How Farm Machinery Rental Services and Off-Farm Work Affect Household Income in China

Abstract

(1) Background: Most nations struggle to close significant income gaps between high and low earners. While the adoption of farm machinery rental services and off-farm employment may be beneficial, it is unclear whether jointly applying both approaches can raise income levels in rural households or help narrow the income gap within the farm sector. This study investigated scenarios involving both participation in farm machinery rental markets and in off-farm work, analyzing their varied impacts on household incomes based on survey data from 1027 rice producers in rural China. (2) Methods: We employed a two-stage econometric procedure encompassing a bivariate ordered probit model with an endogeneity-corrected unconditional quantile regression model. (3) Results: Rice farmers often simultaneously rent farm machinery services and engage in off-farm work. Both activities positively affect their household incomes; however, these effects vary across different income levels. Renting farm machinery provides greater marginal benefits for lower-income households, while off-farm employment has a stronger impact on higher-income households. Farm machinery rental services appear to benefit disadvantaged households more than off-farm employment opportunities do. (4) Suggestions: To enhance the welfare of lower-income households, policymakers should focus on expanding access to farm machinery rental services.

1. Introduction

Narrowing the income gap has become an essential topic attracting attention and extensive discussion worldwide. The farm–non-farm income gap has been discussed by many researchers, and an agreement has been reached that the gap has progressively declined over time [1,2,3]. However, heterogeneity within the farm sector is an important facet of the farm income problem [2]. Enhancing the welfare of rural residents is a long-standing priority for Chinese policymakers, but rural households continue to be disadvantaged by the persistent income gap between urban and rural areas [4]. Especially in rural areas, issues with uneven distribution require serious consideration. Many studies have sought effective solutions.

Some scholars have proposed developing off-farm labor markets, thereby providing rural households with earning opportunities beyond agricultural work [5,6,7,8,9]. In fact, the number of rural migrant workers in China has increased progressively, reaching 285.6 million in 2020, having surged from 15% of the rural labor force in the early 1980s to 39.59%. This shift has created higher wage incomes for farm households but has also caused a shortage of rural labor input in rice production. There is an urgent need to substitute labor with machinery input in rice production to mitigate these changes. From 1995 to 2020, the labor input decreased from 285 man-days/ha to 75 man-days/ha, while the farm machinery expenditure increased from 189.45 CNY/ha to 3008.1 CNY/ha in rice production [10].

Since the beginning of China’s reform and opening up in 1978, the household contract responsibility system has rapidly promoted agricultural production. The system for separating the ownership rights, contract rights, and management rights of rural land has given individual rural households relative autonomy over land use decisions and crop selection. Although this arrangement has stimulated agricultural vitality, land is highly fragmented because of equitable distribution. For example, fragmentation has reached more than 10 plots per household in Sichuan Province, and the average size is only 0.53 ha. As a result, small farmers have little incentive to purchase agricultural machinery. Also, household expenses are hardly covered by incomes from crop planting. So, farmers tend to do part-time work to optimize their income structure. Renting agricultural machinery services to keep farming while taking non-agricultural employment has become a preferred strategy of many Chinese farm households.

China’s farm mechanization has developed towards marketization since the mid-1990s. This period also saw the initiation of farm machinery rental service (FMRS) integration. Market services for plowing, sowing, and rice harvesting have since expanded beyond individual counties or provinces [11]. Since 2004, the Chinese government has launched incentive policies and subsidies to support agricultural mechanization, including subsidizing farm machinery purchases, abolishing the highway tolls for combined harvesters, and providing cross-regional harvest calendars of the service demands in various provinces [12]. Rice production has changed in several key ways after this flourishing of farm equipment rental markets. The number of specialized farm machinery cooperatives providing FMRSs has maintained steady growth, from 8622 in 2008 to 75,449 in 2020 [13]. By 2020, the proportion of the mechanized harvesting area and the mechanized tillage area to the total rice sowing area reached 93.7% and 97.8%, respectively, while the proportion of machine-planted rice area to the total rice sowing area increased from 13.7% in 2008 to 56.3% in 2020 [13].

The effects of shifting input patterns in China’s agricultural sectors have been extensively researched. A common topic is how the income of farm households would be affected. Positive effects of off-farm work include mitigating rural poverty and increasing farm households’ incomes [5,6,7,8,9]. The effects of FMRSs on family income have also been extensively investigated [14,15,16]. However, although the decision-making processes of working off-farm and utilizing FMRSs are concurrent, their joint effects on economic performance have been largely overlooked [17,18]. Little attention has been paid to farmers’ simultaneous decision to farm and conduct off-farm work when examining the association between farm machinery use and farm household incomes [12,19,20,21], apart from Ji et al. [11] and Ma et al. [22]. However, Ji et al. [11] did not examine these dual decisions’ effects on financial performance in farm households. Ma et al. [22] investigated various factors in maize yields and agrochemical expenses without focusing on FMRS use. More importantly, it was assumed that farm households would benefit equally from farm mechanization regardless of existing income levels—an assumption that merits reconsideration.

