Analysis of the Spatial Pattern of Innovation-Driven Productivity at the Intra-Urban Scale in a Megacity Based on Multi-Source Data: A Case Study for Shanghai

Abstract

In the context of accelerating technological and industrial transformation, innovation-driven productivity has garnered significant attention. Based on multi-source data, this study employs the entropy method and spatial pattern analysis to delve into the spatial pattern of innovation-driven productivity. The results are as follows: (1) there is a huge difference in innovation-driven productivity at the street and township level in Shanghai; (2) innovation-driven productivity exhibits global spatial autocorrelation in Shanghai; (3) innovation-driven productivity shows a circle structure with high–high agglomeration at the center and low–low agglomeration at the periphery; (4) innovation-driven productivity hot spots are concentrated in the central region, while cold spots are distributed in a southeast–northwest trend around them. This study is of great significance for Shanghai to achieve an accurate allocation of resources, a coordinated development of industries, and an improvement of urban functions.

1. Introduction

Throughout the long course of human society, productivity has been continuously developing and evolving. In primitive societies, people relied on simple stone tools for hunting and gathering. The invention of the steam engine, internal combustion engine, and others led to a qualitative leap in productivity. Over time, the information technology revolution has pushed productivity to new heights. The emergence of new technologies, such as the internet, big data, artificial intelligence, and others, has profoundly changed people’s production, life, and thinking. Today, innovation-driven productivity, with innovation at its core, integrates cutting-edge scientific and technological achievements, new production factors, and production organization models, gradually becoming the key force in propelling economic and social development into a new phase [1].

The development of productivity theory has generally gone through five stages, namely classical political economy, Marxist productivity theory, neoclassical economics, innovation theory, and modern economic growth theory. During the period of classical political economy, Adam Smith emphasized the importance of the division of labor in improving productivity [2]. In “The German Ideology”, jointly written by Karl Marx and Friedrich Engels in 1846, the objective law of the dialectical development of productive forces and production relations was revealed [3]. Subsequently, Marx further elucidated the decisive role of productive forces in historical development, and proposed the significant proposition that “the revolutionary class itself is the most powerful form of productive force” [4]. During the period of neoclassical economics, Clark put forward the concept of “marginal productivity”, which referred to the increase in output resulting from adding one more unit of a certain factor of production while other conditions remain unchanged [5]. Solow and others introduced technological progress as an exogenous variable into the production function, and found that technological progress was a key factor in long-term economic growth [6]. Schumpeter put forward innovation theory, and believed that innovation was the core force driving the development of productivity. Scholars, such as Romer and Lucas, put forward the endogenous growth theory, emphasizing that elements like knowledge, human capital, and technological innovation interacted with each other and promoted the continuous growth of productivity. The new institutional economics school represented by North believed that institutions were the key endogenous variables that influenced the development of productivity [7]. Up to now, General Secretary Xi Jinping first mentioned the concept of “innovation-driven productivity” and emphasized its importance.

Innovation-driven productivity has emerged as the core driving force for promoting high-quality development in the wave of the new round of scientific and technological revolution and industrial transformation. Scholars generally agree that innovation-driven productivity is a kind of productivity that is innovation-oriented [8]. Innovation-driven productivity is spawned by revolutionary breakthroughs in technology, innovative allocation of production factors, and in-depth transformation and upgrading of industries [9]. From the perspective of supply, China has made significant progress both in the practice and understanding of productivity development, which has provided a solid foundation for the proposal of innovation-driven productivity. From the perspective of demand, China’s modernization drive has an inherent demand for the continuous development of productivity. Meanwhile, changes in both the domestic and international environments, as well as in science, technology, and industry have led to changes in the sources of formation and manifestations of productivity. From the perspective of demand, the requirements of China’s modernization drive for the continuous development of productivity, along with changes in the domestic and international environments and the transformation of the science, technology, and industry, have jointly become the sources for the formation of innovation-driven productivity [10].

