Understanding Economic Resilience Using New Quality Productivity Across Multi-Scale Spatial Locations: Machine-Based Spatio-Temporal Effects

Abstract

Amid intensifying climate crises, widening inequalities, and geopolitical volatility, spatial economic resilience (SER) has become critical for regions facing systemic uncertainty. Traditional land-intensive productivity models prove increasingly untenable as spatial resources become finite and development space constrained. China’s new quality productivity (NQP) has emerged as a strategic response emphasizing innovation-driven structural renewal and territorial coordination. Conceptually, NQP is positioned as a SER-oriented strategy prioritizing adaptability, recoverability, and transformability. However, its actual associations remains theoretically overlooked and empirically untested, with existing research viewing it narrowly as technological upgrading while neglecting institutional dimensions, spatial dependencies, and multi-scalar heterogeneities. This study explores how NQP relates to SER from a spatio-temporal perspective: (1) How do the technological and institutional dimensions of NQP relate to SER? (2) What are the spatial patterns of NQP-SER associations across multi-scale locations? Employing XGBoost-SHAP, spatial generalized difference-in-differences, and Geographical Gaussian Process Regression across provincial, city, and enterprise scales in China, we find that NQP’s two dimensions relate to SER very differently. The technological–industrial dimension is the strongest predictor of SER at the provincial scale, exhibiting threshold-type, non-linear associations, while its predictive salience attenuates at the city and enterprise scales, where industrial structure and firm-specific fundamentals are more strongly associated with resilience. The institutional dimension, by contrast, is not positively associated with above-expectation resilience: once common shocks and provincial heterogeneity are absorbed, higher institutional policy intensity is negatively associated with SER, both within provinces and across neighbouring provinces. Spatially, provincial associations rely on coordination and interregional spillovers, while city associations concentrate in nodal clusters where the strength of association depends on capability–context alignment. The findings provide practical theoretical and analytical guidance for tailored policy-making in structurally diverse Global South facing ongoing uncertainty.

1. Introduction

Amid intensifying climate crises [1,2], widening global inequalities [3], escalating geopolitical volatility, and the increasingly finite nature of land and spatial resources upon which economic development has historically relied, the global economy is entering an era of systemic uncertainty. Traditional productivity forms, characterized by territorial expansion, land-intensive industrialization, and the spatial concentration of resource-extractive activities, are proving increasingly untenable [4,5]. These forms have driven development through the physical extension of industrial zones, the extensive occupation and conversion of land, and the clustering of production activities in space, often exhausting local resource bases and ecological carrying capacities while generating location-specific vulnerabilities. As available land becomes scarcer, development space more constrained, and the environmental and social costs of land-intensive growth more acute, the risks inherent in these spatially expansive models intensify. In response, spatial economic resilience (SER), the capacity of place-based economies to absorb location-specific shocks, adapt their territorial development pathways, and transform their spatial economic structures without relying on continuous land expansion, has become both a critical policy priority [6,7,8] and a growing research frontier across economic geography, regional studies, and spatial planning [9,10,11,12,13]. Crucially, SER is neither uniform nor universally applicable but spatially differentiated and path-dependent, shaped by the locational characteristics of regional industrial structures, place-specific institutional capacities, and territorially embedded governance arrangements [14,15,16].

This highlights the urgent need for spatially differentiated and scale-sensitive policy approaches that recognize how development challenges and capacities vary across multi-scale locations, as documented in international literature on regional industrial path development and evolutionary resilience [17,18,19]—particularly in large developing countries like China, where profound spatial heterogeneity across provinces, cities, and counties challenges one-size-fits-all regional development policies. In this context, the concept of new quality productivity (NQP) has emerged as China’s strategic response to fostering resilience-oriented development [20]. Unlike traditional growth-centric models, NQP emphasizes capacities central to SER: adaptability, recoverability, and transformability. This approach offers a systemic reconceptualization of how productivity is generated and spatially organized through innovation, structural renewal, and multi-scale territorial coordination [21,22]. It reframes productivity by incorporating dimensions of resilience, which are particularly relevant in an era marked by uncertainty. Unlike traditional models like total factor productivity or sustainable productivity, NQP integrates a broader set of indicators, including environmental sustainability, technological innovation, and socio-economic inclusivity, that reflect the evolving challenges of the global economy. In this sense, NQP not only aligns with national development goals but also supports broader ambitions for South–South cooperation and inclusive global transitions. It is reshaping how places reorganize in response to disruption [23,24,25]. When considered as a SER-oriented development strategy, NQP reveals an internally differentiated structure. It not only provides the technological and industrial foundation for enhancing regional flexibility and structural renewal, but also acts as an institutional arrangement that supports policy coordination, regulatory responsiveness, and place-sensitive governance. However, despite its promise, NQP as a SER-oriented development strategy remains theoretically overlooked and empirically untested. Its potential to either genuinely enhance resilience or merely reproduce existing dependencies under the guise of transformation remains an open question. Can NQP truly deliver on its envisioned promise? That is, does it meaningfully enhance SER, or simply mask continuity with the rhetoric of transformation?

Existing theoretical and empirical studies have predominantly viewed NQP as a growth-oriented productivity form, often emphasizing its role only as disruptive technological innovation [26,27]. Some research has started exploring its linear associations with economic resilience at the city level [28]. However, several critical limitations persist. First, existing studies tend to conceptualize NQP through a narrowly technocratic and growth-centric lens, aggregating innovation and output indicators while neglecting the institutional and spatial dimensions fundamental to its SER-oriented rationale. This one-dimensional understanding overlooks NQP’s function not merely as a productivity enhancer, but as a systemic arrangement aimed at restructuring spatial governance and enabling adaptive regional transformation. As a result, NQP remains analytically fragmented and normatively insufficient to guide differentiated, place-based development strategies. Second, while NQP is widely promoted as a new development mode, its actual associations with SER remain empirically uncertain. Existing work rarely transcends rhetorical claims, leaving it uncertain whether and how NQP functions as a resilience enhancer. Third, studies examining NQP and SER relationships often suffer from incomplete indicator selection, nonstationary and biased estimation, neglecting the complex nonlinear relationship between them and the spatial similarities, dependencies, and hierarchical heterogeneities intrinsic to territorial systems. This limits the accuracy and contextual relevance of their findings, constraining a nuanced understanding of NQP’s role across scales and contexts.

To address these gaps, this study aims to examine how NQP relates to SER from a spatio-temporal perspective. Specifically, we focus on two questions: (1) How do the technological and institutional dimensions of NQP relate to SER? (2) What are the spatial patterns of NQP-SER associations across multi-scale spatial locations?

By addressing these questions, we will contribute to the existing knowledge in the following ways: (1) We reconceptualize NQP as a multi-dimensional SER-oriented design, building on but extending beyond existing discussions of NQP as a generic proxy for technological or industrial upgrading. The reframing challenges fragmented views of NQP and highlights its institutional embeddedness and policy relevance in shaping resilience. (2) To the best of our knowledge, we are among the first to systematically bridge the conceptual rationale and empirical consequence of NQP as a SER-oriented design, demonstrating not only how it is theoretically designed to enhance spatial economic resilience, but also how it empirically achieves this across multiple scales and dimensions. This offers analytical clarity and practical relevance, offering potentially transferable conceptual insights for structurally diverse economies navigating structural transitions under uncertainty, with the caveat that direct application requires careful attention to local institutional contexts. (3) We address critical limitations in prior research by explicitly incorporating spatial similarity, spatial dependence, and multi-scalar heterogeneity into the analysis. Methodologically, we overcome small-sample constraints and non-stationarity by employing a spatially interpretable machine learning framework, thereby offering a more robust and nuanced understanding of how NQP is associated with SER across diverse territorial contexts.

2. Literature Review

To build a theoretically integrated framework that positions NQP as a technological and institutional dimension design oriented towards SER, this chapter is structured into two interrelated parts. Section 2.1 reconceptualizes NQP as a multi-dimensional construct fundamentally oriented towards SER. Section 2.2 theorizes its dynamic role in shaping SER outcomes, understanding how NQP’s rationale can be operationalized through regionally embedded evolutionary processes. These sections lay a coherent and theoretically grounded foundation for understanding the pathways through which NQP informs SER, setting the stage for the empirical analyses that follow.

2.1. Understanding of Multi-Dimensional SER-Oriented NQP

NQP is understood as a multi-dimensional, SER-oriented design, not a reactive industrial adjustment, but a forward-looking developmental paradigm. As illustrated in Figure 1, NQP is designed to enhance the three core capacities of SER: adaptability, recoverability, and transformability. Importantly, NQP should not be treated as an exogenous driver of resilience, but as an internally formulated response to the evolving structural demands of regional economies. In contrast to traditional productivity frameworks centered on output maximization and efficiency [29], NQP derives its conceptual significance from its embeddedness in the long-run, path-dependent trajectories of spatial economic evolution. It unfolds along two interdependent dimensions: a technological–industrial foundation rooted in material and innovation dynamics, and an institutional foundation shaped by evolving governance and coordination capacities. Together, these dimensions capture the dual logic of SER and position NQP as an endogenous catalyst of regional transformation.

