Digital infrastructure and local food supply diversity: evidence from 257 Chinese prefecture-level cities

Abstract

Introduction

As food system governance increasingly emphasizes supply diversity, nutrition sensitivity, and sustainability, understanding the role of digital infrastructure in diversifying local food supplies has become important. This study examines whether digital infrastructure is associated with greater diversity across nutritionally relevant food categories in China.

Methods

We construct a Nutritional Food Supply Diversity (NFSD) index and analyze panel data from 257 Chinese prefecture-level cities over 2011 -2023. The analysis examines the association between digital infrastructure and NFSD, potential mechanisms, and heterogeneity across regional development levels, agricultural resource endowments, and urbanization levels. We also address potential endogeneity and conduct robustness checks.

Results

Digital infrastructure development is significantly and positively associated with NFSD. A one-standard-deviation increase in the digital infrastructure index is associated with an average increase of approximately 0.0664 index points in NFSD. This association remains robust after addressing potential endogeneity and conducting robustness checks. Mechanism analyses suggest three potential pathways: encouraging the entry of relevant businesses and service providers, thereby expanding producers’ channels of market participation; improving the matching of digital agricultural services; and strengthening feedback on food demand. The association varies with regional development, agricultural resource endowments, and urbanization.

Discussion

The findings suggest that assessments of the food system effects of digital infrastructure should consider both food quantities and the diversity of local supplies across nutritionally relevant categories. These effects appear to depend on the physical agricultural base, market conditions, and stages of urban development. This study provides an empirical framework for coordinating digital infrastructure development, adjustments to agricultural production structures, and policies promoting the diversity of the local food-supply base.

Introduction

In recent years, the constraints facing global food systems have evolved from insufficient quantities of staple foods into a combination of challenges involving quantity security, nutritional quality, supply chain stability, and sustainability. Climate change, extreme weather, international trade tensions, and public health shocks continue to expose vulnerabilities in traditional food systems arising from concentrated sources, limited distribution reach, and asymmetric supply and demand information. Beyond asking “Is there enough food?”, a growing body of research examines “Which foods can be supplied reliably?” and “Can the food supply structure support diverse and nutritious diets?” Using food inflow data for U.S. metropolitan areas, Gomez et al. (2021) find that diversity in supply chain sources significantly enhances cities' ability to withstand moderate food shocks. The review by Krupitzer and Stein (2024) further shows that digitalization is moving agricultural production, processing, logistics, and retail from separate activities toward system integration. Incorporating diversity in supply structures into food system research therefore concerns both food system resilience and the potential nutritional relevance of a diverse food-supply base.

Changes in China's food consumption structure imply that food system research should consider not only total grain output or self-sufficiency rates, but also the mix of different food categories in local production. This study constructs a local food supply diversity indicator using output shares of grains, vegetables, fruits, meat, and aquatic products to characterize the diversity of cities' local production structures. As a supply-side outcome, this indicator cannot be directly equated with residents' dietary intake, dietary diversity, or health outcomes. Fixed networks, mobile communication networks, and the data connectivity they support are changing how agricultural businesses obtain information about prices, weather, technology, logistics, and demand. Digital infrastructure primarily improves information transmission, service matching, and market coordination; it is not directly equivalent to agricultural machinery, cold chains, or other productive fixed assets. Accordingly, this study no longer interprets digital infrastructure as directly reducing agricultural fixed costs, but examines how it affects local food supply structures through digital market participation, digital agricultural services, and online food accessibility.

Existing studies have demonstrated the important effects of digital technologies on agriculture and food systems from several perspectives. One stream examines the effects of digital technologies, internet use, or digital agricultural services on productivity, technology adoption, and farmers' market participation. Another has begun to discuss the relationships of the digital economy and artificial intelligence with food security, food supply capacity, and agricultural value chain resilience. Recent studies of Chinese cities further situate digital technologies within prefecture-level food system frameworks, finding significant links between digital technologies and food security, and between artificial intelligence applications and urban food supply capacity. These studies provide an important foundation for understanding the food system effects of digitalization. However, their outcome variables remain concentrated on grain output, composite food security indices, supply capacity, or value chain resilience, with comparatively little attention to whether digital infrastructure changes the supply mix across food categories. Meanwhile, agricultural nutrition research has developed a line of inquiry linking production diversity, dietary diversity, and nutritional improvement, although its findings are highly context dependent. The meta-analysis by Sibhatu and Qaim (2018) shows that production diversity is positively associated with dietary diversity on average, but the effects are generally small and vary with market conditions, commercialization, and household characteristics. Using Chinese household data, Yan et al. (2024) further find that crop, livestock and aquaculture, and mixed production diversity help improve household dietary diversity, with subsistence provision and income from agricultural commercialization serving as important channels. This suggests that diversity at the production end has potential nutritional significance. Yet most existing research remains at the farm or household level, with limited examination of how city-level public digital infrastructure, as an external environmental condition, affects food supply structures at a broader spatial scale.

The central question of this study concerns food-supply composition rather than supply scale. “Quantity effects” refer to changes in total food output or aggregate supply capacity, whereas “composition effects” concern changes in the relative shares of different food categories. A proportional increase in output across all categories expands total supply without changing its composition; conversely, changes in category shares can alter supply diversity even when total output remains unchanged. Building on research into the effects of digitalization on productivity, output, and food security, this study asks whether better digital connectivity is associated with a more balanced local production structure across grains, vegetables, fruits, meat, and aquatic products. NFSD is constructed from the relative output shares of these nutritionally relevant categories, making food-supply composition a distinct object of analysis alongside supply scale. This supply-side indicator describes local production composition and does not directly measure residents’ dietary intake or nutritional health outcomes.

Using panel data for 2011–2023 from 257 Chinese prefecture-level cities, this study constructs a local food supply diversity index and applies a model with city and year fixed effects to examine its association with digital infrastructure and the underlying channels. The mechanism analysis focuses on three pathways: producers' participation in digital markets, digital agricultural service accessibility, and online food accessibility. The first concerns whether smallholders, cooperatives, and small and micro businesses can expand market participation through e-commerce platforms, online orders, digital payments, and logistics networks. The second emphasizes whether digital infrastructure promotes the provision and matching of agricultural technical advice, market information, and related digital services. The third examines how access to food information and the convenience of online purchasing and delivery feed consumer demand back into production and distribution.

This study makes three main contributions. First, it moves beyond the conventional approach of assessing food system development through total food output or supply scale and instead examines the association between digital infrastructure and local food supply diversity through diversity in local food production structures. This perspective helps reveal whether digital development shifts local agricultural production and food supply from relatively uniform quantity expansion toward a more diverse allocation of food categories, providing new empirical evidence on the role of digital technologies in food system transformation. Second, the study constructs a composite digital infrastructure index covering fixed-network access, mobile communication connectivity, and telecommunications service provision, measured respectively by household broadband penetration, mobile telephone penetration, and price-deflated telecommunications revenue per capita, and combines these indicators through standardization and entropy weighting. Compared with the common practice of directly dividing fixed broadband subscriptions recorded by household or access account by the total population, this design can partially reduce measurement bias caused by mismatched statistical definitions and more comprehensively reflect cities' digital connectivity and underlying service provision. Third, in measuring mechanisms, the study avoids directly treating general market accessibility, agricultural fixed costs, or total retail sales of consumer goods as transmission channels of digital infrastructure. Instead, it constructs measures using firm entry data: the entry density of agricultural e-commerce firms, digital agricultural service providers, and online food businesses and service providers, representing producers' digital market participation, digital agricultural service accessibility, and online food accessibility, respectively.

Literature review

Digital infrastructure, digital agriculture, and agricultural production transformation

Within a food-system perspective, digital-agriculture research primarily addresses the production side, where digital infrastructure supports technological applications, agricultural services, and production decisions. Early digital agriculture research focused on how precision agriculture, sensors, satellite remote sensing, the agricultural Internet of Things, and decision support systems improve agricultural productivity and resource-use efficiency through more precise information acquisition, optimized input allocation, and closer monitoring of production processes. More recently, the scope has expanded beyond individual production activities to broader agrifood system domains, including digital platforms, mobile internet, online agricultural advisory services, and digital supply chains. In their systematic review of social science research on digital agriculture, smart farming, and Agriculture 4.0, Klerkx et al. (2019) argue that digital agriculture involves more than the application of new production technologies: its development also reshapes agricultural knowledge and innovation systems, agricultural advisory service models, value chain relationships, and interactions among agricultural actors. From a sustainable agricultural systems perspective, Basso and Antle (2020) further argue that digital and geospatial technologies strengthen the monitoring, assessment, and management of soil, climate, and genetic resources. More precise measurement, forecasting, and resource allocation can better coordinate economic, environmental, and social objectives, creating new technological pathways for sustainable agricultural transformation. Examining precision agriculture and the relationship between agricultural production and the environment, Finger et al. (2019) show that sensing, sampling, big data, and precision management enable agricultural decisions that are more spatially and temporally targeted, reducing wasted inputs and environmental externalities. However, adoption and performance also depend on farm size, technological infrastructure, institutional conditions, and actors' capacity to adopt technology. Whether digital technologies translate into productivity gains and food system transformation therefore depends not only on the technologies themselves, but also substantially on connectivity, data access, and related infrastructure. Digital infrastructure should consequently be understood as a fundamental production condition supporting the entry of digital technologies into agricultural production, market distribution, and food supply chains and their effects on resource allocation, rather than merely as an external technological backdrop for digital agriculture.

