Policy Pathways to Digitalization: A Qualitative Comparative Analysis of China’s Construction Industry

Abstract

Understanding how policy mixes influence systemic digital transformation is a key concern in socio-technical transition research. This study explores the configurational effects of policy tools in construction digitalization to address this issue. Based on an analysis of policies issued across 31 Chinese provinces (2020–2025), key policy tools were identified using a Multi-Level Perspective (MLP), followed by fuzzy-set Qualitative Comparative Analysis (fsQCA) to examine their configurational effects. The results show that: (1) policies can be grouped into three categories—technology application, management norms, and industrial ecology—comprising six thematic tools; (2) no single condition is essential for high or low digital development; (3) high-level outcomes are associated with three distinct configurations: “technology-talent driven”, “foundation-governance linkage”, and “system-ecology leading”, while “factor-deficient” and “system-mismatched” types are frequently linked to low-level outcomes. By identifying equifinal pathways and cross-level coordination mechanisms, the study offers configurational insights for designing digital policy mixes. While the theoretical implications are broadly applicable, the specific configurations require validation across different contexts because they depend on China’s provincial policy data.

1. Introduction

The digital transformation of the construction industry has become a global imperative, driven by the spread of digital technologies and the urgent need to improve productivity, sustainability, and resilience [1]. However, the industry remains highly fragmented across both developed and developing economies, characterized by temporary project teams, multiple stakeholders, and deep-rooted traditional practices—all of which collectively impede the spread of digital innovation [2,3]. In this context, digital transformation is not just a technological upgrade but a systemic socio-technical change [4]. It demands significant investment, institutional restructuring, and cross-organizational coordination, costs that are often too high for individual firms, particularly small and medium-sized enterprises, to manage alone [5]. Consequently, policy intervention is widely seen as essential for overcoming path dependence, addressing market failures, and fostering collective action [6].

The widespread adoption of digital technology has driven the construction industry’s digital transformation, making it a key topic of global research. Existing scholarship in this field has mostly concentrated on three main areas: the effects of policy implementation, inter-policy coordination, and the development of policy trends. For example, some studies use policy evaluation frameworks to assess the effectiveness of digitalization policies systematically [7,8]. Other research has focused on policies related to technologies such as BIM, examining barriers to digital technology development and proposing relevant countermeasures [9,10]. Scholars have also highlighted the role of policies, like low-carbon pilot programs, in fostering the combined growth of urban digitalization and greening [11]. Additionally, comprehensive reviews have systematically outlined global policy frameworks and trends for promoting construction industrialization [12]. While these studies have enhanced the understanding of individual policy tools, they mainly adopt a variable-oriented perspective.

However, existing studies mainly treat policies as separate, standalone interventions. In reality, policies are implemented as combinations of multiple tools that can interact in complex, nonlinear ways, producing either synergistic or antagonistic effects [13]. There remains a lack of detailed understanding of how different policy tools are combined to drive systemic transformation, especially regarding the following questions:

Q1: In the context of digital transformation, which types of policy tools usually form the basis of policy mixes?

Q2: Are there any individual policy tools that serve as necessary conditions for achieving high-level development outcomes?

Q3: What configurations of policy tools are sufficient to generate high-level digital development, and conversely, which configurations are linked to low-level outcomes?

These questions address a theoretical concern in policy research and socio-technical transition studies: understanding how policy mixes function to drive systemic change. Answering these questions requires an empirical context where variations in policy approaches and outcomes can be observed. The digitization policies issued by China’s 31 provincial governments during the 14th Five-Year Plan (2021–2025) provide an ideal empirical setting [7]. As the world’s largest construction market, China offers a critical case where rapid digitalization intersects with institutional fragmentation, increasing the importance of policy coordination. The diverse provincial mixes within a unified national framework further create a natural laboratory to examine configurational effects. Although based on Chinese data, the identified mechanisms of equifinality and socio-technical transitions are theoretically applicable more broadly. Operating within a unified national framework, each province developed distinct policy mixes, providing an ideal environment to study configuration effects. While the empirical focus is on China, the insights gained regarding the logic of policy tool combinations can be generalized to other contexts where governments aim to advance industry-wide digital transformation through coordinated efforts. Therefore, this study offers conceptual insights for governments managing industry-wide digital transformation beyond China.

Therefore, this study investigates the complex interactions among policy tools to enhance theoretical understanding of policy mixes and to guide their design. It focuses on digital transformation policies issued by China’s 31 provincial governments during the 14th Five-Year Plan period. The research utilizes a Latent Dirichlet Allocation (LDA) topic model to identify key types of policy tools. Subsequently, it employs fuzzy-set Qualitative Comparative Analysis (fsQCA) to uncover the configuration pathways and synergistic mechanisms arising from different combinations of policy tools.

The rest of the paper is structured as follows. Section 2 reviews the literature; Section 3 introduces the LDA research method and the data; Section 4 discusses the fsQCA research method and presents the data analysis results; Section 5 covers the discussion and limitations; and finally, the paper concludes with the main findings.

2. Literature Review

2.1. Policy-Driven Digital Transformation in Construction

Policy tools are essential means by which governments promote and implement policy measures to achieve policy objectives [14]. Since the 1980s, the theory of policy tools has evolved and become a key research topic in policy science and public administration [15,16,17]. Given that the digital transformation of the construction industry faces core challenges such as high fragmentation, strong cost sensitivity, and a complex regulatory system, its smooth progress depends heavily on systematic, collaborative policy interventions [18].