2.1.1. NQP as the Technological–Industrial Foundation of SER

The technological–industrial dimension of NQP constitutes the material foundation of SER, embedding adaptability, recoverability, and transformability within a continuous and interdependent evolutionary process. This dimension is actualized through three core mechanisms that collectively operationalize NQP’s resilience-oriented function, integrating structural dynamics with systemic learning in complex spatial systems.

First, innovation driving supports adaptability by institutionalizing learning and recombinant capability within regional economic structures. Drawing on evolutionary economic geography, adaptability is conceptualized as a system’s capacity to navigate uncertainty through the continuous generation, recombination, and cross-domain application of knowledge [30]. NQP materializes this capacity by embedding innovation-oriented organizational routines [31] and expanding related variety, thereby widening the regional search space for experimental recombination. In contrast to exogenous models of innovation, NQP internalizes innovation as an endogenous evolutionary function, enhancing structural plasticity and equipping regions with heterogeneous response strategies to external shocks [32]. This process fosters a dynamic equilibrium between exploration and exploitation, crucial for sustaining adaptive potential under volatility. By facilitating this learning process, NQP enables regions to rapidly experiment with new ideas, technology combinations, and organizational practices, which enhances their capacity to absorb and adapt to external disruptions.

Second, resource assurance supports recoverability by constructing the material and organizational redundancies necessary for shock absorption and systemic restoration. While adaptability introduces beneficial variation, it requires structural anchoring to prevent systemic fragmentation under stress. As emphasized in evolutionary resilience theory, resilient systems depend on institutional variety and strategic slack to maintain functionality during disruptions [33]. NQP institutionalizes such redundancy and modularity through diversified infrastructure systems, resilient supply networks, and interoperable digital platforms. This ensures not only rapid recovery but also stability without rigidity, enabling systems to regenerate functional capacity without reverting to obsolete developmental paths. By fostering a diversified infrastructure and supply chains, NQP allows regions to maintain essential functions during disruptions, while facilitating the reconfiguration of these functions post-crisis, ensuring long-term recovery without falling into path-dependency traps.

Third, transformability enables structural reconfiguration within the evolutionary logic of regional development, transcending recovery to facilitate long-term reorientation. Within evolutionary resilience frameworks, transformability entails breaking path dependency through the recombination of heterogeneous knowledge, technologies, and institutional architectures [34]. NQP cultivates this capacity by fostering emergent technological domains and frontier industries, which serve as catalysts for new growth trajectories and structural diversification. These advanced sectors underpin a productivity paradigm oriented toward Schumpeterian renewal, allowing regions to strategically reposition within the global division of labor under conditions of structural uncertainty [35,36]. In this context, NQP transforms episodic disruptions into opportunities for evolutionary re-specialization. Instead of returning to previous growth paths, regions can embrace new trajectories that better align with future global trends and domestic capabilities, thereby fostering long-term sustainability and competitive advantage.

In sum, the technological–industrial dimension of NQP contributes to SER by not only enhancing a region’s adaptability and recoverability but also by fostering structural transformation, enabling regions to effectively manage both immediate disruptions and long-term evolution. These mechanisms collectively ensure that NQP serves as a critical driver of spatial economic resilience, preparing regions to navigate and thrive under dynamic, global conditions.

2.1.2. NQP as the Institutional Foundation of SER

Complementing its material base, the institutional dimension of NQP establishes the systemic architecture that configures and sustains the dynamic capacities of SER across time and scale. This dimension operates through three synergistic institutional mechanisms that underpin NQP’s role as an endogenous driver in the evolutionary dynamics of spatial economies.

First, facilitating science, technology, and talent development forms the foundational institutional layer essential for adaptability. Adaptability is understood not merely as a technical feature but as an institutional process that requires the continuous generation, circulation, and recombination of knowledge across diverse domains. Institutions play a pivotal role in structuring the conditions that allow for knowledge flow, particularly in fostering collaboration between academia, industry, and government. Institutional entrepreneurs, actors capable of reconfiguring rules, incentives, and routines [37], strengthen adaptability by embedding institutionalized mechanisms for R&D investment, talent attraction, and cross-sectoral knowledge exchange. This institutional “thickness” facilitates regional experimentation and innovation, diversifying the knowledge base and fostering local adaptive capacities [38]. Thus, the creation of conducive institutional contexts that support innovation ecosystems becomes a key pathway for improving the resilience of regions facing complex and uncertain environments.

Second, embedding green development and infrastructure constitutes the institutional pillar critical to recoverability. While adaptability ensures that systems can adjust to shocks, without the institutional capacity for recovery, regions may be unable to withstand larger, more disruptive crises. This redundancy, or “strategic slack,” is a central theme in evolutionary resilience theory, which stresses the importance of institutional variety and flexibility to allow regions to “bounce back” [27]. By coordinating multi-level actors in environmental governance, digital infrastructure, and ecological investments, institutions ensure that resilience is not just about restoring the previous status quo but enhancing the system’s ability to function at a higher, more adaptable level post-disruption. Furthermore, these institutional structures create redundant capacities and flexible frameworks that absorb shocks, thus avoiding the risks of lock-in associated with overly rigid or narrowly focused infrastructural systems. In this way, NQP’s institutional dimension helps to embed resilience within the physical and organizational structures of the region, promoting both recovery and long-term stability.

Third, steering structural transformation is, conceptually, the institutional dimension most relevant to the long-term evolution of SER; empirically, however, we find that intensifying institutional policy intensity is not positively associated with above-expectation resilience, and that substantive industrial-innovation capability, rather than institutional policy per se, is what our analysis associates with stronger SER. While adaptability and recoverability are essential, true resilience also requires transformative capacity, the ability to break free from established paths and develop new trajectories for economic and institutional growth. Drawing on the concept of institutional bricolage [39], this mechanism emphasizes the capacity of local governments and other actors to recombine policy tools, legal frameworks, and governance structures to promote the emergence of frontier sectors and new technologies. Through anticipatory regulation, industrial foresight, and mission-oriented governance, institutions provide the guidance and structure needed to facilitate Schumpeterian transformation, enabling the region to create new paths of growth and development in response to shifting global challenges. The key here is that NQP, through its institutional dimension, creates an enabling environment for regional economies to not only resist external shocks but also to actively reinvent themselves by fostering long-term evolutionary renewal. This institutional transformation is essential for overcoming path dependency, allowing regions to reorient their development trajectories toward more sustainable, innovative, and resilient outcomes [29,30].

In summary, the institutional foundation of NQP provides the necessary scaffolding for achieving the core dimensions of spatial economic resilience. By focusing on institutional innovation, multi-level coordination, and structural transformation, NQP enables regions to move beyond mere recovery, offering a strategic framework for continuous adaptation, the creation of new growth trajectories, and the reconfiguration of regional economic landscapes. This institutional foundation supports both the ability to recover from disruptions and the capacity to reinvent regional economies in the face of persistent uncertainty, making it a key driver of evolutionary resilience in complex spatial systems.

2.2. A Framework Linking NQP with SER from a Spatio-Temporal Perspective

NQP, as a composite of technological–industrial capabilities and institutional frameworks, shapes regional development trajectories by creating conditions for regions to resist, adapt to, and recover from external shocks while pursuing structural change that extends beyond mere bounce-back. Rooted in digitalization, green innovation, and platform-based coordination, NQP serves dual roles, acting both as a disruptive shock that challenges existing paths and as a generative catalyst that facilitates reconfiguration, contingent on regional institutional contexts [40]. However, its associations in fostering SER depend on selective filtering, institutional embedding, and regional enactment [30]. This duality underscores the nuanced relationship between innovation and resilience, as the same disruptive forces that challenge established paths can also spark the creative destruction necessary for transformative growth. Thus, SER emerges as a co-evolutionary feedback process between structural disruption and adaptive reordering, where regions are not mere passive recipients of external shocks but active agents in their resilience-building processes [23,35].

NQP opens a critical intervention point within regional evolutionary trajectories by interacting with entrenched path-dependent structures [33,41]. While such structures, anchored in technologies, routines, and institutional logics, offer stability, they also restrict renewal [42]. The interaction between NQP and path-dependent structures creates a paradoxical space: while it may destabilize existing systems, it also holds the potential to create a window of opportunity for reinvention. As Figure 2 illustrates, NQP introduces a window of path plasticity, acting either as a disruptive shock or a generative catalyst [43]. Whether this disturbance leads to transformation hinges on the region’s capacity for productive recombination, institutional embedding, and the formation of new growth paths. This dynamic suggests that resilience is not merely a reaction to shocks but a proactive strategy to reconfigure regional futures in response to new opportunities and challenges.

