Analysis of the Impact of Digital Technology on the Resilience of Anhui’s Agricultural Industry Chain: Threshold Effect of Relative Grain Yield

Abstract

Improving the resilience of the agricultural industry chain to frequent external shocks and the intertwined agricultural production risks is critical for ensuring national food security and sustainable agricultural development. Taking the 16 prefecture-level cities in Anhui Province—a major grain-producing region in China—as the research sample, two composite proxy indices were constructed in this study; namely, an agricultural industry chain resilience index built upon aggregate official statistics (rather than direct observed shock-response outcomes) and an index capturing broad rural digital development instead of purely agriculture-specific digital adoption. Based on balanced panel data from 2015 to 2024, two-way fixed-effect, mediation-effect, and panel-threshold models were adopted in this study to examine the association, mediating mechanism, regional heterogeneity, and threshold characteristics conditioned on the relative grain yield. The empirical results show that the rural-oriented digital development index is significantly and robustly positively associated with our composite agricultural industry chain resilience proxy. Although agricultural TFP exhibits a statistically detectable partial mediating pattern in this correlational relationship, the economic magnitude of this indirect channel is extremely modest. As such, this finding indicates a potential theoretical transmission pathway rather than providing definitive empirical evidence for a causal mechanism. The promotion effect exhibits clear regional heterogeneity, with the strongest effect in Northern Anhui, followed by Central Anhui, while the coefficient is positive but insignificant in Southern Anhui. Furthermore, a single-threshold effect of the relative grain yield is identified; that is, the positive association between rural-oriented digital development and resilience proxy cannot be statistically detected when the grain yield remains below the threshold. Only after the agricultural grain yield increases can digital-related rural development release its resilience-related empowering dividends. We note that both composite indices are subject to measurement constraints linked to the local agricultural economic scale and broad rural digital conditions; these limitations are further discussed in the paper. This study provides empirical correlational evidence for the implementation of differentiated digital agriculture policies in major grain-producing provinces.

1. Introduction

1.1. Research Background

The agricultural industry chain encompasses production, processing, circulation, consumption, and other interlinked segments with extensive coverage and tightly connected nodes. After external shocks, the adverse impacts propagate rapidly along the whole chain. In recent years, frequent global climate anomalies and volatile foreign trade policies have aggravated the inherent vulnerability of agricultural industry chains. Anhui Province, a pivotal grain-producing province in China, registered a total grain output of 83.86 billion jin in 2025. Its agricultural sector currently suffers from multiple coexisting risks: natural disasters, including droughts, floods, and crop pests, break out frequently, while the prices of agricultural production materials and farm products fluctuate drastically. Against this backdrop, it has become an urgent task to strengthen agricultural industry chains’ capacity to resist external disruptions, recover operational functions post-shock, and realize industrial transformation and upgrading.

In recent years, the national government has successively introduced policies to promote the deep integration of digital technologies with agriculture. The 14th Five-Year Plan for the Digital Economy Development designated smart agriculture as a priority project, and four central ministries and commissions jointly issued the Key Guidelines for Digital Rural Development 2025. Following the national strategic arrangements, Anhui Province promulgated the Action Plan for Digital Rural Development of Anhui Province (2022–2025) and the 14th Five-Year Plan for Agricultural and Rural Modernization of Anhui Province, promoting the full coverage of rural broadband and e-commerce penetration and the deployment of the agricultural Internet of Things. Data show that Anhui Province’s rural broadband penetration rate increased from 24.2% to 38.7%, and rural e-commerce transaction volume grew from CNY 8.7 billion to 150 billion.

In this context, studying the impacts of digital technologies on the resilience of agricultural industry chains in Anhui Province can provide both theoretical support and empirical references for Anhui Province and other major agricultural provinces to advance their digital rural strategies, optimize their agricultural policy frameworks, and enhance the resilience capacity of their agricultural industry chains.

1.2. Purpose of the Research

The main research questions of the paper are as follows:

(1)

How can we scientifically construct comprehensive evaluation indicator systems for agricultural industry chain resilience and the rural-oriented digital development level? What are the direction and magnitude of the statistical association between rural-oriented digital development and agricultural industry chain resilience across prefecture-level cities in Anhui Province?

(2)

Through which potential internal transmission channels may rural-oriented digital development relate to agricultural industry chain resilience? Does agricultural TFP function as a mediating variable within this correlational relationship?

(3)

Are there regional disparities in the statistical linkage between rural-oriented digital development and agricultural industry chain resilience? What differences can be observed across Northern, Central, and Southern Anhui, as well as at different grain output performance levels?

(4)

Does the association between rural-oriented digital development and agricultural industry chain resilience display nonlinear features? Is there a threshold effect conditioned on the grain yield efficiency index? What are the observable critical conditions under which the statistical linkage becomes detectable?

In response to the above research questions, the research objectives of this study were as follows:

(1)

Construct comprehensive indicator systems for agricultural industry chain resilience and rural-oriented digital development. Apply the entropy weight method to compute the two composite proxy indices and conduct empirical analysis to explore the direction and magnitude of their statistical association.

(2)

Explore potential transmission pathways through which rural-oriented digital development correlates with agricultural industry chain resilience. Examine whether agricultural TFP acts as a mediating factor, and elaborate the theoretical conceptual chain of “rural-oriented digital development → agricultural TFP → agricultural industry chain resilience.”

(3)

Investigate regional heterogeneity in the statistical linkage. Compare differences across Northern, Central, and Southern Anhui with varying grain output performance to provide a reference for differentiated policy design.

(4)

Identify potential nonlinear patterns in the above-mentioned association. Adopt the grain yield efficiency index as the threshold variable to test for threshold effects and clarify observable critical conditions for the emergence of statistical linkage.

Accordingly, four research hypotheses are proposed in Section 3.

It should be highlighted that our agricultural industry chain resilience indicator is a composite proxy derived from aggregate socioeconomic statistics rather than directly observed post-shock recovery records, and is thus constrained by prefecture-level data availability; similarly, our digital development index reflects overall rural digital development capacity instead of digital tools purely applied within agricultural production. These measurement features shape the hypothesis interpretations, coefficient explanations, and conclusions throughout this paper.

2. Literature Review

2.1. Connotation and Measurement of the Agricultural Industry Chain Resilience

The concept of “resilience” originated from ecology. Holling (1973) [1] was the first author to define resilience as the ability of an ecosystem to absorb changes and maintain a stable state. Reggiani et al. [2] introduced resilience into economics, framing it as an economic system’s evolutionary capacity to revert to its original equilibrium following disruptive shocks. Within the agricultural domain, the Food and Agriculture Organization (FAO) defines agri-food system resilience as the capacity of food systems to secure long-term access to sufficient, safe, and nutritious food for all populations and sustain the livelihoods of the agricultural practitioners amid diverse disruptions [3,4].

The existing literature has yielded rich discussions regarding the definition and quantification of agricultural industry chain resilience. Most scholars define it as the integrated capacity of agricultural chains to withstand risks, restore operational functions, and pursue transformative development when confronted with internal and external shocks [5,6,7]. Two mainstream measurement approaches prevail in the existing literature. The first adopts a one-dimensional proxy, such as the deviation degree of the total output across farming, forestry, animal husbandry, and fishery sectors, to gauge agricultural resilience [8]. The second relies on comprehensive, multi-dimensional evaluation to construct indicator systems covering risk resistance, adaptive recovery, and structural reconstruction capacities [9,10]. Li J et al. [11] further measured the resilience of China’s agricultural industrial and supply chains along three dimensions: resistance, recoverability, and transformative capacity.

2.2. Digital Technologies and Agricultural Development

Existing research generally holds that digital technologies can enhance agricultural productivity by improving information transparency, reducing transaction costs, and optimizing resource allocation [12,13,14,15]. Specifically, rural e-commerce broadens marketing channels for agricultural commodities [16,17]. Digital inclusive finance could alleviate farmers’ financing constraints [18,19], and agricultural IoT enables precision farming and real-time disaster early warning systems [20,21]. Using panel data of the prefecture-level cities in China, empirical studies have revealed that the application of digital technology significantly enhances the resilience level of the agricultural industry chain [22,23].

Furthermore, the linkage between digital technologies and agricultural performance is not a simple linear relationship. Peng D-Y [24] identifies an inverted U-shaped correlation between digital inclusive finance and agricultural resilience. Using panel data covering 30 provincial-level administrative regions in China, Cheng Y-S & Zhang D-Y [25] confirm that digital technologies foster the sound development of farmers’ cooperatives through indirect channels, including stimulating technological innovation, mitigating capital shortages, and improving risk resistance. Their analytical mechanisms provide a critical reference for the present research.