The existing literature primarily focuses on three areas: types of policy tools, policy content, and policy evaluation. (1) Discussions about types of policy tools highlight the diversity of the “policy toolkit” [19,20]. For example, Lowi, T. J. [21] classify policy tools into regulatory and non-regulatory types based on coercion levels; Rothwell [22] separates them into supply-side, demand-side, and environmental types according to decision-making factors such as technological innovation; Howlett and Ramesh [23] categorize them as mandatory, mixed, and voluntary types depending on the level of government involvement; McDonnell and Elmore [24] group them into command, incentive, capacity-building, and system-transforming types based on the needs of those targeted by policy implementation. (2) The discussions around policy content mainly concern the development status of specific policy types. For instance, Xie et al. [25] analyzed policy and standard texts related to BIM technology, clarified the current application status, and explored future directions for BIM in the sustainable, information-based development of architecture. Zhou and Wu [26] compiled policies related to the informatization of the construction industry, used text analysis to reveal the value structure of these policies, and identified development trends in the sector’s informatization. Jin et al. [27] reviewed global policies promoting the industrialization of the construction industry to examine their interrelationships and development trends comprehensively. (3) Research on policy evaluation primarily employs content analysis, which involves the comparative examination of high-frequency words and subject terms within policy texts to assess policies. For example, Chang et al. [28] used content analysis to study China’s policy system aimed at promoting the shift to sustainable buildings; Luo et al. [29] conducted content analysis of policies related to prefabricated construction, assessed policy tools across different stages, and proposed a roadmap to optimize existing policies in this area. A bibliometric analysis by Adekunle et al. [30] suggests that the adoption of digital transformation across construction sectors is uneven, hindered by factors such as cultural resistance, high costs, and skills shortages. A systematic review by Berlato et al. [31] confirms that data fragmentation, limited interoperability, and isolated organizational processes remain systemic barriers to the development of integrated digital ecosystems.

At the policy practice level, governments worldwide have adopted differentiated strategies to tackle these challenges. The European Union advances a dual agenda of digital and green transformation, concentrating on improving value chain transparency through digitalizing permitting processes. Through a literature review covering the progress of circular economy policies and recent research in China and Europe, Zhang et al. [32] highlight that policymakers should consider integrated policy packages when designing policies. This ensures consistency, coherence, and stability in policy formulation and implementation, incorporates the social dimension into circular economy policies, and adapts relevant policies to different circumstances. Research in developing countries shows that infrastructure readiness and institutional capacity are critical bottlenecks hindering policy implementation [33].

However, existing research on digital development policies in the construction industry mainly concentrates on assessing the effectiveness of individual policy tools or categorizing policies in a static way. It rarely investigates the interactive mechanisms and collaborative logic among various tools within complex systems.

2.2. The Perspective of Configuration Analysis of Policies

Policy analysis dominated by linear thinking often fails to systematically examine the interconnections among policy tools, and a non-systematic research paradigm struggles to capture their complex synergistic relationships [34]. To thoroughly reveal the internal mechanisms of policy mixes, it is necessary to introduce the perspective of configurational analysis. Existing research indicates that the effectiveness of policy interventions does not arise from a single instrument but from the interplay of multiple elements [35]. Moreover, traditional symmetric research methods struggle to address complex issues characterized by multiple conjunctural causation [36,37]. Fuzzy-set qualitative comparative analysis (fsQCA) is a set-theoretic approach that examines the necessary and sufficient relationships between antecedent conditions and outcomes. It offers a novel perspective on explaining causal complexity, including multiple concurrent causes, causal asymmetry, and equifinality [38,39].

The construction industry is increasingly adopting fsQCA to examine the configurational effects of digital transformation. Wang et al. [40] utilized fsQCA to identify multiple pathways for integrating digital and low-carbon initiatives that deliver high efficiency. Additionally, fsQCA has been broadly applied in research on digital transformation in manufacturing, urban digital ecosystems, and small and medium-sized enterprises [41,42].

This approach provides a methodological basis that more accurately reflects the complexities of the real world for understanding which policy combinations are effective.

2.3. Multi-Level Perspective Theoretical Framework

A purely configurational analysis risks falling into a descriptive trap. While it can identify which combinations of conditions are effective, it often struggles to explain why these conditions come together and at what level of change they operate collectively. The Organization for Economic Co-operation and Development (OECD) notes in its report “System Innovation: Synthesis Report” that system innovation involves a cross-sectoral policy approach. It aims to build new socio-technical systems through synergistic interactions among system elements [43]. The Multi-Level Perspective (MLP) is a well-established framework for analyzing such transformations within socio-technical systems [44]. In the field of architecture, the MLP has been widely applied to analyze systemic transitions in areas such as smart buildings, digital building permits, and the digitization of energy retrofit processes [45]. Bakhuis et al. [46] systematically reviewed the literature on sociotechnical multi-system innovation frameworks that extend the usual focus on a single sociotechnical system to include influences from multiple systems. The review covers 75 peer-reviewed papers spanning a broad range of energy-intensive systems and primarily builds on the core frameworks of the Multi-level Perspective (MLP) and Technological Innovation Systems (TIS).

The MLP of socio-technical transition theory suggests that transitions are not driven by a single factor; instead, they arise from the combined effects of internal development across three levels and their interactions among them: the macro-level landscape, the meso-level sociotechnical regime, and the micro-level technology niches [47]. The top-down perspective, exemplified by the work of Sabatier and Mazmanian [48], highlights policy goal clarity, hierarchical control, and compliance mechanisms as essential for successful implementation. This view argues that clear policy directives, adequate resource allocation, and effective monitoring systems are vital, especially in government-led digitalization initiatives. In the construction industry, this approach is reflected in the importance of national strategies and regulatory mandates, such as the BIM mandates in the United Kingdom, in promoting industry-wide adoption [49]. Conversely, the bottom-up perspective concentrates on the discretionary power and adaptive actions of frontline implementers, including local governments, industry associations, and enterprises. This outlook stresses the importance of local capacity, stakeholder participation, and the institutional environment in influencing policy outcomes [50]. In the context of building digitization, it explains why uniform national policies can produce significantly different results due to disparities in local administrative capacity, the maturity of digital infrastructure, and industry readiness levels [8].

