Balancing Livelihoods and Sustainable Development: How Does Off-Farm Employment Affect Agricultural Green Total Factor Productivity in China?

Abstract

To contribute to the United Nations’ 17 Sustainable Development Goals (SDGs), this study focuses on improving two specific goals—SDG 2 (Zero Hunger) and SDG 12 (Responsible Consumption and Production)—by examining how off-farm employment affects agricultural green total factor productivity (GTFP) in China, a key link between rural socio-economic transformation and agricultural sustainability. The results show that: First, the part-time operation of farmers significantly reduces the green total factor productivity, and the negative impact is more pronounced for off-farm employment households with higher non-agricultural income shares. It mainly stems from the redundant input of land and machinery elements. Second, the effect showed obvious heterogeneous effects at different stages of family development and land management scale. In addition, the scale effect of continuous agricultural production services and the technological synergy effect driven by the deepening of agricultural division of labor are the key to improving green total factor productivity and alleviating the negative effects of part-time operations. In summary, promoting sustainable agricultural practices requires the government to further deepen the reform of the land property rights system and optimize the agricultural socialization service system to ensure both food security and environmental sustainability.

1. Introduction

In the modern quest for sustainable progress, finding a synergy between financial expansion and nature conservation is increasingly vital. Consequently, moving beyond legacy growth patterns defined by heavy emissions and excessive resource consumption is no longer optional. Guided by the concept of green development, integrating resources, technology, innovation, and other elements to achieve an efficient, low-consumption, clean, and environmentally friendly development model not only improves the quality and resilience of economic development but also builds a solid foundation for sustainable development. In the context of the current agricultural modernization and agricultural transformation and upgrading, high-quality agricultural development not only requires increasing output and ensuring food security, but also requires reducing pollution emissions and developing green agriculture. Agricultural Green Total Factor Productivity (GTFP) measures the effectiveness and efficiency of factor allocation and utilization under resource and environmental constraints [1,2], serving as a critical indicator for quantifying sustainable agricultural development and assessing progress toward environmental sustainability goals [3]. Enhancing agricultural GTFP is strategically important for achieving sustainable food security while maintaining ecological integrity. It also contributes to the establishment of a green, low-carbon, and circular economic system, thereby advancing sustainable agricultural transformation.

China’s agricultural operations are primarily driven by 230 million smallholder farms, which account for 70% of the total agricultural land area. Farm households dominate agricultural production, and their diverse production and management approaches, as well as their attitudes, directly influence agricultural input and the allocation of production factors, thereby impacting agricultural GTFP [4,5]. As industrialization and urbanization advance, the migration of young and middle-aged laborers to non-agricultural sectors has accelerated [6,7], labor migration, which often does not involve the entire family and leaves many members remaining in rural areas [8], has led to the widespread and long-term persistence of a ‘dual role’ agricultural economy in rural areas, characterized by ‘staying on the land while leaving the village’ and ‘off-farm employment and part-time employment’ [9]. From the perspective of direct factor input and output, this phenomenon increases the opportunity cost of agricultural labor, causing land to be primarily cultivated by households with weaker productive capacities or other family groups [10,11]. This phenomenon triggers a simultaneous contraction in the quantity and quality of agricultural labor. Such a trend can lead to resource misallocation [12,13], dragging down the productivity of both smallholders and the industry as a whole. However, some literature argues that working outside the farm increases household income, helping to relax the constraints on farming inputs. These additional funds enable families to adopt new techniques or hire machinery, which effectively compensates for the loss of physical labor and offsets the potential damage to agricultural efficiency [14,15,16].

However, despite these extensive discussions, critical gaps remain in the existing literature regarding the micro-foundations of GTFP. The majority of current studies measure GTFP using macro-level data at the national or provincial scale. While valuable for understanding regional trends, this macro-approach masks the micro-level behavioral logic of farmers in resource allocation and fails to capture the heterogeneous effects across specific crop types. Agricultural decisions are highly context-specific, and aggregate data cannot fully reflect the production realities of specific growers, such as maize farmers. Existing research has not sufficiently investigated the differences between different types of households, particularly regarding the “Family Life Cycle.” Most studies treat off-farm employment as a linear variable or treat farmers as a homogenous group, ignoring that households in the growth, burden, and decline stages possess vastly different resource endowments and labor capabilities. How family-specific traits shape the relationship between labor migration and green productivity remains under-investigated at the micro level. In particular, few researchers have analyzed the synergy between resource scarcity and agricultural outsourcing. It remains unclear how different service combinations interact with household features to stabilize efficiency during the transition to sustainability.

To address these shortcomings, this study bridges these gaps through three primary marginal contributions. First, unlike prior research that focuses on macro-trends, this study shifts the focus to the micro-level by taking maize farm households as the specific research object. Through the construction of a theoretical model of household factor allocation decisions and the application of advanced empirical methods such as the inverse probability weighting (IPW) model and the threshold effect model, the study investigates the direction and underlying logic of the effects, thereby providing new empirical evidence that is often obscured in macro-level studies. Second, this research adopts a life-stage framework, viewing the farming household as a fluid socio-economic entity. By analyzing how demographic shifts, financial needs, and resource availability vary across family phases—ranging from expansion to contraction—the study provides a nuanced assessment of the influence of non-farm labor on green productivity. This lens reveals the underlying diversity in household decision-making. Third, distinguishing between isolated and integrated agricultural supports, the paper examines how these services counteract efficiency drops. We identify two primary channels: the realization of economies of scale through indirect management and the enhancement of technical synergies via professionalized farming tasks.

Against the backdrop of Northeast China’s fertile plains and high mechanization levels, this research utilizes empirical evidence from a survey of 837 maize-growing households. By integrating frameworks such as induced technical change and the division of labor, we evaluate how non-farm labor participation differently influences the GTFP of the agricultural sector. Furthermore, the paper clarifies the underlying drivers of this relationship and suggests practical pathways for promoting sustainable grain cultivation.

2. Theoretical Analysis and Research Hypothesis

Within the framework of the new economics of labor migration, Stark [17,18] identified the household as the core decision-making unit for labor migration. Migration decisions are made collectively by the household, based on resource endowments and economic conditions, to diversify risk, optimize resource allocation, and maximize household income, serving as a fundamental socio-economic strategy for rural livelihood stability. Factors such as age, gender, skill level, and the broader economic environment shape these decisions. Labor allocation within the household often follows a skill-based sorting process. Youthful and more proficient individuals tend to transition into non-agricultural sectors in pursuit of superior financial rewards. Conversely, the elderly or those with fewer technical skills, who often face limited mobility in the broader job market, generally persist in farming to ensure a consistent source of household revenue. Part-time employment consequently provides households with greater economic benefits, ensuring their long-term persistence while reinforcing the tendency toward deeper engagement in non-farm activities as income rises. However, this shift also reallocates household labor time and effort toward non-agricultural sectors, resulting in rising labor costs and reduced availability of labor for farming. Since those who migrate for work are often the relatively more skilled household members [19,20], the quality of agricultural labor input declines. Such changes lead to inefficient factor allocation and ultimately reduce overall GTFP.