The primary objective of this research was to investigate the joint effects of participating in farm machinery rental markets and working off-farm, as well as the heterogeneous impacts of such decisions on household incomes. Survey data from 1027 rice producers in rural China were used. This work attempted to contribute to the literature from three aspects. Firstly, this study considered farm machinery usage through hired services, in contrast to the self-owned machines considered in previous studies. It used a relatively new arrangement of clusters of service providers offering pan-geographic and pan-seasonal outsourcing services [23], as has been the case in China over the past decade. Also, this study focused on the frequency of hiring farm machinery services, as previous studies have mainly focused on dichotomous decisions of FMRS use. Secondly, the heterogeneous effects of FMRS and off-farm work on household income are estimated. Uneven impacts of farm mechanization, stemming from heterogeneous characteristics of rural households, may exacerbate income inequality [10]. It is essential to consider these heterogeneous effects when making policies to increase farmers’ income while promoting farm mechanization and boosting crop yield. Without this knowledge, policymakers may design farm mechanization promotion policies that lead to severe income inequality. Thirdly, relationships among FMRS, income variability, and income distribution are examined in this study. Several studies have explored income variability and inequality [24,25] without referring to the role of farm mechanization. Zhou [18] probed the heterogeneous farm mechanization effects on farm performance but focused on yield variability and inequality in crop production.

The remainder of this paper is organized as follows. Section 2 presents the theoretical analysis. Section 3 describes the data collection and methods used. Results are presented in Section 4, and Section 5 provides further discussion of the results. Section 6 concludes the findings and provides policy implications.

2. Theoretical Analysis

Utilizing FMRSs and working off-farm have mutual effects on farm households. The relative benefits of FMRSs are well-documented—these services relieve smallholders of the burdens of purchasing, owning, and maintaining farm machinery [26,27,28], while increasing access to necessary equipment [20], making them popular in many developing countries [19]. Mechanization can improve farm performance, increasing both crop yields [22,29] and households’ on-farm income [30,31]. Machinery, like tractors, can substitute for labor input in agricultural production [32,33], enabling households to allocate more time to non-farm activities and earn off-farming incomes [15,34]. FMRSs allow farm operators to continue production despite dwindling labor supply and contribute to China’s agricultural transformation [19,29,35], particularly amid the feminization and aging of the rural labor force caused by massive rural-to-urban migration [36]. Thus, it is necessary to examine the effects of off-farm work and FMRS jointly.

Furthermore, there may be heterogeneous effects of using FMRSs and engaging in off-farm work on household incomes. Previous studies have tended to assume that rural workers’ decisions to work off-farm have similar effects on their incomes based on rational self-selection. That is, farmers with higher expected returns in agriculture choose to remain in the farm sector, while those with higher expected returns in off-farm sectors opt to work off-farm [5]. In reality, disadvantaged households may benefit disproportionately from off-farm employment. The effects might also be unequal because wealthier households typically have greater capacity to secure higher-return employment opportunities in the off-farm sector compared to their lower-income counterparts.

Households of varying income levels may also receive different returns from the same usage of farm machinery. FMRSs enable disadvantaged households to allocate more time to off-farm activities, where they may earn better incomes. This increased income can become savings or investments in physical and human capital, in which case, ideally, FMRSs would help to ameliorate income inequalities. However, households cannot benefit equally from FMRSs due to inherent variations in their access to off-farm work. Therefore, further investigation is needed to clarify the heterogeneous effects of FRMS usage on different income-level groups.

Generally, off-farm work and FMRS usage are interdependent and cannot be analyzed separately. Moreover, these two activities have heterogeneous impacts on rural household incomes based on the prevailing wealth distribution. The present study contributes to the literature on FRMS use, off-farm work, and rural household income inequality by examining these effects comprehensively. Understanding these dynamics may provide valuable insights for policymakers seeking to reduce income inequality.

3. Date Collection and Model Specification

3.1. Data Collection and Sample Description

The data supporting this analysis were sourced from a survey of rice farmers in China’s Yangtze River basin, conducted in August 2017. The survey covered various geographical areas and diverse households, providing a robust foundation for this research. Our objectives center on exploring fundamental dynamics among FRMS use, off-farm work, and rural household income inequality, as well as the factors influencing them. It is our understanding that these patterns and factors do not substantially change within short periods of time, especially in China’s agricultural sector. Between 2017 and 2024, there were no significant shifts in relevant agricultural policies or the overall structure of FMRSs and off-farm work opportunities in China.

A total of 12 counties across six provinces were selected for the survey. Sichuan Province represents the upstream region of the Yangtze River within this sample; Hubei, Hunan, and Jiangxi provinces represent the midstream region; and Anhui and Jiangsu provinces represent the downstream region. Provinces are representative of rice production conditions and farm machinery usage in the Yangtze River basin, accounting for 54.7% of the country’s total rice sown area and 55.5% of its rice production.