Integrating the above perspectives shows that successful policy implementation relies on a dynamic balance between policy design and the implementation environment. This balance is shaped by several factors, including inter-organizational coordination, resource sufficiency, and the alignment between policy tools and the characteristics of target groups. System innovation is also a process driven by the combined forces of pressure on the regime level from the socio-technical landscape, challenges and threats posed by niche developments, and the regime’s own internal renewal, which together propel the system’s overall transformation and innovation [51].

Therefore, the MLP offers a clear conceptual framework for understanding points of intervention and the systemic nature of policy actions. Policy tools can be seen as strategic interventions designed to influence a particular level or promote interactions across levels [52]. Integrating the MLP with policy analysis enables us to reframe effective policy tools as coordinated intervention strategies that operate across multiple levels of the MLP. This approach provides a deeper understanding of the systemic logic behind their effectiveness.

3. Quantitative Analysis of the Digital Development Policy of the Construction Industry Based on the LDA Model

The LDA (Latent Dirichlet Allocation) topic model is a text-mining tool based on machine learning and natural language processing [53]. It can effectively identify multiple latent topics and their interrelationships within a text. This method not only has strong data-processing capabilities but also produces highly explainable results, helping to reduce subjective bias that may arise during manual coding [54]. Given that policy texts contain substantial unstructured information and feature complex, multidimensional content, the LDA model is employed to model topics in policy documents. This approach is used to identify policy tool types and conduct quantitative analysis.

3.1. Case Selection

This study employs a subnational comparative design, analyzing China’s 31 provincial-level administrative divisions. Within the theoretical framework of an MLP, the provincial level holds critical theoretical significance, uniquely operating across all three tiers: translating national strategies (macro-context), implementing institutional rules (regime level), and fostering local innovations (niche level). This makes provincial policies an ideal window for observing cross-level interactions. The policy texts focus on the 14th Five-Year Plan period (2021–2025), a phase of accelerated national digitalization. The design leverages a quasi-experimental approach in which institutional homogeneity within a unified national framework coexists with significant policy variation. This enables a controlled comparative analysis of policy-mix effects while reducing interference from confounding factors commonly encountered in other cross-national studies.

Hence, this study examines policies related to the digital transformation of the construction industry that were developed and implemented by the 31 provincial-level governments in China between 1 January 2020 and 30 November 2025. Using keywords such as “digital” OR “digital and intelligent” OR “intelligent” AND “building” OR “construction” for retrieval, a total of 335 relevant policy documents were collected from the State Council’s policy document database, official government websites at various levels, and government open data platforms. To ensure data validity, three researchers independently reviewed and screened the collected materials. Texts unrelated to the research topic were excluded: (1) response-type documents, i.e., documents that merely replied to superior directives or policy inquiries without proposing substantive measures (14 documents); (2) meeting or routine matter notifications, involving procedural matters rather than policy content (9 documents); (3) duplicate or superseded versions, i.e., documents later replaced by new policies (3 documents); (4) documents with only peripheral relevance, texts that only mentioned the construction industry in passing within broader digital economy policies (6 documents). In total, 32 texts were excluded, and 303 texts highly relevant to the digital transformation of the construction industry were retained for analysis. Additionally, a supplementary search using the term “infrastructure” was conducted to ensure thorough coverage. No additional relevant literature was found beyond the initial collection.

3.2. Application Process of the LDA Model

3.2.1. Data Cleaning

To ensure the standardization and purity of the analytical data, this study performed standardized preprocessing on the collected raw Chinese policy texts. First, non-Chinese characters (including special symbols, numbers, and English letters) and unnecessary whitespace were removed from the texts to reduce noise interference. Next, the “Jieba” Chinese word segmentation tool was used to segment the texts. A stop-word list was then applied to filter out function words and purely numeric content that contribute little to meaning, retaining only core semantic units such as nouns and verbs. This processing pipeline cleans the data while maintaining as much of the semantic structure and specific context of the policy texts as possible. Consequently, it ensures that subsequent analytical results are grounded in the Chinese policy context, thereby guaranteeing the objectivity and accuracy of the main analytical process.

3.2.2. Corpus Construction

The preprocessed texts were represented using a bag-of-words model, which ignores word order and grammatical structure, to create a dictionary and a document-term frequency matrix. Each document is converted into a high-dimensional vector, where the dimensions correspond to the unique terms in the vocabulary, and the values indicate the frequency of each term within the document. This process converts the text data into a structured corpus that serves as input to the LDA model.

3.2.3. Determining the Number of Topics

Two metrics, coherence and perplexity, were used to determine the number of topics [55]. The coherence score assesses the semantic relevance and cohesion of words within a topic, with higher scores indicating clearer and more interpretable topic divisions. Perplexity measures clustering quality; lower perplexity indicates stronger model performance and better clustering. The model’s coherence and perplexity were calculated for different numbers of topics. As shown in Figure 1, when the number of topics is 6, the coherence score is relatively highest, and the perplexity is relatively lowest. Therefore, this study ultimately selected 6 topics to ensure the model has high practical significance and interpretability in policy text analysis.

