The Impact of National-Level Modern Agricultural Industrial Parks on County Economies: The Analysis of Lag Effects and Impact Pathways

Abstract

County economies are the cornerstone of China’s economic and social development but face challenges such as a singular industrial structure and the outflow of production factors. As an important policy tool for rural revitalization, the impact mechanism of National-Level Modern Agricultural Industrial Parks (NMAIPs) on county economies remains inadequately explored. This study aims to quantify the dynamic economic effects of the NMAIP policy through rigorous empirical analysis and elucidate the core pathways driving county economic growth. Based on panel data from 44 counties in six central Chinese provinces from 2014 to 2024, this study employs a Multi-Period Difference-in-Differences (DID) model and finds a significant one-year lag effect of the NMAIP policy: in the year following park establishment, county GDP increased by an average of 8.5%, and this positive effect persisted until the fourth year but showed a trend of marginal diminution. Pathway analysis reveals that agricultural scale expansion (measured by gross output value of agriculture, forestry, animal husbandry, and fishery) and production efficiency improvement (measured by the ratio of output value to agricultural expenditure) are the core driving mechanisms, accounting for 48% and 35% of the total effect, respectively. In contrast, the mediating roles of industrial integration (comprehensive index) and industrial structure upgrading (share of agricultural services) were not statistically significant in the short run. The policy lag primarily arises from the conversion cycle of infrastructure investment to economic output, while pathway differences are closely related to the maturity of the county’s agricultural industrial chain and resource allocation efficiency. This study provides robust empirical evidence for optimizing the timing and pathways of the NMAIP policy design: policy effect evaluations require a 1–2 year “window period”; resources should be prioritized for projects that can rapidly enhance scale and efficiency (e.g., scaled planting, technology-driven efficiency gains), laying a solid agricultural foundation before gradually fostering industrial integration. This aligns with the spirit of “avoiding industrial hollowing-out” proposed in the 2024 Central “Thousand Villages Project” and provides the Chinese experience for the policy evaluation and path selection of global agricultural parks.

1. Introduction

China’s county economies support 73% of the national population and 55% of GDP [1], but their development has long faced dilemmas such as a singular industrial structure and the outflow of factors of production. As a core lever of the rural revitalization strategy, modern agricultural industrial parks are endowed with the policy mission of activating county economies. The core logic of this institutional design originates from the explicit orientation in policy documents such as the Notice of the Ministry of Agriculture and Finance on the Creation of National Modern Agricultural Industrial Parks [2]: by clustering modern production factors (e.g., agricultural technology, financial capital, professional talent) and promoting the integration of primary, secondary, and tertiary industries, to achieve an upgrade of the entire “production + processing + service” industrial chain. From the perspective of policy objectives, NMAIPs aim to address the issues of small-scale, fragmented, and weak agriculture—by enhancing production efficiency through the construction of large-scale planting and breeding bases, extending the industrial chain through deep processing, and expanding the diversified value of agriculture combined with leisure agriculture, ultimately forming a county economic growth pole with “agriculture as the foundation, and three industries linked”. Provincial and lower-level parks focus on bridging national strategy with regional realities, for example, Shandong strengthens supply chain resilience with vegetable parks, Sichuan promotes integrated planting and breeding through livestock parks, forming a hierarchical and complementary policy implementation framework. Conceptually, modern agricultural industrial parks constitute a multi-level, multi-type system rather than a single type of economic entity. Among them, NMAIPs are industrial clusters approved by the central government (first batch approved in 2017), dominated by agriculture, integrating planting, breeding, processing, logistics, leisure, cultural tourism, and other links. Their core characteristics are “agriculture as the foundation, integration of three industries, and technology empowerment,” aiming to enhance agricultural competitiveness through scale and intensive operations, driving county economic transformation. Besides the national level, they also include provincial, municipal, and county-level modern agricultural industrial parks. Parks at different levels differ in approval entities, capital investment, industrial scale, and radiation scope: provincial parks are primarily certified by provincial agricultural departments, focusing on regional agricultural characteristics; municipal and county parks focus more on local resource endowments, serving the function of connecting small farmers with large markets. Although these parks at different levels vary in scale, their common feature is that agricultural production is the core foundation, with the leading industry directly related to agricultural production and food processing, rather than being designated as such solely due to their rural location. Their establishment objectives are hierarchical and progressive: NMAIPs aim to create national models of agricultural modernization and explore replicable industrial integration models; local parks focus on solving practical problems such as dispersed agricultural production and short industrial chains.

Globally, similar practices of agricultural industry clustering are widespread, but concepts and models vary due to regional differences. For example, the Netherlands’ “Greenhouse Agriculture Cluster” focuses on the intensive production of high-value-added crops like flowers and vegetables with precision agricultural technology at its core. The US “Agri-Food Innovation Parks” emphasize the cross-border integration of agriculture with technology and finance, promoting the digitalization of the entire chain from production to consumption. Although these international practices do not uniformly employ the term “Modern Agricultural Industrial Park,” they all embody the combination of agricultural dominance and industrial integration—that is, driving the agglomeration of related industries such as processing, logistics, and services by focusing on the agricultural production link, which aligns with the core logic of China’s NMAIPs [3].

By the end of 2021, the central government had approved the construction of parks in five batches, with cumulative investments exceeding CNY 20 billion, driving local matching funds of over CNY 100 billion [4]. However, theoretical debates persist regarding the temporal characteristics and mechanisms of this policy. Existing research predominantly focuses on short-term economic effects. For instance, Xue Qinggen et al. (2022) found a 5.3% GDP growth in the year of park creation using a single-time-point DID model but did not delve into policy lag [5]. Although international experience generally indicates that agricultural parks typically require a 2–3-year cycle from infrastructure investment to industrial agglomeration [6], domestic research lacks systematic empirical testing of its lag effect leveraging the unique policy scenario of “batch creation” (approved in five batches from 2017 to 2021). Regarding pathways, Zhu Yongqi et al. (2024) posited that industrial integration is the core function of the parks, but empirical results showed that its contribution to the county economy was less than 3% [7], and the existing literature lacks quantitative comparison of the specific contribution shares of different pathways like agricultural scale expansion, production efficiency improvement, and industrial integration to county growth. Furthermore, the application of the multi-period DID method in this field is still insufficient, hindering the accurate capture of the dynamic evolution of policy effects. These research gaps constrain the comprehensive understanding and precise evaluation of the NMAIP policy’s effectiveness.

This study aims to address these gaps using a multi-period DID model to quantify the dynamic effects of the NMAIP policy and decompose its mechanism through a mediating effects model. To achieve this goal, this study adopts three core methods: first, exploiting the quasi-natural experiment setting of the batch approval of parks from 2017 to 2021, and constructing a multi-period DID model to compare economic differences between the treatment group (approved counties) and the control group (non-approved counties) across different batches, to precisely identify the temporal characteristics of the policy effect (including immediate impact and lag effects); second, introducing a mediating effects model, using agricultural scale expansion (measured by the gross output value of agriculture, forestry, animal husbandry, and fishery), production efficiency improvement (measured by the ratio of output value to agricultural expenditure), industrial integration (measured by a composite index), and structural upgrading (measured by the share of agricultural services) as transmission pathway variables, and quantifying the contribution weight of each pathway to the total effect; and third, ensuring result robustness through parallel trend tests. The specific objectives are as follows: to verify the lag effect of the NMAIP policy; to reveal its difference from the “project-based” policy immediate effect hypothesis; to identify the effectiveness of pathways like agricultural scale expansion, production efficiency improvement, and industrial integration; and to provide policy recommendations for linking parks with comprehensive rural revitalization in conjunction with the new 2024 Central “Thousand Villages Project” deployment [8]. Theoretically, this study contributes by breaking through the static analysis framework of park economic effects in the existing literature, integrating the temporal dimension into the policy evaluation framework, and enriching the theoretical understanding of agricultural policy economics. Simultaneously, by comparing the effect differences in different pathways, it provides a new perspective for understanding the formation mechanism of agricultural industry clusters—the findings indicate that the synergistic effect of agricultural scale and efficiency improvement is significantly higher than that of industrial integration, aligning with the development pattern characterized by “aggregation first, integration later” [9]. Methodologically, this study pioneers the application of the multi-period DID model to Chinese NMAIP research, effectively identifying causal effects and mitigating regional heterogeneity bias by leveraging the characteristics of batch policy implementation.