Porciello et al. (2022), in their review of digital agricultural services in low- and middle-income countries, find that applications concentrate on digital advice, extension services, weather and price information, and market linkages, with increased knowledge, behavioral change, and improved production decisions being the most common immediate outcomes. Klerkx and Rose's (2020) research on Agriculture 4.0 suggests that platform development, networking, and real-time data exchange change how farmers connect with input suppliers, cooperatives, financial institutions, wholesalers, and consumers. Taking an agrifood chain perspective, Krupitzer and Stein (2024) argue that the central value of digitalization lies in integrating previously fragmented production, processing, logistics, retail, and consumption activities. This provides a direct theoretical basis for examining digital infrastructure through a food system perspective rather than through a single agricultural production activity. McCampbell (2022) discusses the conditions for adopting agricultural digitalization and automation in low- and middle-income countries through ten case studies, providing additional evidence for understanding contextual differences in digital agriculture applications.

In China, the expansion of broadband and mobile communication networks has provided an important foundation for digitalizing agriculture and agricultural product distribution. Numerous studies have examined the effects of internet use, rural e-commerce, digital inclusive finance, and the digital economy on agricultural total factor productivity, green agricultural development, technology adoption, and farmers' incomes. More directly relevant to this study, recent prefecture-level research has begun to examine the effects of digital technologies on food security and food supply capacity. Wei (2026) uses Chinese prefecture-level data to discuss the link between digital technologies and food security, while Bai et al. (2026) study the effects of artificial intelligence applications on urban food supply capacity. These studies show that urban digital conditions can spill over into agriculture and food systems. Nevertheless, their outcomes primarily reflect the “scale” or “security” of supply, without fully addressing whether digital infrastructure promotes a more balanced supply structure across grains, fruits and vegetables, and animal-source foods.

Digital infrastructure, transaction costs, and market integration of small and micro producers

This strand connects production with distribution by examining how market conditions shape producers' participation in trading networks and their ability to sell different food products. Transaction costs and market integration are related but distinct concepts. The former concerns the costs of searching for prices, matching with buyers, negotiating contracts, and executing transactions. The latter emphasizes whether smallholders, cooperatives, and small and micro agricultural businesses can enter digital trading networks. Timely access to order, payment, and logistics information is particularly important for perishable foods. Digital infrastructure can reduce information and coordination frictions, but whether these improvements translate into market performance depends on logistics, the degree of organization, and physical market conditions.

The systematic review by Marion et al. (2024) of interventions to improve agricultural output market access in low- and middle-income countries shows that market information provided through the internet or mobile phones and investments in ICT infrastructure can improve outcomes such as the prices received by farmers, adoption of improved inputs, and agricultural income, although effect sizes and evidence quality differ across studies. Digital connectivity therefore does not automatically improve market performance; its effects are constrained by physical transport, organizational capacity, and local market institutions. This conclusion is particularly relevant to urban food supply. Producers are more likely to expand production of perishable, nutrient-dense foods in response to urban demand when digital infrastructure and transport and logistics networks jointly reduce distribution costs. Conversely, where logistics bottlenecks are severe, better online information may not fully translate into changes in supply structures.

Research on food supply chain diversity also provides relevant evidence. Gomez et al. (2021) use Shannon diversity to characterize the sources of food supplied to U.S. cities and find that more diverse supply chain sources improve cities' ability to withstand food supply shocks. Although their focus is diversity of supply sources across regions, whereas this study examines diversity among major categories of local food production, both emphasize the independent value of “structural diversity” relative to “sufficient quantities.” For Chinese prefecture-level cities, digital infrastructure may change the relative market opportunities for different agricultural products by expanding market reach, reducing transaction frictions, and facilitating information exchange across regions. Local agriculture may consequently shift from concentrating on a few dominant categories toward a more diverse structure aligned with demand across multiple market segments.

Agricultural production diversity, food supply diversity, and nutrition

Research on agricultural production diversity and nutrition links the production side to nutritional considerations, explaining why food-category composition matters beyond aggregate output. The underlying logic is that producing a wider range of agricultural products at the farm or regional level may improve the food mix through two pathways. The first is subsistence provision, which directly expands the range of foods available to households or local markets. The second is commercialization and income, whereby sales of multiple agricultural products increase income and enable purchases of more diverse foods. However, production diversity does not necessarily improve diets in every context. In a meta-analysis of studies from several developing countries, Sibhatu and Qaim (2018) find a positive but limited average effect and show that market access, farm commercialization, and regional conditions significantly affect the relationship between production diversity and dietary outcomes. This finding highlights market conditions as a key moderating link between production diversity and nutritional outcomes.

Research by Chinese scholars provides a more detailed understanding of this relationship. Using Chinese household data, Yan et al. (2024) distinguish crop production diversity, livestock and aquaculture production diversity, and mixed production diversity. They find that crop production diversity improves overall dietary diversity and that different forms of production diversity particularly benefit dietary diversity in animal-source foods, through subsistence provision and income from agricultural commercialization. From a nutritional perspective, “what is grown, what is raised, and which categories are supplied” therefore matters in its own right, beyond total agricultural output value or grain output. This study extends that micro-level logic to urban food systems and constructs NFSD using the relative shares of grains, vegetables, fruits, meat, and aquatic products to assess whether urban supply has a more diverse foundation of food categories.

Food supply diversity and residents' dietary diversity must nevertheless be distinguished. Urban residents can consume substantial quantities of food produced elsewhere through interregional trade, wholesale markets, and e-commerce. Local output structures thus reflect the underlying supply base rather than final intake patterns. Accordingly, this study positions NFSD as a “nutrition-related food system supply outcome,” rather than a direct indicator of household dietary intake, nutritional status, or health. This distinction retains the production diversity literature's concern with nutrition-sensitive agriculture while avoiding a mechanical equivalence between local production and residents' actual consumption. It also leaves room for future research to connect urban production, distribution, and consumption structures more closely.

Online food accessibility and demand feedback

From the demand side of the food system, research on online food accessibility shows how consumption opportunities and demand signals may feed back into production and distribution. Internet penetration and e-commerce development reduce the time and spatial costs of searching for, comparing, and purchasing different foods while increasing access to nutritional and product information. In areas underserved by traditional physical food markets, online channels can expand the range of available foods. For urban residents, fresh-food e-commerce, instant retail, and platform delivery further improve the convenience of purchasing perishable foods. Digital infrastructure therefore changes not only producers' market entry but also generates new demand signals through consumer choices, which feed back into production and distribution through orders, prices, and platform data.

Chinese micro-level studies provide relatively direct support. Using rural household surveys from four Chinese provinces and an endogenous switching model, Ma et al. (2022) find that online food purchasing significantly increases household dietary diversity by an estimated 7.34%. Using the 2021 Chinese rural food consumption survey and an instrumental-variable approach, Shen et al. (2023) further find that e-commerce reduces consumption of staples such as grains and tubers while increasing consumption of legumes, nuts, milk, and dairy products, significantly improving dietary diversity and overall dietary quality. These findings suggest that digital consumption channels can shift food demand from a concentration on staples toward a more diverse mix. Although these studies examine residents' consumption rather than urban production and supply, demand-side changes can be transmitted upstream through market prices, online orders, and retail procurement, providing a basis for considering demand feedback as a potential pathway linking digital connectivity with local supply composition.

The nutritional effects of e-commerce may nevertheless be complex. Online platforms expand food choices and purchasing convenience, but digital food environments may also increase exposure to processed foods high in sugar, salt, and fat and encourage impulse purchases. Digitalization does not inherently imply healthier diets; its nutritional consequences depend on product composition, platform recommendations, delivery conditions, consumer income, and nutritional awareness. This study does not directly assess the net effect of digital infrastructure on healthy diets. Instead, it considers online food accessibility as a potential channel transmitting demand signals to local production and supply. The firm-entry proxy measures the entry of relevant businesses and service providers; it does not directly measure household purchases, dietary intake, or nutritional outcomes.

Digital infrastructure and agricultural system transformation also involve the efficiency of agricultural resource allocation and improvements in management decisions, beyond information transmission and market connectivity. Qin and Sha (2026); analyze how improved anthropogenic management practices enhance vegetation carbon sink capacity in China, showing that optimizing management processes can improve the operation of agricultural and ecological systems. Yang et al. (2026) further show that optimization models and dynamic regulation mechanisms can improve the efficiency of irrigation water allocation under uncertainty, illustrating the importance of information support and decision optimization for agricultural adaptability. Agricultural production is subject to multiple uncertainties associated with climate, water resources, and market fluctuations. Using Monte Carlo methods to examine uncertainty in agricultural water management, Yang et al. (2022) show that decision support based on probabilistic simulation and optimization can improve agricultural resource management. These studies provide additional support for analyzing how digital infrastructure affects local food supply structures by improving the information environment and reducing decision uncertainty.

The development of digital agriculture depends not only on information connectivity but also on technology adoption costs, institutional conditions, and organizational capacity. Lowenberg-DeBoer (2022) identifies the economic benefits of digital automation technologies and barriers to their adoption as important determinants of agricultural digital transformation. Early ICT research also shows that mobile communication and information services reduce the costs of accessing market information and improve agricultural transaction efficiency (Jensen, 2007; Aker, 2010; Muto and Yamano, 2009). In addition, digital agricultural services and platform technologies affect production decisions through information dissemination, technology extension, and market connections (Nakasone et al., 2014; Fabregas et al., 2019). From a global food governance perspective, FAO (2022) further emphasizes the importance of digitalization, innovation, and sustainable agricultural management for food system resilience. Recent research has expanded the links between digital technologies and food system transformation. Zhang and Jiang (2026) and Chen et al. (2026) provide new empirical evidence from urban food systems, digital technology applications, and agricultural resource allocation, suggesting that digital development is becoming a key determinant of food supply capacity and agricultural system adaptability.