Current research tends to focus on the differences at the provincial or large regional levels, while the analysis of the internal spatial pattern of specific cities is relatively lacking. Scholars have revealed that the innovation-driven productivity in China’s provincial regions shows a significant trend of regional imbalance [11]. Some scholars have used kernel density estimation and convergence models to analyze the evolution pattern and convergence of innovation-driven productivity in China’s three major regions. The results show that the level of innovation-driven productivity in the eastern region ranks first, followed by the central region, while that in the western and northeastern regions is relatively low [12]. Some scholars have also applied the entropy weight TOPSIS method to assess the level of innovation-driven productivity in the Yangtze River Delta region and utilized the Dagum Gini coefficient method to analyze the differences among the regions [13]. Overall, the research scale is mostly at the provincial or river basin level, and the Dagum Gini coefficient is mainly adopted as the method for a spatial pattern analysis. However, much of the existing literature implicitly relies on a market-centered or neoliberal explanatory logic, treating the spatial co-agglomeration of high-skilled labor, high-tech industries, and urban amenities as a self-reinforcing outcome of market forces. While such interpretations provide useful insights, they remain analytically insufficient and contextually incomplete for explaining the intra-urban spatial configuration of innovation-driven productivity in China. In particular, they overlook the extent to which state intervention, institutional arrangements, and strategic planning actively shape the spatial organization of innovation factors.

To address these gaps, this study examines the spatial pattern of innovation-driven productivity (IDP) at the township level in Shanghai using a cross-sectional analytical framework. The focus is on intra-urban spatial differentiation rather than temporal evolution, aiming to provide a fine-grained comparison of IDP across streets and townships. This study contributes the following: (1) Research at the township scale and the construction of a refined indicator system. This study analyzes the spatial pattern of innovation-driven productivity within the city from the township scale in Shanghai, which provides decision-making support for the refined planning and management of the city. The indicator system constructed in this article can comprehensively and systematically reflect the performance and development trend of innovation-driven productivity. (2) Multi-source data fusion and precise analysis. The application of data, such as nighttime lights, POI, road networks, and statistical yearbooks, breaks the limitations of single data types and provides a rich data basis for characterizing innovation-driven productivity. (3) Application and development of comprehensive spatial analysis method. Organically combine methods, such as the entropy method, global spatial autocorrelation, local spatial autocorrelation, and optimized hot spot analysis, to comprehensively analyze the spatial pattern of the innovation-driven productivity in Shanghai from a small spatial scale. Furthermore, by taking Shanghai as a global megacity embedded within the Chinese institutional context, this study moves beyond treating it as a mere variant of neoliberal urbanism and instead conceptualizes its development as a state-mediated and institutionally embedded process. It thus provides a context-sensitive analytical framework and offers a theoretically grounded reinterpretation of the spatial heterogeneity of innovation-driven productivity, rather than a simple empirical description.

2. Materials and Methods

2.1. Overview of the Study Area

Shanghai is known as China’s international center for economy, finance, trade, shipping, and technological innovation. It is located at the “T-shaped” intersection of China’s Coastal Economic Belt and the Yangtze River Economic Belt. By 2022, its administrative area was approximately 6340 square kilometers, and the permanent population was nearly 25 million. From a global perspective, Shanghai is located between 120°52′ and 122°12′ east longitude, and between 30°40′ and 31°53′ north latitude. Since the Chongming, Changxing, and Hengsha islands are not connected to the main part of Shanghai, as a whole, and there are significant differences in their development layouts, this study selects 201 streets and townships in the main part of Shanghai as research samples for the study of innovation-driven productivity (Figure 1).

It should be noted that this study adopts a fine-grained spatial analysis within a single megacity. As one of the largest global megacities in the Chinese context, Shanghai exhibits an exceptionally high degree of intra-urban heterogeneity, shaped by the interaction between market forces and strong state intervention. This complexity makes it particularly suitable for an analysis at the street and township level; as such, a fine-scale approach can better capture the micro-spatial differentiation of innovation-driven productivity that is often obscured in studies at coarser spatial levels. All analyses in this study are based on the data from a single year (2022), emphasizing cross-sectional comparability across different spatial units.