The core findings of this study are as follows: County GDP significantly increased by 8.5% in the year following park establishment, and this effect remained statistically significant by the fourth year after establishment but weakened annually (this result contrasts with the “immediate effect” found by Xue Qinggen et al. (2022) [5], revealing the persistent nature of policy impact). Mediating effect tests show that agricultural scale expansion and production efficiency improvement explained 48% and 35% of the total effect, respectively, while the industrial integration and structural upgrading pathways did not reach significance (this finding revises Zhu Yongqi et al. (2024) [7] overemphasis on industrial integration function). Based on these findings, this study proposes the following targeted suggestions: Policy evaluation requires a 1–2-year “window period” to avoid resource misallocation due to short-term performance pressure; simultaneously, priority should be given to supporting agricultural scaling and technology-driven efficiency projects, while delaying blind investment in industrial integration. This highly aligns with the spirit of “consolidating the agricultural foundation and avoiding industrial hollowing-out” proposed in the 2024 Central No. 1 Document [10].

The core indicators selected for this study (county GDP, agricultural total factor productivity, industrial integration level, etc.) were screened based on both policy relevance and data availability: County GDP directly reflects the policy’s effect on activating the county economy, aligning with the core goal of “industrial prosperity” in the Rural Revitalization Strategic Plan [11]. Agricultural total factor productivity and scale indicators precisely correspond to the park’s policy orientation of “improving quality and efficiency”. The industrial integration indicator is used to test the expected function of “industrial chain extension” in policy design. The linkage analysis of these indicators can comprehensively evaluate the park’s achievement level in the three-dimensional goals of “ensuring supply, promoting income increase, and strengthening industry,” providing an objective basis for policy optimization. This study utilizes a sample of 44 counties (cities, districts) in six central Chinese provinces (2014–2024), offering not only a new analytical framework for agricultural policy evaluation but also the Chinese experience for the sustainable development of global agricultural parks.

2. Empirical Methods and Data

2.1. Model Settings

The primary rationale for selecting the Multi-Period Difference-in-Differences (DID) method is that the creation of NMAIPs has the characteristics of batch implementation and non-random assignment (i.e., “quasi-natural experiment” nature). The DID approach effectively controls for unobservable time-invariant confounders and common time trends, thereby more accurately identifying the “net effect” of the policy. Multi-period DID, compared to traditional two-period DID, better addresses the problem of inconsistent policy impact timing across treated units, making it well suited for the policy context of batch approvals from 2017 to 2021 in this study, and allows for the precise characterization of the dynamic trajectory of policy effects (including lag and persistence). Referencing Beck T et al. (2010) [12], Liu Weilin et al. (2021) [13], and Han Liang et al. (2023) [14], the specific model is set as follows:

$$\ln {\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\mathrm{d}\mathrm{i}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+\mathsf{\gamma }\mathrm{t}{\mathrm{x}}_{\mathrm{i}\mathrm{t}}+{\mathrm{v}}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} it$$

(1)

In Equation (1), the dependent variable lngdp_it quantifies local economic development, specifically measured by the logarithm of county-level GDP. did_it is the DID estimator, indicating whether county (city, district) i was approved to establish an NMAIP in year t. Specifically, if county i is approved to establish an NMAIP in year t, then $\mathrm{d}\mathrm{i}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}$ is assigned a value of 1 from year t onwards, and the county is classified into the treatment group; otherwise, it is assigned 0 and classified into the control group. The coefficient β₁ of the DID estimator is the core policy effect of interest in this study, measuring the impact of NMAIPs on county economic development. Additionally, ${\mathrm{x}}_{\mathrm{i}\mathrm{t}}$ represents a series of control variables closely related to local economic development, ${\mathrm{v}}_{\mathrm{i}}$ represents the fixed effect at the county level, ${\mathrm{u}}_{\mathrm{t}}$ represents time fixed effects, and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ represents the regression residual term.

To further examine the dynamic characteristics of the policy effect (especially lag and persistence), an event study model is employed to verify the parallel trends assumption and capture multi-period policy impacts. The model is set as follows:

$${\mathrm{l}\mathrm{n}\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\mathrm{D}}_{\mathrm{i}\mathrm{t}}^{-4}+{\mathsf{\beta }}_2{\mathrm{D}}_{\mathrm{i}\mathrm{t}}^{-3}+\cdots +{\mathsf{\beta }}_8{\mathrm{D}}_{\mathrm{i}\mathrm{t}}^3+{\mathsf{\beta }}_9{\mathrm{D}}_{\mathrm{i}\mathrm{t}}^4+{\mathrm{v}}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(2)

In Equation (2), for a county (city, district) in the j-th year before the implementation of the policy, ${\mathrm{D}}_{\mathrm{i}\mathrm{t}}^{-\mathrm{j}}$ equals 1; for a county (city, district) in the j-th year after the implementation of the policy, ${\mathrm{D}}_{\mathrm{i}\mathrm{t}}^{-\mathrm{j}}$ also equals 1; for all other cases (not j years before or after policy implementation for that county), D is set to 0. Additionally, for counties more than 4 years before implementation or more than 4 years after implementation, D is set to 1. By observing whether the coefficients for the periods before policy implementation (β₁–β₄) are close to 0, one can judge whether the treatment and control groups satisfy the parallel trends assumption before policy implementation (i.e., consistent economic development trends).

To test the pathways through which industrial parks affect county economies, we introduce the causal steps method proposed by Baron and Kenny (1986) [15], using a mediating effects model. The specific models are set as follows:

$$\ln {\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1\mathrm{d}\mathrm{i}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_2{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{v}}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}} i$$

(3)

$${\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1\mathrm{d}\mathrm{i}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\alpha }}_2{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{v}}_{\mathrm{i}}+{\mathrm{u}}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(4)

$$\ln {\mathrm{g}\mathrm{d}\mathrm{p}}_{\mathrm{i}\mathrm{t}}={\mathsf{\theta }}_0+{\mathsf{\theta }}_1\mathrm{d}\mathrm{i}{\mathrm{d}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\theta }}_2{\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\theta }}_3{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathrm{u}}_{\mathrm{t}}{+\mathrm{v}}_{\mathrm{i}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(5)

where ${Mediator}_{it}$ is the mediating variable (agricultural scale expansion, production efficiency, etc.). If ${\mathsf{\beta }}_1$ in Equation (3) is significant, and ${\alpha }_1$ in Equation (4) is significant, and ${\theta }_2$ in Equation (5) is significant, and ${\mathsf{\theta }}_1$ < ${\mathsf{\beta }}_1$, then partial mediation exists. The model verifies the robustness of the mediating effect using the Bootstrap method (500 replications).

Through the above model design, this study can identify the net effect of NMAIPs on county economies and clarify its dynamic characteristics and specific pathways, providing comprehensive empirical evidence for policy evaluation.

