The Effect of Off-Farm Employment on Agricultural Production Efficiency: Micro Evidence in China

Abstract

With ensuring food security becoming a priority for the Chinese government, the prevalence of off-farm employment (OE) may cast a shadow on agricultural productivity. Based on the data of the China Family Panel Studies in 2018, the Tobit model and threshold effect model have been applied to investigate the impact of off-farm employment on agricultural productivity efficiency (APE), measured by data envelopment analysis (DEA). The result has shown that: (1) OE contributes to a low level of APE. (2) Both self-employed off-farm employment (SOE) and wage-based off-farm employment (WOE) result in lower APE levels when endogenous issues are addressed. However, SOE had a greater negative impact on APE than WOE. (3) There exists a threshold for OE associated with a lower level of APE, indicating that the negative effect of OE on APE disappears when the degree of OE is high enough, SOE has a lower threshold than WOE. The study findings have implications for improving agricultural production efficiency in the context of large-scale off-farm employment of Chinese farmers.

1. Introduction

Increasing urbanization and rising domestic income have all contributed to the phenomenon of rural labor migration to the city. According to China’s 2019 monitoring and survey report on migrant workers, by the end of 2019, the number of farmers engaged in part-time employment in China had reached 290 million [1]. Having non-agricultural employment while engaging in agricultural production is the unique landscape of Chinese smallholder agriculture production. In the context of a small-scale peasant economy, off-farm employment (OE) was the inevitable choice of the family division of labor [2]. In order to avoid household economic risks and maximize income, off-farm employment of smallholder farmers has begun to emerge since China’s reform and opening up. Off-farm employment of small-sized farm owners in China has become a common phenomenon in rural areas and gradually expanded from coastal areas to central and western regions [3,4]. In addition, two types of off-farm employment were formed in the process: wage-based off-farm employment (WOE) and self-employed off-farm employment (SOE). The WOE farmers were employed and earning wage incomes. The SOE farmers employed themselves and earn operational income [5,6]. Human capital, social capital, value orientation, and agricultural labor time vary depending on OE type. In the context of the continuous refinement of off-farm employment types, investigating the impact of OE on agricultural production is an important issue.

An indicator of agricultural production needs to be constructed to realize the goal of investigating the impact of OE on agricultural production. Agricultural production efficiency is a farmer’s comprehensive ability to produce a given amount of output using the minimum amount of inputs, such as agricultural land and labor [4,7]. APE is related to the livelihoods of farmers and has an important association with agricultural economic growth [8]. Improving APE is essential in ensuring food security, which is in line with the policy goal of China. Therefore, it is appropriate to employ APE as a proxy indicator of agricultural production. The literature has confirmed that APE may be affected by many factors, such as land transfer [9], land tenure [10], shortage of irrigation [11], interpersonal trust [12], limited farming skills [13], and industrial air pollution [14]. With the rapid urbanization of China, more and more rural labor is migrating to urban areas to engage in off-farm work. Exploring how OE, SOE, and WOE affect APE is not only a supplement to the existing literature, but also provides insight into the policy recommendations to guarantee the high-quality development of China’s agriculture.

Existing studies have paid attention to the transfer of agricultural labor to non-agricultural sectors. The labor transfer leads to changes in household income structure and reallocation of labor resources, which will have a significant impact on agricultural production, and also some policy recommendations are put forward according to the empirical analysis. However, the conclusions were not unified. Some research from developing countries, such as China, Brazil, Pakistan, and Mexico, believes that transferring agricultural labor to the non-agricultural labor market may be beneficial to agricultural production. APE would be improved by the adoption of agricultural technology in order to substitute for labor in the context of agricultural labor migrating to urban areas [15,16]. In addition, the adoption of agricultural technologies and the maintenance of agricultural production facilities would be boosted due to the increase in household income caused by OE [17,18,19], which leads to the improvement of APE [20]. However, other researchers do not believe that OE is beneficial to APE. APE would deteriorate because of the limited production scale due to OE [21]. Dependence on agricultural income and agricultural work time, which are important transmission channels, results in OE reducing APE [22,23]. Some researchers believe that the negative and positive effects of off-farm work exist simultaneously [24]. On the one hand, farmers’ non-agricultural employment will reduce their attention to technological progress and the quality of the labor force. On the other hand, increased access to information and capital due to off-farm work can overcome credit and risk constraints, and improve the mechanization of agricultural production. Finally, the improvement of agricultural mechanization can largely offset the negative impact of off-farm labor transfer.

