Policy Significance Statement

This study highlights the growing trend of algorithmic regulation within the public sector, with China’s social credit system (SCS) serving as a primary example of this development. By conducting a comparative analysis of SCS scoring mechanisms across various Chinese cities, we shed light on the pivotal role that cities play in both implementing and being subject to algorithmic evaluations. Our analysis examines the extent of algorithmic interventionism within the SCS framework, offering insights into its operational dynamics and broader implications for governance. In addition, the study addresses ethical challenges, particularly concerning transparency and fairness, contributing to the ongoing discourse on algorithmic governance. These findings are crucial for informing future policy decisions regarding the design and implementation of algorithmic systems, ensuring they are both effective and ethically sound in navigating complex governance issues.

2. Algorithmic regulation and the SCS

Algorithmic regulation is widely implemented in the business sector in the form of rating or feedback control mechanisms (Dwoskin, Reference Dwoskin2018). Drawing on the experiences of big techs like Uber, algorithmic regulation is proposed as a new form of governance to address policy challenges through innovative problem solving rather than relying on traditional top-down law enforcement approaches (O’Reilly, Reference O’Reilly, Goldstein and Dyson2013; Eyert et al., Reference Eyert, Irgmaier and Ulbricht2022). In the realm of public governance, the concepts of algorithmic regulation (Yeung, Reference Yeung2018) and algorithmic governance (König, Reference König2020; Gritsenko and Wood, Reference Gritsenko and Wood2022) have emerged as novel modes of social coordination and control (Ulbricht and Yeung, Reference Ulbricht and Yeung2022). These terms refer to the management of societal behavior through automated decision-making systems, frequently utilizing scores or rankings as incentives on a population-wide scale (Cristianini and Scantamburlo, Reference Cristianini and Scantamburlo2020). These scores/rankings are computed on the basis of people’s (financial and nonfinancial) behavioral data and are even considered indicative of people’s behavioral traits.

Algorithmic regulation can be understood from two perspectives. From the regulatory perspective, it establishes a control loop in which individuals’ behaviors are directly associated with rewards or punishments and thus adapt their behavior to maximize their utility (O’Reilly, Reference O’Reilly, Goldstein and Dyson2013); from the technical perspective, it emphasizes the utilization of computational algorithms to bolster and streamline the control loop, enabling the monitoring and modification of behavior (Hildebrandt, Reference Hildebrandt2018; Cristianini and Scantamburlo, Reference Cristianini and Scantamburlo2020). According to Yeung (Reference Yeung2018, p. 505), “(a)lgorithmic decision-making refers to the use of algorithmically generated knowledge systems to execute or inform decisions, which can vary widely in simplicity and sophistication.” Our analysis also showed that the sophistication levels of SCSs vary across cities. Some cities claim to employ highly complex algorithmic models that offer real-time assessments, while others use a simpler calculation model based on point additions and deductions (see Section 6). Moreover, the structure of the SCS serves as a cornerstone for credit-rating-based supervision of market entities, which is under development at the moment. This credit-rating-based supervision represents a significant development in algorithmic regulation within the public sector.

As a prominent example within the global trend of algorithmic regulation, the SCS is part of the global shift of “the focus of public law from constitution and rule of law to analytics and algorithm” (Backer, Reference Backer2018, p.130). A key issue in this transition is the evolving degree of intervention that algorithmic regulation introduces (Zhang, Reference Zhang2020a; Zuo, Reference Zuo2020). Dai’s comprehensive study in 2018 concluded that the SCS represents a predominantly state-centered interventionist model for governing societies (Dai, Reference Dai2018). Another important theoretical contribution is from Síthigh and Siems, who devised a framework comprising eight factors to evaluate the extent of state intervention across various rating systems, particularly those bolstered by digital technologies, such as the SCS (see Table 1) (Síthigh and Siems, Reference Síthigh and Siems2019). In their framework, drafter refers to the entity—whether a private entity or state authority—that initiates and drafts the system. The factor user assesses whether participation in the system is mandatory for individual users. Aim inquires whether the scheme has a single, specific objective or encompasses a broader set of objectives across various functions or contexts. Scoring examines whether the system utilizes multiple scores or a single score. Application investigates whether the system’s results are indicative or definitive. Algorithm pertains to the transparency of the system. Enforcement questions who bears the responsibility for enforcing the system. Accountability evaluates the extent to which a separate oversight body monitors the design and operation of the system (Síthigh and Siems, Reference Síthigh and Siems2019). This framework provides a foundation for better understanding the nature of the SCS within the context of regulatory studies and provides a practical tool for conducting a detailed analysis of various SCS implementations at the city level that we draw on in our research.