2.2. Data Sources

2.2.1. Data on the Level of Economic Development

This study has selected the NPP–VIIRS nighttime light image data (Figure 2). The data is sourced from the satellite sensors of the United States National Oceanic and Atmospheric Administration (NOAA). Its imaging principle is to utilize the VIIRS sensor to detect the self-radiation of objects on the Earth’s surface at night and the reflected light signals from natural light sources, such as moonlight. The data have the characteristics of high resolution, wide data coverage, time series, and multi-band information. is the data are widely used in fields such as economic research, urbanization monitoring, population estimation, environmental assessment, and disaster assessment [14,15,16,17]. Through saturation correction, mutual correction, and image continuity correction, with the street vector data of Shanghai as a mask, the effective areas covering the samples were extracted to obtain the nighttime light data for the year 2022. Generally speaking, the stronger the brightness of a city’s lights are, the larger its population will be and the more developed its industries will be [18]. The value of nighttime light brightness per unit area is an excellent substitute indicator for economic development [19]. This article obtains the scale of the digital economy in corresponding regions by combining nighttime light data with the areas of each street and township.

This study utilizes annual composite nighttime light (NTL) radiance data from the NPP–VIIRS sensor provided by the National Oceanic and Atmospheric Administration (NOAA). The dataset has a spatial resolution of 500 m, and radiance values are reported in nW·cm⁻²·sr⁻¹. To improve data quality and to reduce interference from non-luminous sources, a series of preprocessing and correction procedures were applied.

Saturation correction was first conducted to mitigate the residual saturation in urban core areas, improving the linear relationship between observed radiance and actual light intensity. This was followed by intercalibration to address radiometric inconsistencies arising from sensor drift, orbital artifacts, and lunar illumination, thereby enhancing data consistency and comparability. Finally, noise correction was performed to remove background noise, outliers, and transient light sources, such as wildfires and gas flaring, ensuring the robustness and reliability of the dataset.

2.2.2. Point of Interest Data

Based on the AMap Open Platform, this article captured the Point of Interest (POI) data for Shanghai. The time node was December 2022. After the data was cleaned and duplicates were removed, 427,613 valid samples were obtained. Based on different attributes, this study divides the data into the following three categories: transportation facility services; science, education, and cultural services; and the number of high-tech enterprises. The data on transportation facility services are used to reflect the level of infrastructure construction. The data on science, education, and cultural services are used to reflect the skills of workers. The number of high-tech enterprises is used to reflect the level of scientific and technological innovation. Considering the large scale of the dataset, this study takes the data of science, education, and cultural services in Shanghai as an example, and presents it in Figure 3. In order to clarify the density, a quadtree structure is adopted for the background of the graph; that is, the number of points contained in each grid is almost the same regardless of its size.

2.2.3. Road Network Data

The OpenStreetMap platform is the main source for obtaining road traffic network data in this paper. Data can be continuously updated and improved by editing maps and adding information, such as roads and streets, on its website. The road network mainly consists of highways, urban arterial roads, secondary roads, and other roads (Figure 4).

2.2.4. Statistical Data

Statistical data, such as per capita GDP, the proportion of employees in the tertiary industry, and highway mileage are sourced from the Shanghai Statistical Yearbook (2022), the Shanghai Statistical Bulletin (2022), the Shanghai Environmental Status Bulletin (2022), the Shanghai Urban Master Plan (2017–2035), the Shanghai Pudong New Area Statistical Yearbook (2022), the Yangpu Statistical Yearbook (2022), the Qingpu Statistical Yearbook (2022), the Fengxian Yearbook (2022), the Baoshan Yearbook (2022), the Jiading Yearbook (2022), the Jinshan Yearbook (2022), the Songjiang Yearbook (2022), the Xuhui Yearbook (2022), the Shanghai Huangpu District Statistical Yearbook (2022), as well as the statistical yearbooks of various sub-districts in Shanghai.

2.3. Methodology

2.3.1. Determination of the Index Weights of Innovation-Driven Productivity

The concept of entropy originally came from thermodynamics, and was initially used to describe the degree of disorder in a system [20]. Currently, whether in the fields of economy, management, or environment, among other fields, the entropy method can play a role in dealing with the comprehensive evaluation problems involving multiple indicators and multiple objects [21]. This method can make full use of the information contained in each indicator to reflect its relative importance [22]. Suppose the data is a matrix with $m$ rows and $n$ columns. The specific steps are outline below.