3.2.4. Topic and Cluster Word Visualization

After determining the number of topics, Python 3.9.9 was used to train the LDA model with the gensim library. The model was trained with six topics, running fifteen full passes over the corpus. The hyperparameters were set to be learned automatically from the data, and a fixed random seed was used to ensure reproducibility. The pyLDAvis package 3.4.0 was utilized to visualize the topics, as shown in Figure 2. In this figure, each topic is represented by a bubble, with larger bubbles indicating higher frequencies for that topic. The bubbles are evenly distributed across the quadrants without overlap, demonstrating that the model can identify multiple distinct topics in the data and that these topics are strongly independent. Figure 3 displays the feature words and their weights for the top ten high-frequency thematic clusters under topics 1 through 6.

3.3. Policy Tool Analysis from the MLP

According to the MLP framework, systemic innovation emerges from the interaction of three analytical levels: the macro-level landscape (exogenous environment and long-term pressures), the meso-level regime (dominant rules, practices, and institutions), and the micro-level niche (emerging technologies and local innovations). In this study, these three levels serve as an a priori coding framework for systematically mapping the policy themes derived from the textual analysis.

The process of establishing theme levels and classifications employed a hybrid approach that combined inductive content analysis with deductive mapping. First, through close reading of policy documents, inductive thematic analysis, and consultations with three domain experts, the six thematic terms identified earlier were named: digital application, quality control, intelligent review and evaluation, collaborative approval, qualification certification, and industrial upgrading. Next, a clustering operation was carried out, assigning each theme to one of the three levels of the MLP based on its primary domain of action and its theoretical definition in the socio-technical transition literature [52]. Specifically:

Landscape-level themes relate to the broad external pressures and structural shifts that influence the entire industry. Industrial upgrading is classified at this level because it signifies long-term ecological pressures and a macro-level strategic course. Regime-level themes encompass established institutional rules, governance norms, and coordination mechanisms that maintain stability within the system. Tasks such as intelligent review and evaluation, collaborative approval, and qualification certification are assigned to this level because they reshape administrative procedures and regulatory frameworks. Themes at the niche level involve innovative technological applications and localized experiments that diverge from the mainstream regime. Digital applications (like BIM and cloud platforms) and quality control (such as AI-driven defect detection) are situated at this level because they reflect micro-level technological innovations.

Finally, the initial classification results were independently reviewed by the same three experts. For existing disagreements (for example, whether quality control could also be considered a standard at the regime level), consensus was reached through discussion, concluding that its primary manifestation in policy texts is as a tool embedded within niche-level digital applications, rather than as an established institutional rule. The final thematic clustering results, including topic classifications, topic names, and their definitions, are presented in

Table 1. Topic name and its definition.

Topic Level	Topic Classification	Topic Number	Topic Name	Concept Definition
Landscape	Industrial ecology	5	Industrial upgrading	It involves industrial digital upgrading and collaborative innovation. It refers to promoting collaborative and innovative development of the construction industry chain, and facilitating the upgrading of prefabricated construction, industrialization, and intelligence.
Sociotechnical regime	Management norms	3	Intelligent review and evaluation	It involves technical review and evaluation. It refers to establishing a digital system for expert review, survey, and design verification, as well as performance evaluation, to enhance the efficiency and accuracy of review and assessment.
		4	Collaborative approval	It involves multi-departmental collaborative approval. It refers to establishing a digital workflow for joint review and parallel approval by promoting cross-departmental data sharing and business process reengineering, significantly optimizing the approval process for engineering projects.
		6	Qualification certification	It involves qualifications and standards. It refers to establishing market access and evaluation standards compatible with digital capabilities by recognizing and managing the digital qualifications of enterprises and professionals.
Technological niches	Technology application	1	Digital application	It involves data application, leveraging technologies such as big data and IoT to build a platform for the construction industry, enabling data-driven, intelligent construction.
		2	Quality control	It involves the whole-process quality management of the project. It refers to establishing a digital quality supervision system that covers key links such as design, construction, and acceptance, using information technology to achieve whole-process quality traceability and early risk warning, ensuring systematic control of project quality.

.

Based on the six policy themes identified and their classification within the MLP, three analytical dimensions are derived: industrial ecology at the landscape level, management norms at the regime level, and technological applications at the niche level, as illustrated in Figure 4.

Figure 4 depicts a multi-level perspective on digital transformation in construction. At the landscape level, industrial upgrading (ecological pressure) provides external momentum for sector-wide change. The regime level involves management norms—such as intelligent review, collaborative approval, and qualification certification—that restructure institutional rules to create a supportive governance framework. This institutional restructuring is often prompted by external pressure at the landscape level, opening an “opportunity window” for reform. At the niche level, digital applications and quality control emerge as micro-level adoption of technologies like BIM and AI tools by firms. When these innovations accumulate and align with top-down pressure from the landscape, they can challenge and ultimately reshape existing institutional constraints. This interaction sparks a technological and institutional shift, resulting in a new socio-technical system.

In summary, the three levels do not operate in a linear fashion. Instead, they form a dynamically interconnected system characterized by top-down guidance, bottom-up innovation, and cross-level interactive evolution, collectively driving the industry’s digital transformation [56].

4. Analysis of Policy Instrument Combinations Based on fsQCA

Building on the quantitative analysis of policy tools presented earlier, Section 4 explores how various types of policy tools (condition variables) issued by provincial governments work together to produce different performance outcomes across specific digitalization indicators (outcome variables) in the construction industry.