2.3. Mechanisms Underpinning Resilience Enhancement: Technological Progress and Industrial Upgrading

Regarding the mechanisms for enhancing the resilience of agricultural industry chains, the existing research primarily examines this issue from two perspectives: technological progress and industrial upgrading. The induced technological innovation theory posits that the growth in agricultural productivity hinges on technological advancement [26,27].

The application of digital technologies is accelerating the integration of digital tools—such as data, information, and algorithms—into the agricultural industry chain, thereby enhancing its technological supply capacity, driving agricultural TFP [28,29], and strengthening the risk resilience of the agricultural industry chain [30,31,32]. Zheng J-L [22] confirms that agricultural TFP constitutes a vital transmission channel linking digital technologies to agricultural industry chain resilience.

According to the industrial structure evolution theory, technological innovation serves as the core driving force behind the evolution of the industrial structure from a low to high level [33,34]. The application of digital technologies drives the upgrading of traditional rural industries; promotes the deep integration of agriculture, manufacturing, and services; and facilitates the structural upgrading and diversified development of the agricultural sector [35]. The industrial linkage effect mitigates the negative impact of instability factors on the agricultural industry chain and enhances its resilience [36]. Leveraging a double machine learning framework, Huang X [37] verified three pathways to boost agricultural industry chain resilience: advancing technological innovation, deepening agricultural specialization, and upgrading industrial structures.

2.4. Research Gaps

The conceptual measurement of agricultural industry chain resilience, the application effects of digital technologies in the agricultural sector, and the mechanisms and pathways for enhancing resilience have been systematically explored in the existing literature. However, existing research still presents three major gaps.

First, the empirical testing of the mediating mechanism remains insufficient. Empirical studies that incorporate the pathway “digital technologies → agricultural TFP → resilience of the agricultural industry chain” into a unified analytical framework and conduct rigorous mediation effect tests are still relatively limited.

Second, the research perspective on the threshold effect needs to be expanded. How grain yield influences the effectiveness of digital technology empowerment as a threshold condition has rarely been examined in the existing literature.

Third, the criteria used for regional heterogeneity analysis are relatively broad; there is a lack of systematic comparisons across finer geographical subregions within provinces (such as Northern, Central, and Southern Anhui), which makes it difficult to support differentiated policy design at the provincial level.

Building upon the aforementioned gaps, 16 prefecture-level cities in Anhui Province were employed as a sample to examine the threshold effect and regional heterogeneity of grain yield with the aim of supplementing the existing literature.

2.5. Potential Marginal Contributions

This study first elaborates the theoretical mediating transmission path of “rural-oriented digital development → agricultural TFP → agricultural industry chain resilience.” By incorporating agricultural TFP as a mediating variable into an integrated analytical framework, it explores the internal theoretical logic according to the existing literature.

Second, it identifies a statistically significant single-threshold effect conditioned on the grain yield efficiency index. The analysis estimates the threshold value, and the regime-shift pattern for the statistical association between rural-oriented digital development and the resilience proxy is described.

Third, this study refines the relatively coarse regional classification paradigm used in existing heterogeneity research. Granular empirical comparisons were conducted among Northern, Central, and Southern Anhui, and gradient differences in the statistical linkage are documented. The results offer a sub-provincial empirical reference for designing differentiated agricultural policies in major grain-producing provinces.

3. Theoretical Mechanisms and Research Hypotheses

3.1. Concept Definition and Dimension Decomposition of Agricultural Industry Chain Resilience

The resilience of the agricultural industry chain refers to its comprehensive capability to maintain its fundamental operational functions, rapidly return to normal operations, and achieve structural upgrading when confronted with various internal and external disturbances, such as natural disasters, market fluctuations, technological shocks, and policy changes. The resilience of agricultural industry chains was categorized into three core components in this study: resistance capacity, recovery capacity, and transformation capacity. The resistance capacity refers to the chain’s ability to withstand and protect itself when exposed to shocks. The recovery capacity denotes its ability to self-adjust and rebuild after sustaining damage. The transformation capacity describes its ability to achieve enhanced development through technological innovation and structural adjustment following a shock.

3.2. Direct Influence Mechanism of Digital Technology on Agricultural Industry Chain Resilience

The endogenous growth theory posits that digital technologies can reshape agricultural production organizational models, thereby enhancing industry chain resilience from three perspectives. First, the penetration of digital resources into grassroots levels helps break down information barriers, alleviate supply–demand imbalances, reduce distribution costs, buffer against market and natural shocks, and strengthen overall resilience. Second, online platforms connect various business entities, enabling the rapid reallocation of resources to restore operational continuity when the industry chain experiences fluctuations. Third, the integration of big data and artificial intelligence into agriculture drives the industry chain toward higher value-added segments. Digital technologies can address weaknesses within the industry chain, integrate various business entities, and comprehensively enhance risk resilience. Based on this analysis, Hypothesis H1 is proposed.

H1. 

From a theoretical perspective, rural-oriented digital development is positively associated with the composite proxy indicator of agricultural industry chain resilience.

3.3. Mediating Mechanism of Agricultural TFP

The Induced Technological Transformation Theory posits that technological progress is central to enhancing agricultural productivity and strengthening the resilience of the agricultural industry chain against shocks. Digital technologies serve as a key driver of agricultural technological advancement. They facilitate the integration of data with agricultural production resources and optimize resource allocation and models. Continuous upgrades in agricultural technology can enhance land-use efficiency, mitigate production risks, and ensure greater stability and resilience of the industry chain. Based on this rationale, Hypothesis H2 is proposed.

H2. 

Rural-oriented digital development correlates with agricultural industry chain resilience indirectly through changes in agricultural TFP.

3.4. Theoretical Interpretation of the Regional Heterogeneity

The regional economy and agricultural location theory posits that there exist disparities in resource endowments and industrial foundations across different regions. Northern Anhui, as a traditional major agricultural region, possesses a substantial agricultural base and exhibits a high marginal effect of digitalization, thereby exerting a prominent promoting role. Central Anhui, leveraging its position within the provincial capital economic circle, boasts a well-developed economy and digital infrastructure and demonstrates a strong digital empowerment effect. Southern Anhui, characterized primarily by mountainous and hilly terrain, focuses on ecological agriculture and tourism, features a unique industrial structure, and shows relatively weak penetration effects of digital technologies. The divergent objective conditions among these three regions result in graded differences in the impact of digital technologies on resilience. Accordingly, Hypothesis H3 is proposed.

H3. 

The magnitude of the statistical association between rural-oriented digital development and agricultural industry chain resilience exhibits regional heterogeneity across Anhui Province.

3.5. Threshold Effect Mechanism of Grain Yield

The endogenous growth and technology absorption theory posits that the impact of digital technologies is constrained by the absorptive capacity of the subject and exhibits nonlinear characteristics. In regions with low grain yields, the effect of digital empowerment on resilience remains limited due to the constraints in infrastructure development and farmers’ digital literacy. Once the grain yield surpasses a threshold, digital technologies can fully unleash their potential in scenarios such as precision farming and intelligent decision-making as the scale and standardization of agriculture improve. The influence of digital technologies on industry chain resilience consequently follows a threshold-based transition pattern relative to grain yield levels, leading to the formulation of Hypothesis H4.

H4. 

The statistical linkage between rural-oriented digital development and agricultural industry chain resilience has a threshold effect conditioned on the grain yield efficiency index.

The theoretical framework and research hypotheses of this study are visualized in Figure 1.

Figure 1. Influence mechanism of digital technology on agricultural industry chain resilience. Note: Both key variables are measured via composite proxy indices constrained by prefecture-level data availability; see Section 4.1 for detailed measurement qualifications.

4. Research Design

4.1. Variable Selection

(1)

Explained Variable: Agricultural Industry Chain Resilience

Referring to the studies of Hao A-M & Tan J-Y [6] and Zeng X-W et al. [9], a comprehensive evaluation index system was constructed in this study for agricultural industry chain resilience from three dimensions: resistance, recovery capacity, and transformation capacity (Table 1). The entropy weight method was adopted to calculate the weight of each indicator and further measure the annual agricultural industry chain resilience index of prefecture-level cities in Anhui Province.