The standardization of indicators are as follows:

$$x_{ij}^{\prime }={\displaystyle \frac{x_{ij}-x_{j\left(min\right)}}{x_{j\left(max\right)}-x_{j\left(min\right)}}} (\mathrm{Positive})$$

(1)

$$x_{ij}^{\prime }={\displaystyle \frac{x_{j\left(max\right)}-x_{ij}}{x_{j\left(max\right)}-x_{j\left(min\right)}}} (\mathrm{Negative})$$

(2)

where $x_{ij}^{\prime }$ represents the standardized values respectively; xj(max) and xj(min) represent the maximum and minimum values of the j indicator data.

Calculate the proportion $w_{ij}$ of the i data in the j indicator as follows:

$$w_{ij}={\displaystyle \frac{x_{ij}^{\prime }}{{\sum }_{i=1}^m{x}_{ij}^{\prime }}} j=1,2\dots n$$

(3)

Then, calculate the entropy value $e_j$ of each indicator as follows:

$$e_j=-{\displaystyle \frac{1}{\mathrm{l}\mathrm{n}\left(m\right)}}{\sum }_{i=1}^m(w_{ij}\times ln{w}_{ij}) j=1,2\dots n$$

(4)

Calculate the information entropy redundancy degree $d_j$ as follows:

$$d_j=1-e_j j=1,2\dots n$$

(5)

Finally, calculate the weight $W_j$ of the j-th indicator as follows:

$$W_j={\displaystyle \frac{d_j}{\sum _{j=1}{d}_j}}$$

(6)

2.3.2. Methods for Spatial Pattern Analysis

This paper uses Moran’s I to study the spatial pattern model of innovation-driven productivity. Moran’s I is actually the spatial manifestation of the correlation coefficient [23]. It can test the spatial agglomeration trend or dispersion pattern presented by the data [24]. A significant positive value indicates the existence of positive spatial autocorrelation; that is, similar values tend to cluster spatially. A significant negative value indicates the existence of negative spatial autocorrelation; that is, high values and low values tend to be distributed dispersedly in space. If it is close to zero, it indicates that the data are randomly distributed in space [25]. Its calculation formula is as follows:

$$I={\displaystyle \frac{n{\sum }_{i=1}^n{\displaystyle {\sum }_{j=1}^n}{w}_{ij}\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{{\sum }_{i=1}^n{\displaystyle {\sum }_{j=1}^n}{w}_{ij}{\sum }_{i=1}^n{\left(x_i-\overline{x}\right)}^2}}$$

(7)

In the formula, $\overline{x}$ represents the data expectation, and $w_{ij}$ represents the spatial weight matrix. Its significance level is usually determined by the P-test. If the p-value is less than the given significance level α, the null hypothesis that similar observed values are spatially clustered will be rejected; otherwise, the null hypothesis will be accepted. In the analysis of practical problems, the significance level α is usually 0.05.

Professor Luc Anselin from Arizona State University put forward a class of general indicators which he called the local indicators of spatial association (LISA) in 1995 [26]. LISA can reveal the spatial clustering patterns within local areas and avoid the problem that local information is averaged or masked in the global analysis [27]. The spatial clustering patterns can be divided into four types by calculating the local Moran’s I. Both the observed value and its spatial lag are greater than the expectation, belonging to the High–High clustering type (HH). The observed value is greater than the expectation while the spatial lag is less than the expectation, belonging to the High–Low clustering type (HL). The observed value is less than the expectation and the spatial lag is also less than the expectation, belonging to the Low–Low clustering type (LL). The observed value is less than the expectation while the spatial lag is greater than the expectation, belonging to the Low–High clustering type (LH). LISA is usually calculated using the following formula:

$$I_i= {\displaystyle \frac{n\left(x_i-\overline{x}\right)}{{\sum }_{i=1}^n{(x_i-\overline{x})}^2}}{\sum }_{j=1}^n{w}_{ij}(x_j-\overline{x})$$

(8)

where $w_{ij}$ is the value located in the i-th row and j-th column in the spatial weight matrix, $x_{i }$ and $x_j$ represent the observed values of spatial unit i or j, and $\overline{x}$ is the expectation of the observed values.

