Assessing the effect of digital platforms on innovation quality: mechanism identification and threshold characteristics

The paper addresses how digital platforms influence the quality of innovation in China amid rapid advances in big data, cloud computing, AI, and 5G. Despite China’s strong innovation activity and global leadership in international patent applications, innovation quality lags due to low application rates of scientific outputs, limited basic research investment, and scarcity of breakthrough innovations. The study asks whether digital platforms can serve as a key driver for improving innovation quality, through what mechanisms, and whether regional heterogeneity and resource mismatches shape these effects. It posits that digital platforms may both enhance innovation quality through open sharing, information matching, and spillovers, and hinder it via monopolistic behavior, vicious competition, and Matthew-effect dynamics. The paper formulates hypotheses that digital platforms directly affect innovation quality, that resource mismatch mediates this relationship, and that the effect is nonlinear with threshold characteristics governed by the degree of mismatch.

The review covers three strands. (1) Digital platforms (DP): definitions emphasize DP as socio-technical intermediaries and value co-creation nexuses linking multilateral markets; typologies include advertising, online commodity trading, and offline service trading platforms. Measurement work is nascent; a recognized proxy is the log interaction between broadband users and online retail sales. DP effects span technology management, industrial organization, and strategy, with noted risks of monopoly and capacity waste. (2) Digitization and innovation quality (INN): INN encompasses product/service, process, and managerial dimensions. Digitalization reshapes innovation processes and can raise INN via digital services, digital economy development (with spatial heterogeneity), and specific digital technologies (IT, internet, blockchain, big data). Some argue digital transformation favors incremental over quality gains, and studies dissect digital impacts across product, process, organizational, and business model innovation. (3) Resource mismatch (Misa): Misa arises from information imperfections, market failures, and intervention, misallocating factors and lowering productivity. Measurement approaches include TFP-based dispersion and specific R&D capital/personnel mismatches. Misa diminishes innovation output; studies quantify losses from structural, ownership, and sectoral mismatches. Gaps: limited empirical work specifically on DP–INN, contested INN measurement, and insufficient attention to pervasive Misa as a constraint. Contributions: integrate DP, INN, and Misa; employ patent citations to proxy INN; and model Misa as both mediator and threshold.

Design: Provincial panel study of 30 mainland Chinese regions (excluding Tibet, Taiwan, Hong Kong, Macao) from 2011–2021. The study estimates (i) a baseline fixed-effects model with dynamics (lagged INN, GMM for endogeneity), (ii) a mediation (intermediary) model with Misa as mediator, and (iii) a dynamic panel threshold regression using Misa as the threshold variable, estimated by GMM following Hou et al. (2018). Models: (1) Baseline dynamic panel: INN_it depends on INN_it-1, DP_it, controls X_it (gov, urb, com, tech), province fixed effects, and error; estimated via GMM to address endogeneity and pre-dependence. (2) Mediation: Misa_it = β0 + β1 DP_it + βX X_it + μ_i + ε_it; INN_it = ω0 + ω1 DP_it + ω2 Misa_it + ωX X_it + μ_i + ε_it. (3) Dynamic panel threshold: INN_it = φ INN_it-1 + γ1 DP_it I(Misa_it ≤ η) + γ2 DP_it I(Misa_it > η) + θX X_it + μ_i + ε_it, where η is the data-driven threshold identified by least-squares LR statistics. Variables: INN (explained) measured by the number of patent citations, aggregated at the province-year level using data mining of National Patent Office records for listed firms and subsidiaries. DP (core explanatory) proxied by ln(broadband internet users × online retail sales) at the provincial level, capturing online engagement and platform scale. Misa (mediator/threshold) measures capital and labor factor mismatches using Bai and Liu (2018) methods; higher values indicate greater deviation from efficient allocation. Controls: gov (government expenditure share of GDP excluding public health, education, R&D/technology, culture), urb (urbanization level: urban population share), com (openness: total imports and exports over GDP), tech (technology market turnover over GDP). Data source: National Bureau of Statistics of China; logarithmic transformations applied to mitigate heteroscedasticity and multicollinearity; linear interpolation used for limited missing values. Descriptives (N=330): INN mean 8.173 (sd 1.644), DP 4.812 (sd 2.205), Misa 0.437 (sd 0.724), gov 0.269, urb 1.415, com 5.546, tech 0.016. Inference and diagnostics: Hausman test favored fixed effects; threshold tests via bootstrap identified a single significant threshold; AR(1)/AR(2) and Hansen tests support GMM validity. Robustness checks altered dependent timing (lagging citations), dropped control variables, and trimmed regional samples with consistent results.