From the perspective of production capital input, the theory of induced technological change suggests that labor scarcity caused by off-farm employment increases labor costs, which encourages farmers to reduce reliance on manual labor. In pursuit of profit maximization, farmers adjust production strategies in response to changes in factor prices, often by expanding the use of relatively inexpensive machinery or by relying on chemical inputs such as fertilizers and pesticides that deliver short-term efficiency gains [21,22]. However, heavy dependence on chemical inputs can raise production costs, while limited time and attention under part-time operations constrain refined management, resulting in inefficient factor allocation and a decline in GTFP [23]. Heavy dependence on synthetic inputs is fundamentally at odds with the objectives of eco-friendly growth. The intensive application of chemicals triggers a rise in negative externalities, notably greenhouse gas discharges and diffuse pollution. These factors collectively damage ecological health and, as a result, hinder the advancement of GTFP. Although machinery can substitute for labor and maintain crop yields, farming tasks are often undertaken by household members with weaker labor capacity, which may lead to inefficient “self-service” use of small-scale machinery [24] or higher machinery costs. These outcomes undermine the coordination among land, labor, and equipment, ultimately lowering GTFP. In addition, while agricultural machinery services can provide effective labor substitution for part-time farmers [14,25], in highly mechanized regions such as Northeast China, their adoption may not substantially raise GTFP relative to other farmers. Instead, because part-time farmers allocate more effort to non-farm activities, they tend to select labor-saving but more expensive machinery services, driving up production costs without producing significant output gains. Service providers generally prioritize yield security or single-stage efficiency rather than green technologies, and the misalignment of goals between providers and farmers further results in high service fees or limited production benefits. Overall, labor-substituting capital inputs not only fail to maximize profits but also reduce GTFP through rising production costs, increased environmental pollution compromising sustainability, and diminished production efficiency [9].

From the perspective of land input, some farmers reduce agricultural risks by organizing production according to household needs, which may involve leaving part of the land uncultivated or transferring it to others, thereby restricting farming to a smaller scale within their contracted plots. However, the resulting fragmentation often leads to inefficiencies and undermines the long-term ecological foundation [19], causing land input mismatches. Furthermore, when farmland management rights are not transferred or are exchanged only among farmers, the outcome remains a replication of inefficient “smallholder” operations, which entails potential efficiency risks [24].

From the perspective of labor input, part-time employment reduces the amount of household labor available for farming. Although such a shift may raise labor productivity per unit, it can also result in insufficient actual output or require additional inputs of other factors, thereby increasing overall production costs and undesirable outputs. As the degree of part-time engagement deepens, the availability of agricultural labor declines further, amplifying the negative impact on GTFP.

By integrating the new economics of labor migration with the framework of induced technological change, we argue that efficiency loss primarily stems from the misalignment of factor prices. As the opportunity cost of labor rises, households are compelled to replace human labor with intensive chemical applications and ill-suited mechanical services. This shift deviates from the theoretical equilibrium of factor allocation, ultimately leading to a downturn in green productivity. On this basis, we formulate the following hypothesis:

Hypothesis 1 (H1).

Off-farm employment negatively affects GTFP, and the effect intensifies as the degree of part-time engagement increases.

Farm household decision-making reflects the combined influence of multiple factors, with motivations and external drivers of labor allocation and input use differing according to household objectives. With the integration of behavioral economics and sociology into household studies, the theory of “commercial smallholders” argues that farmers’ decisions lie between morality and rationality. Farmers pursue not only profit maximization but also livelihood stability, and their choices are shaped by multiple rational considerations [26]. Although rationality is a common feature among farmers, levels of rationality and pursued goals vary. Consequently, classification of households is essential for meaningful analysis [27]. Building on the theoretical framework of the impact of off-farm employment on GTFP, the study further investigates potential heterogeneity from two perspectives: household development stages and farm operation scale.

For households at different stages of development, the GTFP losses caused by off-farm employment may vary significantly. During the growth and decline stages, household size is relatively small, and off-farm employment often leaves agricultural production to elderly members or women. Constrained by physical capacity and technical skills, they tend to rely heavily on fertilizers and pesticides to compensate for reduced labor input, resulting in environmentally damaging and ecologically inefficient production practices. In contrast, households in the burden stage generally have more abundant labor, and older members often possess agricultural skills and experience. When higher-paying off-farm employment opportunities are limited, such households can maintain agricultural production by allocating labor more efficiently and adopting more scientific and environmentally friendly practices. As a result, reliance on chemical inputs remains limited, and the efficiency losses from non-farm employment of younger members are less pronounced.

Synthesizing the family life cycle theory with the resource endowment perspective, the impact of off-farm employment is moderated by the household’s labor/capital ratio. In the growth and decline stages, the rigid constraint of labor endowment forces a “survival-oriented” substitution strategy (high chemicals), whereas the burden stage possesses sufficient labor buffer to maintain “development-oriented” green management.

Hypothesis 2 (H2).

Off-farm employment exerts stronger negative effects on GTFP in households at the growth and decline stages than in those at the burden stage, where labor is relatively abundant.

In terms of the scale of land operation, generally speaking, the larger the scale of operation of a rural household, the higher the degree of negative impact brought by part-time operation. The reason lies in the essential conflict between the production characteristics of large-scale land operation and the resource dispersion attribute of part-time operation. This conflict is transmitted layer by layer and eventually amplifies the negative effect. From the perspective of labor supply and factor substitution, large-scale agricultural production has much higher requirements for labor in terms of time and space than small-scale operations [28,29]. There are strict key management nodes during the growth periods of different crops. Part-time farmers often need to allocate time and energy between agricultural production and non-agricultural activities. The rigid time constraints of non-agricultural activities can easily lead to the absence of agricultural labor at key nodes. To avoid yield losses caused by labor shortages, farmers have no choice but to use chemical factors to replace labor. The extensive production strategy, such as increasing the input of chemical fertilizers and pesticides, intensifies the negative externalities of the environment [23]. On the other hand, although larger-scale part-time operators can alleviate labor constraints by replacing household labor with agricultural machinery services and theoretically achieve a scale effect of cost reduction per unit area through the continuity of mechanical operations, the transaction costs and management costs hidden in this process are often underestimated [30]. In addition, large-scale land operation places higher demands on the refinement of field management [28], but part-time households, limited by time and energy, frequently face delays in sowing or pest control. Such extensive practices reduce land productivity, increase undesirable emissions, and ultimately lead to losses in GTFP. This outcome highlights the challenge of reconciling large-scale operations with the refined management required for agricultural sustainability.

This phenomenon can be explained by the theory of diseconomies of scale regarding managerial capacity. While large-scale land theoretically offers economies of scale, for part-time farmers, the high transaction costs and monitoring difficulties associated with large-scale management exceed the benefits. The “time poverty” caused by off-farm employment creates a threshold where expanding land scale exacerbates the mismatch of green inputs.

Hypothesis 3 (H3).

The negative effect of off-farm employment on GTFP follows a nonlinear pattern with respect to farm operation scale.

3. Data Sources Model Design, and Variable Setting

3.1. Data Collection and Source Description

Northeast China is the largest grain-producing region in the country, accounting for approximately one-fifth of the national grain output. With abundant agricultural resources and significant potential for yield growth, the region carries a critical responsibility for ensuring national food security. The 2023 policy document Opinions on Further Promoting New Breakthroughs in the Comprehensive Revitalization of Northeast China in the New Era highlighted the importance of leveraging the region’s comparative advantages to enhance grain production capacity, secure supply, and pursue a path of high-quality and sustainable development. The selection of this region is strategic for sustainability research, given its status as the black soil base (a non-renewable ecological resource) and the concurrent rapid out-migration of rural labor (a major socio-economic shift). Therefore, the study focuses on maize farmers in Northeast China. Maize was chosen as the focus crop because it is the country’s most important grain, and the three northeastern provinces are located within the Northeast–Southwest maize belt. The region’s synchronized rainfall and temperature patterns provide highly favorable conditions for crop growth, making it one of China’s primary maize-producing areas. Accordingly, investigating GTFP among maize farmers in this region is both representative and policy-relevant.