The heterogeneity of OE may have different effects on APE. As mentioned above, the heterogeneity of OE is reflected in its type and degree. First, different types of OE have different effects on agricultural production. Due to their social capital, entrepreneurial spirit, and risk-taking ability, SOE farmers are more conducive to agricultural production [25]. Compared to WOE farmers, SOE farmers may have a lower APE due to higher non-agricultural income and less reliance on agricultural production. Second, different degrees of OE have different effects on APE. The agricultural output will be maximized if households rely solely on agricultural income. The APE will be increased to achieve the goal. If households produce on-farm and off-farm at the same time, it is necessary to change off-farm inputs and farm inputs to maximize total income. Therefore, APE should be different at different OE stages and may have threshold values.

The research background and literature review show that it has theoretical and practical significance in studying the influence of OE on APE in China. We identify direct and indirect pathways for OE’s effect on APE based on existing research: the direct pathway is that farmers change their time and resource inputs to agricultural production directly as a result of off-farm employment, whereas the indirect pathway is that off-farm farmers adopt new production technologies, purchase social services, and change the scale of production, thus indirectly affecting APE. In addition, different OE types have different impact paths on APE. SOEs have a higher level of social capital, a value orientation, and a lower reliance on agriculture. As a result, both the direct path of agricultural inputs and the indirect path of technology adoption, service purchase, and land size change for WOE farmers are significantly different. This study aims to determine the OE-APE relationship and investigate how different types and degrees of OE will affect APE. Our study contributes to the literature on the OE-APE relationship in several important ways. First, we investigated the effect of OE on APE. Second, we explored how different OE types affected APE, namely SOE and WOE. Third, we analyzed threshold effects and threshold differences across various OE types in order to gain insight into the relationship between OE and APE.

2. Materials and Methods

2.1. Data Sources

The data used in my analysis was derived from the China Family Panel Studies (CFPS), administered by Peking University’s Institute of Social Science Survey. The survey covers 25 provinces, municipalities, or autonomous regions, representing 95% of the Chinese population, thus, it constitutes a nationally representative sample of Chinese households. Effective use of the rich CFPS data may shed light on China’s contemporary issues. We limited down the survey to the household that engaged in agricultural production, resulting in a final sample of 988 observations.

2.2. APE and OE Measures

2.2.1. Data Envelopment Analysis Model

APE is a subset of the concept of technical efficiency of production, and because economic efficiency of production requires the integration of multiple factors, technical efficiency of agricultural production is defined in this study as the maximum agricultural output that can be achieved with fixed factor inputs, or the least amount of factor inputs required to achieve fixed agricultural output [26]. At present, data envelopment analysis (DEA) and stochastic frontier analysis (SFA) are commonly used in efficiency measurements. In order to capture the variability in production efficiency caused by the heterogeneity of off-farming, it is necessary to measure the difference in efficiency between samples. SFA obtain agricultural production efficiency with a certain production function. CCR and BCC are the two fundamental model forms of DEA. CCR makes the assumption that scale returns remain constant and is capable of evaluating overall efficiency, including scale efficiency. The BCC model is used to determine the DMU’s pure technical and scale efficiency under a variable pay-for-scale scenario. Comprehensive efficiency can be used to describe the technical capability of a decision-making unit in terms of input and output. The APE that needs to be measured in this study is a comprehensive indicator of agricultural production under the assumption of low input costs; thus, the input-oriented CCR model is chosen. In order to avoid the errors caused by the assumed production function, we employed DEA to measure APE. It is difficult to improve the parameter accuracy by using conventional DEA methods because several decision-making units will be most efficient at the same time (efficiency is equal to 1). Therefore, the super-efficiency DEA model was employed to measure APE, which is suitable to analyze that the decision-making unit is 1 at the same time by ignoring the decision-making unit itself. Individual farmers consist of a decision-making unit (DMU). Assuming that there are n DMUs, m input indexes, and q output indexes. The super-efficiency DEA model can be expressed as follows:

