Farmland’s Comprehensive Improvement and Agricultural Total Factor Productivity Increase: Empirical Evidence from China’s National Construction of High-Standard Farmland

Abstract

Farmland improvement has become an overwhelmingly favorable policy in developing countries, being expected to leverage a sustainable agricultural total factor productivity (ATFP) to increase their agricultural competitiveness. Worldwide farmland improvement projects are experiencing an evolution from single goals to comprehensive goals (e.g., comprehensively improve the farmland quality by decreasing farmland abandonment and fragmentation and meanwhile improving soil–water conditions and machinery affordability). However, performances of comprehensive farmland improvement projects have been questioned, especially considering its implementary complexity and regional heterogeneity. This study applies a continuous difference-in-difference (DID) method to China’s provincial panel data (2005–2020) to analyze the impact of the high-standard farmland construction policy (which started China’s national project on comprehensive farmland improvement) on ATFP. Results show the policy significantly increases ATFP by 0.101 units. Moreover, parallel trend and robustness test results indicate the policy effect has stability and continuity. Heterogeneity analysis results show the policy effect is greater in major grain-producing regions than non-major grain-producing regions, the central regions than western or eastern regions, and regions with high disease—pest control and soil—water conservation levels than areas with low levels. Mechanism analysis results show the policy effect is achieved through three paths—operation scale increase (mediating effect size is 16.13%), planting structure adjustment (mediating effect size 12.80%), and agricultural disaster reduction (mediating effect size 13.74%). Thus, this study advocates sustainable and specialized high-standard farmland construction: it suggests post-construction policies maintaining high-standard farmland quality and detailed policies considering different regions’ heterogeneity.

1. Introduction

Developing countries face crucial challenges in improving their agricultural economies by further releasing their agricultural production potential. Take China as an example: although its agricultural economy has recently expanded rapidly, making a sizeable contribution to national growth and poverty reduction, its agricultural competitiveness—especially agricultural productivity—remains far below the OECD frontier. Closing this gap is now framed not only as a growth imperative but also as a prerequisite for meeting SDG 2 (Zero Hunger) and SDG 15 (Life on Land), which jointly call for food-system intensification without further deforestation, land-degradation, or biodiversity loss. Nations worldwide therefore introduce agricultural support policies to raise agricultural productivity; yet, relative to developed countries, most developing economies still lack both the fiscal capacity and the policy diversity to deliver strong, wide-ranging support while simultaneously safeguarding the natural resource base that underpins long-term productivity.

According to the WTO (World Trade Organization) agreement calculation, in 2022, the total agricultural support of developed countries accounted for 15.9% of the total agricultural output value1. Among those developed countries, the United States, Norway, and the European Union had total agricultural support accounting for 60.4%, 62.4%, and 21.2% of their total agricultural output value, respectively. The large amount of agricultural support has effectively promoted their ATFP levels. However, developing countries such as China’s agricultural support accounted for only 11.9% (data in 2022) of its total agricultural output value, seriously lagging behind agriculturally developed countries.

More importantly, agricultural growth should no longer be driven solely by output maximization; it must also advance the Sustainable Development Goals—especially SDG 2 (Zero hunger) and SDG 15 (Life on Land). This implies that the modernization of agriculture must simultaneously enhance resource-use efficiency, reduce non-point-source pollution, and safeguard biodiversity and ecosystem services, thereby forging a synergistic balance between productivity gains and ecological protection. Developing countries also confront widespread land degradation, soil acidification, and habitat loss that directly erode productive capacity. It is noteworthy is that China’s rapid urbanization has caused more than 1/10 of the farmland (roughly 13.4 million hectares) to be abandoned since 1995 [1], disproportionately eroding the most fertile topsoils, permanently lowering the national cultivated land reserve, and also impeding the establishment of sustainable, resource-efficient food systems.

Ahead of the absolute dominance of agricultural support strengths, nowadays, many developed countries have turned to more flexible and comprehensive ways of agricultural support to improve their agricultural competitiveness. For instance, they have begun to change their agricultural support policies from the “yellow boxes” (policies mainly based on direct price subsidies) to the “green boxes” (policies that do not distort the agricultural production prices). During that transformation, “blue boxes” (transitional policies from “yellow boxes” to “green boxes,” being unrestricted with more varieties investing in infrastructure construction and resource utilization) are widely practiced [2]. Farmland is the primary material used in agricultural production; farmland also directly affects the utilization effectiveness of other agricultural input factors and consequently influences agricultural productivity [3]. Therefore, farmland improvement policies become the most representative “blue box” policies [3]. For instance, the United States upgraded its Soil Conservation and Food Security Acts and the European Union implemented the Common Agricultural Policy [4]. Japan also conducted several Land Improvement Projects under its agricultural land law [4]. Developed countries’ farmland policies aim for improvement with comprehensive goals, such as dealing with both landscape issues including farmland abandonment and fragmentation and land condition issues including pollution, biodiversity, soil erosion, drought, and flood disasters, simultaneously [5]. Most developing countries are still pursuing farmland improvement projects with limited goals or a particular goal. For instance, some developing countries focused on landscape issues especially farmland occupation by industrial land and farmland abandonment during the rapid urbanization [6,7].

In 2011, China responded to the widening gaps in agricultural support strength and diversity by launching a national high-standard farmland program that explicitly embeds the principles of SDG 2 (Zero Hunger) and SDG 15. Rather than expanding cultivated area, the policy seeks to lift productivity per hectare through land consolidation, irrigation–drainage retrofits and soil-fertility enhancement, thereby boosting food availability and access in line with SDG 2’s hunger-eradication mandate while curbing the need to convert natural ecosystems. Publicly tendered projects are required to deliver contiguous blocks equipped with modern water-saving infrastructure, wind-break shelterbelts, and organic matter restoration measures that jointly improve disaster resilience, lower agrochemical losses, and sequester soil carbon. By integrating ecological criteria into engineering specifications—removing obstacle sub-soil layers, reshaping field micro-topography to reduce run-off, and mandatory monitoring of soil health2—the initiative operationalizes SDG 15.3’s land-degradation-neutrality target while simultaneously advancing SDG 2’s call for sustainable agricultural intensification that closes yield gaps without compromising future productive capacity, positioning high-standard farmland as the physical platform on which higher total factor productivity and environmental stewardship are pursued in tandem.

By the end of 2022, China has built 67 million hectares (hm²) of high-standard farmland and plans to build 80 million hm² (approximately two-thirds of China’s farmland) by 2030. The Chinese government expects the high-standard farmland will achieve four “high standards”—one for the farmland’s quality, one for the agricultural productivity, one for the disaster resistance capability, and one for the resource-utilizing effectiveness. Yaolin Zhang, China’s Deputy Minister of Agriculture and Rural Affairs, acclaimed on the day that China’s National High-Standard Farmland Construction Plan (2021–2030) was released that “from the actual situation in recent years, after the completion of high-standard farmland, it can significantly improve the efficiency of water and land resource utilization, enhance the ability of grain production and disaster prevention and mitigation, and increase the grain production capacity of the project area by 10% to 20% on average.”3 It can be seen that the Chinese government has made great efforts and placed deep expectations on constructing high-standard farmland. However, questions such as whether and how high-standard farmland construction increases agricultural productivity have rarely been empirically addressed. To verify these questions is crucial especially when facing controversies from multiple channels about high-standard farmland construction projects’ performance.

This study empirically aims to verify the above questions by applying the continuous difference-in-difference (DID) method to China’s provincial panel data (2005–2020). To avoid the limitations of partial productivity indicators, we employ agricultural total factor productivity (ATFP), defined as the ratio of aggregate output to a Divisia index of all observable inputs (land, labor, machinery, fertilizer and intermediate services). Unlike yield or output per hectare, ATFP captures technological progress, scale economies, and efficiency changes, thereby providing a comprehensive measure of productivity growth. Ahead of baseline regression analysis, this study further conducts the parallel trend test, robustness test, and extensive analysis to confirm the results’ reliability and robustness. Heterogeneity and mechanism analyses are also undertaken to enrich and deepen the understanding of the results. This study achieved practical contributions by providing China’s policy insights for developing countries to lift their agricultural competitiveness by improving farmland comprehensively and increasing ATFP.

2. Literature Review

2.1. ATFP

ATFP refers to the amount of factor input consumed by one unit of agricultural production [8]. ATFP is an essential indicator measuring agricultural development’s speed and effectiveness; thus, it has become the recent research focus in development economics [8]. One primary research topic regarding ATFP is its measuring approaches. ATFP measurements commonly include input and output indicators. Input indicators usually include arable land, labor force, capital, machinery power, and pesticide and fertilizer usage [9]. Output indicators usually include gross agricultural product and environmental pollution [10] when measuring the green productivity of agriculture.

Existing measuring methods can be roughly divided into two categories. The first is Stochastic Frontier Analysis (SFA). SFA is a typical parametric analysis method; it determines the production frontier by determining parameters in the frontier production function [11,12]. By SFA, researchers need to determine the specific production function for a specific research object; thus, this measurement has a certain pertinency [11,12]. The second method is Data Enveloping Analysis (DEA). DEA is a typical non-parametric analysis method [13]. It uses linear programming to evaluate comparable units of the same type of research objects. This method does not need to determine the parameters of the production frontier and has the advantage of enabling the processing of multiple indicators. Some researchers have subsequently built several methods based on DEA, such as the ML (Multi-dimensional Learned) index method, GML (global Malmquist–Luenberger) index method, and SBM (Stochastic Block) model [14,15].