A multi-stage random sampling method was used to select survey respondents. Consideration was given to regional economic development, geographic distribution, and differences in land resource endowments. Ten key maize-producing cities in Liaoning, Jilin, and Heilongjiang were identified. From each city, two counties (or county-level cities) were randomly chosen, followed by two townships within each county and three to four villages within each township. In each village, 10–20 maize farmers were randomly selected for the study. The survey was conducted between June and September 2023. Face-to-face interviews were carried out with household heads or the primary agricultural decision-makers.

The questionnaire covered four main areas. First, information was collected on household head characteristics, including gender, age, education level, health status, and off-farm employment. Second, household-level characteristics were recorded, including total family size, the number of agricultural and non-agricultural laborers, and household income composition. Third, farm production and management information was gathered, covering cultivated land area, plot topography, degree of mechanization, and use of agricultural production services, including transaction costs, number of services purchased, and associated expenses. Based on management practices, farming households were classified into three categories: full outsourcing, partial outsourcing, and self-cultivation. Households that fully outsourced production, transferring all management rights to service providers and only receiving income from grain production, were excluded from the analysis, as they made no production decisions and thus had no direct impact on production efficiency or pollution emissions. Fourth, external environmental factors were considered, including whether households received government subsidies for green production, recent exposure to natural disasters, and the distance between the village and the nearest county seat.

To ensure data accuracy, samples with substantial missing information or abnormal values for key variables were excluded. After data cleaning and applying the study’s selection criteria, a total of 837 valid maize-farming households from Liaoning, Jilin, and Heilongjiang were retained. The geographic distribution of the survey areas and the number of samples are presented in

Table A1. Distribution of Data Source Regions.

Province	City	County (or District)	Township (or Town)
Heilongjiang (149)	Harbin (45)	Hulan District	Mengjia Township, Xubao Township, Zhangjiaweizi Town
		Tonghe County	Fengshan Town, Tonghe Town
	Jiamusi (54)	Fujin City	Dayushu Town, Shangjiej iTown
		Huachuan County	Yuelai Town
	Mudanjiang (50)	Hailin City	Hailin Town, Hainan Korean Ethnic Township (Xi’an District)
		Ning’an City	Dongjingcheng Town, Fanjia Township, Ning’an Town
Jilin (301)	Liaoyuan (148)	Dongfeng County	Daxing Town, Lalahe Town
		Dongliao County	Baiquan Town, Lingyun Township
	Siping (81)	Lishu County	Donghe Town, Pingxi Township, Shijiabao Town, Wanfa Town, Yutai Town
		Shuangliao City	Liaonan Subdistrict, Wohu Town, Wolong Town, Yongjia Township
	Changchun (72)	Gongzhuling City	Chaoyangpo Town, Fanjiatun Town, Huanling Township, Qinjia Town
		Nong’an County	Qiangan Township
		Yushu City	Bahao Town
Liaoning (387)	Anshan (82)	Haicheng City	Gengjia Town, Xisi Town
		Tai’an County	Huandong Town, Sanglin Town
	Fuxin (137)	Fuxin Mongol Autonomous County	Furong Town, Guohua Township
		Zhangwu County	Qianfuxingdi Town, Wufeng Town
	Shenyang (90)	Faku County	Mengjia Town, Sijiazi Mongol Township
		Xinmin City	Daliutun Town, Gongzhutun Town
	Tieling (78)	Changtu County	Baoli Town, Laocheng Town
		Tieling County	Cainiu Town, Xintaizi Town

. In addition, to provide a comprehensive understanding of the sample structure, the basic characteristics of the sample, including household head characteristics (

Table A2. Household head characteristics.

Variable	Category	Sample Size (Households)	Percentage (%)	Variable	Category	Sample Size (Households)	Percentage (%)
Gender	Male	794	94.86	Health Status	Very Poor	20	2.39
	Female	43	5.14			92	10.99
Age	<35	11	1.31		Average	130	15.53
	35–44	68	8.12		Good	277	33.09
	45–54	223	26.64		Very Good	318	37.99
	55–64	314	37.51	Working Status	Working Outside	202	24.13
	>65	221	26.40		Not Working Outside	635	75.87
Agricultural Tech Training	Yes	237	28.32	Time at Home	6–12 months	753	89.96
	No	600	71.68		1–5 months	37	4.42
Education Years	6 years or less	315	37.63		None	47	5.62
	7–9 years	410	48.98	At home during busy farming season	Yes	790	94.38
	More than 9 years	112	13.38		No	47	5.62

), farming household characteristics (

Table A3. Characteristics of farming households.

Variable	Category (Persons)	Sample Size (Households)	Percentage (%)	Variable	Category (Persons)	Sample Size (Households)	Percentage (%)
Household Size	2≤	282	33.69	Youth and Labor Force	1≤	248	29.62
	3	187	22.34		2	221	26.40
	4	143	17.08		3	241	28.79
	5	144	17.20		4	112	13.38
	6≥	81	9.68		5≥	15	1.79
Non-Agricultural Income Share	0–10%	470	56.15	Number of Migrant Workers	None	399	47.67
	11–50%	224	26.76		1	314	37.51
	51–90%	127	15.17		2	100	11.95
	>90%	16	1.91		3≥	24	2.87

), and cross-analysis of family size and the number of migrant workers (

Table A4. Relationship Between Household Size and Migrant Worker Count.

		Farming	1 Migrant Worker	2 Migrant Workers	3 or More Migrant Workers	Total
Household Size	≤2	216	56	10	0	282
	3	100	59	22	6	187
	4	73	39	25	6	143
	≥5	107	63	43	12	225
	Total	496	217	100	24
Youth and Labor Force	≤1	195	43	10	0	248
	2	120	69	28	4	221
	3	121	73	33	14	241
	4	52	27	28	5	112
	≥5	8	5	1	1	15
	Total	496	217	100	24

), are detailed. Further, the analysis of family life cycle and factor allocation (

Table A5. Cross-analysis of Family Life Cycle and Factor Allocation.

	Growth Stage	Burden Stage	Decline Stage		Growth Stage	Burden Stage	Decline Stage
Households	312	114	411	Maize Acreage	84.33	68.70	47.13
Part-time Households	191	71	176	Total Income	180,624.5	138,977.7	105,856.5
Total Household Size	3.20	5.26	3.17	Non-agricultural Income	22,809	26,700	14,315
Youth and Labor Force	2.89	2.41	1.46	Degree of Part-time	23%	25%	16%
Number of Migrant Workers	0.64	0.78	0.50

), along with agricultural management characteristics (

Table A6. Characteristics of agriculture management.

