The Effect of Smart City Policies on City Innovation—A Quasi-Natural Experiment from the Smart City Pilot Cities in China

Abstract

The smart city pilot plan plays a pivotal role in the modernization of urban development within Digital China. To accelerate economic growth and establish a nation driven by innovation, it is crucial to examine the impact of smart city pilot programs on urban innovation. This research evaluates the effects of these policies using a multi-period difference-in-differences (DID) model. The findings indicate a significant enhancement in innovation levels within cities that participated in China’s smart city pilot program. These results remain robust even after rigorous validity tests. Mechanism testing reveals that the development of smart city pilot cities primarily boosts urban innovation through industrial upgrading and increased informatization. Additionally, heterogeneity tests show that the influence of smart city policies on urban innovation varies according to the geographical characteristics of different cities. Based on these insights, this study provides recommendations to further strengthen urban innovation by optimizing smart city policies. These include crafting differentiated smart city strategies, promoting the digital transformation of cities, and increasing economic support for smart city initiatives.

1. Introduction

Since the 1950s, global urbanization has been accelerating at an unprecedented rate, bringing about numerous challenges for cities worldwide. These challenges include economic strain, a declining quality of life, and obstacles to sustainable development. As urban populations grow and the pace of urbanization quickens, urban management is confronted with increasingly complex and multifaceted issues. Among these are unchecked urban expansion, excessive resource consumption, a deteriorating standard of living for residents, rising economic costs, and the limited sustainability of urban environments. These problems are prevalent in almost every country undergoing urbanization.

To address these challenges and foster sustainable urban growth, various strategies have been proposed by both nations and international organizations. One of the most prominent strategies that have garnered global attention is the concept of “smart cities”. The notion of a smart city has been evolving since IBM introduced the “Smart Planet” concept in 2008. Smart city management utilizes advanced information technologies to oversee urban operations in a comprehensive, intelligent, and collaborative manner. This approach is aimed at addressing the complexities of urban development, fostering business growth within urban areas, enhancing residents’ quality of life, and promoting the sustainable progress of cities.

Since 2009, countries around the world have embarked on building smart cities or developing strategies tailored to their unique contexts. For instance, Dubuque, in the United States, partnered with IBM to become the country’s first smart city. This initiative interconnected various city resources, including water, electricity, oil, gas, and public services, using IBM’s innovative technology to provide intelligent services through data observation, analysis, and integration.

In South Korea, there are ongoing plans to develop a smart city that combines green technology, digital innovation, and seamless mobile connectivity. Meanwhile, Europe is focusing on integrating information technology into healthcare, transportation, architecture, and ecology to create smart cities that are green, low-carbon, and sustainable. A notable example is Sweden, where IBM’s RFID technology is used to monitor vehicles entering the city center and impose a “road congestion tax” in an effort to reduce traffic congestion in Stockholm [1].

China has actively embraced the development of a “smart city”. In 2012, 90 prefectural and county-level cities were selected for a pilot program, with two additional cities joining later. To guide these initiatives, China introduced the Interim Management Measures for National Smart City Pilots and the National Smart City (District and Town) Pilot Indicator System, which provided standardized guidelines for smart city projects. Over the years, further documents have been released to refine the assessment systems and criteria for smart cities. By 2023, China is projected to have over 700 smart cities, with an estimated market value at RMB 875.44 billion. These pilot cities have modernized and optimized their network infrastructures and public information platforms to meet the country’s requirements for establishing smart cities. The primary goals include integrating advanced technologies, such as the Internet of Things (IoT), cloud computing, and big data, into urban development. This integration aims to promote smart transportation, smart logistics, smart communities, smart finance, and other areas. Additionally, smart security systems, infrastructure, management, and services have progressed more rapidly in pilot cities than in non-pilot cities, contributing to significant advancements in urban innovation.

Research on smart cities and urban innovation can be broadly categorized into the following three groups:

Firstly, innovation is widely recognized as a fundamental component of a smart city and a crucial driver of their growth. Many researchers view enhancing a city’s capacity for innovation as central to the concept of smart cities. Some even argue that boosting innovation should be the primary goal in developing a smart city, though this claim often lacks comprehensive theoretical backing. Key ideas suggest that smart city development can enhance innovation capacity and facilitate democratic decision-making [1]. Innovation is seen as essential to improving the competitiveness of public service provision and stimulating urban economic development [2]. While some scholars consider innovation a means, effect, or goal of smart cities, much of this discourse remains underexplored in terms of evidence. A smart city is an urban form that integrates learning and innovation [3]; it is also a hub where learning, information and communication technology (ICT), and modernization are highly concentrated [4].

Secondly, the development of smart cities is believed to enhance the potential for urban innovation. Scholars argue that building smart cities fosters urban creativity, particularly through the construction of smart cities that drive data-driven innovations. For example, Abella et al. (2015) explain how data-based innovation activities emerge within smart cities [5]. Similarly, Mora et al. (2020) argue that smart cities leverage ICTs to transform social capital structures, relationships, and perceptions. These changes also influence methods of knowledge transfer and sharing, all of which contribute to fostering innovation [6]. Other researchers suggest that the social bonds formed during the construction of smart cities can stimulate creative thinking. ICT is also said to promote urban innovation by reducing transaction costs, increasing citizen participation, facilitating industrial integration, and enhancing management efficiency [7]. Furthermore, empirical studies by Giourka et al. (2020) [8] in European smart cities like Évora and Alkmaar further illustrate how the establishment of Positive Energy Districts (PEDs), along with the energy transition initiatives can boost the innovative potential of smart urban environments. Giourka et al. (2020) also propose that PEDs represent a pioneering approach in energy, transportation, ICT, waste management, civic participation, and the preservation of historical urban fabric. They suggest that the adoption of such programs acts as a catalyst for fostering urban innovation. However, there remains a need for more comprehensive studies on the mechanisms associated with the development of innovative urban environments to support the smart city agenda.