$$\min \left(\mathsf{\theta }-\mathsf{\epsilon }\left({\mathrm{e}}_1^{\mathrm{T}}{\mathrm{S}}^{-}+{\mathrm{e}}_2^{\mathrm{T}}{\mathrm{S}}^{+}\right)\right)$$

(1)

$$\mathrm{s}.\mathrm{t}.\left\{\begin{array}{l} {\displaystyle \sum _{\begin{array}{c}\mathrm{k}=1\\ \mathrm{k}\ne \mathrm{j}\end{array}}^{\mathrm{n}}}{\mathrm{X}}_{\mathrm{k}}{\mathsf{\lambda }}_{\mathrm{k}}+{\mathrm{S}}^{-}={\mathsf{\theta }\mathrm{X}}_{\mathrm{j}}\\ {\displaystyle \sum _{\begin{array}{c}\mathrm{k}=1\\ \mathrm{k}\ne \mathrm{j}\end{array}}^{\mathrm{n}}}{\mathrm{Y}}_{\mathrm{k}}{\mathsf{\lambda }}_{\mathrm{k}}-{\mathrm{S}}^{+}={\mathrm{Y}}_{\mathrm{j}} \\ {\mathsf{\lambda }}_{\mathrm{k}}\ge 0, \mathrm{k}=1, \dots , \mathrm{n} \\ {\mathrm{S}}^{+}\ge 0, {\mathrm{S}}^{-}\ge 0 \end{array}\right.$$

In the above formula:$\mathsf{\theta }$ is the APE of the k-th farmer; $\mathsf{\epsilon }$ is non-Archimedean infinitesimal; S⁺ is the adjustment of input indicators, and S⁻ is the adjustment of output indicators; X_k represents the input of each farmer and Y_k represents the output of each farmer; X_j represents the audited input of the farmers, Y_j represents the audited output of the farmers; λ_k is the weight parameters.

The relative literature discerns several types of input, such as labor, irrigation water, and fertilizer. Agricultural output value and inputs cost are available in CFPS. In this paper, based on the indicator systems used in related APE measurement studies and the data available through CFPS research questionnaires [27,28,29], the input indicators are agricultural inputs (seeds, fertilizers, pesticides), hired labor inputs, machinery inputs, and irrigation inputs, respectively, and the output indicators are planting output value (including sold and own consumption) when assessing the APE of planting farmers via super-efficiency DEA. According to related literature [5,6], we employed the ratio of farmers’ off-farm income to total income to describe OE; SOE is measured by the ratio of self-employed income to total income; WOE is measured by the ratio of income from wages to total income.

2.2.2. Data Description

Evaluating through Equation (1) shows that the average off-farm employment degree of the sample farmers is 40.5% and the production efficiency is 47.3%. Off-farm employment of the farmers was prevalent in China, while the production efficiency of the farmers’ planting industry is still generally low. The distribution frequency of farmers’ off-farm employment (Figure 1a) further shows that the distribution of the overall off-farm employment degree of farmers is polarized. The sample size of farmers with an off-farm employment degree of less than 20% is the largest, accounting for more than 40% of the sample size; the sample size of farmers with an off-farm employment degree of more than 80% is the second largest, accounting for more than 20% of the sample size; and the sample size of farmers with an off-farm employment degree in the range of 20%–80% is insufficient, accounting for 40% of the sample size. Although a significant number of small farmers in China have entered the off-farm sector and a proportion of farmers with a high degree of OE have emerged, farmers with a low degree of OE continue to account for a sizable proportion. APE exhibited an approximately normal distribution (Figure 1b); among them, the number of samples with an agricultural production efficiency of 40%–60% is the largest, accounting for more than half of the total number of samples. Followed by the number of samples with an agricultural production efficiency of 20%–40% and 60%–80%, both accounting for about 20% of the total number of samples. While the number of samples with an agricultural production efficiency of 0%–20% and higher than 80% is the least, both are less than 5% of the total number of samples. With the description of the distribution status of agricultural production efficiency in consideration of the Chinese government’s vigorous efforts to carry out agricultural modernization and mechanization, there are relatively few low-efficiency farmers, but the proportion of high-efficiency farmers is still not high, and there is still a lot of space to improve the production efficiency of small farmers in China.