Variable	Type	Sample Size (Households)	Percentage (%)	Variable	Type	Sample Size (Households)	Percentage (%)
Cultivated Land Area	10 acres or less	100	11.95	Fragmentation Level	5 plots or less	423	50.54
	10–30 acres	320	38.23		6–10 plots	274	32.74
	30–50 acres	170	20.31		11–15 plots	63	7.53
	50–100 acres	150	17.92		More than 15 plots	77	9.20
	More than 100 acres	97	11.59%	Machinery Ownership With machinery	None	507	60.57
Land Leveling Degree	Very Poor	9	1.08		With	330	39.43
	Poor	109	13.02	Agricultural Operation	Manual only	45	5.38
	Average	214	25.57		Manual + Machinery	370	44.21
	Good	333	39.78		Full machinery	422	50.42
	Very Good	172	20.55
Female Participation Rate	<50%	200	23.88
	50%	552	65.87
	>50%	85	10.15

), offers additional insights into the diversity and representativeness of the sample. The analysis of agricultural productive services purchase quantity and participation (

Table A7. Analysis of Agricultural Productive Services Purchase Quantity and Participation.

Variable	Option	Sample Size (Households)	Percentage (%)
Adopted	No	167	19.95
	Yes	670	80.05
Adopted Contiguous Services	No	574	66.45
	Yes	263	33.55
Number of Adopted Stages	0	167	19.95
	1	137	16.37
	2	131	15.65
	3	161	19.24
	4	236	28.20
	5	5	0.60
Adopted Stages	Land Preparation	528	63.08
	Seeding	430	51.37
	Spraying	343	40.98
	Topdressing	16	1.91
	Harvesting	534	63.80

The data source: Results are derived from the organization of survey data.

) complements the overall understanding of the farming households’ engagement with agricultural services.

3.2. Model Design

3.2.1. DEA-SBM Model

In the process of grain production, the use of agricultural chemicals such as pesticides and fertilizers, as well as fuel consumption from machinery, generates environmental pollution and leads to undesirable outputs. Traditional DEA and SFA models are limited in their ability to comprehensively evaluate farm-level GTFP, as they generally consider only desirable outputs. To address this limitation, the study adopts the DEA-SBM model with undesirable outputs, following the framework of Tone Kaoru [31] and Liu et al. [32]. This approach overcomes the shortcomings of conventional DEA and SFA models by simultaneously accounting for both desirable and undesirable outputs in measuring ecological sustainability and resource efficiency. The model is specified as follows:

$$\mathit{TE} = \mathit{min}{\displaystyle \frac{1-\frac{1}{K}{\sum }_{k=1}^K\frac{S_k^x}{x_{k0}}}{1 + \frac{1}{M+1}\left({\sum }_{m=1}^M\frac{S_m^y}{y_{m0}}+{\sum }_{i=1}^I\frac{s_i^u}{u_{i0}}\right)}}$$

(1)

$$s.t. {\sum }_{n=1}^N{z}_n{x}_{\mathrm{nk}}+s_k^x={ x}_{k0}, k=1, 2, ..., K$$

$${\sum }_{n=1}^N{z}_n{y}_{\mathrm{nm}}+{ s}_m^y={ y}_{m0}, m=1, 2, ..., M$$

$${\sum }_{n=1}^N{z}_n{u}_{\mathrm{ni}}+{ s}_i^{ u}=u_{i0}, I=1, 2, ..., I$$

$${\sum }_{n=1}^N{z}_{n }=1$$

$$z_n \ge 0,{ s}_k^{x }\ge 0,{ s}_m^y \ge 0,{ s}_i^u \ge 0$$

In the framework defined by Equation (1), N maize households are considered. Here, k denotes the specific production inputs used, while m and i signify the resulting beneficial outputs and environmental by-products, respectively. TE indicates the calculated GTFP score, which takes values in the range [0, 1]. ${\mathrm{x}}_{\mathrm{k}0 }$ represents the input of maize production, while $y_{m0}$ denotes the desirable output, and ${\mathrm{u}}_{\mathrm{i}0}$ represents the undesirable output associated with maize production. $z_n$ is the weight coefficient. The slack variables ${\mathrm{s}}_{\mathrm{k}}^{\mathrm{x}} \mathrm{and} {\mathrm{s}}_{\mathrm{i}}^{\mathrm{u}}$ capture input excess and output shortfall for grain production, while $s_m^y$ measures the surplus in undesirable outputs. $x_{\mathrm{nk}}, y_{\mathrm{nm}} \mathrm{and} u_{\mathrm{ni}}$ correspond to the actual levels of inputs, desirable outputs, and undesirable outputs in the production process of maize farmers. The DEA-SBM model incorporates the slack variables of inputs and outputs into the objective function, thereby enabling the measurement of GTFP.

3.2.2. Benchmark Regression Model (Tobit Model)

As the GTFP values measured by the DEA-SBM model are restricted to the range of 0–1 and exhibit right-censoring, the dependent variable is constrained. For such truncated regression models, direct estimation using ordinary least squares (OLS) would lead to biased results, as part of the dependent variable is compressed within the boundary. To address this issue, the study follows the approach of Zhou [33], who applies a Tobit regression model, which is appropriate for censored dependent variables, to examine the effect of household off-farm employment on farm-level GTFP. The model is specified as follows:

$${\mathit{gtfp}}_i = {\delta }_0 + {\delta }_{1i}{w}_i +\sum {\rho }_{\mathrm{ji}}{C}_{\mathrm{ij}} + {\epsilon }_i$$

(2)

where gtfp denotes the GTFP calculated in the previous section, w represents the degree of household off-farm employment, and $C_j$ denotes a set of control variables that may influence GTFP. The parameters $\delta \mathrm{and} \rho$ represent the coefficients to be estimated, while $\epsilon$ is the random disturbance term.

3.2.3. IPW Model

The analysis applies the IPW model proposed by Cattaneo [34]. A multiple treatment effects framework is adopted, where the multinomial logit (Mlogit) model is used to estimate the propensity scores of the multi-valued treatment variable. Based on these estimates, the model evaluates and compares the heterogeneous impacts of three types of off-farm employment (single off-farm employment, dual off-farm employment, and non-farm employment) on household GTFP. The estimation procedure follows Cattaneo [34], in which the propensity scores $\widehat{\pi }$ are obtained from the Mlogit model. The average treatment effect under the IPW framework is given by:

$${\widehat{\mathit{ATE}}}_{\mathit{IPW}} = {\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left\{{\displaystyle \frac{T_i{Y}_i}{\widehat{\pi }\left(X_i\right)}}-{\displaystyle \frac{{(1 - T}_i)Y_i}{1 - \widehat{\pi }\left(X_i\right)}}\right\}$$

(3)

3.3. Variable Definition

3.3.1. Dependent Variable

GTFP: In line with Tone Kaoru [31] and Liu et al. [32], production inputs were defined to include capital, labor, and land. Total agricultural output was treated as the desirable output, while carbon emissions and related non-point source pollution generated during production were classified as undesirable outputs. GTFP was estimated using the DEA-SBM model, which allows for a comprehensive assessment of production efficiency by simultaneously accounting for desirable and undesirable outputs.

3.3.2. Core Explanatory Variable

Our analysis treats the household as the primary unit. Families earning income from both crop farming and external work are identified as off-farm households. To quantify this, we use the ratio of non-agricultural wages to total family income (ranging from 0 to 1) as an indicator of livelihood diversification. Following the framework established by the Rural Development Institute of CASS and prior research, we segment households into four distinct categories. These include purely agricultural, primarily agricultural part-time, primarily non-agricultural part-time, and full non-farm households, based on non-farm income thresholds of 10%, 50%, and 90%, respectively.

