Digital transformation is often expected to enhance regional innovation, yet a high level of digitalization does not necessarily translate into real-economy upgrading. This study examines whether digital–real economy integration (DREI) increases Little Giant SME recognitions and whether such effects spill over across cities.
Methods
Using a balanced panel of 276 Chinese cities from 2019 to 2023, DREI is constructed with a coupling–coordination index based on entropy-weighted digital and real-economy subindices. Two-way fixed-effects models and a spatial Durbin model are applied to distinguish local impacts from indirect spillovers. Channel regressions using credit depth and R&D intensity assess the plausible mechanisms, and regime dependence is evaluated with a fiscal-capacity threshold specification. Measurement robustness is assessed using alternative DREI reconstructions.
Result
The results consistently show that higher DREI is associated with more Little Giant recognitions, with indirect spillovers accounting for a large share of the total effect. Credit deepening emerges as the most immediate channel, while short-run innovation mediation is weaker in this short panel. Threshold evidence indicates larger marginal gains in fiscally constrained cities, which is consistent with diminishing returns where fiscal capacity is higher.
Discussion
Policy implications point to diffusion-ready integration and a local implementation plus regional coordination approach, with greater marginal attention to fiscally constrained areas; moreover, the evidence suggests that spillover effects account for a large share of the total impact, indicating that coordination across cities is essential for translating DREI into broader recognition gains.
coupling–coordination degreedigital–real economy integrationLittle Giant SMEsspatial Durbin modelspatial spilloverThe author(s) declared that financial support was not received for this work and/or its publication.section-at-acceptanceSocial PhysicsIntroduction
Digital transformation is increasingly understood as a capability shift in which data, connectivity, and advanced computing reshape how firms organize and innovate, rather than merely expanding the ICT capital [1]. Yet, international evidence shows that digital gains are far from automatic: returns depend on complementary intangible investments such as skills, organizational capital, and governance, so benefits diffuse unevenly across firms and regions with different absorptive capacities [2, 3]. Regional studies further suggest that improved digital access can even widen spatial inequality unless stronger formal and informal institutions help spread gains beyond the early adopters [4]. This puzzle places digital–real economy integration (DREI) at the center of regional innovation debates because what matters is not digital intensity alone but how effectively digital tools are embedded into real-economy processes and networks; China’s pronounced cross-regional variation in digital foundations and innovation gaps offers a useful setting to examine these mechanisms [5].
China’s policy agenda has increasingly shifted from digitizing isolated business functions to embedding digital technologies into real-economy processes that strengthen supply-chain coordination, improve quality management, and support productivity upgrading under the rubric of DREI [6]. This orientation is formalized in central policy documents that emphasize integration as a means of upgrading manufacturing and modern services and improving the efficiency of production networks. In parallel, the Little Giant program identifies and supports specialized and innovation-oriented small and medium enterprises through staged national recognition and complementary policy support, providing an observable certification-type outcome that reflects the upgrading and innovation capability [6, 7]. China also displays pronounced regional differences in digital foundations, industrial structure, and innovation capacity, creating rich within-country variation to examine whether DREI is systematically associated with Little Giant recognition (LGR) and whether any effects extend beyond local boundaries through inter-regional linkages. A central question arises against this backdrop. Does deeper regional DREI increase the likelihood that capable firms attain Little Giant status, and do any effects spill over to neighboring areas?
The study can be placed at the intersection of regional innovation systems, digital transformation, and SME industrial policy. It speaks to production-network research that links technology adoption, quality management, and standards compliance to supplier development and labor mobility and to measurement work that treats integration as a system property rather than as a sectoral-level phenomenon. Methodologically, it belongs to spatial econometrics, which models endogenous interaction across contiguous and economically linked regions, a particularly relevant perspective given cross-jurisdictional computing backbones and supply-chain corridors [6, 8, 9].
Recent studies focus on how the digital economy promotes the development of Little Giant small- and medium-sized enterprises. Most studies treat digitalization as an independent driver and evaluate the effects on firm creation, scaling, and survival. Methodologically, the literature relies mainly on conventional panel estimators such as fixed effects, system GMM, and difference-in-differences. When spatial analysis is included, it typically examines a single variable in isolation, such as mapping the level of the digital economy or a proxy for integration or charting the clustering of Little Giants with Moran’s I and hot-spot analysis, without modeling how integration dynamics translate into recognized outcomes for these firms.
This study addresses three gaps in the literature on digital transformation and regional innovation. First, existing work often proxies integration using the level of the digital economy, which blurs the distinction between digital scale and the effective embedding of digital tools in real-economy upgrading. We, therefore, treat DREI as a joint integration construct that captures coordinated evolution and conversion between digital and real systems, which is consistent with the policy emphasis on deep integration. Second, the regional innovation consequences of digital integration are frequently studied under independence assumptions, with limited attention to cross-regional interaction through production networks and platform connectivity. This omission risks attributing network-based influences to local conditions. Third, research on integration and research on the geography of Little Giants have largely evolved in parallel, leaving open the subject of how integration capability translates into innovation-oriented certification outcomes within a unified empirical framework.
This study makes three contributions. First, it clarifies DREI as an embedding and coordination capability that captures how effectively digital technologies are integrated into real economy upgrading rather than a scale proxy for digitalization or overall development. Second, it reframes the analysis regarding general mechanisms that connect digital transformation to regional innovation outcomes, including coordination along production networks, improvements in the information environment that facilitate financing and compliance, and diffusion through spatial networks. Third, it shows that these mechanisms operate within a spatial system in which local outcomes partly depend on neighboring integration, thereby extending the discussion beyond context-specific features and linking the evidence to broader debates on digital transformation, regional innovation, and spatial spillovers. The study complements these theoretical contributions with a transparent spatial framework and robustness checks for measurements.
Literature review
The specialized, sophisticated, distinctive, and innovative program identifies Little Giant firms as niche leaders with verified specialization, stable supply-chain participation, quality certification, and evidentiary growth potential. Recognitions unlock financing windows, tax preferences, and ecosystem services, thereby turning the count of recognitions into a tractable and externally validated policy outcome. Recent work uses LGR status to evaluate high-quality development among specialized SMEs and study how business environment configurations support capability upgrading [10–12]. Yet, a methodological issue remains. LGR is a form of recognition rather than an outcome of the operating performance, which raises questions about selection, auditing standards, and the role of local administrative capacity. Studies that treat recognitions as outcomes often mitigate these concerns with fixed effects, placebo checks, and by linking recognitions to post-recognition behavior. This literature motivates using LGR counts as an evaluation endpoint while remaining explicit about identification.
Early research on the digital economy often relied on scale proxies such as broadband penetration or platform counts. Recent evidence also highlights information technology and telecommunications infrastructure as a foundational driver of the digital economy, supporting the inclusion of connectivity and infrastructure indicators in digital subindices [13]. More recent work conceptualizes digital–real economy integration as coordination among digital infrastructure, data governance, and production systems and commonly operationalizes it using coupling–coordination frameworks that aggregate multiple subsystems into a comparable index across regions and time [14–17]. In accordance with this coordination view, research on green economy transitions also treats upgrading as a coordinated system process and highlights digital technologies for data collection and processing as an influential driver of sectoral performance improvements [18]. This approach is better suited to certification settings where interoperability, traceability, and standards adoption matter more than the digital volume. Two caveats remain: the indicator choice may embed policy priorities and induce construct drift, and composite coordination measures that mix stocks and flows complicate the interpretation of short-run shocks. Triangulating the coupling–coordination index with micro evidence on platform use and data quality can partly address these concerns and strengthen the validity claims.