2.2. Variable Selection

This study selects county-level GDP (in the logarithmic form, lngdp) as the core dependent variable measuring the county economic development level, primarily based on the indicator’s policy relevance, data representativeness, and international comparability. In terms of policy relevance, GDP is a core indicator reflecting the overall scale and growth vitality of the county economy, directly aligning with the NMAIP policy goal of “driving county economic transformation.” The Notice of the Ministry of Agriculture and Finance on the Creation of National Modern Agricultural Industrial Parks also explicitly lists “county economic growth rate” as a core assessment indicator [2]. In terms of data representativeness, this indicator covers the entire economic sphere (production, circulation, consumption), comprehensively capturing the park’s overall spillover effects on the county economy through pathways like agricultural scale expansion and efficiency improvement, mitigating the limitation of single-industry indicators (e.g., agricultural output value) that struggle to reflect policy spillovers. In terms of international comparability, GDP is a globally recognized measure of economic performance.

To verify result robustness, this study also selects per capita GDP (lnpcgdp) as an alternative dependent variable. This indicator can eliminate the impact of population size differences on total economic output, more accurately reflecting the quality of county economic development and residents’ actual income, and forming complementary verification with the park’s policy orientation of “promoting income increase” (e.g., farmer linkage mechanisms).

Although economic development can be measured in more dimensions such as industrial structure upgrading (e.g., tertiary industry share) and innovation vitality (e.g., number of patents), this study focuses on the policy’s comprehensive growth effect on the county economy, and GDP and its per capita indicator can effectively capture this core objective. Empirically, the regression results for lngdp and lnpcgdp are highly consistent, and they are verified to be significantly correlated with the core pathways, indicating that the selected indicators can fully support the research conclusion on “how the industrial park policy affects county economic growth.” In terms of data availability, data on county-level industrial structure, innovation indicators, etc. have large gaps (especially 2014–2018), while the GDP data are continuous and can be cross-verified through the China County Statistical Yearbook and local statistical bulletins, ensuring the reliability of econometric results.

For the core explanatory variable, this study constructs the policy variable (did) based on the Ministry of Agriculture and Rural Affairs’ annual List of National Modern Agricultural Industrial Parks for Creation: If a county is approved for creation in year t, it is assigned a value of 1 for year t and subsequent years (treatment group); otherwise, it is assigned a value of 0 (control group). Simultaneously, the number of modern agricultural industrial parks within a county (paip) is used as a replacement variable for robustness testing, studying the impact effect of contiguous park development on local economic development.

To exclude interference from other potential factors on local economic development, this study specifically incorporates control variables in two aspects. One aspect is agricultural resource control variables, including the following: ① Agricultural industry development foundation (foad): Agricultural industry development can comprehensively reflect the value of the entire agricultural industrial chain and influences the local economic development levels (Zhao Peihua, 2022) [16]. This study uses the proportion of gross output value of agriculture, forestry, animal husbandry, and fishery in GDP to measure it. ② Agricultural capital input (lnak): The amount of capital input directly affects total agricultural output, thereby significantly impacting the local economic level. Following Cao Fei et al. (2021) [17], this study uses the logarithm of total agricultural machinery power to measure the agricultural capital input level. ③ Agricultural infrastructure (lnele): Rural infrastructure plays a crucial role in agricultural economic development (Qi Wenhao, 2021) [18]. Given that electricity consumption is an important indicator for measuring local infrastructure levels, this study uses the logarithm of rural electricity consumption to measure the agricultural infrastructure level. ④ Agricultural labor input (lnlabor): Studies have pointed out that the number of agricultural laborers plays an important role in promoting the coordinated development of agriculture and the local economy (Liu Weilin et al., 2021) [13]. Therefore, this study uses the logarithm of rural labor resources to measure agricultural labor input. The other aspect is economic environment control variables. Referencing Zhang Guojiang et al. (2019) [19] and Tang Yuehuan et al. (2020) [20], this study selects the following indicators as control variables: ① financial input level (lnfis), quantified using the logarithm of county general public budget expenditure; ② fixed asset investment level (lninvest), reflected using the logarithm of county fixed asset investment; ③ digitalization level (lndigital), measured using the logarithm of the number of county mobile phone users; ④ industrialization level (industry), assessed by calculating the proportion of secondary industry added value in GDP; and ⑤ urbanization level (UR), represented by the county permanent resident urbanization rate. The specific descriptive statistical results of each variable are shown in

Table 1. Descriptive statistical analysis.

Variable Type	Variable Name	N	Mean	SD	Min	Max
Dependent variable	County-level economic development (lngdp)	440	15.48	0.926	11.52	16.32
	Per capita GDP (lnpcgdp)	440	11.36	0.558	9.234	12.88
Core explanatory variable	NMAIP establishment (did)	440	0.378	0.443	0	1
	Number of NMAIPs (paip)	440	0.998	1.684	0	12
Agricultural resource control variable	The foundation of agricultural industry development (foad)	440	31.08	13.57	10.22	66.41
	Agricultural capital input (lnak)	440	12.14	0.985	9.932	14.65
	Agricultural infrastructure (lnele)	440	9.480	0.855	7.178	10.96
	Agricultural labor input (lnlabor)	440	12.80	0.711	11.02	13.74
Economic environment control variable	Level of financial input (lnfis)	440	12.73	0.514	10.66	13.76
	Level of fixed asset investment (lninvest)	440	13.44	0.913	11.75	15.84
	Digitalization level (lndigital)	440	3.876	0.625	1.126	5.232
	Industrialization level (industry)	440	43.85	13.16	15.07	71.84
	Urbanization level (UR)	440	46.65	14.12	20.58	99.46

.

2.3. Data Source

This study selects annual data for relevant variables from 44 county-level administrative divisions in six central Chinese provinces (Shanxi, Anhui, Jiangxi, Henan, Hubei, Hunan) from 2014 to 2024 as the sample. The six central provinces were selected for the following reasons: This region is an important grain-producing region and major agricultural area in China, where the county economy is significantly based on agriculture and faces typical development dilemmas (e.g., industrial structure upgrading pressure, labor outflow). This region is a key area for NMAIP policy implementation (accounting for nearly 30% of the total number of approved parks nationwide), and the experiences from this region are highly representative for understanding the role of agricultural policy in county economies at similar development stages. Additionally, there exists a gradient in economic development across the region (e.g., the average GDP of Hubei counties is significantly higher than that of Shanxi), aiding in observing heterogeneity. The final sample of 44 counties (cities, districts) was obtained through rigorous screening: Data sources include the China County Statistical Yearbook, Shanxi Statistical Yearbook, Anhui Statistical Yearbook, Jiangxi Statistical Yearbook, Henan Statistical Yearbook, Hubei Statistical Yearbook, Hunan Statistical Yearbook, statistical yearbooks of prefecture-level cities in the six provinces, and the 2014–2024 Statistical Communiqués on National Economic and Social Development of each county. Missing data were imputed using interpolation. When compiling statistics, counties where the primary industry’s share of added value was less than 5% were excluded (ensuring the sample is agriculture-dominated), and counties established during the sample period (2014–2024) were excluded (ensuring data continuity). Areas experiencing name changes or administrative adjustments were treated as consistent units. To reduce the potential interference of outliers in county statistical data on estimation results, data cleaning measures were taken, applying a 2% winsorization to all continuous variables to ensure the robustness and reliability of estimation results. The time span (2014–2024) covers the pre-policy implementation period (2014–2016, baseline), the batch implementation period (2017–2021), and the observation period when policy effects are fully manifested (up to 2024, observing lag effects up to 4 years), meeting the DID model’s requirement for comparable data before and after the policy, and providing a data foundation for capturing the complete policy dynamics (including lag effects).