Thirdly, several empirical studies have assessed the impact of smart city development on urban innovation. For instance, Caragliu et al. (2019) analyzed patent applications across 308 EU districts using indices such as smart citizenship, smart economy, smart environment, smart infrastructure, smart governance, and smart mobility [9]. They found a positive correlation between the intensity of these policies and the development of a city’s capacity for innovation. Similarly, Guo et al. (2022) analyzed data from 230 Chinese cities between 2002 and 2019 using the multiple-period difference-in-differences (DID) method, with patents in prefecture-level cities serving as a proxy for urban innovation performance [10]. This study found that the growth of smart cities significantly enhances urban innovation performance, and this positive effect is enduring. Ji et al. (2024) conducted an empirical study on 238 Chinese cities using an interleaved DID model and a mediated effects model, showing that smart city development improves innovation capacity, though the effects vary by region [11]. Further analysis revealed that the impact of smart city development on city innovation differs by area. Wang et al. (2022) examined the influence of smart city policy implementation on urban innovation output [12]. Additionally, Dana et al. (2022) [13] explored the relationship between urban innovation capability and smart city development through a questionnaire survey, considering the role of digital technology. Their findings indicated a bidirectional, positive relationship between smart city development and urban innovation: the construction of smart cities boosts urban innovation capabilities, while the enhancement of these capabilities further contributes to smart city development. The results suggest that the establishment of smart cities improves the innovation capacity of urban areas. While the influence of smart city development on urban innovation capacity has been empirically investigated, the underlying mechanisms remain underexplored and require further research.

In conclusion, this study has explored the benefits of smart city development in enhancing urban innovation capacity. However, the specific impact of smart cities on a city’s ability to innovate remains a critical question, as does whether the innovation effects differ based on the development of specific smart areas. To address these questions, this study employed quasi-natural experiments from smart city pilot programs and utilized a multiple-period DID model to assess their effects on urban innovation.

This paper makes several significant contributions. First, it investigates the theoretical impact of smart city pilot programs on urban innovation, thereby enhancing the broader understanding of how government innovation initiatives, urbanization, informatization, and urban development models influence innovation. Second, by analyzing patent data from 296 prefecture-level cities in China spanning from 2001 to 2021, this study mitigates sample selectivity bias through the use of the PSM-DID method and other techniques. A series of robustness tests were also conducted to ensure the reliability and validity of the findings. Third, this study thoroughly examines the varying impacts of smart city pilot policies across different regions and explores the mechanisms through which smart city construction influences urban innovation, with a focus on informatization levels and industrial structure. The findings provide empirical evidence supporting the continued promotion of smart city development.

The remaining sections of this study are organized as follows: Section 2 presents the testable hypotheses, Section 3 explains the model and related data, Section 4 discusses the results and conducts robustness checks, Section 5 explores heterogeneity, and Section 6 concludes with policy recommendations.

2. Theoretical Mechanisms and Research Hypotheses

A smart city represents a novel approach to urban development that moves beyond conventional paradigms and is fundamentally driven by information technology. While preserving the traditional capacity of cities to foster innovation, smart cities enhance this potential through the construction of intelligent infrastructure, optimized resource allocation, and improving government management through the use of the Internet and the Internet of Things. As a result, smart city development is a crucial strategy for regions transitioning from a traditional, factor-driven model to one that is innovation-driven, embodying a typical innovation ecosystem. Drawing from Schumpeter’s theory of innovation, the construction of smart cities encompasses a comprehensive innovation model that includes technological, organizational, and resource allocation innovation [14].

The construction of smart cities involves deploying a range of sensors and intelligent monitoring tools, providing advanced management capabilities for urban operations. These tools use cutting-edge scientific and technological methods to achieve real-time, comprehensive city supervision. Concurrently, during the development stages of a smart city, a dynamic database of city operations is established through data mining technology, which includes real-time monitoring of the city’s growth and market demands, with data processed and analyzed through cloud computing. This infrastructure supports the efficient flow of innovative elements, facilitates the optimization of resource allocation, and guides talent and knowledge toward the city’s high-tech industries. Furthermore, the creation of a smart city will foster the modernization and transformation of traditional industries. By integrating digital and intelligent technologies into business management and production processes, companies will be able to innovate and upgrade their production processes, modernize their traditional product lines, and enhance the intelligence and informatization of their business models. These advancements lead to improved management efficiency and steer businesses toward innovative growth. Based on these insights, this article proposes its first hypothesis, as follows:

Hypothesis 1. 

The smart city policy enhances urban innovative capabilities.

The distinguishing feature of smart cities compared to traditional cities is that information technology plays a core role in their development. Pilot cities develop intelligent infrastructure, create dynamic databases, and utilize cloud computing technology to process urban information due to the highly specific nature of smart city construction. These efforts lead to higher-quality urban services and the formation of an innovative ecosystem supported by digital technology [15]. Information technology is crucial in major projects related to building smart cities, such as smart finance, smart communities, and smart logistics, all of which elevate the standard of urban informatization. Additionally, the growth of the information industry also demands collaboration with highly skilled professionals, concentrating talent and enhancing knowledge spillover, which further drives urban informatization. Moreover, the development of smart city informatization involves optimizing the innovation environment, integrating information, and sharing data to overcome information transfer barriers [16]. The advancement of information technology contributes to the enhancement of urban innovation levels [17,18]. Consequently, as informatization progresses, entities engaged in innovation activities can leverage advanced information technology to establish open and efficient innovation networks. These platforms facilitate the dissemination of professional knowledge and creativity, improving the spillover effect of knowledge and boosting the innovation capabilities of involved entities. According to the discussion above, the article proposes the following:

Hypothesis 2. 

Constructing a smart city enhances urban innovation through the enhancement of informatization.