Professors Arthur Getis and J.K. Ord jointly proposed another measure of spatial association in 1992—the G statistic [28]. Getis and Ord believed that, by combining the G statistic with Moran’s I, the pattern characteristics that were not revealed by one of the indices could be identified [29]. In particular, the $G_i^{*}$ statistic can detect local “pockets”, which are often referred to as “hot spots” nowadays [30]. In a subsequent study in 1995, they optimized this statistic and put forward the standardized $G_i^{*}$ statistic [31]. Due to the unique properties of the standardized $G_i^{*}$ statistic for high–high clustering and low–low clustering, it has been widely used by scholars in fields such as demography, economic geography, and epidemiological research [32,33]. The $G_i^{*}$ statistic is calculated as follows:

$$G_i^{*}= {\displaystyle \frac{{\sum }_{j=1}^n{w}_{ij}{x}_j}{{\sum }_{j=1}^n{x}_j}}, all j$$

(9)

$$Standardized G_i^{*}={\displaystyle \frac{{\sum }_{j=1}^n{w}_{ij}{x}_j-w_{ij}\overline{x}}{s{(\raisebox{1ex}{$(n{\sum }_{j=1}^n{w}_{ij}^2-({{\sum }_{j=1}^n{w}_{ij})}^2)$}\!\left/ \!\raisebox{-1ex}{$\left(n-1\right)$}\right.)}^{1/2}}}, all j$$

(10)

$$s={[\left({\displaystyle \raisebox{1ex}{${\sum }_{j=1}^n{x}_j^2$}\!\left/ \!\raisebox{-1ex}{$n$}\right.}\right)-{\overline{x}}^2]}^{1/2}$$

(11)

where $w_{ij}$ is the value located in the i-th row and j-th column of the spatial weight matrix, $x_i$ or $x_j$ represents the observed value of spatial unit i or j, and $\overline{x}$ is the expectation of the observed values.

2.3.3. Construction of Innovation-Driven Productivity Indicators

Based on the principle of the three elements of productivity, the innovation-driven productivity at the street and township scales in Shanghai is divided into the following three dimensions: new quality laborers; new quality means of labor; and new quality objects of labor (

Table 1. The index system of innovation-driven productivity at the township level in Shanghai.

Target Layer	Criterion Layer	First-Level Indicators	Second-Level Indicators	Attribute
Innovation-driven productivity in Shanghai	New quality laborers	The skills of laborers	The proportion of the population with a higher education	Positive
			The level of science, education, and cultural services	Positive
		Labor productivity	Per capita GDP	Positive
			Per capita wage	Positive
		The awareness of laborers	The activity level of innovation and entrepreneurship.	Positive
			The proportion of employees in the tertiary industry	Positive
	New means of labor	Scientific and technological innovation	The number of high-tech enterprises	Positive
			The coverage rate of information-based infrastructure	Positive
			Research and development intensity	Positive
			The scale of the digital economy	Positive
		Infrastructure construction	The services of transportation facilities	Positive
			Highway mileage	Positive
			The length of optical fibers	Positive
			The coverage rate of 5G networks	Positive
	The objects of new quality labor	Industrial development	The output value of the new generation of information technology	Positive
			The proportion of new emerging industries	Positive
			The degree of financial service support	Positive
		The ecological environment	Forest coverage rate	Positive
			Wastewater emission	Negative
			Carbon dioxide emissions	Negative

). The indicator system consists of 3 levels and 18 subdivided indicators. Although the research of different scholars has its own focus, there are still commonalities in the selection of key indicators.

For the dimension of new quality laborers, the skills of laborers, labor productivity, and the awareness of laborers are relatively crucial [34]. The proportion of the population with higher education, the intensity of education funding, and the structure of students enrolled in schools are often used as specific indicators in the evaluation [35]. In this paper, the proportion of the population with higher education and the level of science, education, and cultural services are used to measure the skills of laborers. The activity level of innovation and entrepreneurship and the proportion of employees in the tertiary industry are used to measure the awareness of laborers.

For the dimension of the new means of labor, infrastructure construction and scientific and technological innovation are the main directions for scholars in the selection of indicators [36]. In terms of scientific and technological innovation, this paper has selected research and development intensity, the scale of the digital economy, and the number of high-tech enterprises. Traditional infrastructure is measured by highway mileage and the services of transportation facilities. Digital infrastructure is measured by the length of optical fibers and the coverage rate of 5G networks.

The objects of new-quality labor mainly involve two aspects, namely industrial development and the ecological environment [37]. The proportion of new emerging industries is the core indicator of industrial development. The degree of financial service support can objectively reflect the degree of industrial development. Wastewater and carbon dioxide emissions are involved in the evaluation of the ecological environment as negative indicators.

