4)
	West	Northeast	Central	East
lnx0	−0.0052 ***	−0.0048 ***	−0.003 **	−0.002
	(0.001)	(0.002)	(0.001)	(0.001)
age	−0.001	0.031 *	−0.004	0.005
	(0.005)	(0.016)	(0.009)	(0.011)
edu	−0.005	−0.004	−0.007	−0.015 ***
	(0.003)	(0.006)	(0.004)	(0.004)
hea	−0.009 **	0.011 *	0.010 **	−0.003
	(0.004)	(0.006)	(0.004)	(0.004)
size	−0.027 ***	−0.020 ***	−0.025 ***	−0.021 ***
	(0.004)	(0.007)	(0.004)	(0.003)
zd	−0.089 ***	−0.168 ***	−0.072 ***	−0.028 *
	(0.013)	(0.032)	(0.015)	(0.014)
zr	−0.002	−0.024	−0.013	−0.009
	(0.011)	(0.017)	(0.012)	(0.012)
zc	−0.007	−0.051 **	−0.007	0.007
	(0.012)	(0.021)	(0.011)	(0.011)
dg1	−0.044 ***	−0.034 ***	−0.050 ***	−0.033 ***
	(0.004)	(0.007)	(0.004)	(0.004)
net	−0.003	0.003	−0.000	−0.002
	(0.003)	(0.005)	(0.003)	(0.003)
_cons	0.837 ***	−0.517	0.763 ***	0.454
	(0.159)	(0.560)	(0.284)	(0.363)
Year dummy variable	Controlled
N	5046	1426	3544	3567
r2	0.120	0.144	0.170	0.097

Note: *** p < 0.01; ** p < 0.05; * p < 0.1.

show that there is regional variability in the moderating effect of government subsidies on income disparity. It has a significant improvement effect on the income distribution pattern of rural residents in the western and northeastern regions. The coefficient results show statistical significance within the 1% confidence interval. The moderating effect on the eastern region is relatively weak. Its absolute value of the coefficient is at the lowest level and does not pass the significance test.

4.4.3. Age Heterogeneity

Considering the influence of different life cycle groups on the structure of household income distribution, we grouped the farm households based on the age variable. The first is the young group (age 16–35), which accounts for 54.7% of the full sample. The second is the middle-aged and old-aged group (age 35 and above), accounting for 45.3% of the full sample. The two groups were subjected to panel data fixed effects regression, and the results are shown in

Table 8. Age heterogeneity results.

Variable	(1)	(2)
	Youth Group	Middle-Aged and Old-Aged Group
x0	−0.005 ***	−0.003 ***
	(0.001)	(0.001)
edu	0.001	−0.002
	(0.003)	(0.003)
hea	0.001	−0.003
	(0.003)	(0.003)
size	−0.021 ***	−0.028 ***
	(0.002)	(0.003)
zd	−0.077 ***	−0.067 ***
	(0.011)	(0.012)
zr	−0.002	−0.016 *
	(0.009)	(0.009)
zc	−0.011	0.005
	(0.008)	(0.010)
dg1	−0.039 ***	−0.049 ***
	(0.003)	(0.003)
area	−0.024 **	−0.037
	(0.012)	(0.026)
net	−0.005 *	0.002
	(0.002)	(0.003)
_cons	0.708 ***	0.800 ***
	(0.039)	(0.067)
Year dummy variable	Controlled
N	7337	6246
r2	0.136	0.114

Note: *** p < 0.01; ** p < 0.05; * p < 0.1.

. It can be seen that the regression coefficients of the subsidy are negative for both groups, and are significant at the 1% level. The absolute value of the regression coefficient for the youth group is greater than that for the middle-aged and elderly group, indicating that the income redistributive benefits of financial subsidies for young farm families are more obvious, and can more effectively reduce their internal income gap.

4.5. Endogeneity Analysis

We refer to the study of Yang et al. [38] and select the provincial average government subsidy amount as an instrumental variable for endogeneity analysis. The results are shown in

Table 9. Endogeneity test results.

	Benchmark Regression	Instrumental Variables (2SLS)
		First Stage	Second Stage
Government subsidy	−0.004 ***		−0.134 ***
Average provincial government subsidy		0.0001 ***
Durbin–Watson (score) test p-value		0.000
Wu–Hausman test p-value		0.000
Shea’s partial R-squared		0.000

Note: *** p < 0.01.