3.3.3. Control Variables

Control variables were grouped into four categories: household head characteristics, household characteristics, farm operation characteristics, and external environment characteristics. Household head characteristics included age, education, health status, and participation in technical training. Household characteristics included the share of female labor in farming, the proportion of elderly members, and household indebtedness. Farm operation characteristics covered cultivated land area, land quality, and agricultural machinery investment. External environment characteristics included exposure to natural disasters, government support, and village location. Definitions and assignments of these variables are presented in

Table 1. Descriptive Statistics for Variables.

Variable Type	Variable Name	Variables	Obs	Mean	S.D.	Min	Max
Dependent variable	GTFP	GTFP	837	0.723	0.110	0.34	1
Explanatory variable	Off-farm employment	w	837	0.199	0.265	0.00	0.99
Household head characteristics	Age	Age	837	57.779	10.020	27.00	87
	Education	Edu	837	7.853	2.921	0.00	18
	Health status	Health	837	3.933	1.089	1.00	5
	Technical training	Tech	837	0.283	0.451	0.00	1
Household characteristics	Elderly ratio	Elder	837	0.354	0.364	0.00	1
	Female labor ratio	Female	837	0.313	0.190	0.00	1
	Household debt	Debt	837	0.302	0.460	0.00	1
Farm operation characteristics	Land area	Area	837	63.935	114.974	1.00	1000
	Land quality	Land	837	3.657	0.981	1.00	5
	Machinery assets	Mac	837	0.394	0.489	0.00	1
	Production services	Pro	837	0.800	0.400	0.00	1
External environment	Natural disasters	Nature	837	0.826	0.380	0.00	1
	Policy support	Policy	837	0.233	0.423	0.00	1
	Location	Region	837	6.138	4.829	0.01	40

Source: Authors’ survey data.

, based on relevant studies [12,16].

4. Empirical Results and Analysis

4.1. Baseline Regression Results of Off-Farm Employment on GTFP

The regression results in

Table 2. Benchmark Regression Results of Part-time Farming on GTFP.

Variable	(1)	(2)	(3)	(4)	(5)
	GTFP	GTFP	GTFP	GTFP	GTFP
Off-farm employment	−0.030 * (0.014)	−0.029 * (0.014)	−0.032 ** (0.014)	−0.036 ** (0.015)	−0.035 ** (0.015)
Age		0.002 *** (0.000)	0.002 *** (0.000)	0.002 *** (0.000)	0.002 *** (0.000)
Edu		0.003 ** (0.001)	0.004 *** (0.001)	0.004 ** (0.001)	0.003 ** (0.001)
Health		0.006 * (0.004)	0.007 * (0.004)	0.005 (0.004)	0.005 (0.004)
Tech		−0.002 (0.009)	−0.005 (0.009)	−0.004 (0.009)	−0.005 (0.009)
Elder			−0.015 (0.005)	−0.023 * (0.005)	−0.023 * (0.005)
Female			−0.014 (0.021)	−0.001 (0.020)	−0.001 (0.020)
Debt			0.006 (0.008)	0.004 (0.008)	0.004 (0.008)
Area				−0.000 (0.000)	−0.000 (0.000)
Land				0.014 *** (0.004)	0.014 *** (0.004)
Mac				−0.026 *** (0.008)	−0.026 *** (0.008)
Pro				0.006 (0.010)	0.008 (0.010)
Nature					−0.024 *** (0.010)
Policy					0.006 (0.009)
Region					−0.000 (0.001)
Province FE					YES
Constant	0.729 *** (0.007)	0.592 *** (0.034)	0.578 *** (0.037)	0.522 *** (0.038)	0.542 *** (0.039)
Observations	837	837	837	837	837
log likelihood	670.00	670.00	671.30	709.40	712.72

Note: Robust standard errors are reported in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

demonstrate that off-farm employment exerts a significant negative effect on GTFP. Even after the sequential inclusion of household head characteristics, household characteristics, farm operation characteristics, and external environment variables, while controlling for provincial and regional effects, the negative association remains robust, with a coefficient of −0.035. This outcome validates Hypothesis 1, indicating that off-farm employment disperses household production effort, leading to insufficient agricultural inputs or inefficient resource allocation. This result quantifies the critical trade-off between rural socio-economic livelihood diversification and the environmental sustainability of agricultural production. In terms of household head characteristics, GTFP rises with both age and education level. The increase may reflect the greater agricultural experience accumulated by older farmers, as well as the ability of better-educated farmers to acquire and apply agricultural information effectively. These attributes contribute to more efficient and environmentally friendly production decisions that enhance GTFP. With respect to household characteristics, a higher proportion of elderly members corresponds to greater GTFP losses. This effect may stem from the relatively limited physical capacity and production ability of older household members, together with their need for care, which diverts labor and time from farming activities. The result highlights efficiency losses linked to the structural labor shortages created by rural aging, posing a challenge to the social sustainability of agricultural systems. From the perspective of farm operation, households with higher land quality achieve superior GTFP [35], as fertile land improves per-unit productivity and overall efficiency.

Conversely, reliance on self-owned machinery significantly reduces GTFP. This finding may seem counterintuitive given the general benefits of mechanization, but it can be explained by several key economic factors. Agricultural machinery, particularly for smallholder farmers, is often characterized by high asset specificity and significant fixed costs, such as depreciation, maintenance, and storage. For farmers with limited land, the “sunk costs” of purchasing machinery cannot be effectively distributed across a large enough area, leading to a high average cost per unit of output and lowering overall economic efficiency. Additionally, part-time farmers face the issue of “unused capacity,” as the seasonality of agricultural production combined with the rigid time constraints of off-farm employment leaves machinery idle for extended periods. This low utilization means that the capital input is not being converted into productive output, resulting in a waste of resources. Furthermore, land fragmentation in Northeast China exacerbates this inefficiency. Despite relatively flat terrain, household plots are often scattered, requiring the movement of machinery between fragmented areas. This process consumes additional fuel and time without contributing to crop growth, increasing undesirable outputs like emissions and ultimately lowering the green efficiency of the farming system.

Finally, among external environment variables, natural disasters exert a significant negative impact on production efficiency at the 1% level, highlighting their substantial effect on agricultural production, thus undermining the resilience of the farming system.

4.2. Robustness Tests

Several robustness checks were performed by altering the regression model, the measurement of the dependent variable, the calculation method, and the definition of the explanatory variable (

Table 3. Robustness Test.

Variable	(1)	(2)	(3)	(4)
	OLS Regression	Dependent Variable: gtfp1	Dependent Variable: gtfp2	Alternative Explanatory Variable
Off-farm employment	−0.035 ** (0.018)	−0.059 *** (0.017)	−0.042 *** (0.014)
Part-time type				−0.012 ** (0.005)
Control variables	Yes	Yes	Yes	Yes
Number of observations	837	837	837	837

Note: Robust standard errors are reported in parentheses. ** and *** denote significance at the 5% and 1% levels, respectively.

). In the first specification, since GTFP efficiency values are continuously distributed between 0 and 1, the ordinary least squares (OLS)model was applied to re-estimate the relationship between off-farm employment and GTFP. In the second specification, the undesirable output indicator was replaced with non-point source pollution from fertilizer use, producing an alternative measure of GTFP (gtfp1). In the third specification, capital inputs were separated into machinery investment and production material investment to construct another GTFP measure (gtfp2), which was then re-estimated. In the fourth specification, the explanatory variable was replaced with household off-farm employment type, categorizing households into pure agricultural, single part-time, dual part-time, and non-agricultural groups. Across all specifications, the negative effect of off-farm employment on GTFP remained robust, thereby confirming the reliability of the baseline results and underscoring the persistence of the socio-economic factor (off-farm labor) as a driver of diminished ecological efficiency.