3. Results

3.1. Benchmark Estimation Result

The Difference-in-Differences (DID) estimation results demonstrate the actual impact of “creating NMAIPs” on county economic development. Columns (1) and (2) in

Table 2. Benchmark regression estimation results.

Variable	Standard Period		Lag by One Period
	(1) lngdp|(2) lngdp		(3) lngdp|(4) lngdp
did	0.009	0.051	-	-
	(0.25)	(1.29)
did 1	-	-	0.055 ***	0.085 *
			(22.43)	(1.93)
industry	-	0.004	-	0.003
		(0.44)		(0.43)
ur	-	0.001	-	0.001
		(0.45)		(0.34)
Inak	-	0.063 *	-	0.065 *
		(1.73)		(1.77)
Inele	-	0.037	-	0.033
		(0.50)		(0.42)
foad	-	−0.024 ***	-	−0.024 ***
		(−3.21)		(−3.26)
Infis	-	−0.091	-	−0.088
		(−1.27)		(−1.25)
Ininvest	-	−0.031	-	−0.036
		(−1.20)		(−1.38)
Inlabor	-	0.064	-	0.072 *
		(1.57)		(1.99)
Indigital	-	0.072 *	-	0.069 *
		(1.77)		(1.69)
Constant	14.116 ***	14.328 ***	14.116 ***	14.314 ***
	(559.92)	(8.32)	(560.90)	(8.67)
Observations	440	440	440	440
R-squared	0.611	0.975	0.611	0.975
Number of id1	44		44

Note: Robust t-statistic is in parentheses. * and *** indicate significance at the 10% and 1% levels, respectively.

show that, controlling for county and year fixed effects, the “creation of NMAIPs” policy in its implementation year (did) did not exert a significant impact on the economic development of the host county, indicating an insignificant immediate spillover effect of the policy. This differs from the theoretical expectation that “creating NMAIPs can directly promote county economic development.” A potential explanation is that, in the initial stage of policy implementation, NMAIP infrastructure construction remains incomplete, the agricultural industrial chain is underdeveloped, and industrial agglomeration effects have yet to materialize, resulting in limited spillover and driving effects on the county economy, preventing the immediate manifestation of the policy effect [21].

To verify the lag characteristic of the policy effect, this study introduces a one-year lagged policy variable, did1, and re-estimates its impact on county economic development. Column (4) in Table 2 shows that the estimated coefficient of the core explanatory variable did1 on county economic development is positive and statistically significant at the 10% level. As control variables such as agricultural resource endowments (e.g., agricultural capital input, agricultural infrastructure) and economic environmental factors (e.g., financial input level, digitalization level) are incrementally incorporated in the model, the coefficient of did1 increases from 0.055 to 0.085. This indicates that “creating NMAIPs” can significantly promote county economic development in the year following policy implementation, increasing county GDP by approximately 8.5%. This outcome may arise because, one year after policy implementation, the park construction of NMAIPs is more complete, hardware and software facilities are progressively established, and industrial agglomeration effects emerge, fostering advantageous industries, and generating spillover and driving effects for the county economy.

Regarding the control variables, several exhibit significant effects on county economic development, specifically the following variables: The coefficient of agricultural capital input (lnak) is positive and significant at the 10% level, indicating that an increase in capital inputs like total agricultural machinery power can effectively promote county economic development, consistent with the role of capital in enhancing production efficiency in the process of agricultural modernization [22]. The coefficient of agricultural labor input (lnlabor) is significantly positive at the 10% level, indicating that an adequate supply of rural labor resources is an important driver for county economic growth, particularly in agriculture-dominated counties, where labor quantity and quantity directly influence the expansion of agricultural production scale and industry development. The estimated coefficient of digitalization level (lndigital) is significantly positive at the 10% level, suggesting that the diffusion of digital technologies such as mobile phones can foster county economic growth by promoting information flow and optimizing resource allocation. In contrast, the coefficient for the foundation of agricultural industry development (foad) is negative and significant at the 1% level, and this may stem from areas with a strong agricultural base often having a high primary sector share, coupled with less developed secondary and tertiary sectors and insufficient industrial modernization, collectively constraining efficient county economic development [23].

Furthermore, the estimated coefficients of control variables such as financial input level (lnfis), fixed asset investment level (lninvest), industrialization level (industry), and urbanization level (UR) were statistically insignificant. This may be related to the uneven development of sample counties in these aspects, such as the inefficient use of fiscal funds and suboptimal fixed asset investment structures in some counties, preventing their potential growth-promoting effects from being achieved.

In summary, the benchmark regression results not only confirm that NMAIPs have a significant lagged positive effect on county economic development but also reveal multiple driving factors of county economic growth through the analysis of control variables, providing a basis for further exploration of policy pathways.

3.2. Robust Test

The validity of the DID estimator hinges on satisfying the parallel trends assumption. This assumption requires that, before the implementation of the “creation of NMAIPs” policy, the economic development trends of the treatment group (counties approved to create parks) and the control group (non-approved counties) exhibit parallel trends. This is a prerequisite for obtaining unbiased DID estimates. This study uses the event study method for parallel trends testing and employs a model to capture multi-period policy effects (model specification detailed in Section 2.1 Model Design).

Therefore, the software Stata 17.0 is employed for data processing and model estimation. Drawing on prior research, this study will not use the period before the implementation of the policy as the benchmark for comparing other periods or groups, in order to avoid severe multicollinearity issues in model calculation. To test the parallel trends assumption and examine the multi-period dynamic effects of establishing NMAIPs, this study incorporates the interaction terms between year dummies (spanning 4 years before to 4 years after policy implementation in each county/city/district) and the treatment group as explanatory variables in the regression. Subsequently, the estimated coefficients are graphically analyzed (Figure 1) to visually assess the satisfaction of the parallel trends assumption and gain deeper insights into the multi-period impact of the policy implementation.

Figure 1 presents the estimation results of the event study. The vertical axis represents the regression coefficient β for each time point (reflecting the economic development difference between the treatment and control groups), and the horizontal axis indicates the time relative to policy implementation (negative values denote years prior to implementation, 0 denotes the implementation year, and positive values denote years post-implementation). Dashed lines represent the 90% confidence intervals. A confidence interval encompassing 0 indicates no statistically significant difference in economic development levels between the two groups at that time point. Specifically, the coefficients for 4 to 1 year before policy implementation (−4 to −1 periods) are all close to 0, and their confidence intervals include 0, confirming the validity of the parallel trends assumption. The confidence interval for the coefficient at the implementation year (period 0) also includes 0, consistent with the benchmark result of an insignificant effect in the implementation year. The coefficients for 1 to 3 years after implementation (1 to 3 periods) are positive and statistically significant, with confidence intervals excluding zero, indicating a significant lagged policy effect emerging in year 1 and persisting through year 3. The coefficient for period 4 and beyond is statistically insignificant (confidence interval includes 0), indicating a gradual weakening of the policy effect.