3. Results

3.1. Spatial Differentiation of Innovation-Driven Productivity at the Intra-Urban Level

There are obvious differences in the level of innovation-driven productivity among different regions in Shanghai, and the level in the central urban area is significantly higher than that in the suburbs. Hongkou District (such as Quyanglu), Xuhui District (such as Tianlin), and Jing’an District (such as Baoshanlu) show a relatively high level of productivity. However, the levels of innovation-driven productivity in some towns in Baoshan District (such as Yanghang), Fengxian District (such as Haiwan), and Jinshan District (such as Langxia) are relatively low (

Table 2. Classification of levels and representative areas.

Number	New Productivity Level	The Span of Innovation-Driven Productivity	Representative Streets or Townships
1	High	[0.44–1.00]	Quyanglu, Guangzhon lu, Ouyanglu, Tianlin, Kangjianxinqiao, Hunanlu, Baoshanlu, Shimenerlu, Caojiadu.
2	Medium–High	[0.25–0.44)	Changshoulu, Zhenru, Linfenlu, Yichuan lu, Nanjingxilu, Hongqiao, Ganquanlu, Jiangninglu, West Tianmulu.
3	Medium	[0.14–0.25)	Gumei, Tilanqiao, Weifangxincun, Kongjianglu, Quyanglu, Tangqiao, Nanjingdonglu, Sipinglu, Liangchengxincun
4	Medium–Low	[0.06–0.14)	Shihu, Laoximen, Guangfulin, Xinhong, Dinghai lu, Yangjing, Jinyangxincun, Fangsong, Zhoujiadu, Tianlin.
5	Low	[0.00–0.06)	Yanghang, Yuepu, Luodian, Luojing, Zhuqiao, Haiwan, Fengxian, Gucun, Qingcun, Pujiang, Xuxing, Chuanshaxin, Jinhui, Zhelin.

Note: The classification standard adopts the natural breaks method.

).

The levels in Table 2 are classified using the Jenks natural breaks method based on the empirical distribution of innovation-driven productivity in the study area. This method minimizes within-class variance and maximizes between-class differences, thereby revealing the inherent structure of the data. It provides an objective basis for defining both the “new productivity levels” and the “span of innovation-driven productivity”, improving the robustness and interpretability of the classification.

3.2. Analysis of Typical Areas and Influencing Factors

The central urban areas represented by Hongkou District, Xuhui District, and Jing’an District have attracted a large number of innovative enterprises, high-end talents, and high-quality capital by virtue of their superior geographical locations, abundant talent resources, and well-developed infrastructure. The innovation-driven productivity in Jiangsulu of Changning District is as high as 0.71, thanks to its active exploration and investment in the fields of scientific and technological innovation and cultural creativity. The level of innovation-driven productivity in Caojiadu is 0.65, which is related to the high concentration of industries such as financial services and commercial office space.

In the peripheral areas of the central urban areas, the coordinated development of industries has played a crucial role in promoting the improvement of innovation-driven productivity. For example, towns in Putuo District, such as Changzheng, Wanli, and Zhenru, have formed a pattern where the digital industry serves as the core, and related supporting industries develop in a coordinated manner.

Constrained by factors such as geographical location, infrastructure construction, and industrial structure, the levels of innovation-driven productivity in some areas of Baoshan District, Fengxian District, and Jinshan District are relatively low. However, towns represented by Hongqiao also possess great potential. For example, the innovation-driven productivity in Langxia of Jinshan District is 0.03 (Figure 5). The proportion of traditional agriculture in the region is relatively large, while the development of industry and service sectors is relatively insufficient. Accelerating the adjustment, transformation and upgrading of the industrial structure is the key to improving its innovation-driven productivity. The innovation-driven productivity level in Hongqiao is 0.10. Thanks to its advantages in fields such as commerce and trade, and exhibition, and with the in-depth integration of emerging technologies and traditional industries, it has great potential for future development.

3.3. Global Spatial Autocorrelation Analysis

The results of an exploratory spatial analysis show that the innovation-driven productivity in Shanghai present significant spatial autocorrelation characteristics. As shown in Figure 6, Moran’s I is 0.44, the probability density distribution presents a unimodal pattern, the Z-score is 6.87, and the p-value is 0.001. Meanwhile, the Moran scatterplot shows that most of the samples are concentrated in the first and third quadrants.