4.1. Variable Selection and Calibration

4.1.1. Outcome Variable

To reflect the provincial-level development of construction digitalization following policy implementation and to assess its long-term impact, this study selects outcome variables based on key goals and tasks in China’s “14th Five-Year Plan” Construction Industry Development Plan. These include increasing the share of prefabricated buildings in new construction to over 30%, fostering construction-industry internet platforms, developing construction robotics products, and cultivating intelligent construction industrial bases. Additional goals include implementing intelligent construction pilot cities and projects, strengthening R&D, and establishing a standard system for intelligent construction. From these policy targets, three core dimensions are derived: input, output, and innovation indicators. Together, these measures promote provincial digital development in the construction sector. The entropy weight method assigns weights to each indicator, enabling the calculation of a comprehensive digital development index for 31 provinces. This index serves as the outcome variable for the case analysis. Considering the lagged and enduring nature of policy effects, the statistical period for all indicators spans from 1 January 2020 to 30 November 2025.

• Input Indicators: Determined by each province based on the floor area of newly started prefabricated buildings and the proportion of such floor area to the total newly constructed floor area in the previous year. Regions with 30 million square meters or more, or a proportion of 30% or higher, are assigned 5 points; regions with less than 5 million square meters are assigned 3 points; and the remaining regions are assigned 4 points.
• Output Indicators: Points are earned based on the number of “BIM technology demonstration application projects, smart construction sites, intelligent construction pilot projects, intelligent quality projects” in each province, as well as the “pilot cities for digital management reform throughout the entire life cycle of engineering construction projects” and “national intelligent construction pilot cities” announced by the Ministry of Housing and Urban-Rural Development.
• Innovation Indicators: Points are accumulated based on the number of authorized patents related to building digitalization approved in each province.

4.1.2. Condition Variables

Cumulative scoring was performed based on the frequency and strength of mentions of policy topics (Table 1) within provincial policy documents. High-intensity terms were awarded 3 points, including “should”, “must”, “necessary”, “mandatory”, “essential”, “prerequisite”, “without exception”, “full implementation”, “vigorously promote”, and “full coverage.” Medium-intensity terms gained 2 points, such as “need”, “require”, “promote”, “implement”, “advance”, “drive”, “enhance”, and “deepen.” Low-intensity terms received 1 point, including “encourage”, “support”, “advocate”, “guide”, “explore”, “research”, “pilot”, and “layout”. No points were assigned when no relevant terms were mentioned.

Table 2. Keywords related to policy topic.

Policy Topic	Keywords
A1 Digital application	BIM, Building Information Modeling, cloud computing, cloud platform, Internet of Things, IoT, big data, data mining, artificial intelligence, AI, digital twin, Construction Information Modeling, CIM, smart construction site, intelligent construction, digital transformation.
B1 Quality control	Material traceability, material tracking, quality traceability, construction monitoring, process supervision, on-site monitoring, completion acceptance, acceptance filing, engineering acceptance, operation and maintenance, safety investigation, safety supervision.
C1 Intelligent review and evaluation	Intelligent review, automated review, digital review, online review, digitized review, informatization of review, digitalization of review.
D1 Collaborative approval	One-stop service, parallel approval, collaborative approval, engineering construction project approval management system, approval system, approval platform, collaborative approval, joint approval, online approval.
E1 Qualification certification	Qualifications, certifications, certificates, credentials, work permits, bidding bonuses, excellence evaluations, and training.
F1 Industrial upgrading	Prefabricated construction, assembly, green building, energy-saving building, ultra-low energy consumption building, intelligent construction, construction robot, robotic construction, new industrialization of construction, modern industrialization, industrial chain collaboration, industrial chain integration.

displays the keyword categories for each policy topic.

4.1.3. Variable Calibration

The measured variables were calibrated to align with predefined concepts, with the calibrated data indicating the membership degrees of the sets. For the condition variables (policy tools), raw scores range from 0 to 5, indicating increasing frequency and intensity of policy requirements. The distribution shows discrete integer values with natural breaks: 0 = no mention; 1–2 = general or encouraging mentions; 3–4 = moderate to high-frequency requirements; 5 = highest intensity. Due to this discrete and uneven distribution, mechanical percentile thresholds are unsuitable. Instead, we defined theory-informed anchors: full non-membership at raw value 0 (0.0), a crossover point at 2.5 (0.5)—the threshold between encouraging and high-intensity requirements—and full membership at raw value 5 (1.0). For the outcome variable, the data are continuous and relatively evenly distributed (see the data availability statement for specific numerical values). Following standard fsQCA practice, we used the 95th percentile for full membership (0.95), the 50th percentile for the crossover point (0.5), and the 5th percentile for full non-membership (0.05). The fsQCA4.1 software was used to scale the variables to values between 0 and 1, and these results served as the operational data for fsQCA.

4.2. Analysis of Necessity

After calibrating each variable, an analysis was conducted to assess the necessity of conditional variables. A condition is deemed to meet the standard of a necessary condition if its consistency rate exceeds 0.9.

Table 3. Necessity test of conditions.

Conditions	Consistency		Coverage
	High-Level Architectural Digitalization Development (Y1)	Low-Level Architectural Digitalization Development (~Y1)	High-Level Architectural Digitalization Development (Y1)	Low-Level Architectural Digitalization Development (~Y1)
A1	0.686	0.683	0.555	0.581
~A1	0.482	0.477	0.591	0.615
B1	0.690	0.719	0.562	0.615
~B1	0.527	0.488	0.640	0.623
C1	0.621	0.504	0.700	0.598
~C1	0.643	0.748	0.552	0.675
D1	0.785	0.877	0.504	0.592
~D1	0.365	0.266	0.738	0.565
E1	0.651	0.647	0.592	0.618
~E1	0.580	0.573	0.610	0.633
F1	0.707	0.687	0.587	0.600
~F1	0.518	0.527	0.611	0.654

Note: “~” denotes logical negation.

shows that all condition consistency rates are below 0.9, indicating that no single policy instrument is a necessary condition for high or low levels of construction digitalization development. In other words, there is no single “optimal policy”; instead, multiple equivalent “policy configuration paths” can collectively drive the digital transformation of the construction industry.