Table 1. Indicator system of agricultural industry chain resilience.

Dimension (Target Layer)	Indicator Layer	Calculation Method	Attribute
Resistance Capacity	Agricultural Mechanization Level	Total power of agricultural machinery/Cultivated land area	Positive (+)
	Grain Production Capacity	Total grain output	Positive (+)
	Agricultural Product Processing Level	Operating revenue of agricultural product processing industry	Positive (+)
Recovery Capacity	Large-Scale Agricultural Operation	Number of leading enterprises in agricultural industrialization	Positive (+)
	Agricultural Cooperative Coverage	Number of specialized farmer cooperatives	Positive (+)
	Rural Consumption Expenditure	Total consumption expenditure of rural residents	Positive (+)
Transformation Capacity	Level of Scientific Expenditure	Expenditure on research and experimental development	Positive (+)
	Fiscal Support for Agriculture	Expenditure on agriculture, forestry, and water conservancy/General public budget expenditure	Positive (+)
	Quality of Agricultural Labor Force	Average years of schooling of rural labor force	Positive (+)

Notes: All indicators are processed with the entropy weight method. “+” denotes a positive indicator. This index is an aggregate-statistics-based proxy for agricultural industry chain resilience.

It is important to highlight measurement limitations for this composite resilience index. Conceptually, agricultural industry chain resilience captures shock resistance, post-shock recovery, and transformative capacity. However, this study was constrained by the lack of direct, prefecture-level observational data on shock losses and disaster responses in publicly available statistical yearbooks. The resilience index was therefore constructed using aggregate socioeconomic statistical indicators. Consequently, the composite resilience proxy inevitably correlates with regional agricultural economic scale. Each indicator in Table 1 serves as a reasonable indirect proxy for resilience dimensions: total grain output provides a material buffer against supply-side shocks; revenue from agricultural product processing reflects the processing sector’s capacity to absorb raw-material price volatility; the number of agricultural leading enterprises and farmer cooperatives represents organizational resources for factor reallocation after disturbances; and R&D expenditure, fiscal agricultural support, and rural labor quality capture innovation resources for upgrading the industry chain. Potential confounding links between this resilience proxy and the local agricultural development scale are further addressed via model specification in the subsequent subsection.

(2)

Core Explanatory Variable: Digital Technology Index

The digital technology index reflects the application level of regional digital technology in the agricultural sector. Following the research frameworks of Zheng J-L [22] and Wu Y-T et al. [38], the indicators used in this study were selected from three dimensions; namely, digital application development capacity, digital technology adoption, and digital infrastructure construction level (Table 2).

Table 2. Indicator system of digital technology index.

Primary Indicator	Secondary Indicator	Indicator Definition	Attribute
Digital Application Development Capacity	Rural Postal and Communication Service Level	Average population served per postal outlet (10,000 people)	Negative (−)
	Digital Technology Purchasing Power	Rural residents’ per capita disposable income/Regional gross domestic product (CNY 10,000/CNY 100 million)	Positive (+)
	Digital Service Consumption Level	Per capita transportation and communication consumption expenditure of rural households (CNY)	Positive (+)
	Digital Technology R&D	Number of granted digital invention patents	Positive (+)
	Digital Technology Embedding	Rural e-commerce transaction value	Positive (+)
Digital Technology Application	Internet Penetration Rate	Number of internet broadband access users/Total population × 100%	Positive (+)
	Mobile Phone Penetration Rate	Number of mobile phone users/Total population × 100%	Positive (+)
	Proportion of Internal R&D Expenditure in Regional GDP	Internal expenditure on research and experimental development (R&D)/Regional gross domestic product (GDP) × 100%	Positive (+)
Digital Infrastructure Level	Rural Delivery Routes	Length of rural delivery routes (km)	Positive (+)
	Rural Broadband Penetration Ratio	Number of rural broadband access users/Total internet broadband access users × 100%	Positive (+)
	Rural Energy Density	Rural electricity consumption/Rural population (10,000 people)	Positive (+)

Note: Reverse min–max normalization was applied for negative indicators before entropy weight calculation. This index measures rural-oriented digital development rather than pure agriculture-specific digital adoption.

Several data-driven caveats apply to this digital technology index. Due to data constraints for Anhui’s prefecture-level units, there are no official statistical indicators that purely measure digital technology adoption specifically within agricultural production. Therefore, our composite index characterizes overall rural-oriented digital development capacity, which constitutes a precondition for digital penetration into agricultural industry chains, rather than a pure agriculture-specific digital adoption metric. Although most selected indicators focus on rural contexts, this index still partly reflects general rural economic and infrastructure conditions. The indicator of digital technology purchasing power (ratio of rural residents’ disposable income to regional GDP) was introduced to partially mitigate confounding bias caused by city-level economic scale. The negative indicator “average population served per postal outlet” was treated with reverse min–max normalization before entropy weight calculation.

(3)

Mediating Variable: agricultural TFP

The DEA–Malmquist index method was adopted. The input indicators were crop sown area, agricultural labor input, and fertilizer input, and the output indicator was agricultural added value [39]. The changes in agricultural total factor productivity of each prefecture-level city over the years were calculated.

Agricultural total factor productivity reflects the overall technological progress level of agriculture. Based on the agricultural input–output accounting data, it was calculated without using any of the original underlying indicators in the resilience index or digital technology index, meeting the requirement that the intermediate variables should be independent of the explanatory variables and the dependent variables.

(4)

Threshold Variable: Grain Yield Efficiency Index

The grain yield efficiency index was adopted as an outcome-based proxy for agricultural grain yield, following the empirical framework proposed by Peng D-Y [24]. It should be noted that this index does not directly measure inherent grain production capacity, the magnitude of shock severity, or the degree of risk exposure. Instead, it reflects ex post production performance after regions are exposed to various agricultural production-risk shocks.

Given data limitations for prefecture-level cities in Anhui, consistent city-level panel indicators of direct risk measures (including disaster-caused yield losses, frequency of meteorological hazards, and annual yield volatility) are unavailable over 2015–2024. Regions repeatedly exposed to floods, droughts, pests, and other production risks tend to experience stagnation or decline in grain yield relative to the benchmark year. Therefore, we constructed the grain yield efficiency index, taking 2015 as the base year (base value = 100). A value below 100 indicates reduced yield compared with the baseline, which may indirectly signal adverse production-related pressures; a value above 100 reflects positive relative grain output growth against each city’s own 2015 baseline. Crucially, higher index values represent realized output growth relative to a city’s own historical baseline and should not be interpreted as equivalent to higher intrinsic grain production capacity. This threshold variable helps identify whether the statistical association between rural-oriented digital development and composite resilience proxy shifts based on this ex post grain output outcome.

(5)

Control Variables

To mitigate the estimation bias caused by omitted variables, a set of control variables were selected in accordance with existing studies [22,24]. The selected control variables are specified as follows. Fiscal expenditure scale (lnfis) is measured by the natural logarithm of expenditures on agriculture, forestry, and water conservancy in local general public budget expenditures (ten thousand CNY). Financial development level (FD) is defined as the proportion of the balance of deposits and loans of financial institutions at the year-end in regional gross domestic product (%). Transportation infrastructure (road) is represented by highway density, calculated as the ratio of highway mileage (kilometers) to administrative area (kilometers per hundred square kilometers). Agricultural employment ratio (agemp) is measured by the proportion of employees in the primary industry in the total regional employment (%).

We adopted several empirical strategies to distinguish the association between digital development and resilience proxy from time-invariant cross-city agricultural disparities, given that both composite indices are partially correlated with the general level of local agricultural and rural economic development. First, city-level fixed effects absorbed all time-invariant heterogeneity, including geographical conditions, long-term agricultural resource endowments, and persistent gaps in agricultural economic scale across prefectures. Second, time fixed effects removed province-wide common shocks. Third, we included a set of control variables capturing fiscal support for agriculture, financial development, transportation infrastructure, and agricultural employment share to partial out concurrent agriculture-related conditions. Fourth, heterogeneity analysis and threshold regression were employed to further test whether the digital resilience association varies based on grain production conditions, rather than merely mirroring static cross-city agricultural gaps. Nevertheless, we could not fully eliminate confounding effects driven by time-varying unobserved agricultural development factors, so our evidence should be treated as correlational rather than definitive causal evidence.