Table 1.

Degree of interventionism in rating systems

Note: This table is developed on the basis of the work of Síthigh and Siems (Reference Síthigh and Siems2019).

The deployment of algorithms in the public sector potentially increases the efficiency and efficacy of public services, such as selecting the optimal policy alternatives and directing interventions toward specific profiles with precision (OECD, 2015). In contrast, such precise interventions also trigger significant concerns, not only regarding their capabilities, processes, and practices (Veale and Brass, Reference Veale, Brass, Yeung and Lodge2019), but also ethical considerations over privacy, self-datafication, transparency, fairness, and issues regarding legitimacy and surveillance (Chen et al., Reference Chen, Engelmann and Grossklags2023b; Veale and Brass, Reference Veale, Brass, Yeung and Lodge2019; Gerhard and Hepp, Reference Gerhard and Hepp2018; Dencik et al., Reference Dencik, Hintz, Redden and Warne2018; Chen and Cheung, Reference Chen and Cheung2017). Existing literature consistently calls for transparency to uncover arbitrary assessments, inaccurate characterizations, and biases within the scoring systems (Zarsky, Reference Zarsky2016; Citron and Pasquale, Reference Citron and Pasquale2014). There are also social initiatives, such as OpenSchufa, demanding more transparency in the credit scoring systems. However, the ideal of transparency remains under debate. Its various limitations are discussed, for example, in the context of privileging seeing over understanding, creating false boundaries, and causing professional boundary work (Ananny and Crawford, Reference Ananny and Crawford2018). Fairness garners significant attention when algorithms are employed to facilitate decision-making in high-stakes contexts such as hiring (Köchling and Wehner, Reference Köchling and Wehner2020). This emphasis on fairness becomes even more critical when algorithms are utilized for automated decision-making in social governance. Likewise, the pursuit of fairness should always be assessed together with other factors such as prediction accuracy (Heidari et al., Reference Heidari, Ferrari, Gummadi and Krause2018) and group discrimination (Citron and Pasquale, Reference Citron and Pasquale2014).

Ethical inquiries and critical examination extend to the SCS as well. The existing literature concludes that the system lacks algorithmic transparency (Kobie, Reference Kobie2019; Síthigh and Siems, Reference Síthigh and Siems2019), offers limited transparency from the perspective of foreign companies (Chen et al., Reference Chen, Bogner, Becheva and Grossklags2023a), and showcases asymmetries in transparency between SCS blacklists and redlists (Engelmann et al., Reference Engelmann, Chen, Fischer, Kao and Grossklags2019). In these studies, the SCS was typically regarded as a single, overarching system, while its fragmentation and complexity were addressed to a lesser extent (except Liu, Reference Liu2019; Tsai et al., Reference Tsai, Wang and Lin2021). This oversight has left a substantial research gap, particularly regarding transparency variations across different SCS implementations at the city level, further compounded by their diverse structures. In contrast to the extensive focus on transparency, investigations into fairness concerns arising from the SCS remain relatively sparse. This research aims to address these gaps by examining the current state of algorithmic regulation development at the city level in China, and by analyzing the transparency of the computational algorithms employed in the SCSs, as well as their societal fairness implications.