Assessment of the literature

Viewed together, these strands describe connected parts of a food system: digital agriculture concerns production capabilities and decisions; market-access research connects producers with distribution networks; production-diversity research explains the nutritional relevance of food-category composition; and e-commerce research examines consumption opportunities and demand feedback. Their connections suggest that changes in production capabilities may influence supply composition, while distribution conditions determine which products reach markets and consumer demand may guide producers' choices. However, these links are often examined separately, leaving limited evidence on how city-level digital connectivity is associated with local food-supply composition across nutritionally relevant categories. This study integrates these strands within a production–distribution–consumption framework. It considers digital agricultural service accessibility on the production side, producers' digital market participation at the production–distribution interface, and online food accessibility on the demand side. In the empirical analysis, these are proxied by the entry density of digital agricultural service providers, agricultural e-commerce firms, and online food businesses and service providers, respectively. The framework links the literature to the theoretical mechanisms and empirical proxies while retaining a clear measurement boundary: NFSD describes local production composition, not residents' actual food purchases, dietary intake, or nutritional health.

Theoretical analysis and research hypotheses

Digital infrastructure and local food supply diversity

The proposed conceptual sequence is: digital infrastructure → more diversified local production and supply → broader structural availability of food categories → potential nutritional relevance. Here, structural availability refers to the diversity and balance of the local production base across the included food categories, as represented by their relative output shares. This is a conceptual pathway, not an empirically established chain of nutritional improvements. The analysis examines the association between digital infrastructure and supply composition; it does not directly measure household access to food or consumption. Whether this supply base translates into improved diets, nutritional status, or health depends on distribution, affordability, food choices, and intake, and lies outside the direct scope of this study.

Agricultural production and food distribution commonly face information asymmetries, inadequate supply–demand matching, and demand uncertainty. Food categories differ markedly in production cycles, price fluctuations, storage and transport conditions, and preservation requirements. When adjusting their mix of crops, livestock, or business activities, producers need timely access to multidimensional information on weather changes, market prices, pest and disease warnings, technical guidance, and order demand. Broadband and mobile communication development can reduce information search, communication and coordination, and market comparison costs, improving producers' ability to identify opportunities across food categories, compare operating options, and adjust production. Digital infrastructure does not directly replace agricultural machinery, cold-chain equipment, or storage facilities; its primary role is to improve the external information environment on which production and distribution decisions depend.

Digital infrastructure can also improve information matching among producers, buyers, logistics providers, and consumers. Applications such as online orders, digital payments, location services, and logistics tracking help reduce informational uncertainty and performance risks during transactions, making it easier for producers to meet market demand across regions, product categories, and transaction scales. When digital connectivity complements roads, cold chains, and storage facilities, interregional trade and stable sales of perishable foods such as fruits, vegetables, meat, and aquatic products become more feasible. Digital infrastructure may therefore be positively associated with diversity in local food supply structures by reducing transaction frictions, improving supply–demand matching, and expanding market opportunities.

Viewed through a technology–organization–environment lens, these channels depend on technological connectivity, the organizational capacity of producers, service providers and market intermediaries, and the surrounding production and market environment. Zhou (2026), using grounded theory and fsQCA, identifies alternative combinations of technological, institutional, cognitive and social-structural conditions associated with AI adoption in environmental governance. Although the outcome differs from food supply, this configurational logic provides a general rationale for treating digital connectivity as an enabling condition whose implications depend on complementary capacities and contexts. In the present study, digital market participation, agricultural service accessibility and online demand feedback require production resources, organizational coordination and functioning distribution networks to support supply adjustment. This is a conceptual interpretation; the empirical models do not directly test a complete TOE configuration.

Data signals generated by consumers' online searches, purchases, reviews, and orders can also feed back into production and distribution through retail procurement, platform orders, and price changes. Digital infrastructure improves supply-side access to information and service matching while increasing the visibility and transmission speed of food demand signals and the efficiency of market responses. This creates conditions for producers to adjust their product mix to consumer preferences and expand diverse food supplies. Based on this analysis, the following hypothesis is proposed:

H1: Digital infrastructure development is significantly and positively associated with local food supply diversity.

Digital market integration of small and micro producers

Smallholders, family farms, cooperatives, and small and micro agricultural businesses often face limited sales channels, unstable orders, insufficient price information, and weak logistics organization. Unlike general market accessibility, producers' digital market participation concerns whether these actors can use digital connectivity to enter previously inaccessible trading networks. This involves establishing stable connections with a broader range of buyers, retailers, and consumers through online product displays, order matching, mobile payments, customer communication, and logistics services. The mechanism thus focuses on whether digital technologies expand the opportunities and scope for small and micro producers to participate in market transactions, rather than on traditional market distance or road accessibility itself.

Digital infrastructure provides the underlying connectivity that allows small and micro producers to access e-commerce platforms, online wholesale markets, digital payment networks, and logistics information systems. Improvements in network coverage and access quality help reduce informational separation between producers and markets, lower barriers to order communication, transaction matching, and delivery coordination, and improve access to information about customer demand, product prices, and sales opportunities. Compared with measuring market conditions solely through road density or conventional freight volumes, this mechanism more specifically captures the role of digital connectivity in expanding market participation among small and micro businesses.

Access to more customer, order, and price information and relatively stable online trading relationships can increase the expected returns and commercial viability of producing and selling specialized and multiple food categories. Wider market reach helps alleviate the constraints imposed by limited local demand on sales of specialty and perishable foods. More comprehensive order and price information also reduces uncertainty when producers adjust their product mix. Digital infrastructure may therefore expand the effective market boundaries of small and micro producers through greater digital market participation, promoting a more diverse local food supply structure. Accordingly, the following hypothesis is proposed:

H2: Digital infrastructure promotes local food supply diversity by increasing producers' participation in digital markets.

Digital agricultural service accessibility

Producing multiple food categories generally requires a broader range of technical knowledge, risk warnings, and management support. Food categories differ in cultivation and husbandry techniques, pest and disease control, production cycles, quality standards, and marketing practices. Businesses seeking to adjust or expand their product categories therefore need corresponding professional information and services. Digital infrastructure does not automatically translate into agricultural machinery, cold-chain facilities, or other productive fixed assets, so this study does not interpret it as directly reducing agricultural fixed costs. A more plausible transmission mechanism is that network connectivity improves access to online agricultural technical advice, pest and disease warnings, price information, demand information, and outsourced agricultural services, thereby enhancing digital agricultural service accessibility.

Greater digital agricultural service accessibility makes it easier for businesses to obtain technical guidance, risk information, and service resources suited to different food categories, reducing information search, service matching, and operational coordination costs. For example, producers can use digital channels to obtain timely information about new varieties, technologies, pest and disease warnings, and changing market demand, and then adjust cultivation, husbandry, and product combinations. Digital infrastructure does not necessarily cause changes in production structures: its actual effects continue to depend on local agricultural resource endowments, service provision capacity, physical infrastructure, and market demand.

At the city level, supply diversity does not require every farmer or business to increase all food categories simultaneously. It may instead involve a more appropriate division of production across actors according to their comparative advantages. Digital agricultural services primarily enhance businesses' ability to identify comparative advantages, access professional services, and allocate production resources, creating conditions for more diverse local food supply structures. Accordingly, the following hypothesis is proposed:

H3: Digital infrastructure promotes local food supply diversity by improving digital agricultural service accessibility.

Online food accessibility and demand feedback

Food supply structures are shaped not only by production conditions and market organization but also by continuing signals of consumer demand. Digital infrastructure can reduce the time and spatial costs of searching for, comparing, and purchasing different foods, improving access to online food retail, delivery services, and product information and making it easier for consumers to encounter and choose different food categories. This study no longer equates an increase in total retail sales of consumer goods directly with upgraded food consumption, because growth in general consumption does not necessarily increase food expenditure or directly represent improvements in food consumption diversity or nutritional quality.

As online food channels expand the set of foods that consumers can search for, compare, and purchase, their searches, browsing, purchases, reviews, and orders generate more timely demand signals. These signals can feed back into distribution and production through retail procurement, platform orders, price changes, and supply chain coordination, helping businesses identify demand for different food categories and adjust procurement, sales, and production accordingly. Digital infrastructure may therefore improve the transmission of demand signals to the supply side by enhancing online food accessibility.

Online food accessibility is only one condition enabling demand feedback. Its actual effects also depend on delivery systems, food product composition, residents' incomes, nutritional awareness, and offline market conditions. This study therefore treats online food accessibility as a potential channel transmitting demand to adjustments in local food supply structures, rather than directly equating it with improved dietary diversity among residents. Accordingly, the following hypothesis is proposed:

H4: Digital infrastructure promotes local food supply diversity by improving online food accessibility and demand feedback.