From

Table 1. Distribution of agricultural production efficiency and three types of off-farm employment.

APE	Average Degree of OE	Average Degree of SOE	Average Degree of WOE
0–20%	46.0%	7.4%	38.6%
20–40%	41.4%	5.6%	35.8%
40–60%	41.2%	4.0%	37.2%
60–80%	37.2%	3.7%	33.5%
Above 80%	36.2%	2.7%	33.5%

, the degree of OE, SOE, and WOE is lower in higher APE intervals. The mean of OE gradually decreases from 46.0% to 36.2%. The mean of SOE gradually decreased from 7.4% to 2.7%. The mean of WOE gradually decreased from 38.6% to 33.5%. The results indicate that there may exist heterogeneity in the relationship between OE and APE, which needs a deeper understanding of the potential impact of OE on APE.

2.3. Empirical Strategy

2.3.1. Benchmark Regression Model Setting and Variable Selection

The ordinary least squares (OLS) regression results will be biased because the minimum value of planting agricultural production efficiency calculated by the super-efficiency DEA model is 0, the so-called limited dependent variable. Therefore, we applied the Tobit model to identify the impact of OE on APE. The following econometric model was developed:

$${\mathrm{APE}}_{\mathrm{i}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{OE}}_{\mathrm{i}}+{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(2)

$${\mathrm{APE}}_{\mathrm{i}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{SOE}}_{\mathrm{i}}+{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(3)

$${\mathrm{APE}}_{\mathrm{i}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{WOE}}_{\mathrm{i}}+{\mathsf{\beta }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(4)

$${\mathrm{APE}}_{\mathrm{i}} = \left\{\begin{array}{c}0,{\mathrm{APE}}_{\mathrm{i}}\le 0\\ \ln \mathrm{APE},{\mathrm{APE}}_{\mathrm{i}}>0\end{array}\right.$$

In Equations (2)–(4), APE_i is the agricultural production efficiency of farmer i, which is obtained from Formula (1); OE_i, SOE_i, and WOE are the core independent variables, OE_i is the ratio of the farmers’ off-farm income to total income; SOE_i is the indicator of the degree of SOE, which is measured by the ratio of self-employed income to total income; WOE_i is the indicator of the degree of WOE, which is measured by the ratio of income from wage to total income. X_ki is a vector of control variables. Specifically, AGE is the age of the head of household; Gen is the gender of the head of household, female = 0, male sex = 1; EDU refers to the education level of the head of the household, ranging from illiterate/semi-illiterate to PhD, with a value of 1–8; FNI is the family net income, in 10 thousand yuan; PCA is the per capital assets of households, including total household assets (operating assets, real estate, durable consumer goods, agricultural machinery, cash assets, and financial assets). The unit is 10 thousand yuan; AM is the value of household agricultural machinery, including water pumps, tractors, and machine-driven agricultural tools measured by 10 thousand yuan; FE is the social interaction expenditure of household, which is calculated by the physical and cash human courtesy of the family, and the unit is 10,000 yuan; β₀ is the constant term of the regression equation; β₁–β₈ is the regression coefficient of each variable; µ_r indicates regional control effect; $\epsilon$ is the error term. The descriptive statistics of variables of the econometric model are shown in

Table 2. Descriptive statistics.