On the one hand, due to the industrial agglomeration effect, similar high values tend to have spatial agglomeration. For example, a large number of high-tech enterprises and national scientific research institutes are clustered in Jiading District. There are as many as 5 national scientific research institutes and 230 high-tech enterprises just within the “Jiading Science and Technology Innovation Core”. The agglomeration of industries is conducive to the sharing of infrastructure, technology, and talent resources among enterprises, reducing production costs and improving production efficiency. The agglomeration of industries not only facilitates the sharing of infrastructure, technology, and talent resources among enterprises but contributes to the reduction of production costs and the enhancement of production efficiency. On the other hand, due to the uneven spatial distribution of innovative resources, the low observed values also show a tendency of agglomeration. Innovation is the core driving force of innovation-driven productivity. Universities, scientific research institutions, and innovation platforms in Shanghai are unevenly distributed in space, mainly concentrated in areas like Yangpu and Minhang.

3.4. Local Spatial Autocorrelation and Clustering Patterns

The measurement results of local spatial autocorrelation divide Shanghai into three layers with Lujiazui as the core and spreading outward (Figure 7). The central layer shows a high–high agglomeration. The spatial structure of the middle layer is not significant. And the peripheral layer mostly presents a low–low agglomeration. Areas such as Beixinjing, Jiangsulu, and Caojiadu show a spatial pattern where high values are surrounded by high values; while areas like Xinyang, Xuanqiao, and Zhoupu present a spatial pattern in which low values are surrounded by low values (

Table 3. Local spatial autocorrelation clustering patterns and representative areas.

Number	Pattern	Representative Streets or Townships
1	HH	Beixinjing, Jiangsulu, Caojiadu, Caoyangxincun, Zhoujiaqiao
2	LL	Xinyang, Xuanqiao, Zhoupu, Kangqiao, Datuan, Xinyang, Sanlin, Jinhui
3	NS	Yushan, Zhaoxiang, Fangsong, Xujing, Sijing, Xiayang, Jiuting

).

The high–high clustering (HH) is represented by Beixinjing Street (Table 3). Thanks to its core geographical location, it has attracted a large number of innovative resources. Meanwhile, the well-developed surrounding infrastructure, especially the transportation and communication networks, also provides strong support for the development of innovation-driven productivity. The street represented by the low–low value clustering (LL) is Xinyang. The geographical location is relatively remote and transportation is not convenient enough, resulting in relatively high costs for information exchange and transportation. The industrial structure of the surrounding towns and villages is mostly dominated by traditional agriculture or low-end manufacturing, making it difficult to form an industrial environment that is attractive to innovation-driven productivity.

3.5. Hot Spot Analysis of Innovation-Driven Productivity

The results of an optimized hot spot analysis show that the hot spots are concentrated in the central urban area, and the cold spots are distributed around it in a southeast–northwest trend. Districts like Huangpu and Jing’an, as the cores of Shanghai, enjoy prosperous commerce and a developed financial sector. A large number of corporate headquarters, financial institutions, and high-end service industries are gathered here. Some areas in its surrounding regions have also witnessed extremely active innovation activities, thanks to the spillover effect of the core area. The distribution of hot spots and cold spots is mainly influenced by the industrial structure and location factors. From the perspective of the industrial structure, the hot spot areas are mainly dominated by modern service industries, finance, and high-tech industries. These industries have high added value and are highly innovative or have strong innovation capabilities. However, traditional agriculture and low-end manufacturing in the cold spot regions have restricted the development of innovation-driven productivity. From the perspective of location factors, the city center enjoys convenient transportation, rapid information flow, and abundant talent resources. The outer suburbs lag behind in the development of innovation-driven productivity due to their remote geographical location, inconvenient transportation, and insufficient talent attraction.

4. Discussion

This study is based on cross-sectional data from a single year, and focuses on examining the spatial differentiation of new quality productivity at a specific point in time. The entropy weight method is employed to determine indicator weights, effectively reducing subjective bias and enhancing the objectivity and robustness of the evaluation results. This approach enables a more accurate identification of the relative contributions of different factors to innovation-driven productivity (IDP), thereby providing a solid foundation for analyzing its compositional structure at the intra-urban (subdistrict–township) scale in Shanghai. Furthermore, by integrating global spatial autocorrelation, local spatial autocorrelation, and optimized hot spot analysis, this study reveals the spatial distribution patterns and clustering characteristics of IDP from multiple spatial perspectives.