The specific coefficient estimates further corroborate this pattern: The coefficients for the pre-treatment periods (−4 to −1 periods) are −0.012, 0.008, 0.015, and 0.006 (all insignificant at p > 0.1), indicating no statistically significant pre-treatment differences in trends, satisfying the parallel trends assumption, and supporting the validity of the DID estimates. The coefficient at implementation year (period 0) is 0.021 (p = 0.23), again verifying the conclusion of “no significant effect in the policy implementation year”. The coefficient for year 1 after implementation (period 1) is 0.085 (p < 0.01), for year 2 (period 2) is 0.072 (p < 0.05), and for year 3 (period 3) is 0.053 (p < 0.1), indicating that the positive effect of the policy on the county economy appears with a 1-year lag and gradually weakens but remains statistically significant over the next 2 years. The coefficient for year 4 after implementation (period 4) is 0.028 (p = 0.31), which is no longer significant, indicating that the policy effect tends to disappear by year 4. Figure 1 clearly shows that coefficients for periods before policy implementation (−4 to −1) are not significantly different from 0, strictly satisfying the DID parallel trends assumption, and ensuring the reliability of the benchmark regression results. The dynamic pattern of the policy effect (appearing with a 1-year lag, persisting until year 3, and weakening in year 4) is empirically robust.

Potential explanations for the diminishing policy effect include the following: On the one hand, NMAIPs’ development relies on county-level resources, and the limited land, labor, and other resource availability constrains continuous industrial scale expansion, leading to diminishing marginal returns from the policy. On the other hand, as the parks’ demonstration effect diffuses, neighboring counties may emulate their development models, diluting the policy spillover effects for the original treatment group and attenuating its direct contribution. Regarding long-term sustainability, future policy adjustments that alleviate resource constraints (e.g., cross-county resource integration, technology-driven improvements in factor efficiency) could stabilize or even slightly rebound the policy effect beyond year five. Conversely, persisting with existing production models without upgrading would likely lead to continued weakening until stabilization.

Note that Figure 1 depicts the average treatment effect of the policy on the 44 counties, and does not mean that the results for each county are completely identical. Owing to heterogeneity across counties in resource endowments (e.g., agricultural foundation, infrastructure) and industrial structure, the intensity and duration of policy effects may vary. However, for the purpose of overall policy evaluation, this study focuses on the average treatment effect of the “creation of NMAIPs” policy—despite individual differences, its overall trajectory (appearing with a 1-year lag, lasting about 3 years, gradually weakening) provides robust insights for policy design.

Furthermore, concerning the impact of policy on park sustainability, sustained national-level support is crucial. Current policies encompass not only initial construction funding (central cumulative investment exceeding CNY 20 billion) but also guide operations through dynamic evaluation mechanisms (e.g., park annual assessments). However, the specific role of management models has not been directly evaluated in this study. Practically, parks adopting a hybrid management model (government guidance, enterprise leadership, farmer participation) generally exhibit longer-lasting policy effects compared to purely government-operated models (a preliminary observation suggests that, in the three sample counties utilizing the hybrid model, the policy effect remained significant until year five; however, the small sample size necessitates verification with larger samples). This suggests that flexible management models may extend the policy impact cycle. Future research should further evaluate the synergistic effects of different management models and policy support for optimizing long-term operational strategies.

In summary, the parallel trends test results show that the conclusion in the benchmark regression that “the industrial park policy has a 1-year lag effect and long-term marginal diminution” is robust, and its long-term impact may be jointly affected by resource constraints and management mechanisms, providing a more comprehensive basis for dynamic policy adjustment.

3.3. Analysis of Pathways Through Which NMAIPs Boost County Economic Development

After confirming that NMAIPs have a significant lagged positive effect on the county economy, to further examine how the “creation of NMAIPs” policy boosts county economic development [24], this study draws on the mediating effects model approach for corresponding tests to reveal the pathways of policy implementation (model specification detailed in Section 2.1 Model Design). This study uses the logarithm of gross output value of agriculture, forestry, animal husbandry, and fishery; the proportion of output value of agricultural services in the gross output value of agriculture, forestry, animal husbandry, and fishery; the level of integration of the three industries; and the logarithm of the ratio of gross output value of agriculture, forestry, animal husbandry, and fishery to agricultural expenditure as proxy variables for agricultural industry scale expansion (lnagr), agricultural production efficiency improvement (lnefficiency), enhancement of integrated development level of the three industries (aiq), and upgrading of agricultural industrial structure (aiu), respectively.

3.3.1. Mediating Variable Selection

(1) Upgrading of agricultural industrial structure (aiu): The literature commonly employs the ratio of tertiary industry output value to secondary industry output value as the measurement standard [25], reflecting the degree of optimization and upgrading of the industrial structure. In the agricultural field, this study draws on the studies by Gan Chunhui (2011) [26] and Ma Yuting (2023) [27] and, considering data availability, uses the share of the output value of agricultural services in the gross output value of agriculture, forestry, animal husbandry, and fishery to measure the agricultural industrial structure. An increase in this share reflects a rise in agricultural servitization, thereby indicating the optimization and upgrading of the agricultural industrial structure.

(2) Agricultural industry scale expansion (lnagr): Agricultural output level is a core indicator for measuring agricultural development scale and is also closely linked to local economic development. The literature lacks a consensus on how agricultural industry scale expansion affects local economic development. This study refers to Zhang Bosheng et al. (2021) [28], utilizing the logarithm of the county-level gross output value of agriculture, forestry, animal husbandry, and fishery to measure agricultural industry scale.

(3) Agricultural production efficiency (lnefficiency): Numerous studies exist on analyzing China’s agricultural production efficiency. Estimates of China’s agricultural production efficiency vary depending on the input–output indicators used. This study draws on Wu Yuhuan et al. (2022) [29] and Zheng Hongyun et al. (2022) [30], employing the logarithm of the ratio of gross output value of agriculture, forestry, animal husbandry, and fishery to government financial expenditure on agriculture, forestry, and water conservancy to measure agricultural production efficiency.

(4) Level of integrated development of the three industries (aiq): The level of three industry integration is closely linked to the overall agricultural production efficiency. A key focus of this study is whether NMAIP creation promotes industrial integration through enhanced agricultural modernization, thereby affecting local economic development. Currently, no universally accepted indicator system exists for measuring the level of industrial integration. This study refers to Chen Xueyun (2018) [31] and Cao Fei et al. (2021) [17], applying the entropy weight method to comprehensively consider three dimensions—industrial scale, growth situation, and performance indicators—to measure the level of integrated development between the primary industry and secondary/tertiary industries.

In information theory, entropy measures the disorder or uncertainty in a system. Generally, a larger entropy value means the system is more disordered and contains less information; conversely, a smaller entropy value indicates a more ordered system carrying more information [32]. Due to differences in indicator levels and positive/negative indicators, to derive a more objective evaluation level of county-level three-industry integration development, this study employs the entropy weight method to calculate the weight of each indicator. The entropy weight method determines indicator weights based on their information entropy, yielding relatively objective weights. The specific method is as follows:

The raw data for positive and negative indicators are standardized separately:

$$X_{ij}^{\prime }=\left\{\begin{array}{c}{\displaystyle \frac{X_{ij}-\mathrm{m}\mathrm{i}\mathrm{n}{X}_{ij}}{\mathrm{m}\mathrm{a}\mathrm{x}{X}_j-\mathrm{m}\mathrm{i}\mathrm{n}{X}_j}},\hspace{1em}\mathrm{position}\mathrm{indicator}\\ {\displaystyle \frac{\mathrm{m}\mathrm{a}\mathrm{x}{X}_j-X_{ij}}{\mathrm{m}\mathrm{a}\mathrm{x}{X}_j-\mathrm{m}\mathrm{i}\mathrm{n}{X}_j}},\hspace{1em}\mathrm{negative}\mathrm{indicator}\end{array}\right.$$

(6)

The entropy value for each indicator is calculated as follows:

$${\mathrm{e}}_{\mathrm{j}}=-{\displaystyle \frac{1}{\mathit{ln}n}}{\sum }_{i=1}^n{p}_{ij}\mathit{ln}\left(p_{ij}\right),\hspace{1em}0\le e_j\le 1$$

(7)

The differentiation coefficient is calculated as follows:

g_i = 1 − e_j

(8)

The final weight of each indicator is calculated to compute the comprehensive score of the three-industry integration development level for each county:

$${\mathrm{W}}_{\mathrm{j}}={\displaystyle \frac{{\mathrm{g}}_{\mathrm{j}}}{\sum _{\mathrm{i}=1}^{\mathrm{m}}{\mathrm{g}}_{\mathrm{j}}}},\hspace{1em}j=1,2,3\cdots ,m$$

(9)

In summary, the three-industry integration development level evaluation system set in this study is shown in

Table 3. An evaluation index system for the development level of the integration of the primary, secondary, and tertiary industries.