Recent international research further emphasizes that digital gains depend on complementary capabilities such as skills, organizational capital, and governance, which help explain persistent heterogeneity across firms and regions [1–3]. This perspective encourages interpreting DREI as an embedding capability rather than a proxy for aggregate development, and it reinforces the need to distinguish integration from scale in both measurement and inference [4].
A large body of evidence connects digital expansion to deeper and better-allocated credit. Studies document that richer verifiable information reduces screening costs, improves model-based risk pricing, and broadens access to working capital and investment loans for capable SMEs [19–21]. At the regional scale, digital infrastructure is associated with stronger credit markets, more effective collateralization, and lower misallocation, which are precisely the conditions under which specialized suppliers upgrade to meet certification thresholds. This mechanism satisfies three criteria for mediation analysis. First, integration predicts credit depth after absorbing city and year heterogeneity. Second, credit depth associates with recognition outcomes that are conditional on integration. Third, the product of the two effects is economically meaningful relative to the total effect. A recurring advantage of the finance route is its adjustment speed. Banking capacity, registries, and risk models respond within policy cycles that are shorter than typical innovation budgeting. The literature, therefore, expects a pronounced mediated effect through credit in short and medium horizons. Digital–real coordination can reduce the user cost of research and development through shared platforms, accessible data, and standardization of interfaces. Aggregate relationships between the digital economy and innovation are positive across several settings, yet they display lag structures and spatial concentration in a limited set of hubs [22, 23]. Innovation spending often follows multi-year plans and may require sustained coordination before observable changes occur. This implies a secondary role for R&D mediation in short panels. Two boundary conditions appear repeatedly. First, absorptive capacity conditions the translation of integration into research spending. Second, public innovation programs can crowd in or crowd out private R&D, which complicates identification. These features motivate treating R&D as a complementary channel and interpreting mediated estimates with attention to timing.
Studies in spatial econometrics argue that policy and market interactions generate interdependence across regions. The spatial Durbin model (SDM) has become a preferred workhorse because it nests SAR, SEM, and SLX specifications and enables impact decomposition into direct, indirect, and total effects, which aligns with inference under interdependence in regional policy settings [24–26]. Work on small- and medium-sized enterprises highlights diffusion channels that matter for recognitions and capability formation, including shared certification services, supplier development programs, and cross-jurisdictional finance linkages documented in recent analyses of the Little Giant initiative and the related SME ecosystems [10–12]. This diffusion logic is also consistent with the view that some digital technologies spread in a quasi-viral manner through rapid adoption and feedback effects, which can amplify cross-regional spillovers beyond administrative borders [27].
Recent research frames thresholds as capacities and governance conditions that influence how digital development translates into measurable SME outcomes. In the public finance strand, fiscal design and decentralization shape the resources and screening capacity available for innovation support and certification. City–year evidence shows that fiscal arrangements can alter innovation incentives and the effectiveness of public investment, with regime shifts that are detectable in short panels using panel threshold methods [28–30]. On the industrial structure side, studies utilizing the SDMs report that digital development fosters structural upgrading and that these effects exhibit spatial propagation and nonlinearity along the development path [29]. Together, these literatures justify treating public finance and industrial structure as threshold variables when explaining recognition-type outcomes for specialized SMEs. For estimation and interpretation under interdependence, the SDM with impact decomposition remains standard, while panel thresholds identify regime-dependent slopes associated with observable fiscal readiness and structural sophistication [26, 31].
The literature reveals several interconnected gaps. Coordination-based measures of digital–real integration have rarely been linked directly to LGRs, and mechanism studies rarely evaluate credit deepening and R&D intensity within a unified framework to assess their relative contributions. Evidence on regime dependence is largely conceptual and is not tested with observable thresholds grounded in public finance and industrial structure. Identification risks from networked endogeneity and timing are frequently under-addressed, and there has been limited sensitivity analysis on the construct validity of coordination indices. Building on these gaps, the present study links a coordination-based DREI index to LGR counts; implements parallel mediation through finance and R&D; estimates SDMs with direct, indirect, and total impacts under alternative weight matrices; and tests panel thresholds using public finance and industrial structure, while strengthening identification and measurement checks.
Theoretical framework and hypothesis
We conceptualize DREI as a coordination-and-embedding capability. Conceptually, digital intensity captures the level of digital inputs, such as infrastructure, platforms, and digital activity, whereas DREI captures the extent to which these digital inputs are embedded into real-sector production and upgrading processes. Accordingly, high digital intensity can coexist with weak integration when complementary capabilities are lacking, leading to “digital islands” that do not translate into operational upgrading. The coupling–coordination index is, therefore, used to operationalize alignment and joint evolution between the digital and real sub-systems rather than the level of either sub-system in isolation. It captures how effectively digital infrastructure, data-driven services, and platform-based tools are integrated into real-economy production, logistics, and market processes rather than reflecting the sheer scale of ICT investment or the general level of economic development. This distinction matters because recent international research on digital transformation highlights strong complementarities, where digital technologies generate larger and more persistent returns only when coupled with skills, organizational capital, and governance capacity, implying heterogeneous outcomes across regions with different absorptive capacities [1–3]. In this regard, DREI approximates the conversion efficiency through which digital inputs translate into real-sector upgrading, making it closer to a mechanism-based construct than a simple development proxy.
Building on this literature and the policy logic embedded in the Little Giant program, we argue that DREI influences LGRs through five interrelated mechanisms. To begin with, stronger integration can directly facilitate upgrading by reducing coordination costs and information frictions along production networks, strengthening traceability and standards compliance, and improving quality management and delivery reliability. These capability improvements align closely with the evaluation dimensions used in LGRs, making DREI a plausible determinant of recognition outcomes even after accounting for background development differences.
In addition, DREI may operate through a finance (credit-depth) mechanism. Integration can strengthen local financial intermediation by improving the information environment and risk-assessment capacity, thereby expanding credit availability for SMEs investing in certification, equipment upgrading, and compliance. Because this channel can respond relatively quickly within policy cycles, it is likely to be particularly salient over the short- to medium-term horizon [19, 21]. A further channel concerns innovation inputs. By lowering experimentation and coordination costs, enabling data reuse, and reinforcing platform complementarities, integration can stimulate innovation investment and capability accumulation. However, R&D budgets and knowledge accumulation typically adjust more slowly and are often spatially concentrated. As a result, the innovation-input channel may exhibit lagged and heterogeneous effects, especially in short panels [22, 23]. Beyond within-region effects, DREI may also shape outcomes through a spatial-network mechanism. Integration is not confined within administrative boundaries: supply-chain linkages, shared platforms, mobile talent, and cross-regional market access can transmit integration benefits to neighboring areas, generating spillovers in recognition outcomes. Where digital backbones and industrial networks are interconnected, the relevant counterfactual is, therefore, not “within-region only” but is a system in which local outcomes depend partly on neighboring integration [24, 26]. Finally, we expect these mechanisms to depend on the regime. Complementary public inputs and structural conditions shape whether digital tools can be effectively embedded into real-sector upgrading. Fiscal capacity (FC) affects the provision of public services, certification support, and complementary infrastructure, while industrial structure determines whether integration translates into certifiable upgrading rather than isolated “digital islands.” These considerations motivate testing for nonlinear or threshold effects using public finance and industrial upgrading [28–30].
Guided by the literature reviewed in Section 2 and the theoretical framework as shown in Figure 1 in Section 3, we test the following hypotheses.