4. The SCS implementation in Chinese cities: an overview

The implementation of the SCS at the city level is driven by the model city program. Selected cities, which are called SCS construction model cities, conduct trials of SCS construction and implementation to accumulate experiences, which are later copied by or promoted to other parts of China (National Development and Reform Commission and People’s Bank of China, 2017), manifesting the policy experimentation feature in Chinese public policy (Heffer and Schubert, Reference Heffer and Schubert2023). Four groups of model cities were selected between 2017 and 2023, resulting in a total of 130 model cities and districts in China (see Credit Zhejiang, https://www.cnfin.com/xy-lb/detail/20230830/3923862_1.html). These cities or districts range from the sub-provincial level (e.g., Hangzhou, Shanghai-Pudong, and Tianjin-Binhai) to the county level (e.g., Yiwu, Jinhua, and Rongcheng). They are distributed unevenly across different provinces. For instance, in Zhejiang province, all 11 prefecture-level cities are designated as SCS model cities, whereas in Qinghai, there is no SCS model city.

From the policy perspective, SCS-related regulations and methods are issued at both the national and local levels. At the highest level, strengthening the SCS is highlighted as a critical part of a unified market that should adhere to high standards in the 14th Five-Year Plan (2021–2025) for National Economic and Social Development (National Development and Reform Commission, 2021). The State Council has continuously issued national policies about the SCS construction (e.g., State Council, 2014; State Council, 2019; State Council, 2022). Recently, the 2024–2025 Action Plan for the Establishment of the Social Credit System was issued to guide the development of the system in the next step. These policies serve more as principles for the construction of the SCS. Guided by these policies, Chinese cities and provinces have developed corresponding work plans, regulations, and measures for local SCS implementation (e.g., The People’s Government of Suzhou, 2022; Shanghai People’s Congress Standing Committee, 2021).

Among the various local SCS policy documents, local social credit regulations hold particular significance, serving as a cornerstone for the implementation of the system in cities. It is noteworthy that documents of Social Credit Regulation from various cities and provinces consistently provide similar definitions for “social credit” and “social credit information” (e.g., Shanghai People’s Congress Standing Committee, 2021; Tianjin People’s Congress Standing Committee, 2021; Guangdong People’s Congress Standing Committee, 2021; Hangzhou People’s Congress Standing Committee, 2021). It is surprising that these fundamental concepts, which are essential for comprehending the SCS, lack explicit elaboration in the various national SCS policies. This further underscores the importance and necessity of examining the SCS at the city level. “Social credit” refers to “the state in which natural persons, legal persons, and unincorporated organizations with full capacity for civil conduct abide by legal obligations or perform agreed obligations in social and economic activities.” “Social credit information” refers to objective data and materials that can be used to identify, analyze, and assess the compliance and performance status of information subjects, including public credit information and market credit information. Public credit information refers to the social credit information generated or obtained by state organs and organizations authorized by laws and regulations to manage public affairs in the process of providing services in accordance with the law. Market credit information refers to the social credit information generated or obtained by market credit service agencies, credit service industry organizations, other enterprises, public institutions, and social organizations in their production, operation, and social service activities. Thus, while “social credit” emphasizes on the legal and contractual obligations, “social credit information” is collected from almost all perspectives of life.

Cities also exhibit variety in their specific SCS implementation methods. This diversity is evident in, among others, how they administer credit-related rewards and punishments, which are often customized based on the social and economic environment of each city. For example, Hangzhou’s rewarding mechanism is characterized by granting priority in policy support and incubation measures to enterprises engaged in innovation and startup activities (Hangzhou People’s Congress Standing Committee, 2021). Shenzhen and Nanjing exhibit a relatively lenient approach to implementing punishments. This leniency includes exceptions to the application of penalties in cases where credit subjects promptly rectify minor instances of dishonest behavior without causing harm, or when it is their first-time offense and the consequences are minimal (Shenzhen People’s Congress Standing Committee, 2022; Nanjing People’s Congress Standing Committee, 2020).