Subsystem	First-Level Indicator	Secondary Indicators	Unit	Indicator Attribute	Index Weight
The primary industry	Scale indicator	The proportion of the added value of the primary industry in GDP	%	Positive	0.172
	Growth indicator	The growth rate of added value of the primary industry	%	Positive	0.052
	Performance indicators	Grain yield per unit area	Kilograms per hectare	Positive	0.093
The secondary industry	Scale indicator	The proportion of the added value of the secondary industry in GDP	%	Positive	0.684
	Growth indicator	The growth rate of added value of the secondary industry	%	Positive	0.285
	Performance indicators	Labor productivity in the secondary industry	CNY ten thousand per person	Positive	0.029
The third industry	Scale indicator	The proportion of the added value of the tertiary industry in GDP	%	Positive	0.172
	Growth indicator	The growth rate of added value of the tertiary industry	%	Positive	0.193
	Performance indicators	Labor productivity in the tertiary industry	CNY ten thousand per person	Positive	0.634

Note: labor productivity = industrial added value/number of employees.

:

Table 4. Descriptive Statistics of Mediating Variables.

Variable Type	Variable Name	N	Mean	SD	Min	Max
Dependent variable	County-level economic development (lngdp)	440	14.48	0.847	11.62	16.32
Explanatory variable	The establishment of the National Modern Agricultural Industrial Park (did)	440	0.325	0.469	0	1
Mediating variable	The level of integration of primary, secondary and tertiary industries (aiq)	440	0.464	0.0997	0.289	0.727
	Upgrading of the agricultural industrial structure (aiu)	440	4.832	3.220	0.39	12.43
	Expansion of agricultural industry scale (lnagr)	440	13.19	0.712	11.09	14.39
	Agricultural production efficiency (lnefficiency)	440	2.239	0.635	−0.0731	3.940

reports the descriptive statistical results for the main variables (sample size N = 440). For the dependent variable, the mean of county economic development (lngdp) is 14.48 (log form), and standard deviation is 0.847, with a minimum of 11.62 and a maximum of 16.32. This data feature reflects moderate differences in the economic scale of sample counties, consistent with the reality of uneven economic development in central counties. The mean of the core explanatory variable park creation (did) is 0.325, meaning that 32.5% of sample counties were approved as NMAIPs during the observation period. This proportion aligns with the actual implementation of the batch advancement policy, providing a reasonable grouping basis for subsequent policy effect evaluation. The values of mediating variables are as follows: Agricultural industry scale (lnagr), measured by the logarithm of gross agricultural output value, has a mean of 13.19, a standard deviation of 0.712, and a range of 11.09 to 14.39, showing significant differences in agricultural scale among counties. Agricultural production efficiency (lnefficiency) has a mean of 2.239, a standard deviation of 0.635, and a wide distribution range of −0.073 to 3.94, implying obvious gradient differences in technology application and management levels among counties. Industrial integration (aiq) is an index indicator, with a mean of 0.464, a standard deviation of 0.0997, and data concentrated in the 0.289 to 0.727 interval, indicating that the overall industrial integration level of central counties is in the medium range. Agricultural structure upgrading (aiu) has a mean of 4.832, but a high standard deviation of 3.220, and a range of 0.39 to 12.43, revealing huge differences among counties in the process of agricultural servitization transformation.

Further analysis of these data yields the following insights: The range of county GDP logarithm is 4.7, equivalent to an actual GDP difference of about 110 times, highlighting the necessity of controlling fixed effects in this study to exclude individual heterogeneity interference. About one-third of counties received park policy support, providing a sufficient basis for comparison between treatment and control groups in the DID model. Judging from pathway variable characteristics, the standard deviations of agricultural industry scale (lnagr) and production efficiency (lnefficiency) reach 54% and 28% of their means, respectively, suggesting these variables may be highly sensitive to economic impact. The coefficient of variation for agricultural structure upgrading (aiu) is 66.6%, significantly higher than other variables, which also provides a possible explanation for its failure to pass the mediation test. Industrial integration (aiq) data are relatively stable, echoing the “short-term difficulty in showing results” conclusion in the benchmark regression. Particularly noteworthy is the negative value for agricultural production efficiency (lnefficiency) (minimum: −0.073), indicating that some counties experience inverted agricultural input–output ratios, and providing a realistic basis for the park policy’s focus on “technology-driven efficiency gains.” Simultaneously, the agricultural structure upgrading indicator (aiu) shows a right-skewed distribution (mean > median), implying that a few advanced counties raise the overall level, while most counties are still in the early stages of transformation.

3.3.2. Pathway Test Results

Table 5. Estimation results of the role path of modern agricultural industrial parks in promoting county-level economic development 1.

Variable	Expansion of Agricultural Scale		Improvement of Agricultural Production Efficiency		County-Level Economic Development
	(1)	(2)	(3)	(4)	(5)	(6)
	lnagr	lnagr	lnefficiency	lnefficiency	lngdp	lngdp
did1	0.067 *	0.084 **	0.182 **	0.171 **	0.044 *	0.055 *
	(1.73)	(2.49)	(2.09)	(2.32)	(1.69)	(1.75)
Inagr	-	-	-	-	1.087 ***	-
					(4.40)
lnefficiency	-	-	-	-	-	0.154 ***
						(3.37)
Inak	-	−0.001	-	−0.086	−0.014	0.042
		(−0.01)		(−0.46)	(−0.28)	(0.95)
industry	-	0.005 *	-	−0.060	−0.005	0.832 ***
		(1.78)		(−0.13)	(−0.90)	(2.89)
Ininvest	-	−0.071	-	−0.116 *	0.016	−0.015
		(−1.43)		(−1.93)	(0.71)	(−0.43)
Indigital	-	−0.059	-	−0.050	0.001	−0.144
		(−1.00)		(−0.45)	(1.39)	(−0.24)
foad	-	0.485 **	-	0.019 ***	−0.041 ***	−0.028 ***
		(2.14)		(3.47)	(−6.00)	(−4.70)
Inele	-	0.038	-	−0.016	−0.007	0.062
		(0.45)		(−1.21)	(−0.13)	(0.79)
ur	-	−0.002	-	0.016	−0.001	−0.002
		(−0.37)		(1.64)	(−0.16)	(−0.77)
Infis	-	−0.281	-	0.617	−0.032	−0.043
		(−1.14)		(1.48)	(−0.86)	(−0.63)
Inlabor		0.001	-	−0.116	−0.024	0.002
		(0.58)		(−0.82)	(−0.42)	(0.26)
Constant	12.927 ***	9.372 ***	2.372 ***	5.655 ***	2.256 ***	6.858 ***
	(629.02)	(4.34)	(43.66)	(4.28)	(2.93)	(2.89)
Observations	440	440	440	440	440	440
R-squared	0.690	0.974	0.090	0.817	0.989	0.978
Number of id1	44		44

Note: Robust t-statistic is in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

and

Table 6. Estimation results of the role path of modern agricultural industrial parks in promoting county-level economic development 2.