2.2. Land and ATFP

Researchers have investigated many operations and conditions that influence ATFP. These operations and conditions include but are not limited to agricultural machinery, agricultural informatization, rural infrastructure construction, labor migration, and geographical conditions (see, e.g., [16,17,18]). Among those operations and conditions, land, as a natural economic complex and one of the essential inputs in the agricultural production process, has received much practical and academic focus. Researchers find that farmland’s quality and utilization mode determine its productivity level, affecting the total ATFP [19]. Reference [20] finds that the land’s quality and climate status explain 20% of the agricultural output in most countries and regions.

There are also more in-detailed studies. For instance, reference [21] identifies four indicators reflecting land quality—soil quality, the proportion of arable land to agricultural land, the proportion of effective irrigated area to total area, and average rainfall. Results show these four indicators all have a significantly positive correlation with ATFP. Some researchers also found that when land intensification reaches a certain level, production factors can exert the agglomeration effect, which positively impacts ATFP and is heterogeneous in different regions [22].

2.3. Land Policies and ATFP

It is worth nothing that the farmland policy, as a governmental operation lifting farmland’s quality and improving its utilization effectiveness, also catches academic interest nowadays. Researchers claim that the farmland policy can strengthen farmland’s quality management and ecological control, promoting sustainable agricultural development and increasing ATFP [23]. With China’s recent advance in releasing agricultural policies, quite a few policies have been examined, such as policies on agricultural cooperative subsidies [24], rights verification and product certification [25], and food subsidies [26]. Ahead of those policies based on price subsidies, China’s agricultural policies are experiencing a transformation from “yellow boxes” to “green boxes.” “Blue boxes” of agricultural policies, such as farmland policies, occupy an essential practical rationale during this transformation.

China’s farmland policies have experienced a process consisting of different stages, from farmland protection (Stage 1) to farmland development, consolation, and reclamation (Stage 2), stepping into a new stage of overall farmland’s comprehensive improvement (Stage 3) [27]. High-standard farmland construction is the new era’s most essential farmland policy. This policy promotes high-standard farmland construction projects, the largest single fiscal expenditure project in China’s agriculture industry. However, this farmland policy still has not received due attention from the academic community. Despite research identifying its multi-facet spillover effect on fertilizer application, agricultural film recycling and agroecological efficiency [28], adoption of water-saving irrigation technology and food production [29], grain production adjustment [30], and rural poverty reduction and revitalization [31], extant discussions on whether (and how) this policy influences ATFP are far from sufficiency.

Specifically, the existing few discussions offer theoretical hypotheses about the positive effects of high-standard farmland construction on ATFP, usually based on the expectation that optimizing agricultural factor input structure and quality will improve agricultural total factor productivity. Recent studies frame the high-standard farmland program as a three-pronged catalyst for ATFP growth. First, by leveling fragmented plots and inserting permanent field roads and irrigation trunks, the policy drastically enlarges the feasible operational scale. With contiguous blocks that can be traversed by large machinery, households rent in or swap adjacent parcels until the holding size matches the locally available mechanical capacity; the associated economies of scale raise output per unit of land and labor, pushing observed ATFP upward [28,29]. Second, the same infrastructure lowers the downside risk of droughts and floods [32], prompting farmers to reallocate working capital away from precautionary chemical overuse toward quality seed and precision-machinery services [33], an efficiency gain that further lifts ATFP. Third, because mechanized grain production is the least labor-intensive option on enlarged fields, producers switch from scattered cash-crop patches to consolidated grain blocks—an endogenous planting structure adjustment that reduces input complexity, simplifies crop management, and reinforces the aggregate productivity gain [34]. Together, scale expansion, input reallocation and grain-oriented restructuring constitute the inter-locking channels through which high-standard farmland construction is expected to shift the agricultural productivity frontier outward.

Controversy remains. While many studies credit high-standard farmland (HSF) construction with raising productivity, others warn that regrading and replowing can disrupt the land’s natural layout and lower ATFP [35]. Investment fragmentation, repeated earthworks and absence of plot-specific design further erode returns [36]. Empirical evidence from outside China is mixed: reference [37] found zero TFP gains after plot consolidation in Ethiopia when complementary inputs are missing, and similarly, reference [38] found no significant productivity improvements following land consolidation in Slovenia, attributing the neutral effect to persistent small farm sizes and limited post-consolidation investments. Reference [39] records yield losses in Romanian land-reaggregation projects caused by temporary soil compaction and delayed replanting. These cases underscore that high-standard farmland construction benefits are conditional on timely provision of drainage, machinery access, and farmer training; without them, the same earth-moving operations can leave ATFP unchanged or even negative.

The sustainability (that the project can perform well and stably in a time range) is also suspected, considering that the project leads a split between organizations taking the project to construct the farmland and farmers utilizing and maintaining the farmland [36]. Empirical examinations are still lacking in dealing with such disputes on high-standard farmland construction policy performance. Further, although some researchers interpreted that the inner mechanisms of high-standard farmland construction positively influence ATFP as agricultural scale and machinery level rise and agricultural infrastructure and ecological environment improve [40], a holistic mechanism complex has not been established nor empirically verified.

Based on the above considerations, this paper uses China’s provincial panel data from 2005 to 2020 to analyze the effect of high-standard farmland construction on ATFP. It discusses the effect’s heterogeneity and inner mechanism to provide an empirical reference for comprehensive farmland improvement and agricultural production efficiency lift.

3. Research Model and Hypothesis

3.1. Research Model

This study establishes a research model as shown in Figure 1. Implementing high-standard farmland construction policy influences ATFP through three diverse paths—operation scale increase, planting structure adjustment, and disaster resistance.

3.2. Research Hypothesis

3.2.1. Path of Operation Scale Increase

China’s household contract responsibility system4 could easily generate farmland fragmentation, decentralizing rural land management and impeding agricultural machinery [29,41]. Such a situation is not conducive to scale effects forming in agricultural production, resulting in low input efficiency of agricultural production factors and seriously restricting the ATFP improvement [29]. High-standard farmland construction might improve the ATFP by facilitating the formation of scale effects in agricultural operations. First, through land integration measures, high-standard farmland construction reduces farmland fragmentation, promotes land scale expansion, and facilitates centralized contiguous management; thus, agricultural productions can benefit from scale effects and centralized management [42]. Second, through land consolidation measures, high-standard farmland construction can reduce soil compaction degree and soil water evaporation and improve soil moisture retention capacity and organic matter content [43], thus reducing the frequency of agricultural production activities (e.g., agricultural irrigation, soil loosening, and fertilization), and finally optimizing agricultural labor factor’s input efficiency [44]. Third, through its road construction measures, high-standard farmland construction can optimize farmland’s layout and facilitate agricultural machinery, thus reducing farmers’ operation burdens, optimizing agricultural labor factor’s input efficiency [45], and further encouraging farmers to expand their farmland’s scale [46]. When the operation scale expands, farmers will adopt soil improvement measures to improve farmland’s quality, making it possible to reduce fertilizer input, thereby optimizing agricultural factor’s input efficiency, increasing agricultural production, meanwhile saving agricultural production costs, and ultimately promoting ATFP improvement [47]. Nevertheless, scale expansion may also induce moral hazards: larger plots could tempt farmers to over-mechanize or adopt monoculture to minimize supervision cost, which may erode biodiversity and long-run soil health, and eventually offset part of the ATFP gains [35].

3.2.2. Path of Planting Structure Adjustment

One of the most essential goals of the high-standard farmland construction policy is to ensure grain production stability. The high-standard farmland construction policy provides sufficient profit space for grain planting by the following three measures, enables agricultural machinery’s cross-regional and large-scale operation, and thus promotes the planting structure towards the “grainization5” [28,30]. First, through the land leveling measures, the high-standard farmland construction has changed the status of previously uneven farmland, eliminated the farmland quality difference between different areas, and facilitated the convenience of agricultural machines’ operation [48]. Given that food crops are more geologically suitable for mechanized production, the high-standard farmland construction promotes farmers’ preference for planting food crops over cash crops. Second, as mentioned in Section 3.2.1, the high-standard farmland construction expands the operation scale. According to [49], planting crops will result in a labor shortage and the household’s marginal income declines when the operation scale exceeds the rural household’s labor force supply. Such situations push farmers to seek more labor-saving operations and thus turn to agricultural machinery. Also, considering food crops’ fitness for mechanized production, the operation scale expansion will probably change farmers’ planting preferences from cash crops to food crops [50]. Compared with cash crops, ecological conservation crops such as rice, corn, and wheat are more likely to achieve fertilizer reduction [51]. In addition, farmers can more easily apply operations such as “the good cultivar combined with good technique” and “agricultural machinery integrated with agronomy” when planting food crops. Altogether, such planting structure adjustment can reduce agricultural factors’ input, lift agricultural factors’ input efficiency, and increase ATFP [52], rent-seeking behavior may emerge when local cadres steer subsidies toward grain crops regardless of soil suitability, leading to yield losses and undermining the anticipated ATFP improvement [28].

3.2.3. Paths of Agricultural Disaster Resistance

High-standard farmland construction reduces agricultural disasters and protects farming production through three channels. First, high-standard farmland construction makes up for the shortcomings of extant agricultural infrastructure [53]. This project builds farmland to guarantee high and stable yields even after drought and flood [54]. As aforementioned, such goals should be achieved by improving rural roads, promoting farmland leveling, and promoting irrigation and drainage backbone projects [55]. Acquiring water sources through the construction of irrigation infrastructure could improve irrigation effectiveness and reduce disaster harm [56]. Second, high-standard farmland construction enhances soil–water loss control effectiveness. This project utilizes afforestation to consolidate soil and trap precipitation, thus strengthening the soil’s infiltration capacity [57]. In addition, various soil–water conservation measures are adopted to enhance farmland’s water storage capacity and improve farmland’s ecological environment. Thus, agricultural disasters’ concurrence rates can be reduced [58]. Third, high-standard farmland construction also promotes the socialization of agricultural services. This project encourages professional institutions to provide farmers with excellent varieties and advanced techniques for improving crop planting characteristics (i.e., flood, drought, and insect tolerance, [59]). Meanwhile, farmers can outsource and entrust professional organizations to carry out standardized pest control operations, thus reducing the harms of agricultural disasters [60]. Reducing agricultural disasters can improve ATFP by maintaining stable and safe agricultural production [54,58]. However, ecological degradation—such as over-extraction of groundwater for irrigation or excessive simplification of landscape structure—may offset the disaster reduction benefits and jeopardize long-run ATFP [28].