Research design

Sample and data

The sample comprises panel data for 257 Chinese prefecture-level cities from 2011 to 2023. During data preparation, cities with major administrative boundary changes or prolonged missing data for core variables that prevented comparability across years were excluded, yielding a balanced panel of 257 prefecture-level cities and 3,341 city–year observations. The data primarily come from successive editions of the China City Statistical Yearbook, statistical yearbooks of provinces, autonomous regions, and municipalities, prefecture-level statistical yearbooks, and city statistical bulletins on national economic and social development. Digital infrastructure and urban socioeconomic characteristics are mainly drawn from the China City Statistical Yearbook. Food production data for grains, vegetables, fruits, meat, and aquatic products are primarily compiled from provincial and prefecture-level statistical yearbooks, with city statistical bulletins used to supplement selected missing observations and cross-check the data. Firm registration records required for the mechanism variables are obtained from Qichacha, Tianyancha, and the CSMAR firm registration database. Establishment dates, registered addresses, and business scopes are extracted and matched to the city–year panel. Data from different sources are matched according to consistent statistical definitions and temporal continuity, with harmonized units and calculation methods.

Variable definitions

Dependent variable

Nutritional Food Supply Diversity (NFSD). This study defines NFSD as the diversity and balance of urban food production and supply structures across major food categories. Here, “nutrition-sensitive supply” denotes a production base containing nutritionally relevant food categories. NFSD measures diversity in that supply base and does not measure dietary intake, nutritional status, or health. The baseline indicator includes five categories: grains, vegetables, fruits, meat, and aquatic products. Where complete data are available, eggs and dairy are added to construct a seven-category indicator.

First, the share p jit of the output Q jit of food category j in city i in year t in the total output of the included food categories is calculated: p jit = Q jit ∑ j = 1 J Q jit (1)

Using these food supply shares, the Shannon diversity index is employed to construct the nutritional food supply diversity indicator: NFSD it = − ∑ j = 1 J p jit ln ( p jit ) (2)

Here, J denotes the number of food categories included. A higher NFSD indicates more balanced supply shares across categories and a more diverse urban food production and supply structure. Balance in physical output shares does not establish nutritional adequacy, household food access, or improved dietary intake. To examine whether the findings depend on the specific form of the Shannon index, the robustness analysis constructs an alternative dependent variable, Food Diversity Index (FDI), using the reverse Herfindahl index: FDI it = 1 − ∑ j = 1 J p jit 2 (3)

A higher FDI likewise indicates a more diverse food supply structure. NFSD and FDI measure local food production and supply structures rather than residents' actual dietary intake, and are therefore interpreted as nutrition-related supply-side food system indicators.

Core explanatory variable

A composite digital infrastructure index is constructed across three dimensions: fixed-network access, mobile communication connectivity, and telecommunications service provision. Household broadband penetration is measured by the ratio of internet broadband access subscribers to the total number of households at year-end. Mobile communication connection density is measured by the ratio of mobile telephone subscribers to the permanent resident population at year-end. Telecommunications service provision intensity is measured by price-deflated telecommunications revenue per capita. After standardizing the three indicators, entropy weights are calculated and the indicators are aggregated to form a city-level composite digital infrastructure index. The indicators are presented in Table 1.

Construction of the composite digital infrastructure index.

Dimension | Indicator | Symbol | Calculation | Interpretation | Direction
Fixed-network access capacity | Household broadband penetration | BroadbandHH | Year-end internet broadband subscribers ÷ year-end total households × 100 | Fixed broadband connections per 100 households; reflects fixed-network access conditions and household connectivity. | Positive
Mobile communication connectivity | Mobile communication connection density | MobilePen | Year-end mobile telephone subscribers ÷ year-end permanent resident population × 100 | Mobile communication connections per 100 residents; reflects the intensity of residents' mobile-network connectivity. | Positive
Telecommunications service provision intensity | Telecommunications revenue per capita | TelecomService | Price-deflated telecommunications revenue ÷ year-end permanent resident population; natural logarithm used in the baseline specification | Reflects the scale of telecommunications service provision and use in the city; deflation to constant 2011 prices using the provincial consumer price index. | Positive

Control variables

Considering urban development and the principal determinants of agricultural food supply, this study controls for economic development (lnGDPpc), urbanization (Urban), fiscal support for agriculture (FiscalAgri), transport infrastructure (Road), industrial structure (Tertiary), population density (lnPopDen), and agricultural mechanization (Mech). Their definitions are provided in the variable definition table.

Mechanism variables

First, entry density of agricultural e-commerce firms (EcomEntry). This measure is constructed as ln(1 + x), where x is the number of newly established firms in a given year whose business scopes contain both agricultural terms, such as “agriculture,” “agricultural products,” or “agricultural by-products,” and online trading terms, such as “e-commerce,” “internet sales,” or “online sales,” divided by employment in the primary sector and multiplied by 10000. The indicator proxies the entry of agricultural digital trading businesses that smallholders, family farms, cooperatives, and small and micro businesses can connect with, thereby reflecting external conditions for producers' digital market participation.

Second, entry density of digital agricultural service providers (AgriServiceEntry). This measure is constructed as ln(1 + x), where x is the number of newly established firms in a given year whose business scopes contain terms such as “smart agriculture,” “agricultural informatization,” “agricultural Internet of Things,” “agricultural big data,” “agricultural software,” “agricultural remote sensing,” “agricultural technical services,” “agricultural technical advice,” or “pest and disease monitoring,” divided by employment in the primary sector and multiplied by 10,000. This indicator proxies locally available digital agricultural information and service provision conditions; it is not equated with agricultural machinery, cold-chain facilities, or actual investment in fixed assets.

Third, entry density of online food businesses and service providers (FoodOnlineEntry). This measure is constructed as ln(1 + x), where x is the number of newly established firms in a given year whose business scopes contain both food-related terms, such as “food,” “fresh food,” “agricultural products,” “agricultural by-products,” or “catering,” and online operation or delivery terms, such as “internet sales,” “e-commerce,” “online sales,” “takeaway,” “delivery,” or “instant delivery,” divided by the permanent resident population at year-end and multiplied by 10,000. This indicator proxies entry conditions for online food businesses and delivery services and characterizes the supply-side foundation of online food accessibility.

The definitions of the variables are presented in Table 2.

Variable definitions.

Variable type | Variable | Symbol | Measurement
Dependent variable | Local food supply diversity | NFSD | Shannon index based on local output shares of five major food categories; reflects production structure rather than residents' diets.
Core explanatory variable | Reconstructed digital infrastructure index | DII | Fixed-network access, mobile connectivity, and telecommunications service provision are combined using entropy weights and standardized.
Control variables | Economic development | lnGDPpc | Natural logarithm of GDP per capita.
Urbanization rate | Urban | Urban population as a share of total population.
Agricultural fiscal expenditure | FiscalAgri | Agriculture, forestry, and water expenditure as a share of general public budget expenditure.
Transport infrastructure | Road | Total road length (km) ÷ administrative area (km2).
Tertiary-sector share | Tertiary | Tertiary-sector value added as a share of regional GDP.
Population density | lnPopDen | Natural logarithm of population density.
Agricultural mechanization | Mech | Total agricultural machinery power (kW) ÷ cultivated land area (ha).
Mechanism variables | Entry density of agricultural e-commerce firms | EcomEntry | ln[1 + (new agricultural e-commerce firms in the year ÷ primary-sector employment) × 10 000]; business scopes contain both agricultural terms and terms such as e-commerce, internet sales, or online sales.
Entry density of digital agricultural service providers | AgriServiceEntry | ln[1 + (new digital agricultural service firms in the year ÷ primary-sector employment) × 10 000]; business scopes include terms such as smart agriculture, agricultural informatization, agricultural IoT, or agricultural technical advice.
Entry density of online food businesses and service providers | FoodOnlineEntry | ln[1 + (new online food business and service firms in the year ÷ year-end permanent resident population) × 10 000]; business scopes contain both food-related terms and online operation or delivery terms.

Model specification

Baseline regression model

Following related city-panel studies, a model with city and year fixed effects is specified to identify the relationship between digital infrastructure development and nutritional food supply diversity: NFSD it = α + βDI I it + γ ′ X it + μ i + λ t + ε it (4)

Here, NFSD it denotes nutritional food supply diversity in city i in year t; DI I it denotes digital infrastructure; X it is a vector of control variables; μ i denotes city fixed effects, controlling for time-invariant city characteristics such as natural endowments, location, and long-term institutional conditions; λ t denotes year fixed effects, absorbing macroeconomic policies, nationwide market changes, and common technological trends; and ε it is the random disturbance term. β is the central coefficient of interest. A significantly positive β indicates a significant positive relationship between digital infrastructure development and greater diversity and balance in local food-category output shares, as measured by NFSD. Standard errors in all baseline estimations are clustered at the city level to address serial correlation and heteroskedasticity within cities.

Mediation models

To further examine the channels through which digital infrastructure affects NFSD via the entry density of agricultural e-commerce firms, the entry density of digital agricultural service providers, and the entry density of online food businesses and service providers, a two-step mediation approach is adopted. First, the effect of digital infrastructure on the mechanism variable is examined: M it ( m ) = α m + a m DI I it + θ m ′ X it + μ i + λ t + ν it ( m ) (5)

Second, digital infrastructure and the mechanism variable are jointly included in the baseline model: NFSD it = α m ∗ + c m ′ DI I it + b m M it ( m ) + δ m ′ X it + μ i + λ t + ξ it ( m ) (6)

Here, M it ( m ) denotes mechanism variable m, where m ∈ {EcomEntry, AgriServiceEntry, FoodOnlineEntry}; a m measures the effect of digital infrastructure on the corresponding mechanism variable; b m measures the relationship between the mechanism variable and NFSD after controlling for digital infrastructure; and c m ′ is the digital infrastructure coefficient after including the mechanism variable. If both α m ∗ and b m are statistically significant, and the absolute value of c m ′ is lower than that of β in the baseline model, the corresponding channel is consistent with partial mediation.

