Innovation-led productivity lifts urban energy utilization efficiency via policy and governance

Jing Zhou 1

Zhongwu Zhang 1

Xiaodong Chang 2✉ Emailchangxd674@nenu.edu.cn

Kunwei Zhang 1 Emailzhangzw@sxnu.edu.cn

Lifen Qin 1

1 School of Geographic Science Shanxi Normal University 030031 Taiyuan China

2 School of Geographic Science Northeast Normal University 130024 Changchun China

Jing Zhou ¹, Zhongwu Zhang*¹, Xiaodong Chang *², Kunwei Zhang¹ & Lifen Qin¹

¹ School of Geographic Science, Shanxi Normal University, Taiyuan 030031, China

² School of Geographic Science, Northeast Normal University, Changchun 130024, China

* Email: zhangzw@sxnu.edu.cn

* Email: changxd674@nenu.edu.cn

Abstract

The global pursuit of sustainable urban development is often undermined by the classic rebound effect where gains in energy utilization efficiency are paradoxically offset by increased consumption. This challenge underscores the limitations of isolated technological fixes and calls for a more systemic framework. Under this background, China's national strategy of "New Quality Productive Forces" (NPRO) presents a novel paradigm, yet its causal impact on urban energy utilization efficiency (TE) remains empirically underexplored. This study addresses this critical gap using a panel of 280 Chinese cities from 2010–2022, employing Double Machine Learning (DML) for robust causal inference and spatial econometric models to quantify spillover effects. The findings are as follows: first, the spatial analysis uncovers a pattern of asynchronous evolution. NPRO generation exhibits strong geographic stickiness in eastern coastal hubs, whereas its TE benefits display a broad diffusion pattern inland, driven by positive spatial spillovers to economically interconnected neighboring regions. Second, causal evidence indicates that NPRO significantly enhances urban energy efficiency; a one-standard-deviation increase in the NPRO index is associated with a 0.644-standard-deviation improvement in TE. Third, positive effect is primarily channeled through two institutional mechanisms: the intensification of low-carbon policies and the enhancement of governance transparency. Finally, this positive impact is heterogeneous, proving most pronounced in cities that prioritize environmental protection, possess advanced industrial structures, are in the eastern region, or have a lower degree of government intervention. Ultimately, this study offers robust causal evidence that a systemic, state-led productivity framework can mitigate traditional efficiency paradoxes, providing crucial implications for designing spatially-aware energy transition policies in emerging economies.

Keywords:

New quality productive forces

Energy utilization efficiency

Causal inference

Double machine learning

Spatial spillover effect

China

1. Introduction

The global imperative to reconcile economic growth with environmental sustainability has positioned the enhancement of energy utilization efficiency (TE) as a cornerstone of the green transition and a critical pathway to achieving the United Nations Sustainable Development Goals (SDGs) (Adetomi Adewnmi et al., 2023; Luo et al., 2024). Cities play a pivotal role in this transition, accounting for nearly 75% of global energy consumption and about 80–85% of China’s commercial energy use (Dhakal, 2009; Yang et al., 2024). Improving urban TE is therefore indispensable for sustainable development. However, the nexus between technological progress and net energy savings is fraught with complexity, most notably illustrated by the classic "rebound effect." This paradox suggests that efficiency gains, by lowering the effective price of energy services, can stimulate greater consumption, thereby offsetting or even eliminating anticipated savings (Wei and Liu, 2017). This phenomenon underscores the limitations of purely technology-centric approaches. A successful green transition requires a systemic framework that integrates technological innovation with institutional incentives, industrial upgrading, and optimized governance.

In this context, China’s national strategy of "New Quality Productive Forces" (NPRO) emerges as a large-scale quasi-natural experiment to address this systemic challenge. Conceptually, NPRO constitutes a paradigm shift away from traditional growth models. It aims to achieve a systemic enhancement in productivity, underpinned by foundational technological innovation. Specifically, NPRO emphasizes the optimized allocation of three factors of production: (1) highly skilled labor force (new-type laborers), (2) digital and intelligent infrastructure (advanced means of labor), and (3) strategic emerging industries that treat data as a core production element (novel objects of labor) (Lin et al., 2024). By coordinating technology, labor, and capital within a green-oriented framework, NPRO seeks to transcend the path dependence of energy-intensive growth models (Chin et al., 2025). Its systemic nature offers a potential solution to the rebound effect, providing an empirical setting to investigate how innovation-driven productivity frameworks can mitigate entrenched energy and environmental constraints with broader global implications.

Despite its theoretical importance, the sustainability outcomes of NPRO remain underexplored in the empirical literature. While a nascent body of work has begun to examine its economic impacts, robust causal evidence on its effect on TE is lacking. This motivates the research questions of this study: (1) What is the net causal effect of NPRO on TE? (2) Through which specific mechanisms does this impact manifest? (3) Does the effect exhibit heterogeneity and spatial spillovers across different regional and institutional contexts? To answer above questions while accounting for high-dimensional confounding factors, this study employs a Double Machine Learning (DML) approach to facilitate robust causal inference, complemented by spatial econometric analysis to capture spillover effects.

This study contributes to the literature on innovation-driven green transitions in three ways. First, this study provides a systematic causal assessment of the NPRO paradigm, extending the discourse from discrete technologies to a national-scale systemic strategy. Second, this research elucidates the pivotal role of institutional arrangements by identifying a synergistic institutional combination effect. This study finds that NPRO’s positive impact on TE is most pronounced in regions where effective government (proxied by strong environmental oversight) is coupled with efficient markets (characterized by minimal direct intervention). Third, this study identifies a phenomenon of spatially asynchronous evolution, whereby NPRO generation is geographically concentrated in innovation hubs while its TE benefits diffuse broadly across regions, offering a new perspective on the cross-regional transmission of innovation dividends. The study framework is illustrated in Fig. 1. Section 2 reviews the literature and develops the hypotheses. Section 3 describes the methodology, followed by Section 4, which reports the empirical results. Finally, Sections 5 and 6 discuss the findings and conclude the study.

Fig. 1

Research framework.

2. Literature review and hypotheses development

2.1. Measuring urban energy efficiency and its determinants

The scholarly approach to measuring energy utilization efficiency (TE) has evolved significantly, moving from simple single-factor indicators to sophisticated total-factor frameworks that account for environmental externalities. Early work conceptualized TE through the lens of energy intensity (energy consumption per unit of GDP), a metric whose primary limitation is its inability to account for the contributions of capital, labor, or technological substitution (He et al., 2024; Wu et al., 2023). To address this, Total Factor Energy Efficiency (TFEE), pioneered by Hu and Wang (2006), was developed to situate energy within a multi-factor production function. A critical subsequent refinement led to Green Total Factor Energy Efficiency (GTFEE), which incorporates undesirable outputs (e.g., pollutants) into the efficiency assessment, a methodological advancement largely based on the directional distance function and Data Envelopment Analysis (DEA) approaches for handling negative externalities (Chen et al., 2021). By systematically accounting for environmental costs, GTFEE offers a more accurate measure of the trade-offs between economic output and sustainability. Given its capacity to provide a robust and comprehensive assessment, this study adopts the GTFEE concept (hereafter TE) as its dependent variable (X. Li et al., 2024).

Building upon above measurement, the extant literature identifies a diverse set of TE determinants. These can be broadly categorized into four dimensions: institutional drivers, such as environmental regulations and market-oriented reforms (Chen et al., 2021; X. Li et al., 2024); technological catalysts, which highlight green innovation and R&D investment (Song et al., 2024a); allocative factors, where the optimal reallocation of resources is crucial (He et al., 2024); and structural shifts, such as industrial upgrading and digitalization (Cui and Cao, 2024; Wu et al., 2023). However, these determinants are often analyzed in isolation, resulting in a fragmented analytical landscape that overlooks the potential for synergistic or conflicting interactions between them. This fragmentation is particularly problematic when confronting the core theoretical challenge in this field.

2.2. The innovation-efficiency paradox and the need for a systemic response

Among the determinants of TE, technological innovation is widely regarded as the core engine for catalyzing a green transition (Song et al., 2024b). However, the relationship between technological innovation and net energy savings is fundamentally constrained by the "rebound effect," a paradox first conceptualized by Jevons (1865) and extensively corroborated in modern energy economics(Gillingham et al., 2016). This effect posits that efficiency gains, by lowering the effective price of energy services, can paradoxically stimulate greater energy consumption, thereby partially or even entirely offsetting anticipated conservation benefits. The persistence of this phenomenon underscores a critical limitation of relying solely on technology-centric strategies (Sorrell, 2009).

This paradox reveals that a successful green transition cannot be anchored in discrete technological advancements alone. Instead, it demands a systemic framework that endogenously integrates innovation with institutional incentives, industrial upgrading, and optimized governance. Such a holistic approach, consistent with theories of socio-technical transitions which emphasize the co-evolution of technology and institutions, is necessary to mitigate or overcome the rebound effect (Geels, 2004).

