The distinction the authors draw is subtle but important. A city could expand every food category proportionally and increase total output without changing its supply composition at all; the NFSD index would remain unchanged. Conversely, shifts in the relative shares of grains, vegetables, fruits, meat, and aquatic products alter diversity even when total output is flat. Because the index is built from relative output shares, a positive change signals a more balanced production structure rather than simple output growth. The authors are careful to note that the 6.05 percent figure is a scale comparison against the sample mean, not an equivalent percentage improvement in food output, dietary diversity, or nutritional health.

The dataset is a balanced panel of 3,341 city-year observations. Food production data were compiled from provincial and prefecture-level statistical yearbooks and city statistical bulletins, while digital infrastructure was measured across three dimensions: household broadband penetration, mobile telephone penetration, and price-deflated telecommunications revenue per capita, deflated to constant 2011 prices. These were standardized and combined using entropy weighting, a design the authors argue reduces measurement bias compared with simply dividing broadband subscriptions by population. Descriptive statistics show substantial variation: NFSD ranges from 0.631 to 1.575, and the standardized digital index spans from −2.468 to 3.233.

The baseline model uses city and year fixed effects, controlling for time-invariant city characteristics such as natural endowments and location, and for common national shocks such as macroeconomic policy and technological trends. The digital infrastructure coefficient declines from 0.1011 in a simple specification to 0.0664 with full controls and two-way fixed effects, but its direction and significance remain stable. Standard errors are clustered at the city level to address serial correlation and heteroskedasticity within cities.

Because digital infrastructure and food supply diversity could plausibly influence each other, the authors confronted endogeneity directly. Lagged tests show one-period and two-period lags of the digital index carry positive coefficients of 0.0498 and 0.0417 respectively, both significant at the 1 percent level, though the authors caution these cannot be read as causal effects. They then constructed an instrumental variable by interacting historical city communications infrastructure with the national digitalization trend, exploiting the fact that cities with stronger pre-existing communication networks could upgrade at lower marginal cost when national broadband expansion arrived. The first-stage coefficient is 0.7794 with a city-clustered Wald F statistic of 3,037.24, indicating strong relevance. The second-stage estimate of 0.0657 is nearly identical to the baseline, though the authors note the model is exactly identified, so overidentification tests cannot verify the exclusion restriction.

Robustness checks reinforce the finding. Replacing the Shannon-based NFSD with a reverse Herfindahl index yields a coefficient of 0.0370; substituting a single mobile-network connectivity measure gives 0.0632; winsorizing the main variables at the 1 percent tails gives 0.0660; excluding the 26 largest cities gives 0.0660; and adding region-by-year fixed effects gives 0.0663. All remain positive and significant at the 1 percent level, suggesting the result is not driven by extreme observations, major cities, or regional shocks.

The mechanism analysis is where the study breaks new ground. Rather than using generic proxies such as road density or total retail sales, the authors measured three channels through new firm registration data drawn from Qichacha, Tianyancha, and the CSMAR database. The first channel is producers’ digital market participation, proxied by the entry density of agricultural e-commerce firms. The second is digital agricultural service accessibility, proxied by entry of providers in fields such as smart agriculture, agricultural informatization, and the agricultural Internet of Things. The third is online food accessibility, proxied by entry of online food businesses and delivery service providers. In each case, the digital infrastructure index is significantly and positively associated with firm entry, and adding the mechanism variable to the outcome equation reduces the digital index coefficient while leaving it significant, a pattern consistent with partial mediation.

The economic logic behind these channels is grounded in the particular challenges of perishable foods. Vegetables, fruits, meat, and aquatic products have short shelf lives and demand timely transport and logistics coordination. When producers cannot obtain demand, price, and logistics information promptly, they tend to concentrate on a few low-risk categories. By reducing search, matching, and distribution frictions, digital connectivity makes more food categories commercially viable and widens the effective market reach of smallholders, cooperatives, and small and micro businesses. On the demand side, easier online searching, ordering, and delivery generate consumption signals that feed back upstream through prices, platform orders, and retail procurement, giving producers incentives to adjust their product mix.

Heterogeneity analyses reveal that the association is broad but conditional. The estimated coefficient is 0.0649 in eastern cities and 0.0671 in central and western cities; 0.0674 in cities with high agricultural resource endowments and 0.0652 in low-endowment cities; and 0.0698 in highly urbanized cities versus 0.0631 in less urbanized ones. All are significant at the 1 percent level, but none of the between-group differences is statistically significant. The authors interpret this through a technology-organization-environment lens: digital connectivity is an enabling condition whose effects depend on the physical agricultural base, transport and cold-chain logistics, and the stage of urban market development. A further test adding a squared term found a positive and significant coefficient of 0.0022 at the 5 percent level, indicating no evidence of diminishing marginal effects at the current stage of development.

The authors are explicit about the limits of their findings. NFSD measures the diversity of local production, not what residents actually eat; urban consumers obtain much of their food through interregional trade, wholesale markets, and e-commerce, so local output structure is a supply foundation rather than a dietary outcome. Whether this structural diversity translates into better diets, nutritional status, or health requires household consumption and nutrition data that lie outside the study’s scope. The digital infrastructure index also does not capture network quality, platform applications, or actual usage behavior. The policy implication the authors draw is coordination: continue expanding connectivity where it is weak, but pair digital networks with agricultural production capacity, cold-chain logistics, storage, and market service networks so that better information can actually be converted into a more diverse and nutritionally relevant local food supply.

Subject of Research: The association between digital infrastructure and nutritional food supply diversity in Chinese prefecture-level cities

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

Article References: Li, C., & Chen, C. (2026). Digital infrastructure and local food supply diversity: evidence from 257 Chinese prefecture-level cities. Frontiers in Sustainable Food Systems, 10, Article 1987334. https://doi.org/10.3389/fsufs.2026.1987334

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1987334

Keywords: digital infrastructure, food supply diversity, Shannon diversity index, China, prefecture-level cities, agricultural e-commerce, digital agricultural services, online food accessibility, nutrition-sensitive food systems, panel data, instrumental variable, food system transformation