- Baseline effects: Nationally, DP significantly increases INN (coefficient 0.490, t=14.46). By region, effects are all positive and significant with magnitudes Eastern 0.552 (t=6.76), Central 0.559 (t=5.59), Western 0.432 (t=9.09). International competition (com) negatively relates to INN nationally (−0.325, significant). - Mediation via Misa: DP reduces Misa (−0.057, t=−2.25) and directly raises INN when controlling for Misa (0.603, t=23.76). Misa negatively impacts INN (−0.179, t=−3.25). The calculated indirect effect is 0.010 (= −0.057 × −0.179), accounting for 1.605% of DP’s total effect on INN. Sobel Z=1.848 (p=0.065) and bootstrap Z=2.030 (p=0.042) support mediation. - Threshold nonlinearity: A single Misa threshold is significant (F=16.27, p=0.020); double-threshold is not. Estimated threshold η=0.399 with 95% CI [0.393, 0.401]. When Misa ≤ 0.399, DP’s effect on INN is strong (0.222, p<0.001); when Misa > 0.399, the effect attenuates (0.030, p=0.029). Thus, high mismatch markedly constrains the innovation-quality gains from DP. - GMM diagnostics: AR(1) p=0.000, AR(2) p=0.465; Hansen test supports instrument validity. Robustness tests (lagged citations, altered controls, sample trimming) preserve sign and significance of core coefficients.

The results confirm that digital platforms are a significant driver of regional innovation quality in China, directly enhancing the efficiency and impact of innovation through open sharing, improved information matching, and knowledge/technology spillovers. However, the benefits are uneven across regions, aligning with disparities in endowments and infrastructure (Eastern > Central > Western). Critically, resource mismatch weakens the DP–INN linkage both as a mediator—where DP helps alleviate mismatch and thereby improves innovation quality—and as a nonlinear moderator with a clear threshold: beyond a mismatch level of approximately 0.399, the marginal effectiveness of platforms in raising innovation quality drops sharply. This demonstrates that optimizing factor allocation (capital and labor) and reducing price distortions and lock-in effects are prerequisites for fully realizing DP-induced gains in innovation quality. The findings underscore the importance of complementary market reforms and platform governance to unlock the full potential of digital platforms for high-quality innovation.

Using provincial panel data (2011–2021), the study demonstrates that digital platforms significantly and directly promote innovation quality, with stronger effects in more developed eastern regions. Resource mismatch serves both as a mediating channel—DP reduces mismatch and thus raises innovation quality—and as a threshold constraint: when mismatch is low (≤0.399), DP’s positive effect is large; when mismatch is high (>0.399), the effect is much smaller. Extensive robustness checks corroborate these conclusions. Policy implications include: (1) supporting DP-centric, digital technology-based enterprises and projects, while curbing platform monopolistic practices and vicious competition to ensure a safe, efficient innovation environment; (2) implementing dynamic resource allocation reforms that reduce administrative distortions, improve factor price systems, and facilitate interregional mobility of talent, capital, and technology to alleviate mismatch and lock-in; and (3) fostering firm strategies that balance innovation quantity with quality, with improved evaluation systems and increased inputs in science, technology, and high-skilled talent. Future work should deepen analysis of platform governance (privacy, antitrust) and unpack multi-faceted mechanisms through frameworks such as TOE.

The study focuses on the DP–INN relationship under resource mismatch but does not fully address governance challenges such as data privacy protection, antitrust, and emerging digital monopolies that can shape platform impacts. The measurement of innovation quality, while improved via patent citations, may not capture all facets or time lags of innovation value realization. Mechanisms are likely more complex than modeled; future research will construct richer mediating frameworks (e.g., TOE-based) and explore how to enhance innovation quality while sustaining sufficient innovation quantity.