Variable	The Level of Integration of Primary, Secondary, and Tertiary Industries		Upgrading of the Agricultural Industrial Structure		County-Level Economic Development
	(1)	(2)	(3)	(4)	(5)	(6)
	aiq	aiq	aiu	aiu	Ingdp	Ingdp
did1	−0.021	−0.016	0.472	0.587	0.072 **	0.081 **
	(−1.63)	(−1.49)	(0.68)	(0.83)	(2.03)	(2.09)
aiq	-	-	-	-	−0.221	-
					(−0.52)
aiu	-	-	-	-	-	−0.005
						(−0.77)
industry	-	0.003 **	-	−0.007	0.004	0.004
		(2.36)		(−0.22)	(0.59)	(0.44)
ur	-	0.001	-	−0.074	0.001	0.000
		(0.91)		(−0.69)	(0.41)	(0.19)
Inak	-	0.004	-	1.855 *	0.066 *	0.074
		(0.23)		(1.91)	(1.71)	(1.63)
Inele	-	−0.016	-	−0.049	0.030	0.034
		(−0.91)		(−0.07)	(0.36)	(0.43)
foad	-	−0.001	-	0.002	−0.025 ***	−0.024 ***
		(−1.39)		(0.07)	(−3.16)	(−3.23)
Infis	-	−0.002	-	−0.707	−0.088	−0.096
		(−0.08)		(−0.60)	(−1.26)	(−1.32)
Ininvest	-	−0.008	-	0.414	−0.038	−0.035
		(−0.66)		(−0.80)	(−1.54)	(−1.02)
Inlabor	-	0.042 **	-	−0.083	0.082 *	0.072 *
		(2.29)		(−0.12)	(1.96)	(1.99)
Indigital		−0.008	-	0.074	0.067 *	0.070 *
		(−0.66)		(0.12)	(1.75)	(1.72)
Constant	0.423 ***	0.116	3.329 ***	0.396	14.340 ***	14.313 ***
	(44.22)	(0.25)	(10.56)	(0.02)	(8.50)	(8.34)
Observations	440	440	440	440	440	440
R-squared	0.619	0.780	0.238	0.698	0.975	0.975
Number of id1	44		44

Note: Robust t-statistic is in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

present the mediating effect test results. Table 5 focuses on agricultural scale expansion and production efficiency improvement pathways. Table 6 focuses on industrial integration and agricultural structure upgrading pathways.

Table 5 indicates that agricultural scale expansion and production efficiency improvement are significant mediating pathways: The coefficient of did1 on lnagr is positive and significant at the 5% level (0.085), indicating that the parks significantly promoted agricultural scaling. After controlling for lnagr, the direct effect of did1 on county GDP decreased from 0.085 to 0.044, with the mediating effect accounting for approximately 48%, confirming the presence of economies of scale effect formed through land concentration and leading enterprise driving. Regarding production efficiency improvement, the coefficient of did1 on lnefficiency is positive and significant at the 5% level (0.171), indicating that parks improved production efficiency through technology promotion. After controlling for inefficiency, the direct effect of did1 on GDP decreased to 0.055, with the mediating effect accounting for approximately 35%, reflecting cost-saving and efficiency gains of technologies like smart irrigation and agricultural machinery integration.

Table 6 shows statistically insignificant mediating roles of industrial integration and agricultural structure upgrading did not reach significance, as judged by p-values exceeding 0.10: For industrial integration (aiq), the coefficient of did1 on aiq is −0.016 (p = 0.149), indicating insignificant promotion of industrial integration by the parks. The coefficient of aiq on GDP is −0.221 (p = 0.52), further supporting the ineffectiveness of this pathway, potentially due to weak industrial foundations and insufficient cold chain logistics support in central counties. For agricultural structure upgrading (aiu), the coefficient of did1 on aiu is 0.587 (p = 0.83), and the coefficient of aiu on GDP is −0.005 (p = 0.77), suggesting that the transition towards agricultural servitization is lagging and has not yet emerged as an economic driver.

3.3.3. Results’ Discussion

The observed pathway differences are closely linked to the industrial development stage of central counties: In the initial stage of policy implementation, resources are more concentrated in the agricultural production segment, rapidly enhancing agricultural productivity through scale expansion and efficiency gains, generating a direct boot for the county economy in the short run. In contrast, industrial integration and industrial structure upgrading depend on the long-term development of supporting industrial chain infrastructure (e.g., processing bases, cold chain logistics) and service ecosystems (e.g., e-commerce platforms, cultural tourism services). These conditions remain underdeveloped in the central counties, hindering significant short-term effects.

This finding offers targeted insights for policy optimization—priority should be accorded to supporting production-side initiatives such as scaled planting, improved seed breeding, and agricultural machinery promotion, accumulating technology and capital through the consolidation of the agricultural production base. Only after the production capacity stabilizes should efforts to foster industrial integration be gradually intensified, focusing on improving industrial chain support and cultivating integration vehicles, mitigating policy inefficiency stemming from resource misallocation.

4. Discussion and Conclusions

4.1. Discussion

This study systematically examines the impact mechanism of NMAIPs on county economies through multi-period DID and mediating effects models. The core findings complement and refine existing research. Regarding the temporal characteristics of policy effects, this study confirms that the economic effect of the park policy has a 1-year lag period, persisting until the fourth year but exhibiting marginal diminution (e.g., GDP growth of 8.5% in year 1, declining to 5.3% by year 3). This contrasts with the “immediate effect” identified by Xue Qinggen et al. (2022) [5], revealing the cyclical nature of agricultural parks from infrastructure investment to industrial cluster formation (typically requiring 2–3 years). This result aligns with international experience (e.g., the development cycle of Dutch greenhouse clusters [33]), suggesting that agricultural policy evaluation necessitates sufficient “window periods” to avoid resource misallocation driven by short-term performance pressures. Regarding pathways, the findings indicate that agricultural scale expansion (48% contribution) and production efficiency improvement (35% contribution) constitute the core driving pathways, whereas the mediating roles of industrial integration and structural upgrading were insignificant. This difference is closely linked to the characteristics of the central counties, characterized by a solid production-side foundation but a weak integration ecosystem. On the one hand, parks achieve scale production, through land concentration and leading enterprise facilitation, reducing unit costs (e.g., contiguous planting increases fertilizer utilization by 15–20%). On the other hand, technology promotion (e.g., smart irrigation, improved seed varieties) directly enhances resource allocation efficiency, increasing output per mu by 8–12%. However, constrained by weak industrial foundations (average secondary industry share in sample counties of 41.56%) and insufficient cold chain logistics coverage (<30%), industrial integration proves difficult to achieve in the short term. This contrasts with Zhu Yongqi et al.’s (2024) [7] expectation of “industrial integration as the core function,” suggesting the need for phased policy advancement. Furthermore, this study revealed that management models significantly influence the persistence of policy effects: Parks adopting a hybrid model (government guidance, enterprise leadership, farmer participation) sustained significant policy effects until year five, whereas purely government-operated models exhibited weakening by year three. This suggests that flexible management models can extend the policy impact cycle, offering insights for long-term operations.