Variable Name	Variable Code	Variable Definition	Mean Value	Standard Deviation
Agricultural Production Efficiency	APE	Continuous Variable	0.473	0.209
Degree of overall off-farm employment	OE	Continuous Variable	0.405	0.366
Degree of self-employed off-farm employment	SOE	Continuous Variable	0.044	0.169
Degree of employed off-farm employment	MOE	Continuous Variable	0.361	0.359
Age	AGE	Continuous Variable, Year	53.569	12.408
Gender	GEN	Female = 0 Male = 1	0.607	0.489
Education level	EDU	Literate/semi-illiterate = 1 Primary school = 2 Junior high school = 3 High school/technical secondary school/technical school/vocational school = 4 Junior college = 5 Undergraduate = 6 Master = 7 Doctor = 8	2.443	1.034
Net household income	FNI	Continuous Variable, 10,000 yuan	4.453	4.639
Per capital household assets	PCA	Continuous Variable, 10,000 yuan	7.851	11.558
Total value of household agricultural machinery	AM	Continuous Variable, 10,000 yuan	0.344	1.230
Family affection expenditure	FE	Continuous Variable, 10,000 yuan	0.329	0.412

.

2.3.2. Threshold Effect Regression Model Setting and Variable Selection

From descriptive analysis, there may be a threshold value in the relation of OE-APE. We applied the threshold effect regression model to investigate the threshold by the approximate distribution of the least squares estimator of the threshold [30]. The threshold effect model is developed as follows:

$${\mathrm{APE}}_{\mathrm{ai}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_{1\mathrm{a}}{\mathrm{OE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{OE}}_{\mathrm{i}}\le \mathrm{r})+{\mathsf{\alpha }}_{1\mathrm{b}}{\mathrm{OE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{OE}}_{\mathrm{i}}>\mathrm{r})+{\displaystyle \sum }_2^{\mathrm{k}}{\mathsf{\alpha }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(5)

$${\mathrm{APE}}_{\mathrm{ai}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_{1\mathrm{a}}{\mathrm{SOE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{SOE}}_{\mathrm{i}}\le \mathrm{r})+{\mathsf{\alpha }}_{1\mathrm{b}}{\mathrm{SOE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{SOE}}_{\mathrm{i}}>\mathrm{r})+{\displaystyle \sum }_2^{\mathrm{k}}{\mathsf{\alpha }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(6)

$${\mathrm{APE}}_{\mathrm{ai}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_{1\mathrm{a}}{\mathrm{WOE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{WOE}}_{\mathrm{i}}\le \mathrm{r})+{\mathsf{\alpha }}_{1\mathrm{b}}{\mathrm{WOE}}_{\mathrm{i}}{ * \mathrm{I}(\mathrm{WOE}}_{\mathrm{i}}>\mathrm{r})+{\displaystyle \sum }_2^{\mathrm{k}}{\mathsf{\alpha }}_{\mathrm{k}}{\mathrm{X}}_{\mathrm{ki}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(7)

In Equations (5)–(7), OE_i, SOE_i, and WOE_i are the core independent variables. Where r is the threshold value to be estimated, I is the indicator function, which equals 1 when satisfying its condition, otherwise, it is 0; the control variable in the formula is the same as above.

3. Results and Discussions

3.1. Baseline Model Regression Results

The regression results of the benchmark model (Equations (2)–(4)) are shown in

Table 3. Impact of off-farm employment on agricultural production efficiency.