2.3. New quality productive forces as a systemic framework

The "New Quality Productive Forces" (NPRO) paradigm, an emerging national strategy in China, can be conceptualized as a direct response to this call for a systemic framework. Officially introduced as a strategic initiative to foster high-quality development, NPRO is delineated as an advanced productivity paradigm propelled by revolutionary technological breakthroughs, innovative factor allocation, and profound industrial upgrading. It is characterized by the integration of three novel production factors: a highly skilled labor force (new-type laborer), digital and intelligent infrastructure (advanced means of labor), and strategic emerging industries with data as a core production element (novel objects of labor) (Lin et al., 2024)

NPRO's inherently systemic nature is the pivotal attribute that distinguishes it from prior, fragmented policy approaches. By simultaneously orchestrating advancements in technology, human capital, and industrial structure within a green-oriented framework, the NPRO paradigm is theorized to possess an intrinsic defense against the rebound effect. While early empirical work has linked NPRO to high-quality economic development and carbon mitigation (Dai and Zheng, 2025; Yuan et al., 2025; Zou et al., 2024), its direct, causal impact on TE remains a critical and underexplored area. This theoretical potential provides the foundation upon which the following hypotheses are built.

2.4. Hypotheses development

Based on the preceding theoretical framework, this study proposes the three hypotheses:

First, given NPRO's dual capacity to propel green technological advancement while orchestrating factor reallocation towards sustainable sectors, it is expected to directly curtail energy consumption per unit of output (Li et al., 2025; Gang and Zhao, 2025). The systemic integration of technology, production factors, and industrial structures is hypothesized to furnish a robust defense against the "rebound effect," leading to an unambiguous net positive impact on urban TE (Luo et al., 2025).

NPRO exerts a significant and positive causal effect on urban TE.

Second, the translation of NPRO's potential into tangible TE gains is likely mediated by an enabling institutional environment. The literature suggests that effective policy implementation and governance quality are critical for translating innovation into environmental performance. The strategic impetus of NPRO is expected to compel sub-national governments to implement more stringent low-carbon policies (Chen et al., 2021). Concurrently, its data-driven architecture necessitates a modernization of governance that enhances transparency and resource allocation efficiency (Cui and Cao, 2024; Wu et al., 2023).

The intensification of low-carbon policies and the enhancement of governance effectiveness serve as significant mediating mechanisms in the relationship between NPRO and urban TE.

Third, in an era of deepening regional integration, the effects of NPRO are unlikely to be confined within administrative borders. The core components of NPRO-knowledge, technology, and capital-are inherently networked and diffusive, creating the potential for spatial spillovers (Inkpen and Tsang, 2016). This study theorizes that these spillovers manifest through interconnected channels consistent with regional economic theory, including knowledge diffusion, industrial linkages, and policy learning (Shipan and Volden, 2008).

The development of NPRO in a given city exerts a positive spatial spillover effect on the TE of its economically interconnected neighbors.

3. Method and data

3.1. Model specification

3.1.1. Spatial autocorrelation

To explore the spatial distribution characteristics of NPRO and TE, this study initially employs univariate global spatial autocorrelation. This analysis is conducted to ascertain whether each variable exhibits significant spatial clustering, randomness, or dispersion across the geographical units under investigation (Q. Li et al., 2024). Subsequently, to examine the spatial interplay between these two variables, bivariate global and local spatial autocorrelation analyses are utilized. These techniques allow for a nuanced investigation of the spatial correlation and potential agglomeration patterns, revealing whether regions with high levels of NPRO are spatially associated with regions exhibiting high (or low) levels of TE. The formal specifications for these spatial autocorrelation indices are as follows:

$\:{\text{I}}_{\text{(}\text{x}\text{)}}\text{=}\frac{\text{n}{\sum\:}_{\text{i}\text{=1}}^{\text{n}}\sum\:_{\text{j}\text{=}\text{1}}^{\text{n}}{W}_{\text{ij}}\text{(}{\text{x}}_{\text{i}}\text{−}\stackrel{\text{-}}{\text{x}}\text{)}}{{\sum\:}_{\text{i}\text{=1}}^{\text{n}}{\sum\:}_{\text{j}\text{=1}}^{\text{n}}{\text{W}}_{\text{ij}}{\sum\:}_{\text{j}\text{=1}}^{\text{n}}{\text{(}{\text{x}}_{\text{j}}\text{−}\stackrel{\text{-}}{\text{x}}\text{)}}^{\text{2}}}$

$\:{\text{I}}_{\text{(}\text{x}\text{,}\text{y}\text{)}}\text{=}\frac{{\sum\:}_{\text{i}\text{=1}}^{\text{n}}\sum\:_{\text{j}\text{=}\text{1}}^{\text{n}}{W}_{\text{ij}}\text{(}{\text{x}}_{\text{i}}\text{−}\stackrel{\text{-}}{\text{x}}\text{)(}{\text{y}}_{\text{j}}\text{−}\stackrel{\text{-}}{\text{y}}\text{)}}{{\text{s}}^{\text{2}}{\sum\:}_{\text{i}\text{=1}}^{\text{n}}{\sum\:}_{\text{j}\text{=1}}^{\text{n}}{\text{W}}_{\text{ij}}}$

$\:{\text{I}}_{\text{i}}\text{=}{\text{Z}}_{\text{i}}^{\text{x}}{\sum\:}_{\text{j}\text{=1}}^{\text{n}}{\text{W}}_{\text{ij}}{\text{Z}}_{\text{j}}^{\text{y}}$

Where

$\:{\text{I}}_{\text{(}\text{x}\text{)}}\text{}$

and

$\:{\text{I}}_{\text{(}\text{x}\text{,}\text{y}\text{)}}$

denote the univariate and bivariate global Moran's I, respectively, which are employed to quantify the overall spatial autocorrelation between variables

$\:\text{x}$

and

$\:\text{y}$

. The terms

$\:{\text{x}}_{\text{i}}$

and

$\:{\text{y}}_{\text{j}}$

represent the observed values for the independent and dependent variables in cities

$\:\text{i}$

and

$\:\text{j}$

, respectively;

$\:\text{n}$

is the total number of city samples;

$\:{\text{s}}^{\text{2}}$

denotes the sample variance; and

$\:{\text{w}}_{\text{ij}}$

is the element of the spatial weight matrix. Furthermore,

$\:{\text{I}}_{\text{i}}$

signifies the local spatial association between the independent and dependent variables for city

$\:\text{i}$

, while

$\:{\text{Z}}_{\text{i}}^{\text{x}}$

and

$\:{\text{Z}}_{\text{j}}^{\text{y}}$

represent the standardized values of the new quality productive forces for city

$\:\text{i}$

and energy utilization efficiency for city

$\:\text{j}$

, respectively.

3.1.2. Double Machine Learning

To accurately identify the causal effect of NPRO on TE, this study needs to address potential biases arising from high-dimensional confounding and complex, nonlinear relationships where traditional linear models may be inadequate. We therefore employ the Double Machine Learning (DML) framework, a state-of-the-art approach specifically designed for robust causal inference in such settings (Chernozhukov et al., 2018). DML leverages the predictive power of flexible machine learning algorithms to partial out the effects of a large set of control variables from both the treatment (NPRO) and the outcome (TE).

This is achieved through a process of orthogonalization and cross-fitting. This dual process effectively isolates the residual variation in NPRO that is orthogonal to the complex confounding structure, enabling robust causal inference on our parameter of interest. The estimation procedure is as follows:

Step 1: Specify the main regression model to identify the impact of NPRO on TE:

$\:{\text{TE}}_{\text{it}\text{+1}}\text{=}{\theta}_{\text{0}}{\text{NPRO}}_{\text{it}}\text{+}\text{g}\left({\text{X}}_{\text{it}}\right)\text{+}{\text{U}}_{\text{it}}$

$\:\text{E}\left({\text{U}}_{\text{it}}|{\text{NPRO}}_{\text{it}}\text{,}{\text{X}}_{\text{it}}\right)\text{=0}$

Step 2: Specify an auxiliary regression model to control for endogeneity issues:

$\:{\text{NPRO}}_{\text{it}}\text{=}\text{m}\left({\text{X}}_{\text{it}}\right)\text{+}{\text{V}}_{\text{it}}$

$\:\text{E}\left({\text{V}}_{\text{it}}|{\text{X}}_{\text{it}}\right)\text{=0}$

Where the subscripts

$\:\text{i}$

and

$\:\text{t}$

denote city and year, respectively;

$\:{\text{TE}}_{\text{it}\text{+1}}\text{}$

represents energy utilization efficiency;

$\:{\text{NPRO}}_{\text{it}}$

signifies the development level of the New Quality Productive Forces;

$\:{\theta}_{\text{0}}$

is the coefficient of interest, which assesses the causal impact of the New Quality Productive Forces on energy utilization efficiency;

$\:{\text{X}}_{\text{it}}$

is a vector of high-dimensional control variables;

$\:\text{g}\left({\text{X}}_{\text{it}}\right)$

and

$\:\text{m}\left({\text{X}}_{\text{it}}\right)$

are unknown functional forms estimated via machine learning algorithms;

$\:{\text{U}}_{\text{it}}$

and

$\:{\text{V}}_{\text{it}}$

are the stochastic error terms.