Although the preceding discussion has highlighted potential risks—such as moral hazard arising from scale expansion, rent-seeking in planting structure adjustment, and ecological degradation linked to disaster-proof infrastructure—these costs are either transitory or amenable to mitigation through improved governance and complementary investments. On balance, the dominant share of empirical evidence and the policy design features that embed monitoring, extension services, and environmental safeguards suggest that the positive productivity channels outweigh the negative distortions. Accordingly, we maintain that the net impact of high-standard farmland construction on agricultural productivity is expected to be positive. Altogether, this paper proposes the following hypotheses:

Hypothesis 1.

Implementing the high-standard farmland construction policy contributes to the ATFP improvement.

Hypothesis 2.

Implementing the high-standard farmland construction policy improves ATFP through paths including (a) increasing agricultural operation scale, (b) adjusting agricultural planting structure, and c) reducing agricultural disaster and protecting agricultural production.

4. Materials and Methods

4.1. Measurements of ATFP

Considering that ATFP is affected by many uncontrollable factors, this paper chooses SFA instead of the DEA method to measure ATFP. As mentioned, DEA is a typical non-parametric method that does not need to set any specific form of the production function. However, the DEA fails to consider the impacts of random factors. SFA has an advantage in considering random factors’ influences, although it is a parametric method that needs to set the specific production function and estimate all parameters.

This study applied the commonly used panel data SFA models—the BC92 and BC95 models to measure ATFP. The model is as follows:

$${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}=\mathrm{f}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}},\beta \right)\exp \left({\mathrm{v}}_{\mathrm{i}\mathrm{t}}-{\mathrm{u}}_{\mathrm{i}\mathrm{t}}\right),\mathrm{i}=\mathrm{1,2},\dots ,\mathrm{N};\mathrm{t}=\mathrm{1,2},\dots ,\mathrm{T}$$

(1)

When taking logarithms of both sides in Formula (1), the model is written as follows:

$${\mathrm{l}\mathrm{n}\mathrm{Y}}_{\mathrm{i}\mathrm{t}}=\mathrm{l}\mathrm{n}\mathrm{f}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}},\beta \right)+{\mathrm{v}}_{\mathrm{i}\mathrm{t}}-{\mathrm{u}}_{\mathrm{i}\mathrm{t}},\mathrm{i}=\mathrm{1,2},\dots ,\mathrm{N};\mathrm{t}=\mathrm{1,2},\dots ,\mathrm{T}$$

(2)

In Formula (2), ${\mathrm{Y}}_{\mathrm{i}\mathrm{t}}$ was the agricultural output value of the ith province in the tth year. $\mathrm{f}\left({\mathrm{X}}_{\mathrm{i}\mathrm{t}},\beta \right)$ was the best output level of the ith province in the tth year under the given input. ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ was the agricultural output value of the ith province in the tth year, and $\beta$ was the parameter to be estimated. ${\mathrm{v}}_{\mathrm{i}\mathrm{t}}$ was the noise term, which was used to judge the influence of measurement error and random interference factors and followed the normal distribution of ${\mathrm{v}}_{\mathrm{i}\mathrm{t}}$~$($0,${{\sigma }_{\mathrm{v}}}^2)$. ${\mathrm{u}}_{\mathrm{i}\mathrm{t}}$ was the technical inefficiency term, reflecting the distance of the ith province from the efficiency front in the tth year. According to [61], we assumed the technical inefficiency term changes with changes in provinces and time simultaneously, i.e., ${\mathrm{u}}_{\mathrm{i}\mathrm{t}}={\mathrm{u}}_{\mathrm{i}}\mathrm{e}\mathrm{x}\mathrm{p}[-\mathrm{ŋ}(\mathrm{t}-\mathrm{T}\left)\right]$. ${\mathrm{u}}_{\mathrm{i}\mathrm{t}}$ was independent of ${\mathrm{v}}_{\mathrm{i}\mathrm{t}}$, and its distribution, ${\mathrm{u}}_{\mathrm{i}\mathrm{t}}~\mathrm{N}+{(\mathrm{u}}_{\mathrm{i}\mathrm{t}},{{\sigma }_{\mathrm{u}}}^2)$, met with the normal distribution on the left side. $\mathrm{ŋ}$ was the parameter to be estimated. Reference [62] further proposed a technical inefficiency impact model, which expressed ${\mathrm{u}}_{\mathrm{i}\mathrm{t}}$ as a function of specific factors affecting productivity. When estimated using the maximum likelihood method, the ATFP of the ith province in the tth year (${\mathrm{t}\mathrm{e}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$) is expressed as follows:

$${\mathrm{t}\mathrm{e}\mathrm{l}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{{\mathrm{Y}}_{\mathrm{i}\mathrm{t}}}{\mathrm{e}\mathrm{x}\mathrm{p}\left[\mathrm{f}\right({\mathrm{X}}_{\mathrm{i}\mathrm{t}},\beta )+{\mathrm{v}}_{\mathrm{i}\mathrm{t}}]}}=\mathrm{e}\mathrm{x}\mathrm{p}(-{\mathrm{u}}_{\mathrm{i}\mathrm{t}})$$

(3)

The value range of ${\mathrm{t}\mathrm{e}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$ was [0, 1]. The closer it was to 1 (0), the higher (lower) the value of ATFP.

This study further constructed the stochastic frontier production function based on the C-D production function. Unlike the traditional C-D production function, which assumed that each factor’s substitution elasticity was set to 0 or 1, SFA used the more flexible translogarithm production function instead of specifying the substitution elasticity of each factor. The ATFP function was set as follows:

$$\begin{gathered} \mathrm{l}\mathrm{n}{\mathrm{Y}}_{\mathrm{i}}={\beta }_0+{\beta }_{\mathrm{K}}\mathrm{l}\mathrm{n}\mathrm{F}+{\displaystyle \frac{1}{2}}{\beta }_{\mathrm{K}\mathrm{K}}{\left(\mathrm{l}\mathrm{n}\mathrm{F}\right)}^2+{\beta }_{\mathrm{M}}\mathrm{l}\mathrm{n}\mathrm{M}+{\displaystyle \frac{1}{2}}{\beta }_{\mathrm{M}\mathrm{M}}{\left(\mathrm{l}\mathrm{n}\mathrm{M}\right)}^2+{\beta }_{\mathrm{M}}\mathrm{l}\mathrm{n}\mathrm{E}+{\displaystyle \frac{1}{2}}{\beta }_{\mathrm{M}\mathrm{M}}{\left(\mathrm{l}\mathrm{n}\mathrm{E}\right)}^2+{\beta }_{\mathrm{K}}\mathrm{l}\mathrm{n}\mathrm{L}+{\displaystyle \frac{1}{2}}{\beta }_{\mathrm{L}\mathrm{L}}{\left(\mathrm{l}\mathrm{n}\mathrm{L}\right)}^2+ \\ {\beta }_{\mathrm{S}}\mathrm{l}\mathrm{n}\mathrm{S}+{\displaystyle \frac{1}{2}}{\beta }_{\mathrm{S}\mathrm{S}}{\left(\mathrm{l}\mathrm{n}\mathrm{S}\right)}^2+{\beta }_{\mathrm{K}\mathrm{M}}\left(\mathrm{l}\mathrm{n}\mathrm{F}\right)\left(\mathrm{l}\mathrm{n}\mathrm{M}\right)+{\beta }_{\mathrm{K}\mathrm{S}}\left(\mathrm{l}\mathrm{n}\mathrm{F}\right)\left(\mathrm{l}\mathrm{n}\mathrm{E}\right)+{\beta }_{\mathrm{K}\mathrm{L}}\left(\mathrm{l}\mathrm{n}\mathrm{F}\right)\left(\mathrm{l}\mathrm{n}\mathrm{L}\right)+{\beta }_{\mathrm{K}\mathrm{S}}\left(\mathrm{l}\mathrm{n}\mathrm{F}\right)\left(\mathrm{l}\mathrm{n}\mathrm{S}\right)+{\beta }_{\mathrm{M}\mathrm{L}}\left(\mathrm{l}\mathrm{n}\mathrm{M}\right)\left(\mathrm{l}\mathrm{n}\mathrm{E}\right)+ \\ {\beta }_{\mathrm{M}\mathrm{L}}\left(\mathrm{l}\mathrm{n}\mathrm{M}\right)\left(\mathrm{l}\mathrm{n}\mathrm{L}\right)+{\beta }_{\mathrm{M}\mathrm{S}}\left(\mathrm{l}\mathrm{n}\mathrm{M}\right)\left(\mathrm{l}\mathrm{n}\mathrm{S}\right)+{{\beta }_{\mathrm{M}\mathrm{L}}\left(\mathrm{l}\mathrm{n}\mathrm{E}\right)\left(\mathrm{l}\mathrm{n}\mathrm{L}\right)+{\beta }_{\mathrm{M}\mathrm{L}}\left(\mathrm{l}\mathrm{n}\mathrm{E}\right)\left(\mathrm{l}\mathrm{n}\mathrm{S}\right)+\beta }_{\mathrm{L}\mathrm{S}}\left(\mathrm{l}\mathrm{n}\mathrm{L}\right)\left(\mathrm{l}\mathrm{n}\mathrm{S}\right)+{\mathrm{v}}_{\mathrm{i}}-{\mathrm{u}}_{\mathrm{i}} \end{gathered}$$

(4)

In Formula (4), agricultural fertilizer input ($\mathrm{F}$) referred to the investment in fertilizers; agricultural machine input ($\mathrm{M}$) was the sum of agricultural machinery’s costs in fuels, repairs, and social services; agricultural electricity input ($\mathrm{E}$) was the sum of rural electricity consumption; agricultural labor input ($\mathrm{L}$) was the input amount converted from labor use while agricultural land input ($\mathrm{S}$) was that from farmland area.