In addition, the IPW method was employed to estimate the average treatment effects of different household off-farm employment types, using pure agricultural households as the reference group. The results passed the overlap test. As shown in

Table 4. Heterogeneous Effects of Farmer Part-time Operations on GTFP.

Variable	Off-Farm Employment Type	GTFP
		ATE/ATT	Standard Error
IPW	Single part-time	−0.006	0.009
	Dual part-time	−0.225 ***	0.131
	Non-agricultural	−0.241 ***	0.045

Note: *** denote significance at the 1% level.

, a greater part-time engagement is associated with larger GTFP losses. The effect of single part-time households is not statistically significant because agricultural income still dominates in these households, leading to greater attention and reliance on farming. When making decisions, such households continue to prioritize agricultural input allocation, thereby avoiding substantial GTFP losses. In contrast, dual-part-time and non-agricultural households place agriculture in a “low-priority” position, tending to adopt less efficient extensive farming practices, which result in significant GTFP losses. The quantified losses (ATT of −0.225 and −0.241 for dual-part-time and non-agricultural households, respectively) demonstrate the acute environmental cost associated with advanced socio-economic livelihood diversification.

The statistical dominance of the negative effect in dual-part-time and non-agricultural groups can be explained by several interconnected mechanisms. The high opportunity cost of labor leads to a “siphoning effect,” where non-agricultural income becomes the primary livelihood source, prompting the allocation of high-quality labor almost exclusively to off-farm sectors. As a result, agricultural production is left to elderly members, who have lower physical and cognitive capacity to adopt green technologies. Additionally, these households often adopt a “short-termism” approach to land management, perceiving land more as a safety net than as a productive asset. This mindset discourages investment in long-term soil health, such as the use of organic fertilizers, and instead leads to a reliance on excessive chemical fertilizers to maintain basic yields with minimal effort. Furthermore, a “moral hazard” arises from the principal–agent relationship in outsourced services. Because these households are frequently absent from the village, they are unable to effectively oversee service providers. As a result, providers may engage in opportunistic behaviors—such as shallow plowing or uneven spraying—that reduce technical efficiency and worsen environmental pollution.

The regression results in the previous text indicate that, on the whole, off-farm employment hurts GTFP, but there may be self-selection bias or bidirectional causal relationships among variables, leading to deviations in the estimation results. Therefore, this paper further employs the 2SLS instrumental variable method to explore the relationship between off-farm employment and GTFP. In existing studies, scholars mostly use the non-agricultural employment network as an instrumental variable for farmers’ part-time operations to address endogeneity issues [33], The reason is that instrumental variables are highly correlated with endogenous variables and have no direct relationship with the explained variables [27], when there are many farmers around adopting this business model and the market is relatively active, farmers will exhibit following behavior. This behavior does not directly influence farmers’ GTFP, which aligns with the principles of selecting instrumental variables. Therefore, this study uses the proportion of non-agricultural income from other households in the village, excluding the one under study, as an instrumental variable for farmers’ part-time business activities. The instrumental variable passed both the LM test and the Wald-F test, confirming its suitability. The regression results in

Table 5. Two Stage Least Squares Estimation Results.

Variable	First Stage	Second Stage
Off-farm employment		−0.191 *** (0.058)
Instrumental variable (non-agricultural income share)	0.574 *** (0.0743)
Control variables	Controlled
LM statistic	40.491
Wald-F statistic	84.963
KP Wald-F statistic	59.697
Sample size	837

Note: Robust standard errors are reported in parentheses. *** denote significance at the 1% level.

show that the negative impact of off-farm employment on GTFP remains significant at the 1% level. Even after controlling for endogeneity, this continued negative relationship supports the conclusion that off-farm employment, as a social driver, imposes a real and measurable ecological cost on the agricultural system.

4.3. Mechanism Analysis

The previous analysis confirms that household off-farm employment leads to GTFP losses. To provide a deeper understanding of the transmission mechanism, this study further investigates the impact of off-farm employment on the allocation of input factors, specifically labor, land, and capital. The “Factor Surplus Rate”—derived from the input slack variables in the DEA-SBM model—serves as a core indicator of resource allocation efficiency. It accurately captures issues such as factor idleness and inefficient utilization caused by part-time operations. Variations in the surplus rate directly reflect the transmission effect of factor distortion on GTFP.

As shown in

Table 6. Empirical Results on the Impact of Off-Farm Employment on Factor Allocation.

	(1) Labor Surplus Rate	(2) Land Surplus Rate	(3) Capital Surplus Rate
Part-time Degree	−0.094 ***	0.145 ***	0.013 **
	(0.024)	(0.021)	(0.008)
Control Variables	Yes	Yes	Yes
Sample Size	837	837	837

Note: Robust standard errors are reported in parentheses. ** and *** denote significance at the 5% and 1% levels, respectively.

, off-farm employment has a significant positive impact on the surplus rates of land and capital, while significantly reducing the labor surplus rate. This indicates that off-farm employment exacerbates the redundancy of land and capital inputs. These redundant inputs lead to lower land utilization efficiency and extensive capital configuration, resulting in insufficient overall output and increased emissions, thereby negatively affecting GTFP and intensifying factor misallocation. Conversely, due to the labor-draining nature of off-farm employment, the redundancy of household labor decreases; however, this is offset by the substitution of capital and the inability to maintain optimal yield levels.

The analysis above establishes that off-farm employment increases land and capital redundancy. To further pinpoint the source of capital redundancy—whether it stems from machinery investment (substituting for labor) or production materials (e.g., fertilizers and pesticides)—we performed a further test using the specific slack variables for machinery and materials (

Table 7. Empirical Results on the Impact of Off-Farm Employment on Detailed Capital.

	(1) Labor Surplus Rate	(2) Land Surplus Rate	(3) Machinery Input Surplus Rate	(4) Production Material Input Surplus Rate
Part-time Degree	−0.016	0.128 ***	0.066 ***	−0.004
	(0.020)	(0.039)	(0.019)	(0.008)
Control Variables	Yes	Yes	Yes	Yes
Sample Size	837	837	837	837

Note: Robust standard errors are reported in parentheses. *** denote significance at the 1% level.

).

The results indicate that the capital redundancy is primarily driven by the surplus of machinery input (Coefficient = 0.066, significant at 1%), while the impact on production material surplus is not significant. This finding aligns with the theoretical analysis: part-time farmers, facing time constraints, tend to over-invest in machinery or purchase mismatched machinery services to replace labor. In contrast, the application of materials like fertilizers is more influenced by farmers’ habits or peer effects rather than by their off-farm employment status directly. Therefore, the mechanism through which off-farm employment suppresses GTFP is primarily the “siphoning effect” on labor, which triggers a compensatory but inefficient “redundancy effect” in land and machinery inputs.

5. Extended Discussion and Analysis

5.1. Heterogeneity Analysis

5.1.1. Household Development Stage Perspective

Household life cycle theory suggests that farm households at different development stages differ in member composition, livelihood objectives, and resource demands, which may lead to heterogeneous patterns of resource allocation in agricultural production. In line with this framework, households were divided into three stages: growth, burden, and decline. Growth-stage households are typically composed of young couples with children or relatively small family units. Burden-stage households are characterized by the dual responsibility of raising children and supporting elderly members. Decline-stage households generally consist of young couples living with elderly parents or families in which all members are over 60 years old. The empirical results demonstrate clear heterogeneity in the effect of off-farm employment on GTFP across these stages (

Table 8. Heterogeneous Effects of Family Life Cycle on GTFP.