3.1.3. Spatial econometric model

This study employs Spatial Panel Durbin Model to examine the spatial spillover effects of NPRO on TE. The model incorporates spatial lags of both the dependent and independent variables (Feng and Wang, 2020) and is defined as follows:

$\:{\text{Y}}_{\text{it}}\text{=}\text{β}{\text{X}}_{\text{it}}\text{+}\text{ρ}\sum\:_{\text{j}\text{=1}}^{\text{n}}{\text{W}}_{\text{ij}}{\text{Y}}_{\text{jt}}\text{+}\text{γ}\sum\:_{\text{j}\text{=1}}^{\text{n}}{\text{W}}_{\text{ij}}{\text{X}}_{\text{jt}}\text{+}{\text{μ}}_{\text{i}}\text{+}{\text{σ}}_{\text{t}}\text{+}{\text{ε}}_{\text{it}}$

Where

$\:{\text{Y}}_{\text{it}}$

and

$\:{\text{Y}}_{\text{jt}}$

are the observed values of the dependent variable for regions

$\:\text{i}$

and

$\:\text{j}$

, respectively;

$\:{\text{X}}_{\text{it}}$

and

$\:{\text{X}}_{\text{jt}}$

represent the vectors of explanatory variables, which encompass both core explanatory and control variables;

$\:{\text{W}}_{\text{ij}}$

is the element of the spatial weight matrix;

$\:\text{β}$

denotes the coefficient vector for the explanatory variables;

$\:\text{ρ}$

is the spatial autoregressive coefficient of the dependent variable;

$\:\text{γ}$

represents the spatial lag coefficient of the explanatory variables;

$\:{\text{μ}}_{\text{i}}$

and

$\:{\text{σ}}_{\text{t}}$

capture the individual (spatial) and time-fixed effects, respectively; and

$\:{\text{ε}}_{\text{it}}$

is the stochastic error term.

3.2. Variable description

3.2.1. Explanatory variable

The principal explanatory variable in this study is the level of NPRO, which is quantified using a comprehensively constructed composite index. Grounded in the theoretical framework of NPRO, this study establishes a multidimensional evaluation system designed to capture its multifaceted nature. The system encompasses three primary dimensions: new-quality labor, new-quality means of production, and new-quality objects of production(Liu et al., 2025; Zhou et al., 2025), and disaggregated into 15 specific indicators (Table 1).

To aggregate these indicators objectively, this study employs the entropy weight method (EWM), which determines weights based on the inherent informational value within the data, thereby mitigating the subjectivity of expert-based schemes (Jiang et al., 2025). The index construction process begins with data normalization, where all indicators are scaled to ensure comparability after any inverse indicators are converted. Next, the information entropy of each normalized indicator is computed to measure its dispersion; a lower entropy value signifies greater data variability and thus a higher informational weight. Finally, these entropy values are used to derive objective weights, and the composite NPRO index is calculated as the weighted sum of the normalized indicators, providing a comprehensive measure of NPRO.

Table 1

Evaluation index of NPRO.
Dimensionality	Constituent elements	Explanation of indicators and sources	Direction
New quality labor force	The number of employees in emerging industries	The total number of employees of listed companies in strategic emerging industries and future industries is summarized by their registered location to the prefecture-level city, sourced from the annual reports of the enterprises.	་
	Personal capacity of the labor force	Average salary of on-the-job employees (yuan), source: "China Statistical Yearbook"	་
	High quality level of the labor force	The number of regular institutions of higher learning (institutions), source: "China Education Statistical Yearbook"	་
New types of labor objects	Infrastructure	Internet broadband access users (thousand households), source: China Statistical Yearbook	་
	Infrastructure	Total volume of telecommunications services (in billions of yuan), source: China Statistical Yearbook	་
	Future development	The installation density of robots, source: "Executive summary world robotics 2018 industrial robots" (https://www.universal-robots.com)	་
	Ecological environment	Investment in environmental pollution control (billion yuan), source: "China Environmental Statistical Yearbook"	་
		Carbon trading, energy consumption rights trading, and pollution discharge rights trading (in billions of yuan), disclosed and collated from the official websites of each prefecture-level city	་
		Harmless treatment rate of domestic waste (%), source: "China Urban Construction Statistical Yearbook"	་
New quality labor materials	Technology research and development	The proportion of scientific expenditure in local fiscal expenditure, source: "China Statistical Yearbook"	་
	Innovation output	The number of inventions applied for that year was sourced from the State Patent Office	་
	Innovation output	The number of utility model applications filed in that year is sourced from the National Patent Office	་
	Intelligentization	The number of artificial intelligence enterprises is sourced from the Tianyancha platform (https://www.tianyancha.com/)	་
	Technical foundation	The number of green invention patents filed in that year; data sourced from the China National Intellectual Property Administration	་
	Technical foundation	The number of utility-model patents filed in that year; data sourced from the China National Intellectual Property Administration	་
	Data element	Level of data element utilization, measured by the frequency of listed companies’ disclosures of data-asset-related terms. Specifically, we count the occurrence of relevant keywords in corporate reports, aggregate them by the registered city, and then calculate the average frequency at the city level as a proxy indicator. Data from Yuan et al. (2022)	་
	Data element	Whether a city has a data exchange market. If present, the value is coded as 1; if absent, the value is coded as 0.	་

3.2.2. Dependent variable

The dependent variable is urban energy utilization efficiency (TE), which this study construct using a dynamic, two-stage approach (Li and Chen, 2021). First, we employ the super-efficiency Slacks-Based Measure (SBM) model to assess static efficiency levels. This non-radial model is specifically chosen for its ability to handle multiple inputs and outputs while explicitly incorporating undesirable outputs. Subsequently, the Malmquist-Luenberger (ML) productivity index is applied to capture the intertemporal changes, allowing us to construct a dynamic TE index linked to a base year. This framework is operationalized using three inputs: labor (total urban employment), capital (stock estimated via the perpetual inventory method), and energy (total regional consumption). The model further distinguishes between a desirable output (real GDP) and a vector of undesirable outputs, namely industrial wastewater, SO₂, and smoke/dust emissions.

3.2.3. Mechanism variable

To empirically investigate the transmission channels of NPRO's effect on TE, this study examines two core mechanism variables representing distinct policy and governance pathways. The first, low-carbon policy intensity, captures the stringency of municipal environmental regulations. It is constructed as a city-level index via text analysis of government work reports, which systematically quantifies the implementation strength of various low-carbon policy instruments (Wang et al., 2011; Yang et al., 2023). The second mechanism, governance transparency, serves as a proxy for the quality of local governance and is measured by each city's fiscal transparency score, reflecting the openness of public financial information (Jiang and Huang, 2025). Due to data availability, this latter variable is constrained to the 2013–2022 period for mechanism analysis.

3.2.4. Control variable

To isolate the net effect of NPRO on TE and mitigate omitted variable bias, this study incorporates a comprehensive set of city-level control variables, guided by the established literature (Cui and Cao, 2024; Wu et al., 2023). The control variables are specified as follows: 1) Economic development level (lnPGDP): Proxied by the natural logarithm of regional per capita GDP. 2) Population size (lnPEO): Quantified as the natural logarithm of the registered population at the year-end. 3) Foreign direct investment (FDI): Measured as the ratio of actual FDI inflows to the city's GDP. FDI data, originally denominated in U.S. dollars, are converted to RMB using the average annual exchange rate. 4) Industrial structure (STR): Defined as the proportion of the value-added from the tertiary industry relative to the city's total GDP. 5) Government intervention (GOV): Captured by the ratio of general government fiscal expenditure to GDP, reflecting the extent of governmental influence on the local economy. 6) Technological expenditure (TEC): Measured as the share of local government expenditure on science and technology within its total fiscal expenditure, indicating the local commitment to technological advancement. 7) Financial development (FIA): Proxied by the ratio of the year-end balance of loans from financial institutions to GDP. 8) Carbon emissions (lnCO₂): The natural logarithm of total urban carbon dioxide emissions.

To mitigate potential heteroskedasticity and normalize data distributions, variables characterized by large scales or skewed distributions (per capita GDP, population, carbon emissions) are transformed using a natural logarithm. Furthermore, to capture potential non-linear relationships, the quadratic term for each control variable is also included in our model (Zhang and Li, 2023). Finally, the specification is augmented with a two-way fixed effects structure, incorporating both city and year fixed effects. This comprehensive strategy, which combines key covariates, data transformations, and fixed effects, substantially strengthens the reliability of estimates by controlling for a wide range of observable and unobservable confounders (Wang and Zou, 2025).

3.3. Data source

The sample for this study covers 280 prefecture-level and above cities in China from 2010 to 2022 (excluding the Hong Kong, Macao, and Taiwan regions, as well as cities with significant data gaps). Data for each variable are sourced as follows: Input-output data for energy utilization efficiency are primarily from the China Urban Statistical Yearbook and the China Energy Statistical Yearbook (https://www.stats.gov.cn/sj/ndsj/). For the mechanism variables, the low-carbon policy index is sourced from the Peking University Law Database (https://www.pku.edu.cn/), while the fiscal transparency scores are derived from research reports published by Tsinghua University's School (https://www.tsinghua.edu.cn/). Among the control variables, carbon emissions data are obtained from the China Emissions Accounting Database (CEADs), while the remaining data are primarily sourced from the “China Urban Statistical Yearbook”. Minor data gaps were addressed using linear interpolation. Descriptive statistics for all variables are presented in Table 2.