A multistage stratified sampling procedure was used to select the sampled farms. In the first stage, the six provinces were purposely chosen for their intensive rice production systems. In the second stage, random sampling was used to select two counties from each province, three towns per county, three villages per town, and ten farm households per village. This resulted in data from 1080 farm households across 108 villages in six provinces in China. There were 53 samples excluded due to incomplete information. Face-to-face interviews were conducted with the primary decision-maker in the household by well-trained interviewers.

The sample distribution is summarized in

Table 1. Sample distribution of FMRS usage and off-farm work.

FMRS		Number	Percentage
0	Did not use FMRS	313	30.5%
1	Use FMRS in one production stage	122	11.9%
2	Use FMRS in two production stages	291	28.3%
3	Use FMRS in three production stages	175	17.0%
4	Use FMRS in four production stages	126	12.3%
Stage of rice production a
	Land plowing	628	61.1%
	Rice transplanting	226	22.0%
	Fertilizer and pesticide application	170	16.6%
	Harvesting	709	69.0%
Off-farm work b
0	No off-farm work	260	25.3%
1	Works off-farm less than 1/3 of full time	253	24.6%
2	Works off-farm between 1/3 and 2/3 of full time	413	40.2%
3	Works off-farm more than 2/3 of full time	101	9.8%

^a: 1 = if FMRS is used in this stage of rice production; 0 = otherwise. ^b: Share of days worked off the farm to total days worked.

. Among sampled farm households, 25.6% did not use FMRSs in rice production. Most (28.3%) used FMRSs in two production stages, and some (17.0%) in three production stages. Only 12.3% of households used FMRSs in all four stages of rice production, namely, land plowing, rice transplanting, fertilizer and pesticide application, and harvesting.

Off-farm work is denoted by the ratio of off-farm man-days to total man-days (both off- and on-farm), following previous studies [24,37]. Consistent with other research [38,39], household income per capita was applied to compare income levels across households. A household’s income is the sum of farm earnings, off-farm earnings, and income received from other sources such as transfers and rents.

Table 2. Definition and descriptive statistics.

Variables	Definition	Mean (SD)
Dependent variables
FMRS	Decision of farm households to hire FMRS (0–4)	1.69 (1.12)
Off-farm work	Decision of farm households to work off-farm (0–3)	1.35 (0.96)
Household income	Per capita household income (1000 Yuan/year)	8.99 (5.31)
Independent variables
Age	Age of household head (years)	56.75 (10.23)
Education	Education of household head (years)	6.05 (3.32)
Gender	Gender of household head (1 = male, 0 = female)	0.86 (0.41)
Average age	Average age of labors within the household (years)	48.55 (10.62)
Highest education	Years of a household member taking highest level of education (years)	9.26 (3.59)
Training	1 = if the household receiving training in farming techniques, 0 = otherwise	0.65 (0.47)
Household size	Number of household members	4.78 (1.92)
Share of labors	Proportion of labors in household	0.69 (0.22)
Share of male labors	Proportion of male labors to total household labor	0.35 (0.18)
Cultivated land area	Cultivated land areas (Mu)	12.07 (7.11)
Agricultural assets	Total present value of self-owned agricultural machinery (1000 Yuan)	5.96 (40.88)
Village cadre	Number of village cadre within the household	0.13 (0.38)
Credit access	1 = if the farm household has access to financial credit, 0 = otherwise	0.14 (0.34)
Road	1 = if there is a road for tractors leading to farmlands, 0 = otherwise	0.59 (0.49)
Terrain	1 = if the farms located in the plain area, 0 = otherwise	0.39 (0.48)
Price	Average price of the machinery service at the village level (yuan/mu)	105.65 (17.42)
Distance	Distance between the surveyed village and its affiliated county center (in km)	23.15 (10.75)
Number of farm machines	Number of farm machines can provide production service (village level)	45.14 (25.23)
Ratio of off-farm workers	Share of farmers working off-farm (village level)	0.29 (0.12)

Notes: 1 mu = 1/15 hectare. Yuan is the Chinese currency; 1 USD = 6.7518 Yuan in 2017.

presents the definitions and summary statistics of the variables. The mean of FMRS is 1.69, indicating that most of the sampled farm households used FMRSs in at least one production stage. Most farm households described engaging in off-farm work less than 1/3 of full-time. Other characteristics were controlled with reference to previous studies [35,36,37,38,39].

3.2. Joint Decision Model of FMRS Use and Off-Farm Employment

Decisions to use FMRSs and engage in off-farm work are probably affected by unobserved factors, such as the household’s motives and innate abilities, which are also likely to influence household income. This may have created an endogeneity problem. To correct for this, we followed the two-stage estimation procedure used in the conventional instrument variable approach [24,40]. In the first stage, a bivariate-ordered probit model was estimated to investigate the relationship between FMRS usage and off-farm work. In the second stage, a quantile regression model was applied to explore how these two decisions affect farm households’ incomes heterogeneously. After considering endogeneity issues, the predicted values of FMRS and off-farm work were calculated based on the first stage estimation. They were then used to replace the original values of the two variables in the second stage.