4.3. Conditional Configuration Analysis

4.3.1. Configuration Paths for High-Level Architectural Digitalization Development

Based on the distribution of cases, solution consistency, and coverage, the original consistency threshold was set at 0.8, the case frequency threshold at 1, and the PRI at 0.65 [57]. Subsequent analyses were conducted using these established parameters. Using fsQCA4.1, three solution types were generated: complex, intermediate, and parsimonious. The intermediate solution served as the basis for analysis, and the conditional variables present in both the intermediate and parsimonious solutions were identified as core conditions. In contrast, those appearing solely in the intermediate solution were classified as peripheral conditions [58,59]. As shown in

Table 4. Configuration analysis of high-level architectural digitalization development.

Configuration Variables	Solution 1	Solution 2	Solution 3	Solution 4
A1	●	●	●	●
B1	⊗	🞄	🞄	🞄
C1		🞄	⊗	⊗
D1	⊗	●	🞄	🞄
E1	●	⊗	●	🞄
F1	⊗	⊗	⊗	●
Raw Coverage	0.212	0.152	0.169	0.184
Unique coverage	0.101	0.023	0.036	0.052
Consistency	1.000	0.854	0.833	0.930
Solution coverage	0.567
Solution consistency	0.825

Note: ● and 🞄 indicate that the condition presents; ⊗ and ⊗ indicate that the condition is absent; Blank spaces indicate that a condition may be either present or absent. ● and ⊗ as the core condition.

, four conditional configurations were identified as collectively contributing to the development of high-level architectural digitalization. These configurations demonstrated a solution consistency of 0.825, indicating strong explanatory power for the outcomes. Additionally, the solution coverage was 0.567, indicating that approximately 57% of high-level cases were explained, confirming good representativeness. Regarding the configurational conditions, corresponding to different market conditions and development focuses, the four configurations are summarized and defined as the “technology-talent driven type”, the “foundation-governance linkage type”, and the “system-ecology-leading type”.

(1)

Technology-talent driven type. Solution 1 combines a strong niche-level digital technology application (A1) with active restructuring of talent qualification systems at the regime level (E1), forming a “technology implementation and talent support” synergy. Guided by the national intelligent construction strategy (landscape), Shangxi Province established an intelligent construction evaluation system. This included creating an expert database and providing full lifecycle technical templates to enhance micro-level diffusion. Simultaneously, it implemented a talent training system through university-enterprise collaboration to cultivate compound talents with both digital and engineering skills. This pathway reflects the MLP logic of “landscape guidance, niche innovation, and institutional adaptation”, advancing the industry’s digital development.

(2)

Foundation-governance linkage type. Solutions 2 and 3 mainly depend on the systematic enhancement of internal governance at the institutional level. They create a closed-loop structure where niche-level digital technology applications (A1) are supported by institutional arrangements for quality control (B1), collaborative approval (D1), and qualification certification (E1). This fosters endogenous transformation even when the dominant industrial ecology (~F1) is not yet fully developed. For example, Shandong Province strives for comprehensive coverage of intelligent construction scenarios across its 16 cities. It has established a complete chain system integrating “technology application, process supervision, and approval collaboration”, actively promotes BIM and IoT for quality and safety oversight, and develops intelligent supervision platforms to improve multi-department coordination. This setup demonstrates how institutions can proactively embed grassroots innovations within a stable environment and propel regional digital growth through standardization and process re-engineering, supporting a gradual “institution-led, niche-embedded” transformation model.

(3)

System-ecology Leading Type. Solution 4 is driven by the development of an industrial ecosystem at the landscape level (F1). It systematically combines institutional elements—such as intelligent review, collaborative approval, and qualification certification—while empowering niche-level technological applications (A1). This establishes a three-tier linkage of “ecological leadership, institutional coordination, and technology implementation”. For example, Shanghai’s digital transformation plan explicitly guides the development of an intelligent construction system covering R&D, standards, scenarios, and enterprise clusters. Within this framework, advanced technologies like BIM, construction robots, and digital twins are integrated throughout the engineering process. This pathway represents an advanced form of construction digitalization, evolving from scattered pilots towards systematic, ecological, and large-scale development.

4.3.2. Configurational Pathways of Low-Level Architectural Digitalization Development

Low-level architectural digitalization development cannot simply be viewed as the reverse of high-level development [60]. To better understand the complex causal relationships and asymmetric effects in architectural digitalization, this study concurrently analyzed the configurational pathways of low-level architectural digitalization development.

Table 5. Configuration analysis of low-level architectural digitalization development.

Configuration Variables	Solution 1	Solution 2	Solution 3	Solution 4
A1	●	⊗	⊗	⊗
B1	⊗		⊗	⊗
C1	⊗	⊗	⊗	⊗
D1		●	●	●
E1	⊗	⊗		🞄
F1	⊗	⊗	⊗	⊗
Raw Coverage	0.291	0.220	0.214	0.164
Unique coverage	0.132	0.029	0.026	0.037
Consistency	0.875	0.779	0.835	0.956
Solution coverage	0.713
Solution consistency	0.803

Note: ● and 🞄 indicate that the condition presents; ⊗ and ⊗ indicate that the condition is absent; Blank spaces indicate that a condition may be either present or absent. ● and ⊗ as the core condition.

presents four configurations for low-level architectural digitalization development, with a solution coverage rate of 71% and a solution consistency of 0.803, indicating strong explanatory power. The results demonstrate that outcomes differ between low-level architectural digitalization development and high-level development, aligning with the concept of “causal asymmetry”.