One of the primary objectives of constructing smart city pilots, as outlined by the Chinese government in the Interim Management Measures for National Smart City Pilots, is to facilitate the optimization and upgrading of industries. Research by Deng et al. (2013) highlights that smart city initiatives provide a significant competitive advantage for developing modern and high-tech service industries [19]. Other studies have confirmed that the smart city pilot policy has a pronounced effect on accelerating the optimization of the industrial framework, particularly in advancing and rationalizing industrial structures [11,20]. The advancement of cutting-edge technologies, such as big data, 5G, cloud computing, artificial intelligence, blockchain, and the Internet of Things, drives the upgrading of the industrial framework and raises standards for innovation in smart city development. It is evident that information-based sectors are closely linked to the progress of smart cities [21,22]. Building smart cities also supports the modernization and marketization of traditional industries, enabling them to adapt more readily to changing consumer demands. Furthermore, this development fosters the development of new business models and innovative approaches for restructuring and modernizing industries, as well as for establishing new information-based industries [23]. Traditional businesses can transition from labor-intensive to more intelligent modes of management and production. The transition to technology- or capital-intensive industries, alongside diverse operational models, not only enhances operational efficiency but also transforms developmental paradigms, significantly increasing innovation in these areas. Based on this, the research proposes the following hypothesis:

Hypothesis 3. 

Building smart cities facilitates the modernization of the industrial structure, which supports the growth of urban innovation potential.

The study framework for this paper is shown in Figure 1.

3. Results

3.1. Model Development

The standard DID method is typically applied to policy analysis at a single point in time. In contrast, the smart city pilot project follows a phased implementation over several years. Consequently, this research uses a multiple-period DID model to quantitatively assess the impact of smart city development on city innovation. This model designates a group dummy variable, assigning a value of 1 to the experimental group (pilot cities) and a value of 0 to the control group (non-pilot cities). The policy year is marked with a value of 1 for the pilot districts and for the years following the policy implementation, while other years are assigned a value of 0. This creates a dummy variable for the policy implementation period (dt). Following the approach of Beck et al. (2010), the bidirectional fixed-effects regression model is specified as follows [24]:

$${lninx}_{i,t}={\alpha }_0+{\alpha }_1{dt}_{i,t}+\sum {\gamma }_j{X}_{i,t}+{\mu }_i+{\eta }_t+{\epsilon }_{i,t}$$

(1)

(1)

In this model, ${lninx}_{i,t}$ represents the innovation capability of city i in year t (Kou et al., 2017). The binary variable ${dt}_{i,t}$ denotes whether city i was designated as a pilot smart city in year t; it takes the value of 1 if the city is a pilot and 0 otherwise.

(2)

$X_{i,t}$ is a control variable that accounts for factors potentially influencing the city’s innovation capability. The individual fixed effects are denoted by ${\mu }_i$, the time fixed effects are denoted by ${\eta }_t$, and the random error term is represented by ${\epsilon }_{i,t}$. The core coefficient of this paper is ${\alpha }_1$, which measures the net impact of smart city construction on a city’s innovation capability. A positive value of ${\alpha }_1$ indicates that the construction of smart cities enhances the level of city innovation. Control variables selected based on Huang et al. (2020) [25] are detailed in Section 3.2.

3.2. Data Descriptions and Descriptive Statistics of Variables

3.2.1. Data Descriptions

The focus of this study is on prefecture-level cities, as the smart city pilot projects are rolled out in phases across various administrative levels. Considering the availability of data, this article utilizes panel data from 296 prefecture-level cities in China, covering the years from 2001 to 2021. Data for the experimental group of explanatory variables were collected from the list of smart city pilots in China for the years 2012, 2013, 2014, and 2015.

The data for the explanatory variables were derived from Kou et al. (2017), who used the city innovation index provided by the “China City and Industry Innovation Report 2017” to evaluate the level of city innovation [26]. Control variable data were sourced from the “China City Statistical Yearbook”. To ensure data integrity, missing data were addressed through interpolation, and continuous variables were adjusted by 1% to minimize the influence of outliers.

3.2.2. Variable Descriptions

(1) Explanatory Variables

In this study, the natural logarithm of the urban innovation index is used as the explanatory variable. Urban finance, science, and technology inputs do not always fully translate into innovation capacity, making it challenging to evaluate the degree of city innovation based solely on these factors. Currently, the primary indicators for assessing innovation at the district level include the number of patents granted and the area’s investment in science and technology. While counting patents offers a more straightforward method for gauging innovation capacity, it is not a reliable measure of innovation level if the quality of the patents is not considered. This work references Kou et al. (2017), using data from the China City and Industry Innovation Report 2017 to assess municipal innovation levels [26]. To provide a more accurate measure of city innovation, the report constructs the city innovation index primarily based on patent data, taking into account the value of patents of varying ages.

(2) Core Explanatory Variables

In this research, the dummy variables representing the smart city pilot policy serve as the primary explanatory factors. These variables are coded as 1 in the year when the smart city pilot policy was implemented and for subsequent years in pilot cities, while other years are coded as 0.

(3) Control Variables

Consistent with the findings of previous theoretical and empirical studies, including those by Huang and Zhang (2020), this paper includes various variables that influence urban innovation. These variables include:

(1)

The degree of financial development (lnfinan), measured by the ratio of regional GDP to the year-end balance of deposits and loans in financial institutions;

(2)

The degree of openness (lnind), determined by the ratio of GDP to foreign direct investment;

(3)

The degree of cultural development (lncul), calculated by the number of public library books per 100 residents, relative to regional GDP;

(4)

The degree of human capital (lnhum), assessed by the number of students enrolled in higher education institutions in each region relative to the city’s population;

(5)

Science and technology expenditures (lntech), represented by the proportion of science and technology expenditures to local government spending;

(6)

The pace of economic growth (lnecon), which reflects the rate of economic advancement in the city;

(7)

The degree of economic improvement (lnecon), determined by the ratio of foreign investment usage to GDP.

3.3. Descriptive Statistics

The principal variables are presented in statistical form in

Table 1. Descriptive statistics.