4.2. Model Setting and Method Choosing

4.2.1. The Overall Method Design

In 2011, China issued the National Agricultural Comprehensive Development of High-standard Farmland Construction Plan (2011–2020), the first document regulating high-standard farmland construction. The document leads to different tasks and efforts in the construction work in different provinces. The construction-completed area of high-standard farmland in each province continued to increase along with the duration of construction after this starting point. Therefore, we assumed two differences existed—the high-standard farmland area in each province was different before and after the document release, and the area was different among different provinces in any year. Considering this complex situation, this study took the proportion of high-standard farmland area as the policy variable, thus dividing the experimental and control groups as provinces with relatively high and low areas, respectively.

Compared with previous studies on China’s agricultural total factor productivity, the method utilized in this study has the following features: First, we employ a continuous difference-in-differences (DID) estimator that exploits province-level variation in the cumulative area of high-standard farmland, thereby allowing the policy effect to evolve smoothly over time and avoiding the arbitrary post-period assumption that characterizes earlier DID applications. Second, we go beyond the conventional “east/central/west” split and examine whether the policy effect varies with (i) major-grain-region status, (ii) economic–geography zone, (iii) pest and disease control level, and (iv) soil–water conservation capacity. This study is among the earliest ATFP study on China that simultaneously tests functional, geographical, and agroecological heterogeneities, revealing a Matthew effect. Third, we quantify the causal mediation of the policy effect through three empirically measured channels—operation-scale expansion, planting structure adjustment and disaster-risk reduction—using a structural-equation framework, providing the first mechanistic decomposition of high-standard farmland impacts on ATFP for China.

Province-level clustering is adopted because the high-standard farmland program is administered and funded at the provincial level, inducing within-province serial correlation in both the treatment variable and the error term. This clustering scheme follows [63], yielding conservative standard errors that account for arbitrary intra-province autocorrelation and heteroskedasticity across years.

4.2.2. Baseline Regression Model

To identify the impact of the high-standard farmland construction policy on ATFP, we set the baseline model as follows:

$${\mathrm{t}\mathrm{e}\mathrm{l}}_{\mathrm{i}\mathrm{t}}={\beta }_0+{\beta }_1{\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}+{\beta }_2{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}+{\eta }_{\mathrm{i}}+{\gamma }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(5)

In Formula (5), the explained variable ${\mathrm{t}\mathrm{e}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$ was ATFP in the ith province and the tth year. ${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$ was the core explanatory variable, in which ${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}$ was the proportion of high-standard farmland area, ${{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$ was the policy time point variable whose value was 1 or 0, and 2011 was set as the time point. ${\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}$ represented a set of control variables. ${\eta }_{\mathrm{i}}$ and ${\gamma }_{\mathrm{t}}$ represented the regional and time effects in the model. ${\epsilon }_{\mathrm{i}\mathrm{t}}$ was the random error term. ${\beta }_0$ was a constant term while ${\beta }_1$ and ${\beta }_2$ were the coefficients to be estimated.

4.2.3. Parallel Trend Test and Policy Dynamic Impact

We tested the parallel trend of the DID model, that is, to examine whether ATFP in the experimental and control groups changed similarly before the policy’s release. We set the following model:

$${\mathrm{t}\mathrm{e}1}_{\mathrm{i}\mathrm{t}}={\beta }_0+{\sum }_{\mathrm{t}=2005}^{2017}{\beta }_{\mathrm{t}}{\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {\mathrm{D}}_{\mathrm{t}}+{\beta }_2{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{\mathrm{i}\mathrm{t}}+{\eta }_{\mathrm{i}}+{\gamma }_{\mathrm{t}}+{\epsilon }_{\mathrm{i}\mathrm{t}}$$

(6)

In Formula (6), ${\mathrm{D}}_{\mathrm{t}}$ was the year dummy variable, and other variables were set unchanged. If the policy release could significantly improve ATFP, the estimated coefficient of the interaction term (${\beta }_{\mathrm{t}}$) in the model before the policy implementation should be relatively stable, and there should be a significant upward trend of ${\beta }_{\mathrm{t}}$ after the policy’s implementation. Formula (6) also estimated the policy’s dynamic impact on ATFP.

4.3. Variable Descriptions

4.3.1. Explained Variable

BC92 and BC95 methods of SFA were used to measure the explained variable (ATFP). Descriptive statistics of each agricultural input are listed in

Table 1. Agricultural productivity’s measurement index system.

Abbreviation	Variable	Measuring Unit	Mean Value	Standard Deviation
Fertilizer	Fertilizer usage	Thousand tons	1794.672	1442.111
Machine	Agricultural machine usage	Million kilowatt-hours	29.776	28.198
Electricity	Rural electricity consumption	Billion kilowatt-hours	23.250	35.701
Labor	Rural farmers	Million persons	15.237	11.909
Sowing	Agricultural area of sowing	Thousand hectares	5203.333	3693.602

. It is worth noting that we also used agricultural output value per capita as an alternative measurement of ATFP in the robustness test.

4.3.2. Core Explanatory Variable

For the core explanatory variable—the high-standard farmland construction policy—this study used the interaction term (${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$) as aforementioned. This study also used the interaction term ($\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$) in the robustness test. $\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}$ was the comprehensive development investment per unit area.

4.3.3. Mediator Variables

This paper selected the proportion of cultivated land circulation area, planting structure, and disaster rate as the mediator variables, corresponding to the agricultural operation scale path, planting structure adjustment path, and agricultural disaster reduction path, respectively, in the theoretical hypotheses.

4.3.4. Control Variables

The literature commonly acknowledge that farmers’ human capital and agricultural support from techniques and finances all significantly impact farmers’ adoption of new farm techniques, which consequently influence ATFP. Therefore, this study included farmers’ average education years as a control variable in the model to represent the human capital, and similarly, the number of agricultural skills per 1000 people, and the ratio of expenditure on agriculture, forestry, and water affairs to the total budget expenditure, to represent agricultural support from techniques and finances, separately. In addition, farmland’s irrigation status influenced the efficiency of agricultural input factors. Thus, this study included farmland’s irrigation status (represented by the proportion of effectively irrigated farmland area in the total farmland area) as a control variable in the model. A higher industrialization rate, which meant a lower proportion of agricultural output value and a higher outflow of agricultural labor, was usually associated with a lower ATFP. Therefore, this study included the industrialization rate (measured by the ratio of secondary industries’ added value to the total output value) as a control variable in the model. The rural Engel coefficient was also included as the higher the rural Engel coefficient value, the less income farmers afforded to expand their agricultural production and the lower adoption of agricultural machines. Additionally, average daily sunshine duration, annual temperature, and daily rainfall were selected as climate variables, also included in the model.

4.4. Data Sources

This study used panel data from 31 provinces from 2005 to 2020 to evaluate the effect of high-standard farmland construction policy on ATFP. We collected data on high-standard farmland’s construction area and the comprehensive development investment per unit area from the China Financial Statistical Yearbooks6. The data on farmers’ average education years, the number of rural labor force, the amount of plastic film used, and the amount of fertilizer applied were derived from the China Rural Statistical Yearbooks7. Data on cultivated land area, soil erosion control area, ratio of effective irrigation area, and comprehensive prevention and control rate of pests and diseases were obtained from the China Statistical Yearbooks8. The average daily sunshine duration, annual temperature, and daily rainfall were obtained from the China Meteorological Administration9. Variables, measurements, and their descriptive results are displayed in

Table 2. Variable selection and descriptive statistics.

Abbreviation	Variable	Measuring Method	Measuring Unit	Mean Value	Standard Deviation
field	Proportion of high-standard farmland construction area	High-standard farmland construction area/total farmland area	%	0.3684	0.2373
efficiency1	Agricultural productivity	Calculated by SFA model: bc95	%	0.8542	0.0908
efficiency2	Agricultural productivity	Calculated by SFA model: bc92	%	0.8457	0.0941
value	Agricultural output value per capita	Total agricultural output value/rural farmers	thousand RMB per capita	19.651	13.951
land	Agricultural output value per unit of farmland area	Total agricultural output value/total farmland area	thousand RMB/hm2	74.690	55.542
invest	Comprehensive development investment per unit of farmland area	Total comprehensive development investment//total farmland area	thousand RMB/hm2	6.777	8.011
education	Average years of education of rural labor force	(∑(number of people in each age group × average years of education in each age group)/total population	years	8.6474	1.2073
irrigation	Proportion of effective irrigated area	Total effective irrigated area/total farmland area	%	0.5132	0.2304
technician	Technical service for agriculture	Number of agricultural technicians per thousand farmers	person per thousand capita	1.064	0.710
fiscal	Fiscal support for agriculture	Expenditure on agriculture, forestry and water affairs/General budget expenditure	%	0.051	0.178
industry	Industrialization level	Value added of secondary industry/Gross regional product	%	0.4352	0.0830
rengel	Rural Engel coefficient	Rural population’s Expenditure on food, tobacco and alcohol/Rural population’s total expenditure	%	0.3905	0.0728
sun	Daily sunshine duration	Annual sunshine duration/365	h/day	5.6515	1.3609
temperature	Average annual temperature	Sum of daily temperature/365	°C	13.1259	5.7189
rain	Average daily rainfall	Annual rainfall/365	mL/day	2.5943	1.2517
food	Major grain-producing areas	1 = major grain-producing area; 0 = non-major grain-producing area	—	0.4194	0.4941
location	location	1 = eastern regions; 2 = central regions; 3 = western regions	—	2.0323	0.8618
disease	Pest control level	Pest control area/pest occurrence area	%	1.0986	0.3132
conservation	Soil conservation level	Soil conservation area/total farmland area	%	0.4469	0.3010
circulation	The proportion of farmland circulation area	Farmland circulation area/area of farmland being contracted by farmers	%	0.2064	0.1645
structure	Planting structure	Grain-planting area/total planting area	%	0.6555	0.1300
hazard	Hazard level	Hazarded farmland area/total farmland area	%	0.1090	0.0866

.