	(1)	(2)	(3)	(4)	(5)
VARIABLES	APE	APE	APE	APE	APE
OE	−0.0444 **	−0.0424 **	−0.0551 ***
	(0.0181)	(0.0184)	(0.0190)
SOE				−0.0664
				(0.0413)
WOE					−0.0446 **
					(0.0199)
AGE		0.000353	0.000585	0.000695	0.000670
		(0.000591)	(0.000595)	(0.000595)	(0.000594)
GEN		−0.00632	−0.00333	0.000926	−0.00258
		(0.0143)	(0.0142)	(0.0141)	(0.0142)
EDU		−0.0137 *	−0.0149 **	−0.0134 *	−0.0154 **
		(0.00699)	(0.00700)	(0.00704)	(0.00703)
FNI			0.00353 **	0.00186	0.00352 **
			(0.00163)	(0.00159)	(0.00167)
PCA			−0.0000608	0.000209	−0.000239
			(0.000585)	(0.000609)	(0.000591)
AM			−0.00649	−0.00464	−0.00630
			(0.00568)	(0.00567)	(0.00570)
FE			0.0488 ***	0.0533 ***	0.0478 ***
			(0.0169)	(0.0170)	(0.0170)
Constant			0.473 ***	0.445 ***	0.465 ***
			(0.0429)	(0.0417)	(0.0429)
LR chi2	5.99 **	12.33 **	29.29 ***	23.51 ***	25.96 ***
Pseudo R2	−0.0208	−0.0427	−0.1015	−0.0814	−0.0899
DWH test significance			0.0188	0.0095	0.0088
First stage F value			36.0989 ***	34.5998 ***	11.5301 ***

Standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

. The LR test depicts that the independent variables are joint significantly (p < 0.01 or p < 0.05), and the fitness degree of the model is good. Model 1 only sets OE as the independent variable, model 2 adds characteristics of the individual household head, and model 3 adds characteristics of the household. The regression results of the three models show that the OE negatively affects agricultural APE and the results are robust. It indicates that the comprehensive impact of OE on APE is negative.

Models 4 and 5 were used to explore how different types of OE affect APE. WOE and SOE negatively affected APE, but only WOE was statistically significant. Regarding family factors, higher education was associated with lower APE. Well-educated people tend to place more value on high-paying off-farm jobs and neglect agricultural production. Although, it has been noted that education can help farmers improve their APE by increasing their access to information and adopting new technologies [31]. Education, on the other hand, contributes positively to OE [32], and it has been demonstrated to have a negative effect on APE in this study. Thus, farmer education should have a bidirectional effect on APE, which has been demonstrated in this paper to be negative. Higher household income is associated with higher APE, indicating improving APE caused by the adoption of agriculture technologies with more expenditure. A higher level of social capital contributes to higher APE. The ability to access information on agricultural production may be a channel through which social capital increased APE.

3.2. Endogenous Discussion and IV Regression Correction

However, the possibility of endogeneity arises in our baseline model of the effect of OE on APE, including the possibility of omitted variable bias, measurement error, and reverse causality. The bias of omitted variables will render the estimated coefficient of OE either to be biased upward or downward. In particular, one key limitation of the baseline model estimate is that it cannot rule out some unobserved factors that may simultaneously impact OE and APE. For instance, social support may not only affect OE but also APE. Another potential endogeneity source is a measurement error in estimating OE, especially when households do not accurately recall their income, possibly due to their different levels of socioeconomic status and cognitive abilities. There may also exist endogeneity problems, such as reverse causality. For instance, farmers’ income may be a channel through which APE impacts farmers’ decisions on OE.

To rule out these endogeneity issues, we ran a two-step method (control function approach) using the communication expenditure (postal and telecommunication charges) of the household as the IV instruments, under the implicit assumption that an increase in communication expenditure indicates a person left home for a long time and over a long distance, which raises the likelihood of OE.

The Durbin-Wu-Huasman test indicated that core variables are not exogenous (Model 3 OE: 0.0188; Model 4 SOE: 0.0095; Model 5 WOE: 0.008). In Table 3, we present the results from our two-stage analysis. The F statistics of the first stage are much larger than the critical value set by Stock & Yogo [33], suggesting no weak IV instrumentation. After correcting for endogeneity issues, the relationship between OE (including SOE and WOE) and APE are negative and statistically significant.

In

Table 4. IV Tobit estimates.