Variables	N	Mean	Sd	Min	Max
lninx	6200	−0.0703	2.228	−5.087	5.552
DID	6258	0.154	0.361	0	1
lninst	6258	0.156	0.474	−1.179	1.574
lntech	6255	−4.967	1.038	−7.154	−2.664
lnhum	6201	10.22	1.451	6.174	13.55
lncul	6258	3.488	0.866	1.609	6.225
lnecon	6071	2.290	0.586	0	3.549
lnfinan	6258	−0.202	0.521	−1.443	1.190
lninfo	6250	−2.433	1.259	−5.791	−0.0730
lnind	6193	−6.584	1.468	−11.20	−4.046

. The logarithm of the city innovation index ranges from a maximum value of 5.552 to a minimum value of −5.078, indicating significant variation in innovation capacity across Chinese cities.

4. Empirical Testing and Analysis of Results

4.1. Benchmark Regression Results

The regression outcomes, presented in

Table 2. Benchmark regression of smart city construction to urban innovation ability.

	(1)	(2)	(3)	(4)
Variables	lninx	lninx	lninx	Lninx
DID	2.215 ***	0.171 ***	0.801 ***	0.091 ***
	(30.42)	(7.12)	(17.89)	(4.00)
lnecon			0.079 ***	0.040 ***
			(3.01)	(3.88)
lntech			0.797 ***	0.283 ***
			(43.05)	(26.31)
lnfinan			0.348 ***	0.250 ***
			(9.73)	(9.21)
lnind			−0.151 ***	−0.027 ***
			(−13.29)	(−4.14)
lnhum			0.698 ***	0.051 ***
			(51.07)	(3.33)
lncul			0.326 ***	0.215 ***
			(14.38)	(12.10)
Constant	−0.415 ***	−2.386 ***	−5.606 ***	−2.060 ***
	(−14.45)	(−88.54)	(−23.43)	(−11.75)
Observations	6200	6200	5903	5903
R-squared	0.130	0.917	0.711	0.930
Year fixed effect	No	Yes	No	Yes
City fixed effect	No	Yes	No	Yes

Note: in parentheses are the regional level cluster robustness standard errors; *** represent that they are significant at the statistical levels of 1%.

, are used to assess the hypotheses proposed earlier, prior to conducting the benchmark regression. In Model 1, the regression coefficient is significantly positive, considering only the smart city pilot dummy variable as the explanatory factor. Several reasons could explain this result: first, the time trend effect, which suggests that innovation levels in both control and experimental cities fluctuate annually; second, selectivity bias, which may account for the high levels of innovation observed in the selected smart city; and third, the inherent impact of smart city construction itself in raising the standards for urban innovation.

After adjusting for fixed effects, as shown in Column (1), the regression outcomes in Column (2) still demonstrate a positive coefficient, indicating that urban innovation continues to improve yearly, albeit at a different rate. Further adjustments for fixed factors based on Column (1) yield the regression outcomes in Column (3), where the coefficient remains significantly positive at the 1% confidence level. After accounting for both time and individual factors, as shown in Column (4), the results show no significant changes, thus confirming Hypothesis 1.

4.2. Parallel Trend Test

The DID method assumes parallel trends, which posits that in the absence of the pilot policy’s effects, the trends in urban innovation for both the experimental and control organizations would follow similar trajectories. However, the implementation of the smart city pilot policy involves a buffer period, resulting in a delay in policy effects. This delay arises due to various influencing factors, such as the intensity of policy implementation, the foundational elements of policy execution, and the adjustment of production components. To address this, the research incorporates insights from current literature [27,28,29] and employs event analysis, with the formulation as follows:

$${lninx}_{i,t}=\alpha +\sum _{k\ge 5}^5{\beta }_k{D}_{i,t}^k+\sum {\gamma }_j{X}_{i,t}+{\mu }_i+{\eta }_t+{\epsilon }_{i,t}$$

(2)

where ${lninx}_{i,t}$ is the logarithmic value of the level of urban innovation of the explanatory variable; ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{\mathrm{k}}$ denotes the “event” to build a “smart city”, which is the dummy variable. The specific allocation rule of ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{\mathrm{k}}$ is as follows: ${\mathrm{S}}_{\mathrm{i}}$ denotes the certain year to establish the “smart city” pilot, if ${\mathrm{t}-\mathrm{S}}_{\mathrm{i}}\le -5$, ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{-5}=1$, or else ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{-5}=0$ (to avoid multicollinearity, ${\mathrm{t}-\mathrm{S}}_{\mathrm{i}}=-20,\dots \dots ,-5$ is subsumed into $-5$); if ${\mathrm{t}-\mathrm{S}}_{\mathrm{i}}=\mathrm{k}$, then ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{\mathrm{k}}=1$, otherwise ${\mathrm{D}}_{\mathrm{i},\mathrm{t}}^{\mathrm{k}}=0$.

In the research, we adopt the graphical method to test the parallel trend and dynamic influences of the policy. The innovation levels before and after the pilot deployment are compared using a graphical technique. As illustrated in Figure 2, the parallel trend is satisfied, showing no difference in urban innovation between the experimental and control groups before the pilot. Figure 2 further supports this observation. Following the implementation of smart city initiatives, the policy effect becomes apparent. Throughout the pilot program, the policy effect initially increases but eventually declines and stabilizes. This pattern indicates that the impact of the smart city pilot on urban innovation is not driven by delays, and the policy effect stabilizes over a specific period.

4.3. Regression Analysis Based on PSM-DID Methodology

The smart city pilot policy is not a standard trial. Although the double-difference approach isolates the common treatment effect of the pilot policy, a Selection Effect (SELECTION) may still arise when analyzing the study data. To address this, the multi-temporal PSM-DID model employed in this work mitigates sample selectivity bias.