4.2. Data Sources and Processing

(1)

Sample Selection

In this study, 16 prefecture-level cities in Anhui Province were taken as the research sample; namely, Hefei, Huaibei, Bozhou, Suzhou, Bengbu, Fuyang, Huainan, Chuzhou, Lu’an, Maanshan, Wuhu, Xuancheng, Tongling, Chizhou, Anqing, and Huangshan. The research period spanned from 2015 to 2024, covering a 10-year interval. Accordingly, a balanced panel dataset consisting of 160 observations (16 cities × 10 years) was constructed for empirical analysis.

(2)

Data Sources

The primary data sources of this study are presented in Table 3. For individual missing values in certain years, the linear interpolation method was employed for data supplementation to ensure the integrity and continuity of the panel data.

Table 3. Data sources of main variables.

Variable Category	Variable Name and Abbreviation	Data Source
Explained Variable	Agricultural Industry Chain Resilience (Resilience)	Calculated based on relevant indicators from Anhui Statistical Yearbook and China Rural Statistical Yearbook
Core Explanatory Variable	Digital Technology Index (Digital)	Constructed using indicators from Anhui Statistical Yearbook, China Science and Technology Statistical Yearbook, and the Peking University Digital Inclusive Finance Index
Control Variables	Fiscal Expenditure Scale (fis)	Expenditure on agriculture, forestry, and water conservancy in Anhui Statistical Yearbook
	Financial Development Level (FD)	Balance of deposits and loans of financial institutions and regional GDP from Anhui Statistical Yearbook
	Transportation Infrastructure (road)	Highway mileage and administrative area from Anhui Statistical Yearbook
	Agricultural Employment Ratio (agemp)	Employees in the primary industry and total employed population from Anhui Statistical Yearbook
Mediating Variable	Prefecture-Level Agricultural Total Factor Productivity (agricultural TFP)	Crop input and output indicators from Anhui Statistical Yearbook
Threshold Variable	Grain Yield Efficiency Index (yield)	Calculated and sorted from grain output data in Anhui Statistical Yearbook

Notes: The citation website of Anhui Statistical Yearbook is: https://tjj.ah.gov.cn/ssah/qwfbjd/tjnj/index.html (accessed on 18 August 2026); The citation website of China Rural Statistical Yearbook is: https://data.cnki.net/yearBook/single?id=N2026010296&pinyinCode=YMCTJ (accessed on 18 August 2026); The citation website of China Science and Technology Statistical Yearbook is: https://data.cnki.net/yearBook/single?id=N2026030259&pinyinCode=YBVCX (accessed on 18 August 2026); The citation website of the Peking University Digital Inclusive Finance Index is: https://idf.pku.edu.cn/zsbz/bjdxszphjrzs/index.htm (accessed on 18 August 2026).

(3)

Descriptive Statistics of Variables

The sample size, mean value, standard deviation, minimum value, and maximum value of all variables involved in this study are presented in Table 4.

Table 4. Descriptive statistics of main variables.

Variable	N	Mean	Std. Dev.	Min	Max
Agricultural Industry Chain Resilience (Resilience)	160	0.448	0.092	0.112	0.649
Digital Technology Index (Digital)	160	0.337	0.074	0.223	0.664
Fiscal Expenditure Scale (lnfis)	160	2.476	0.345	1.526	3.114
Financial Development Level (FD)	160	2.751	0.520	1.764	4.389
Transportation Infrastructure (road)	160	157.402	38.300	54.643	216.519
Agricultural Employment Ratio (agemp)	160	9.916	4.501	2.541	22.587
Agricultural TFP (ATFP)	160	0.745	0.280	0.049	1.067
Grain Yield Efficiency Index (yield)	160	98.712	8.360	82.014	118.312

As shown in Table 4, each variable exhibits considerable variation during the sample period, which provides a reliable data foundation for the subsequent econometric analysis. The mean value of agricultural industry chain resilience is 0.448, with a standard deviation of 0.092, a minimum value of 0.112, and a maximum value of 0.649, indicating obvious disparities in agricultural industry chain resilience across prefecture-level cities in Anhui Province. The digital technology index has a mean value of 0.337, a minimum value of 0.223, and a maximum value of 0.664, reflecting an uneven application level of digital technology in the agricultural sector of Anhui Province and laying a solid foundation for exploring its heterogeneous effects.

Table 5 presents the sub-regional statistical results of the digital technology index and agricultural industry chain resilience index. The results reveal that both the level of digital technology application and agricultural industry chain resilience follow a distinct gradient distribution pattern, ranked as Southern Anhui > Central Anhui > Northern Anhui. There exists a significant positive correlation between the mean values of the two indicators.

Table 5. Regional descriptive statistics of the digital technology index and industry chain resilience index.

Region	Mean of Digital Technology Index	Mean of Industry Chain Resilience
Southern Anhui	0.490	0.612
Central Anhui	0.412	0.579
Northern Anhui	0.341	0.435

Figure 2 presents the kernel density distribution of the digital technology index in Anhui Province. The digital technology application index across the province generally follows a unimodal distribution, with the peak value mainly concentrated in the range of 30 to 50. This finding suggests that the values of the digital technology index were relatively concentrated from 2015 to 2024, with no obvious polarization during the sample period. The overall distribution curve shifts rightward over time, and the central position of the distribution continues to move upward with the advancement of the study years. This trend further verifies the steady improvement of digital technology application levels across Anhui Province and reflects the sound and synchronized development of digital agriculture in the whole region. For the sake of the aesthetic presentation of the graph, the numerical values of the digital technology index in Figure 2 are equivalent to the numerical values of the digital technology index in Table 4 multiplied by 100.

Figure 2. Kernel density estimation of the digital technology index in Anhui Province.

4.3. Model Specification

(1)

Benchmark Regression Model

Based on the foregoing theoretical analysis, a benchmark regression model was constructed to empirically verify the research hypotheses proposed in this study. The model is specified as follows:

Resilience_it = α₀ + α₁Digital_it + α₂C_it + μ_i + v_t + ε_it

(1)

In Equation (1), Resilience_it denotes the level of agricultural industry chain resilience; Digital_it represents the digital technology level; C_it refers to the set of control variables affecting agricultural industry chain resilience; μ_i indicates the city fixed effect; v_t denotes the time fixed effect; ε_it is the random error term; α₀ is the constant term; and α₁ represents the influence coefficient of digital technology on agricultural industry chain resilience.

(2)

Mediating Effect Model

A mediating effect model was constructed, following the research paradigm of Jiang T [40], to further explore the internal mechanism through which digital technology affects agricultural industry chain resilience, as presented below:

lnTech_agri_it = β₀ + β₁Digital_it + β₂C_it + μ_i + v_t + ε_it

(2)

Resilience_it = δ₀ + δ₁Digital_it + δ₂C_it + δ₃lnTech_agri_it + μ_i + v_t + ε_it

(3)

In Equation (2) and Equation (3), lnTech_agri_it denotes the mediating variable of agricultural TFP, β₀ is the constant term, and β₁ is the estimated coefficient of digital technology on the mediating variable. The definitions of other variables are consistent with those in Equation (1).

(3)

Threshold Effect Model With Agricultural Grain Yield

A panel threshold regression model was adopted to verify the nonlinear threshold effect of agricultural grain yield on the relationship between digital technology and agricultural industry chain resilience. The single-threshold model was constructed as follows:

Resilience_it = β₀ + β₁Digital_itI(yield_it ≤ γ) + β₂Digital_itI(yield_it > γ) + δC_it + μ_i + γ_t + ε_it

(4)

In Equation (4), yield_it is the threshold variable measured with the grain yield efficiency index, which is used to characterize agricultural grain yield. γ represents the threshold value, and I is the indicator function. Specifically, when yield_it is less than or equal to the threshold γ, the influence coefficient of digital technology on agricultural industry chain resilience is β₁; when yield_it exceeds the threshold γ, the corresponding influence coefficient is β₂. For all two-way fixed-effect panel regressions in this paper, standard errors are clustered at the prefecture–city level to account for serial correlation within each city over years.

This panel threshold model follows Hansen’s panel threshold framework. We implemented bootstrap resampling with 500 replications to test for the existence of threshold effects. Standard errors are consistently clustered at the prefecture–city level for all regression specifications in this paper, including baseline two-way fixed-effect regressions, mediating-effect models, regional subgroup heterogeneity regressions, and panel-threshold regressions, to account for serial correlation of disturbances within each city across years. Note that testing whether there exists a threshold is distinct from a post-estimation Wald test evaluating whether regression coefficients differ across the threshold-separated regimes. It should also be noted that our sample only contains 16 independent prefecture-level cities. The number of city-level clusters becomes even smaller for regional subgroup analyses, which may reduce the reliability of asymptotic cluster-robust inference. Therefore, subgroup regression results should be interpreted with appropriate caution.