Table 2

Variable descriptive statistics.
Variable	Symbol	Observed value	Mean value	Standard deviation	Minimum value	Maximum value
New quality productivity forces	NPRO	3640	0.342	0.136	0.103	1.177
Energy utilization efficiency	TE	3640	0.052	0.075	0.002	0.618
Economic development level	lnPGDP	3640	1.533	0.580	-0.592	3.066
Population size	lnPEO	3640	5.760	0.914	1.619	8.100
Foreign direct investment	FDI	3640	0.015	0.019	-0.219	0.328
Industrial structure	STR	3640	1.061	0.601	0.109	5.652
Degree of government intervention	GOV	3640	0.198	0.098	0.044	1.485
Level of science and technology expenditure	TEC	3640	0.017	0.018	0.001	0.207
Degree of financial development	FIA	3640	2.545	1.218	0.504	21.297
Degree of carbon emissions	lnCO₂	3640	16.890	0.991	13.982	19.479

4. Result

4.1. Temporal evolution of NPRO and TE

From 2010 to 2022, both NPRO and TE in China exhibited a general upward trend, yet their spatiotemporal evolutionary trajectories were markedly distinct (Fig. 2). The distribution curve for NPRO demonstrates a consistent rightward shift over the study period, indicating a steady enhancement in urban innovation-driven productivity across China. This evolution can be delineated into two phases. During the 2010–2018 period, the distribution's morphology became progressively leptokurtic, evidenced by a sharpening central peak and a narrowing spread, which indicates a phase of rapid growth accompanied by a convergence that narrowed regional disparities. However, post-2018, the distribution's central peak flattened and its width broadened, signifying a deceleration in the overall growth momentum and a re-emergence of widening urban disparities. Concurrently, the gradual extension of the distribution's right tail indicates that a subset of cities has continued to advance into high-level tiers. This dynamic suggests a pronounced "flying geese" pattern, wherein coastal urban agglomerations, particularly those in the east, have emerged as the primary growth poles for innovation.

In contrast, the evolutionary pattern of TE exhibits greater complexity. While its distribution also shifted rightward, signifying an overall improvement in energy efficiency, it transitioned from a unimodal to a bimodal structure. This structural transformation, coupled with a continuously widening peak, points to an intensifying regional polarization in energy utilization efficiency. The emergence of a bimodal shape suggests that Chinese cities are increasingly bifurcating into two distinct clubs: one comprising high-efficiency performers and the other consisting of cities lagging.

Fig. 2

The distribution and evolution of China's NPRO and TE from 2010 to 2022.

4.2. Spatial distribution of NPRO and TE

The spatial distribution of NPRO and TE reveals divergent evolutionary pathways between 2010 and 2022, suggesting distinct underlying drivers and diffusion mechanisms. The spatial pattern of NPRO is characterized by a persistent gradient, with higher values concentrated in the east and lower values in the west (Fig. 3). During the initial phase (2010–2015), high-NPRO clusters gradually consolidated in the three major coastal urban agglomerations: the Yangtze River Delta, the Pearl River Delta, and the Beijing-Tianjin-Hebei region, while high-value areas in central and western China remained sporadic. From 2015 to 2020, a notable inland radiation effect emerged, with these core clusters beginning to diffuse towards the interior. This was particularly evident in the rapid development of central provincial capitals, which contributed to a temporary narrowing of regional disparities. However, by 2022, a slight contraction of these high-value zones was observed, a phenomenon potentially attributable to external economic shocks and the lingering impacts of the COVID-19 pandemic. Overall, the spatial configuration of NPRO demonstrates substantial stability and path dependence, underscoring the critical role of pre-existing innovation ecosystems and robust industrial foundations in its development.

Fig. 3

Evolution of spatial distribution pattern of urban NPRO level in China in 2010, 2015, 2020 and 2022.

The spatial evolution of TE exhibits a more dynamic and pronounced diffusion pattern (Fig. 4). In 2010, high-TE cities were relatively scattered, primarily located in the Northeast and along the eastern coast. By 2015, distinct high-efficiency corridors had formed along the coastal belts of the Yangtze and Pearl River Deltas. This diffusionary trend continued to penetrate inland, and by 2020, key central cities such as Wuhan and Changsha had emerged as new growth nodes for energy efficiency. By 2022, the spatial structure had evolved into a multi-polar pattern, with high-efficiency zones coexisting in both eastern and central regions. Notably, the Shandong Peninsula and the Middle Yangtze River Urban Agglomeration have solidified their positions as emerging high-efficiency poles. In summary, the spatial evolution of TE follows a clear trajectory from points to corridors and ultimately to multi-polar structure. This dynamic suggests that TE improvements are more spatially fluid and can be effectively transmitted across regions through mechanisms such as policy diffusion, inter-regional technology transfer, and coordinated industrial upgrading.

Fig. 4

Evolution of spatial distribution pattern of urban TE in China in 2010, 2015, 2020 and 2022.

4.3. Spatial autocorrelation of NPRO and TE

In the spatial interdependence between NPRO and TE, univariate analysis reveals that the global Moran's I for both NPRO and TE remains significantly positive at the 1% level throughout the study period, providing robust evidence of strong and persistent spatial clustering at the urban scale (Table 3). Temporally, these indices trace a distinct inverted U-shaped trajectory, with agglomeration intensity peaking around 2018 before receding. This dynamic arc suggests that Chinese cities experienced a phase of spatial convergence followed by divergence in both innovation capacity and energy efficiency, a later-stage divergence potentially attributable to macroeconomic shocks like the COVID-19 pandemic and the solidification of regional endowment disparities.

Bivariate analysis corroborates and deepens this finding. The bivariate Moran's I for NPRO and TE is also consistently positive and significant, mirroring the univariate trend of an "intensify-then-decelerate" pattern. This confirms a significant synergistic clustering effect, where innovation-driven productivity gains in one city are spatially co-located with enhanced energy efficiency in its neighbors. While this affirms that NPRO can indeed catalyze regional TE improvements, the non-linear trajectory of this linkage-characterized by periodic fluctuations rather than sustained linear strengthening-underscores the complex and non-stationary nature of regional collaborative development.

Table 3

Global Moran's I of NPRO and TE.
Year	Univariate Moran's I		Bivariate Moran's I
Year	NPRO	TE	NPRO and TE
2010	0.079^***	0.414^***	0.160^***
2012	0.099^***	0.463^***	0.189^***
2014	0.122^***	0.407^***	0.175^***
2016	0.158^***	0.402^***	0.182^***
2018	0.197^***	0.470^***	0.231^***
2020	0.194^***	0.474^***	0.220^***
2022	0.157^***	0.444^***	0.185^***
Note: * and represent significant levels of 1% and 5% respectively.

To further dissect the sources of these spatial disparities, we employ bivariate Local Indicators of Spatial Association (LISA) to map the local clustering patterns in Fig. 5. The analysis indicate that High-High (H-H) clusters, representing areas where both NPRO and TE are high, have been persistently concentrated in the eastern coastal regions. Initially centered on the Shandong and the Yangtze River Delta in 2010, these clusters progressively expanded and became contiguous, extending to the Beijing-Tianjin-Hebei region by 2020. This pattern underscores the pivotal role of coastal urban agglomerations as the core drivers of synergy between innovation and energy utilization efficiency. Although the H-H region experienced a slight contraction by 2022, the Beijing-Tianjin-Hebei area and the northern Shandong maintained their high-value status, indicating the formation of resilient growth poles with strong path dependency.

Fig. 5

Bivariate LISA clustering of China's NPRO and TE in 2010, 2015, 2020 and 2022.

In contrast, the northwest and southwest regions remained entrenched in Low-Low (L-L) patterns, exhibiting a classic "low-level equilibrium trap" or "lock-in" effect. Notably, the Low-High (L-H) clusters-cities with low NPRO but high TE-predominantly appear on the periphery of the eastern H-H zones. This spatial configuration reveals a complex interplay between the "siphon effect" of innovation from the core hubs and a potential "trickle-down effect" of energy utilization efficiency benefits (Chen and Wang, 2022; Ma et al., 2024). Overall, the persistent coexistence of H-H and L-L clusters indicates significant spatial inertia in China's urban development. Meanwhile, the emergence of L-H clusters in these peripheral zones highlights the structural tensions arising from unbalanced regional development.

4.4. Empirical analysis

4.4.1. Benchmark regression

To ascertain the net causal effect of NPRO on TE, this study employs the DML methodology within a partially linear model framework. The Random Forest algorithm is utilized as the base learner to flexibly capture the nonlinear relationships between the outcome (TE), the treatment (NPRO), and a comprehensive set of control variables. To mitigate the risk of overfitting and ensure the validity of the causal inference, the four fold cross-fitting procedure is implemented and presents the estimation results from a series of model specifications. These specifications systematically and sequentially incorporate a set of linear control variables, city and year fixed effects, and quadratic terms for the control variables to ensure the robustness of the findings.