The bivariate-ordered probit model comprises two univariate-ordered probit equations, one for FMRS usage and a second for off-farm work. Let $y_{1i}$ and $y_{2i}$ be the observed discrete variables of these two choices, where J1 and J2 are categories that each discrete variable fall into (in this case, J1 = 4 and J2 = 3). As per Sajaia and Chang and Mishra [24,41], the bivariate-ordered probit model was formulated as follows:

$$\begin{gathered} \begin{array}{l}{y}_{1i}^{*}=X_i{\beta }_1+Z_{1i}{\gamma }_1+{\epsilon }_{1i}\\ y_{2i}^{*}=X_i{\beta }_2+Z_{2i}{\gamma }_2+{\epsilon }_{2i}\end{array} \\ {y}_{1i}=\left\{\begin{array}{l}1\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{y}_{1i}^{*}\le U_{11}\\ 2\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{U}_{11}<y_{1i}^{*}\le U_{12}\\ 3\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{U}_{12}<y_{1i}^{*}\le U_{13}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\\ 4\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{U}_{13}<y_{1i}^{*}\hspace{1em}\end{array}\right. \mathrm{and} y_{2i}=\left\{\begin{array}{l}1\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{y}_{2i}^{*}\le U_{21}\\ 2\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{U}_{21}<y_{2i}^{*}\le U_{22}\\ 3\hspace{1em}\hspace{1em}\hspace{1em}if\hspace{1em}{U}_{22}<y_{2i}^{*}\hspace{1em}\end{array}\right. \end{gathered}$$

(1)

In Equation (1), $y_{1i}^{*}$ and $y_{2i}^{*}$ are unobserved latent variables of FMRS usage and off-farm work, respectively, for household i. The kernel variable, FMRS usage, was measured as a count variable. Each respondent was asked to identify every stage of rice production when FMRS was used, including land plowing, rice transplanting, fertilizer and pesticide application, and harvesting stage. Each stage was coded as a binary variable (1 = yes; otherwise = 0) and the sum of the production stages was used to represent the intensity of FMRS usage for a given farm household, following previous studies [18,22]. X_i is a vector of exogenous factors influencing both decisions. The vectors Z_1i and Z_2i are the instrument variables that directly influence the use of FMRSs and engagement in off-farm work indirectly influencing household income. The corresponding parameters to be estimated for the two equations are β₁, γ₁, β₂ and γ₂. U_si denotes unknown parameters of the threshold points to be estimated. We assume that the error terms ${\epsilon }_{1i}$ and ${\epsilon }_{2i}$ are standard and normally distributed across individuals. The correlation coefficient of the error terms and the consistent estimators (β₁, γ₁, β₂, γ₂, ρ) are estimated using maximum likelihood with the log-likelihood function specified as follows [42]:

$$LnL={\displaystyle \sum _{i=1}^N{\displaystyle \sum _{j_1=1}^{J_1}{\displaystyle \sum _{j_2=1}^{J_2}I(F_{1i}=j_1,F_{2i}=j_2)\times \ln \mathrm{Pr}(F_{1i}=j_1,F_{2i}=j_2)}}}$$

(2)

where I(·) is a binary indicator specifying one of the J₁ × J₂ categories into which each farm household may be classified. The null hypothesis that there is no correlation between the two error terms (i.e., ρ = 0) was tested using the Wald test to determine whether the decisions to use FMRSs and to work off-farm are independent.

3.3. Heterogeneous Effects of FMRS Use and Off-Farm Work on Income

At the second stage, both conditional quantile regression (CQR) and the unconditional quantile regression (UQR) are potential models that can be used to estimate the varying effect of FMRS usage and off-farm work on household income. However, CQR can only be used to estimate a covariate’s effect on a specific quantile conditioned on the mean value of other independent variables [5,42]. With UQR, the effect is evaluated marginally over the distribution of the independent variables and does not change with the set of covariates; therefore, it can be used to measure the full impact of FMRSs and off-farm work on household incomes and changes in household incomes at the quantiles.

Following Mishra et al. and Khanal et al. [5,43], the UQR approach in this study uses the re-centered influence function (RIF) to measure how changes in the underlying distribution of household income influence the distributional statistics such as the median, different quantiles, variance, and the Gini coefficient of household income. The marginal effect for a given distribution statistics of interest can be obtained by averaging the RIF regression function in the case of the change in the distribution of the covariates [43]. The linear RIF regression function is specified as:

$$E\left[RIF\right(v_i:q_{\tau }|X,Y_1,Y_2)]=X_i\alpha +{\stackrel{̑}{Y}}_{1i}{\zeta }_1+{\stackrel{̑}{Y}}_{2i}{\zeta }_2+u_i$$

(3)

where v_i represents the income for household i; α, ${\zeta }_1$, and ${\zeta }_2$ are parameters that define the marginal effect of FMRS and off-farm work on the relevant distributional statistic, respectively. The quantile is denoted as q_τ. Xi is the exogenous factor from Equation (1). ${\widehat{Y}}_{1\mathrm{i}}$ and ${\widehat{Y}}_{2i}$ are the predicted values for FMRS and off-farm work from the first stage, and u_i is the stochastic error term. Firpo et al. [44] noted that parameter estimates from the RIF regression have a similar interpretation as those from the OLS regression. The parameters measure the unconditional quantile marginal effects associated with a slight change in each covariate on household income at the t^th quantile. According to Frölich and Melly [44], the RIF estimator is $\sqrt{n}$ consistent, asymptotically normal, and efficient.