. Firstly, we tested whether the instrumental variable is a weak instrumental variable. The Robust F value is greater than 10, which rejects the original hypothesis of “the existence of weak instrumental variables”. The p-value of the Durbin–Watson test and the Wu–Hausman test is less than 0.01. The 2SLS estimation results show that the average provincial government subsidies and the individual government subsidies are positively significant at the 1% level, which shows that the instrumental variables are valid. The results of the two-stage regression show that the absolute value of the coefficient of government grants becomes larger compared to the baseline regression. Suggesting that the fiscal expenditure variables in the model do have endogeneity problems.

4.6. Robust Tests

Considering the endogeneity problem in the model, this paper draws on the studies of Shen et al. [71] and Niu et al. [72] by using the recentered response function regression (RIF) method proposed by Rios-Avila [73] to make the empirical results more robust. The results are shown in

Table 10. Recentered response function regression results.

Variable	(1)	(2)	(3)
	Q90–Q10 Quantile	Gini Coefficient	Variance
lnx0	−0.027 ***	−0.001 ***	−0.026 ***
	(0.008)	(0.000)	(0.007)
age	0.020 ***	0.000 ***	0.009 ***
	(0.003)	(0.000)	(0.002)
edu	−0.047 **	−0.000	0.008
	(0.021)	(0.000)	(0.013)
hea	−0.059 **	−0.001 **	−0.012
	(0.025)	(0.000)	(0.018)
size	−0.011	0.000	0.018
	(0.015)	(0.000)	(0.013)
zd	0.158	0.003 *	0.142 **
	(0.099)	(0.002)	(0.069)
zr	−0.137 **	−0.003 **	−0.060
	(0.069)	(0.001)	(0.057)
zc	−0.052	−0.000	−0.005
	(0.068)	(0.001)	(0.040)
dg1	−0.491 ***	−0.009 ***	−0.310 ***
	(0.024)	(0.000)	(0.020)
net	−0.034	−0.001 ***	−0.028 *
	(0.021)	(0.000)	(0.016)
area	−0.088 ***	−0.002 ***	−0.046 **
	(0.023)	(0.000)	(0.020)
r2	0.062	0.073	0.044
r2_a	0.061	0.073	0.043

Note: *** p < 0.01; ** p < 0.05; * p < 0.1.

. Columns (1), (2), and (3) show that the estimated coefficients of government grants in the regression of the recentered impact function based on the Q90–Q10 quantile distance, the Gini coefficient, and the variance composition are −0.027, −0.001, and −0.026, respectively. All are significantly negative at the 1% level. It indicates that the subsidy is effective in suppressing rural household income disparity and the results are robust.

In order to exclude the impact of the extreme values of the variables on the stability of the regression results, we further trimmed the net household income, y1, in the baseline regression sample for the top and bottom five percent of household income in the income rankings. The conclusions drawn remain basically consistent with the previous section. The results are shown in

Table 11. Benchmark regression model after excluding extremes.

Variable	(1)	(2)
	Net Household Income	Income Inequality Among Farm Households
lnx0	0.014 ***	−0.004 ***
	(5.83)	(−5.42)
age	0.008	0.000
	(0.67)	(0.07)
edu	0.022 ***	−0.007 ***
	(3.11)	(−3.39)
hea	0.004	−0.001
	(0.56)	(−0.25)
size	0.073 ***	−0.023 ***
	(10.28)	(−11.06)
zd	0.240 ***	−0.073 ***
	(7.93)	(−8.06)
zr	0.029	−0.010
	(1.31)	(−1.58)
zc	0.010	−0.006
	(0.50)	(−1.01)
dg1	0.148 ***	−0.043 ***
	(19.73)	(−19.17)
net	0.008	−0.002
	(1.37)	(−1.11)
area	0.111 ***	−0.031 ***
	(2.97)	(−2.81)
_cons	9.119 ***	−0.004 ***
	(22.64)	(−5.42)
Year dummy variable	Controlled
N	13,583	13,583
r2	0.228	0.117
F	159.1	79.98

Note: *** p < 0.01.

below.

5. Discussion and Conclusions

5.1. Discussion

The findings of this study reveal a subtle relationship between government subsidies and farm household income inequality. While subsidies have a significant positive effect on raising the overall level of rural income, their impact on mitigating income inequality seems to depend on the design and implementation mechanism of the subsidy policy [74]. For example, our study found that, for every 1% boost in subsidy, the difference in help for the two age groups differed by 0.2%. This coincides with the study of Guo and Yu [75], which emphasizes pinpointing the poor. Further, the concluding data showed that, for each year of continued government subsidy, farmers’ incomes increased by 1.9%, and the level of inequality decreased by 0.4%. Based on a 2024 per capita rural disposable income of RMB 23,100, this equates to an annual increase of RMB 438.9. The amount would cover 41 percent of the cost of basic health insurance (based on the full cost of RMB 1070). Although its effect is smaller than that of technical progress in agriculture (coefficient 0.076 [76]), it is more inclusive for vulnerable groups.