H1. DREI is positively associated with the LGR.
H2a. The effect of DREI on recognition is mediated by credit depth (CD).
H2b. The effect of DREI on recognition is mediated by R&D intensity (RD).
H3. There is a positive spillover from neighboring DREI to local recognition.
H4. The effect of DREI on recognition is nonlinear and varies across regimes of FC.
Theoretical framework.
Flowchart illustrating the integration of the digital economy and real economy, showing pathways from each to digital and real economy integration. Arrows lead to financial deepening, innovation input, little giant recognition, spatial spillover, and threshold fiscal capacity, with hypotheses labeled H1 to H4.MethodologyVariables and measurementDependent variable
The empirical setting is a balanced panel of 276 Chinese cities from 2019 to 2023. LGRs are compiled from official batch lists released by the Ministry of Industry and Information Technology and related agencies. It is a certification-type policy outcome that reflects policy-verified upgrading and innovation readiness, but recognition counts may also partly capture local implementation capacity in screening and supporting applications [32, 33]. LGRs are measured as the annual number of nationally recognized, specialized, sophisticated, distinctive, and innovative SMEs in each region. The baseline model uses raw counts and considers log (1 + y).
Recognitions are highly uneven across space, forming clear coastal and metropolitan clusters, as shown in Figure 2. The darkest cities are concentrated in the Beijing–Tianjin–Hebei core, the Yangtze River Delta, and the Guangdong–Hong Kong–Macao Greater Bay Area. Secondary hot spots appear around Chengdu–Chongqing, Wuhan, Zhengzhou, the Shandong Peninsula, and parts of Fujian. Most provincial capitals and sub-provincial cities occupy higher bins than the surrounding prefectures. In contrast, much of the west and parts of the northeast fall into the lightest bins, indicating few or no recognitions. White areas denote no data or zero counts. Overall, the pattern indicates strong alignment with developed urban belts and innovation hubs, with notable interior heterogeneity.
City-level Little Giant recognitions description in 2023.
Choropleth map of China displays regions shaded in five gradients of green, representing values ranging from zero to just over four thousand. Darker colors indicate higher values concentrated in central, eastern, and northeastern areas. Compass rose and a scale bar are present. Inset map highlights southeastern islands.Independent variable
DREI is measured as a synergy index. Two subindices, the digital economy and the real economy, are built with the entropy method from a hierarchical indicator system. To justify indicator selection, we operationalize DREI as the coordination between a digital subsystem and a real-economy subsystem using city-level statistics with consistent definitions and annual coverage. The digital subsystem captures infrastructure, digital industry activity, and the enabling environment, while the real-economy subsystem captures demand, openness, and industrial upgrading. Some dimensions, such as platform adoption, data governance, and firm-level digital transformation, are not included due to limited comparable city–year data. To reduce concerns regarding overlap across indicators, we follow the CCD structure and triangulate the results using alternative operationalizations and weighting schemes, including equal weights and principal component analysis (PCA)-based reconstruction. Indicator definitions and weights are reported in Table 1 [34, 35]. All indicators are min–max normalized by year, entropy weights are computed, and the two subindices are synthesized via the coupling–coordination degree to obtain the DREI regressor used in the models [15, 35]. The coupling–coordination measure increases only when both the digital subsystem and the real-economy subsystem improve in a coordinated manner; regions with high digital development but lagging real-sector upgrading obtain a lower coordination score by construction. This property is consistent with our conceptualization of DREI as reflecting embedding and conversion efficiency rather than digital intensity.
Indicator system for DREI and entropy weights.
First-level
Second-level
Third-level
Unit
Direction
Digital economy
Digital infrastructure
Long-distance fiber-optic cable length
km
+
Broadband access ports
10,000 ports
+
Number of domain names
10,000
+
Number of web pages
10,000 pages
+
IPv4 addresses
10,000
+
Digital industry development
Telecom business volume
100 million yuan
+
Postal business volume
100 million yuan
+
Express deliveries
10,000 pieces
+
Digital environment
Mobile phone penetration
Per 100 persons
+
R&D expenditure
100 million yuan
+
Number of patents
Piece
+
Real economy
Consumption demand
Total retail sales of consumer goods
100 million yuan
+
External openness
Total imports and exports
10,000 USD
+
Industrial upgrading
Tertiary/secondary value added (IU)
%
+
Indicators are min–max normalized by year. Weights are derived via the entropy method. “Direction” denotes polarity; all indicators here are positive.
Let Dit and Rit denote the entropy-weighted digital and real subindices, respectively. Following the coordination literature, the coupling degree is calculated as Equation 1:Cit=2DitRitDit+Rit.
The coordination index is Tit = α Dit + β Rit, with α = β = 0.5, and the coupling–coordination degree is CCDit=CitTit.
To clarify that DREI is not a proxy for digital intensity alone, the study provides a simple discriminant check. A quadrant comparison based on high versus low digital and real-economy subindices indicates that regions with high digital development but weak real-economy upgrading exhibit substantially lower DREI (mean 0.1696) than regions that are high on both dimensions (mean 0.3273). This pattern directly illustrates that “high digital intensity” can coexist with low integration and that DREI captures coordinated embedding and conversion rather than the level of digital development in isolation, as shown in Table 2.
DREI by digital–real economy quadrants.
Quadrant (by DE and RE)
N
Mean of DREI (D)
Median of DREI (D)
High DE and low RE
55
0.1696
0.175
High DE and high RE
223
0.3273
0.3073
Low DE and high RE
16
0.2072
0.2085
Low DE and low RE
107
0.1522
0.154
Total (quadrant sample)
401
0.2541
0.2511
Figure 3 maps the CCD-based DREI in 2023. High values concentrate along the eastern coastal corridor that runs from Beijing–Tianjin–Hebei through Shandong, Jiangsu, Shanghai, and Zhejiang to Fujian and Guangdong, with Liaoning also in the upper tier. The Chengdu–Chongqing area forms a clear inland highland. A broad middle tier covers parts of the middle Yangtze and Huai basins, including Hubei, Hunan, Anhui, Jiangxi, and Henan. Low values form a largely contiguous belt in the northwest that includes Xinjiang, Tibet, Qinghai, Gansu, Ningxia, and Inner Mongolia, with parts of the southwest, such as Guizhou and Yunnan, closer to the lower middle class. The overall picture exhibits an east–west gradient, with coastal cores and interior satellites.
DREI in 2023.
Choropleth map of China showing administrative regions shaded in gradients of green to represent values grouped into five ranges from 0.116723 to 0.590137, with a legend, scale bar, inset map, and compass rose.
Digital–real integration shows a clear coastal–inland gradient. The highest values cluster along the Yangtze River Delta, the Guangdong–Hong Kong–Macao Greater Bay Area, and the Beijing–Tianjin–Hebei corridor, with additional interior pockets around Chengdu–Chongqing, Wuhan, Zhengzhou, and the Shandong peninsula. Most provincial capitals and sub-provincial cities lie in the upper quantiles relative to their surrounding prefectures. Large parts of the northwest, the Qinghai–Tibet area, and some northeast prefectures lie in the lower bins. Overall, the pattern indicates pronounced spatial clustering consistent with developed urban belts and node-type hubs.
To control background heterogeneity and set up the mechanism analysis, the baseline specification includes a compact set of controls and two mediators. The controls include economic development, urbanization, and population density, which were compiled from provincial statistical yearbooks and the NBS portal for 2019–2023. The mediators capture two transmission channels: credit depth, reflecting financial intermediation that can alleviate SME financing frictions, and R&D intensity, reflecting innovation investment and capability accumulation, as commonly adopted in finance–growth–innovation studies. Full definitions, units, and transformations are reported in Table 3.