3.4. Income Variance and Gini Coefficient

In addition to evaluating how decisions to use FMRSs and participate in off-farm work affect household income, we also estimated the influence of these decisions on income variance and the distribution of income variance measured by the Gini coefficient. In line with previous studies [8,18,42], the distributional variance of household income can be computed as:

$$Variance={\sigma }_V^2=\int (v-{\mu }_V{)}^{2}f(v)dv$$

(4)

where $V=[v_1,v_2,\dots ,v_n]$ represents household income, ${\mu }_V$ is the average household income, and $f(v)$ is a probability density function of income.

The Gini coefficient is specified as:

$$Gin{i}_V=1-{\displaystyle \frac{2}{{\mu }_V}}{R}_V$$

(5)

where $R_V={\displaystyle \underset{0}{\overset{1}{\int }}G{L}_V(p)dp}$ with $p=F_V(v)$ and $G{L}_V(p)={\displaystyle \underset{-\infty }{\overset{F_V^{-1}(p)}{\int }}v}d{F}_V(v)$ which is the generalized Lorenz ordinate of $F_V$ representing the cumulative distribution function.

Concerning the variables in Equation (1), the identification strategy depends on selecting relevant variables associated with FMRS use and off-farm work decisions without directly affecting household income. The number of farm machines available for hire at the village level are used as the instrumental variable for FMRS. It is rational because the more farm machinery that is available, the easier it is for farmers to access FMRS. However, the number of farm machines in the village is not expected to affect household income directly. For the off-farm work decision model, the share of farmers working off-farm at the village level (i.e., the ratio of off-farm to total workers) is used as an instrument. Empirical studies have shown that village out-migrating and networking can influence farm households’ decisions to engage off-farm work but not directly affect their economic performance [45,46].

To test the validity of our instruments, two-ordered probit models (one for FMRS usage and a second for off-farm work) and one OLS regression model for income were estimated with instrumental variables included. The results are displayed in

Table A1. Parameter estimates: test on the validity of the selection instruments.

Variables	FMRS	Off-Farm Work	Household Income
Number of machines	0.116 *** (0.021)		0.012 (0.010)
Ratio of off-farm workers		0.188 *** (0.036)	0.579 (0.483)
Control Variables	yes	yes	yes
Province	yes	yes	yes
Constant			3.756 *** (0.026)
cut1	−0.937 * (0.497)	−3.920 *** (0.526)
cut2	−0.365 (0.496)	−2.836 *** (0.523)
cut3	1.413 *** (0.499)	−1.026 ** (0.518)
cut4	2.483 *** (0.503)
Observations	1027	1027	1027

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. Two ordered probit models estimated for FMRS and off-farm work.

and show that the number of farm machines exerts a significant and positive impact on FMRS usage. Similarly, the coefficient of the ratio of off-farm to total workers in the village is positive and statistically significant at the 1% level. However, as the OLS model shows, neither instrument significantly affects household income. These results indicate that the instruments used in Equation (1) are valid.

4. Results

4.1. Joint Estimation of FMRS Use and Off-Farm Work Decision

Table 3. Joint estimation of farm machinery service use and off-farm work.

Variable	FMRS		Off-Farm Work
	Coefficient	SE	Coefficient	SE
Age	−0.001	0.005	−0.046 ***	0.007
Education	−0.013	0.015	−0.015	0.013
Gender	−0.090 **	0.038	0.312 ***	0.093
Average age	0.009 *	0.005	−0.086 ***	0.007
Highest education	0.014	0.013	0.038 ***	0.014
Training	0.424 ***	0.092	−0.061	0.088
Household size	0.011	0.027	0.079 ***	0.028
Share of labors	−0.071 ***	0.025	0.394 ***	0.103
Share of male labors	−0.062	0.077	0.206 ***	0.016
Cultivated land area	0.016 **	0.008	−0.018 *	0.009
Agricultural assets	−0.002 **	0.001	−0.008 ***	0.002
Village cadre	0.004	0.090	0.106 **	0.049
Credit access	0.080	0.116	−0.161	0.116
Road	0.262 ***	0.087	0.540 ***	0.086
Terrain	0.360 **	0.143	0.392 ***	0.151
Price	−0.017 ***	0.003	0.012	0.009
Distance	0.008 **	0.004	−0.004	0.004
Number of machines	0.113 ***	0.022
Ratio of off-farm workers			0.197 ***	0.035
Province dummy variable	Yes	yes	Yes	yes
Cut points
cut1	−0.935 *	0.536	−3.915 ***	0.544
cut2	−0.358	0.538	−2.831 ***	0.538
cut3	1.417 **	0.543	−1.027 *	0.540
cut4	2.478 ***	0.555
RHO	0.131 ***	0.042
Specification test
Test for RHO a	9.73 ***
Log pseudolikelihood	−2071.203
Wald chi2(23)	425.39 ***