In contrast to existing studies that primarily rely on spatial econometric models, panel regression, or difference-in-differences approaches to identify causal relationships and to assess policy effects [38,39], this study adopts a different analytical pathway. Previous research has constructed composite indices using the entropy weight method and combined them with DEA and spatial econometric models to analyze spillover effects [40], or has employed panel data models to explore nonlinear relationships and mediating mechanisms between innovation and regional disparities [41]. These studies have mainly emphasized causal identification and mechanism testing at the macro or firm level. By comparison, this paper focuses on the intra-urban scale and develops an integrated analytical framework based on multi-source data and multiple spatial analysis techniques.

At the results level, the significant spatial autocorrelation and the “core–periphery” structure of new quality productivity identified in this study are broadly consistent with prior findings on spatial dependence and regional inequality [38,39]. However, rather than conducting direct causal inference, this study interprets the observed spatial patterns as the combined outcome of multiple underlying mechanisms, including industrial agglomeration, resource allocation, and locational conditions. This perspective provides a spatially grounded analytical framework for understanding how these mechanisms operate within megacities. Moreover, unlike studies that focus on single indicators or the trade-off between the quantity and quality of innovation [40], this study constructs a multidimensional indicator system to comprehensively characterize IDP, thereby offering a more systematic understanding of its spatial heterogeneity.

Despite these contributions, several limitations should be acknowledged. Although the entropy weight method enhances objectivity, the results remain sensitive to data quality and indicator selection. In addition, while spatial autocorrelation and hot spot analysis are effective in identifying spatial patterns, they are limited in capturing complex spatial interactions, particularly due to constraints associated with the specification of the spatial weight matrix. Owing to data limitations, certain sociocultural and institutional factors influencing IDP are not incorporated into the analysis. Future research could expand the data sources by incorporating indicators that better reflect institutional and policy environments, and combine higher spatiotemporal resolution data with more advanced spatial econometric or causal inference models [41] in order to more thoroughly uncover the dynamic evolution and underlying mechanisms of IDP and provide more precise support for urban governance and sustainable development.

5. Conclusions

Examining innovation-driven productivity at the street and township level is essential for advancing sustainable development, regional coordination, and the competitiveness of megacities, Under rapid globalization and urbanization, achieving intra-urban balance has become a common challenge [42]. As a key driver of growth, fine-scale innovation-driven productivity helps reveal intra-urban disparities and supports more effective urban planning and governance. Enhancing and balancing productivity across local units is therefore central to alleviating pressure in urban cores, optimizing spatial structure, and promoting urban–rural integration, insights which are applicable across megacities.

Using Shanghai as an illustrative case, innovation-driven productivity follows a pattern declining from central areas toward the urban fringe. This pattern indicates that talent concentration and infrastructure advantages reinforce central dominance while peripheral areas depend more on industrial diversification and synergy. Such a “core concentration–peripheral differentiation” structure typifies uneven spatial development in megacities.

Spatially, innovation-driven productivity exhibits strong clustering rather than random distribution, forming a “high-core, low-periphery” pattern, with more volatile dynamics in transitional zones. This reflects pronounced spatial dependence shaped by both core spillovers and local conditions. The interplay of agglomeration, spillover, and heterogeneity provides a useful framework for interpreting intra-urban inequality.

Hot spots and cold spots arise from interacting factors, including natural conditions, industrial structure, and technological capacity. This underscores the need for place-based strategies rather than uniform policies, alongside stronger cross-regional coordination. These insights are relevant for other megacities undergoing structural transformation.

The Shanghai case shows that the interaction between state intervention and market forces produces complex spatial differentiation in megacities. The contribution of this study lies not only in its empirical evidence, but in proposing a transferable analytical framework for understanding similar dynamics, particularly in the megacities of developing economies. By integrating a micro-scale spatial analysis with agglomeration, industrial synergy, and institutional factors, this framework advances an understanding of the spatial mechanisms underpinning innovation-driven development.