According to the results in Table 3, the negative effect of the two higher income groups reveals a higher degree of part-time employment. When subsidies induce factor inputs in agriculture, competition for labor hours leads to the crowding out of non-farm employment opportunities by agricultural operations. Overall, it instead results in a net loss of total household income. Notably, direct income subsidies (e.g., poverty alleviation grants) exhibit stronger redistributive effects, benefiting low-income households in particular by reducing absolute income disparities. However, project-based subsidies (e.g., agricultural technology adoption or infrastructure investment) tend to disproportionately benefit households with higher initial resource endowments, which can exacerbate relative inequality. Such differences are consistent with the theory of the Matthew effect, whereby pre-existing differences in human capital, social networks, and access to information may lead to different efficiencies in the utilization of subsidies.

We see three sources of divergence in farmers’ response to government subsidies. The first is the differentiation of credit response driven by resource endowment. For large-scale entities (e.g., cooperatives), it is easier to leverage credit resources through government subsidies, forming a virtuous cycle of subsidizing first, then mortgaging assets, and then lending and reinvesting. Similarly, large-scale crop trusteeship and mechanization have improved the value of land, making it more eligible for bank loan collateral [77]. The root cause of why subsidized funds are converted more into living consumption than production investment among small farmers is their lack of credit support mechanisms. This liquidity constraint limits small unit growers. Second is the amplification of differences in education levels. Highly educated farmers are more absorptive of technical extension services. As our analysis yielded, an increase of 1 unit of educational attainment reduces the inequality gap by 0.7%. The implication is that increasing intellectual capital can transform subsidies into a technological dividend. The third is the structural incentive bias in the design of the system. As the existing subsidy delivery mechanism is more suitable for large-scale business entities, the bias will encourage them to carry out reforms, such as land transfer. The different access to subsidy support among farmers results in different levels of response.

The empirical results further suggest that the mitigating effect of subsidies on inequality is affected by regional heterogeneity. In less developed regions, subsidies mainly play the role of a safety net, which can effectively reduce the incidence of poverty but have a limited impact on structural inequality. By contrast, in economically developed rural areas, subsidies are more likely to play the role of a springboard, widening income differentiation through market-oriented production incentives. This spatial variation highlights the importance of designing subsidy policies within local socio-economic ecosystems.

Some limitations also exist in our analysis. The shortcomings are confined to limited data, and in analyzing the types of government subsidies, only 2014 cross-sectional data are analyzed. In addition, after the implementation of the “one-card” reform of financial subsidies in China, farmers themselves do not have a clear conceptualization of government subsidies. In our case, we focus on the impact of a unified government subsidy on household income inequality, while the impact of individual government subsidies on income inequality remains to be explored. We believe that over-reliance on government subsidies without complementary institutional support (e.g., skills training, market access facilitation) may lead to policy inefficiencies. The next step of our study will focus on the long-term dynamic effects of subsidy-induced behavioral changes (e.g., labor allocation decisions, intergenerational human capital investments). Since more refined data are needed to accurately measure the ultimate effects of subsidies, we will locate the Northeast China region to distribute the questionnaire based on the existing foundation. Possible models to be used are double-difference methods and quasi-experimental designs to clarify the causal mechanisms between specific subsidy types and allocation outcomes.

5.2. Conclusions

The main conclusions can be summarized by the following three points.

First, government subsidies can notably increase the income of the population and effectively reduce the income gap, but the economic benefits vary among different income classes. Compared with high-income farmers, the marginal benefits of the subsidy policy are more significant for low-income groups, indicating that the income-raising mechanism is more effective for vulnerable sectors. Factors such as education level, family size, land transfer, number of migrant workers, and geographic location all significantly affect their income level and income disparity. Second, agricultural inputs, as intermediate variables, play an important role in the process of income distribution of subsidy effects. It changes the production and management decisions of households, increases the intensity of agricultural resource allocation, and leads to wealth accumulation effects that improve the internal distribution of rural capital. In terms of the transmission path, the direct increase in transfer income and the incentive to increase agricultural investment together optimize the pattern of rural income allocation. Finally, heterogeneity exists in the impact of subsidies on income disparities. Productive assistance targeting agriculture is more conducive to distributive justice than policy guarantees. Grants have the most significant income equalizing effect in Northeast China, and are relatively weaker in the East; the level of utility is stronger than that of the middle-aged and older groups because it more closely bridges the youth entrepreneurial capital gap.