According to the MLP framework, the four low-level configurations listed in Table 5 can be divided into two types based on whether core conditions are present or absent and the resulting cross-level mismatch patterns. The first type is the “factor-deficient type”, characterized by having only one condition at the niche level (digital application) while lacking all other conditions at the regime and landscape levels. The second type is the “system-mismatched type”, characterized by having only one condition at the regime level (collaborative approval) while missing the technological foundation at the niche level and strategic support at the landscape level. These two categories represent two different mechanisms of configuration failure: one appears as isolated technological applications without institutional embedding, while the other appears as institutional reforms lacking technological and ecological support.

• Factor-deficient type. Solution 1 corresponds to a configuration based on A1~B1~C1~E1~F1, characterized by the presence of digital application (A1) alongside the absence of quality control (~B1), intelligent review (~C1), qualification certification (~E1), and industrial upgrading (~F1). This “single-point advancement with systemic absence” pattern reflects micro-level technological innovation that lacks both institutional adaptation and strategic guidance at the landscape level. Here, the lack of regime-level instruments (~C1, ~E1) is not just a passive gap but an active causal factor: without institutional rules to standardize and legitimize niche innovations, technological experimentation remains fragmented and fails to scale. Therefore, it does not progress from isolated experimentation to a recognized practice.

• Case evidence shows that such policy arrangements, which focus on promoting technology without adequate institutional coordination, often lead to disconnections between digital initiatives and the existing institutional environment. This mismatch hinders synergistic effects and significantly reduces the practical effectiveness of the technology. The pattern demonstrates how a lack of cross-level coordination within the MLP framework obstructs digital transformation.

2.

System-mismatched type: In solutions 2, 3, and 4, this type is based on D1*~A1*~C1*~F1, which indicates collaborative approval (D1) combined with the absence of digital application (~A1), quality control (~B1), intelligent review (~C1), and industrial upgrading (~F1). While these policy designs emphasize collaborative approval to streamline governance, they generally lack the necessary supporting environment. This includes deficiencies in digital infrastructure, quality assurance systems, intelligent review capabilities, and a clear focus on industrial upgrading. In this setup, the lack of niche-level technological foundations (~A1, ~B1) acts as a key constraint: collaborative approval mechanisms cannot function effectively without the digital infrastructure and data interoperability supplied by niche-level tools.

• This configuration represents an attempt at proactive adaptation at the institutional level. However, due to weak technical capacity in the technological niches and insufficient ecological pressure at the landscape level, the institutional innovation lacks a solid foundation for implementation. The result is a structural mismatch characterized by “top-level promotion, mid-level inertia, and bottom-level weakness”.

4.4. Robustness Test

Robustness checks were performed using set-theoretic methods, adjusting the consistency and case frequency thresholds—two complementary and widely accepted approaches for evaluating the stability of the identified configurations [61]. It should be clarified that modifying calibration anchors (i.e., mechanical percentile thresholds), although common in some fsQCA studies, is not appropriate for the calibration in this study, which is based on substantive policy content and the characteristics of the data distribution. The original consistency threshold was raised from 0.8 to 0.85, and the case frequency was set to 2. After these changes, although the number of configurations for high-level and low-level building digitalization development altered, the subset relationship with the original results remained clear. The parameter differences remained within a narrow range, insufficient to produce different substantive interpretations, thus confirming the robustness of the original configuration analysis results.

5. Research Discussion and Limitations

5.1. Research Discussion

5.1.1. High-Level Architectural Digitalization Development in the Construction Industry

The results of the policy configuration analysis show that there is no single optimal path to achieving high-level digital development in the construction industry. Instead, three distinct yet functionally equivalent configurations have emerged: the “technology-talent driven”, “foundation-governance linkage”, and “system-ecology leading”. Each represents a unique combination of policy instruments that collectively contribute to successful outcomes. This finding supports the principle of “equifinality”, demonstrating that within a policy-mix framework, multiple causal pathways can lead to the same result [62].

The underlying logic of each configuration can be understood through interactions across different levels within the MLP. In the “technology-talent-driven” pathway, digital applications at the niche level (e.g., BIM, AI-based quality control) are combined with institutional support at the regime level (e.g., qualification certification), allowing localized technological experiments to expand upward. The “Foundation-Governance Linkage” pathway follows a top-down logic: pressure for industrial upgrading at the landscape level prompts governance reforms at the regime level (e.g., intelligent review, collaborative approval), which then creates institutional space for technology adoption at the niche level. The “System-Ecology Leading” pathway shows a balanced integration across all three MLP levels, where the industrial ecosystem (landscape), management norms (regime), and technological applications (niche) form a self-reinforcing cycle. Notably, although “digital technology application” appears in all three configurations, it never functions alone. Each configuration relies critically on supportive conditions at either the regime or landscape level. This interdependence explains why single policy measures often fail to achieve systemic change [44].

The three configurations correspond with regional differences in developmental stages and institutional capabilities. For example, provinces with advanced digital infrastructure and a skilled talent pool are more prone to follow the “technology-talent-driven” path. Simultaneously, regions with strong administrative coordination tend to adopt the “foundation-governance linkage” path. This pattern shows that successful policy combinations are not universal standards but instead configurations that can utilize existing regional advantages and adapt to local conditions.

Theoretically, these findings expand the MLP framework by illustrating how policy mixes act as cross-level coordination mechanisms. The configurational evidence suggests that effective digital transformation requires the coordinated effort of policies across three levels, rather than treating the landscape, regime, and niche as separate layers. Practically, this means policymakers should go beyond isolated “point-based” interventions and adopt an integrated “systemic” policy approach that ensures complementarity among technological, institutional, and ecological instruments.