5. Empirical Results and Analysis

5.1. Correlation Analysis and Multicollinearity Test

Table 6 presents the results of the correlation analysis among variables. The correlation coefficient between digital technology and agricultural resilience is 0.721*, indicating a statistically significant positive correlation at the 10% level.

Table 6. Correlation analysis results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
(1) Resilience	1.000
(2) Digital	0.721 *	1.000
	(0.001)
(3) FD	0.379 *	0.425 *	1.000
	(0.000)	(0.000)
(4) road	0.521 *	0.279 *	0.147	1.000
	(0.000)	(0.036)	(0.075)
(5) lnfis	0.091	0.219 *	0.172 *	0.097	1.000
	(0.265)	(0.000)	(0.045)	(0.278)
(6) agemp	0.321 *	0.517 *	0.145	0.096	0.678 *	1.000
	(0.000)	(0.000)	(0.112)	(0.448)	(0.000)
(7) agricultural TFP	0.322 *	0.041	0.176	0.486 *	0.379 *	0.589 *	1.000
	(0.028)	(0.574)	(0.144)	(0.000)	(0.000)	(0.000)
(8) yield	0.327	−0.265 *	−0.086	0.389 *	0.220	0.273 *	0.574 *	1.000
	(0.079)	(0.037)	(0.173)	(0.475)	(0.158)	(0.008)	(0.000)

Notes: *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at prefecture–city level) are reported in parentheses.

Table 7 presents the results of the multicollinearity test; the average VIF is 2.819, and the VIF values for all variables are <10, indicating no severe multicollinearity exists. Thus, we could proceed with the regression analysis.

Table 7. Multicollinearity test results.

	VIF	1/VIF
agemp	5.668	0.176
agricultural TFP	4.331	0.231
Digital	3.268	0.306
lnfis	1.769	0.565
yield	1.687	0.593
road	1.581	0.633
FD	1.432	0.698
Mean VIF	2.819	.

5.2. Panel Model Selection Test

Before conducting a benchmark regression, it is necessary to determine the appropriate model using the Hausman test.

Table 8 presents the panel model selection test results: the Hausman test yields χ²(5) = 30.125, p = 0.0023 < 0.01. The null hypothesis for the random-effects model is rejected. Therefore, we adopted the two-way fixed-effects model controlling for both city fixed effects and year fixed effects for all subsequent regressions.

Table 8. Panel model selection test results.

	Value
Chi-square statistic (χ2)	30.125
Degrees of freedom (df)	5
p-value	0.0023

5.3. Baseline Regression Results

Table 9 illustrates the detailed results of the baseline regression. Column (1) presents the estimation outcomes without any control variables. The coefficient of digital technology is 0.141 and significantly positive at the 5% level, indicating that a one-unit increase in the digital technology index contributes to an average increase of 0.141 units in the agricultural industry chain resilience index. Column (2) incorporates all control variables into the regression model. The coefficient of digital technology rises to 0.341 and remains statistically positive at the 5% significance level, which lends support to the robustness of our core statistical finding.

Table 9. Baseline regression results.

	(1) Resilience No Controls	(2) Resilience With Controls
Digital	0.141 **	0.341 **
	(0.0427)	(0.136)
FD		0.0147
		(0.0189)
road		0.0000312
		(0.00021)
agemp		0.00419
		(0.00393)
lnfis		0.0412
		(0.0291)
Constant	0.311 ***	0.219 **
	(0.0142)	(0.0901)
Individual fixed effects	Yes	Yes
Time fixed effects	Yes	Yes
Observations	160	160
R2	0.812	0.809

Notes: *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at prefecture–city level) are reported in parentheses.

In terms of the control variables, the regression coefficients of financial development level (FD), transportation infrastructure (road), fiscal expenditure scale (lnfis), and agricultural employment ratio (agemp) fail to pass the significance test. The financial development, fiscal expenditure, and agricultural employment ratio show positive coefficients, suggesting that financial support, fiscal subsidies for agriculture, and the scale of agricultural workers can marginally promote agricultural industry chain resilience, although such driving effects have not yet reached statistical significance in the current stage. The coefficient of transportation infrastructure is close to zero and insignificant. This finding can be explained by the mature highway network construction across prefecture-level cities in Anhui Province, leading to a limited marginal improvement space of transportation facilities in enhancing agricultural resilience.

The baseline result suggests a significant positive association between our rural-oriented composite digital index and the composite proxy for agricultural industry chain resilience after absorbing time-invariant city characteristics and aggregate year-specific shocks. We emphasize again that the resilience index is built on aggregate statistics and partly reflects agricultural economic scale; the digital index captures broad rural digital development instead of purely agriculture-specific digital adoption. Therefore, Hypothesis H1 is confirmed.

5.4. Mediating Effect Regression

Table 10 reports the empirical results of the mediating effect test. Column (1) presents the regression results of digital technology on agricultural TFP. The coefficient of digital technology is 0.8307 and is statistically significant at the 5% level, indicating that digital technology can substantially promote the advancement of agricultural technology. Digital technology effectively improves agricultural mechanization and actual production efficiency, which is achieved by supporting agricultural research and development and facilitating the dissemination and popularization of professional agricultural knowledge. Column (2) incorporates the mediating variable, agricultural TFP, into the benchmark model. The regression coefficient of agricultural TFP is 0.0081 (p < 0.1), suggesting that agricultural TFP shows a statistical correlation with agricultural industry chain resilience.

Table 10. Mediation effect regression results.

	(1) Agricultural TFP	(2) Resilience
agricultural TFP	—	0.0081 ***
		(0.00217)
Digital	0.8307 ***	0.315 ***
	(0.231)	(0.0279)
lnfis	−0.212	0.0372
	(0.153)	(0.0092)
FD	−0.00983	0.0131
	(0.129)	(0.0172)
road	0.00837	0.000293
	(0.00215)	(0.00019)
agemp	0.142 ***	0.00387
	(0.0120)	(0.00271)
_cons	2.210 ***	0.137 **
	(0.218)	(0.0713)
N	160	160

Notes: *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at the prefecture–city level) are reported in parentheses.

The stepwise regression results in Table 10 provide preliminary evidence for a mediating pathway, yet sequential significance tests cannot formally validate the indirect effect. We therefore conducted a bootstrap-based formal mediation test with 500 resampling replications, and the decomposition results are presented in Table 11.

Table 11. Mediation effect decomposition.

Effect	Coefficient	Bias-Corrected 95% CI
Path a (Digital → agricultural TFP)	0.8307	[0.412, 1.241]
Path b (agricultural TFP → Resilience)	0.0081	[0.0034, 0.0129]
Indirect effect (a × b)	0.0067	[0.0022, 0.0118]
Direct effect (Digital → Resilience)	0.315	[0.2610, 0.3690]
Total effect	0.3217	[0.2681, 0.3753]
Mediating proportion (Indirect/Total)	2.08%	—

Notes: Bootstrap resampling times =500. CI refers to the bias-corrected bootstrap confidence interval. CI excluding zero supports the existence of a statistically significant indirect mediating effect.

As shown in Table 11, path coefficient a (Digital → agricultural TFP) equals 0.8307, and path coefficient b (agricultural TFP → Resilience) equals 0.0081. The estimated indirect effect a × b is 0.0067, and its bias-corrected 95% bootstrap confidence interval is [0.0022, 0.0118]. Since the confidence interval does not contain zero, the partial mediating effect of agricultural TFP is statistically confirmed. The direct effect of digital technology on resilience is 0.3150, and the total effect reaches 0.3217. The share of the mediating channel accounts for only 2.08% of the total effect.

It should be noted that although the indirect effect is statistically significant, its economic magnitude is very small. This indicates that digital technology improves agricultural industry chain resilience predominantly through its direct channels such as information transparency, resource reallocation, and platform construction. Agricultural TFP functions only as a supplementary partial transmission mechanism. The limited magnitude of the mediating effect may be attributable to time lags in agricultural technology diffusion, farmers’ uneven digital literacy across prefecture-level cities, and the fact that many local digital agricultural projects were still in early construction phases during 2015–2024 and have not yielded large-scale productivity gains. Therefore, it is assumed that H2 is verified.