Variable definitions and descriptions.
Variable
Abbreviation
Role
Measurement method
Source
Little Giant recognition
LGR
Dependent
ln (1 + annual count)
[32]
Digital–real integration
DREI
Independent
Coupling–coordination (CCD) index
[15, 34–36]
Economic development
ED
Control 1
ln (GDP per capita)
[36–38]
Urbanization
Urban
Control 2
Urban population share (%)
Population density
PD
Control 3
ln (population/land area)
Credit depth
CD
Mediator 1
End-year loan balance/GDP (%)
[19, 21, 36]
R&D intensity
RD
Mediator 2
Intramural R&D expenditure/GDP (%)
Descriptive statistics
Table 4 reports the summary statistics. LGR shows ample between- and within-city variation suitable for FE, while DREI varies mainly across cities with modest within-city changes, and the controls behave as expected.
Descriptive statistics.
Variable
Mean (overall)
SD (overall)
Min
Max
SD (between)
SD (within)
LGR
3.3216
1.9512
0
8.3052
1.3678
1.3935
DREI
0.2111
0.0675
0.1031
0.5901
0.0662
0.0139
ED
11.0895
0.4619
9.7063
12.4819
0.4499
0.1071
Urban
0.4873
0.1719
0.1611
1.0472
0.1714
0.0155
PD
0.0515
0.0737
0.0005
0.8908
0.0738
0.0012
CD
1.3732
0.7500
0.4053
12.8171
0.6824
0.3134
RD
0.0034
0.0031
0.0002
0.0216
0.0030
0.0009
Analysis resultBaseline specification
We estimate a two-way fixed-effects model (Equation 2) to identify the association between digital–real integration and LGRs:LGRit=βDREIit+γControlit+ui+τt+εit,where i represents cities and t represents years. ui and τt are fixed effects of cities and year, respectively; Control contains controls (ED, urban, and PD). As a compact presentation of the stepwise build-up, columns (1)–(4) add controls sequentially.
Table 5 reports the estimates. Across columns (1)–(4), the coefficient on DREI is positive and statistically significant at conventional levels. The magnitude declines moderately as controls are added, yet it remains economically meaningful. Using column (4) as the preferred specification, the point estimate 5.503 indicates that stronger digital–real integration is associated with more LGRs; for example, a 0.10 increase in DREI is associated with an approximately 0.55 increase in ln (1+LGR), which implies an approximately 73% increase. Controls behave as expected. ED is negative and significant in columns (2)–(4), which is consistent with convergence effects or tighter recognition thresholds in richer cities after accounting for fixed effects and other covariates. Urban share is positive but statistically indistinguishable from zero in columns (3)–(4), implying that a higher urban share is not mechanically linked to recognitions once common shocks and provincial heterogeneity are absorbed. PD is also statistically indistinguishable from zero in column (4), which aligns with its limited within-city time variation over the sample. Model fit is high and remains essentially unchanged as controls are added, within R2 ranging from 0.881 to 0.882. City and year effects are jointly important, supporting the use of a two-way fixed-effects design.
Baseline regressions result.
Variables
Dependent variable: LGR
(1)
(2)
(3)
(4)
DREI
6.226***(2.289)
5.651**(2.327)
5.485**(2.312)
5.503**(2.311)
ED
−0.926**(0.446)
−0.915**(0.449)
−0.908**(0.449)
urban
2.314 (1.738)
2.223 (1.740)
PD
4.128 (12.259)
Constant
2.005***(0.483)
12.390**(5.109)
11.228 (5.239)**
10.928**(5.316)
Observations
1,380
1,380
1,380
1,380
R-squared
0.881
0.882
0.882
0.882
Cities FE
Yes
Yes
Yes
Yes
Year FE
Yes
Yes
Yes
Yes
Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, and *p < 0.1.
Robustness test
Table 6 presents a series of robustness exercises. First, trimming the distribution of regressors via Winsorization (col. 1) leaves the estimated effect of DREI positive and statistically significant, indicating that outliers do not drive the baseline result. Second, replacing contemporaneous DREI with its one-period lag yields a similar and significant coefficient, which mitigates concerns about contemporaneous reverse causality. Third, lagging the control variables by one period produces a slightly larger estimate, suggesting that the timing of covariates does not overturn the main finding. To address concerns that recognition outcomes may partly reflect local implementation capacity, we replace urbanization with public fiscal expenditure as a proxy for local public inputs and policy support intensity, and the estimated association between DREI and LGRs remains positive and statistically significant, indicating that the baseline finding is not driven by this choice of control.
Robustness test.
Variable
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Winsorized
DREI lagged by one period
Controls lagged by one period
Alternative controls (urban replaced by FC)
Placebo 1: Lead-1 of DREI
Placebo 2: Lead-2 of DREI
Placebo 3: Time-shift DREI
DREI
5.571**(2.36)
6.780*** (2.26)
4.669** (2.39)
−1.483 (1.366)
L.DREI
5.108*** (1.94)
F1.DREI
7.071**(2.54)
F2.DREI
5.87 (4.56)
Controls
YES
YES
YES
YES
YES
YES
YES
Cities FE
YES
YES
YES
YES
YES
YES
YES
Year FE
YES
YES
YES
YES
YES
YES
YES
Robust standard errors in parentheses. ***p < 0.01; **p < 0.05. Standard errors in columns (1)–(3) are clustered at the city level. Placebo columns (5)–(7) utilize two-way clustering by city and year to account for common annual shocks.
We further conduct falsification tests using placebo designs. When future leads of DREI are included as regressors, the 1-year lead is statistically significant, while the 2-year lead is not. This pattern is consistent with the strong persistence of index-type measures and potentially with short-horizon anticipation, and it cautions against relying on lead placebos alone as a definitive validity check. We, therefore, complement the lead tests with a time-shift placebo that breaks the true within-city time alignment of DREI while preserving the panel structure. Under this design, the estimated effect does not replicate the baseline positive association and becomes small and negative. In addition, the main finding remains stable when a key control is replaced with an alternative proxy. Taken together, these results show that the positive relationship between DREI and LGRs is robust to alternative control choices and timing perturbations and that the placebo evidence, especially the time-shift design, reduces concerns that the baseline two-way fixed-effects estimates are driven by generic trends or index construction artifacts.
To strengthen the construct validity of DREI, we re-estimate the baseline specification using four alternative operationalizations that capture integration from complementary perspectives (Table 7). First, we construct a purged DREI measure by residualizing the baseline CCD based index on key development fundamentals, such as urbanization, population density, and industrial upgrading, while accounting for city and year fixed effects. This procedure removes the component of DREI that is mechanically associated with broad development conditions. Second, we use a complementarity proxy defined as the product of standardized digital economy and real economy indices, which captures the joint embedding of digital capabilities with real economy foundations without imposing a specific coupling coordination functional form. Third, we reconstruct the digital economy and real economy indices using principal component analysis and then recompute the coupling coordination degree-based DREI, providing a data-driven alternative to entropy-based aggregation. Fourth, we implement an equal-weight scheme that assigns uniform weights to indicators within each subsystem (after normalization) and recomputes the CCD-based DREI.
Robustness to alternative operationalizations of DREI.