Descriptive statistics

Table 3 reports descriptive statistics for the main variables. The sample covers 257 Chinese prefecture-level cities from 2011 to 2023, yielding 3,341 city–year observations. Local food supply diversity (NFSD) has a mean of 1.096, a standard deviation of 0.135, and minimum and maximum values of 0.631 and 1.575, respectively, indicating notable differences in local food production structures across cities. After entropy-weighted aggregation and standardization, the digital infrastructure index (DII) has a mean of 0, a standard deviation of 1, and a range from −2.468 to 3.233, demonstrating substantial variation in digital connectivity development across sample cities. Among the three component indicators, household broadband penetration, mobile communication connection density, and telecommunications revenue per capita (logarithm) have means of 37.117, 99.392, and 5.982, respectively. The mechanism variables—the entry density of agricultural e-commerce firms, digital agricultural service providers, and online food businesses and service providers—have means of 1.869, 1.800, and 1.918 and standard deviations of 0.621, 0.501, and 0.579, respectively, indicating differences in the entry of relevant digital businesses and service providers across cities. The other controls also display substantial sample variation, providing a data foundation for subsequent baseline regressions, mechanism tests, and heterogeneity analyses using city and year fixed effects.

Descriptive statistics.

Variable | N | Mean | Standard deviation | Minimum | Maximum
Local food supply diversity | 3,341 | 1.096 | 0.135 | 0.631 | 1.575
Reconstructed digital infrastructure index | 3,341 | 0.000 | 1.000 | −2.468 | 3.233
Household broadband penetration | 3,341 | 37.117 | 14.103 | 8.060 | 83.106
Mobile communication connection density | 3,341 | 99.392 | 19.989 | 46.830 | 160.180
Telecommunications revenue per capita (log) | 3,341 | 5.982 | 0.358 | 4.832 | 7.301
Entry density of agricultural e-commerce firms | 3,341 | 1.869 | 0.621 | 0.362 | 4.146
Entry density of digital agricultural service providers | 3,341 | 1.800 | 0.501 | 0.422 | 3.496
Entry density of online food businesses and service providers | 3,341 | 1.918 | 0.579 | 0.401 | 4.072
Economic development | 3,341 | 10.718 | 0.284 | 9.746 | 11.642
Urbanization rate | 3,341 | 0.497 | 0.152 | 0.120 | 0.879
Agricultural fiscal expenditure | 3,341 | 0.091 | 0.019 | 0.042 | 0.207
Transport infrastructure | 3,341 | 0.590 | 0.094 | 0.314 | 0.976
Tertiary-sector share | 3,341 | 0.445 | 0.064 | 0.215 | 0.663
Population density | 3,341 | 6.476 | 0.532 | 4.518 | 7.780
Agricultural mechanization | 3,341 | 1.415 | 0.310 | 0.718 | 2.825

Empirical results

Baseline regressions

Table 4 reports the baseline regression results for the effect of the digital infrastructure index on nutritional food supply diversity. Column (1) includes only the standardized digital infrastructure index (DII). Its estimated coefficient is 0.1011, positive and significant at the 1% level, suggesting that cities with more developed digital infrastructure generally have higher nutritional food supply diversity. Column (2) adds controls for economic development, urbanization, agricultural fiscal expenditure, transport infrastructure, the tertiary-sector share, population density, and agricultural mechanization. The DII coefficient decreases to 0.0736 but remains positive and significant at the 1% level. Thus, although part of the correlation reflects cities' development foundations, a significant positive relationship persists after controlling for observable characteristics. Column (3) additionally includes city fixed effects, yielding a DII coefficient of 0.0693, positive and significant at the 1% level. Column (4) adds year fixed effects, producing the two-way fixed-effects baseline model used in subsequent analyses. Its DII coefficient is 0.0664, positive and significant at the 1% level. Although the coefficient declines as controls and fixed effects are progressively added, its direction and significance remain stable. This indicates that the positive association between digital infrastructure and nutritional food supply diversity is not entirely driven by persistent city characteristics or common year shocks, providing baseline evidence for H1. This association concerns the local production mix across food categories and provides no direct evidence of changes in household diets, nutritional status, or health.

Baseline regression results.

Variable | (1) Baseline | (2) With controls | (3) City fixed effects | (4) Two-way fixed effects
Digital infrastructure index | 0.1011***(0.0026) | 0.0736***(0.0024) | 0.0693***(0.0016) | 0.0664***(0.0017)
Economic development | | 0.0618***(0.0146) | 0.0555***(0.0083) | 0.0280***(0.0087)
Urbanization rate | | −0.0021(0.0415) | 0.2178***(0.0420) | 0.0836*(0.0447)
Agricultural fiscal expenditure | | −0.0608(0.1192) | −0.0550(0.0650) | −0.0593(0.0637)
Transport infrastructure | | 0.1087***(0.0398) | 0.0215(0.0171) | −0.0101(0.0179)
Tertiary-sector share | | 0.1283**(0.0501) | 0.0999***(0.0284) | 0.0486*(0.0287)
Population density | | 0.0001(0.0090) | 0.1832***(0.0607) | −0.2095***(0.0735)
Agricultural mechanization | | 0.0250***(0.0073) | 0.0203***(0.0044) | 0.0092**(0.0046)
City fixed effects | No | No | Yes | Yes
Year fixed effects | No | No | No | Yes
Clustering level | City | City | City | City
N | 3,341 | 3,341 | 3,341 | 3,341
R2 | 0.559 | 0.605 | 0.771 | 0.466

City-clustered robust standard errors are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

Regarding magnitude, DII is standardized before estimation. The coefficient of 0.0664 in column (4) means that, holding other conditions constant, a one-standard-deviation increase in digital infrastructure is associated with an average increase of approximately 0.0664 index points in NFSD. Relative to the sample mean NFSD of 1.096, this change is approximately 6.05%, suggesting that the positive association has economic significance.

The estimated NFSD change of 0.0664 corresponds to approximately 6.05% of its sample mean of 1.096, providing a benchmark for interpreting its magnitude. Its economic meaning should also be understood in terms of the index’s structural nature. NFSD reflects relative output shares across food categories: proportional growth in all categories increases total output but leaves the index unchanged. A positive change therefore indicates a more diverse and balanced composition across the categories examined, rather than output expansion alone. Adjustments to producers’ product mixes and related marketing arrangements may take time, so a seemingly modest index change can still have substantive implications for supply composition. However, the percentage relative to the sample mean is a scale comparison, not an equivalent percentage improvement in food output, dietary diversity, or nutritional health. The baseline estimate alone also does not establish how long the compositional adjustment persists.

Several economic explanations may account for this positive effect. First, improved digital infrastructure reduces agricultural producers' costs of obtaining price, demand, weather, and technical information, enabling them to identify changes in demand for different food categories more promptly. It also reduces informational uncertainty when shifting from single-product production to multiple crop and livestock categories, increasing flexibility in production adjustments. Second, digital connectivity strengthens information matching among production, wholesale distribution, and consumer markets and expands the effective market reach of agricultural products. This is particularly relevant for vegetables, fruits, meat, and aquatic products, which have short shelf lives and require strong logistics coordination and market information. Digital trading and information platforms can reduce search costs, matching costs, and distribution frictions, allowing more food categories to enter urban markets. Third, by promoting e-commerce, fresh-food retail, and digital platforms, digital infrastructure may strengthen the feedback of residents' demand for diverse and higher-quality foods to producers. When information about demand for different food categories reaches producers and distributors more promptly, food supply structures are more likely to shift from relatively uniform quantity provision toward a more diverse mix of food categories with potential nutritional relevance.

Endogeneity analysis

Although the baseline results in Table 4 show a significant positive association between digital infrastructure and nutritional food supply diversity, the two-way fixed-effects model may still face endogeneity arising from reverse causality, omitted variables, and measurement error. Digital infrastructure may improve food production, distribution, and market matching, thereby increasing nutritional food supply diversity. Conversely, cities with more developed food industries and more diverse market demand may have stronger demand for digital infrastructure investment. Local governance capacity, historical commercial traditions, existing informatization, and regional market integration may also affect both digital infrastructure development and food supply structures. To examine the dynamic features of the results, one- and two-period lags of digital infrastructure are first used in estimation, followed by instrumental-variable estimation to provide additional evidence.

Lagged explanatory variable tests

Table 5 reports estimates using one- and two-period lags of the digital infrastructure index. In column (1), the estimated coefficient on one-period-lagged DII is 0.0498, positive and significant at the 1% level. In column (2), the coefficient on two-period-lagged DII is 0.0417 and is positive and significant at the 1% level. Compared with the positive contemporaneous DII coefficient in column (4) of Table 4, the lagged terms also exhibit positive associations.

Tests using lagged digital infrastructure indices.

Variable | (1) One-period lag | (2) Two-period lag
Digital infrastructure index (one-period lag) | 0.0498***(0.0022) |
Digital infrastructure index (two-period lag) | | 0.0417***(0.0027)
Economic development | 0.0567***(0.0102) | 0.0537***(0.0114)
Urbanization rate | 0.1373**(0.0533) | 0.1872***(0.0596)
Agricultural fiscal expenditure | −0.1418*(0.0779) | −0.1453*(0.0879)
Transport infrastructure | 0.0004(0.0226) | −0.0076(0.0241)
Tertiary-sector share | 0.0306(0.0352) | 0.0408(0.0378)
Population density | −0.3239***(0.0924) | −0.3212***(0.1023)
Agricultural mechanization | 0.0050(0.0060) | 0.0015(0.0063)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 3,084 | 2,827
R2 | 0.249 | 0.165

These results indicate that, in the current sample, the relationship between digital infrastructure and nutritional food supply diversity includes positive contemporaneous and lagged associations. The positive coefficient on the two-period lag should also not be interpreted directly as a causal effect. It may reflect dynamic adjustments in infrastructure investment, strong correlations among lagged variables, and reductions in the effective sample size. Including the lagged variables reduces the sample to 3,084 and 2,827 observations, respectively. Lagged-variable tests therefore cannot substitute for causal identification; the baseline results still require cautious interpretation alongside exogenous shocks or credible instruments.