In the literature, two approaches are commonly used to address the challenges of combining using Propensity Score Matching (PSM) with cross-sectional data and DID with panel data. The first approach involves constructing a cross-sectional PSM by treating panel data as cross-sectional and performing matching. The second approach uses period-by-period matching, as detailed by a previous article [30]. Despite their drawbacks—such as potential “self-matching” issues when converting panel data into cross-sectional data, and possible inconsistent matching between pre- and post-policy objects with period-by-period matching—both methods are preferable under current conditions [31]. Therefore, this study employs both the period-by-period matching method and panel data transformation for PSM. This study selects matching variables based on city characteristics, including urban economic growth, government fiscal expenditure, financial development, openness to the outside world, human capital, and cultural level, drawing on the methodology of Heckman et al. (1998) [32]. The research produces two distinct datasets: first, a cross-sectional PSM dataset created using the nearest-neighbor matching method to identify the best control group for smart city pilot cities, and second, a dataset created by removing non-common support areas. Subsequently, a period-by-period matching method is applied, matching samples annually. Each year’s matching data are combined vertically into a panel dataset for regression analysis. After balancing the two matching datasets and verifying the matching effects, a multi-time DID approach is used to reassess the impact of smart city pilot strategies on urban innovation potential.

The balance test results are presented in Figure 3, which reflects the panel data matching process that treats the data as if it were cross-sectional. One key outcome meets the requirement that the standardized mean deviation remains below 10%. As shown in Figure 3, the standardized mean deviation for each matching variable is under 10% after matching, marking a notable improvement compared to the deviation before matching. The propensity score values for samples outside the main range are primarily clustered around 0.2 and 0.6, as illustrated in Figure 4. This suggests that the majority of samples in both the experimental and control groups fall within the same range of values. Although the cross-sectional Propensity Score Matching (PSM) reveals certain imbalances, the results should be interpreted as indicative rather than conclusive.

To determine whether there is a systematic bias in the matched variables between the two teams, the comparison must be conducted within the same year because the year-by-year PSM is performed on an annual basis. The method outlined by Xie et al. (2021) was applied to compare the logit regression results for each year before and after matching [31]. Specifically, if the coefficient values of each matching variable decrease and become insignificant, and if the pseudo-R² values decrease significantly after matching, it indicates there is no systematic bias in the matching variables between the two groups for each year. A comparison of

Table 3. Yearly Balance Test (Before Matching).

	2001b	2002b	2003b	2004b	2005b	2006b	2007b	2008b	2009b	2010b	2011b	2012b	2013b	2014b	2015b	2016b	2017b	2018b	2019b	2020b	2021b
lnecon	0.0344	0.2175	0.5068 *	0.0657	−0.0619	−0.1883	−0.0735	0.2448	−0.1756	−0.1345	0.094	−0.2927	−0.3937	−0.1364	−0.3360 *	−0.1408	−0.1417	−0.0188	0.0506	−0.0227	−0.3803 *
	−0.1426	−0.9064	−1.7606	−0.2716	(−0.2829)	(−0.9621)	(−0.3403)	−1.087	(−0.8014)	(−0.6022)	−0.482	(−1.3255)	(−1.5135)	(−0.6614)	(−1.6521)	(−0.7485)	(−0.6897)	(−0.0903)	−0.2557	(−0.1184)	(−1.7098)
lntech	−0.3157	−0.2908	−0.1722	−0.0839	−0.1499	−0.0518	0.1909	0.0257	−0.0216	0.0572	0.2013	0.1428	0.1795	0.3189	0.4577 **	0.2446	0.3887 **	0.3839 **	0.3791 **	0.3403 **	0.3306 **
	(−1.3894)	(−1.2423)	(−0.7047)	(−0.3422)	(−0.6030)	(−0.2189)	−0.859	−0.1204	(−0.1064)	−0.285	−0.9718	−0.6904	−0.8194	−1.6141	−2.2809	−1.3873	−2.2887	−2.5156	−2.5722	−2.4111	−2.2463
lnfinan	−0.2867	−0.2679	−0.3219	−0.0731	−0.1062	0.002	0.0617	0.1071	0.1341	0.1977	0.106	0.0569	0.0936	0.0573	0.2549	0.1628	0.1587	0.0731	0.0945	0.0963	−0.0093
	(−0.9433)	(−0.8738)	(−0.9362)	(−0.2626)	(−0.3634)	−0.0069	−0.2123	−0.3957	−0.4851	−0.6991	−0.3848	−0.198	−0.2943	−0.1922	−0.7859	−0.5342	−0.526	−0.2416	−0.2997	−0.2969	(−0.0286)
lnind	−0.2255 **	−0.0935	−0.2168 **	−0.2590 ***	−0.2805 ***	−0.2664 ***	−0.2466 **	−0.2117 **	−0.1248	−0.2378 **	−0.2093 *	−0.3184 ***	−0.1936 *	−0.2128 **	−0.0841	−0.0603	−0.083	−0.1365	−0.1399	−0.1478 *	−0.1423
	(−2.1803)	(−0.9353)	(−2.1623)	(−2.8389)	(−2.9584)	(−2.6256)	(−2.2614)	(−1.9775)	(−1.2101)	(−2.2528)	(−1.8841)	(−2.9161)	(−1.8400)	(−2.0413)	(−0.8259)	(−0.6642)	(−0.9701)	(−1.5890)	(−1.5148)	(−1.6673)	(−1.5874)
lnhum	0.2442 **	0.2701 **	0.3054 **	0.2174 **	0.3089 ***	0.2545 **	0.1978	0.2471 *	0.1529	0.1551	0.1285	0.1801	0.1325	0.106	0.0287	0.0493	0.0271	0.0533	0.0929	0.0967	0.0909
	−2.0247	−2.3935	−2.5145	−2.0316	−2.7836	−2.2766	−1.6108	−1.9489	−1.2059	−1.2729	−1.0707	−1.4419	−1.1196	−0.8858	−0.2303	−0.3897	−0.2074	−0.416	−0.6988	−0.7702	−0.6959
lncul	0.285	0.1877	0.2402	0.3291	0.2454	0.1149	−0.0625	0.0102	0.091	0.0932	0.1047	0.0658	0.2293	0.1219	−0.0109	0.0906	0.0559	0.0672	0.0664	0.1407	0.1369
	−1.2837	−0.8028	−1.0077	−1.6269	−1.1797	−0.6025	(−0.3039)	−0.0534	−0.4678	−0.4701	−0.5318	−0.3305	−1.0305	−0.6007	(−0.0494)	−0.4534	−0.2892	−0.3634	−0.3676	−0.8193	−0.8117
Pseudo R2	0.0367	0.032	0.0498	0.0392	0.0432	0.0298	0.0225	0.0237	0.0143	0.025	0.0212	0.0341	0.0378	0.0303	0.0357	0.0188	0.0281	0.0296	0.035	0.0348	0.0396