Absent institutional absorptive capacity, NQP may destabilise rather than productively transform established structures, unsettling but unproductive. In such contexts, NQP dismantles existing paths without viable alternatives, leading to organizational decline, labor displacement, or innovation lock-out [44]. This triggers a negative feedback loop that destabilizes the system while failing to enable adaptive renewal, thereby deepening structural lock-in or stagnation through over-reliance on eroded growth mechanisms [23]. In these cases, SER is undermined less by the shock itself than by the absence of internal mechanisms for constructive transformation. The role of institutional absorptive capacity, therefore, is not merely reactive but is essential for transforming external shocks into opportunities for growth, learning, and adaptation. Without these mechanisms, regions risk falling into a cycle of decline rather than renewal, highlighting the centrality of institutional resilience in the face of disruption.

Only when embedded within sufficiently dynamic institutional systems can NQP activate the positive feedback mechanisms that support path transformation. Regions with vibrant innovation ecosystems and flexible governance can reconfigure their economies through knowledge recombination, institutional experimentation, and resource mobilization [45]. These capacities align with the notion of constructive resilience [38], enabling economies to transcend inherited structures. In such contexts, NQP facilitates systemic transitions by generating new growth mechanisms that reinforce existing capabilities and establish positive feedback loops, thereby overcoming path dependencies and enabling sustainable renewal [46,47]. This process of reconfiguration not only supports the development of new growth paths but also reinforces the institutional foundations necessary for future resilience. As such, NQP plays a dual role in both enabling and accelerating the evolution of regional innovation systems, underscoring the importance of flexible and adaptive institutional frameworks for sustainable development.

NQP does not prescribe resilience outcomes but defines the institutional opportunity space through which regions may navigate structural disruption [48]. SER reflects a region’s capacity to convert shocks into opportunities for path creation and institutional innovation, rather than restoration [49]. Whether regions stagnate or transform depends on their capacity to mediate between NQP-induced disruption and their institutional and cognitive infrastructures [36]. These evolving spatial capacities, including absorptive institutions, learning mechanisms, and collective agency, render resilience an evolutionary accomplishment [50]. The challenge, therefore, lies in the ability of regions to leverage NQP as a catalyst for renewal rather than destruction. By doing so, they not only strengthen their immediate resilience but also lay the groundwork for long-term adaptive capacity, which is vital for navigating future challenges and opportunities. Thus, resilience becomes an ongoing, iterative process, constantly shaped by both external shocks and internal responses.

Building on this evolutionary logic, the framework yields scale-specific expectations that we examine empirically rather than rationalise post hoc. We anticipate three patterns. First, at the provincial and enterprise scales, the technological–industrial dimension is expected to show threshold-type, non-linear associations with SER, such that the predictive contribution of NQP becomes pronounced only once a region or firm surpasses a minimum capability level. Second, at the city scale, the association between NQP and SER is expected to be context-dependent, varying with industrial structure and education levels, so that otherwise similar cities exhibit divergent patterns. Third, at the provincial scale, the institutional dimension is expected to display spatial dependence, whereby policy intensity in neighbouring regions co-varies with local outcomes. We also keep open the possibility that NQP exhibits ambiguous, selective, or even negative associations with SER where institutional or absorptive conditions are limited. These anticipated, scale-differentiated patterns motivate the multi-scale empirical design and structure the comparative discussion that follows. We emphasise that these are predictive and associational expectations.

3. Methodology

3.1. Data and Variables

The explanatory variable in this study is NQP, which we measure through a multi-scale, multi-dimensional system reflecting both technological–industrial and institutional foundations for SER. Operationally, NQP is defined in this study as a composite of technological–industrial capabilities (innovation driving, resource assurance, structural transformation) and institutional arrangements (science/talent support, green/infrastructure embedding, structural transformation guidance) that enable regions and firms to absorb shocks and reconfigure their development trajectories. This definition distinguishes NQP from related but narrower concepts: unlike Total Factor Productivity (TFP), which aggregates input efficiency, NQP incorporates forward-looking structural and institutional dimensions; unlike the Regional Innovation System (RIS) framework, which centres on the localised institutional and network arrangements that generate and diffuse innovation within a single region, NQP emphasises multi-scalar territorial embedding and an explicit orientation toward resilience-enhancing structural transformation; unlike green productivity, it integrates both technological renewal and institutional governance. NQP can thus be understood as partly extending the RIS tradition while remaining conceptually distinct in its multi-scalar, resilience-oriented framing.

NQP is reconceptualized as a SER-oriented design, with the measurement framework designed to capture how NQP directs SER across two dimensions. The technological–industrial foundation dimension focuses on both provincial/city and enterprise scales, with distinct indicators and data sources at each level. These indicators, sourced from official statistical yearbooks and enterprise databases, are aggregated into composite NQP measures at multi-scales using the entropy weighting method. Specifically, all indicators are min–max normalised to the [0, 1] interval; entropy weights are computed using the information entropy method; missing values are imputed by linear interpolation where fewer than two consecutive years are missing, and larger gaps are excluded; and the NLP scoring applies a structured three-dimension rubric (science-and-technology/talent, green/infrastructure, and structural transformation) to each sentence in provincial government work reports, with dimension scores averaged across sentences.

Due to data constraints, the institutional foundation dimension is measured only at the provincial scale. The multi-scale framework is therefore partial and asymmetric: the technological–industrial dimension is analysed at the provincial, city, and enterprise scales, whereas the institutional dimension is analysed only at the provincial level. We make this scope explicit and refrain from claiming that the institutional mechanism is equally analysed at all spatial levels. Although the policy was launched nationally in 2023, its intensity varies regionally. The provincial panel used for the spatial generalized difference-in-differences analysis spans 2018–2023, with 2023 as the single post-treatment year following the national NQP rollout. We adopt a data-intelligent approach by applying large-scale text mining and natural language processing (NLP) techniques to thousands of local government policy documents, thereby constructing a continuous measure of NQP policy intensity that reflects SER-oriented institutional arrangements (

Table 1. NQP measurement system at multi-dimensions and multi-scales.

Dimensions	Scales	1-Level Indicators	2-Level Indicators	Definition	Orient of SER
Technological–industrial foundation	Province & City	Innovation Driving	employee competence	Average salary of employees (RMB yuan)	Adaptability
			Educated level of staff	Number of Colleges and Universities
		Resource Assurance	New quality infrastructure	Internet broadband access subscribers (103)	Recoverability
				Total value of telecommunications services
			Pollution Control	Environmental pollution control investment (108 RMB yuan)
				Harmless treatment rate of domestic waste (%)
		Structural Transformation	Innovative outputs	Proportion of scientific expenditure in local fiscal expenditure (%)	Transformability
				Number of invention patents applied for in the current year
				Number of utility model patents applied for in the current year
			Intelligent level Green development	Number of AI companies
				Number of green invention patents applied for in the current year
				Number of green utility model patents applied for in the current year
			Data factors	Number of data trading platforms
	Enterprise	Innovation Driving	R&D salary	Proportion of R&D salary on operating income (%)	Adaptability
			R&D person	Proportion of R&D personnel (%)
			Educated level of staff	Proportion of employees with bachelor’s degree or above (%)
		Resource Assurance	Fixed asset	Proportion of fixed asset on total asset (%)	Recoverability
			Manufacturing expenses	Proportion of manufacturing expenses on total cost (%)
		Structural Transformation	Hard technology	Proportion of R&D depreciation amortization (%)	Transformability
				Proportion of R&D leasing costs (%)
				Proportion of R&D direct input (%)
			Soft technology	Proportion of intangible assets (%)
				Total asset turnover rate (%)
				Reciprocal of the equity multiplier
Institutional foundation	Province	Institutional Support for S&T and Talent	Technology leadership	Reflects policy support for scientific leadership and talent-driven innovation, facilitating rapid regional adaptation	Adaptability
			Talent & innovation
		Institutional Support for Green and Infrastructure	Environmental governance	Reflects institutional support for environmental resilience and digital infrastructure, ensuring system-wide recovery	Recoverability
			Infrastructure support
		Institutional Guidance for Structural Transformation	Industrial upgrading	Captures how policies guide regional economies toward future-oriented industries and technological trajectories	Transformability
			Technological shift

). This NLP-based measure offers several advantages over conventional proxies for institutional policy commitment. Unlike simple policy counts, binary pilot-designation indicators, or raw keyword frequencies, the NLP scoring approach captures the semantic intensity and thematic emphasis of policy discourse across provinces and years, providing a continuous and temporally dynamic index of institutional engagement with NQP. The scores have been cross-validated against official NQP pilot designation lists, yielding broadly consistent geographic patterns that support the measure’s construct validity. One inherent caveat is that intensity scores reflect the prominence of NQP-related language in strategic documents and may not fully capture implementation depth, which is discussed further in Section 5.