5. Results

5.1. Characteristic Descriptions

This study first conducted a preliminary analysis of the relationship between the high-standard farmland construction policy and ATFP. Figure 2 (both 2a and 2b contain the same curve of ATFP for convenient comparison) shows that the ATFP grows gently before rapidly after the implementation of the high-standard farmland construction policy (2011–2017); the high-standard farmland area (shown in Figure 2a) and the proportion of high-standard farmland area in China’s total farmland area (shown in Figure 2b) also increased accordingly.

5.2. Baseline Regression Results

This study examined the baseline regression model as shown in Formula (1).

Table 3. Baseline regression estimation results.

Variable	Standard Error by Provincial Clustering	Common Standard Error	Robust Standard Error	Standard Error by Provincial Clustering	Bootstrap1000 Times
	(1)	(2)	(3)	(4)	(5)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.222 ***	0.101 ***	0.101 ***	0.101 **	0.101 **
	(0.0441)	(0.0168)	(0.0218)	(0.0383)	(0.0465)
education	—	0.0063 ***	0.0063 ***	0.0063 ***	0.0063 ***
		(0.0009)	(0.0013)	(0.0011)	(0.0012)
irrigation	—	0.0482 ***	0.0482 ***	0.0482 **	0.0482 *
		(0.0110)	(0.0137)	(0.0227)	(0.0260)
technician	—	0.0043 **	0.0043 ***	0.0043 ***	0.0043 **
		(0.0021)	(0.0012)	(0.0015)	(0.0017)
fiscal	—	−0.0092 *	−0.0092 ***	−0.0092	−0.0092
		(0.0049)	(0.0028)	(0.0109)	(0.0128)
industry	—	−0.0388 **	−0.0388 *	−0.0388	−0.0388
		(0.0171)	(0.0192)	(0.0409)	(0.0465)
rengel	—	−0.0896 ***	−0.0896 ***	−0.0896 **	−0.0896
		(0.0305)	(0.0185)	(0.0356)	(0.0688)
sun	—	−0.0135 **	−0.0135 **	−0.0135 *	−0.0135 **
		(0.0065)	(0.0053)	(0.0067)	(0.0069)
temperature	—	0.0079 ***	0.0079 ***	0.0079 ***	0.0079 ***
		(0.0015)	(0.0012)	(0.0012)	(0.0014)
rain	—	−0.0188 ***	−0.0188 ***	−0.0188 **	−0.0188 **
		(0.0045)	(0.0031)	(0.0073)	(0.0075)
Constant	0.553 ***	0.692 ***	0.692 ***	0.692 ***	0.709 ***
	(0.0097)	(0.0508)	(0.0447)	(0.0605)	(0.0778)
R-squared	0.866	0.937	0.937	0.937	0.976

Notes: * p < 0.1; ** p < 0.01; *** p < 0.001.

listed the estimation results. Column (1) listed the estimation results of the model without any control variables (under provincially clustered standard errors). Columns (2)–(5) listed the estimation results of the model with mediators and control variables under different standard errors (i.e., common, robust, provincially clustered, bootstrap1000 standard errors). Suppose the province and year effects are simultaneously controlled. In that case, the impact of high-standard farmland construction on ATFP is significant at the 1% level, no matter which standard error was adopted, indicating the estimations’ robustness. In addition, the estimated coefficient is 0.101. These results altogether indicated a good policy effect. Specifically, the high-standard farmland construction policy significantly improved ATFP; other conditions were unchanged, and the high-standard farmland construction policy lifted ATFP by 0.105 units on average. Thus, H1 was verified.

Estimations on mediators and control variables under different standard errors are basically the same. Farmers’ average education years, proportion of effectively irrigated area, technical support, financial support, and average annual temperature (whose estimated coefficients are significantly positive) positively influence ATFP. The industrialization rate, rural Engel coefficient, average daily sunshine duration, and average daily rainfall (whose estimated coefficients are significantly negative) negatively impact ATFP.

5.3. Parallel Trend Test and Policy Dynamic Effect

As shown in Figure 3, the regression coefficients’ confidence intervals basically contained zero before the high-standard farmland construction policy’s release (the policy year: 2011) while excluding zero (above zero) after the policy’s release, indicating the non-significance of those coefficients before the policy’s release but positive significance after. These results verified the assumption of the stationary trend before the policy’s release and implied that the high-standard farmland construction policy continuously caused the increase in ATFP.

Additionally,

Table 4. Dynamic regression estimation results.

Variable	No Control Variables Included	Control Variables Included
	(1)	(2)
field × 2006	0.0075	−0.0161
	(0.0203)	(0.0182)
field × 2007	−0.0036	−0.0107
	(0.0293)	(0.0304)
field × 2008	0.0145	0.0089
	(0.0363)	(0.0268)
field × 2009	0.0119	−0.0005
	(0.0292)	(0.0274)
field × 2010	0.0102	−0.0105
	(0.0284)	(0.0252)
field × 2011	0.270 ***	0.168 ***
	(0.0389)	(0.0249)
field × 2012	0.255 ***	0.155 ***
	(0.0424)	(0.0309)
field × 2013	0.250 ***	0.149 ***
	(0.0461)	(0.0347)
field × 2014	0.243 ***	0.146 ***
	(0.0462)	(0.0302)
field × 2015	0.222 ***	0.138 ***
	(0.0503)	(0.0281)
field × 2016	0.193 ***	0.125 ***
	(0.0580)	(0.0322)
field × 2017	0.188 ***	0.112 ***
	(0.0597)	(0.0340)
Control variable	NO	YES
Regional effect	YES	YES
Time effect	YES	YES
Constant	0.550 ***	0.694 ***
	(0.0116)	(0.0679)
R-squared	0.868	0.935

Note: *** p < 0.001.

reported the dynamic estimation results of the influence of high-standard farmland construction policy on ATFP. Column (2) in Table 4 showed that the interaction term’s regression coefficients are significantly positive after the policy’s release. Specifically, the interaction terms’ regression coefficient is 0.168 in the year of the policy release (2011) and then decreased year by year to 0.112 in the sixth year after (2017), implying a weakening yet still positive effect of the policy. One possible explanation of the weakening effect was the decreasing marginal output of improving farmland quality by the high-standard farmland construction policy.

5.4. Robustness Test

5.4.1. Placebo Test by Advancing the Policy’s Release Time Point

To exclude the alternative situation that the increase in ATFP was affected by factors other than the high-standard farmland construction policy (released in 2011), this study conducted the placebo test by fictionalizing the policy release time. The most-used operation is to advance the policy release time. If the new regression coefficient after the policy release was insignificant, it could be implied that the policy release time is not random and caused the explained variable’s dynamics; otherwise, the policy had no effect. We took 2008 and 2010 as fictitious policy release time points. Columns (1) and (2) in

Table 5. Robustness test results.

Variable	Advancing the Policy’s Release Time Point		Alternating the Dependent Variable or Its Measurement		Alternating Core Explanatory Variables	Alternating Control Variables	Alternating Models
	Year 2008	Year 2010	bc92 Method	Agricultural Output Value per Capita	${\mathbf{i}\mathbf{n}\mathbf{v}\mathbf{e}\mathbf{s}\mathbf{t}}_{\mathbf{i}}\times {\mathbf{I}}_{\mathbf{t}}^{\mathbf{p}\mathbf{o}\mathbf{s}\mathbf{t}}$	Lagging by One Stage	Spatial Error Model
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.0068	−0.0416	0.128 ***	4.217 ***	—	0.101 **	0.619 ***
	(0.0205)	(0.0369)	(0.0407)	(1.0420)		(0.0383)	(0.0125)
${\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}}_{\mathrm{i}}\times {\mathrm{I}}_{\mathrm{t}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	—	—	—	—	0.0199 **	—	—
					(0.0093)
Control variable	YES	YES	YES	YES	YES	YES	YES
Regional effect	YES	YES	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES	YES	YES
Constant	0.626 ***	0.705 ***	0.680 ***	5.913 **	0.710 ***	0.692 ***	1.451 ***
	(0.0417)	(0.0555)	(0.0668)	(2.1570)	(0.0642)	(0.0605)	(0.0215)
R-squared	0.972	0.954	0.938	0.909	0.935	0.937	0.954

Notes: ** p < 0.01; *** p < 0.001.

showed that the interaction term’s regression coefficients were insignificant, further verifying the policy effect’s robustness.