The benchmark results, presented in Table 4, reveal a positive and statistically significant coefficient for NPRO at the 1% level. This finding remains stable across all specifications, including the comprehensive model with full fixed effects and quadratic controls (Column 4). In terms of economic magnitude, the result suggests that a one-standard-deviation increase in the NPRO index is associated with a 0.644 standard-deviation improvement in TE, holding all other factors constant. This provides empirical support for Hypothesis H1, confirming that the development of new productive forces is a significant driver of urban energy efficiency.

Table 4

Benchmark regression results: the impact of NPRO on TE.
Variables	(1)	(2)	(3)	(4)
Variables	TE	TE	TE	TE
NPRO	0.384***	0.630***	0.642***	0.644***
NPRO	(0.074)	(0.119)	(0.121)	(0.122)
Constant	-0.002	0.000	0.001	0.000
Constant	(0.001)	(0.001)	(0.001)	(0.001)
Observations	3640	3640	3640	3640
Control the primary term of the variable	Yes	Yes	Yes	Yes
Control the quadratic term of the variable	No	No	No	Yes
Year FE	No	No	Yes	Yes
City FE	No	Yes	Yes	Yes
Note: *, , and * indicate 1%, 5%, and 10% significance levels, respectively, with robust standard errors in parentheses. The following tables are identical.

4.4.2. Robustness and endogeneity

To corroborate the reliability and validity of benchmark findings, this study conduct a comprehensive battery of robustness checks. These tests are designed to address potential concerns related to model specification, variable measurement, sample selection, and endogeneity. The results are summarized in Tables 5–6 and Fig. 6.

(1)

To address model specification concerns, this study assess the sensitivity of results to the choice of the DML base learner. The benchmark model is re-estimated by substituting the Random Forest algorithm with alternative machine learning techniques, namely Gradient Boosting Machines (Gradboost), Neural Networks (Nnet), Lasso regression (Lassocv), and Support Vector Machines (SVM). Furthermore, this study employs a Generalized Random Forest (GRF) model, which is specifically designed for robust causal inference. Across all alternative specifications, the coefficient of NPRO remains positive and statistically significant, confirming that our core finding is not contingent upon a specific algorithmic choice.

(2)

To mitigate potential variable measurement issues, this study re-evaluates the dependent variable and refine the model controls. TE is recalculated using a traditional radial Data Envelopment Analysis (DEA) model (the CCR model). Concurrently, we re-estimate our main specification with more stringent "province×year" fixed effects to absorb any unobserved, time-varying provincial-level shocks. The coefficient of NPRO retains its significance and positive sign, indicating that our conclusion is insensitive to both the specific measurement of the outcome variable and the method of controlling for unobserved heterogeneity.

(3)

To ensure that results are not driven by potential outliers or influential subsamples, this study performs two additional checks. This study excludes municipalities directly under the central government, which may possess unique political and economic characteristics. Additionally, this study apply 1% winsorization to all continuous variables to curtail the influence of extreme values. The promotional effect of NPRO on TE remains highly significant in both tests, suggesting that our findings are not skewed by a few megacities or extreme data points.

(4)

To more formally address endogeneity concerns arising from potential reverse causality or omitted variables, this study employs two distinct instrumental variable (IV) strategies within a two-stage least squares (2SLS) framework. The first strategy utilizes the lagged values of the explanatory variable as internal instruments. The second strategy employs the heteroscedasticity-based identification method proposed by Lewbel (2012), which constructs valid instruments from the model's residuals. The results from both IV approaches consistently support a positive and statistically significant causal effect of NPRO on TE. Collectively, this battery of robustness checks and endogeneity tests consistently reinforces the validity of core finding that new productive forces significantly enhance urban energy utilization efficiency.

Table 5

The robustness test.
Variables	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Gradboost	Nnet	Lassocv	Svm	K = 3	K = 8
NPRO	0.414***	1.344***	1.588***	0.473***	0.656***	0.640***
NPRO	(0.082)	(0.230)	(0.297)	(0.058)	(0.105)	(0.117)
Constant	0.000	0.011***	-0.000	0.025***	0.000	0.001
Constant	(0.002)	(0.002)	(0.001)	(0.004)	(0.001)	(0.001)
Observations	3640	3640	3640	3640	3640	3640
Control the primary term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Control the quadratic term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
Province # Time	No	No	No	No	No	No

(Continued Table).

Fig. 6

The distribution of the disposal effect of NPRO on TE under different number of decision trees.

Variables	(1)	(2)
Variables	IV = L.NPRO	IV = Lewbel
NPRO	0.657***	1.202***
NPRO	(0.147)	(0.267)
Constant	0.002	0.000
Constant	(0.001)	(0.001)
Observations	3360	3640
Control the primary term of the variable	Yes	Yes
Control the quadratic term of the variable	Yes	Yes
Year FE	Yes	Yes
City FE	Yes	Yes

4.5. Mechanism analysis

Building on the established positive effect of NPRO on urban TE, this section investigates the underlying mechanisms. The two-step mediation analysis framework proposed by Jiang (2022) is employed to empirically test two specific transmission pathways: the intensification of low-carbon policies and the improvement of transparent governance.

(1)

The analysis first reveals that low-carbon policy intensity serves as a key transmission pathway. As shown in Column (1) of Table 7, NPRO has a positive effect on the intensity of these policies. This finding suggests that the development of NPRO enhances the capacity and motivation of local governments to implement more stringent environmental regulations. Consistent with the literature, such policies are instrumental in boosting TE by curbing the expansion of energy-intensive industries, fostering the growth of clean and service-oriented sectors, and leveraging digital tools for more precise policy implementation (Bai and Liu, 2024).,

(2)

Transparent governance is identified as the second crucial mediating variable. The results in Column (2) indicate that NPRO significantly enhances the effectiveness of transparent governance. This implies that the advancement of NPRO both necessitates and propels greater openness and accountability in local administration. As suggested by prior research, improved governance-particularly fiscal transparency-provides essential institutional safeguards for green development. It mitigates resource misallocation and rent-seeking by strengthening budgetary constraints (Mo and Zhang, 2023) and helps optimize public expenditure by channeling funds toward energy conservation and green innovation.

Overall, the confirmation of these two pathways provides robust empirical support for the second hypothesis (H2). The findings elucidate how the development of NPRO translates into tangible gains in TE through the channels of strengthened environmental policy and enhanced governance.

Table 7

The influence mechanism test of NPRO on TE.
Variables	(1)	(2)
Variables	Low-carbon policy intensity	Government fiscal transparency
NPRO	9.798** (4.571)	47.736*** (13.623)
Constant	0.044 (0.057)	-0.150 (0.273)
Observations	3640	3640
Control the primary term of the variable	Yes	Yes
Control the quadratic term of the variable	Yes	Yes
Year FE	Yes	Yes
City FE	Yes	Yes

5.6 Heterogeneity analysis

Recognizing that the impact of NPRO on TE may vary across different regional and institutional contexts, this study conducts a comprehensive heterogeneity analysis. We explore these differential effects across four key dimensions: geographical location, industrial structure, environmental regulation, and government intervention. The analysis employs both subsample regressions and interaction term models to ensure the robustness of our findings, with the results presented in Tables 8 and 9.

(1)

Geographical location. The sample is bifurcated into eastern and central-western regions to account for China's well-documented regional development disparities. The results indicate that the positive effect of NPRO on TE is statistically significant and substantially more pronounced in the eastern cities. This regional disparity can likely be attributed to the eastern region's more advanced innovation ecosystems, higher-quality human capital, and more efficient factor markets (Huang et al., 2025; Shuai et al., 2025). These conditions are benefit for the rapid conversion of technological advancements inherent in NPRO into tangible improvements in TE.

(2)

Industrial structure. To examine the role of industrial composition, cities are categorized based on the median share of their tertiary sector in GDP. The analysis reveals that the positive impact of NPRO is markedly stronger in cities with a more advanced industrial structure (i.e., a higher share of the tertiary sector). This finding is plausible, as modern service industries and high-tech manufacturing sectors, which constitute NPRO, possess a greater intrinsic capacity for process innovation and digital transformation. These sectors are better positioned to adopt energy-saving technologies and optimize production processes, enabling more direct and significant reductions in energy consumption per unit of output (Tao et al., 2024).

(3)

Environmental intervention. This study investigates whether the stringency of environmental policy moderates the effect of NPRO by grouping cities based on their designation as “key environmental protection cities”. The findings show that the promotional effect of NPRO on TE is significantly more potent in cities subject to stricter environmental regulations. This suggests that a stringent regulatory framework acts as a catalyst, compelling firms to internalize environmental externalities and actively seek innovative solutions (Cheng, 2025). Consequently, it amplifies the positive impacts of NPRO on TE, thereby accelerating the regional green transition.