	(6)	(7)	(8)
VARIABLES	APE	APE	APE
OE	−0.288 ***
	(0.108)
SOE		−0.639 ***
		(0.244)
WOE			−0.526 **
			(0.233)
Controls	Controlled	Controlled	Controlled
Constant	0.608 ***	0.473 ***	0.719 ***
	(0.0768)	(0.0471)	(0.133)
Wald chi2	25.62 ***	24.59 ***	18.46 **

Standard errors in parentheses; *** p < 0.01, ** p < 0.05.

, the bias-corrected estimates from the IV Tobit are larger than our baseline estimates, and the negative effect of SOE on APE passes the significance test in model 7. This implies that our baseline estimates are downward biased. The sign on the coefficients of OE from the IV Tobit estimates is consistent with the baseline specification and thus confirms the general conclusion that OE (including SOE and WOE) has a significant and negative effect on agricultural APE. Moreover, after correcting endogeneity, the negative effect of SOE on APE was slightly larger than WOE on APE. This is due to the farmers pursuing the maximization of their own interests as the microeconomic subjects in the market. The essence of the impact of OE on agricultural production is the change of production and operation objectives and the structure of production factors; the reason for the change is to adapt to the characteristics of the farmers’ own labor and capital income, and the ultimate goal of the change is to maximize the overall interests of the family. Therefore, compared with WOE, SOE shows a more declined dependence on agricultural income, which may be the reason why SOE had a stronger effect on APE.

3.3. Threshold Effect Model Results and Discussion

As mentioned above, there may exist a threshold effect in the impact of OE on APE.

Table 5. Threshold effect test.

	Single Threshold	F Value	Critical Value-1%	Citical Value-5%	Critical Value-10%
OE	0.542	4.149 *	7.782	4.682	3.230
SOE	0.640	2.416 *	6.361	3.310	2.317
WOE	0.945	9.368 ***	7.100	5.082	3.220

Standard errors in parentheses; *** p < 0.01, * p < 0.1.

shows the result that three indicators of off-farm work have single threshold values (OE: 0.452; SOE: 0.640; WOE: 0.945) and are statistically significant by using bootstrap to run 300 simulations.

To ensure that our results are robust to endogeneity, the IV Tobit has been applied to analyze the threshold effect. Taking the threshold value as the boundary, our threshold effect analysis is carried out for the higher and lower threshold values, respectively (

Table 6. Result of threshold effect analysis.

	(9)		(10)		(11)
	OE-APE	OE-APE	SOE-APE	SOE-APE	WOE-APE	WOE-APE
Threshold	<0.542	>0.542	<0.64	>0.64	<0.945	>0.945
Coefficient	−1.178 *	−1.263	−3.428 *	−0.225	−0.569 **	33.78
	(0.714)	(1.132)	(2.054)	(1.409)	(0.238)	(166.9)
Controls	Controlled	Controlled	Controlled	Controlled	Controlled	Controlled
Constant	0.778 ***	1.439	0.459 ***	0.934	0.745 ***	−33.21
	(0.212)	(0.894)	(0.0711)	(1.105)	(0.138)	(167.4)
Wald chi2	3.07 *	1.56	4.83 **	0.02	6.00 **	0.21

Standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

). The result of the threshold effect model implies that the impact of OE, SOE, and WOE on APE has a threshold. If it is above the threshold value, the effect of OE, SOE, and WOE on agricultural APE will disappear.

This is because peasant households lost a small part of their labor force and saw a slight rise in total income during the initial stage of the OE. However, the remaining labor force can still maintain family agricultural production, and the additional income brought by OE is not enough to have a subversive impact on the production and life of farmers before the threshold value. Farmers may not increase production investment, seek to acquire mechanization services, or update production technology at this time in order to maximize total household income. Therefore, factors conducive to the improvement of APE may not appear in the primary stage of OE of farmers [34]. At the same time, the initial OE of members of a peasant household will also lead to a decline in the proportion of agricultural income and the aging of the agricultural population, which will reduce the importance and dependence of agricultural production in the household. The limited and aging labor force will adopt more extensive production methods in order to complete agricultural production [7], which has a negative impact on APE [35]. Therefore, as the OE of households increases but does not reach the threshold, the productivity of family farming will decline.

