The theory of factor substitution posits that the migration of rural labor from agricultural production to non-farm sectors inevitably induces agricultural labor shortages, thereby adversely affecting production outcomes (Xie and Lu, 2017; Zhang et al., 2020; Zhang et al., 2021). As boundedly rational decision-making agents, farmers typically respond to this challenge by adopting modern agricultural technologies to maintain production stability. Specifically, they tend to intensify inputs of modern agricultural elements such as chemical fertilizers and pesticides to compensate for productivity losses caused by labor scarcity (Shi et al., 2016). Empirical observations reveal that migrant workers predominantly consist of younger, better-educated individuals, leaving behind an aging agricultural workforce characterized by limited awareness of ecological conservation and persistent reliance on extensive farming practices (Ashley et al., 2018). This demographic shift creates a propensity for excessive agrochemical application on remaining cultivated land, resulting in dual consequences of diminished production efficiency and environmental degradation. Simultaneously, rural labor outflow exerts upward pressure on agricultural wages through labor market tightening (Gallardo and Sauer, 2018). The resultant changes in relative factor prices create economic incentives for farmers to substitute relatively abundant production factors (e.g., chemical inputs and machinery) for scarce labor resources (Nguyen et al., 2019; Su et al., 2016). This substitution behavior manifests in increased procurement of yield-enhancing inputs including pesticides, fertilizers, and mechanization services (Guo et al., 2022), ostensibly aimed at improving production efficiency and maximizing agricultural income. However, such compensatory strategies carry significant environmental externalities. The persistent intensification of chemical inputs, while potentially effective in the short term, tends to culminate in the chronic over-application of agrochemicals. This practice constitutes a primary driver of ANPP, presenting critical challenges to sustainable agricultural development. Based on this theoretical framework and empirical evidence, Hypothesis 1 (H1) is formulated.

Rural labor transfer represents a significant phenomenon in the industrialization and urbanization processes of developing nations, particularly in China, where it exerts profound impacts on rural economic structures, farmers’ income sources, and agricultural production patterns. From an economic perspective, rural labor transfer serves as a crucial mechanism for farmers to obtain non-farm income and enhance household earnings (Caulfield et al., 2019). The substantial increase in non-farm income has led to notable changes in farmers’ production decisions and resource allocation, subsequently influencing ANPP. The primary economic effect of transfer manifests in the significant augmentation of non-farm income. According to data from China’s National Bureau of Statistics, wage income constituted over 42% of farmers’ household income in 2022, emerging as the principal driver of income growth. This diversification of income sources reduces reliance on singular agricultural income and enhances household economic resilience. Concurrently, it alleviates financial constraints in agricultural production, enabling farmers to invest more in chemical inputs such as pesticides and fertilizers, thereby compensating for labor shortages resulting from migration (Li et al., 2022; Luan et al., 2016). Furthermore, the increase in non-farm income has enhanced farmers’ risk-bearing capacity, prompting a shift in cultivation patterns from traditional food crops to higher-value cash crops that require intensive chemical inputs (e.g., vegetables and fruits) (Luo, 2017; Yang et al., 2025). This transition has led to increased utilization of chemical substances in agricultural production. Moreover, the growth of non-farm income has facilitated the transformation of agricultural production from labor-intensive to capital-intensive modes. Farmers increasingly rely on agricultural mechanization and chemical inputs to enhance production efficiency, rather than traditional labor-intensive methods. While this transformation has improved agricultural productivity, it has simultaneously exacerbated ANPP through the intensified use of modern machinery and chemical inputs (Wu and Liu, 2017; Shao and Li, 2020). As a result, rural labor transfer, by increasing non-farm income, alters farmers’ production decisions and resource allocation patterns, ultimately contributing to the aggravation of ANPP. Hence, Hypothesis 2 (H2) is proposed.

Agricultural socialized services refer to a series of comprehensive services provided by multiple entities such as government public service institutions, enterprises, and social organizations for agricultural production and operation entities, covering the entire process from pre-production, production to post-production of agriculture. The rise in non-farm income enhances the overall household income level of farmers, enabling them to access diversified agricultural socialized services (Li et al., 2023; Yang et al., 2020). This development partially mitigates the issue of farmers relying exclusively on chemical inputs such as pesticides and fertilizers to compensate for declining agricultural productivity caused by labor shortages. In practice, various aspects of agricultural production can be outsourced to specialized service providers (Zhang and Luo, 2019), leading to a scenario where “professionals handle professional tasks.” The procurement of agricultural socialized services offers significant cost advantages compared to the direct purchase of chemical inputs like pesticides and fertilizers. Consequently, driven by cost efficiency and higher profitability, households primarily dependent on non-farm income are more inclined to delegate the pre-production, production, and post-production stages of agriculture to third-party service organizations. Service providers, in delivering these services, tend to reduce the usage of chemicals such as pesticides and fertilizers to minimize input costs and maximize profits (Yang et al., 2020), thereby contributing to the reduction of ANPP. Additionally, service providers possess the advantage of accessing cutting-edge technology and market data. They can disseminate application standards, environmental awareness, and efficient pesticide and fertilizer application techniques to farmers through their services. This transfer of knowledge and skills has enhanced farmers’ understanding and proficiency in the scientific application of pesticides and fertilizers, transforming their previous over-reliance on chemical inputs. Thus, the ANPP resulting from increased non-farm income can be alleviated. Based on this reasoning, Hypothesis 3 (H3) is proposed.