5.4.2. Robustness Test by Alternating the Dependent Variable or Its Measurement

Considering that ATFP obtained by different measuring methods might affect the estimation results, this study conducted the robustness test using other methods (i.e., the BC92 method) to measure ATFP. ATFP measured by the BC92 method was brought into the regression model. Column (3) in Table 5 showed the estimated result. This paper additionally brought the agricultural output value per capita into the model as a substitute variable, with Column (4) in Table 5 showing the corresponding estimated result. Both estimated results showed that the interaction term’s regression coefficients were still significantly positive at the 1% level, indicating the robustness of the baseline regression results.

5.4.3. Robustness Test by Alternating Core Explanatory Variables

This study used the amount of investment in comprehensive agricultural development to reflect the extent of high-standard farmland construction, considering the financial investment into improving farmland’s quality consists a crucial component of the high-standard farmland construction project. The core explanatory variables (the interaction term between the high-standard farmland area and the dummy policy variable, ${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$) is alternated by the interaction term between the investment amount in agricultural comprehensive development and the dummy policy variable. As shown by Column (5) in Table 5, the new interaction term (${\mathrm{i}\mathrm{n}\mathrm{v}\mathrm{e}\mathrm{s}\mathrm{t}}_{\mathrm{i}}\times {\mathrm{I}}_{\mathrm{t}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$)’s estimated coefficient is significantly positive at the 1% level, further verifying the robustness of the baseline regression results.

5.4.4. Robustness Test by Alternating Control Variables

Considering the probable existence of the endogenous problem caused by the mutual causality between the explained and control variables, this study applied the operation as alternating control variables by their one-stage lags. As shown by column (6) in Table 5, the interaction term’s estimated coefficients were all significantly positive at the 1% level, further verifying the baseline regression results’ robustness.

5.4.5. Robustness Test by Alternating Models

Given that agricultural policies and conditions are often spatially related, this study also employs a spatial estimation model (specifically, the spatial error model) for estimation. Pre-estimation LM tests indicated that spatial autocorrelation resides mainly in the error rather than in a spatially lagged dependent variable, and the administrative assignment of the program leaves little room for policy spill-overs, so the spatial error specification is preferred to SAR (Spatial Autoregressive Model) or SDM (Spatial Durbin Model) alternatives. The results are presented in column (7) of Table 5. After accounting for spatial effects, the estimated coefficients of the interaction term remain significantly positive at the 1% level, further indicating that the impact of the high-standard farmland construction policy remains relatively stable.

5.5. Heterogeneity Analysis

5.5.1. Heterogeneity in the Agricultural Function Deployment

Considering the policy effect might differ in different functional deployment of agricultural production regions, we divided the sample into two groups—provinces belonging to the major grain-producing region and not. Columns (1) and (2) in

Table 6. Heterogeneity in the agricultural production’s functional deployment and economic geography.

Variable	Agricultural Production’s Functional Deployment		Economic Geography
	Major Grain-Producing Regions	Non-Major Grain-Producing Regions	Eastern Regions	Central Regions	Western Regions
	(1)	(2)	(3)	(4)	(5)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.2010 ***	0.0866 **	0.0775 ***	0.238 **	0.213 **
	(0.0240)	(0.0407)	(0.0261)	(0.0898)	(0.0760)
Control variable	YES	YES	YES	YES	YES
Regional effect	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES
Constant	0.457 ***	0.724 ***	0.611 ***	0.4030	0.924 ***
	(0.1480)	(0.0589)	(0.0852)	(0.2580)	(0.0704)
R-squared	0.944	0.947	0.965	0.946	0.943

Notes: ** p < 0.01; *** p < 0.001.

showed the estimated results. The interaction term’s estimated coefficients in and out of major grain-producing regions were both significantly positive (0.2010 and 0.0866, respectively) at the 1% level, indicating the policy effect on ATFP existed in regions with different functions. The policy had a greater effect on the major grain-producing regions, possibly because these regions had higher levels of specialization and scale management and obtained more agricultural support in technique and finance. Therefore, the co-existence of high-standard farmland construction and major grain-producing region policies brought superimposed effects on ATFP in major grain-producing regions.

5.5.2. Heterogeneity in Economic Geography

The policy effect may be influenced by the agricultural and economic development levels varying geographically. Therefore, this study separately estimated the effect of high-standard farmland construction policy on ATFP in different geographic locations—the eastern, central and western regions. Estimated results shown by Columns (3)–(5) in Table 6 indicated the policy has significantly positive impacts on ATFP in the eastern, central, and western regions; however, such policy effects were the largest on the central region, followed by the western region, and was the least on the eastern region. The reason might be that the central region was an important food-producing region with better accumulations of agricultural infrastructures, and its large-scale farmland could better facilitate farmland management professionalism. Although the eastern region had the highest levels of industrial and commercial development, the industrial and commercial economy might “crowd out” the agricultural economy, leading to a relatively gentle policy effect.

5.5.3. Heterogeneity in Pest and Disease Control Levels

This study also focused on heterogeneity caused by different levels of pest and disease controls. In China, pest and disease control is experiencing transformations from chemical measures to using physical, biological, and green pesticide technologies. Such changes influence agricultural output. Therefore, this study estimated the effect of high-standard farmland construction policy on ATFP in regions with different pest and disease control levels by dividing areas with a control level greater than or equal to 50% into the high-control-level region and areas less than 50% into the low-control group. Columns (1) and (2) in

Table 7. Heterogeneity in the pest and disease control levels and soil and water conservation levels.

Variable	Pest and Disease Control Levels		Soil and Water Conservation Levels
	High-Level Regions	Low-Level Regions	High-Level Regions	Low-Level Regions
	(1)	(2)	(3)	(4)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.1453 ***	0.077 ***	0.121 ***	0.0841 ***
	(0.0210)	(0.0229)	(0.0294)	(0.0212)
Control variable	YES	YES	YES	YES
Regional effect	YES	YES	YES	YES
Time effect	YES	YES	YES	YES
Constant	0.692 ***	0.728 ***	0.667 ***	0.601 ***
	(0.0469)	(0.0924)	(0.0660)	(0.0661)
R-squared	0.948	0.959	0.965	0.948

Note: *** p < 0.001.

showed that the high-standard farmland construction policy positively impacts ATFP in high-control and low-control groups; however, such policy effects are larger in the high-control group. One natural explanation was that pests and diseases in the low-control group harm crops’ growth, reduce the crop yielding amount and quality, and even lead to crop extinction under some extreme situations, which greatly offset the policy effect.

5.5.4. Heterogeneity in Soil–Water Conservation Levels

The farmland’s soil–water environment is critical to agricultural production. The soil water environment can easily suffer from extreme climates such as floods; therefore, soil–water conservation might be crucial for improving ATFP. This study estimated the effects of high-standard farmland construction policy on ATFP in regions with different soil–water conservation levels by dividing the regions with soil–water conservation levels greater than or equal to 50% into the high-conservation-level group, and those with less than 50% into the low-conservation-level group. Columns (3) and (4) in Table 7 indicate that the high-standard farmland construction policy positively impacts ATFP in both high- and low-conservation groups. Still, the impact on the high-conservation-level group is greater. One natural explanation was that farmland’s soil carbon sequestration capacity in the high-conservation-level group was stronger, meaning higher agricultural output.

5.6. Mechanism Analysis

This study further verified the proposed three paths of high-standard farmland construction policy influencing ATFP. This study established several regression models based on the baseline regression model.

Table 8. Mechanism analysis results.

Variable	te1	Agricultural Operation Scale		Planting Structure Adjustment		Agricultural Disaster Reduction
		Scale	Te1	Structure	Te1	Hazard	Te1
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.101 **	0.524 **	0.109 **	0.273 ***	0.106 ***	−0.298 ***	0.116 ***
	(0.0383)	(0.0401)	(0.0424)	(0.0586)	(0.0379)	(0.0579)	(0.0408)
scale	—	—	0.0403 ***	—		—	—
			(0.0022)
structure	—	—	—	—	0.0572 ***	—	—
					(0.0179)
hazard	—	—	—	—	—	—	−0.0621 *
							(0.0286)
Control variable	YES	YES	YES	YES	YES	YES	YES
Regional effect	YES	YES	YES	YES	YES	YES	YES
Time effect	YES	YES	YES	YES	YES	YES	YES
Constant	0.692 ***	−50.43 *	0.674 ***	1.215 ***	0.753 ***	−0.1970	0.702 ***
	(0.0605)	(27.9100)	(0.0644)	(0.1260)	(0.0773)	(0.1380)	(0.0601)
R-squared	0.937	0.94	0.938	0.902	0.938	0.961	0.938

Notes: * p < 0.1; ** p < 0.01; *** p < 0.001.

showed all estimated results. Specifically, the baseline regression model (Model 1) contained the interaction term (between the high-standard farmland area and the dummy policy variable, ${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$) as the explanatory variable, and ATFP as the explained variable. Model 1 also included all the aforementioned control variables and controlled the provincial and time effects, as did the other models. Models 2, 4, and 6 aimed to verify the association between the interaction term and the mediator (i.e., the farmland management area per capita, planting structure, and hazard rate, responsively), using the interaction term as explanatory variable, while using the mediator as explained variable, separately. Models 3, 5, and 7 separately included each mediator into Model 1. Table 8 showed the estimation results of these models (each column displayed the estimation result of the corresponding model). Column (1) in Table 8 showed that policy significantly impacted ATFP (β = 0.101, t = 0.0383). Examinations of the three mediators were as follows.