(4)

Government intervention. To assess the interplay between market mechanisms and government action, cities are grouped by their level of government intervention, measured by the ratio of fiscal expenditure to GDP. The analysis reveals that NPRO's positive effect is stronger and more statistically significant in cities with lower levels of government intervention. This finding suggests that a more market-oriented environment facilitates a more efficient allocation of resources, allowing the dividends from NPRO-driven innovation to translate more effectively into TE gains. Conversely, excessive government intervention may distort factor prices and lead to resource misallocation, thereby undermining the efficiency-enhancing effects of NPRO. This result does not negate the importance of an “effective government” in correcting market failures, but rather highlights that over-intervention can create institutional frictions that impede the efficient diffusion of technology and the optimal reallocation of production factors (Chen et al., 2023).

Table 8

Heterogeneity analysis (geographical location and industrial structure).
Variables	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Eastern region	Central and western regions	Interaction item	High industrial structure	Low industrial structure	Interaction item
NPRO	0.716***	0.401***		0.461***	0.251***
NPRO	(0.179)	(0.176)		(0.120)	(0.112)
NPRO×Area			0.392*** (0.112)
NPRO×Str						0.559*** (0.120)
Constant	0.001	-0.000	0.000	-0.001	0.000	0.000
Constant	(0.003)	(0.001)	(0.001)	(0.002)	(0.001)	(0.001)
Observations	1118	2522	3640	1820	1820	3640
Control the primary term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Control the quadratic term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes

Table 9

Heterogeneity analysis (environmental protection city type and degree of government intervention).
Variables	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Key city for environmental protection	Non-key city for environmental protection	Interaction item	Low government intervention	High government intervention	Interaction item
NPRO	0.477***	0.469***		0.453***	0.075***
NPRO	(0.108)	(0.311)		(0.105)	(0.239)
NPRO×City			0.500*** (0.117)
NPRO×Gov						0.365*** (0.090)
Constant	0.000	0.001	0.001	0.001	0.001	0.001
Constant	(0.002)	(0.001)	(0.001)	(0.002)	(0.002)	(0.001)
Observations	1443	2197	3640	1820	1820	3640
Control the primary term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Control the quadratic term of the variable	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes

4.6. Spatial spillover effect analysis

The global Moran’s I for both NPRO and TE was consistently significant at the 1% level from 2010 to 2022, confirming the presence of spatial autocorrelation and justifying the application of spatial econometric models. Subsequent model specification tests, as presented in Table 10, revealed that the Likelihood Ratio (LR) and Wald tests, alongside the Lagrange Multiplier (LM) and robust LM tests, consistently favored the Spatial Durbin Model (SDM) over the Spatial Autoregressive (SAR) and Spatial Error (SEM) models. Accordingly, this study employs a two-way fixed-effects SDM, with the spatial spillover effects decomposed using the partial derivative method proposed by LeSage (2008) (Table 11).

Table 10

Spatial econometric model tests.
Inspection	Statistical value	P value
LM-Test
Spatial error:
Moran's I	2.296	0.022
Lagrange multiplier	5.035	0.025
Robust Lagrange multiplier	22.786	0.000
Spatial lag:
Lagrange multiplier	16.099	0.000
Robust Lagrange multiplier	33.849	0.000
LR-Test
SDM-SAR	(chi2)59.43	0.000
SDM-SEM	(chi2)60.39	0.000
Wald-Test	(chi2)76.77	0.000

The spatial decomposition confirms NPRO as a key driver of Total-factor TE, exhibiting significant positive direct, indirect, and total effects. This dual impact-enhancing local efficiency through technological superiority while simultaneously generating positive spillovers via inter-regional linkages-provides robust support for Hypothesis three. This core relationship is, however, embedded within a complex geography of development. The analysis of control variables reveals competing spatial dynamics: while factors like Foreign Direct Investment (FDI) foster regional synergy through positive spillovers, the economic growth of core cities can create the siphon effect, benefiting local TE at the expense of neighboring areas. This underscores that the overall impact of NPRO is shaped by the interplay between cooperative networks and competitive agglomeration forces.

Table 11

The spatial effect decomposition of NPRO on TE
	(1)	(2)	(3)
Variables	Total effect	Direct effect	Indirect effect
NPRO	2.179***	1.252***	0.927***
NPRO	(0.286)	(0.106)	(0.312)
GOV	0.508***	-0.066*	0.574***
GOV	(0.122)	(0.035)	(0.118)
TEC	-0.872*	-0.534***	-0.338
TEC	(0.494)	(0.155)	(0.449)
InPEO	-0.123	-0.182***	0.059
InPEO	(0.110)	(0.030)	(0.105)
FIA	0.008	0.012***	-0.004
FIA	(0.010)	(0.003)	(0.009)
FDI	1.081***	0.003	1.078***
FDI	(0.357)	(0.093)	(0.337)
STR	-0.053**	-0.020***	-0.033
STR	(0.021)	(0.006)	(0.020)
InCO₂	-0.035	-0.032***	-0.003
InCO₂	(0.042)	(0.011)	(0.039)
InPGDP	-0.039	0.048***	-0.086***
InPGDP	(0.029)	(0.011)	(0.028)

5. Discussion

5.1. NPRO as a paradigm for sustainable transition

The central finding-that NPRO significantly enhances TE-offers a novel empirical solution to the long-standing "rebound effect" paradox, which posits that isolated technological efficiency gains are often offset by induced consumption Sorrell, (2009). The results suggest that NPRO represents a paradigm shift beyond discrete technological improvements. Its inherently systemic nature-endogenously integrating technological breakthroughs with institutional incentives and optimized factor allocation-appears to be the key mechanism that mitigates or counteracts this paradox. This perspective deepens the Porter Hypothesis (Porter and Linde, 1995) by elevating it from a firm-level response to specific regulations to a macro-level outcome of a proactive, state-led systemic productivity framework. The research demonstrates that a coherent national strategy, by bundling innovation policy with environmental goals, can systemically foster a virtuous cycle between innovation, regulation, and environmental performance.

5.2. Spatial stickiness of innovation and diffusion of its benefits

A particularly striking finding is the asynchronous geography between the generation of NPRO and the diffusion of its TE benefits. On one hand, NPRO generation exhibits strong "geographic stickiness," concentrating within the mature innovation ecosystems of China's eastern coastal hubs. This spatial clustering aligns with theories of "club convergence" (Quah 1996), challenging traditional models of uniform regional economic convergence and affirming the persistence of an innovation core. On the other hand, the enhancement of TE exhibits a pronounced diffusion pattern, with clear spatial spillovers to inland cities. This finding does not invalidate the classic "core-periphery" model (Krugman, 2009) but adds a crucial dynamic layer: while innovation generation remains spatially concentrated, its dividends in TE can be effectively transmitted to the periphery. The positive spatial spillovers identified in the econometric models suggest that dense economic linkages and institutional coupling-rather than mere geographic proximity-are the critical transmission channels (Rodríguez-Pose and Crescenzi, 2008). This transforms the periphery from a passive recipient into an active participant in a networked system, highlighting that targeted inter-regional collaboration can effectively disseminate the benefits of innovation, even when the innovation engines themselves remain geographically clustered.

This perspective both aligns with and extends Porter’s Hypothesis (Porter and Linde, 1995). While Porter argued that well-designed environmental regulations can spur firm-level innovation, this study findings highlight the pivotal role of a proactive, state-led productivity framework like NPRO. NPRO appears to function as a powerful macro-level catalyst, translating innovative potential into tangible, widespread energy savings. This study extends Porter’s original hypothesis from firm level to a national scale. Unlike single policy studies (e.g., Lanoie et al., 2011, on pollution charges), this research shows that a coherent national strategy-bundling innovation policy with environmental goals-can foster a systemic virtuous cycle of innovation, regulation, and environmental performance.

5.3. Governance and innovation as primary drivers of green transition

The study finds that technological innovation and high-quality governance serve as the primary transmission mechanisms for NPRO's effect on TE, with their explanatory power overshadowing that of traditional industrial structure upgrading. This suggests a complementary, and potentially faster, pathway to environmental improvement than the one conventionally described in the Environmental Kuznets Curve (EKC) literature (Grossman and Krueger, 1991). The EKC hypothesis has been critiqued for its deterministic view that environmental quality is a passive, and often slow, byproduct of economic development (Stern, 2004). The analysis, however, indicates that a systemic framework like NPRO may enable an economic "leapfrogging" effect. Systemic and disruptive innovations, such as digitalization and AI, can directly transform energy use patterns within existing industries, aligning with the concept of "green growth" (Acemoglu et al., 2012). Furthermore, the mechanism analysis reveals that high-quality governance acts as a critical "accelerator" in this process. Enhanced fiscal transparency and effective low-carbon policies do not merely support innovation; they amplify its conversion into tangible energy savings. This synthesizes the arguments for an "effective government" and an "efficient market" (Chen, 2018), demonstrating that their synergistic combination constitutes the optimal institutional foundation for a green transition.