Rural labor transfer fundamentally represents a reallocation of household labor resources (Zhang et al., 2021). In the initial stages of rural labor transfer, when the scale remains limited, its impact on household labor division and income generation is relatively marginal, and the continuity of agricultural production can be maintained. However, as the magnitude of rural labor transfer expands and the depletion of household labor resources intensifies, significant transformations occur in agricultural production patterns. This transition manifests as a shift from labor-intensive agricultural practices to modern production factor-oriented approaches. Concurrently, agricultural producers increasingly resort to intensive application of chemical inputs, particularly pesticides and fertilizers, to ensure production stability and maintain income levels (Li et al., 2023). This practice has been empirically demonstrated to exacerbate ANPP (Hou and Yao, 2019). Nevertheless, when rural labor transfer reaches its saturation point or exceeds optimal thresholds, the agricultural production system stabilizes at a new equilibrium. Within this context, the marginal returns from chemical input-intensive production structures exhibit diminishing trends, while environmental degradation intensifies, creating incentives for behavioral change among agricultural producers. In addition, the emergence of green production technologies and agricultural socialized service organizations has fundamentally altered traditional production paradigms. These developments have provided viable alternatives to chemical input-dependent production methods. Therefore, under the dual influence of national policies promoting chemical input reduction and the availability of sustainable alternatives, agricultural producers demonstrate increased willingness to adopt socialized agricultural services and green production technologies. This behavioral shift, motivated by rational economic considerations, facilitates the transition toward sustainable agricultural practices, thereby creating opportunities for effective management of ANPP. Under the influence of these dynamic factors, the marginal impact of rural labor transfer on ANPP demonstrates significant variation across different stages and scales of implementation. Hence, Hypothesis 4 (H4) is proposed.

Building upon the aforementioned theoretical analysis and the formulation of relevant hypotheses, this study establishes a comprehensive conceptual framework that examines the interrelationships among rural labor transfer, non-farm income, agricultural socialized services, and ANPP, as illustrated in Figure 1.

To investigate the relationship between rural labor transfer and ANPP, this study employs a series of econometric models, including a two-way fixed effects model, a mediating effect model, a moderating effect model, and a threshold model. The two-way fixed effects model controls for unobserved regional heterogeneity and time-varying trends in panel data, accurately identifying the causal impact of rural labor transfer on ANPP by mitigating endogeneity from omitted variables. The mediating effect model uncovers the transmission pathways through which rural labor transfer influences ANPP, enhancing theoretical clarity on causal mechanisms. The moderating effect model examines the role of agricultural socialized services, identifying how agricultural socialized services change the magnitude and direction of the relationship between non-farm income and ANPP. The threshold model detects nonlinear dynamics, estimating critical points where rural labor transfer’s impact on ANPP shifts abruptly. Collectively, these models integrate causal inference, mechanism analysis, contextual variation, and nonlinearity, providing a comprehensive framework to decode the multifaceted link between labor mobility and ANPP. The specifications of these models are detailed as follows.

Panel data of 30 provinces (excluding Tibet, Hong Kong, Macao, and Taiwan) from 2005 to 2022 are selected as research samples. The data is collected from 2006—2023 National Statistical Yearbook, China Rural Statistical Yearbook, National Rural Economic Statistics, China Rural Management Statistical Annual Report, China Rural Policy and Reform Statistical Annual Report, China Population and Employment Statistical Yearbook, China Tertiary Industry Statistical Yearbook, China Water Statistical Yearbook, and provincial statistical yearbooks. A small number of missing data are filled by linear interpolation method. Logarithms are taken for ANPP, rural labor transfer, output value of agricultural socialized services, per capita GDP, the average education level of rural residents and drainage methods to reduce data volatility and address issues like heteroscedasticity. Descriptive statistics of each variable are shown in Table 1.