A point warranting explanation is the relationship between the negative coefficient of agricultural industry development foundation (foad) in the benchmark regression and the significant positive agricultural scale expansion (lnagr) pathway in the mediating effect test. Superficially, the negative and significant coefficient of foad (share of gross agricultural output in GDP) appears to suggest that counties with a high agricultural share slower growth (Table 2), seemingly conflicting with the finding that parks drive economic growth by promoting agricultural scale expansion (lnagr) (Table 5). In-depth analysis reveals that the negative foad effect captures the static structural characteristics of the county economy: Prior to policy implementation, a higher primary industry share is often associated with underdeveloped secondary/tertiary sectors having insufficient industrial modernization. This static structure inherently constrains efficient county economic development. The positive mediating effect of lnagr, however, reveals the dynamic incremental change driven by park policy intervention: Through policy interventions (e.g., land transfer support, leading enterprise introduction, technology investment), parks effectively enhance the scale and intensification of agricultural production (reflected in lnagr) within the existing initial structure, significantly driving county economic growth via economies of scale and efficiency gains. In essence, the negative foad effect reflects the constraining static initial industrial structure, while the positive lnagr effect reflects the dynamic breakthrough in optimizing agricultural production methods achieved through policy intervention. This finding underscores that the core value of the park policy lies in overcoming initial structural disadvantages, achieving agricultural productivity gains through factor reorganization and technological progress, thereby laying the groundwork for subsequent structural upgrading.

4.2. Conclusions

This study achieved its three core objectives through empirical analysis. First, it confirmed the policy lag effect: Confirming that the positive effect of NMAIPs on county economies has a 1-year lag, and persists until the fourth year, while exhibiting a trend of marginal diminution, and revising the conventional understanding of “policy immediate effect.” Second, it identified pathways by quantifying the mediating contributions of agricultural scale expansion and production efficiency improvement, revealing the lagging nature of industrial integration and structural upgrading, and providing a basis for targeted policy interventions. Third, it proposed policy recommendations, forming a phased and differentiated optimization path tailored to central counties’ characteristics, which also aligns with the “consolidating the agricultural foundation” objective of the 2024 Central “Thousand Villages Project”.

4.3. Policy Recommendations

Based on the findings and the current status of industrial parks, this study offers key insights for global agricultural park policy design: Prioritizing investments in production-side scale and efficiency represents a successful pathway common to agricultural parks in developing countries; Policy evaluation necessitates a conversion window period (e.g., 2–3 years). Industrial chain integration should be fostered gradually following scale maturity. Based on these insights, the specific recommendations proposed are as follows:

• Optimize the Policy Evaluation Cycle: A 1–2 year “window period” should be established for NMAIP policy evaluation, focusing on assessing economic growth effects in years 2–4 post-establishment (i.e., lag 1–3 years), to avoid prioritizing construction over operation due to short-term performance pressures.
• Prioritize Production-Side Capacity Building: Given that empirical results identify agricultural scale expansion and production efficiency improvement as the core pathways for parks to drive county economic growth with significant short-term effects (1–3 years), while industrial integration’s direct contribution is insignificant, policy resources in the initial stage (1–3 years after approval) should prioritize initiatives that rapidly enhance scale and efficiency. Support for scaled planting (e.g., subsidies for contiguous land transfer), improved seed breeding base construction (subsidy CNY 100–200 per mu), and integrated application of agricultural machinery (increase agricultural machinery purchase subsidy ratio to 30%) should be prioritized, consolidating agricultural scale and efficiency advantages.
• Phase in the Cultivation of the Integration Ecosystem: Once the production foundation is solidified (e.g., year 3 onwards), efforts to cultivate industrial chain extension and integrated formats leveraging scale and efficiency advantages should be gradually intensified. In the initial phase (1–3 years), primary processing facilities at origin (e.g., build cold storage above 500 tons) and e-commerce logistics support (cultivate 2–3 agricultural product e-commerce platforms per county) should be improved. In the medium term (3–5 years), “agriculture + research/education” and “agriculture + cultural tourism” models (e.g., creating 1–2 rural tourism demonstration points) should be explored, gradually extending the industrial chain.
• Innovate Management Models: A “government planning + enterprise operation + farmer shareholding” model, clearly delineating the roles and responsibilities of the government (infrastructure investment), enterprises (market operation), and farmers (production participation and shareholding), should be promoted. Farmers’ share should be substantially increased in agricultural added value through mechanisms like “contract farming” and “equity cooperation.”
• Implement Differentiated Support Strategies: In grain-producing areas (e.g., parts of Henan, Anhui), scaled production facilities (e.g., high-standard farmland construction) should be focused on. In characteristic agricultural areas (e.g., parts of Jiangxi, Hubei), brand cultivation (e.g., subsidies for geographical indication certification) and market expansion (e.g., support for participation in national agricultural exhibitions) should be focused on.
• Strengthen Linkages with “Thousand Villages Project”: NMAIPs should be positioned as the industrial anchor of the “Thousand Villages Project,” leveraging the parks’ scaled bases to drive surrounding villages to develop supporting cultivation (e.g., contract vegetable planting). The parks’ cold chain logistics networks should be utilized to cover village-level storage points, establishing a “core park + thousand village base” production–marketing linkage, and promoting deep integration between rural industrial revitalization and county economic upgrading.

4.4. Development Vision and Expectations

Over the next 5–10 years, the upgrading of National-Level Modern Agricultural Industrial Parks (NMAIPs) will deepen in terms of three dimensions—production efficiency enhancement, industrial chain extension, and regional synergy—amidst the global wave of agricultural modernization. Drawing on international best practices while adapting to China’s county-level realities, a clear evolutionary pathway can be formulated.

In production-side upgrading, the core direction lies in the deep integration of digitalization and scale. Internationally, agricultural parks in the US Midwest have significantly boosted production efficiency by adopting digital technologies like satellite remote sensing and smart irrigation. The EU, through cooperative models, has synergized scale and digitalization by uniting smallholders—exemplified by grape grower alliances in Bordeaux, France, which leverage shared digital platforms to reduce input costs and enhance productivity. Building on these examples, China’s NMAIPs can expand the application scope of IoT and smart agricultural machinery, extending coverage from core zones to surrounding bases. Concurrently, deepened land transfer and enterprise-led initiatives can elevate large-scale operations, reducing unit production costs and optimizing resource allocation to strengthen production-side competitiveness.

For industrial chain extension, the global trend is shifting from “singular production” to “full-chain value addition.” The Netherlands, by integrating deep processing of flowers with global logistics networks, has substantially increased added value and fortified its international market position. Japan’s “One Village, One Product” model expands industrial value through the fusion of agriculture with cultural tourism and e-commerce. Future NMAIPs in China should prioritize enhancing agro-processing capabilities, extending industrial chains by improving cold chain logistics and aligning with upgraded consumer demand. Simultaneously, exploring integrated formats like “Agriculture + Research Education” and “Agriculture + Branded E-commerce” can gradually increase income shares from processing and services, forming a holistic profit model of “production as the foundation, processing for value addition, and services for expansion”.

Regarding regional synergy, building a clustered pattern of “core leadership with peripheral linkage” is key to elevating overall agricultural efficiency. Denmark’s network layout—“National Innovation Center + Regional Demonstration Bases”—enables the efficient diffusion of agricultural technologies and resources, offering valuable insights for China. In central and western regions, NMAIPs can act as hubs, facilitating the flow of resources such as improved seeds, technologies, and cold chain facilities to neighboring counties. This creates a division of labor where core zones focus on deep processing while peripheral areas supply raw materials. Through resource sharing and complementary strengths, policy spillover effects can be amplified, positioning these parks as critical growth poles that integrate central county economies into the global agricultural division of labor.

In conclusion, as a core vehicle for rural revitalization, the value of NMAIPs extends beyond short-term economic stimulation to provide sustainable industrial support for high-quality county economic development through the trajectory of “production foundation–efficiency improvement–integration upgrading.” This process necessitates respecting objective economic laws and persistent effort.