Note: in parentheses are the regional level cluster robustness standard errors; ***, **, * represent that they are significant at the statistical levels of 1%, 5%, and 10%, respectively.

and

Table 4. Yearly Balance Test (After Matching).

	2001b	2002b	2003b	2004b	2005b	2006b	2007b	2008b	2009b	2010b	2011b	2012b	2013b	2014b	2015b	2016b	2017b	2018b	2019b	2020b	2021b
lnecon	0.0344	0.2175	0.5068 *	0.0657	−0.0619	−0.1883	−0.0735	0.2448	−0.1756	−0.1345	0.094	−0.2927	−0.3937	−0.1364	−0.3360 *	−0.1408	−0.1417	−0.0188	0.0506	−0.0227	−0.3803 *
	−0.1426	−0.9064	−1.7606	−0.2716	(−0.2829)	(−0.9621)	(−0.3403)	−1.087	(−0.8014)	(−0.6022)	−0.482	(−1.3255)	(−1.5135)	(−0.6614)	(−1.6521)	(−0.7485)	(−0.6897)	(−0.0903)	−0.2557	(−0.1184)	(−1.7098)
lntech	−0.3157	−0.2908	−0.1722	−0.0839	−0.1499	−0.0518	0.1909	0.0257	−0.0216	0.0572	0.2013	0.1428	0.1795	0.3189	0.4577 **	0.2446	0.3887 **	0.3839 **	0.3791 **	0.3403 **	0.3306 **
	(−1.3894)	(−1.2423)	(−0.7047)	(−0.3422)	(−0.6030)	(−0.2189)	−0.859	−0.1204	(−0.1064)	−0.285	−0.9718	−0.6904	−0.8194	−1.6141	−2.2809	−1.3873	−2.2887	−2.5156	−2.5722	−2.4111	−2.2463
lnfinan	−0.2867	−0.2679	−0.3219	−0.0731	−0.1062	0.002	0.0617	0.1071	0.1341	0.1977	0.106	0.0569	0.0936	0.0573	0.2549	0.1628	0.1587	0.0731	0.0945	0.0963	−0.0093
	(−0.9433)	(−0.8738)	(−0.9362)	(−0.2626)	(−0.3634)	−0.0069	−0.2123	−0.3957	−0.4851	−0.6991	−0.3848	−0.198	−0.2943	−0.1922	−0.7859	−0.5342	−0.526	−0.2416	−0.2997	−0.2969	(−0.0286)
lnind	−0.2255 **	−0.0935	−0.2168 **	−0.2590 ***	−0.2805 ***	−0.2664 ***	−0.2466 **	−0.2117 **	−0.1248	−0.2378 **	−0.2093 *	−0.3184 ***	−0.1936 *	−0.2128 **	−0.0841	−0.0603	−0.083	−0.1365	−0.1399	−0.1478 *	−0.1423
	(−2.1803)	(−0.9353)	(−2.1623)	(−2.8389)	(−2.9584)	(−2.6256)	(−2.2614)	(−1.9775)	(−1.2101)	(−2.2528)	(−1.8841)	(−2.9161)	(−1.8400)	(−2.0413)	(−0.8259)	(−0.6642)	(−0.9701)	(−1.5890)	(−1.5148)	(−1.6673)	(−1.5874)
lnhum	0.2442 **	0.2701 **	0.3054 **	0.2174 **	0.3089 ***	0.2545**	0.1978	0.2471 *	0.1529	0.1551	0.1285	0.1801	0.1325	0.106	0.0287	0.0493	0.0271	0.0533	0.0929	0.0967	0.0909
	−2.0247	−2.3935	−2.5145	−2.0316	−2.7836	−2.2766	−1.6108	−1.9489	−1.2059	−1.2729	−1.0707	−1.4419	−1.1196	−0.8858	−0.2303	−0.3897	−0.2074	−0.416	−0.6988	−0.7702	−0.6959
lncul	0.285	0.1877	0.2402	0.3291	0.2454	0.1149	−0.0625	0.0102	0.091	0.0932	0.1047	0.0658	0.2293	0.1219	−0.0109	0.0906	0.0559	0.0672	0.0664	0.1407	0.1369
	−1.2837	−0.8028	−1.0077	−1.6269	−1.1797	−0.6025	(−0.3039)	−0.0534	−0.4678	−0.4701	−0.5318	−0.3305	−1.0305	−0.6007	(−0.0494)	−0.4534	−0.2892	−0.3634	−0.3676	−0.8193	−0.8117
Pseudo R2	0.0367	0.032	0.0498	0.0392	0.0432	0.0298	0.0225	0.0237	0.0143	0.025	0.0212	0.0341	0.0378	0.0303	0.0357	0.0188	0.0281	0.0296	0.035	0.0348	0.0396

Note: in parentheses are the regional level cluster robustness standard errors; ***, **, * represent that they are significant at the statistical levels of 1%, 5%, and 10%, respectively.

reveals that most coefficient values of matching variables decrease and become insignificant after matching and that the pseudo-R² values for all regressions decrease significantly. This suggests that, to a certain extent, there is no systematic bias in the matching variables between the two groups across different years, and the results meet the balance test requirements.