5.1.2. Low-Level Architectural Digitalization Development in the Construction Industry

The analysis of the results highlights two emerging patterns linked to low-level digital development: the “factor-deficient” type and the “system mismatch” type. These configurations offer vital counterfactual evidence, supporting the theory derived from high-level pathways.

The “factor-deficient” configuration is characterized by the presence of technology applications (niche-level tools) but a lack of corresponding regime-level regulations or landscape-level ecosystem conditions to support them. This pattern reveals the risk of formalistic digitalization, in which the pace of technology procurement outstrips institutional integration. Without complementary governance mechanisms, such as data standards, interoperability requirements, or procurement incentives, technology applications will remain fragmented and fail to generate systemic improvements. The underlying mechanism stems from the breakdown of cross-level synergy, in which innovations at the niche level are introduced without corresponding adjustments at the regime level, resulting in underutilized platforms and wasted investment.

In contrast, the “system mismatch” configuration occurs when regime-level governance tools (such as AI review systems or collaborative approval platforms) are deployed. However, sufficient niche-level technological foundations or landscape-level industrial ecosystem support are lacking. This highlights a critical dependency: governance reforms rely on the prior or simultaneous development of digital infrastructure (such as BIM platforms, IoT networks, and data-sharing protocols) to operate effectively. When these foundational conditions are absent, governance tools become administratively burdensome without delivering efficiency gains. These low-level configurations, on the other hand, validate the findings from the high-level pathways. They show that the same policy instruments that contribute to success in well-coordinated configurations can lead to failure when used in isolation or mismatched combinations. The contrast between high-level and low-level development pathways emphasizes the importance of functional coupling among policy measures, indicating that policy instruments are not inherently effective or ineffective. Their effectiveness depends critically on how they are combined and whether supporting conditions from other layers of the socio-technical system are present.

These findings provide insights for research on policy mixes, suggesting that a key factor behind low policy performance may be a mismatch in configuration, rather than a lack of any single policy tool. For practitioners, this highlights the importance of sequencing and coordinating policy implementation. Before implementing complex governance reforms, regions must first establish foundational digital infrastructure and ensure that technological development progresses alongside institutional development.

5.2. Research Limitations

This study has certain inherent boundaries. First, the fsQCA results show that the high-level architectural digitalization development configurations identified in this study achieve a solution coverage of about 57% (see Table 4). In configurational research that emphasizes causal complexity over explaining variance, this level of explanatory power is acceptable [61], but it also indicates that the model does not account for other important antecedent conditions. Possible variables include regional economic development levels, local government fiscal capacity, and the path dependence of industrial structures. Additionally, this study focuses solely on the characteristics of policy texts, which reflect government attention allocation and policy commitments, without considering moderating factors such as implementation effectiveness and execution capacity. These elements may all impact the relationship between policy mixes and outcomes. Future research could incorporate multiple data sources, such as fiscal investment intensity, implementation status, and third-party evaluation data, to improve measurement accuracy.

Second, the empirical evidence in this study is entirely based on digital transformation policies issued by China’s 31 provincial governments during the 14th Five-Year Plan period (2021–2025). Although this dataset provides comprehensive spatial coverage and temporal consistency, the findings are limited by China’s institutional, economic, and political context. While the theoretical logic concerning policy-equivalent pathways, multi-level policy coordination from the MLP, and configurational mismatch is generally applicable, the specific configurations identified and their empirical performance need further validation in other national and industrial contexts. Future research could extend the configurational approach to cross-national comparisons, include demand-side and implementation variables, and adopt mixed-methods designs to trace causal processes.

Third, while this study identifies the configurational pathways of policy tools across the levels of the MLP, it does not provide a quantitative estimate of the strength or statistical significance of these cross-level interactions. Future research could consider employing quantitative methods to complement these configurational findings, thereby further testing and refining the interactive relationships among the landscape, regime, and niche conditions. This would offer a more comprehensive perspective for understanding the dynamics of the MLP.

6. Conclusions

This study explores the configurational effects of policy tools on digital transformation within the construction industry. Using digital transformation policies released by China’s 31 provincial governments during the 14th Five-Year Plan period (2020–2025), an LDA topic model was applied to identify key policy tools from an MLP, followed by fsQCA to investigate how different combinations of these tools lead to high or low digital development outcomes. The primary conclusions are as follows:

(1)

Policy tools are grouped into three categories aligned with MLP. Six thematic instruments were identified, corresponding to technology application (niche level), management norms (regime level), and industrial ecology (landscape level). This classification provides a practical link between policy analysis and socio-technical transition theory.

(2)

There is no single policy tool to promote high-level or low-level digital development. The necessity analysis indicates that, regardless of whether any individual tool exists, the results cannot be reliably determined.

(3)

High-level development follows three equifinal pathways. The “technology-talent driven”, “foundation-governance linkage”, and “system-ecology leading” each achieve high digital development through different combinations of cross-level policy interventions. This demonstrates that multiple causal routes exist, and success depends on contextual alignment rather than a single optimal model. Low-level development results from configurational misfit. The “factor-deficient” and “system-mismatched” explain low-level performance. These configurations reaffirm the logic of the high-level pathways, showing that the same tools can fail when used in isolation or in mismatched combinations.

(4)

Effective policy mixes require coordination across different levels. In all successful setups, digital technology uses consistently depend on supporting conditions at the regime or landscape levels. This highlights that policy design should go beyond isolated actions towards integrated strategies that align technological, institutional, and ecological aspects.

(5)

Theoretical logic extends beyond the empirical context. The empirical evidence is derived from China’s provincial policies; thus, although the specific configurations mirror local institutional conditions, the core mechanisms (equifinality, cross-level coordination, and configurational misfit) are conceptually applicable to other contexts where governments aim to coordinate industry-wide digital transformation.