The explained (dependent) variable in this study is SER. At the provincial and city scales, SER is captured through a single-indicator approach, reflecting not a static outcome but a dynamic process in which regions continually adjust structures and adapt to shocks. Specifically, it is measured as the proportional deviation between actual and counterfactual GDP growth, with the latter proxied by national real GDP growth. It should be emphasised that SER is, by definition, the spatial and multi-scalar manifestation of economic resilience: it denotes economic (output) resilience only, and does not purport to measure social, ecological, or institutional resilience. Operationalising economic resilience as the deviation of actual regional growth from an expected national benchmark is a well-established and widely adopted approach in the regional economic resilience literature [43,51,52]. This output-based measure directly captures the resistance-and-recovery performance through which a region’s adaptability, recoverability, and transformability are ultimately realised in economic terms, while providing a consistent and comparable indicator across the provincial, city, and enterprise scales. Employment resilience, structural diversification, and output volatility are conceptually distinct constructs—respectively a separate type of resilience, a determinant of resilience, and a different statistical property—so aggregating them into a single composite would conflate outcomes, determinants, and distinct resilience types and blur the conceptual boundaries of the construct. The calculation is as follows:

$${\mathrm{SER}}_{\mathrm{i},\mathrm{t}}={\displaystyle \frac{\left({\Delta \mathrm{E}}_{\mathrm{it}}\right)-{({\Delta \mathrm{E}}_{\mathrm{it}})}^{\mathrm{expected}}}{\mid {({\Delta \mathrm{E}}_{\mathrm{it}})}^{\mathrm{expected}}\mid }}$$

(1)

where ${\mathrm{SER}}_{\mathrm{i},\mathrm{t}}$ is the spatial economy resilience, and i is the region; ${\Delta \mathrm{E}}_{\mathrm{it}}$ is the real GDP growth of region $\mathrm{i}$ in year $\mathrm{t}$; ${({\Delta \mathrm{E}}_{\mathrm{it}})}^{\mathrm{expected}}$ is a counterfactual GDP growth. When the SER is greater than 0, it means that the economic resilience of the region is higher than the national average, and the higher the value, the stronger the city’s economic resilience, and vice versa. At the enterprise level, SER incorporates both growth and volatility: three-year cumulative sales growth captures performance, while the standard deviation of monthly stock returns reflects volatility. These are aggregated via entropy weighting. SER here also denotes an adaptive, evolving capacity, not a fixed outcome, through which Enterprises respond to uncertainty and maintain growth.

To address potential omitted variable bias and improve robustness, we also control for a set of characteristics at the multi-scale (

Table 2. Descriptions of the selected control variables.

Scale	Variables	Definitions
Province and city	IS	Ratio of secondary industry added value to tertiary industry added value
	Edu	Share of education expenditure in general fiscal budget expenditure
	Fin	Ratio of the sum of deposit balances and loan balances in financial institutions to GDP
	Cons	Ratio of total retail sales of consumer goods to GDP
	Road	The area of roads per capita
	Temp	Average annual temperature of a region
	Prec	Annual precipitation of a region (mm)
Enterprise	Size	Natural logarithm of total assets
	Lev	Total liabilities divided by total assets (Debt-to-asset ratio)
	ROA	Net profit divided by total assets (Return on assets)
	ROE	Net profit divided by shareholders’ equity (Return on equity)
	Cashflow	Operating cash flow divided by total assets
	Balance	Current ratio: current assets divided by current liabilities
	TobinQ	Market value of firm divided by replacement cost of assets (Tobin’s Q)
	FirmAge	Years since establishment of the firm
	INST	Institutional ownership share (%)
	Mshare	Management ownership share (%)
	Employ	Number of employees (logged)
	Cap	Capital intensity: ratio of fixed assets to total assets

).

Spatiotemporal mapping across multiple scales (Figure 3, Figure 4 and Figure 5) reveals a clear scale effect and spatial disparity in the distribution of NQP and SER. At the provincial level, both variables exhibit relatively continuous spatial patterns. However, as the scale narrows to the city and firm levels, their spatial configurations become increasingly fragmented and discontinuous. This fragmentation reflects more than the finer spatial granularity of smaller units; it is an economically meaningful outcome of uneven territorial development. Administrative boundary effects and the uneven distribution of innovation resources across the urban hierarchy concentrate NQP capabilities in core cities. Agglomeration economies further reinforce advantages in provincial capitals and special-economic-zone cities while peripheral cities lag. Meanwhile, the polycentric structure of the Chinese urban system produces high-NQP nodes embedded within lower-NQP surroundings. The discrete nodal clusters observed at the city scale should therefore be read as a substantive signature of spatially selective capability accumulation rather than merely a statistical artefact of scale. Temporally, NQP maintains a persistent pattern of agglomeration along the eastern coastal regions, while economic resilience displays greater spatial randomness, with pockets of high-high clusters emerging more frequently in central and western regions.

3.2. Method

This study employs three complementary analytical methods to address its two research questions. RQ1 (how do the technological and institutional dimensions of NQP relate to SER?) is addressed by two complementary methods: (1) XGBoost-SHAP, which provides predictive associations between NQP’s technological–industrial dimension and SER at the provincial, city, and enterprise scales, yielding scale-differentiated predictive inference; and (2) the Spatial Generalized Difference-in-Differences model (SGDID), which examines the associations of NQP’s institutional policy intensity with SER at the provincial scale under a quasi-experimental design that controls for province and year fixed effects and spatial spillovers, yielding associational inference. RQ2 (what are the spatial patterns of NQP-SER associations across multi-scale locations?) is addressed by Geographical Gaussian Process Regression with GeoShapley (GGPR-GeoShapley), which characterises spatially varying predictive associations at the provincial and city scales, providing spatially continuous, uncertainty-quantified geographic inference.

In this study, we first employ the XGBoost–SHAP method to examine how the technological–industrial dimension of NQP is associated with SER across multiple spatial scales. This relationship is inherently complex, nonlinear, and interactive, as NQP comprises diverse components, such as human capital, innovative actors, and digital production platforms, that are associated with SER in entangled, non-additive ways. Yet, prior studies have tended to oversimplify this complexity [28], often relying on linear modeling frameworks or facing constraints due to small sample sizes and nonstationarity, thereby limiting their explanatory power. In contrast, the XGBoost algorithm, a gradient boosting decision tree method, is well-suited to detect nonlinear patterns and feature interactions without presupposing specific functional forms. The XGBoost models are estimated with 200 boosting trees, a maximum tree depth of 4, a learning rate of 0.05, and 80% subsampling of observations and features per tree. Moreover, it performs effectively with limited sample sizes and is robust to non-stationary distributions, which is particularly advantageous for capturing the complex, spatially uneven dynamics through which NQP is associated with SER across diverse territorial scales. Coupled with SHAP (SHapley Additive exPlanations) values for interpretability, this approach provides a novel opportunity to systematically unravel the multifaceted associations of NQP with SER, overcoming prior methodological constraints. This model can be formally expressed as:

$$\widehat{\mathrm{S}\mathrm{E}\mathrm{R}}={\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}} {\mathrm{\alpha }}_{\mathrm{k}}{\mathrm{h}}_{\mathrm{k}}\left(\mathrm{x}\right)$$

(2)

$$\mathrm{S}\mathrm{H}\mathrm{A}{\mathrm{P}}_{\mathrm{i}}={\mathrm{\varphi }}_{\mathrm{i}}\left(\mathrm{f}\right)=\sum _{\mathrm{S}\subseteq \mathrm{N}\setminus \mathrm{i}} {\displaystyle \frac{\left|\mathrm{S}\right|!\left(\left|\mathrm{N}\right|-\left|\mathrm{S}\right|-1\right)!}{\left|\mathrm{N}\right|!}}\left[\mathrm{f}\left(\mathrm{S}\cup \mathrm{i}\right)-\mathrm{f}\left(\mathrm{S}\right)\right]$$

(3)

where $\widehat{\mathrm{SER}}$ is the predicted outcome of SER, $\mathrm{K}$ is the number of trees, ${\mathrm{\alpha }}_{\mathrm{k}}$ is the weight of the kth tree, and ${\mathrm{h}}_{\mathrm{k}}\left(\mathrm{x}\right)$ is the prediction made by the kth tree. The model is trained on diverse regional attributes, including policy and control variables. $\mathrm{N}$ is the set of all features, $\mathrm{S}$ is a subset of features, and $\mathrm{f}\left(\mathrm{S}\right)$ represents the prediction made by the model using features from subset $\mathrm{S}$. This facilitates transparent predictive interpretation and mechanistic insights into the associations between NQP and SER.