According to the cross-sectional PSM and the year-by-year PSM procedures, the kernel density maps before and after matching between the experimental and control groups are displayed in Figure 5, Figure 6, Figure 7 and Figure 8. These maps partially demonstrate that both the cross-sectional PSM and year-by-year PSM methods reduce sample selection bias by producing treatment effects. As illustrated, while there is a noticeable deviation between the two kernel density curves before and after matching, the distance between the average lines is reduced, and the curves align more closely after matching.

Table 5. Comparison of cross-sectional PSM-DID regression consequences.

	(1)	(2)	(3)	(4)
Variables	lninx	lninx	lninx	lninx
DID	0.783 ***	0.138 **	0.147 **	0.141 **
	(19.01)	(2.01)	(2.18)	(2.08)
Control variables	Yes	Yes	Yes	Yes
City fixed effect	No	Yes	Yes	Yes
Year fixed effect	No	Yes	Yes	Yes
Constant	−5.972 ***	−3.071 ***	−3.041 ***	−3.140 ***
	(−24.96)	(−5.39)	(−5.44)	(−5.70)
R-squared	0.719	0.921	0.927	0.922
Observations	5903	5903	4157	5882

Note: in parentheses are the regional-level cluster robustness standard errors; ***, **, represent that they are significant at the statistical levels of 1%, 5% respectively.

and

Table 6. Comparison of PSM-DID regression results year by year.

	(1)	(2)	(3)
Variables	lninx	lninx	lninx
DID	0.128 *	0.130 *	0.138 **
	(1.92)	(1.95)	(2.01)
Control variables	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes
Constant	−2.747 ***	−3.088 ***	−2.703 ***
	(−4.54)	(−5.60)	(−4.51)
R-squared	0.928	0.926	0.927
Observations	4086	5710	5983

Note: in parentheses are the regional-level cluster robustness standard errors; ***, **, * represent that they are significant at the statistical levels of 1%, 5%, and 10%, respectively.

present the results of additional regression analyses conducted after validating the reasonableness of the two approaches. Column (4) of Table 5 and Column (2) of Table 6 display the outcomes of the multi-temporal PSM-DID using two different methods. The results indicate that the development of smart cities has enhanced the level of urban innovation, reaffirming the robustness of the benchmark regression estimates. The coefficient of lninx is positive, consistent with the results of the benchmark regression, further supporting the conclusion that smart cities contribute to enhancing urban innovation.

4.4. Robustness Tests

4.4.1. Placebo Test

To strengthen the argument that the smart city plan positively impacts urban innovation, this study employs a placebo test, drawing on methods from previous research, to strengthen the argument that the smart city plan contributes to an increase in urban innovation. Specifically, the paper utilizes a random sampling approach, conducting 1000 iterations to randomly assign experimental and control organizations from the sample set. This process assesses the impact of the smart city policy on the control cities by introducing a random shock.

The baseline regression model is repeatedly estimated to derive 1000 coefficients for the core explanatory variables. This iterative process ensures that the impact of the smart city policy on urban innovation is not confounded by other variables. After randomization, the regression coefficient for city innovation is determined to be 0.0907. Figure 9 illustrates the kernel density distribution, with the dashed line representing the actual estimated coefficient from the PSM-DID analysis. The significant difference between this value and the placebo test coefficient suggests that the observed increase in city innovation, attributed to smart city policies, is unlikely to be due to unobservable factors.

4.4.2. Controlling for Shocks from Other Similar Policies

The national-level big data comprehensive pilot zone and broadband pilot policies implemented during the study period are relevant to this research. To control for their potential impact on the assessment outcomes, the baseline regression model includes dummy variables for the years these two policies were introduced: the broadband pilot and the national-level big data comprehensive experimental zone. The results are presented in

Table 7. Eliminate other policy interferences.

	(1)	(2)	(3)	(4)
Variables	lninx	lninx	lninx	lninx
DID	0.1121 **	0.1060 ***	0.1305 **	0.1051 *
	(2.0660)	(4.3242)	(2.0568)	(1.8277)
DID1	−0.0036	−0.0087
	(−0.0570)	(−0.3554)
DID2			−0.2437 ***	−0.2111 ***
			(−3.3246)	(−3.1188)
Control variables	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes	Yes
Constant	−0.5527 ***	−1.1725 ***	−1.2116 ***	−2.0230 ***
	(−20.5258)	(−5.6600)	(−38.6239)	(−3.8654)
R-squared	0.8960	0.9070	0.9103	0.9206
Observations	3226	3113	4291	4136

Note: in parentheses are the regional level cluster robustness standard errors; ***, **, * indicate that they are significant at the statistical levels of 1%, 5%, and 10%, respectively.

, columns (1) through (4). The estimation outcomes show that the coefficient for the dummy variable related to innovative city construction remains significantly positive, even when accounting for the two policies. This indicates that the smart city pilot policy significantly enhances cities’ capacity for innovation, reinforcing the robustness of the paper’s findings.

5. Heterogeneity Analysis

5.1. Economic Regional Heterogeneity

China’s vast geographic size results in diverse regions with varying capacities to support innovation. These variations can lead to differences in the effectiveness of smart city initiatives. This paper categorizes cities into eastern, central, and western regions based on their geographic locations and performs regression analysis for each region separately, with results presented in

Table 8. Regional heterogeneity analysis.

	(1)	(2)	(3)
Variables	East	Middle	West
DID	0.0318	0.1052 **	0.3071 ***
	(0.8145)	(2.5461)	(8.9309)
Control variables	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes
Constant	−0.1988	−3.2599 ***	−0.7196 ***
	(−0.6270)	(−11.9492)	(−2.7688)
R-squared	0.9518	0.9177	0.9403
Observations	2058	1844	2001

Note: in parentheses are the regional level cluster robustness standard errors; ***, **, indicate that they are significant at the statistical levels of 1%, 5%, respectively.