*** p < 0.01, ** p < 0.05, * p < 0.1. ^a LR test was used. Ho: RHO = 0. The critical value is ×2 (0.95, 1) = 3.84.

presents the estimates of the bivariate ordered probit model for households’ decisions to adopt FMRSs and participate in off-farm work. The estimated correlation coefficient (RHO) is significant at the 1% level, which justifies the joint estimation of these two equations in improving the statistical efficiency of the parameter estimates. Results show that farm households’ joint decisions to use FMRS in rice production and off-farm work appear to be mostly dependent on their labor and land characteristics. These characteristics are captured by variables like average age, share of labor among household members, endowment of agricultural assets, and prevailing land conditions of the area.

Human capital characteristics of rice producers play an important role in both the decision to work off-farm and to employ FRMSs. Results are consistent with observed migration trends where educated young and male farm laborers migrate to non-agricultural and urban areas [45,47]. However, the household labor shortage is being mitigated by FMRS usage; we also found that households headed by women or with more elderly members, who are less likely to engage in farm labor, are more likely to use FMRSs in rice production. Previous studies have made similar observations [15,48].

Households’ land endowment is another key factor. The cultivated land area has a positive and statistically significant association with the use of FMRS. This can be explained by the inverse relationship between farm size and the average cost of using mechanized services. The result is consistent with observations by Lai et al. [49], who reported a positive association between cultivated land area and machinery used in China. On the other hand, the cultivated land area negatively and significantly impacts off-farm work, implying that households with larger farms also need to spend more time managing their own resources than seeking work elsewhere. Beyond the land scale, the condition of arable land is important in terms of household decisions to use FMRSs or seek off-farm work. Both the road conditions and terrain characteristics of farmlands are meaningful in this way. For example, if there is a road leading to their land that is suitable for tractors, farmers are more likely to use FMRSs. The same is true if the household’s land is in plain regions, as is the case for the areas we sampled (arable land for rice farming in the Yangtze River basin is also called the Jianghan Plains). The conditions there are favorable and convenient for mechanized farming, which further explains the positive and significant relationship between land conditions and off-farm work. Better land conditions make farm machinery use more convenient and popular, which also saves time and labor to pursue off-farm work opportunities.

4.2. Heterogeneous Impacts of FMRS and Off-Farm Work on Household Income

The effects of FMRS usage and off-farm work on household income are detailed in

Table 4. Unconditional quantile regression estimates for income: Estimations corrected for endogeneity issues.

Variables	OLS	Quantiles
		0.10	0.25	0.50	0.75	0.90
FMRS a	0.504 ***	0.561 ***	0.523 ***	0.280 ***	0.191 ***	0.102 ***
	(0.186)	(0.085)	(0.077)	(0.081)	(0.056)	(0.021)
Off-farm work b	2.030 ***	1.070	1.527 ***	2.668 ***	3.289 ***	3.738 ***
	(0.068)	(0.992)	(0.430)	(0.213)	(0.163)	(0.122)
Contral Variables	yes	yes	yes	yes	yes	yes
Province	yes	yes	yes	yes	yes	yes
Constant	3.747 ***	3.883 ***	3.368 ***	6.541 ***	1.308 *	1.719 **
	(0.029)	(0.118)	(0.743)	(1.088)	(0.789)	(0.845)

^a. 1 = if FMRS is used in this stage of rice production; 0 = otherwise. ^b. Share of days worked off the farm to total days worked”. Note: *** p < 0.01, ** p < 0.05, * p < 0.1; Standard errors in parentheses are calculated on the bootstrap method with 2000 replications. Household income is measured in 1000 Yuan per capita.

. Estimation results at the 10th, 25th, 50th, 75th, and 90th quantiles are reported for interpretation. Results from OLS and UQR methods do not show significant differences. As shown in Figure 1, we also plotted the estimation results to visualize the estimated coefficients on the 5th to 95th quantiles. The solid line traces the coefficient estimates, and the dashed line represents the upper and lower bounds of the 95% confidence interval.

The results (Figure 1a) indicate that the effect of FMRS usage is positive and significant across the entire household income distribution, and access to FMRSs in rice production is associated with increased income across farm households. These findings are consistent with Adu-Baffour’s results on Zambia [31], where farmers who receive tractor services tend to earn higher incomes through farming. However, the positive impact in our results is more pronounced at lower quantiles of household income. For instance, the effects are 0.561 and 0.102 at the 10th and 90th quantiles (Table 4), indicating that the marginal effect of using FMRSs on per capita income is much stronger for low-income households. This disparity may be because low-income households are more dependent on agriculture, so the same income increase constitutes a larger share of their total household earnings. Using FMRSs can increase on-farm income while allowing household laborers, who would otherwise work at home, to pursue jobs elsewhere.