5.6.1. Mechanism of Agricultural Operation Scale

Column (2) indicated that the policy significantly positively impacted farmland management area per capita (β = 0.524, t = 0.0401). Combined with results in Column (3), in which both the interaction term (β = 0.109, t = 0.0424) and farmland management area per capita (β = 0.040, t = 0.0022) significantly and positively impact ATFP, we inferred that farmland management area per capita partially mediates the policy effect (mediating effect = 0.021, 16.13% of the total effect). Operation-scale expansion accounts for 16.13% of the ATFP effect of high-standard farmland construction, corresponding to a 1.63% productivity gain. H2a was verified.

5.6.2. Mechanism of Planting Structure Adjustment

Column (4) showed that the policy significantly impacted the planting structure (β = 0.273, t = 0.0586). Combined with results in Column (5), in which both the interaction term (β = 0.106, t = 0.0379) and planting structure (β = 0.057, t = 0.0179) significantly and positively impacted ATFP, we inferred that planting structure partially mediates the policy effect (mediating effect = 0.016, 12.80% of the total effect). Planting structure adjustment explains 16.13% of the ATFP boost from high-standard farmland construction, translating into a 1.63% gain in productivity. H2b was verified.

5.6.3. Mechanisms of Agricultural Disaster Reduction

Column (6) showed that the policy negatively impacted the hazard rate (β = −0.298, t = 0.0579). Combined with results in Column (7), in which both the interaction term (β = 0.116, t = 0.0408) and hazard rate (β = −0.062, t = 0.0286) significantly impact ATFP, we inferred that hazard rate partially mediates the policy effect (mediating effect = 0.018, 13.74% of the total effect). Agricultural disaster reduction contributes 16.13% of the ATFP effect driven by high-standard farmland construction, equivalent to a 1.63% increase in productivity. H2c was verified.

5.7. Extensive Analyses: By Supplementing Data from 2018 to 2020

5.7.1. Data Supplement Through Extrapolation and Reclamation Proportion Methods

China’s Bureau of Statistics published data on high-standard farmland construction area up to 2017. To investigate the policy effect after 2017, we applied two different operations (i.e., extrapolation and reclamation proportion methods) to supplement the missing data from 2018 to 2020. We chose the extrapolation method because the data was missing at one end. We also calculated the data through the reclamation proportion method, a proportional addition of high-standard farmland reclamation area released by each provincial reclamation system10. Figure 4 showed that the growth trend of high-standard farmland’s construction area from 2005 to 2020 is relatively stable, and the overall agricultural productivity is also on the rise.

5.7.2. Baseline Estimation Results

This study then carried out baseline regression analysis using the expanded data set of 2005–2020, in which data for high-standard farmland construction area in 2018–2020 were calculated by extrapolation and reclamation proportion methods, respectively. Estimated results are listed in

Table 9. Extensive analysis results by supplementing data of 2018–2020.

Variable	Extrapolation Method	Reclamation Proportion Method
	(1)	(2)
${\mathrm{f}\mathrm{i}\mathrm{e}\mathrm{l}\mathrm{d}}_{\mathrm{i}}\times {{\mathrm{I}}_{\mathrm{t}}}^{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$	0.252 ***	0.248 ***
	(0.0215)	(0.0238)
Control variable	Control	Control
Regional effect	YES	YES
Time effect	YES	YES
Constant	0.643 ***	0.652 ***
	(0.0456)	(0.0432)
R-squared	0.956	0.957

Note: *** p < 0.001.

. Effects of high-standard farmland construction policy on ATFP were significantly positive under the extrapolation method (β = 0.252, t = 0.0215) and reclamation proportion method (β = 0.248, t = 0.0238). Results indicated that the policy effect still existed, considering a longer time range of 2005–2020.

5.7.3. Parallel Trend Test

This study then carried out the parallel trend test using the expanded data set in 2005–2020. Figure 5 displayed results using data supplemented by extrapolation and reclamation proportion methods, respectively. Before the policy’s release (year 2011), the interaction term’s estimated coefficients were non-significant, and their confidence intervals fluctuated around zero. After the policy’s release and up to nine years after, the interaction term’s estimated coefficients became significant, and their confidence intervals were all above zero. Such results verified the assumption of the stationary trend before the policy’s release. They implied that the increase in ATFP was caused by the continuous high-standard farmland construction policy until 2020.

6. Discussion

6.1. Key Findings

First, this study moves beyond prior theoretical conjectures by providing empirical evidence that the high-standard farmland construction policy generates two distinct layers of productivity gains. Using a continuous difference-in-differences design on 2005–2017 data, we estimate an average treatment effect of 10.1%. This figure implies that, during the year 2005–2017, on average, counties with newly completed high-standard farmland projects produce 10.1% more agricultural output per unit of aggregated inputs than their control counterparts in the same year. Because treated plots are retained for every post-completion year up to 2017, this coefficient already embeds the accumulation possible within the observed window; it is therefore best interpreted as a medium-short-run gain rather than a one-year jump. A battery of robustness checks—alternative timing assumptions, respecified variables, and allowance for serial correlation—confirms that this upward shift is not a statistical artifact. When the observation window is extended to 2020, however, the cumulative effect keeps widening rather than decaying. The additional, gradual uplift is consistent with a durable, outward movement of the production frontier that unfolds only after the initial efficiency dividend has been realized. Thus, the policy delivers both: an immediate medium-short-run boost to technical efficiency and a slower-building, yet seemingly permanent, rise in the economy-wide productivity trend. Whether the latter will continue indefinitely or will eventually plateau remains an open question; tracking the trajectory beyond 2020 is therefore an essential task for future work.

Second, through a series of heterogeneity analyses, this study further elucidates variations in the policy effect’s size, reflecting the policy effect’s intrinsic authenticity in different cases within China. Results show the policy effect is more evident in major grain-producing regions in central China, with higher levels of disease–pest control and soil–water conservation. Such a finding implies that the high-standard farmland construction policy better affects ATFP when integrating with other approaches such as major grain-producing region deployment, agricultural machinery, and investment in farmland’s environment quality, such as disease–pest control and soil–water conservation. Such heterogeneity might possibly cause the Matthew effect in high-standard farmland construction. Specifically, regions already holding better geographic economics, disease and pest controls, and soil–water conditions might be more enthusiastic about high-standard farmland construction since they will cost less and benefit more from investing in the construction project. On the other hand, regions with poorer geographic economics, disease–pest control, and soil–water conditions might be reluctant to invest in this construction project under the fear that their investment might be worthless. Meanwhile, for each region, it could be an economic (with a higher output/input ratio) choice to transform the farmland with better conditions to high-standard farmland, which might leave the most challenging problem—transforming the poorer-conditioned farmland to high-standard farmland—to the latter stage of the construction project. Such a situation implies perhaps an over-optimistic estimation of the policy effect of high-standard farmland construction at its early stage. From the perspective of SDG 15 (Life on Land), the observed Matthew effect signals a risk that the most degraded or ecologically fragile parcels—those whose restoration would yield the greatest biodiversity and soil carbon dividends—could be deferred indefinitely, thereby delaying progress toward land-degradation neutrality and ecosystem-service recovery.

Third, this study also verifies the mechanisms of the high-standard farmland construction policy affecting ATFP. This study points out three paths: operation scale increase, planting structure adjustment, and disaster reduction and production protection partially mediating the policy effect. Specifically, when mediating effects of operation scale increase, planting structure adjustment, and disaster reduction and production protection occupy 16.13%, 12.80%, and 13.74% of the total effect, respectively. By conducting this, the current study integrates and verifies previous hypotheses on these paths that comprehensive farmland improvement can result in operation scale increase [29], planting structure adjustment [28,30], and disaster reduction [53,54,57,59,60]. Moreover, the operation scale increase explains marginally more of the policy effect, followed by disaster reduction and planting structure adjustment. Thus, in line with the extant literature [29], the current study emphasizes the path of operation scale increase. Despite the path of disaster reduction being mostly attuned in previous research, this study empirically finds that this path should be further strengthened. Viewed through SDG 2 (Zero Hunger), the dominance of scale expansion over input-saving or pollution-mitigation channels suggests that future policy design could more explicitly reward yield-enhancing yet sustainable practices—such as precision-input adoption, drought-resilient varieties and soil health improvements—thereby coupling ATFP gains with measurable progress toward ending hunger and ensuring food security for all.

Additionally, although ATFP is the core performance metric, the estimated 10.1% policy gain should be read against its environmental co-benefits: the possible consolidated blocks reduced field-edge habitat fragmentation and water-saving irrigation, cut pumping-related CO₂ emissions, improved drainage, and lowered nitrogen run-off concentration [64,65,66]. These indicators together could suggest that the productivity uplift did not come at the expense of ecological carrying capacity, but rather reinforced it—implying a higher long-run ATFP trajectory than the single-year effect implies.

6.2. Policy Implications

The key findings of this study could provide insights for policy-making regarding high-standard farmland construction. First, based on the confirmation of the continually positive policy effect of high-standard farmland construction, policymakers should continue to create a preferable environment and offer the necessary facilities to support the construction project. Beyond productivity, the policy generates co-benefits and trade-offs through environmental and social channels that must be weighed in any appraisal. Land leveling and channel hardening curtail soil erosion immediately, yet the accompanying removal of field ridges, shelterbelts, and small wetlands can erode biodiversity and ecosystem services such as pollination; a long-term tilt toward monoculture amplifies this risk [67]. Wider farm roads and tractor lanes improve machinery access but enlarge the compacted and sealed area, potentially slowing soil carbon accretion [67]. Irrigation upgrades safeguard yields in drought years, yet without pumping quotas they may trigger groundwater over-draft and downstream externalities [67]. Future project design should therefore bring “ecological infrastructure” into the cost–benefit calculation—e.g., retaining vegetated field margins, installing permeable drainage ditches and seasonal wetlands—to sustain the balance between ecological functions and agricultural output.