Variable
(1)
(2)
(3)
(4)
Purged DREI (residualized)
Complementarity proxy (D × R interaction)
DREI-PCA
DREI-EQ
DREI_purged
3.7128** (1.822)
DREI_int (D*R)
0.0621** (0.027)
DREI-PCA
1.980** (1.010)
DREI-EQ
4.577***(1.554)
Controls
YES
YES
YES
YES
Cities FE
YES
YES
YES
YES
Year FE
YES
YES
YES
YES
Robust standard errors in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. Column (1) uses a purged measure obtained by residualizing the baseline integration index with respect to development fundamentals (urbanization, population density, and industrial upgrading) while absorbing city and year fixed effects. Column (2) uses the product of standardized DE and RE (zDE × zRE) as a complementarity proxy. Column (3) reconstructs DE and RE using PCA and recalculates the coupling coordination degree (CCD_pca) while absorbing city and year fixed effects. Column (4) constructs DE and RE using equal weights (simple averages of normalized indicators within each subsystem) and recalculates the CCD-based DREI.
The results are consistent across these measurement strategies. The estimated effect remains positive and statistically distinguishable from zero when using the purged measure and the complementarity proxy, and it remains positive and marginally significant under the principal component analysis-based DREI. Taken together, these findings indicate that the main association is not driven by a particular weighting scheme or index definition, and they reinforce the interpretation of DREI as an embedding and coordination capability rather than a generic proxy for overall development (Table 7).
Endogeneity tests
To address potential endogeneity of DREI, an instrumental-variable two-stage least squares (IV–2SLS) strategy is implemented, using the national Broadband China pilot as an exogenous policy source. The excluded instrument is constructed to indicate whether a city was selected into the Broadband China pilot, and MU is a city–year measure of local broadband supply intensity. This interaction captures time-varying, policy-linked heterogeneity in effective broadband expansion, which shifts DREI after conditioning on standard covariates and city and year fixed effects. The IV–2SLS results are consistent with a positive effect of DREI on LGRs: under the BBC_MU instrument, the second-stage coefficient on instrumented DREI is 20.9737, while the first stage shows a significant relationship between BBC_MU and DREI. Identification and instrument strength diagnostics support the relevance (Kleibergen–Paap rk LM = 13.086, p = 0.0003; Cragg–Donald F = 36.375; Kleibergen–Paap rk Wald F = 11.730). As a robustness check, using lagged DREI as an alternative excluded instrument yields a similarly positive second-stage estimate with strong first-stage performance and stronger weak-ID diagnostics (Kleibergen–Paap rk LM = 43.183, p < 0.001; Cragg–Donald F = 274.303; Kleibergen–Paap rk Wald F = 60.153). Because each specification is exactly identified, over-identification tests (Hansen J) are not informative; overall, the IV evidence supports the main conclusion that higher DREI is associated with higher LGRs after accounting for fixed effects and controls, thus mitigating concerns about reverse causality and omitted-variable bias (Table 8).
Two-stage least squares estimates.
Variable
First-stage DREI
Second-stage LGR
First-stage DREI
Second-stage LGR
BBC_MU
−0.0017** (0.0007)
DREI.L
0.4459*** (0.0575)
DREI
20.9737** (9.8781)
11.3156 (3.7961)
Constant
0.8304*** (0.1201)
−4.8453 (10.6001)
0.6548*** (0.0846)
5.0644 (6.3352)
Controls
Yes
Yes
Yes
Yes
Year FE
Yes
Yes
Yes
Yes
City FE
Yes
Yes
Yes
Yes
N
1,380
1,380
1,104
1,104
R2
0.9774
0.8995
0.9887
0.9429
Kleibergen–Paap rk LM
13.086 [p = 0.0003]
43.183 [p = 0.0000]
Cragg–Donald Wald F
36.375
274.303
Kleibergen–Paap rk Wald F
11.730
60.153
Hansen J
0.000 (exactly identified)
0.000 (exactly identified)
Robust standard errors in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.
Regional heterogeneity
The estimates reveal pronounced variation across regions (Table 9). The marginal effect of DREI on LGR is strong and positive in the east and remains positive, though smaller, in the central region. In the west, the effect is statistically indistinguishable from zero. Pairwise tests show that the west differs markedly from both the east and central regions, while the central–east difference is not statistically meaningful. Taken together, digital–real integration delivers clear gains where market access and infrastructure are more developed, whereas limited absorptive capacity in the west dampens the impact.
Heterogeneity test.
Regions
Coef
p-value
East
11.245
0.013**
Central
7.383
0.019**
West
−7.102
0.171
Differences in slopes
F-stat
p-value
Central–east
0.830
0.362
West–east
10.610
0.001***
West–central
7.770
0.006***
*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
Threshold specification
To examine nonlinear responses with respect to DREI, a threshold two-way fixed-effects model is estimated, and two theoretically grounded state variables are utilized to capture potential nonlinearity in the effect of DREI on LGRs. FC, which is proxied by the ratio of public finance expenditure to GDP, reflects a city’s ability to provide complementary public inputs. The specification is given in Equation 3.LGRit=a+β1DREIit*Iqit≤θ1+β2DREIit*Iθ1<qit≤θ2+⋯+βnDREIit*Iθn−1<qit≤θn+δ′Controlit+ui+τt+εit,where i represents cities and t represents years. ui and τt denote the city and year fixed effects, respectively. Controlit includes ED, urban, and PD. Predetermined threshold variables include the public finance-to-GDP ratio (FC). The threshold θ is obtained by grid search, and critical values are computed from bootstrap replications. The LR profile for the FC threshold is shown in Figure 4.
Single-threshold effect of FC.
Line graph showing LR statistic on the y-axis and First Threshold on the x-axis, with values ranging from 0.1 to 0.5. A red dashed horizontal line marks a reference level around 7.
It is evident that across regimes defined by FC, DREI exerts a positive and significant impact on LGRs, yet the high-threshold regime exhibits a smaller coefficient than the low-threshold regime. Specifically, when FC ≤ 0.174, the marginal effect of DREI is stronger; once these cutoffs are crossed, the marginal effect declines markedly, confirming a pattern of “positive but diminishing across thresholds.” The mechanism is twofold. First, in fiscally constrained cities, incremental digital investment and public services yield higher marginal returns, translating more efficiently into recognition; in fiscally stronger cities, saturation, project homogeneity, and diminishing returns attenuate the payoff (Table 10).
Threshold test.
Threshold variable
FC
θ1
0.174
DREIit*I (qit≤θ1)
24.127***(2.387)
DREIit*I (qit>θ1)
21.439***(2.390)
Control variables
Yes
Observations
1380
R-squared
0.578
Robust standard errors in parentheses. ***p < 0.01, **p < 0.05, and *p < 0.1.
Mechanism and spatial spillover effects analysisMechanism analysis
This section examines whether DREI affects LGRs through two theoretically grounded channels, credit deepening (CD) and innovation input measured by R&D intensity (RD). The mediator equations are specified as below (Equations 4, 5) [39]:CDit=a0+a1DREIit+a2′Controlit+ui+τt+εit,RDit=β0+β1DREIit+β2′Controlit+ui+τt+εit,where credit depth (CDit) and R&D intensity (RDit) are the mediators; DREIit denotes digital–real economy integration; Controlit is the control vector; ui and τt are the city and year fixed effects, respectively; εit represents the disturbance terms.