Before using Equation (1) for benchmark regression, the applicability of the model is tested. The random effects model is disregarded by Hausman’s test (chi2(11) = 15.29, Prob > chi2 = 0.0001), indicating that a fixed effects model is preferable for estimation. Robust standard errors are employed for estimate in all subsequent regression models. Furthermore, unit root tests are performed using Levin Lin Chu (LLC) test, Bruiting test, Im Peraran Shin (IPS) test, and Augmented Dickey-Fuller (ADF) test.¹ All variables pass at least the above three panel unit root tests, suggesting that the selected variables are stationary.

Table 2 reports the empirical results on the impact of rural labor transfer on ANPP. Model (1) demonstrates the regression results without control variables, revealing a statistically significant coefficient of 0.212 at the 1% level, indicating a positive association between rural labor transfer and ANPP. The inclusion of control variables in Model (2) yields a slightly attenuated yet statistically robust coefficient of 0.207, maintaining significance at the 1% level. This consistent positive relationship suggests that rural labor transfer contributes to the exacerbation of ANPP, further validating the research findings of (Luan and Han, 2021). To ensure the robustness of our findings, Model (3) incorporates Bootstrap robust standard errors to address potential estimation biases. The results confirm the stability of both the coefficient magnitude and direction, thereby reinforcing the validity of our initial findings. Consequently, we conclude that rural labor transfer significantly exacerbates ANPP, generating negative externalities in agricultural production and substantiating H1.

The analysis of control variables in Model (2) reveals several noteworthy patterns. It is evident from the table that among the economic variables, agricultural technological innovation, financial expenditure on agriculture, and the level of grain yield per unit area exert a significant positive impact on ANPP. This could be attributed to the fact that while agricultural technological progress enhances production efficiency, it has not yet fully achieved a green transformation in the short term. Meanwhile, the utilization of fiscal funds for agricultural support may indirectly stimulate an increase in inputs such as chemical fertilizers and pesticides. Additionally, the improvement in grain yield per unit area is often accompanied by greater investment in agricultural production factors, thereby exacerbating ANPP. In contrast, the multiple cropping index has a significant negative impact on ANPP. This is because appropriate multiple cropping optimizes the crop planting structure and drives the rational adjustment of the structure of agricultural input factors, thereby alleviating ANPP.

Population variables, which include rural population aging, urbanization rate, and the average educational level of rural residents, are used to capture the impact of changes in population structure and quality on ANPP. In the benchmark regression, none of the aforementioned population control variables showed significant effects. This may be due to the fact that, at the current stage, the promoting effect of rural population aging on the adoption of green production technologies, the resource crowding-out and pollution-increasing effects brought by the urbanization rate, and the facilitating effect of improved educational levels of rural residents on environmentally friendly production methods offset each other.

Environmental variables, including precipitation, temperature, and drainage methods, reflect the impact of natural conditions and infrastructure on ANPP. These environmental control variables in the table also did not exhibit significant effects. Although, in theory, drainage methods can alleviate ANPP through centralized treatment of domestic and agricultural wastewater, and climatic factors such as precipitation and temperature may affect the diffusion and occurrence of ANPP through natural processes, the role of these factors in this regression model has not reached a statistically significant level.

To test whether rural labor transfer act indirectly on ANPP through farmers’ non-farm income, the mediating effect of non-farm income is empirically estimated using Equations (2) and (3). Table 3 reports the results of the mediating effect of non-farm income. Model (1) reflects the overall effect of rural labor transfer on ANPP. Model (2) reflects the impact of rural labor transfer on non-farm income. The transfer of rural labor force changes by 1 percentage point, and non-farm income can significantly increase by 0.064 percentage points, indicating the existence of mediating effect. Model (3) reflects the impact of non-farm income on ANPP after controlling for the direct effect of rural labor transfer on ANPP. And it is found that the coefficients of rural labor transfer and non-farm income on ANPP are 0.184 and 0.368, respectively, and that pass the test at the 1% significance level, indicating that the indirect effect is obvious. Based on the above results, it can be seen that the promoting effect of rural labor transfer on ANPP can follow the path of “rural labor transfer → non-farm income →ANPP,” that is, non-farm income plays a mediating role in the impact of rural labor force transfer on ANPP, thereby verifying H2. It also supports the viewpoints of Wu and Liu (2017) and Shao and Li (2020). The test of Sobel and Bootstrap are used to examine the mediating effect to further confirm the reliability of the findings. As shown in Table 3, the Z-value of Sobel test is 2.686, which is significant at the 1% level, indicating that non-farm income acts as a mediator. The findings of the Bootstrap test of the mediation effect after 500 iterations of self-sampling are displayed in Table 4. It demonstrates that both direct and indirect impact coefficients are statistically significant in the positive direction, and the test for the mediating effect does not contain a value of 0 within the 95% confidence interval, effectively also suggesting the presence of the mediating effect (Hayes and Preacher, 2014). It is calculated that the mediating effect of non-farm income is 11.11% as the direct effect coefficient is 0.184 and the indirect effect coefficient is 0.023.