In addition, to further characterise the associations of NQP’s institutional dimension with SER, we treat its nationwide rollout in 2023 as a quasi-natural experiment. While traditional Difference-in-Differences (DID) models can identify policy effects, they assume spatial independence and thus overlook spatial spillovers, a critical limitation when analyzing innovation diffusion, talent mobility, and digital infrastructure, which often cross administrative boundaries. Ignoring these linkages risks biased estimates and underestimating spillover magnitude.

Therefore, we adopt a Spatial Generalized Difference-in-Differences (SGDID) framework that accounts for both temporal variation and spatial interdependence. By interacting the policy intensity measured by NLP with a post-time dummy, we create a generalized treatment variable that captures heterogeneous policy exposure. To improve coefficient interpretability, the policy intensity variable and the NQP level variable are each standardised by their cross-sectional standard deviation. We estimate a spatial Durbin model with two-way fixed effects (province and year): province effects absorb time-invariant regional heterogeneity, year effects absorb common shocks such as the aggregate change in the policy year, and the spatially lagged covariates capture cross-province spillovers. Because the policy was launched nationwide in 2023 with no untreated control group, identification relies on cross-provincial variation in policy intensity, and we interpret the estimates as associations rather than causal effects. This design moves beyond binary treatment assumptions and allows for more nuanced characterisation of spatially differentiated policy associations. To mitigate potential confounding effects from the technological–industrial dimension of NQP, we explicitly include it in the model. To incorporate spatial spillovers, our empirical specification explicitly models both the dependence of SER on neighboring regions’ outcomes and the spatial diffusion of policy associations. Specifically, our SGDID model incorporates spatial dependence by including spatially lagged dependent variables as well as spatially lagged explanatory variables, capturing both the associations of neighboring regions’ SER and the spatial spillover of policy intensity and control factors. This approach addresses spatial autocorrelation and controls for unobserved spatial heterogeneity, improving estimation accuracy and avoiding bias from ignored spatial interdependence. Formally, the model can be expressed as follows:

$${\mathrm{SER}}_{\mathrm{it}}={\mathrm{\rho }\mathrm{WSER}}_{\mathrm{it}}+{\mathrm{\beta }}_1\left({\mathrm{Post}}_{\mathrm{t}}\times {\mathrm{Intensity}}_{\mathrm{i}}\right)+{\mathrm{\beta }}_2{\mathrm{Intensity}}_{\mathrm{i}}+{\mathrm{\beta }}_3{\mathrm{NQP}}_{\mathrm{it}}+{\mathrm{\beta }}_4{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\theta }\mathrm{WX}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\lambda }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(4)

where $\mathrm{W}$ is the spatial weight matrix, capturing the spatial dependence among regions. ${\mathrm{WSER}}_{\mathrm{it}}$ accounts for the associations of neighboring regions’ SER on region $\mathrm{i}$. ${\mathrm{Intensity}}_{\mathrm{i}}$ is the continuous measure of policy intensity in region $\mathrm{i}$. ${\mathrm{Post}}_{\mathrm{t}}$ is a time dummy indicating the post-policy period. ${\mathrm{Post}}_{\mathrm{t}}\times {\mathrm{Intensity}}_{\mathrm{i}}$ is the interaction term representing the heterogeneous policy intensity. $\mathrm{Controls}$ denotes the control variables. ${\mathrm{\mu }}_{\mathrm{i}}$ and ${\mathrm{\lambda }}_{\mathrm{t}}$ denote city and time fixed effects, respectively, and $\mathrm{\epsilon }$ is the error term. This specification explicitly accounts for spatial dependence in both the outcome and explanatory variables, which helps mitigate bias caused by spatial spillovers and unobserved spatial heterogeneity. We decompose the total policy effect into direct (own-province) and indirect (spatial-spillover) components using the standard spatial-effects decomposition, reported alongside the main coefficients in

Table 3. Results of SGDID at province scale.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
VARIABLES	Main	Wx	Total	Direct	Indirect	Spatial	Variance
DID	−0.802 ***	−1.410 ***	−2.341 ***	−0.819 ***	−1.522 ***
	(0.189)	(0.488)	(0.579)	(0.191)	(0.528)
NQP	−0.231 *	−0.798 ***	0–1.089 ***	−0.241 *	−0.848 ***
	(0.124)	(0.300)	(0.351)	(0.128)	(0.316)
IS	−0.290	−1.702 ***	−2.109 ***	−0.241 *	−0.848 ***
	(0.207)	(0.300)	(0.664)	(0.206)	(0.613)
Edu	0.063	2.467	2.678	0.093	2.585
	(3.959)	(11.76)	(13.717)	(3.991)	(12.860)
Fin	0.105	1.078 ***	1.252 ***	0.118	1.134 ***
	(0.152)	(0.391)	(0.453)	(0.152)	(0.408)
Cons	−0.090	−1.502	−1.685	−0.108	−1.577
	(1.800)	(3.123)	(3.860)	(1.807)	(3.380)
Road	0.111 **	−0.243 **	−0.140	0.108 **	−0.248 **
	(0.000)	(0.110)	(0.129)	(0.046)	(0.120)
Temp	0.039	0.104	0.151 *	0.040	0.111
	(0.036)	(0.068)	(0.085)	(0.036)	(0.074)
Prec	−0.000 *	0.000	−0.000	−0.000 *	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
rho						0.055
						(0.067)
sigma2_e							0.737 **
Observations	186	186	186	186	186	186	186
R-squared	0.788	0.788	0.788	0.788	0.788	0.788	0.788
Number of code	31	31	31	31	31	31	31
Province FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Note: Standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

; the model thereby incorporates spatial interconnections and improves the robustness of the estimates. Due to data limitations, our analysis is conducted at the provincial scale, which provides an initial but meaningful perspective on the spatiotemporal variation in NQP policy associations.

However, the XGBoost-SHAP model does not explicitly account for spatial dependence or heterogeneity, which are critical for capturing regional variations in economic resilience. While traditional spatial regression models, such as Geographically Weighted Regression (GWR), address spatial heterogeneity, they often require subjective parameter tuning and provide limited probabilistic inference, constraining their interpretability and robustness.

To overcome these limitations and better characterize the spatially varying associations of NQP with SER, we employ Geographical Gaussian Process Regression (GGPR). GGPR incorporates geographic coordinates directly into its kernel functions, enabling it to model spatial autocorrelation as a continuous function of distance. By combining a Matérn kernel with a spatial similarity kernel, GGPR effectively captures both broad regional trends and localized deviations, while providing rigorous uncertainty quantification for predictions. In this framework, the SER index is modeled as a realization of a Gaussian process with spatially structured covariance, allowing for smooth interpolation and inference across both observed and unobserved locations [5]. Formally, the SER index f(x) is treated as a realization from a Gaussian process:

$$\mathrm{f}\left(\mathrm{x}\right)~\mathrm{gp}\left(\mathrm{m}\left(\mathrm{x}\right),\mathrm{k}\left(\mathrm{x},{\mathrm{x}}^{\prime }\right)\right)$$

(5)

where $\mathrm{m}\left(\mathrm{x}\right)$ is the mean function and $\mathrm{k}\left(\mathrm{x},{\mathrm{x}}^{\prime }\right)$ is the kernel function representing spatial similarity.

To further enhance spatial interpretability, we integrate GeoShapley, an extension of the conventional SHAP attribution method that embeds spatial information within the GGPR kernel [49]. Unlike standard SHAP, which treats location as an ordinary feature, GeoShapley enables location-aware decomposition of predictions, attributing feature effects and their interactions in a spatially continuous manner. This approach generates smooth spatial surfaces that visualize how individual components of NQP are associated with economic resilience across space, alongside corresponding predictive uncertainties.

The combined GGPR–GeoShapley framework offers several key advantages for analyzing NQP-SER associations across multiple spatial scales. First, it provides a transparent and rigorous attribution mechanism within a flexible, spatially continuous modeling structure, balancing interpretability and predictive accuracy. Second, it captures spatial non-stationarity by jointly modeling global spatial trends and local heterogeneities. Third, by integrating spatial autocorrelation with feature similarity, the approach improves robustness and granularity, offering deeper insights into spatial dependence patterns and underlying mechanisms.