Following Jiang [39], we do not implement a full mediation decomposition that conditions the outcome equation on post-treatment mediators because doing so may introduce bad control bias in observational settings. Instead, we treat channel analysis as mechanism-consistent evidence and examine whether DREI systematically shifts plausible determinants of LGR by estimating reduced-form mediator regressions. We acknowledge that, even with rich controls and two-way fixed effects, the mediator regressions may still be exposed to unobserved time-varying confounders at the city level, such as policy intensity or administrative capacity; therefore, the channel evidence should be interpreted cautiously. In Table 11, DREI is strongly and positively associated with credit deepening. This relationship is robust and remains significant when credit depth is lagged by one period, suggesting that integration is accompanied by persistent improvements in local financial intermediation. Economically, this pattern is consistent with lower financing frictions through better information infrastructures, improved risk assessment enabled by data industry linkage, and wider access to credit products for specialized SMEs, which can translate more quickly into certification-type outcomes. Because cities are interdependent, part of the channel may also operate through cross-city diffusion; we address such spatial interdependence in the SDM analysis, and we do not interpret the reduced-form mediator regressions as capturing all network transmission.
Mechanism analysis.
Variable
Credit depth (CD)
Credit depth lag one period (L.CD)
R&D intensity (RD)
R&D intensity (L.RD) lag one period
DREI
2.569***(3.39)
2.597***(3.63)
0.006* (1.67)
0.0003 (0.07)
ED
−0.245 (−0.64)
−0.419***(−3.98)
−0.001 (−1.46)
−0.001 (-1.34)
Urban
0.938 (0.97)
0.359 (0.79)
0.001 (0.24)
−0.002 (−0.60)
PD
9.647* (1.66)
12.541**(2.51)
0.004 (0.11)
0.024 (0.69)
Constant
2.610 (0.57)
4.642***(3.74)
0.015 (1.53)
0.131 (1.58)
Observations
1,380
1,104
1,380
1,104
R-squared
0.843
0.980
0.920
0.935
Cities FE
YES
YES
YES
YES
Time FE
YES
YES
YES
YES
Robust t-statistics in parentheses. ***p < 0.01, **p < 0.05, and *p < 0.1.
City-level measurement and horizon: R&D intensity measured at the city level is an aggregate proxy and may attenuate firm-level innovation responses that are most relevant for Little Giant upgrading. Innovation activity is typically concentrated in a subset of firms, so aggregation can dilute variation and introduce measurement error. Moreover, innovation inputs often respond with lags due to multi-year budgeting and project cycles, which may not be fully captured over the 2019–2023 window. We, therefore, interpret the weak short-run R&D mediation as consistent with delayed and heterogeneous innovation dynamics rather than as evidence of an absent innovation channel.
Spatial spillover effects analysis
To motivate the spatial specification and verify that cities are not independent observations, spatial autocorrelation is tested using a weighting scheme kept identical across diagnostics and regressions.
Let dij denote binary adjacency between cities,αij=1,if i≠j0,ifi=j,and the row-standardized weights areWij=αij∑i≠jαij,∑jwij=1,wii=0.
Using a binary adjacency matrix W, global Moran’s I is computed annually for the dependent variable (LGRs) and CCD-based DREI. Results show positive and statistically significant spatial autocorrelation in every year for both series, as shown in Table 12.
Global Moran’s I.
Year
LGR
DREI
Moran’s I
Z
p-value
Moran’s I
Z
p-value
2019
0.1083
2.6650
0.0077
0.2649
6.4855
0.0000
2020
0.2527
6.1135
0.0000
0.2968
7.2575
0.0000
2021
0.3357
8.1014
0.0000
0.3312
8.0897
0.0000
2022
0.4359
10.4881
0.0000
0.3558
8.6856
0.0000
2023
0.4750
11.4176
0.0000
0.3525
8.6029
0.0000
Across all years, both variables display positive and statistically significant global spatial autocorrelation (p = 0.0000 in all cases, except for LGR in 2019 with p = 0.0077). Moran’s I for LGR increases from 0.1083 in 2019 to 0.4750 in 2023, indicating a steady strengthening of regional clustering in recognitions. Moran’s I for DREI remains persistently positive and increases from 0.2649 to 0.3525 over the sample period; the associated z-scores are consistently large, which is consistent with stable and slightly intensifying spatial agglomeration in digital–real integration.
At the beginning of the period, DREI is the more clustered process, but by 2023, LGR overtakes DREI. This reversal indicates that recognitions have become increasingly concentrated among contiguous cities, plausibly reflecting network effects in supply chains, certification resources, and access to financing that amplify local advantages over time. The persistence and upward trend in global Moran’s I justifies the use of spatial econometric specifications in subsequent analysis.
Following standard practice in spatial econometrics [24, 25], model choice begins with Lagrange multiplier (LM) diagnostics for spatial dependence. LM-lag and LM-error, together with their robust versions, are applied to the residuals from a baseline OLS panel. If LM-lag is significant and LM-error is not, the spatial lag model (SAR) is adopted. If the LM-error is significant and the LM-lag is not, the spatial error model (SEM) is adopted. When both statistics are significant, the SDM is estimated as the encompassing specification, and the likelihood-ratio or Wald tests are used to determine whether SDM can be reduced to SAR (θ = 0) or to SEM (common-factor restriction).
The LM diagnostics strongly reject the null of no spatial dependence. Global Moran’s I on the residuals is large (155.274, p = 0.000). Both the error and lag LM tests are highly significant, as are their robust counterparts, all with p = 0.000. These results indicate that a non-spatial specification is inadequate; spatial error correlation is quantitatively dominant, while a non-negligible spatial lag component is also present [24, 25] (Table 13).
LM test.
Test
Statistic
df
p-value
Spatial error: Moran’s I
155.274
1
0.000
Spatial error: Lagrange multiplier
2.1e+04
1
0.000
Spatial error: Robust LM
1.2e+04
1
0.000
Spatial lag: Lagrange multiplier
9725.814
1
0.000
Spatial lag: Robust LM
357.988
1
0.000
Given significance on both dimensions, estimation proceeds with the SDM with two-way fixed effects and the row-standardized contiguity matrix W. The SDM nests common alternatives and avoids a priori misspecification: if θ = 0, it collapses to SAR; if ρ = 0, it reduces to SLX; and under the common-factor restriction, it is observationally equivalent to SEM. Post-estimation joint tests on Wx terms are reported to assess possible reductions; all main results are presented using impact decomposition, which is the appropriate basis for inference under SDM.
The empirical specification is a SDM with city and year fixed effects as below (Equation 6).LGRit=ρ∑j=1NωijLGRjt+βDREIit+θ∑j=1NωijDREIit+γ‘Controlit+ξ′∑j=1NωijControljt+αi+τt+uit.
LGRit: LGRs in city i and year t, defined as ln (1+recognitions) to accommodate zeros.
DREIit: CCD-based digital–real integration index, which is the key regressor.
Controlit′: Economic development (ln GDP per capita), urbanization rate, and population density (ln population density).
αi: City fixed effects capture time-invariant heterogeneity across cities.
τt: Year fixed effects capture shocks common to all cities in year t.
ρ: Spatial autoregressive coefficient on the dependent variable.
θ: Spatial spillover coefficient for DREI.
ξ: Spatial spillover coefficients for the controls.
Table 14 indicates strong spatial dependence, with spillovers as the dominant transmission channel. The spatial parameter is large and precisely estimated across alternative spatial weight matrices, suggesting that LGRs are strongly interdependent across connected cities. With the city and year fixed effects absorbed, the local DREI coefficient is identified from within-city changes over time, net of time-invariant city characteristics and common shocks. Both the local DREI regressor and its spatial lag are positive and highly significant in the contiguity and economic–geographic specifications, implying that DREI is associated with recognitions within a city and, more strongly, through inter-city linkages.
SDM results under alternative spatial weight matrices.