Considering the policy effect was declining in the early years, we should strengthen the management of soil–water conservation in the whole process, reduce surface disturbance and vegetation damage, effectively control soil–water loss that may be caused, and improve the efficiency of water and water resource utilization. Moreover, considering the policy effect’s hysteric feature, there is a declining trend in the earlier stage but a rising trend later (and probably another declining trend after that since all policies have contemporary boundaries). The construction project’ post-evaluation should be conducted based on an appropriate evaluating time point. To align with SDG 15, post-construction financing mechanisms could earmark a fixed share of proceeds for continuous soil carbon monitoring, biodiversity corridor upkeep, and drought-resilient vegetative cover, ensuring that productivity gains do not inadvertently reverse ecological benefits after the initial investment cycle.

During implementation, land requisition, crop-compensation payments, and construction noise can ignite intra-village disputes; opaque standards may entrench existing power asymmetries. The policy’s grain-first priority can push smallholders out of higher-value cash crops, narrowing income diversity and increasing vulnerability when grain prices swing [67]. Devolving design authority to village collectives through “participatory blueprints” can strengthen community trust, cut post-construction maintenance costs and generate shared gains via collective machinery purchase. In addition, policymakers should allocate adequate resources during the post-construction stage to maintain the policy effect. Since this project needs constant investment during and after construction, it is required to support and guide local communities to raise funds in innovate ways and actively explore diversified financing mechanisms, ahead of receiving central government’s investments, to guarantee the post-construction management and maintenance. Additionally, policymakers should activate the enthusiasm of farmers and village-level organizations participating in the post-construction management and maintenance work, exploring a long-term and diversified post-construction management and maintenance system.

Second, policymakers should be aware of the Mathew effect existing in the heterogeneity of the high-standard farmland construction project. It should be a foreseen challenge that the policy’s implementation would face obstructions in the later stage despite it going smoothly earlier. The central government must take the dominant proportion of the investment instead of letting the regional government self-invest in their construction project, especially in the regions with poorer geographic economics, disease–pest control, and soil–water conditions. More importantly, the government should also design more specialized and concise policies based on the trade-offs of efficiency and equality. For poorer-conditioned regions, it might be a multi-dimensional decision (considering not only the ATFP but also environmental sustainability and household livelihood) between a preferential policy (to stimulate the region’s farmland improvement) or a less preferential policy (to allocate more resources to regions with higher ATFP increases).

Third, it is implied that the mechanism of operation scale increase contributes more than disaster reduction and planting structure adjustment to ATFP. Policymakers could consider releasing more policies to support the operation’s scale increase, especially land circulation. For instance, policies stimulating the agricultural machinery [30] and land circulation market construction and governance [34] could leverage the effect of operation scale increase on ATFP increase. Compared with the operation scale increase, more potential capabilities should be formed through disaster reduction and planting structure adjustment paths. Facilitating policies include (but are not limited to) technological support on disaster–pest prevention, farmland water conservancy infrastructure construction and renovation, agricultural insurance, and grain-planting subsidies.

Fourth, by empirically clarifying how high-standard farmland construction boosts agricultural total factor productivity, this study equips us to benchmark it more rigorously against the EU’s Common Agricultural Policy (CAP) and Japan’s Land Improvement Projects (LIP). Compared with the EU CAP, China’s high-standard farmland program resembles the post-2020 “eco-scheme” pillar in that both condition public payments on plot-level environmental outcomes such as soil carbon stocks and biodiversity strips. A key difference, however, is governance: CAP payments are decoupled from production and require cross-compliance with 17 Statutory Management Requirements, whereas high-standard farmland construction funds are still coupled to area consolidation and are administered through competitive public bidding rather than annual per-hectare entitlements. Consequently, CAP tends to reward maintenance practices, while high-standard farmland construction incentivizes capital-intensive lumpy investments (land leveling, drainage, drip pipes) that generate one-off productivity jumps. Turning to Japan’s LIP, the similarity lies in the engineering approach—both schemes finance irrigation retrofits, sub-soiling and contiguous reshaping. Yet LIP is embedded in a multi-functional rural-development statute that earmarks 30% of funds for amenity preservation and paddy ecosystem services; high-standard farmland construction, by contrast, allocates < 5% of its budget to ecological infrastructure and still lacks a statutory biodiversity quota. The Japanese experience thus suggests that inserting a mandatory “green infrastructure” tranche (e.g., vegetative buffer strips, pollinator corridors) into the high-standard farmland construction cost-list could further raise ATFP through ecosystem-service inputs while meeting SDG 15 targets. Overall, integrating CAP-style decoupled eco-payments and LIP-style multi-functional covenants into the next high-standard farmland construction phase would reconcile productivity growth with weak-sustainability critiques leveled at both CAP and China’s high-standard farmland construction.

Fifth, while the current decomposition attributes 42.7% of the ATFP gain to scale, sowing structure, and disaster-risk channels, a non-negligible share is likely to operate through technological transfer and digitalization. Recent cross-country evidence shows that rural digitization raises agricultural TFP mainly via improved technical efficiency rather than scale expansion. High-standard farmland plots are prerequisites for precision-seeding drones, IoT (Internet of Things) soil moisture sensors and block-chain food-traceability platforms; these tools reduce unit cost variability and accelerate knowledge spill-overs [67]. Reference [33] further demonstrates that agricultural machinery investment exhibits threshold effects on farmer poverty only when complemented by digital extension services, implying that the productivity dividend of mechanization emerges more strongly under an “equipment + digital know-how” package. Integrating digital advisory apps, remote-sensing yield diagnostics and on-farm sensor networks into the post-construction maintenance fund would therefore capture an additional pathway through which the policy lifts ATFP while keeping nitrogen surplus and diesel use below critical limits [68].

Additionally, to safeguard these indirect ATFP drivers, we recommend embedding three environmental performance metrics into the post-2025 project evaluation protocol: (i) field-edge biodiversity index; (ii) carbon intensity of irrigation water; (iii) surplus nitrogen balance. Meeting quantitative thresholds for these indicators should be a prerequisite for central government maintenance subsidies, ensuring that future ATFP gains remain consistent with SDG 2 without breaching SDG 15 boundaries.

7. Conclusions, Limitation and Future Research

This study verifies the role of high-standard farmland construction in increasing ATFP, based on applying continuous DID analysis to China’s provincial panel data from 2005 to 2020. We find that every one-percentage-point increase in the share of land converted to high-standard farmland increases in provincial agricultural total factor productivity (ATFP) by 0.101 percentage-point—an elasticity that survives parallel trend and placebo tests. The gain is not uniform: major grain-producing provinces reap an effect roughly twice that of non-major provinces, central China outperforms the west and east, and areas with ≥50% pest–disease control or soil–water conservation capture significantly larger boosts. About 43% of the total effect is channeled through three mechanisms—farm-size expansion (16.13%), crop-mix optimization (12.80%), and disaster-risk reduction (13.74%)—indicating that capital-intensive land consolidation transmits to productivity mainly via scale, diversification and resilience rather than through input deepening alone. These results demonstrate that large-scale, engineering-based farmland consolidation can raise agricultural total factor productivity without expanding cultivated area, offering a replicable model for other developing countries seeking to balance food security targets (SDG 2) with land-degradation-neutrality commitments (SDG 15). By integrating digital extension, eco-scheme payments and tenure-safe land circulation, governments can convert one-off capital investments into sustained, inclusive productivity gains while maintaining soil carbon stocks and biodiversity corridors.

This study still has several limitations which can enlighten future research directions.

First, more granular time-stamped data are needed to trace the post-construction productivity trajectory; satellite-based monitoring of soil organic carbon, nitrogen surplus, and biodiversity indices over the next decade will reveal whether the observed ATFP gains are compatible with long-term ecological carrying capacity (SDG 15).

Second, provincial panels smooth away within-province heterogeneity across smallholders, large operators and agroecological zones; future work can exploit farm-level micro-data linked to high-standard farm plot boundaries to re-estimate distributional impacts, examine tenure-specific farmer responses—input reallocation, technology adoption and off-farm labor supply that cannot be identified with provincial aggregates. Once farm-level micro-data become accessible, future studies should revisit the policy effect along per capita income and land-tenure dimensions (contract, transfer, cooperative ownership) to uncover household-level heterogeneity that cannot be identified with provincial panels. Research could also test external validity for Tibet, Qinghai, Ningxia and contrasting landscapes such as Heilongjiang reclamation areas versus hilly south-east provinces.

Third, instead of a single “share of high-standard area”, multi-dimensional indicators—field roads, irrigation–drainage capacity, soil–water conservation, and ecological protection levels—should be introduced.

Fourth, continuous provincial series on biodiversity, nitrogen surplus and carbon emissions are currently unavailable; incorporating these variables once OECD-compatible environmental accounts are released will allow direct estimation of “green total factor productivity” gains and verify whether ecosystem-service improvements underpin the observed ATFP uplift.

Fifth, a cost–benefit analysis that incorporates maintenance expenditure and climate-extreme probabilities will clarify the fiscal sustainability of scaling the program to the remaining 60% of cultivable land, thereby guiding policy transfer to other developing countries striving to balance productivity growth with environmental integrity.

Finally, the mediation model omits the “technology transfer & digitalization” channel due to absent province-level data on drones, IoT, and digital extension; future work should add these variables to create a fourth path and test whether digital know-how is the threshold that converts machinery investment into measurable ATFP gains (cf. [33]). Incorporating this path would yield a conceptually complete model and align the evaluation framework with SDG 9 (industry, innovation and infrastructure) while reinforcing SDG 2 and SDG 15.