6. Conclusion and policy recommendations

6.1. Conclusion

This study provides robust causal evidence on the role of China's NPRO in enhancing urban TE, revealing a complex interplay of institutional and spatial dynamics. Including: (1) The spatial analysis reveals a pattern of asynchronous evolution. NPRO generation exhibits strong geographic "stickiness," concentrating in eastern coastal hubs. In contrast, the resulting TE benefits display a broad diffusion pattern inland, a dynamic driven by significant positive spatial spillovers to economically interconnected neighbors. This spatial decoupling of innovation generation from its efficiency gains underscores a complex geographic interplay. (2) The causal evidence confirms that NPRO significantly enhances urban energy efficiency. The benchmark regression, employing a Double Machine Learning approach, indicates that a one-standard-deviation increase in the NPRO index is associated with a 0.644-standard-deviation improvement in TE. (3) This positive effect is primarily transmitted through institutional channels. The mediation analysis demonstrates that NPRO improves TE by strengthening the intensity of local low-carbon policies and enhancing governance transparency, highlighting the critical role of institutional arrangements. (4) The impact of NPRO is context-dependent and heterogeneous. The efficiency-enhancing effects are most pronounced in cities that prioritize environmental protection, possess advanced industrial structures, are located in the eastern region, or have a lower degree of government intervention.

6.2. Policy recommendation

Based on the above findings, this study proposes the following policy recommendations to maximize the potential of NPRO in driving urban green transitions:

(1)

Adopt a differentiated, network-based regional strategy. The asynchronous geography of NPRO's innovation and benefits necessitates a shift from uniform national policies to a dual-pronged regional approach. For eastern innovation hubs, policy should focus on accelerating frontier green technology development to amplify their role as innovation engines. For central and western regions, the strategic priority must be to enhance their absorptive capacity. This requires targeted measures-such as technology transfer funds and business environment optimization-to internalize spillovers. Critically, because these spillovers are transmitted through economic networks, networked governance mechanisms like cross-city "Green Collaboration Alliances" should be established to facilitate the flow of capital, talent, and best practices beyond rigid administrative borders.

(2)

Redefine governance from direct intervention to strategic market enablement. This study finding that NPRO's impact is strongest where an "effective government" and an "efficient market" coexist calls for a strategic redefinition of the state's role. Rather than direct intervention, which can be counterproductive, the government should focus on market enablement. Its primary functions should be to provide stable, long-term low-carbon policy signals and enhance fiscal transparency to reduce investment uncertainty. This approach-clarifying the boundary between state and market while avoiding both inaction and overreach-creates the optimal institutional environment for the market to efficiently allocate resources and fully realize NPRO's potential in driving the green transition.

6.3. Limitations and future research

While this study provides valuable insights, its limitations delineate clear avenues for future research. First, this study utilizes a composite index for NPRO; future research could disaggregate this framework into its constituent components-such as intelligent manufacturing, the digital economy, and green finance-to investigate their distinct causal pathways and provide more granular policy guidance. Second, this study city-level analysis establishes a macro-level causal link; future work using firm-level micro-data is needed to uncover the underlying behavioral mechanisms, particularly how heterogeneous firms adjust their R&D, technology adoption, and production processes in response to the NPRO framework. Finally, as the findings are situated within China's unique institutional setting, this external validity warrants investigation. Cross-country comparative studies are essential to test the generalizability of our framework and to explore how different institutional and economic contexts moderate the relationship between systemic productivity strategies and energy efficiency.

Data Availability

The datasets generated and/or analyzed during the current study are not publicly available at this time as they form an integral part of ongoing research investigations. However, the data are available from the corresponding author (Zhongwu Zhang ) upon reasonable request.

References

Acemoglu, D., Aghion, P., Bursztyn, L. & Hemous, D. The Environment and Directed Technical Change. Am. Econ. Rev. 102, 131–166. https://doi.org/10.1257/aer.102.1.131 (2012).

Adetomi Adewnmi, K. A., Olu-lawal, C. E., Okoli, F. O., Usman, G. S. & Usiagu Sustainable energy solutions and climate change: A policy review of emerging trends and global responses. World J. Adv. Res. Rev. 21, 408–420. https://doi.org/10.30574/wjarr.2024.21.2.0474 (2023).

Bai, D. & Liu, X. Can green fiscal policy improve energy efficiency? Evidence from comprehensive demonstration cities for fiscal policies to promote energy conservation and emission reduction. Contemp. Finance Econ. 1–16. https://doi.org/10.13676/j.cnki.cn36-1030/f.20241022.003 (2024).

Chen, L. & Wang, K. The spatial spillover effect of low-carbon city pilot scheme on green efficiency in China’s cities: Evidence from a quasi-natural experiment. Energy Econ. 110, 106018. https://doi.org/10.1016/j.eneco.2022.106018 (2022).

Chen, Y. Mature Market Economy-A Combination of Effective Government and Efficient Market. Arch. Bus. Adm. Manag. 1 https://doi.org/10.29011/2642-3243.100015 (2018).

Chen, Y., Zhang, M., Wang, C., Lin, X. & Zhang, Z. High-Tech Industrial Agglomeration, Government Intervention and Regional Energy Efficiency: Based on the Perspective of the Spatial Spillover Effect and Panel Threshold Effect. Sustainability 15, 6295. https://doi.org/10.3390/su15076295 (2023).

Chen, Z., Song, P. & Wang, B. Carbon emissions trading scheme, energy efficiency and rebound effect – Evidence from China’s provincial data. Energy Policy. 157, 112507. https://doi.org/10.1016/j.enpol.2021.112507 (2021).

Cheng, Z. Can environmental information disclosure promote urban carbon emission reduction? Quasi-experimental evidence from China. J. Clean. Prod. 489, 144698. https://doi.org/10.1016/j.jclepro.2025.144698 (2025).

Chernozhukov, V. et al. Double/debiased machine learning for treatment and structural parameters. Econometrics J. 21, C1–C68. https://doi.org/10.1111/ectj.12097 (2018).

Chin, T., Li, Z., Huang, L. & Li, X. How artificial intelligence promotes new quality productive forces of firms: A dynamic capability view. Technol. Forecast. Soc. Chang. 216, 124128. https://doi.org/10.1016/j.techfore.2025.124128 (2025).

Cui, H. & Cao, Y. Do smart cities improve energy efficiency? A test of spatial effects and mechanisms. Sustainable Cities Soc. 101, 105124. https://doi.org/10.1016/j.scs.2023.105124 (2024).

Dai, D. & Zheng, Y. The New Quality Productive Force, Science and Technology Innovation, and Optimization of Industrial Structure. Sustainability 17, 4439. https://doi.org/10.3390/su17104439 (2025).

Dhakal, S. Urban energy use and carbon emissions from cities in China and policy implications. Energy Policy. 37, 4208–4219. https://doi.org/10.1016/j.enpol.2009.05.020 (2009).

Feng, Y. & Wang, X. Effects of urban sprawl on haze pollution in China based on dynamic spatial Durbin model during 2003–2016. J. Clean. Prod. 242, 118368. https://doi.org/10.1016/j.jclepro.2019.118368 (2020).

Gang, H. & Zhao, F. Research on the coupling and harmonization degree of new productive force and high-quality economic development. Finance Res. Lett. 84, 107684. https://doi.org/10.1016/j.frl.2025.107684 (2025).

Geels, F. W. From sectoral systems of innovation to socio-technical systems. Res. Policy. 33, 897–920. https://doi.org/10.1016/j.respol.2004.01.015 (2004).

Gillingham, K., Rapson, D. & Wagner, G. The Rebound Effect and Energy Efficiency Policy. Rev. Environ. Econ. Policy. 10, 68–88. https://doi.org/10.1093/reep/rev017 (2016).

Grossman, G. & Krueger, A. Environmental Impacts of a North American Free Trade Agreement (No. w3914) (National Bureau of Economic Research, 1991).

He, W., Li, W., Wang, C., Wang, S. & Yang, Y. Does energy resource misallocation affect energy utilization efficiency? Evidence from Chinese provincial panel data. Energy 288, 129544. https://doi.org/10.1016/j.energy.2023.129544 (2024).

Hu, J. L. & Wang, S. C. Total-factor energy efficiency of regions in China. Energy Policy. 34, 3206–3217. https://doi.org/10.1016/j.enpol.2005.06.015 (2006).

Huang, Z., Zhang, Q., Zheng, Y., Huang, F. & Guo, R. Heterogeneous impact of climate policy uncertainty on regional economic development in China: Evidence from double machine learning. J. Environ. Manage. 393, 127046. https://doi.org/10.1016/j.jenvman.2025.127046 (2025).

Inkpen, A. C. & Tsang, E. W. K. Reflections on the 2015 Decade Award—Social Capital, Networks, and Knowledge Transfer: An Emergent Stream of Research. AMR 41, 573–588. https://doi.org/10.5465/amr.2016.0140 (2016).

Jevons, W. S. 1865. On the Variation of Prices and the Value of the Currency since 1782. J. Stat. Soc. Lond. 28, 294. https://doi.org/10.2307/2338419

Jiang, L., Xuan, Y. & Zhang, K. Enhancing sustainable development in Chinese cities: The impact of artificial intelligence technology and spatio-temporal evolution analysis. Sustainable Cities Soc. 131, 106747. https://doi.org/10.1016/j.scs.2025.106747 (2025).