Model (4) in Table 3 presents the moderating effect of agricultural socialized services. To mitigate potential multicollinearity, the output value of agricultural socialized services and non-farm income were standardized before regression analysis using Equation (4). Subsequently, interaction terms were constructed and incorporated into Model (4). The results show that the regression coefficient of agricultural socialized services is 0.005, which does not reach statistical significance, indicating that its direct impact on ANPP when acting alone is not obvious. However, the coefficient of its interaction term with non-farm income is −0.230, and it is significant at the 1% level, indicating that the interaction has a significant impact on ANPP. Meanwhile, the coefficient of non-farm income dropped from 0.368 to 0.317 (−13.86%). This change indicates that agricultural socialized services have weakened the mediating role of non-farm income in the relationship between rural labor transfer and ANPP. Further analysis through the marginal effect equation reveals that when the output value of agricultural socialized services reaches 396.799 million yuan, the “pollution-increasing effect” of non-farm income will reverse. It is that before this critical point, non-farm income may drive the intensification of ANPP. When the output value of agricultural socialized services exceeds this figure, the role of non-farm income on ANPP will shift from “promotion” to “inhibition.” Based on the data in Table 1, the minimum value of the output value of agricultural socialized services is 1.087 and the maximum value is 7.126. This indicates that in most provinces of China, agricultural socialized services have reached a level that can reverse the pollution-increasing effect of non-farm income, and their pollution reduction role has been actually demonstrated, thereby validating H3.

In the benchmark regression estimation, there may be errors when choosing method and variable leading to biased results, and to ensure the reliability of the benchmark regression results, the robustness tests are conducted as follows.

The potential issue of endogeneity arising from mutual causality may exist in the relationship between rural labor transfer and ANPP. To address this concern and ensure the unbiased nature of our model estimates presented in Table 5, two methodological approaches are employed to examine the endogeneity of model.

Firstly, drawing upon the methodological approach established by Ashley et al. (2018), we implement a temporal adjustment by replacing the current values with the independent variable lagged by one period and the dependent variable advanced by one period to mitigate potential endogeneity concerns in models (5) and (6). Secondly, following the methodological approach of Lee and Wang (2022), the lagged variable of rural labor transfer is selected as an instrumental variable. The paper subsequently employs the Two-Stage Least Squares method (IV_2SLS) to estimate the effect of rural labor transfer on ANPP. The robustness of the instrumental variable approach is supported by several diagnostic tests as follows. The first-stage F statistic (F = 29.762) in model (7) exceeds the conventional threshold of 10, indicating a statistically strong correlation between the endogenous variables and the instrumental variables. Furthermore, as evidenced in model (8), the Cragg-Donald Wald F statistic (778.556) surpasses the critical value at the 10% significance level, thereby dismissing concerns regarding weak instrumental variables. Additionally, the Kleibergen-Paap rk LM statistics consistently reject the null hypothesis of underidentification at the 1% significance level, while the Hansen J statistic indicates the absence of over-identification issues, collectively confirming the appropriateness of the selected instrumental variable. The regression results from models (5) through (8) in Table 5 demonstrate that the significance level and directional impact of rural labor transfer on ANPP remain consistent with the findings from model (2). This consistency persists even after accounting for potential endogeneity, suggesting that rural labor transfer continues to exert a positive influence on ANPP when endogeneity considerations are incorporated into the analysis.

To examine the threshold effect of rural labor transfer, this section employs Equation (5) for empirical testing.” before the sentence” As evidenced by the results presented in Table 8, the corresponding F-value achieves statistical significance at the 5% level, confirming the existence of a single threshold effect in the relationship between rural labor transfer and ANPP, with a threshold value of 3.967 ( q = 3.967 ). The threshold regression results, as detailed in Table 9, demonstrate distinct patterns of influence across different scales of rural labor transfer. When the logarithmic value of rural labor transfer remains below or equal to the threshold of 3.967, the estimated coefficient for its impact on ANPP is 0.199, significant at the 1% level. Conversely, when the logarithmic value exceeds this threshold, the estimated coefficient decreases to 0.141, while maintaining statistical significance at the 1% level. These findings suggest a non-linear relationship between rural labor transfer and ANPP, wherein the marginal effect of rural labor transfer on pollution intensity diminishes as the scale of transfer increases beyond the identified threshold. This empirical evidence provides robust support for H4, indicating that while rural labor transfer generally exacerbates ANPP, its promoting effect attenuates as the scale of transfer expands beyond a critical point.