Item
Contiguity matrix (adjacency W)
Geographic matrix (distance W)
Economic-geographic W
Main: DREI
4.646*** (1.763)
3.125** (1.611)
3.718** (1.640)
Wx
20.855*** (2.540)
5.876* (3.444)
22.290*** (2.952)
Spatial parameter
0.719*** (0.018)
0.911*** (0.022)
0.755*** (0.021)
Error variance
0.407*** (0.016)
0.338*** (0.013)
0.384*** (0.015)
Direct effect
11.271*** (1.804)
3.501** (1.624)
7.332*** (1.666)
Indirect effect
77.200*** (4.724)
97.271*** (19.871)
99.090*** (6.452)
Total effect
88.471*** (5.383)
100.772*** (20.158)
106.422*** (6.889)
Standard errors in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Reported direct, indirect, and total effects are average impacts from SDM impact decomposition.
The impact decomposition further quantifies this pattern. Under the contiguity matrix, the estimated indirect effect is 77.200 and the direct effect is 11.271 (both p < 0.01), yielding a total effect of 88.471. Spillovers, therefore, account for approximately 87% of the total effect, while the within-city component contributes approximately 13%, reinforcing that spatial transmission is central rather than incidental in this setting. This conclusion is robust to alternative connectivity structures. Under the distance matrix, the indirect effect is 97.271 compared with a direct effect of 3.501, and under the economic–geographic matrix, the indirect effect is 99.090 compared with a direct effect of 7.332, indicating that spillovers remain the major component of the total effect across weight matrices. The variance estimate is stable, which is consistent with a well-behaved specification.
In substantive terms, higher DREI levels are associated with more LGRs primarily via intercity channels, while within-city responses are comparatively modest. From a policy perspective, the dominance of indirect effects implies that the main payoff from raising local DREI is realized through its impact on connected cities. Local levers remain important, but they yield the highest returns when designed for diffusion and complemented by coordination. Practical examples include building interoperable data and platform governance, expanding cross-city credit and guarantee linkages, and providing shared certification and supplier-development services. Because these benefits are not fully internalized by any single locality, coordination arrangements that align standards and co-fund common platforms help transform spillovers into broader recognition gains.
Discussion
The analysis evaluated whether DREI increases policy-recognized Little Giant SMEs, LGRs, and clarified how such effects operate across space and through measurable channels. The results show that higher DREI is associated with a greater number of recognitions. The impact decomposition indicates that this relationship is driven largely by intercity spillovers. Mediation tests identify credit deepening as the primary conduit, with R&D intensity playing a weaker supporting role. Threshold estimates further indicate that returns to integration are regime-dependent.
The spillover-dominated impact profile is consistent with a networked transmission mechanism in which integration upgrades supply-chain visibility, data governance, and certification readiness that are relevant to the recognition criteria [12, 22]. The mediation evidence suggests that DREI first strengthens local financial intermediation by improving information infrastructures, risk analytics, and collateralization. This enhanced credit capacity is then translated into recognitions, which is consistent with the view that finance is a proximate lever linking coordination to firm upgrading [19, 21, 39]. The weaker R&D pathway is congruent with short-horizon adjustment dynamics and the spatial concentration of research activity, where multi-year budgeting and accumulation processes may not be fully captured at the city–year scale [23]. Importantly, spillovers substantially exceed direct effects, indicating that integration benefits propagate along adjacency networks through supplier development, shared certification practices, and interoperable data platforms [24, 26, 27]. The threshold results sharpen this interpretation. Below a coordination minimum, digital inputs tend to accumulate as fragmented “islands” with limited translation into recognition. Above the threshold, complementarities across platforms, standards, and financial analytics raise marginal returns, which is consistent with coordination theories that imply nonlinearity once a critical mass is reached.
The dominance of spillovers provides further evidence that the digital economy produces interregional externalities in innovation, green upgrading, and productivity [20, 22, 40]. More generally, these patterns align with the international view that digital transformation generates innovation gains primarily when digital tools are embedded into real-sector routines and inter-firm networks, and that such gains often diffuse unevenly across space, making spatial-dependence central rather than incidental. The finance-mediated pathway is also in line with research linking digital infrastructure to financial development and downstream firm outcomes [19, 21]. In contrast, the weaker short-run R&D response accords with findings that innovation effects materialize with lags and are concentrated in hubs [23]. Relative to studies of LGR ecosystems that emphasize business-environment configurations [10, 12, 15], the present study advances three fronts. First, it operationalizes coordination-based integration via a coupling–coordination construct and embeds it in a spatial Durbin framework that reports direct, indirect, and total impacts, responding to calls to move beyond coefficient-level interpretation in interdependent settings [24–26]. Second, it identifies mechanisms using panel-consistent mediation evidence that highlights a dominant credit-deepening channel and a secondary R&D channel over the sample horizon. Third, it documents threshold-governed marginal effects, extending panel-threshold methods to a spatial digital–industrial context and clarifying when integration yields the largest upgrading dividends.
Conclusion
This study shows that regional digital–real integration is associated with higher LGRs and that a substantial share of the effect propagates to neighboring cities. We contribute by measuring integration as coordination rather than scale, decomposing spatial effects, and documenting tier-specific heterogeneity, where the impacts shift from negative at low integration to positive at higher tiers. Policy should, therefore, prioritize credit market deepening, adopt a local implementation plus regional coordination approach to convert spillovers into certification gains, and tailor integration programs to local fiscal and structural conditions. Three limitations remain. The time-span is constrained by outcome availability, as comparable LGR data are only consistently available from 2019 onward, so the short window may understate longer-run innovation effects. In addition, because LGR is a certification-type policy outcome, recognition counts may partly reflect differences in local implementation capacity and screening intensity. Third, innovation inputs are measured at the city level, which may dilute firm-level heterogeneity and mask lagged R&D and patenting responses among potential Little Giant candidates. Future work can extend the time-span, link to firm-level microdata, and test spatial weights informed by supply-chain ties.
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material; further inquiries can be directed to the corresponding author.