As more and more peasants work in cities and towns, the cost of agricultural labor has gradually risen, and this, along with China’s vigorous promotion of agricultural mechanization, has jointly promoted the rise of China’s agricultural social services. At this time, when the degree of OE of a peasant household reaches a certain stage, the additional income brought by OE has occupied an absolute share in the total production of the family. Agricultural production has only become the survival guarantee of farmers, and the remaining agricultural labor force also cannot complete the existing family farming production. Therefore, in order to release more labor for non-agricultural production and maximize the total family income, peasant households began to purchase social services in agricultural production, which is conducive to the improvement of family APE. In addition, the continuous improvement of OE has also increased the supply of agricultural land in the land transfer market, thus realizing the further concentration of land, and the farmers who have flowed into the land have realized a more rational and effective use of land resources by virtue of their large-scale and mechanized production [36,37]. To sum up, the factors that are not conducive to APE still exist (the decline of the importance of agricultural production, the reduction in the agricultural population, and the aging of the population), but the factors that are conducive to APE (the purchase of social services and the scale of land) began to appear when the OE rose to the threshold. Therefore, the negative effect of OE on APE is offset, resulting in the impact of OE on APE no longer being significant after the threshold value.

In addition, the thresholds for different types of OE are not the same. SOE has a lower threshold value than WOE, which has a higher threshold. WOE is less innovative, less ambitious, and more traditional, making it harder to adopt modern agricultural techniques and buy agricultural services that are better for APE improvement. However, SOE has a lower threshold for their impact on APE due to their higher income, more social capital, and longer-term visions.

4. Conclusions and Policy Implications

China’s rapid urbanization and the widespread participation of farmers in non-farm employment pose significant challenges to China’s agricultural sustainable development. Given the urgent priority of improving APE for smallholder farmers in China, there is still a need to explore its various determinants. Despite much attention to the determinants of APE, empirical research provides few insights into the association between OE and APE, particularly in China. Our analysis of nationally representative data from the CFPS is designed to shed light not only on the OE-APE relation in China but also on the heterogeneity and threshold effect of the OE impact on APE.

The analysis yields several key findings. Firstly, OE contributes to a low level of APE. This demonstrates that the negative impact of OE on APE (the aging of the agricultural population, the decline in agricultural attention, the lag in new technology updates, etc) is greater than the positive impact. Secondly, both SOE and WOE lead to lower levels of APE with killing endogenous problems. However, the negative effect of SOE on APE was greater than WOE. This is because SOE is less reliant on agriculture and has a greater negative effect on APE than WOE when the objective is to maximize total household income. Thirdly, there exists a threshold for OE associated with a lower level of APE, indicating that the negative effect of OE on APE disappears when the degree of OE is high enough, thus SOE has a lower threshold than WOE. This is because as OE approaches the threshold value, favorable factors for APE (increased agricultural investment, expansion of land scale) begin to emerge, partially offsetting the negative impact of OE on APE. Additionally, because SOE possesses greater social capital and is less reliant on agriculture, its threshold point appears to be earlier than that of WOE.

Our findings have important implications for policy. (1) Above all, as Chinese farmers’ current non-farm employment behavior is not conducive to increasing agricultural production efficiency, it is necessary to promote agricultural labor force development to achieve nearby non-farm employment, and to continue attracting talented people to return to their hometowns and cultivating new types of professional farmers. This encourages returning capable people and new professional farmers to provide agricultural socialization. (1) Given the heterogeneity of OE, agricultural production subsidies, and agricultural technology adoption subsidies should favor low-level OE and WOE. This encourages the purchase of agricultural production custody services by high-level OE and SOE farmers. Promoting land consolidation to new business entities with higher production efficiency could effectively use idle OE land to provide an external guarantee for agricultural production efficiency improvement in the face of OE spread.