It should be stressed that, despite the statistically significant indirect effect from the bootstrap test, the indirect channel accounts for merely 2.08% of the total association magnitude, which means its economic contribution is very limited. This statistical partial mediation pattern is suggestive of a potential theoretical transmission pathway; however, given the remaining endogeneity concerns, we cannot treat it as a fully demonstrated causal transmission mechanism. Our observational panel data cannot rule out alternative confounding processes that may produce this same statistical pattern. Therefore, this mediating result should be interpreted with caution and viewed as correlational evidence rather than proof of causal mediation.

5.5. Robustness Test

Considering the potential reverse causality between digital technology and agricultural resilience, several robustness tests were conducted to verify the stability of the baseline estimates. We note that strict causal identification is constrained by sample availability, which is further discussed in Section 6.4.

(1)

Shortened Time Interval

To verify the stability and reliability of the benchmark regression conclusions, a robustness test was conducted by shortening and subdividing the sample time interval. The full sample was divided into two subsample periods—namely, 2015–2019 and 2020–2024—for grouped regression analysis. The corresponding test results are presented in Table 12.

Table 12. Results of subsample regressions by time period.

	2015–2019 Resilience	2020–2024 Resilience
Digital	0.543 *	0.612 **
	(0.221)	(0.231)
lnfis	0.0279	0.0796
	(0.0380)	(0.0674)
FD	0.0251 **	0.0445 **
	(0.00779)	(0.0145)
road	0.0000479	0.00170
	(0.000112)	(0.00179)
agemp	0.00578 **	0.0263 *
	(0.00246)	(0.0131)
_cons	0.211 ***	0.151
	(0.0387)	(0.365)
N	80	80

Notes: Standard errors are clustered at the prefecture–city level. *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at the prefecture–city level) are reported in parentheses.

In Table 12, the coefficients of digital technology are significantly positive in both subsamples, which is consistent with the direction of the benchmark regression results. Specifically, the coefficient is 0.543 and significant at the 1% level during 2015–2019, while the coefficient reaches 0.612 and is significant at the 5% level during 2020–2024. The signs and significance levels of the core variables do not undergo fundamental changes, which strongly confirms the robustness of the benchmark regression findings.

(2)

Exclusion of the Provincial Capital City Sample

Considering that the provincial capital city exhibits systematic differences from other prefecture-level cities in terms of economic development level and digital infrastructure construction, its sample inclusion may interfere with the overall regression results. Therefore, a further robustness test was conducted by excluding the provincial capital sample, and the corresponding estimation results are reported in Table 13.

Table 13. Results of robustness test excluding Hefei.

	(1) Full-Sample Resilience	(2) Resilience Excluding Hefei
Digital	0.341 **	0.420 **
	(0.136)	(0.168)
lnfis	0.0513 *	0.0411
	(0.0214)	(0.025)
FD	0.0610 ***	0.0396
	(0.0157)	(0.073)
road	0.000210	0.000223
	(0.000156)	(0.00015)
agemp	0.00879 **	0.00681
	(0.00361)	(0.0012)
_cons	0.00863	0.0310
	(0.0025)	(0.0115)
N	160	150

Notes: Standard errors are clustered at the prefecture–city level. *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at the prefecture–city level) are reported in parentheses.

The coefficient of the core explanatory variable Digital remains statistically significant at the 5% level as a positive value (0.420), and its sign is entirely consistent with the results of the baseline regression. This indicates that the positive promoting effect of digital technology on the resilience of the agricultural industry chain has not been altered by the sample adjustment, demonstrating strong robustness of the core conclusion.

Both of the aforementioned robustness tests support the baseline conclusion, demonstrating that digital technology’s positive impact on the resilience of the agricultural industry chain is robust and reliable.

5.6. Heterogeneity Analysis

(1)

Heterogeneity Analysis Based on Grain Production Levels

The agricultural production foundation may influence the empowerment effect of digital technologies. The study sample was divided into high-grain-producing regions (yield> median) and low-grain-producing regions (yield_it ≤ median) based on the median value of the grain yield efficiency index (yield), and separate regressions were conducted for each group; the results are presented in Table 14. In high-grain-producing regions, the Digital coefficient is 1.461 (p <0.01); in low-grain-producing regions, the coefficient is 0.479 (p <0.01). The difference between the coefficients across the two groups is statistically significant (as confirmed using the SUEST test for the pseudo-uncorrelated model, p = 0.038), indicating that a stronger agricultural production foundation leads to a more pronounced enhancing effect of digital technologies on resilience. A possible reason for this phenomenon is that, in high-grain-producing regions, agriculture is larger-scale and better organized, with a wider range of application scenarios for digital technologies—such as leading advancements in precision fertilization, drone-based plant protection, and smart irrigation—resulting in higher marginal returns.

Table 14. Heterogeneity analysis by grain output level.

	(1) High Grain Output Resilience	(2) Low Grain Output Resilience
Digital	1.461 ***	0.479 **
	(0.169)	(0.135)
lnfis	0.00532	0.0798 *
	(0.0031)	(0.058)
FD	0.0201 *	0.0576 *
	(0.012)	(0.073)
road	0.000047	0.00039
	(0.000089)	(0.00014)
agemp	0.000382	0.0196 **
	(0.00074)	(0.0093)
_cons	0.173 *	0.0198
	(0.112)	(0.012)
N	80	80

Notes: Standard errors are clustered at the prefecture–city level. *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at the prefecture–city level) are reported in parentheses.

(2)

Regional Heterogeneity

According to the geographical division of Anhui Province, the 16 prefecture-level cities in the sample were classified into 3 sub-regions: Northern Anhui (six cities), Central Anhui (four cities), and Southern Anhui (six cities). Specifically, Northern Anhui includes Huaibei, Bozhou, Suzhou, Bengbu, Fuyang, and Huainan; Central Anhui covers Hefei, Chuzhou, Lu’an, and Anqing; and Southern Anhui comprises Wuhu, Maanshan, Tongling, Xuancheng, Chizhou, and Huangshan. Grouped regression analyses were conducted separately for the three sub-regions, and the empirical results are reported in Table 15.

Table 15. Regional heterogeneity analysis results.

Variable	Northern Anhui Resilience	Central Anhui Resilience	Southern Anhui Resilience
Digital	1.712 ***	0.518 ***	0.716
	(0.161)	(0.0301)	(0.605)
lnfis	0.00691	0.0455	0.0602
	(0.00855)	(0.0206)	(0.0296)
FD	0.0112	0.0513 **	0.0617
	(0.00731)	(0.00899)	(0.0799)
road	0.0000196	0.000211	0.000603
	(0.000312)	(0.000531)	(0.000402)
agemp	0.00136	0.0163 ***	0.00412
	(0.00203)	(0.00236)	(0.00763)
_cons	−0.125 *	0.0631	−0.0801
	(0.0501)	(0.0701)	(0.0656)
Number of cities	6	4	6
Observations (city–year)	60	40	60

Notes: Standard errors are clustered at the prefecture–city level. The regional subgroups feature a small number of independent city-level clusters (4 cities for Central Anhui); hence, subgroup estimates should be interpreted cautiously. *** p < 0.01, ** p < 0.05, * p < 0.10. Cluster-robust standard errors (at the prefecture–city level) are reported in parentheses.

North Anhui Region: The regression coefficient for digital technologies is 1.712 (p < 0.01), the highest among the three groups. North Anhui is a major traditional agricultural region with a long agricultural industry chain, yet its digital infrastructure remains relatively weak; consequently, its starting point for digital transformation is low, but its potential is substantial, making the marginal contribution from digital technologies in this region the highest among all regions.

Central Anhui Region: The regression coefficient for digital technology is 0.518 (p < 0.01), placing it in the middle range. Central Anhui includes Hefei, the provincial capital, so this region is economically developed with well-established digital infrastructure. However, agriculture accounts for a relatively low proportion of its industrial structure, leaving limited room for enhancing resilience.

Southern Anhui Region: The regression coefficient for digital technology is 0.716, but the standard error is relatively large (0.605), with a p-value > 0.1, indicating no statistical significance. Southern Anhui is predominantly mountainous and hilly, and its agriculture is characterized by ecological agriculture, the tea industry, and the integration of cultural tourism. Consequently, the resilience of its grain supply chain is constrained by factors such as topography, industrial models, and the level of digital penetration, and the enabling effect of digital technology has not yet been fully realized.