In Figure 1b, the intensity of off-farm work is significant across the entire income distribution except for the 5th and 10th quantiles. Coefficients of off-farm work are more pronounced at higher quantiles, which indicates that the marginal effect of off-farm work on household income per capita becomes stronger from the lower tail to the upper tail of the income distribution. For instance, the estimates in the 75th and 90th quantiles are approximately three times higher than those in the 10th quantile (Table 4). While many previous studies have verified the contribution of off-farm work to increasing households’ incomes [18,23,50], few have observed this widening of the income gap. Previous studies have, however, found that income inequality might be primarily generated by the amount of time allocated to off-farm labor [51]. It is possible that this result is due to lower-income rice producers in the sampled area having less access to off-farm labor markets.

To check the robustness, we also applied the generalized quantile regression (GQR), which can estimate unconditional quantile treatment effects given one or more treatment variables (for a detailed example, see Powell, 2020 [52]). The results are presented in

Table A2. Generalized quantile regression estimates for household income: Estimations corrected for endogeneity issues.

Variables	0.10	0.25	0.50	0.75	0.90
FMRS	0.299 ***	0.201 ***	0.148 ***	0.147 ***	0.089 ***
	(0.002)	(0.019)	(0.023)	(0.026)	(0.019)
Off-farm work	1.215	1.404 ***	1.874 ***	2.289 ***	2.830 ***
	(0.903)	(0.023)	(0.026)	(0.041)	(0.018)
	(0.001)	(0.001)	(0.002)	(0.006)	(0.006)
Control Variables	yes	yes	yes	yes	yes
Province	yes	yes	yes	yes	yes
Constant	4.531 ***	6.031 ***	6.926 ***	1.651 *	1.127 **
	(0.068)	(0.462)	(1.154)	(0.997)	(0.528)

Note: *** p < 0.01, ** p < 0.05, * p < 0.1. Instrument variables for FMRS and off-farm work are the number of machines available and the ratio of off-farm workers to total workers at village level, respectively. Total draws: 10,000; Burn-in draws: 3000.

, which are in line with results from the UQR model estimation.

4.3. Impacts on the Distributional Variance and Equality of Household Income

The effects of FMRS use and off-farm work on the distributional variance and equality of income were further investigated. Income levels were estimated using the distributional variance and Gini coefficient of income as dependent variables. The results are shown in

Table 5. Effect of farm mechanization and off-farm work on household income variance and Gini coefficient.

Variables	Variance	Variance	Gini Coefficient	Gini Coefficient
FMRS a	35.150 * (18.526)	37.925 ** (17.947)	−0.045 * (0.026)	−0.044 ** (0.021)
Off-farm work a	27.844 (44.462)	28.141 (42.213)	0.245 ** (0.103)	0.209 ** (0.106)
FMRS × Off-farm work		1.279 (1.208)		0.011 ** (0.005)
Control Variables	yes	yes	yes	yes
Province	yes	yes	yes	yes
Constant	−7.107 (35.891)	−8.611 (35.430)	0.267 (0.164)	0.268 * (0.159)
Adjusted R2	0.102	0.078	0.208	0.165

Note: ** p < 0.05, * p < 0.1. Standard errors in parentheses are calculated on the bootstrap method with 2000 replications. ^a Predicted value.

.

The variance of the income distribution is high among households using FMRSs relative to those who do not. This could be due to the uncertainty in the supply of farm mechanization services in our sampled area. During peak production seasons, the timely availability of farm machinery is not guaranteed. In contrast, households with their own farm machinery effectively have assured availability.

The estimated effect of off-farm work on the Gini coefficient is positive and significant, suggesting that participating in off-farm work increases income inequality among rice producers in the Yangtze River Basin. In contrast, promoting FMRSs reduces income inequality, as indicated by a negative estimate of FMRSs on the Gini coefficient. This finding can be explained by our earlier results, where the marginal effect of FMRS usage on household income is higher for disadvantaged households, while the marginal impact of off-farm work on income is greater for families in the upper tail of the income distribution. In other words, lower-income households employing FMRSs may help reduce income disparity within the agricultural sector. This may also help to offset the adverse impact of off-farm employment on the income gap. These findings align with previous studies suggesting that FMRS usage helps reduce income inequality [18].

To better understand the income distribution, we plotted the Lorenz curve (Figure 2) to visualize the degree of inequality among rural households. This curve and the Gini coefficient (0.429) imply that the income distribution for the sampled rural households is relatively unequal.

Furthermore, an interactive effect of FMRS and off-farm work was investigated to reveal whether households who do both of those two activities are better off or not. The results indicate that the decision to use FMRSs and to work off-farm simultaneously increases income inequality significantly. This could be due to the positive effects of off-farm work being more pronounced than the adverse effects of FMRS usage on income inequality.

5. Further Discussion

Using survey data from 1027 farm households in rural China, this study analyzed the heterogeneous combined effects of utilizing FMRSs and participating in off-farm work on household per capita income. We employed a bivariate ordered probit model and an endogeneity-corrected unconditional quantile regression model to account for the possible joint decisions of FMRS usage and off-farm work engagement. We focused on the impacts of those two activities on the distributional variance of income and Gini coefficients.