. The regression results indicate that while the impact of smart city development on innovation is not significant in the eastern region, it significantly enhances city innovation in the central and western regions. Recent strategies, such as the Yangtze River Economic Belt and the “Belt and Road” initiative, as well as the Chengdu-Chongqing Urban Agglomeration, the Wuhan Metropolitan Area, and the Central Plains Urban Agglomeration, have begun to yield positive results. These strategies, along with the agglomeration of innovation resources in these areas, outperform the eastern regions, which have benefited from more advanced smart policies. This highlights the pronounced regional heterogeneity of smart city initiatives and underscores the importance of smart city development in the central and western regions.

5.2. Mechanism Analysis

This section empirically examines how the development of smart cities impacts the level of city innovation based on the preceding analysis.

The regression results presented in columns (1) and (2) of

Table 9. Mechanism test.

	(1)	(2)	(3)	(4)
Variables	lninx	lninx	lninx	Lninx
DID	0.1641 ***	0.0868 ***	0.3511 ***	0.1849 ***
	(6.8584)	(3.8366)	(7.3283)	(4.0871)
DIDXinst	0.2096 ***	0.1644 ***
	(5.3883)	(4.4427)
DIDXinfo			0.1181 ***	0.0595 **
			(4.1083)	(2.2069)
Control variables	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes	Yes
Constant	−2.3828 ***	−2.0359 ***	−2.4696 ***	−1.8416 ***
	(−87.6114)	(−11.6243)	(−40.2874)	(−10.8086)
R-squared	0.9169	0.9303	0.9188	0.9319
Observations	6200	5903	6193	5897

Note: in parentheses are the regional level cluster robustness standard errors; ***, **, represent that they are significant at the statistical levels of 1%, 5%, respectively.

reveal that the cross-multiplier of the smart DID and the industrial structure (lnist) in the baseline model has a significant positive regression coefficient at the 1% significance level. This finding indicates that the expansion of smart cities promotes the modernization of a city’s industrial structure, which, in turn, enhances the city’s capacity for innovation. Consequently, Hypothesis 3 is supported.

This research also integrates the interaction terms between the city informatization level variable (lninfo) and the smart DID into the baseline regression model. Previous studies have identified the per capita number of Internet users in a city as a measure of city informatization (lninfo). Whether or not control variables are included, the coefficients of the interaction terms remain significantly positive, as demonstrated in Columns (3) and (4) of Table 9. This indicates that the introduction of smart city initiatives serves as an exogenous shock that effectively enhances the level of city informatization. This enhancement promotes more efficient information exchange within the city and accelerates information processing, thereby fostering urban innovation. Given that smart city projects are supported by advanced data and communication technologies, such as the Internet, Hypothesis 2 is also validated.

6. Conclusions and Policy Recommendations

6.1. Conclusions

The development of “smart cities” has profoundly influenced urban development models by addressing the challenges associated with urban growth over years of practice. This study utilizes a quasi-natural experiment of smart city pilots and panel data from 296 cities spanning from 2001 to 2021. By employing multiple-period DID, PSM-DID, and other methods, this study empirically examines the impact of smart city initiatives on urban innovation, grounded in theoretical analysis. The key findings are as follows: (1) The construction of smart cities contributes to increased urban innovation. (2) Smart city initiatives help improve the city’s industrial structure, thereby fostering greater innovation. (3) It can advance the level of city informatization, which also promotes innovation. (4) The impact of smart city initiatives varies across geographic regions. Over time, the initial substantial enhancement in innovation driven by pilot projects will diminish and stabilize, leading to a steady policy impact.

6.2. Policy Recommendations

The findings of this study suggest several key policy implications:

Firstly, the development of “smart cities” has had a substantial impact on urban development models by addressing longstanding challenges in urban growth. This research employs a quasi-natural experiment of smart city pilots, using panel data from 296 cities between 2001 and 2021. Through the application of multiple-period DID and PSM-DID methods, this study empirically investigates the effect of smart city initiatives on urban innovation, grounded in theoretical analysis. The primary findings are as follows: (1) Smart city construction significantly enhances urban innovation. (2) It improves the industrial structure of cities, thereby promoting greater innovation. (3) It advances the level of informatization within cities, further encouraging innovation. (4) It creates a heterogeneous mix of geographic regions. Over time, the initial surge in innovation driven by the pilot projects diminishes and stabilizes, leading to a steady and enduring policy impact.

Secondly, a gradual and hierarchical approach should be adopted to effectively advance smart city construction, tailored to the specific characteristics of each city. For smaller cities with lower economic development, higher levels of human capital, stronger financial development, and better informatization, it is essential to fully leverage the potential of smart city initiatives to enhance urban innovation capacity while maintaining their existing developmental advantages. This strategy ensures the full realization of policy benefits. Conversely, for larger cities with higher economic development, lower levels of human capital, inadequate financial development, and weaker informatization, it is crucial to closely monitor and evaluate the policy’s effects. Regular assessments of the economic and environmental impacts of smart city initiatives should be conducted to facilitate timely adjustments and optimizations of the strategy.

Thirdly, it is vital for the government to fully commit to increasing expenditures on science and technology, as well as on upgrading industrial structures. The government should enhance support for urban innovation activities by developing and improving platforms for innovation and research and development. This includes increasing the proportion of science and technology expenditures within the overall government budget to ensure sufficient financial backing for these innovation efforts. Simultaneously, the optimization of the city’s industrial structure should be prioritized by advancing high-tech industries to achieve comprehensive coverage of the Internet of Things. Additionally, fostering new forms of productivity focused on intelligent industries should be a priority.

6.3. Study Restrictions

The primary limitations in this paper are related to sample matching and variable selection, which warrant further investigation for potential improvements. Additionally, the paper does not examine how varying intensities of smart city policies (i.e., different levels of intervention) affect urban innovation capabilities. Furthermore, the main objective of the follow-up study is to develop an optimal design framework and advancement methodology for creating smart cities. This methodology will classify the building process, layer advancement, and ongoing demands from the perspective of “constructing a new urban operating environment”.