Variable	(1)	(2)	(3)
	Growth Stage	Burden Stage	Decline Stage
Off-farm employment degree	−0.075 *** (0.025)	0.063 ** (0.030)	−0.039 * (0.003)
Control variables	Controlled	Controlled	Controlled
Number of observations	312	114	411

Note: Robust standard errors are reported in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

). Before the burden stage, the negative effect of off-farm employment on GTFP gradually weakens, and during the burden stage, off-farm employment even exerts a positive effect on GTFP. However, once households transition into the decline stage, the negative impact of off-farm employment intensifies. A plausible explanation is that during the burden stage, household labor supply expands, mitigating the risk of agricultural underinvestment caused by off-farm employment. Furthermore, elderly members, though less competitive in external labor markets, often retain sufficient physical capacity to meet basic agricultural labor requirements. Critically, their contribution is amplified by two key mechanisms: the “experience dividend” and “flexible scheduling.” Senior members bring with them a wealth of tacit knowledge about local soil variability and microclimates, accumulated over years of farming. This deep understanding allows them to implement precise field management strategies, such as timing fertilizer application to align with rainfall patterns, which helps minimize chemical waste. In addition, unlike younger workers who are often tied to rigid industrial schedules, elderly members face low opportunity costs and have flexible time availability. This enables them to devote more time to labor-intensive yet environmentally friendly practices, such as manual weeding or multiple low-dose fertilizations, which serve as sustainable alternatives to the heavy chemical inputs often relied upon by time-constrained off-farm workers. This positive effect reveals a crucial pathway where socio-economic diversification (off-farm income) successfully translates into enhanced agro-ecological efficiency, achieving a temporary synergy between livelihood goals and environmental sustainability. By contrast, in the decline stage (including both support and empty-nest phases), the primary labor force shifts from prime-age adults to elderly members, and household size diminishes. As high-quality labor is increasingly absorbed into non-agricultural employment, agricultural practices tend to become extensive and reliant on traditional, low-efficiency methods, ultimately leading to a decline in GTFP. This decline highlights the deep challenge rural aging poses to sustainable resource management.

5.1.2. Land Operation Scale Perspective

Threshold effect analysis of land operation scale identifies a double-threshold relationship between off-farm employment and GTFP (

Table 9. Heterogeneous Effects of Land Scale on GTFP.

Variable	(1)
Land scale ≤ 6 mu	0.0423 *** (4.40)
6 mu < Land scale ≤ 25 mu	−0.0108 ** (−2.22)
Land scale > 25 mu	−0.0328 *** (−5.72)
Control variables	Controlled
Sample size	837

Note: Robust standard errors are reported in parentheses. ** and *** denote significance at the 5% and 1% levels, respectively.

). To understand the economic implications of these specific limits, it is necessary to relate them to the mechanization levels and management capacities typical of Northeast China. When farmland exceeds 6 mu, the negative effect of off-farm employment on GTFP intensifies as scale expands, with efficiency losses becoming more pronounced once the scale surpasses 25 mu. Economically, the 6 mu threshold roughly corresponds to the per capita arable land area in the region, representing the “subsistence line.” Below this limit, farming is a “garden-style” activity easily managed by household members in their spare time using small tools. However, for households operating between 6 and 25 mu, a “mechanization mismatch” occurs. This scale is too large for efficient manual or small-machine management by part-time workers, yet too small to justify the high fixed costs of large-scale, high-efficiency machinery. Consequently, these households fall into an “investment trap,” facing high costs and low efficiency. For households operating between 6 and 25 mu, the limit reflects the “management ceiling” of a typical dual-role household. the underlying reason likely lies in “diseconomies of scale,” where the marginal cost of mechanization exceeds the marginal return. Off-farm employment diverts labor away from agriculture, aggravating extensive rather than intensive cultivation and thereby reducing GTFP. For households cultivating more than 25 mu, the higher specialization required for large-scale production cannot be met under part-time conditions. Managing over 25 mu requires strict adherence to agricultural timing (e.g., precise planting and harvesting windows). Limited capacity to manage extensive farmland leads to reliance on low-efficiency practices and resource misallocation, which further undermines GTFP. This dual-threshold pattern strongly suggests that unmanaged scale expansion, driven by socio-economic factors like labor outflow, can exacerbate environmental efficiency losses, thereby threatening the principle of sustainable land management. A different pattern emerges among small-scale households cultivating less than or equal to 6 mu. Agricultural production at this scale primarily aims at subsistence, and available household labor is typically sufficient to meet basic production needs even when high-quality members participate in off-farm work. Household income in these cases depends largely on wage earnings, and off-farm employment intensity is relatively high. Furthermore, such households often reinvest part of their non-agricultural income into agricultural activities and green technology adoption, thereby improving GTFP.

5.2. Analysis of Productive Service Effects

According to the theory of agricultural division of labor and existing studies [24,36], agricultural production services may mitigate the negative effect of off-farm employment on GTFP through three potential mechanisms: labor substitution, technological synergy, and economies of scale. These mechanisms do not necessarily occur simultaneously. Labor substitution serves as the fundamental function of service adoption, whereas technological synergy and economies of scale require systematic integration of multi-stage service models or spatial contiguity of farmland operations. Based on field survey evidence, households were categorized into four groups according to their production modes: manual operations, self-owned machinery operations, adoption of single-stage services, and adoption of multi-stage services. To better understand the operational differences, ‘single-stage service’ typically refers to the ad hoc outsourcing of a discrete production task, most commonly mechanical harvesting. In this model, the farmer retains control over other inputs, often leading to a mismatch between planting density and harvester specifications. In contrast, ‘multi-stage service’ involves a bundled contract where a provider handles a sequence of related tasks (e.g., deep plowing followed by precision sowing and fertilization). This integration ensures technical consistency across stages, reducing resource waste. The analysis examined the effect of off-farm employment on GTFP across these groups and explored the role of production services in generating labor substitution and technological synergy. The results in

Table 10. The Results of Technological Synergy Effects in Producer Services.

Variable	Manual Operations	Self-Owned Machinery Operations	Adoption of Single-Stage Services	Adoption of Multi-Stage Services
Off-farm employment	−0.119 ** (0.051)	−0.112 * (0.060)	−0.064 * (0.038)	−0.017 (0.018)
Control variables	Controlled	Controlled	Controlled	Controlled
Number of observations	45	122	137	533

Note: Robust standard errors are reported in parentheses. * and ** denote significance at the 10% and 5% levels, respectively.

show that in Northeast China, where mechanization levels are already relatively high, adopting production services can alleviate part of the GTFP loss caused by off-farm employment, although the effect remains limited. By contrast, households adopting multi-stage services achieve a deeper division of labor across production links, which generates stronger technological synergy and significantly offsets the negative impact of off-farm employment on GTFP. This finding underscores the role of multi-stage services as a critical infrastructure for fostering green technological diffusion and achieving system-level sustainability (efficiency and reduced environmental externalities) within smallholder agriculture.