The results of this study have both similarities and differences with previous research. First, in this paper, some factors are found to influence farmers’ joint decisions to use FMRS and to work outside their farms, which diverges from previous studies that focused on FMRS and on-farm employment separately. Factors like human capital (i.e., age, gender, education) [45,46,47,48] and land endowments [48,49] can significantly influence farmers’ decisions to pursue off-farm work and to use FMRSs, respectively. Second, heterogeneous effects of FMRS and off-farm work on household incomes were estimated in this research. Other scholars have found that uneven impacts of farm mechanization, stemming from the varied characteristics of rural households, can exacerbate income inequalities [10]. Third, relationships among FMRS, income variability, and income distribution were detected in this study. The results are consistent with previous studies [24,25]. For example, Song et al. [10] explored the heterogeneity in factors affecting rural household income in different quantiles, finding that agricultural mechanization services increased rural household incomes and helped to narrow the income gap among higher- and lower-income rural households.

6. Conclusions, Implications and Limitations

6.1. Conclusions

The conclusions of this study are as follows. First, farm households’ decisions to use FMRSs are significantly correlated with their choice to work off-farm due to unobserved factors captured in the estimated error terms. These joint decisions are largely dependent on specific labor and land characteristics, as assumed, through unobserved effects captured in the estimated error terms. Farm households’ joint decisions about FMRS use and off-farm work are mostly dependent on their labor and land characteristics.

Second, both FMRS usage and off-farm employment are associated with higher per-capital household incomes; however, these effects vary across existing income levels. The marginal benefit of hiring farm machinery is greater for disadvantaged households, while the income-enhancing impact of off-farm work is more significant for higher-income households but at varying levels. The marginal effect of hiring farm machinery tends to be inversely related to household income per capita. In contrast, the income-increasing impact of off-farm work is more significant for households at higher per capita income levels.

Third, using FMRSs benefits disadvantaged households more than off-farm employment does. Specifically, the adoption of FMRSs by lower-income farm households helps reduce income disparities within the agricultural sector. However, the positive effect of FMRS usage on reducing income inequality can be offset by off-farm work engagement. In addition, the simultaneous decision to use FMRSs and work off-farm may be poor compared with off-farm employment. Namely, the application of FMRSs by farm households at lower income levels helps to reduce income disparity within the agricultural sector. The impact of FMRS use on reducing income inequality can be mitigated by off-farm work. Also, the decision to use FMRSs and to work off-farm simultaneously may increase increase existing income inequality significantly.

6.2. Implications

Our findings may have important policy implications for rural development and welfare. Public policies that support the development of the FMRS market can help raise living standards and economic wellbeing among rural households. For the majority of developing countries, governments should formulate differentiated purchase subsidy policies according to the type, function, and market demand of agricultural machinery. Especially for efficient and environmentally friendly medium and large agricultural machinery, subsidies should be increased to encourage farmers and agricultural machinery service organizations to upgrade and improve agricultural production efficiency. It is also important to build agricultural machinery service information platforms and service supervision platforms to achieve information transparency, improve service efficiency and quality, and promote the optimal allocation of agricultural machinery resources. In addition, governments should provide substantial relief support in terms of agricultural machinery fuel, maintenance, road tolls, etc., to reduce the economic burden of agricultural machinery service organizations. The establishment of a coordination mechanism for agricultural machinery cross-regional operation can be encouraged to reduce administrative barriers and to promote the cross-regional flow of agricultural machinery services.

Additionally, providing vocational skills training and off-farm job opportunities may be a good choice, but promoting FRMSs for lower-income households tends to be more effective than encouraging farmers to go off-farm when income inequality is taken into account. In other words, it is essential to consider the heterogeneous effects of FMRS adoption by households at varied income levels when making policies to increase farmers’ income through promoting farm mechanization and boosting crop yield. Without this knowledge, policymakers may design farm mechanization promotion policies that lead to severe income inequality. Therefore, different agricultural machinery promotion strategies should be designed for families with different income levels. For low-income households, greater subsidies, preferential loans, or leasing services can be provided to lower the threshold to adopt agricultural machinery. Also, farmers are encouraged to set up or join agricultural machinery service cooperatives to share agricultural machinery resources, reducing the cost burden on individual families. Last but not least, attention should be paid to the contribution migrant workers make to farmers’ income, except for agricultural machinery services. Policymakers can explore how to establish a complementary relationship between migrant workers and agricultural machinery services, for example, by providing rural migrant workers with technical support, such as farming and planting or whole-process agricultural machinery services, so that they can effectively manage farmland while they are away from home.

6.3. Limitations

Results and conclusions should be extended to other fields or countries with caution. It should be noted that the empirical analysis was conducted with data collected from rice producers in a river basin in China, which limits the application of the interpretation to similar situations in other countries or plant producers. Therefore, further studies focusing on crops other than rice and other regions/countries are necessary to better understand the heterogeneous effects within a broader context. Additionally, more updated evidence is needed to discuss this topic. Further research design could distinguish the heterogenous income effects of FMRSs used in the four stages (i.e., land plowing, rice transplanting, fertilizer and pesticide use, and harvesting).