It should be emphasized that these subgroup regression results are subject to important statistical caveats. The full sample only includes 16 independent prefecture-level cities; after splitting into the three geographical subgroups, the number of city-level clusters dropped to 6 (Northern Anhui), 4 (Central Anhui), and 6 (Southern Anhui). A small number of clusters may impair the performance of cluster-robust standard-error inference. Consequently, these regional heterogeneous patterns constitute empirical evidence rather than strong, definitive statistical conclusions. The empirical pattern is consistent with Hypothesis H3.

5.7. Threshold Effect of Agricultural Grain Yield

The grain yield efficiency index (yield) was adopted as the proxy indicator of agricultural grain yield and the threshold variable for empirical estimation to verify Hypothesis H4 that the statistical linkage between rural-oriented digital development and agricultural industry chain resilience displays a threshold effect conditioned on this index [24]. Based on the panel threshold model proposed by Hansen [41], a threshold existence test was first conducted. The bootstrap method with 500 repeated sampling times was applied to test the significance of the threshold effect, and the corresponding empirical results are shown in Table 16.

Table 16. Threshold effect existence test results.

Number of Thresholds	F-Statistic	p-Value	10% Critical Value	5% Critical Value	1% Critical Value
Single threshold	29.13	0.026	23.66	27.31	34.26
Double threshold	13.55	0.315	17.95	22.36	30.99
Triple threshold	6.77	0.617	14.23	17.96	27.81

Table 16 shows that the single threshold effect is statistically significant at the 5% level (F = 29.13, p = 0.026), while the double and triple threshold effects are insignificant. Accordingly, the empirical model possesses a significant single threshold of grain yield efficiency. Following the principle of minimizing the residual sum of squares, the estimated threshold value is τ = 103.21, with a 95% confidence interval of [103.06, 104.31].

The sample was divided into two intervals based on the threshold value (yield ≤ 103.21 and yield > 103.21), with the impact coefficients of digital technology on resilience estimated separately. The results are presented in Table 17.

Table 17. Panel threshold regression results.

Variable	Coefficient	Robust Std. Err.	t-Value	p-Value
Digital I(yield ≤ 103.21)	0.398	0.312	1.28	0.176
Digital I(yield > 103.21)	0.865 ***	0.179	4.83	0.001
Control variables	Yes
Observations	160

Notes: *** p < 0.01, ** p < 0.05, * p < 0.10.

As shown in Table 17, the estimated single-threshold value τ is 103.21. With this index, crossing this threshold means a city achieves 3.21% relative growth in grain output compared to its own 2015 baseline level. When the grain yield efficiency index is lower than or equal to the threshold of 103.21 (i.e., grain output grows by no more than 3.21% relative to the city-specific 2015 baseline), the regression coefficient of the rural-oriented digital development proxy is 0.398 and statistically insignificant (p = 0.176). This statistical result shows that there is no statistically significant detectable association between rural-oriented digital development and our composite resilience proxy in city–year observations where grain output has not yet realized this degree of relative growth against its own 2015 baseline.

In contrast, when the Grain Yield Efficiency Index exceeds the threshold value (yield > 103.21), the coefficient rises sharply and becomes statistically significant at 0.865 (p < 0.01); that is, among observations where cities have achieved more than 3.21% grain output growth relative to their individual 2015 baseline, rural-oriented digital development shows a strong positive statistical linkage with the composite proxy of agricultural industry chain resilience.

We further conducted a post-estimation Wald test for the null hypothesis H₀: β₁ = β2 (equal coefficients across the two regimes). The Wald test yields χ² = 4.72, p = 0.030, rejecting the null hypothesis. This provides direct statistical evidence that the association coefficients for the two regimes are significantly different from each other. It should be emphasized that observing significance in one regime and insignificance in the other alone cannot justify the conclusion of coefficient difference; therefore, this direct coefficient difference test is required.

It is important to emphasize the interpretive boundaries of this threshold finding. The grain yield efficiency index is an ex post outcome indicator measuring realized grain output change relative to each city’s own historical baseline. It does not directly measure intrinsic agricultural production capacity, pre-existing infrastructure conditions, farmers’ digital literacy, or the severity of production risk shocks. We observe a statistical regime shift based on this realized output outcome, but we cannot infer from this threshold result alone that weak infrastructure, low digital literacy, or high-risk exposure are the causal drivers behind the insignificant coefficient below 103.21.

The observed pattern demonstrates that the strong statistical association between rural-oriented digital development and the resilience proxy only emerges for city–years where the city has attained substantial relative grain output growth compared with its 2015 level. In summary, the threshold model estimation results are consistent with Hypothesis H4. Accordingly, the promotion of digital agriculture should avoid uniform policy implementation.

Note that this threshold finding is based on our composite proxy indices; it reflects a regime shift in statistical association, rather than a strict causal threshold for agriculture-specific digital investment.

6. Conclusions and Policy Implications

6.1. Research Conclusions

First, our rural-oriented composite digital development index shows a significant and robust positive association with the composite proxy index of agricultural industry chain resilience across Anhui prefecture-level cities. It should be noted that the resilience proxy was constructed from aggregate official statistics (and correlates with local agricultural scale), while our digital index reflects broad rural digital development capacity instead of digital adoption solely within agricultural production. After controlling for city and time fixed effects, the regression coefficient of the digital index was 0.341 and was statistically significant at the 5% level. This positive association holds across robustness tests.

Second, a statistically detectable partial mediation pattern is observed for agricultural TFP within the correlational linkage between our rural-oriented digital development proxy and the composite resilience proxy. Crucially, the indirect pathway via agricultural TFP accounts for merely 2.08% of the total association magnitude, so its substantive economic contribution is extremely limited. Our statistical outputs show that the rural-oriented digital development proxy has a coefficient of 0.8307 in relation to agricultural TFP. After introducing agricultural TFP into the regression model, both the direct statistical association of the digital development proxy and the statistical coefficient of agricultural TFP remain statistically distinguishable from zero. This observed statistical pattern is consistent with the theoretical conceptual chain of “digital-related development investment → agricultural TFP improvement → higher composite proxy for agricultural industry chain resilience.” Nevertheless, given the unresolved reverse causality risks and measurement constraints of prefecture-level composite proxy indices, this pattern only offers theoretical clues and cannot be regarded as a demonstrated causal transmission mechanism. The bulk of the observed positive linkage operates through other direct statistical association channels.

Third, the statistical association between the rural-oriented digital development proxy and the composite agricultural industry chain resilience proxy presents significant regional heterogeneity across Anhui Province. Grouped regional regressions demonstrate that the magnitude of this positive association is the most prominent in Northern Anhui, followed by Central Anhui, while the coefficient remains positive but statistically insignificant in Southern Anhui. These gradient differences reflect regional disparities in agricultural resource endowments, digital transformation foundations, and industrial structural characteristics. Again, these empirical patterns are derived from our aggregate composite proxies: Northern Anhui shows large marginal statistical correlation relying on its scale advantage as a traditional major agricultural region; the association magnitude in Central Anhui is constrained by the relatively low proportion of agriculture in the overall industrial structure; and Southern Anhui fails to reveal a statistically meaningful correlation under the current proxy measurement framework due to its eco-agriculture and agritourism-oriented industrial system.

Fourth, the statistical association between our rural-oriented digital development proxy and the composite agricultural industry chain resilience proxy is characterized by a significant single-threshold effect conditioned on the grain yield efficiency index. The estimated threshold value equals 103.21, which represents 3.21% growth in grain output relative to each city’s own 2015 baseline. When the grain yield efficiency index is below or equal to this threshold value, the coefficient for the digital development proxy equals 0.398 and is statistically insignificant. When the index crosses this critical threshold, the coefficient rises sharply and becomes statistically significant at 0.865 (p < 0.01). We emphasize that the grain yield efficiency index is an ex post outcome-based indicator of realized grain output change relative to each city’s historical baseline; it is not a direct measure of inherent grain production capacity or shock risk exposure. The result shows that the strong positive statistical linkage between rural-oriented digital development and the resilience proxy only emerges for city–year observations that have achieved this degree of relative grain output growth. We note that this threshold regime-switching pattern is observed based on aggregate proxy indices, and should be interpreted as a nonlinear statistical association rather than a strict causal threshold for agriculture-specific digital investments.