4. Results and Discussion

4.1. The Multi-Dimensional Associations of NQP with SER

4.1.1. The Technology–Industrial Dimension: Considering Nonlinearity and Small Sample

At the provincial scale, XGBoost–SHAP results reveal that NQP, measured through its technological–industrial dimension, emerges as the most influential predictor of SER within the model, exceeding conventional factors such as Edu, Fin, IS, and Road. This finding challenges the traditional assumption that regional resilience stems primarily from factor accumulation. Instead, it suggests that resilience depends more fundamentally on regions’ capacity to reconfigure their economic structures through knowledge recombination and technological upgrading. The magnitude of NQP’s SHAP values highlights its role as a mechanism of adaptive renewal, consistent with Boschma’s [46] framing of evolutionary resilience. The upward trend of its marginal effect across quantiles suggests a threshold effect, where SER accelerates once regions accumulate sufficient cognitive and technological resources, which Martin and Sunley [30] describe as critical transitions. Regions below this threshold may find that investments in physical infrastructure or financial services yield limited resilience benefits without concurrent upgrading of their technological capabilities. In contrast, provinces that have developed strong innovation ecosystems, through research clusters, advanced manufacturing zones, or technology transfer mechanisms, experience compounding returns that strengthen their adaptive capacity. The significant NQP × Edu interaction further supports the idea that adaptive capacity emerges from co-evolving knowledge infrastructures and absorptive institutions. While IS and Fin remain relevant, their relatively weaker and flatter contributions imply that static endowments alone do not produce resilience unless embedded in dynamic learning systems. This pattern aligns with the emphasis on related variety [30,53], where structural diversity gains evolutionary significance only through recombinatory potential. Meanwhile, the negative SHAP values for climate variables reflect geographically embedded constraints that limit the scope for institutional adaptation, reaffirming that SER is conditioned by local path dependencies. In sum, the technological–industrial dimension of NQP at the provincial scale not only surpasses legacy factors but serves as a resilience-generating mechanism, enabling adaptive reorientation rather than mere recovery (see Figure 6).

At the city scale (Figure 7), NQP no longer dominates as the primary driver of SER, suggesting that the mechanisms operating at provincial and municipal levels differ fundamentally in scale and composition. However, NQP’s interactions with IS and Edu remain statistically significant, indicating that its predictive importance varies with local structural contexts rather than operating as a universal catalyst. SHAP dependence plots reveal a pronounced non-linear threshold pattern: below a certain accumulation level, NQP contributes minimally to SER; above it, marginal returns rise sharply. This threshold behavior reflects what Grillitsch et al. [35] describe as path shaping, the idea that regional transformation requires surpassing critical capability thresholds before self-reinforcing dynamics take hold. Cities below this threshold face a capability trap: investments in new quality development may generate limited resilience returns because the local system lacks sufficient complementary assets to internalize and diffuse new knowledge. Conversely, cities that have crossed this threshold, often those with established industrial clusters, research universities, or mature innovation networks, demonstrate accelerating returns as NQP investments trigger cascading processes through existing institutional and cognitive infrastructures.

The significant NQP × IS interaction at the city scale deserves particular attention. It suggests that the same level of NQP investment produces divergent resilience outcomes depending on industrial composition. Cities with diversified yet related industrial structures appear better positioned to leverage NQP for resilience enhancement, as technological upgrading in one sector can spill over into adjacent activities. In contrast, cities with either highly specialized or fragmented industrial bases may struggle to translate NQP into adaptive capacity, even when investment levels are comparable. This finding resonates with Asheim et al.’s [53] argument that innovation assets must be recombined through context-specific routines to yield adaptive outcomes. Simply possessing technological capabilities is insufficient without industrial structures that facilitate their diffusion and application. Moreover, the persistent importance of the NQP × Edu interaction at the city scale reinforces that human capital serves as a critical mediating factor. Cities with stronger educational foundations demonstrate greater capacity to absorb and deploy new technologies, suggesting that workforce adaptability functions as a binding constraint on how effectively NQP translates into resilience. This is particularly relevant for second-tier cities attempting to upgrade their economic structures: without concurrent investment in educational infrastructure and skill development, technological upgrading initiatives may fail to take root.

The diminished standalone effect of NQP at the city scale, combined with its significant interactions, reveals an important structural reality: municipal-level resilience emerges from the interplay between new capabilities and existing endowments. Unlike provinces, which may rely more heavily on policy coordination and large-scale resource mobilization, cities depend on finer-grained institutional complementarities and localized knowledge networks. This suggests that city-level strategies for resilience enhancement must be more diagnostically targeted, focusing on identifying and strengthening the specific structural bottlenecks, whether industrial composition, educational capacity, or institutional coordination, that constrain NQP-SER associations in particular urban contexts.

At the enterprise scale (Figure 8), NQP is not among the dominant predictors of firm-level SER; firm fundamentals such as Size, Growth, and Employ carry the largest predictive contributions, while the marginal contribution of NQP is comparatively limited. Together with the city-scale results, where NQP likewise does not dominate, this reveals a clear scale-heterogeneity pattern: the predictive salience of NQP for SER is concentrated at the provincial scale and attenuates at the finer city and firm scales, where local industrial structure and firm-specific fundamentals become the primary correlates of resilience. The threshold pattern observed at the firm level carries distinct implications from those at higher spatial scales. For enterprises, the threshold represents an organizational capability boundary: firms below this point possess insufficient absorptive capacity to effectively internalize and deploy new knowledge, even when external resources are available. Once firms cross this threshold, typically through accumulated R&D experience, skilled workforce development, or sustained innovation investment, they enter a regime where additional NQP investments trigger disproportionate resilience gains. This is consistent with the capability accumulation model in evolutionary economics [30], where incremental learning builds upon prior knowledge stocks in a path-dependent manner.

The significant interactions between NQP and both Employ and Size reveal that firm-level resilience is fundamentally relational rather than additive. Larger firms with more employees do not automatically translate NQP into resilience advantages. Instead, the outcome depends on how these organizational resources are configured to support knowledge absorption and recombination. This finding challenges simplistic assumptions that scale alone confers adaptive advantages. In practice, we observe considerable heterogeneity: some large firms with high NQP demonstrate exceptional resilience through institutionalized innovation routines and diversified knowledge bases, while others remain rigid despite comparable resource endowments. Conversely, smaller firms with modest but strategically focused NQP investments, often in niche technologies or specialized capabilities, can achieve disproportionate resilience through organizational agility and rapid reconfiguration capacity. Meanwhile, this heterogeneity in firm-level resilience under similar external shocks illustrates a critical insight: SER at the enterprise level reflects not just exposure to common macroeconomic conditions, but distinct trajectories of internal capability formation. Firms within the same city, facing identical policy environments and market disruptions, exhibit divergent resilience outcomes based on their accumulated technological competencies and organizational learning processes.

Importantly, firm-level resilience is not formed in isolation but is conditioned by the city- and provincial-level environments in which firms are embedded, including access to regional innovation resources, infrastructure endowments, talent pools, and institutional support for NQP. We also clarify the theoretical rationale for extending the SER concept to the micro scale: although SER is conventionally a territorial construct, firm-level performance resilience—measured as the deviation of firm output growth from its expected benchmark—captures the micro-foundations of regional resilience, since aggregate territorial resilience ultimately emerges from the adaptive capacities of constituent firms. We acknowledge that this firm-level adaptation of SER is a simplification and discuss its conceptual limits in Section 5. This pattern helps explain the spatially uneven development observed at higher scales—regional and urban resilience ultimately aggregates from heterogeneous firm-level adaptive capacities, which themselves depend on long-term investment in knowledge-intensive capabilities.

The enterprise-level findings also shed light on why city-scale associations appear more muted: municipal resilience represents a composite of highly varied firm capabilities, where averaging across heterogeneous enterprises may obscure the strong associations visible at both macro (provincial coordination) and micro (firm implementation) levels. For policy, this suggests that effective resilience-building requires attention not only to aggregate regional conditions but to the distribution of capabilities across the firm population. Strategies that raise the floor, helping lagging firms cross critical capability thresholds, may be as important as those that push the frontier through support for leading innovators.

4.1.2. The Institutional Dimension: Considering Spatial Spillover Effect

Table 3 reports a SGDID model with two-way fixed effects (province and year). Both the policy-intensity treatment and the NQP level are standardised so the coefficients are directly interpretable. Year fixed effects absorb the common 2023 shock; the SDM includes spatially lagged covariates to capture cross-border spillovers. Under this specification, the institutional policy-intensity term is negatively associated with SER both within the province and across neighbours: direct effect −0.819 (p < 0.01), indirect effect −1.522 (p < 0.01), total effect −2.341 (p < 0.01). The NQP level shows a similar negative pattern (total effect −1.089, p < 0.01). Road infrastructure is positively associated within the province (direct effect 0.111, p < 0.05). The spatial autoregressive parameter is insignificant (rho = 0.055, p = 0.41), and model fit is high (pseudo R² = 0.788). We interpret these as associations, not causal effects. Together with the provincial-scale predictive evidence in Section 4.1.1, where the technological–industrial dimension of NQP is the strongest predictor of SER, these findings point to the importance of substantive industrial-innovation capability rather than institutional policy intensity per se.