Author contributions
YG: Formal Analysis, Methodology, Data curation, Writing – original draft, Writing – review and editing, Conceptualization, Investigation, Resources. OL: Supervision, Writing – review and editing, Visualization.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
ReferencesVerhoefPCBroekhuizenTBartYBhattacharyaAQi DongJ (2021). Digital transformation: a multidisciplinary reflection and research agenda. J Business Res, 122, 889–901. 10.1016/j.jbusres.2019.09.022AnghelBBunelSBijnensGBotelhoVFalckELabhardVDigitalisation and productivity: a report by the ESCB expert group on productivity, innovation and technological change. Occasional Paper Series (2024). Available online at: https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op339∼f67b6981a9.en.pdf (Accessed September 25, 2025).CorradoCCriscuoloCHaskelJHimbertAJona-LasinioC. New evidence on intangibles, diffusion and productivity [2021/10] OECD science, technology and industry working papers, issue (2021). Available online at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2021/07/new-evidence-on-intangibles-diffusion-and-productivity_f15ac91a/de0378f3-en.pdf (Accessed September 25, 2025).AntoniettiRBurlinaCRodríguez-PoseA. Digital technology and regional income inequality: are better institutions the solution?Pap Reg Sci (2025) 104(2):100079. 10.1016/j.pirs.2025.100079DuZ-YWangQ. Digital infrastructure and innovation: digital divide or digital dividend?J Innovation and Knowledge (2024) 9:100542. 10.1016/j.jik.2024.100542State Council of the People's Republic of China. Plan focuses on digital economy development during 14th five-year plan period (2022). Available online at: https://english.www.gov.cn/policies/latestreleases/202201/12/content_WS61de9a35c6d09c94e48a385f.html (Accessed September 25, 2025).State Council Information Office. Briefing on stepping up support for enterprises (2024). Available online at: https://english.scio.gov.cn/m/pressroom/2024-12/09/content_117595452_6.html (Accessed September 28, 2025).Ministry of Industry and Information Technology of the People's Republic of China. Industrial internet innovation and development action plan (2021–2023) (2020). Available online at: https://www.gov.cn/zhengce/zhengceku/2021-01/13/content_5579519.htm (Accessed September 28, 2025).National Development and Reform Commission. Implementation opinions on further promoting “Eastern Data Western Computing” project to accelerate construction of national integrated computing power network (2023). Available online at: https://www.gov.cn/zhengce/zhengceku/202401/content_6924596.htm (Accessed September 28, 2025).BrownA. Small firms join the fray of China's techno-industrial drive. CURRENT HISTORY (2024) 123:40–9. 10.1525/curh.2024.123.S1.40TangGWangLZhengTWuW. What types of business environment fosters the emergence of more specialized and sophisticated “little giant” enterprises? An empirical study based on the TOE framework and configuration adaptation theory. Managerial Decis Econ (2024) 45(3):1557–72. 10.1002/mde.4063ZhangZFengHWangLYangL. How to achieve high-quality development of SRDI enterprises—A study of the TOE framework-based configuration. Plos one (2024) 19(6):e0304688. 10.1371/journal.pone.030468838829914HienLTTamPT. Impacting information technology and telecommunications infrastructure on the digital economy. Emerging Sci J (2025) 9(2):1013–36. 10.28991/ESJ-2025-09-02-026LiuY. Coupling and coordination evaluation of digital economy and green development efficiency in eight urban agglomerations in China. Scientific Rep (2025) 15(1):26829. 10.1038/s41598-025-07967-740701977TangS (2024). Coupling coordination measurement and spatiotemporal evolution characteristics of digital economy, environmental regulation, and carbon emission intensity—Empirical evidence from Chinese provinces [Original Research]. Front Environ Econ3–2024. 10.3389/frevc.2024.1511080WangSWangALiuSZhangCQiaoLLiX. Research on the coupling coordination relationship between the digital economy and high-quality energy development: evidence from China. Heliyon (2024) 10(2):e24637. 10.1016/j.heliyon.2024.e2463738312570WangZLinSChenYLyulyovOPimonenkoT. The digital economy and real economy: the dynamic interaction effect and the coupling coordination degree. Sustainability (2024) 16(13):5769. 10.3390/su16135769SaparovaDSaginovaSBelgibayevaKBekbussinovaGKasenovaATleulinC. New development model of the agro-industrial complex under green economy conditions. Emerging Sci J (2025) 9(5):2619–42. 10.28991/ESJ-2025-09-05-019SongSWenJLiYLiL. How does digital economy affect green technological innovation in China? New evidence from the “Broadband China” policy. Econ Anal Policy (2024) 81:1093–112. 10.1016/j.eap.2024.01.008ZhangZMaoRZhouZZengZ. How does digital finance affect green innovation? city-Level evidence from China. Finance Res Lett (2023) 58:104424. 10.1016/j.frl.2023.104424ZhuXAsimakopoulosSKimJ. Financial development and innovation-led growth: is too much finance better?J Int Money Finance (2020) 100:102083. 10.1016/j.jimonfin.2019.102083SunGFangJLiJWangX. Research on the impact of the integration of digital economy and real economy on enterprise green innovation. Technol Forecast Social Change (2024) 200:123097. 10.1016/j.techfore.2023.123097SunHYangXTangXPengF. How innovation funding leads enterprises to engage in research and development: small and medium enterprises’ perspective. Plos One (2023) 18(7):e0289166. 10.1371/journal.pone.028916637490503AnselinL (2022). Chapter 6: spatial econometrics handbook of spatial analysis in the social sciences. 101–22. Cheltenham, United Kingdom: Edward Elgar Publishing. 10.4337/9781789903942.00014AnselinLBeraAKFloraxRYoonMJ. Simple diagnostic tests for spatial dependence. Reg Sci Urban Econ (1996) 26(1):77–104. 10.1016/0166-0462(95)02111-6LeeL-feiYuJ. (2010). Estimation of spatial autoregressive panel data models with fixed effects. J Econom, 154(2), 165–85. 10.1016/j.jeconom.2009.08.001MelnykLVasaLSinghSKubatkoOKalinichenkoL. Quasi-viral technologies as the drivers of the economy digital transformation towards sustainability. HighTech Innovation J (2025) 6(1):183–200. 10.28991/HIJ-2025-06-01-013HuYGaoYWangX. Chinese-style fiscal decentralization, government innovation investment, and regional innovation. Sustainability (2024) 16(24):11100. 10.3390/su162411100LuYHuJJ. Spatial impact of digital economy on the upgrading of industrial structure: evidence from Chinese cities. Sage Open (2024) 14(1):21582440241233940. 10.1177/21582440241233940MiaoKLiKSuL. Panel threshold models with interactive fixed effects. J Econom (2020) 219(1):137–70. 10.1016/j.jeconom.2020.05.018ElhorstJP. Spatial econometrics: from cross-sectional data to spatial panels, 479. Springer (2014). 10.1007/978-3-642-40340-8Ministry of Industry and Information Technology of the People's Republic of China. Announcement of the first batch of “Little Giant” enterprises (2019). Available online at: https://www.gov.cn/xinwen/2019-06/18/content_5401193.htm (Accessed October 15, 2025).Ministry of Industry and Information Technology of the People's Republic of China. Public list of the third batch of “Little Giant” enterprises (2021). Available online at: https://www.ncsti.gov.cn/kjdt/tzgg/202107/P020210720361507502455.pdf (Accessed October 15, 2025).BaiDLiMWangYMallekSShahzadU. (2024). Impact mechanisms and spatial and temporal evolution of digital economy and green innovation: a perspective based on regional collaboration within urban agglomerations. Technol Forecast Social Change, 207, 123613. 10.1016/j.techfore.2024.123613LiuBHuangYChenMLanZ. Towards sustainability: how does the digital-real integration affect regional green development efficiency?Econ Anal Policy (2024) 83:42–59. 10.1016/j.eap.2024.06.008National Bureau of Statistics of China. China statistical yearbook (2019). Available online at: https://www.stats.gov.cn/sj/ndsj/ (Accessed October 15, 2025).ChenYGuoFWangJCaiWWangCWangK. Provincial and gridded population projection for China under shared socioeconomic pathways from 2010 to 2100. Scientific Data (2020) 7(1):83. 10.1038/s41597-020-0421-y32152299ShangYLeeHMaJ. A study on the spatial effects of the digital economy on regional economic growth in China. Sustainability (2025) 17(5):2259. 10.3390/su17052259JiangT. Mediation and moderation effects in empirical research on causal inference. China Ind Econ (2022)(05) 100–20. 10.19581/j.cnki.ciejournal.2022.05.00535154431LiHLuZYinQ. The development of fintech and SME innovation: empirical evidence from China. Sustainability (2023) 15(3):2541. 10.3390/su15032541
Edited by:Grigorios L. Kyriakopoulos, National Technical University of Athens, Greece
Reviewed by:Omid Aminoroayaie Yamini, K. N. Toosi University of Technology, Iran
Shikun Qin, Southwestern University of Finance and Economics, China