6.2. Policy Implications

First, it is essential to continue deepening the integration and application of digital technologies across the entire agricultural industry chain and to establish a synergistic development mechanism centered on “Digital + Resilience.” Given the significant positive impact of digital technologies on the resilience of the agricultural industry chain, it is recommended that Anhui Province further increase investment in digital infrastructure within the agricultural sector, with a focus on advancing rural broadband network upgrades, expanding the coverage of the Agricultural Internet of Things (IoT), and building a rural e-commerce ecosystem. Local authorities should also be encouraged to establish digital platforms for the agricultural industry chain to integrate data resources from all stages—including production, processing, distribution, and consumption—thereby comprehensively enhancing the industry chain’s ability to withstand external shocks and its efficiency in recovery.

Second, it is important to strengthen the intermediary driving role of agricultural TFP and streamline the transmission chain from “digital empowerment → technological progress → enhanced resilience.” The government should establish a special fund for the research and promotion of agricultural digital technologies, supporting the R&D and demonstration applications of technologies such as intelligent agricultural machinery, precision fertilization and irrigation, and drone-based plant protection. Simultaneously, new types of agricultural entities—such as agricultural cooperatives and family farms—should be leveraged to enable the provision of digital agriculture training, thus enhancing farmers’ capacity to adopt and apply these technologies. By establishing a technology extension network composed of “research institutions + enterprises + farmers,” we can accelerate the transformation of digital technologies into agricultural productivity, thereby indirectly enhancing the resilience of the agricultural industry chain.

Third, differentiated regional digital agriculture policies should be implemented to enhance the resilience levels of various regions according to local conditions. In the northern Anhui region, where the marginal effect of digitalization is high but its foundational infrastructure remains weak, priority should be given to addressing gaps in digital infrastructure, focusing on the promotion of applicable technologies such as intelligent agricultural machinery and precision agriculture and establishing digital agriculture pilot counties to develop replicable experience models. The radiating and driving role of the provincial capital economic circle in the central Anhui region should be leveraged to promote the deep integration of digital technologies with high-value-added sectors—including agricultural product processing, cold-chain logistics, and brand marketing—thereby creating a resilience-led zone. In the southern Anhui region, efforts should be made to explore an integrated development path of “digital + ecology + cultural tourism,” integrating the characteristics of ecological agriculture and cultural tourism industries and, thereby, increasing the penetration rate of digital technologies within specialty agriculture and gradually unleashing their enabling potential.

Fourth, support for improvements in realized grain output performance should be prioritized in regions that have not yet attained relative grain output growth compared with their own historical baseline to create conditions under which rural digital-related development correlates with higher resilience proxy values. The threshold effect results document that the strong statistical association between rural-oriented digital development and our composite resilience proxy emerges only after cities have achieved sufficient relative grain output growth against their 2015 baseline. In jurisdictions where grain output has not yet realized this relative growth milestone, our empirical pattern suggests that policymakers should focus first on interventions aiming to improve grain output performance, such as farmland water conservancy construction, improved seed promotion, and integrated pest management. After local grain output achieves this relative growth benchmark, digital agriculture investments may be pursued. This approach avoids a one-size-fits-all digital agriculture rollout. However, we stress that this policy suggestion is derived from observed statistical associations, and it cannot be established at this time that improving grain output causes digital development to become more effective.

6.3. Further Discussion

This study empirically investigates how rural-oriented digital development correlates with agricultural industry chain resilience based on prefecture-level panel data for Anhui Province. Our finding of a positive statistical association aligns with conclusions reported by Zheng J-L [22]. Consistent with the prior literature, rural-oriented digital development may reduce information frictions, optimize resource allocation, and foster novel agricultural business formats, which in theory could strengthen system-level resilience to external disturbances.

Nevertheless, most existing relevant empirical evidence is derived from national or provincial-level samples. This study documents notable intra-provincial gradient heterogeneity; in particular, the magnitude of the positive statistical linkage is strongest in Northern Anhui and moderate in Central Anhui, while the coefficient remains positive but statistically insignificant for Southern Anhui. This intra-provincial heterogeneous pattern has rarely been reported in previous studies, which implies that uniform digital agriculture policies may not be equally suitable for all sub-regions within one province.

Moreover, a partial mediating statistical pattern related to agricultural TFP was detected in this study, echoing the theoretical mechanism proposed by Zheng J-L [22]. Complementing the nonlinear insights offered by Peng D-Y [24], we further identify a single-threshold effect conditioned on the grain yield efficiency index. Rural-oriented digital development does not exhibit a detectable resilience-relevant statistical association when grain yield efficiency remains at or below the threshold value; a strong positive statistical linkage emerges only after grain output achieves sufficient relative growth above this threshold.

It is worth emphasizing a general interpretive caveat for our mediating-effect results. The partial mediation pattern we detected represents a statistical association derived from observational panel data; as such, while it aligns with ex ante theoretical expectations, it does not constitute a demonstrated real-world transmission mechanism. Endogeneity and measurement limitations prevent us from firmly establishing causal ordering among rural-oriented digital development, agricultural TFP, and agricultural industry chain resilience. Readers should distinguish between the documented statistical patterns in our observational data and causal mechanism conclusions that would require stronger identification strategies.

6.4. Limitations and Future Research

The impact of digital technology on agricultural industry chain resilience was empirically investigated in this study using a two-way fixed-effect model based on panel data of 16 prefecture-level cities in Anhui Province from 2015 to 2024. Although the model controls for observable confounding variables, including fiscal expenditure, financial development, and transportation infrastructure, and absorbs time-invariant city-specific attributes and common temporal shocks via city- and year-fixed effects, it cannot fully address endogeneity originating from reverse causality. Cities with higher agricultural industry chain resilience generally possess stronger fiscal capacity and more complete infrastructure, which enables greater investment in digital construction and yields higher digital technology indices. In other words, superior chain resilience may reciprocally boost digital technology adoption and induce estimation bias. Valid exogenous instrumental variables or credible quasi-experimental shocks are difficult to obtain due to the constraints of our sample, which only covers 16 prefecture-level cities within a single province. Our empirical findings should accordingly be interpreted as correlational evidence rather than definitive causal effects. Even though multiple robustness checks, including subsample regressions and excluding Hefei City, support our main conclusions, caution is required for causal inference.

Second, measurement limitations originate from our two core composite indices, as highlighted in Section 4.1. Our agricultural industry chain resilience index relies on aggregate socioeconomic indicators rather than direct observational records of shock exposure and post-shock recovery behavior at the city level. Consequently, this proxy inevitably co-varies with the overall scale of the local agricultural economy. Our digital technology index similarly measures comprehensive rural digital development capacity, rather than digital technology usage exclusively in agricultural production links. Although we applied two-way fixed effects and observable control variables to partial out static cross-city agricultural disparities, we could not perfectly disentangle the digital resilience association from time-varying unobserved agricultural development gaps. Future research could employ micro-level farm survey data, disaster loss statistics, and agriculture-specific digital adoption indicators to construct more precise measurements for both variables.

Third, regarding endogeneity concerns, our study has measurement limitations constrained by prefecture-level aggregate statistics. Our digital technology index reflects overall rural digital development capacity instead of purely agriculture-specific digital adoption. In addition, the agricultural industry chain resilience index is built upon aggregate statistical indicators. Although we supplement theoretical justification for each proxy variable, agricultural economic scale cannot be fully separated from genuine shock-coping capacity. Future research may use farm-level micro-survey data or disaster-induced yield loss indicators to construct more targeted measurements for agricultural digital adoption and industry chain resilience.

Fourth, our full-sample dataset only covers 16 independent prefecture-level cities. Even though all regression models adopt city-level clustered standard errors to handle within-city serial correlation, cluster-robust inference may exhibit size distortion when the number of clusters is small. The number of independent city-level clusters in the regional subgroup heterogeneity analysis was reduced to as low as four (Central Anhui subgroup). This imposed limits on the statistical power and reliability of subgroup-specific coefficient estimates. Therefore, our subgroup-based heterogeneous findings should be regarded as tentative evidence. Future research could expand the sample to county-level panel data with substantially more cross-sectional units to further verify these heterogeneous effects.

Several avenues are available for future research based on the above-mentioned limitations. First, the dataset can be expanded to a nationwide county-level panel. More rigorous causal identification strategies, such as difference-in-differences (DID) built upon digital rural pilot-policy shocks and instrumental variable approaches, can be employed to mitigate reverse causality bias and omitted-variable bias. Second, future work can further distinguish heterogeneous influences exerted by different categories of digital tools on agricultural industry chain resilience.