In assessing the labor substitution effect and economies of scale from production services, households were divided into four categories: manual operations, self-owned machinery operations, adoption of services without contiguous land use, and adoption of contiguous services. Contiguous operation in this context entails a ‘virtual scale’ arrangement. Unlike traditional land consolidation, which requires transferring ownership rights, contiguous services involve multiple smallholders in adjacent plots agreeing to unified mechanical operations. Service providers remove temporary boundary ridges and apply uniform planting standards (same variety, same timing) across the combined block, thereby eliminating the inefficiencies of frequent machinery turning and edge effects. The findings in

Table 11. The Results of Scale Effects in Producer Services.

Variable	Manual Operations	Self-Owned Machinery Operations	Service Adoption Without Contiguous Operations	Service Adoption with Contiguous Operations
Off-farm employment	−0.119 ** (0.051)	−0.112 * (0.060)	−0.051 ** (0.024)	0.003 (0.025)
Control variables	Controlled	Controlled	Controlled	Controlled
Sample size	45	122	407	263

Note: Robust standard errors are reported in parentheses. * and ** denote significance at the 10% and 5% levels, respectively.

suggest that when households adopt contiguous services, which facilitate large-scale and spatially integrated operations beyond original land boundaries, the GTFP losses associated with off-farm employment are substantially alleviated. Under this service mode, the coefficient of off-farm employment on GTFP even becomes positive. This achievement of ecological efficiency is attributed to the realization of virtual scale economies, which successfully minimizes resource waste and energy consumption typically associated with fragmented, part-time farming, providing a concrete policy pathway towards sustainable food systems.

The results indicate that in regions with a high level of mechanization, such as Northeast China, the labor substitution advantage generated solely by production services remains limited. In contrast, contiguous services and multi-stage services enhance GTFP through indirect scale effects and specialized production pathways. These service models have emerged as the primary mechanisms for mitigating the GTFP losses caused by off-farm employment and represent promising directions for future development, aligning economic growth with environmental conservation.

6. Research Conclusions and Policy Implications

6.1. Research Conclusions

First, the overall evidence demonstrates that off-farm employment reduces GTFP, with households relying more heavily on non-agricultural income (dual part-time households and non-agricultural households) experiencing greater losses. This result quantifies the conflict between household socio-economic development (livelihood diversification) and ecological sustainability (GTFP). In Northeast China, agricultural production remains dominated by small-scale, fragmented household operations. This structure limits the efficient allocation of agricultural resources and, compounded by the aging and weakening of remaining farm operators, which raises concerns about social sustainability, hinders the adoption of advanced agricultural technologies, thereby constraining improvements in GTFP.

Second, the mechanism analysis shows that GTFP losses from off-farm employment are primarily driven by redundant land and capital inputs. Specifically, larger-scale households require stable and sustained labor input, making them more vulnerable to the inefficiencies of extensive part-time operations. Although these households often reduce land inputs to smaller or contracted plots, the scale still approaches the second threshold of 25 mu, beyond which efficiency losses intensify, reinforcing the negative effect on GTFP. Redundant capital input arises mainly from mechanical investment intended to substitute for labor. While mechanization helps address labor shortages, dual part-time and non-agricultural households often purchase labor-intensive but costly machinery services. Weak field management further reduces their ability to respond to natural disasters, amplifying GTFP losses and posing a threat to the environmental integrity of the production system.

Third, the effect of off-farm employment on GTFP differs across household types. From a household life cycle perspective, productivity losses are not uniform. Households in the burden stage, supported by relatively strong labor quality and elderly members who retain productive capacity, can simultaneously meet childcare needs and contribute agricultural experience to improve production efficiency. This reveals a pathway where socio-economic stability can create a synergy with agro-ecological efficiency. In contrast, households in the growth and decline stages suffer varying degrees of GTFP loss due to part-time operations. From a land scale perspective, small-scale households (≤6 mu), whose production is largely subsistence-oriented, do not incur GTFP losses from off-farm employment. However, larger-scale households remain more exposed to efficiency losses when engaged in part-time operations.

Fourth, in highly mechanized regions such as Northeast China, the labor substitution effect from adopting production services alone is limited. In contrast, contiguous services that expand operational scale and multi-stage services that deepen specialization and create technological synergy prove more effective in mitigating the adverse impact of off-farm employment on GTFP.

In summary, this study distinguishes itself from existing macro-level literature by deepening the micro-level understanding of the relationship between off-farm employment and agricultural GTFP. Unlike previous studies that often treat farm households as a homogeneous group, this work provides a novel contribution by revealing that the environmental cost of labor migration is not uniform but is strictly contingent on the household’s developmental stage and the mode of service adoption. By integrating the perspectives of the family life cycle and socialized service heterogeneity, we deconstruct the ‘black box’ of efficiency loss, demonstrating that while off-farm work generally strains green efficiency, specific combinations of ‘burden-stage’ labor management and ‘multi-stage’ technical services can convert this socio-economic shift into a sustainable synergy.

6.2. Policy Implications

First, land tenure reform should be deepened and rural land contract relations stabilized to guide part-time households toward moderate-scale operations. Accelerated confirmation, registration, and certification of land contract rights can clarify ownership and fully safeguard farmers’ rights of disposal and income, enabling part-time households to release part or all of their production rights. Improving land transfer registration and supervision systems can further enhance market transparency and efficiency. For households with a high degree of off-farm employment, policy should encourage a transition toward agricultural specialization or promote land transfer to pure farmers, professional farmers, family farms, or cooperatives. At the same time, green production constraints should be incorporated into scale expansion to prevent excessive enlargement and ensure that land transfer contributes to long-term sustainability and ecological preservation.

Second, the agricultural service system should be optimized, and contiguous trusteeship and whole-industry-chain service models should be fostered to strengthen the connecting role of new agricultural entities. Closer integration between land transfer markets and agricultural services can promote specialization and intensification. Capacity building for service providers should be reinforced, and service organizations should be encouraged to extend into pre-production, production, and post-production stages, thereby improving both the scope and quality of supply. Farmers should be guided to cooperate and adopt contiguous and multi-stage services, reducing purchase costs, expanding service coverage, and generating scale and agglomeration effects that alleviate factor misallocation.

Developing Social Security and Green Incentives Based on Family Life Cycle (Targeting Vulnerable Groups): Ensuring social equity and ecological security for elderly farming households: For households in the decline stage, rural pension systems should be integrated with land tenure systems to encourage complete land trusteeship or withdrawal. The government can provide ecological compensation payments to support these households in entrusting their land to professional green service organizations, thereby ensuring sustainable land resource management while safeguarding the social welfare of elderly farmers. Incentivizing green behavior among households in the “burden stage”: For burden-stage households with labor advantages, green technology training and certification should be provided to encourage the adoption of low-carbon farming practices. For example, specialized financial credit support can be offered to incentivize them to become new-type green professional farmers.

Fourth, Strengthening Green Technology Promotion and Policy Coordination (Targeting System Resilience). First, establish a collaborative mechanism for green technology promotion. When promoting agricultural technologies, technical training must be closely integrated with socialized service systems to ensure that the diffusion of green technologies no longer relies on the learning capacity of aging farmers, but rather is achieved through efficient and standardized applications by professional service organizations. Second, enhance the environmental resilience of agricultural systems. Given the significant negative impact of natural disasters on GTFP, policies should increase fiscal investment in agricultural green insurance, particularly in specialized insurance products related to climate change and black soil conservation, to enhance the sustainable resilience of agricultural systems in the face of environmental shocks and boost market confidence.