Instrumental-variable estimation

This study constructs the interaction “historical communications infrastructure × national digitalization trend” as an instrument for digital infrastructure. Historical city communications infrastructure characterizes the pre-existing information and communication network endowments of different regions before digital infrastructure expansion. The national digitalization trend reflects common exogenous shocks arising from national technological progress, communication network upgrades, and digital diffusion during the sample period. Their interaction generates different levels of exposure to digital infrastructure expansion across cities according to their pre-existing communications infrastructure, creating variation across cities and over time.

The instrument is selected based primarily on relevance and exogeneity. First, digital infrastructure development exhibits strong historical path dependence. Cities with more developed historical communication networks generally have stronger communications conduits, network nodes, technical personnel, and user bases. When national broadband networks and digital technologies diffuse rapidly, these cities can typically complete network upgrades at lower marginal construction costs. The interaction “historical communications infrastructure × national digitalization trend” should therefore be strongly correlated with current city digital infrastructure development. The first-stage results in Table 6 show an estimated instrument coefficient of 0.7794, positive and significant at the 1% level, supporting its relevance to DII.

Re-estimation using an instrumental variable.

Variable | (1) First stage: DII | (2) Second stage: NFSD
Instrumental variable | 0.7794***(0.0141) |
Digital infrastructure index | | 0.0657***(0.0023)
Economic development | 0.3950***(0.0667) | 0.0282***(0.0087)
Urbanization rate | 0.9791***(0.3537) | 0.0836*(0.0447)
Agricultural fiscal expenditure | −0.8768*(0.4925) | −0.0606(0.0637)
Transport infrastructure | −0.0923(0.1522) | −0.0101(0.0179)
Tertiary-sector share | -0.1000(0.2592) | 0.0488*(0.0286)
Population density | −1.1321**(0.5514) | −0.2108***(0.0735)
Agricultural mechanization | −0.0542(0.0376) | 0.0091**(0.0046)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 3,341 | 3,341
First-stage clustered Wald F | 3037.240 |

Second, regarding exogeneity and the exclusion restriction, the national digitalization trend in the instrument is determined by national technological progress and communication infrastructure expansion, rather than solely by the nutritional food supply structure of an individual prefecture-level city in a given year. Historical communications infrastructure was established before the baseline study period, and city fixed effects can largely absorb its time-invariant direct effects. After additionally controlling for economic development, urbanization, fiscal support for agriculture, transport infrastructure, industrial structure, population density, agricultural mechanization, and year fixed effects, the interaction thus primarily affects nutritional food supply diversity through the current penetration of digital infrastructure.

Regarding instrument validity, Table 6 shows that the city-clustered first-stage Wald F statistic for the excluded instrument is 3037.240, indicating strong conditional relevance between the instrument and DII; this statistic cannot verify the exclusion restriction. Since one excluded instrument is used for one endogenous explanatory variable, the model is exactly identified, preventing Hansen J or Sargan overidentification tests. The exclusion restriction must therefore be interpreted cautiously in conjunction with the theoretical argument that historical communications infrastructure does not directly affect NFSD through other pathways.

Column (2) of Table 6 reports the second-stage results. The estimated DII coefficient is 0.0657, positive and significant at the 1% level, and similar in magnitude to the baseline estimate of 0.0664 in column (4) of Table 4. Thus, under the instrumental-variable specification, digital infrastructure remains significantly and positively associated with nutritional food supply diversity, providing additional evidence for the baseline results.

Robustness checks

Table 7 examines robustness in five ways: replacing the dependent variable, replacing the core explanatory variable, treating extreme values, adjusting the sample, and controlling more strictly for regional time shocks. All models use the two-way fixed-effects model in column (4) of Table 4 as the reference and retain the main controls.

Robustness checks.

Variable | (1) Alternative dependent variable | (2) Alternative explanatory variable | (3) 1% winsorization | (4) Excluding large cities | (5) Region × year fixed effects
Digital infrastructure index | 0.0370***(0.0009) | | 0.0660***(0.0016) | 0.0660***(0.0018) | 0.0663***(0.0017)
Mobile communication connection density | | 0.0632***(0.0017) | | |
Economic development | 0.0156***(0.0049) | 0.0294***(0.0087) | 0.0251***(0.0089) | 0.0345***(0.0094) | 0.0277***(0.0088)
Urbanization rate | 0.0467*(0.0249) | 0.0726(0.0453) | 0.0797*(0.0445) | 0.0936**(0.0473) | 0.0856*(0.0450)
Agricultural fiscal expenditure | −0.0327(0.0355) | −0.0700(0.0658) | −0.0543(0.0656) | −0.0528(0.0669) | −0.0523(0.0641)
Transport infrastructure | −0.0056(0.0099) | −0.0118(0.0185) | −0.0101(0.0181) | −0.0058(0.0187) | −0.0120(0.0179)
Tertiary-sector share | 0.0265*(0.0159) | 0.0569**(0.0288) | 0.0365(0.0284) | 0.0592**(0.0293) | 0.0459(0.0288)
Population density | −0.1177***(0.0410) | −0.2196***(0.0786) | −0.1780**(0.0729) | −0.2375***(0.0763) | −0.2066***(0.0745)
Agricultural mechanization | 0.0051**(0.0025) | 0.0105**(0.0048) | 0.0077*(0.0046) | 0.0075(0.0049) | 0.0093**(0.0045)
City fixed effects | Yes | Yes | Yes | Yes | Yes
Year fixed effects | Yes | Yes | Yes | Yes | Region × year
Clustering level | City | City | City | City | City
N | 3,341 | 3,341 | 3,341 | 3,003 | 3,341
R2 | 0.466 | 0.437 | 0.463 | 0.463 | 0.465

Column (1) replaces Shannon-based NFSD with a food supply diversity indicator constructed using the reverse Herfindahl index. The estimated DII coefficient is 0.0370, positive and significant at the 1% level, indicating that the positive association does not depend on the specific construction of the dependent variable. Column (2) replaces the composite DII with a standardized mobile-network connectivity indicator, yielding a coefficient of 0.0632, positive and significant at the 1% level. The findings therefore also hold when city digital conditions are measured through a single mobile communication connectivity dimension.

Column (3) winsorizes the main continuous variables at the lower and upper 1% tails. The estimated DII coefficient is 0.0660, positive and significant at the 1% level. In column (4), after excluding the 26 largest cities ranked by baseline population, the DII coefficient is 0.0660 and remains positive and significant at the 1% level. These results suggest that the baseline finding is not driven by a few extreme observations or major cities.

Column (5) additionally controls for region × year fixed effects to absorb common shocks affecting different regions in each year. The estimated DII coefficient is 0.0663, positive and significant at the 1% level, and close to the baseline estimate in Table 4. Across the five checks, the core explanatory variable retains a significant positive association after changing variable measures, treating extreme values, adjusting the sample, and strengthening fixed-effects controls, indicating that the baseline finding is robust.

Mechanism analysis

To avoid the conceptual mismatches among the previous proxies for market accessibility, production structure diversity, and total consumption, this study constructs mechanism variables using firm entry data. The entry density of agricultural e-commerce firms proxies producers' digital market participation; that of digital agricultural service providers proxies digital agricultural service accessibility; and that of online food businesses and service providers proxies online food accessibility. The mechanism regressions first examine the effect of DII on each entry density and then jointly include DII and the corresponding mechanism variable in the NFSD equation. Tables 8–10 report mechanism test results based on firm entry data, together with estimates for the controls.

Mechanism test: entry density of agricultural e-commerce firms.

Variable | (1) Mechanism variable | (2) NFSD
Digital infrastructure index | 0.3683***(0.0086) | 0.0566***(0.0021)
Entry density of agricultural e-commerce firms | | 0.0264***(0.0035)
Economic development | 0.0874*(0.0476) | 0.0257***(0.0086)
Urbanization rate | −0.2643(0.2318) | 0.0906**(0.0444)
Agricultural fiscal expenditure | 0.4843(0.3324) | −0.0721(0.0628)
Transport infrastructure | −0.1279(0.0973) | −0.0067(0.0174)
Tertiary-sector share | 0.0010(0.1613) | 0.0486*(0.0284)
Population density | −0.1069(0.3754) | −0.2067***(0.0723)
Agricultural mechanization | −0.0196(0.0275) | 0.0097**(0.0045)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 3,341 | 3,341
R2 | 0.475 | 0.477

Producer participation in digital markets

The first channel examined is producers' participation in digital markets. In Table 8, when the entry density of agricultural e-commerce firms is the dependent variable, the estimated DII coefficient is 0.3683, positive and significant at the 1% level. Improved digital connectivity is therefore significantly associated with greater entry of agricultural e-commerce firms. Theoretically, broadband networks and digital platforms improve producers' access to information about prices, orders, logistics, and market demand and increase matching efficiency among producers, wholesalers, retailers, and consumers, thereby expanding the effective market reach available to agricultural producers.