The SGDID estimates add an important qualification to the theoretical expectation. Once provincial heterogeneity and common temporal shocks are absorbed, heightened institutional policy intensity is on average negatively associated with above-expectation SER, suggesting that the enabling potential of institutional policy does not materialise uniformly across provinces. Within the spatio-temporal framework of policy-led path creation [41], institutional policy intensity is theorised to support capability upgrading and regional resilience, though this enabling potential is acknowledged to be conditional on regional absorptive conditions. The SGDID estimates give empirical content to this conditionality. Once provincial heterogeneity and common temporal shocks are absorbed, heightened institutional policy intensity is on average negatively associated with above-expectation SER. This pattern may reflect measurement limits of the NLP index, given that government work reports are strategic documents whose intensity scores may capture rhetorical prominence rather than substantive implementation depth [33], or a genuine mismatch between formal policy signals and the absorptive capacity required for productive capability reconfiguration. Consistent with the institutional entrepreneurship perspective [37], the results further reveal substantial spatial heterogeneity in the direction and magnitude of these associations, underscoring that the relationship between institutional policy intensity and SER is highly context-dependent.

The path creation literature [41] anticipates that institutional policy design can create enabling conditions and redirect regional development trajectories, while simultaneously recognising that this enabling potential is conditional on regional absorptive capacity. In regions where institutional foundations are weak, higher policy intensity may instead induce resource misallocation, reduce adaptive diversity, or generate friction between imposed development trajectories and entrenched local path structures [54,55]. One plausible, though not directly tested, mechanism involves a crowding-out effect whereby heavy-handed policy signals suppress the endogenous experimentation and local knowledge generation that underpin genuine adaptive capacity. This concern echoes long-standing arguments in evolutionary economics regarding the limits of institutional engineering, since policy can direct resources and set strategic priorities but cannot fully substitute for the organic processes of capability-building and knowledge recombination that emerge from decentralised learning [41,54]. Regions that over-rely on policy mandates therefore risk constructing brittle resilience structures that perform well under anticipated conditions but lack the flexibility to respond to unexpected disruptions.

The results are consistent with the presence of spatial spillover patterns, suggesting that policy intensity in neighbouring regions may be associated with local SER outcomes. This spatial interdependence operates through multiple channels. First, successful policy implementation in adjacent provinces generates demonstration effects, providing templates that neighboring regions can adapt to their own contexts. Second, interregional knowledge flows, facilitated by labor mobility, supply chain linkages, and collaborative innovation networks, allow policy-induced capabilities in one region to diffuse across administrative boundaries. Third, competitive pressures emerge as regions observe neighbors’ resilience gains, spurring emulation or differentiation strategies. These spatial spillovers underscore that NQP operates not as a purely localized intervention but within broader interregional innovation systems. The associations of provincial policies with SER depend partly on the institutional landscape of surrounding regions, suggesting that resilience-building is fundamentally a multi-scalar and networked process. This has implications for policy coordination that isolated provincial initiatives may achieve suboptimal results compared to coordinated strategies that account for spillover dynamics and complementarities across administrative units.

Importantly, the NQP indicator representing the technological–industrial foundation remains a consistently strong predictor of SER throughout the SGDID models. These two components reveal an important asymmetry that the endogenous technological–industrial NQP level is positively associated with SER, while exogenous institutional policy intensity is negatively associated, pointing to a divergence between substantive capability and formal policy intensity. Regions with high baseline technological–industrial NQP demonstrate stronger adaptive capacity through internalized processes of knowledge recombination and structural renewal, largely independently of formal policy intensity. Even in the absence of formal policy intervention, regions with high endogenous NQP levels exhibit greater adaptive capacity. This reflects internalized processes of knowledge recombination, path branching, and structural renewal [23,30]. Such regions have developed institutional routines and organizational capabilities that enable continuous adaptation, independent of top-down policy stimuli. This suggests that while NQP policies can accelerate resilience-building, they function most effectively as catalysts for regions that have already initiated capability development, rather than as substitutes for foundational investments in knowledge infrastructure. The combined evidence from technological–industrial (Section 4.1.1) and institutional (Section 4.1.2) dimensions reveals NQP as a multi-faceted construct whose two components relate to SER in empirically opposite directions. This asymmetry underscores that substantive capability accumulation, rather than policy intensity per se, is what the evidence most consistently associates with stronger SER, and that context-sensitive policy design focused on genuine absorptive capacity is essential.

4.1.3. Robustness and Validity Checks

To assess the stability of these associations, we conduct two checks. First, a parallel-trend test regressing SER on a Year-trend by High-NQP-intensity interaction term for the pre-treatment period (2018–2023) yields an interaction coefficient of 0.012 (p = 0.465), indicating that high- and low-intensity provinces did not follow divergent trends before the 2023 policy year. Second, the negative institutional-intensity association is robust to the choice of spatial weight matrix: replacing the spatial matrix with an economic-distance matrix yields a direct effect of −0.606 (p < 0.001), close to the baseline. Consistent with our predictive framing, we treat these spatial-panel results as associations rather than causal estimates.

4.2. Spatial Patterns of NQP’s Associations with SER

At the provincial scale, GGPR results (Figure 9) reveal that NQP, particularly through its spatial interaction term (NQP × GEO), emerges as the dominant predictive factor for spatially differentiated SER. GeoShapley rankings indicate that both the standalone and spatially embedded associations of NQP surpass those of traditional variables such as IS and Fin, corroborating the XGBoost–SHAP findings from Section 4.1.1. This strengthens the evidence that NQP emerges as a primary predictor of adaptive capacity. It suggests that NQP’s contribution to resilience is not merely a function of its absolute level within a region, but depends critically on how that capability is embedded within specific territorial contexts. Regions with comparable NQP investments may exhibit divergent resilience outcomes depending on their geographic positioning, accessibility to knowledge networks, and proximity to complementary capabilities in neighboring areas. This spatially contingent pattern substantiates the evolutionary economic geography perspective that adaptive transformation is inseparable from the territorial embedding of knowledge and capability systems [31,44]. The results also reveal a boundary effect in NQP’s marginal contribution: its association with SER declines after reaching a certain threshold. This pattern suggests that capability accumulation alone does not guarantee proportional resilience gains indefinitely. Beyond a critical point, additional NQP investments may encounter diminishing returns, potentially reflecting saturation in absorptive capacity or the emergence of new forms of lock-in when recombinatory potential plateaus [30]. Regions at this threshold face a distinct challenge: maintaining resilience requires not simply more of the same capabilities, but qualitative shifts in how existing knowledge stocks are recombined and redeployed.

The spatial distribution of NQP’s associations reveals clear regional clustering patterns. Coastal provinces and economically advanced regions show the strongest NQP-SER associations, while interior and peripheral provinces show more modest contributions. This spatial heterogeneity reflects the uneven distribution of cognitive, infrastructural, and institutional resources across China’s economic geography. Regions with dense innovation networks, diversified industrial structures, and well-developed institutional frameworks appear better positioned to translate NQP investments into resilience outcomes. However, the findings also point to emerging spatial dynamics that complicate this core-periphery pattern. Certain central and western provinces demonstrate NQP-SER associations that exceed what their traditional economic indicators would predict, suggesting that targeted capability-building in these regions is beginning to yield resilience dividends. This indicates that while spatial advantages remain consequential, they are not entirely deterministic—regions can partially compensate for geographic disadvantages through strategic investments in knowledge infrastructure and institutional capacity. The spatial clustering of strong NQP-SER associations also highlights the role of agglomeration economies and knowledge spillovers in amplifying resilience. Provinces situated within or adjacent to major economic zones benefit not only from their own NQP investments but from proximity to complementary capabilities in neighboring regions. This creates reinforcing dynamics where spatial concentration of knowledge-intensive activities generates collective resilience that exceeds the sum of individual regional contributions. Conversely, isolated provinces with limited connectivity to innovation networks struggle to fully leverage their NQP investments, as the benefits remain localized without broader diffusion mechanisms.

These findings demonstrate that NQP’s associations with SER are shaped fundamentally by structural coupling between innovation inputs and spatial configurations. The same level of technological capability or institutional support produces different resilience outcomes depending on where it is deployed and how it interacts with surrounding geographic contexts. This geographic embeddedness suggests that resilience-building strategies cannot be standardized across regions but must account for position-specific advantages and constraints within broader spatial systems. The boundary effect and spatial clustering patterns together reveal a tension in regional development: while capability accumulation remains essential, its marginal returns depend on both internal thresholds and external spatial relationships. Regions approaching saturation in standalone NQP associations may need to reorient their strategies toward enhancing connectivity and knowledge flows with other regions, effectively leveraging spatial interdependencies to sustain resilience trajectories. Meanwhile, peripheral regions must balance investments in building foundational capabilities with efforts to overcome spatial disadvantages through improved linkages to core innovation networks.

