The study explores the relationship between human capital and urban innovation, particularly focusing on the often-overlooked spatial spillover effects within urban agglomerations. Human capital, encompassing skills, knowledge, and experience, acts as a catalyst for technological innovation and improved urban innovation capacity. In China, the concentration of highly skilled individuals in urban areas like the Beijing-Tianjin-Hebei and Yangtze River Delta regions creates a strong spillover effect, boosting regional innovation. The Yangtze River Economic Belt (YREB), a vast area encompassing 11 provinces and 108 cities, presents a compelling case study due to its significant economic contribution (46.6% of China's GDP in 2020), large population (600 million), substantial talent pool (18 million professional and technical talents, 20.2 million highly skilled talents), and the presence of similar talent incentive policies. However, the YREB also exhibits significant internal disparities in development levels, hindering transboundary cooperation and creating challenges for balanced development. Developed areas, such as the Yangtze River Delta region, benefit from the positive impacts of high-level human capital, while underdeveloped and poor areas suffer from a lack of talent and slow development. This study aims to analyze the interactions between heterogeneous human capital and regional innovation, quantify the spatial spillover effects using the Spatial Durbin Model (SDM), and provide policy recommendations to optimize human capital allocation and achieve balanced development within the YREB.

Existing literature supports the positive correlation between human capital and innovation, emphasizing the role of knowledge spillover and communication in strengthening this relationship. Studies indicate that improved human capital enhances technology adoption and regional innovation. However, the impact of human capital on innovation varies across regions depending on the level of economic development. Research suggests that in developing countries with low human capital levels, improvements in even low-level human capital can enhance innovation. Previous studies have used methods like spatial stochastic frontier methods and spatial econometric models to analyze spatial spillover effects. The Spatial Durbin Model (SDM) is frequently chosen for its ability to address endogeneity and spatial spillover effects. However, the SDM's limitation in addressing heterogeneity of spatial spillover effects necessitates the use of geographically weighted regression (GWR). This study addresses gaps in the literature by focusing on heterogeneous human capital (low, intermediate, and high levels) and by adopting a city-level analysis, complementing macro and micro-level studies. The focus on the 108 diverse cities within the YREB allows for an exploration of individual characteristics and regional differences in the context of balanced regional development.

The study employs a Cobb-Douglas production function to model the relationship between innovation inputs and outputs. The dependent variable is urban innovation capability, measured by per capita patent authorization. Independent variables include three levels of human capital (low, intermediate, and high), measured by the number of students per capita in primary, secondary, and tertiary education, respectively. Control variables include foreign direct investment (FDI) per unit of GDP, science and education investment as a proportion of total financial expenditure, industrial structure (ratio of tertiary to secondary industry GDP), and population density. The study employs a spatial Durbin model (SDM) to account for spatial spillover effects. A spatial autocorrelation test using Moran's I is conducted to determine the presence of spatial dependence. Model selection involves Lagrange Multiplier (LM) tests to determine appropriate spatial econometric models (SLM, SEM, SDM). Wald and LR tests are then conducted to assess if the SDM should be simplified to SLM or SEM. Finally, a geographically weighted regression (GWR) is applied to explore the spatial heterogeneity of the impacts across upstream, midstream, and downstream cities in the YREB. Panel data from 2011-2020 were collected from multiple sources (China Urban Statistical Yearbook, China Statistical Yearbook, provincial and city statistical yearbooks). Missing values were estimated using the growth rates and benchmark data of previous years. Data were logarithmically transformed to address issues with scale.

Analysis of the spatial-temporal evolution of innovation in the YREB (Figure 3) reveals that patent authorization has significantly increased over time and shows a westward diffusion pattern. Innovation is highest in downstream cities, exhibiting a decreasing trend from downstream to upstream. Results from the Spatial Durbin Model (Table 6) show that: 1. Low-level human capital has a negative direct effect on urban innovation (-5.44% decrease in patent authorization for a 1% increase in low-level human capital). 2. Intermediate and high-level human capital have positive direct effects (3.56% and 1.36% increase in patent authorization for a 1% increase in intermediate and high-level human capital, respectively). 3. The indirect effects of human capital are the opposite of the direct effects: low-level human capital has a positive indirect effect, while intermediate and high-level human capital have negative indirect effects. Regarding control variables (Table 6), foreign investment and science and education investment show negative indirect effects. Industrial structure has a negative indirect effect, while population agglomeration has negative direct and indirect effects. An ANOVA test (Table 7) confirms that the GWR model fits the data significantly better than the OLS model. Spatial heterogeneity analysis (Figure 4) reveals that the impact of low-level human capital changes from negative to positive over time and diffuses from downstream to upstream. The effect of intermediate human capital decreases over time, becoming negative in upstream regions. High-level human capital shows a spatial polarization effect in 2020, decreasing in downstream and midstream areas while increasing slightly in upstream areas. A robustness test using an economic distance matrix (Table 8) generally supports the findings from the spatial distance matrix analysis, although some individual effects differ.

The findings confirm the significant impact of heterogeneous human capital on regional innovation. The negative impact of low-level human capital in the early stages of development suggests a mismatch between skill levels and the demand for innovation. The positive impact of intermediate and high-level human capital underscores the importance of education in fostering innovation. The opposite trends observed for direct and indirect effects suggest that the flow of high-skilled talent can have a negative spillover effect on surrounding areas, possibly due to competition for resources or talent drain. The negative impact of excessive foreign investment, science and education investment, and population agglomeration highlights the importance of optimized resource allocation. The spatial heterogeneity analysis emphasizes the need for differentiated policies tailored to the specific needs of different regions within the YREB. The GWR results demonstrate the variation in the impact of different factors on innovation capacity across the region and over time, calling for location-specific policies.

This study highlights the significant but heterogeneous impact of human capital on regional innovation within the YREB. Low-level human capital hinders innovation, whereas intermediate and high-level human capital fosters it. Spatial spillover effects are context-dependent, varying across regions and over time. The study suggests policy recommendations focusing on enhancing talent spillover, optimizing human capital investments based on regional needs, and improving basic education in upstream and midstream cities, while reforming higher education in downstream cities. Future research could explore the use of average years of education as a more comprehensive human capital indicator, and the findings could be applied to similar economic belts to enhance cross-regional comparative analysis.

The study's reliance on the number of students per capita as a proxy for human capital presents a limitation, as it may not fully capture the quality or skills of the human capital. Data limitations from utilizing multiple public sources might introduce uncertainty and influence the interpretation of causality. The focus on patent authorization as a measure of innovation may overlook other aspects of innovation output. Further, the cross-sectional nature of the data may not capture fully the dynamic effects of human capital on innovation.