This mechanism is relevant to the local supply of multiple food categories. Compared with relatively storable products such as grains, vegetables, fruits, meat, and aquatic products are generally more perishable and require more timely transport. If producers cannot obtain demand, price, and logistics information promptly, their decisions are more likely to concentrate on a few food categories with relatively low transaction risks. By reducing market information frictions, digital infrastructure increases the likelihood that different foods can enter markets, creating larger potential markets for different food categories previously constrained by distance, information asymmetry, and limited sales channels.

When the entry density of agricultural e-commerce firms is added to the NFSD equation, its estimated coefficient is 0.0264, positive and significant at the 1% level. The DII coefficient declines from the baseline value of 0.0664 to 0.0566 but remains positive and significant at the 1% level. This is consistent with a partial transmission pathway of “digital infrastructure–producer participation in digital markets–food supply diversity.” Part of the role of digital infrastructure may operate through encouraging agricultural e-commerce firm entry and improving producers' connections with markets. Greater market reach and lower transaction frictions increase the expected returns from bringing different food categories to market, creating more favorable conditions for expanding local production across food categories such as fruits, vegetables, meat, and aquatic products. H2 is supported.

The economic meaning of producers' digital market participation thus extends beyond “better networks, more food.” By encouraging agricultural e-commerce firm entry and reducing informational and transaction frictions in food distribution, digital infrastructure makes more food categories commercially viable, promoting a shift from supply concentrated in a few categories toward the coexistence of multiple categories.

Digital agricultural service accessibility

The second mechanism concerns digital agricultural service provision. In Table 9, when the entry density of digital agricultural service providers is the dependent variable, the estimated DII coefficient is 0.2888, positive and significant at the 1% level. Improved digital infrastructure is therefore significantly and positively associated with greater entry of digital agricultural service providers.

Mechanism test: entry density of digital agricultural service providers.

Variable | (1) Mechanism variable | (2) NFSD
Digital infrastructure index | 0.2888***(0.0073) | 0.0617***(0.0020)
Entry density of digital agricultural service providers | | 0.0162***(0.0032)
Economic development | −0.0195(0.0463) | 0.0283***(0.0087)
Urbanization rate | −0.3593*(0.2131) | 0.0894**(0.0444)
Agricultural fiscal expenditure | 0.0279(0.3151) | −0.0597(0.0635)
Transport infrastructure | 0.0693(0.0968) | −0.0112(0.0179)
Tertiary-sector share | 0.1774(0.1544) | 0.0457(0.0286)
Population density | −0.4433(0.3843) | −0.2023***(0.0731)
Agricultural mechanization | 0.0018(0.0263) | 0.0091**(0.0045)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 3,341 | 3,341
R2 | 0.358 | 0.470

Mechanism test: entry density of online food businesses and service providers.

Variable | (1) Mechanism variable | (2) NFSD
Digital infrastructure index | 0.3448***(0.0072) | 0.0603***(0.0019)
Entry density of online food businesses and service providers | | 0.0176***(0.0033)
Economic development | 0.0309(0.0478) | 0.0274***(0.0087)
Urbanization rate | 0.1311(0.2064) | 0.0813*(0.0448)
Agricultural fiscal expenditure | 0.1170(0.3183) | −0.0613(0.0638)
Transport infrastructure | 0.0516(0.1050) | −0.0110(0.0177)
Tertiary-sector share | −0.0697(0.1521) | 0.0498*(0.0287)
Population density | 0.0927(0.3823) | −0.2111***(0.0726)
Agricultural mechanization | 0.0007(0.0277) | 0.0091**(0.0045)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 3,341 | 3,341
R2 | 0.442 | 0.471

This result is consistent with the information-based decision mechanism proposed earlier. In traditional agriculture, growing and raising multiple product categories typically requires more complex technical knowledge, market assessment, and management. Producers must understand production techniques for different varieties and assess future prices, market demand, and sales channels. When information acquisition is costly, they may prefer a few familiar crops or established products to reduce operating risks. Improved digital infrastructure provides more timely information on prices, weather, pests and diseases, agricultural technologies, and end-user demand. Online technical advice and outsourced agricultural services also reach producers more easily, lowering the informational and coordination costs associated with producing multiple categories.

When the entry density of digital agricultural service providers is added to the NFSD equation, its estimated coefficient is 0.0162, positive and significant at the 1% level. The DII coefficient decreases from 0.0664 to 0.0617, remaining positive and significant at the 1% level. This suggests that digital agricultural service accessibility may be one channel linking digital infrastructure to food supply diversity. Digital infrastructure increases service provider entry and service matching efficiency, enabling producers to obtain technical, early-warning, and market information more easily and to adjust cultivation, husbandry, and product combinations accordingly. The actual effects nevertheless depend on local agricultural resources, service provision capacity, physical infrastructure, and market demand. H3 is supported.

Online food accessibility and demand feedback

The third pathway originates on the demand side of the food system. In Table 10, when the entry density of online food businesses and service providers is the dependent variable, the estimated DII coefficient is 0.3448, positive and significant at the 1% level. Cities with greater digital connectivity thus generally experience more entry by online food businesses and delivery service providers.

Digital infrastructure may affect residents' food demand in two main ways. First, e-commerce, online retail, and instant delivery reduce the costs of searching for, comparing, and purchasing different foods, bringing foods previously constrained by retail outlet coverage and spatial distance into residents' choice sets. Second, the internet accelerates the dissemination of food, nutrition, and health information, making it easier for consumers to develop demand for diverse foods, including fruits, vegetables, meat, eggs, dairy, and aquatic products. Digitalization may therefore change not only “what producers can produce,” but potentially also “what consumers are willing to buy.”

When the entry density of online food businesses and service providers is included in the NFSD equation, its estimated coefficient is 0.0176, positive and significant at the 1% level. The DII coefficient declines to 0.0603, remaining positive and significant at the 1% level. Part of the food supply structural effect of digital infrastructure may therefore stem from improved online food accessibility and associated demand feedback. As residents can more easily search for, compare, purchase, and receive delivery of foods, new consumption signals feed back into distribution and production through prices, online orders, retail sales, and market procurement. These signals increase producers' and businesses' incentives to expand supplies across food categories. H4 is supported.

This result has particular significance for food systems. Entry of online food businesses and service providers may improve the conditions for food information access, purchasing, and delivery. The proposed feedback mechanism—“improved digital connectivity–greater online food accessibility–stronger demand signals–adjustments in production and supply structures”—is consistent with a potential demand channel associated with diversification in local food-supply composition.

Taken together, the three mechanisms suggest that digital infrastructure may affect nutritional food supply diversity through three interconnected stages: producer participation in digital markets, digital agricultural service accessibility, and online food accessibility. Agricultural e-commerce firm entry expands producers' channels of market participation; digital agricultural service provider entry improves the matching of technical and information services; and online food business and service provider entry strengthens the feedback of demand signals into production and distribution. Rather than operating independently, these channels are more likely to form a food system cycle of “information connectivity–market participation–production adjustment–demand feedback.”

Heterogeneity analysis

The results above reflect average effects across 257 prefecture-level cities. Whether digital infrastructure translates into improved food supply structures also depends on local agricultural production foundations, market conditions, and stages of urban development. Heterogeneity is therefore examined across three dimensions: regional development, agricultural resource endowments, and urbanization. These dimensions are interpreted within the technology–organization–environment perspective outlined in Section 3.1: they represent overlapping production, organizational and market contexts, rather than independent or exhaustive measures of the three dimensions.

Regional development heterogeneity: eastern versus central and western regions

Table 11 reports the regional regression results. The estimated coefficient on the digital infrastructure index (DII) is 0.0649 for eastern cities and 0.0671 for central and western cities, both positive and significant at the 1% level. After controlling for other relevant factors, the positive relationship between digital infrastructure and nutritional food supply diversity is therefore not confined to one region but appears across regions with different development stages and resource conditions. The estimated coefficient for eastern cities is slightly lower than that for central and western cities, although the between-group difference is not statistically significant. Theoretically, eastern regions generally have more developed transport, cold-chain storage, logistics and delivery, and market service systems. These can complement digital connectivity, improving information transmission, order matching, and interregional distribution and facilitating market entry for a diverse range of food categories.

Regional development heterogeneity.

Variable | (1) Eastern | (2) Central and western
Reconstructed digital infrastructure index | 0.0649***(0.0027) | 0.0671***(0.0021)
Economic development | 0.0282**(0.0124) | 0.0263**(0.0122)
Urbanization rate | 0.1711**(0.0709) | 0.0294(0.0577)
Agricultural fiscal expenditure | −0.0637(0.1018) | −0.0416(0.0835)
Transport infrastructure | 0.0163(0.0279) | −0.0327(0.0234)
Tertiary-sector share | 0.0748(0.0459) | 0.0272(0.0373)
Population density | −0.2518**(0.1094) | −0.1700*(0.1009)
Agricultural mechanization | 0.0112(0.0077) | 0.0079(0.0056)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 1,352 | 1,989
R2 | 0.460 | 0.470

Agricultural resource endowment heterogeneity: high versus low

Table 12 reports regressions grouped by agricultural resource endowments. The estimated digital infrastructure index (DII) coefficient is 0.0674 in cities with high agricultural resource endowments and 0.0652 in those with low endowments, both positive and significant at the 1% level. Thus, the positive relationship between digital infrastructure and local food supply diversity exists under different agricultural resource conditions. By comparison, the coefficient is slightly larger in high-endowment cities, although the between-group difference is not statistically significant. Theoretically, the role of digital infrastructure in promoting diversity may depend more heavily on physical production conditions, including land, water resources, agricultural labor, mechanization, and the existing agricultural industrial base. Digital connectivity improves information acquisition, market matching, and service provision, but translating these advantages into supplies across multiple food categories still requires production capacity and a resource base.