At the city level (Figure 10), NQP retains the highest GeoShapley value, confirming its continued importance in explaining SER at finer spatial scales. However, the nature of its associations shifts notably compared to the provincial level. Marginal returns to NQP diminish more sharply at the city scale, with SHAP dependence plots revealing a decelerating effect beyond mid-level quantiles. This pattern indicates that urban systems encounter tighter constraints on capability recombination than their provincial counterparts. Several mechanisms likely account for this more pronounced diminishing return. Cities possess more limited combinatorial depth than provinces; their industrial bases are typically narrower, their knowledge networks less diverse, and their institutional capacities more specialized. As a result, cities may reach saturation points more quickly, where additional NQP investments struggle to find complementary assets for productive recombination. This constraint reflects a fundamental tension in urban resilience: while cities serve as primary sites of innovation and economic dynamism, their smaller scale simultaneously limits the scope for continuous capability expansion within bounded territorial systems.

Spatially, NQP’s associations with SER at the city level show greater fragmentation than at the provincial scale. High-performing zones concentrate in coastal innovation hubs such as Shenzhen, Suzhou, and Qingdao, alongside emergent inland cities including Chengdu and Changsha. These locations form discrete but functionally strategic nodal clusters within broader regional innovation systems. The clustering pattern suggests that city-level resilience depends not only on internal capabilities but on positioning within multi-city networks where knowledge spillovers, supply chain linkages, and institutional coordination generate collective adaptive capacity. The emergence of inland cities like Chengdu and Changsha as high-NQP zones is particularly noteworthy. It indicates that geographic position, while influential, is not entirely deterministic. These cities have developed specialized technological strengths and institutional innovations that partially compensate for their distance from coastal markets and global knowledge networks. Their inclusion among high-performing clusters demonstrates that strategic capability-building can enable cities to overcome certain spatial disadvantages, though the threshold effects and sharper diminishing returns observed at the city scale suggest these advantages remain more constrained than at provincial levels.

The spatial distribution also reveals that functional heterogeneity intensifies at the city scale. Cities with comparable NQP levels demonstrate divergent resilience outcomes based on differences in industrial composition, institutional quality, and connectivity to broader innovation networks. Shenzhen and Suzhou, for instance, both exhibit strong NQP-SER associations but through distinct mechanisms—Shenzhen through entrepreneurial dynamism and technological frontier-pushing, Suzhou through advanced manufacturing integration and foreign technology absorption. This heterogeneity underscores that cities vary not just in NQP capacity, but in their ability to leverage those capabilities for adaptive transformation. Some cities effectively use their NQP investments to open new development trajectories and diversify their economic structures, exhibiting the path-branching dynamics central to evolutionary resilience. Others, despite similar capability investments, experience limited transformation, either stabilizing along existing trajectories without substantive upgrading, or encountering premature lock-in where accumulated capabilities reinforce rather than challenge established development paths. This divergence emphasizes that SER at the city scale is not a linear function of NQP investment levels, but emerges from complex interactions between endogenous capabilities and the specific territorial contexts in which they are deployed.

Comparing across spatial scales, the provincial and city-level findings reveal both scalar asymmetries and structural differentiations in how NQP is associated with SER. At the provincial level, NQP operates primarily through coordination mechanisms, aligning diverse urban centers, facilitating interregional knowledge flows, and providing institutional frameworks that enable collective adaptation. Its associations are more sustained across capability levels, reflecting provinces’ greater combinatorial depth and institutional scope. At the city level, NQP functions more as a targeted catalyst—its associations with SER depend critically on precise matching with local conditions, and its marginal returns decline more rapidly as cities encounter bounded opportunities for further capability recombination within their limited territorial scope. These scalar differences carry implications for understanding regional resilience as a multi-level phenomenon. Provincial resilience emerges partly from the capacity to coordinate heterogeneous cities and balance their divergent trajectories within integrated regional systems. City resilience, conversely, depends more on exploiting specialized advantages and maintaining connectivity to external knowledge networks that compensate for internal combinatorial limits. The cases of Chengdu and Changsha illustrate this dynamic: their resilience gains derive not from isolated capability accumulation but from strategic positioning within national innovation networks and policy frameworks that connect them to coastal knowledge centers. The spatial patterns through which NQP is associated with SER thus differ across scales. At provincial levels, NQP enables coordination and knowledge diffusion across diverse urban centers; at city levels, it catalyzes specialized capability-building constrained by local combinatorial limits. The nodal clustering around cities demonstrates that resilience is not passively inherited from static endowments but actively constructed through ongoing processes of capability formation, institutional adaptation, and strategic positioning within multi-scalar spatial systems.

5. Conclusions

This study demonstrates that NQP’s technological–industrial dimension is positively associated with SER, while its institutional dimension, as measured by policy intensity, shows a negative or non-positive association. This suggests that substantive capability matters more than policy rhetoric. Specifically, the following key findings emerge from our analysis.

First, NQP’s dual dimensions relate to SER through distinct mechanisms, with the technological–industrial dimension exhibiting threshold-driven nonlinearity and the institutional dimension showing a negative association in our estimates. The technological–industrial dimension functions as a catalyst for capability accumulation and adaptive transformation. At the provincial scale, NQP emerges as the most influential predictor of SER, exceeding conventional factors through threshold patterns where resilience accelerates once regions accumulate sufficient cognitive and technological resources. These associations operate through knowledge recombination, path-branching dynamics, and interactions with educational infrastructure. At the city scale, NQP’s associations become more conditional and interaction-dependent, with its contribution to resilience mediated by local industrial structures and absorptive capacities. At the enterprise level, NQP re-emerges as a dominant predictor, surpassing financial indicators and demonstrating that firm-level resilience depends fundamentally on organizational capabilities for knowledge absorption and recombination rather than resource endowments alone. The institutional dimension, by contrast, is not positively associated with above-expectation resilience. In our spatial panel estimates, higher institutional policy intensity is negatively associated with SER both locally and across neighbours. We interpret these as associations rather than causal effects. This pattern underscores that substantive industrial-innovation capability matters more than institutional policy intensity per se, and that context-sensitive policy design, accounting for local absorptive capacities, is essential.

Second, NQP’s spatial associations operate through scalar asymmetries: provincial associations rely on coordination and spillovers, while city associations concentrate in nodal clusters where local conditions moderate the strength of associations. At the provincial scale, GGPR results demonstrate that NQP, particularly through its spatial interaction term (NQP × GEO), dominates the explanation of spatially differentiated SER. The spatial distribution shows clear regional clustering, with the strongest associations in economically advanced coastal provinces and emerging associations in certain inland regions. This clustering reflects the uneven distribution of cognitive, infrastructural, and institutional resources, while also revealing that strategic capability-building can partially compensate for geographic disadvantages. Importantly, spatial spillover patterns indicate that policy intensity in neighboring regions is significantly associated with local resilience outcomes, underscoring that NQP operates within broader interregional innovation systems rather than as purely localized interventions. At the city scale, NQP’s spatial associations become more fragmented yet strategically concentrated. High-performing zones cluster in coastal innovation hubs, forming discrete nodal clusters within broader regional systems. However, marginal returns to NQP diminish more sharply at the city scale than at the provincial level, indicating that urban systems encounter tighter constraints on capability recombination due to narrower industrial bases and more specialized institutional capacities. This scalar asymmetry reveals that provincial resilience operates primarily through coordination mechanisms across diverse urban centers, while city resilience depends more on exploiting specialized advantages and maintaining connectivity to external knowledge networks. The functional heterogeneity intensifies at finer scales: cities with comparable NQP levels demonstrate divergent resilience outcomes based on differences in industrial composition, institutional quality, and network positioning, emphasizing that the strength of NQP-SER associations depends on precise alignment between capabilities and territorial contexts.

Despite these insights, several limitations should be acknowledged, most stemming from the recent introduction of the NQP policy and the resulting scarcity of long-term firm- and city-level data. First, the short post-policy period (since 2023) means that the SGDID associations should be interpreted as predictive, not causal. Second, due to data constraints, SHAP and GeoShapley values reflect predictive feature contributions, not causal estimates; causal identification awaits longer panels. Third, the unavailability of city- and firm-level institutional indicators—a direct result of the policy’s recent implementation—limits the institutional dimension to the provincial scale, and the three scales are analysed separately. Fourth, the NLP-based institutional policy intensity index measures the rhetorical prominence of NQP-related language in provincial government work reports, which may not fully reflect the depth or consistency of actual policy implementation. Provinces with stronger administrative communication capacity may score higher, irrespective of substantive implementation intensity. Independent validation against objective implementation-outcome data, such as enterprise-level NQP adoption records or third-party policy assessments, would further strengthen the robustness of the institutional dimension findings and remains a direction for future research.