Jiang, N. & Huang, J. Can transparency governance promote interregional coordinated carbon emission reduction? A new perspective from China provincial transparency index. Energy Strategy Reviews. 58, 101649. https://doi.org/10.1016/j.esr.2025.101649 (2025).

Jiang, T. Mediating effects and moderating effects in causal inference. China Industrial Econ. 100–120. https://doi.org/10.19581/j.cnki.ciejournal.2022.05.005 (2022).

Krugman, P. The Increasing Returns Revolution in Trade and Geography. Am. Econ. Rev. 99, 561–571. https://doi.org/10.1257/aer.99.3.561 (2009).

Lanoie, P., Laurent-Lucchetti, J., Johnstone, N. & Ambec, S. Environmental Policy, Innovation and Performance: New Insights on the Porter Hypothesis. Econ. Manag Strategy. 20, 803–842. https://doi.org/10.1111/j.1530-9134.2011.00301.x (2011).

LeSage, J. P. An Introduction to Spatial Econometrics. rei 19–44. (2008). https://doi.org/10.4000/rei.3887

Li, Q., Zhang, J., Feng, Y., Sun, R. & Hu, J. Towards a high-energy efficiency world: Assessing the impact of artificial intelligence on urban energy efficiency. J. Clean. Prod. 461, 142593. https://doi.org/10.1016/j.jclepro.2024.142593 (2024).

Li, X. et al. Practical effects of carbon emissions trading system on energy efficiency. Sci. Rep. 14, 279. https://doi.org/10.1038/s41598-023-50621-3 (2024).

Li, X., Li, S., Wu, Q. & Cheng, J. Influence Mechanism of Digital Economy on Urban Green Development Efficiency: A Perspective on New Quality Productive Forces. Sustainability 17, 4539. https://doi.org/10.3390/su17104539 (2025).

Li, Y. & Chen, Y. Development of an SBM-ML model for the measurement of green total factor productivity: The case of pearl river delta urban agglomeration. Renew. Sustain. Energy Rev. 145, 111131. https://doi.org/10.1016/j.rser.2021.111131 (2021).

Lin, L., Gu, T. & Shi, Y. The Influence of New Quality Productive Forces on High-Quality Agricultural Development in China: Mechanisms and Empirical Testing. Agriculture 14, 1022. https://doi.org/10.3390/agriculture14071022 (2024).

Liu, S., Zeng, Z. & Kong, Y. The Impact of New Quality Productive Forces on Common Prosperity: Evidence from Chinese Cities. Sustainability 17, 7703. https://doi.org/10.3390/su17177703 (2025).

Luo, Z. et al. Green Revolution vs. Digital Leap: Decoding the Impact of Environmental Regulation on New Quality Productive Forces in China’s Yangtze River Basin. Sustainability 17, 7216. https://doi.org/10.3390/su17167216 (2025).

Ma, M., Wang, Y. & Ke, S. Spatial spillover effect and driving factors of urban carbon emissions in the Yellow River Basin using nighttime light data. Sci. Rep. 14, 19672. https://doi.org/10.1038/s41598-024-70520-5 (2024).

Mo, L. & Zhang, X. Legalization of government budget, fiscal transparency, and local fiscal efficiency: Evidences based on the implementation of the new budget law. Contemp. Finance Econ. 39–51. https://doi.org/10.13676/j.cnki.cn36-1030/f.2023.05.004 (2023).

Porter, M. E., Linde, C. & van der Toward a New Conception of the Environment-Competitiveness Relationship. J. Economic Perspect. 9, 97–118. https://doi.org/10.1257/jep.9.4.97 (1995).

Prabhakar, A. Sustainable and Inclusive Economic Development: A Global Imperative. Sustainability Policy. https://orcid.org/0000-0002-1061-6509 (2025).

Quah, D. T. Empirics for economic growth and convergence. Eur. Econ. Rev. 40, 1353–1375. https://doi.org/10.1016/0014-2921(95)00051-8 (1996).

Rodríguez-Pose, A. & Crescenzi, R. Research and Development, Spillovers, Innovation Systems, and the Genesis of Regional Growth in Europe. Reg. Stud. 42, 51–67. https://doi.org/10.1080/00343400701654186 (2008).

Shipan, C. R. & Volden, C. The Mechanisms of Policy Diffusion. Am. J. Political Sci. 52, 840–857. https://doi.org/10.1111/j.1540-5907.2008.00346.x (2008).

Shuai, K., Chang, H., He, F. & He, X. Analysis of spatial correlation network and influencing factors of green energy efficiency in urban agglomerations. Sci. Rep. 15, 32137. https://doi.org/10.1038/s41598-025-17905-2 (2025).

Song, A., Rasool, Z., Nazar, R. & Anser, M. K. Towards a greener future: How green technology innovation and energy efficiency are transforming sustainability. Energy 290, 129891. https://doi.org/10.1016/j.energy.2023.129891 (2024a).

Sorrell, S. Jevons’ Paradox revisited: The evidence for backfire from improved energy efficiency. Energy Policy. 37, 1456–1469. https://doi.org/10.1016/j.enpol.2008.12.003 (2009).

Stern, D. I. The Rise and Fall of the Environmental Kuznets Curve. World Dev. 32, 1419–1439. https://doi.org/10.1016/j.worlddev.2004.03.004 (2004).

Tao, M., Poletti, S., Wen, L. & Sheng, M. S. Modelling the role of industrial structure adjustment on China’s energy efficiency: Insights from technology innovation. J. Clean. Prod. 441, 140861. https://doi.org/10.1016/j.jclepro.2024.140861 (2024).

Wang, R., Liu, W., Xiao, L., Liu, J. & Kao, W. Path towards achieving of China’s 2020 carbon emission reduction target—A discussion of low-carbon energy policies at province level. Energy Policy. 39, 2740–2747. https://doi.org/10.1016/j.enpol.2011.02.043 (2011).

Wang, S. & Zou, J. The impact of smart cities on residents’ well-being: Based on quasi-natural experiments & spatial spillover effects analysis. Sustainable Cities Soc. 130, 106565. https://doi.org/10.1016/j.scs.2025.106565 (2025).

Wei, T. & Liu, Y. Estimation of global rebound effect caused by energy efficiency improvement. Energy Econ. 66, 27–34. https://doi.org/10.1016/j.eneco.2017.05.030 (2017).

Wu, J., Lin, K. & Sun, J. Improving urban energy efficiency: What role does the digital economy play? J. Clean. Prod. 418, 138104. https://doi.org/10.1016/j.jclepro.2023.138104 (2023).

Yang, G. et al. A comprehensive city-level final energy consumption dataset including renewable energy for China, 2005–2021. Sci. Data. 11, 738. https://doi.org/10.1038/s41597-024-03529-0 (2024).

Yang, S., Jahanger, A. & Hossain, M. R. Does China’s low-carbon city pilot intervention limit electricity consumption? An analysis of industrial energy efficiency using time-varying DID model. Energy Econ. 121, 106636. https://doi.org/10.1016/j.eneco.2023.106636 (2023).

Yuan, B., Gu, R., Wang, P. & Hu, Y. How Does New Quality Productive Forces Affect Green Total Factor Energy Efficiency in China? Consider the Threshold Effect of Artificial Intelligence. Sustainability 17, 7012. https://doi.org/10.3390/su17157012 (2025).

Yuan, Z., Yu, X. & Li, M. Data asset information disclosure, institutional investor heterogeneity and enterprise value. Mod. Finance Economics-Journal Tianjin Univ. Finance Econ. 42, 32–47. https://doi.org/10.19559/j.cnki.12-1387.2022.11.003 (2022).

Zhang, M., Zhang, D. & Xie, T. Balancing urban energy considering economic growth and environmental sustainability through integration of renewable energy. Sustainable Cities Soc. 101, 105178. https://doi.org/10.1016/j.scs.2024.105178 (2024).

Zhang, T. & Li, J. Network infrastructure, inclusive green growth, and regional inequality: From causal inference based on double machine learning. J. Quant. Technological Econ. 40, 113–135. https://doi.org/10.13653/j.cnki.jqte.20230310.005 (2023).

Zhou, J., Guo, J. & Xu, W. Construction of big data comprehensive pilot zones, new quality productive forces and transformation of watershed resource–based cities: Double machine learning approach. Sustainable Cities Soc. 120, 106144. https://doi.org/10.1016/j.scs.2025.106144 (2025).

Zou, C., Li, S., Xiong, B., Liu, H. & Ma, F. Revolution and significance of Green Energy Transition in the context of new quality productive forces: A discussion on theoretical understanding of Energy Triangle. Pet. Explor. Dev. 51, 1611–1627. https://doi.org/10.1016/S1876-3804(25)60564-7 (2024).

Author Contribution

J.Z.: methodology, analysis and investigation, writing-original draft. Z.Z.: data, validation, writing- review and editing. X.C: investigation, supervision. K.Z: validation, analysis. L.Q: validation, analysis. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by the Project of Basic Research Program of Shanxi Province (No.202303021222184), Shanxi Province Philosophy and Social Sciences Planning Project (No. 2024QN058).

Declarations

Competing interests

The authors declare no competing interests.

Additional information

Correspondence and requests for materials should be addressed to Z.Z.

Yes