Agricultural resource endowment heterogeneity.

Variable | (1) High agricultural endowment | (2) Low agricultural endowment
Reconstructed digital infrastructure index | 0.0674***(0.0022) | 0.0652***(0.0025)
Economic development | 0.0235*(0.0130) | 0.0331***(0.0122)
Urbanization rate | 0.0764(0.0621) | 0.0894(0.0658)
Agricultural fiscal expenditure | −0.0195(0.0937) | −0.1036(0.0855)
Transport infrastructure | −0.0075(0.0243) | −0.0113(0.0271)
Tertiary-sector share | 0.0060(0.0348) | 0.0901**(0.0455)
Population density | −0.2122**(0.1054) | −0.2078*(0.1071)
Agricultural mechanization | 0.0076(0.0060) | 0.0109(0.0070)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 1,677 | 1,664
R2 | 0.490 | 0.440

In cities with relatively weak agricultural resource endowments, improved digital networks may enhance market information, technical services, and transaction matching. However, limited land resources, a weak agricultural industrial base, or a lack of practical conditions for producing multiple categories can restrict the scope for translating digital benefits into adjustments in local food production structures. Digital infrastructure therefore cannot replace the physical agricultural base; it is more likely to complement agricultural resource endowments and production conditions in jointly promoting local food supply diversity.

Urbanization heterogeneity: high versus low

Table 13 reports regressions grouped by urbanization. The estimated digital infrastructure index (DII) coefficient is 0.0698 for highly urbanized cities and 0.0631 for less urbanized cities, both positive and significant at the 1% level. The positive association between digital infrastructure and nutritional food supply diversity therefore appears across different urbanization stages. The coefficient is slightly larger in highly urbanized cities, possibly because their retail networks, logistics and delivery systems, and market information channels are more developed. These can form stronger complementarities with digital connectivity, improving information transmission, order matching, and supply–demand coordination and facilitating adjustments in food supply structures.

Urbanization heterogeneity.

Variable | (1) High urbanization | (2) Low urbanization
Reconstructed digital infrastructure index | 0.0698***(0.0024) | 0.0631***(0.0023)
Economic development | 0.0229*(0.0128) | 0.0338***(0.0120)
Urbanization rate | 0.0933(0.0602) | 0.0663(0.0663)
Agricultural fiscal expenditure | −0.1301(0.0915) | −0.0177(0.0868)
Transport infrastructure | −0.0089(0.0240) | −0.0097(0.0274)
Tertiary-sector share | 0.0458(0.0411) | 0.0474(0.0401)
Population density | −0.1935*(0.1029) | −0.2176**(0.1061)
Agricultural mechanization | 0.0139**(0.0057) | 0.0031(0.0075)
City fixed effects | Yes | Yes
Year fixed effects | Yes | Yes
Clustering level | City | City
N | 1,677 | 1,664
R2 | 0.490 | 0.440

Less urbanized cities can likewise improve information acquisition and transaction matching through digital infrastructure, but the translation of these benefits may be constrained by small markets, incomplete logistics systems, and insufficient relevant service providers. The between-group coefficient differences therefore reflect possible complementarities between digital connectivity and existing market conditions and do not by themselves establish that one type of city has a significantly stronger marginal effect.

Combining the heterogeneity tests for regions, agricultural resource endowments, and urbanization, no stable, statistically significant between-group differences in the effects of digital infrastructure on nutritional food supply diversity have been identified. The advantages of digital technologies in market access, information services, and demand feedback require support from an agricultural resource base, transport and logistics conditions, and urban market development. Digital infrastructure, the physical agricultural base, and the stage of urban development thus work together to shape diversity in local food supply structures.

A configurational perspective helps connect these contextual explanations. Zhou (2026) shows that AI adoption in environmental governance is associated with alternative combinations of technological and complementary institutional, cognitive and social conditions. Yan et al. (2026), combining necessary condition analysis and fsQCA, similarly identify multiple configurations associated with persistent green patenting in textile firms and find no individual condition to be necessary. Applied conceptually here, these studies suggest that stronger connectivity may support food-supply adjustment when combined with suitable production resources, service capacity and distribution arrangements; different combinations may provide different opportunities or constraints. However, the subgroup estimates do not identify these combinations as sufficient or necessary conditions, establish interactions among them, or demonstrate uniformly significant between-group differences. The cited studies therefore provide a theoretical and methodological reference for conditionality, not direct evidence of food-system effects or proof of configurational causality in this panel. The relevant outcome remains local food-category output composition, rather than household diets or health.

Further analysis: do digital infrastructure effects diminish at the margin?

The preceding urbanization heterogeneity analysis shows a slightly larger estimated digital infrastructure coefficient in highly urbanized cities than in less urbanized cities. This motivates a further question: as digital infrastructure development increases, does its positive association with nutritional food supply diversity exhibit diminishing marginal effects?

To test this nonlinear relationship, the squared term of the centered digital infrastructure index (DII) is added to the model with city and year fixed effects. Table 14 shows an estimated coefficient of 0.0653 on the linear DII term, positive and significant at the 1% level. The coefficient on the squared term is 0.0022 and is positive and significant at the 5% level. This result does not contradict the preceding urbanization heterogeneity analysis. A relatively larger DII coefficient in highly urbanized cities merely indicates that average associations between digital infrastructure and nutritional food supply diversity may vary across stages of urban development; it does not directly establish a significant nonlinear relationship in the full sample. The squared term is significant at the 5% level, indicating positive curvature in the relationship between digital infrastructure and nutritional food supply diversity and providing no support for diminishing marginal effects.

Digital infrastructure nonlinearity test.

Variable | (1) Nonlinear model
Centered digital infrastructure index | 0.0653***(0.0017)
Squared digital infrastructure index | 0.0022**(0.0010)
Economic development | 0.0282***(0.0088)
Urbanization rate | 0.0856*(0.0446)
Agricultural fiscal expenditure | −0.0620(0.0637)
Transport infrastructure | −0.0112(0.0179)
Tertiary-sector share | 0.0495*(0.0288)
Population density | −0.2044***(0.0733)
Agricultural mechanization | 0.0089*(0.0045)
City fixed effects | Yes
Year fixed effects | Yes
Clustering level | City
N | 3,341
R2 | 0.467

At the current stage of development, digital infrastructure therefore retains the potential to improve local food supply structures. There is no evidence that its positive association has generally entered a clearly saturated or declining phase. Policy should continue to improve digital infrastructure in areas with relatively weak connectivity while coordinating digital networks with agricultural production capacity, cold-chain logistics, and market service networks to improve the efficiency with which digital infrastructure supports diversity in local food-supply composition.

Study limitations

This study places digital infrastructure and local food supply diversity within a common analytical framework, emphasizing that network access may affect local food production structures through information, services, markets, and demand feedback. Compared with studies focusing only on total food quantities, this perspective gives greater attention to the structural mix of food categories, but it cannot directly establish residents' dietary or nutritional health outcomes. “Supply diversity” must nevertheless be carefully distinguished from “improved nutritional intake.” NFSD is constructed from cities' local output of major foods and measures structural diversity on the supply side rather than residents' final dietary intake. The relationship between local output and final consumption is also shaped by interregional shipments, wholesale market systems, cold-chain logistics, prices, and residents' purchasing power. Accordingly, this study does not interpret NFSD directly as a nutritional health outcome, but as the supply foundation of nutrition-related food systems. Testing whether this structural diversity translates into better diets, nutritional status, or health requires household consumption or nutrition data and lies outside the direct scope of the present analysis. A second limitation concerns the measurement of digital infrastructure. The composite index uses household broadband penetration, mobile communication connection density, and telecommunications revenue per capita. Although it better approximates the multidimensional concept of fixed access, mobile connectivity, and telecommunications service provision than a single broadband subscriber measure, it still does not fully capture network quality, platform applications, data capabilities, or actual usage behavior.

Conclusions and policy implications

Conclusion

Using a balanced panel of 257 cities for 2011–2023, this study constructs a local food supply diversity index and remeasures digital infrastructure across household broadband access, mobile communication connectivity, and telecommunications service provision. The two-way fixed-effects results show that the standardized digital infrastructure index (DII) is significantly and positively associated with local food supply diversity, with an estimated coefficient of 0.0664, significant at the 1% level. The findings remain consistent when using an alternative dependent variable, an alternative mobile-network connectivity measure, winsorization, sample adjustments, and region × year fixed effects, indicating a robust positive association between digital infrastructure and local food supply diversity.

The mechanism tests further show that digital infrastructure is significantly and positively associated with the entry density of agricultural e-commerce firms, digital agricultural service providers, and online food businesses and service providers. All three mechanism variables also have significant positive relationships with local food supply diversity in the outcome equations. Digital infrastructure may therefore promote a more diverse local food supply structure by expanding producers' digital market participation, strengthening the provision and matching of digital agricultural services, and improving conditions for online food businesses and the transmission of demand information.

Overall, the food system effects of digital infrastructure extend beyond expanding information connectivity or improving general transaction efficiency. By influencing business entry, agricultural service provision, and online food distribution, it may also be associated with a more diverse mix of local food categories. However, these effects continue to depend on physical conditions, including agricultural resource endowments, production capacity, cold-chain logistics, and market service systems. Policies should therefore continue to im