At present, all cities in China, irrespective of their designation as model cities, should have integrated the SCS into their governance frameworks. The COVID-19 pandemic played a role in accelerating the adoption and expansion of social credit systems across Chinese cities (Knight and Creemers, Reference Knight and Creemers2021; Engelmann et al., Reference Engelmann, Chen, Dang and Grossklags2021). Many cities have tailored credit scoring systems with a distinct focus on evaluating both legal entities and individuals. The subsequent sections delve into the intricate workings of enterprise credit rating systems and personal credit scoring systems using the framework from Síthigh and Siems (Reference Síthigh and Siems2019), offering insights into the implementation of algorithmic regulation across cities. In addition, these sections scrutinize the level of transparency embedded within these systems, aiming to uncover the extent to which these systems reveal their operational mechanisms and criteria.

5. Enterprise credit scoring systems across cities

At the national level, the Chinese enterprise credit rating and scoring systems typically follow a sector-oriented approach. Various government departments have developed distinct methods for rating enterprise credit across sectors such as food and medicine, real estate, intellectual property, taxation, human resources, and environmental protection (e.g., National Development and Reform Commission, 2016; State Taxation Administration, 2014). In the implementation of sector-based scoring systems, enterprises undergo scoring and rating processes consistently across cities within the same sector. Data from various governmental sources are gathered and presented on the National Enterprise Credit Information Publicity System (see https://www.gsxt.gov.cn/index) that is designed as a key national network for market supervision, as well as other platforms such as the national SCS platform—Credit China—at different administrative levels and the departments’ websites. Thus, they function as comprehensive repositories of credit-related information.

At the city level, enterprise credit scoring systems are initiated and implemented by local governments, each with its own unique design and approach. All locally registered enterprises are subject to these local systems. Therefore, the degree of interventionism is high from the perspectives of both the drafter and the user. Figure 1 offers a visual representation depicting the algorithmic regulatory framework designed for public credit scores for enterprises at the local level (this flowchart has been derived through a detailed policy analysis. It serves the purpose of providing a strategic tool for conceptual understanding). Government departments collect and process data regarding locally registered enterprises, and generate scores or ratings within their respective departments. They also place enterprises on SCS blacklists or redlists if they perform any defined “bad” or “good” behaviors, respectively. Evaluation results from this step are directly related to punishments and rewards under the SCS framework. In the next step, a public credit score is automatically calculated. Cities have developed various scoring models, which will be analyzed in detail in the next paragraphs. The enterprise credit score is deemed to reflect an enterprise’s overall creditworthiness/trustworthiness and is then linked to supervision at different levels, establishing an advanced approach for governance in the public sector—the credit-rating-based supervision system. In this system, higher ratings for enterprises often facilitate expedited administrative procedures, streamlined access to government procurement opportunities, and reduced regulatory scrutiny. Conversely, lower ratings may result in restrictions on government procurement, reduced access to financial support, and a heightened level of regulatory inspections (Changde Development and Reform Commission, 2022b; Yongzhou Municipal Government Office, 2020).

Figure 1.

The algorithmic regulatory framework for enterprises’ public credit scores.

In general, the enterprise credit system aims to achieve several socio-economic goals across multiple levels simultaneously. At the enterprise level, its objective is to reinforce or alter enterprise compliance and prosocial behavior; at the city level, it aims to promote an overall environment of creditworthiness; and at the regulatory level, the system contributes to a new governance approach based on credit ratings. Thus, the interventionism degree is regarded as a medium from the perspective of aim for the enterprise credit scoring systems at the city level.

Guided by SCS policies at the national level, cities have developed various scoring and rating methods for locally registered enterprises. To better illustrate this situation, we present the enterprise credit scoring structures from one province and three cities as an example in Supplementary Table A1. These cases are selected for the following reasons. Zhejiang took the lead by publishing provincial-level scores in 2019 (Zhejiang Development and Reform Commission, 2019), while Guangzhou represents a recent case (Guangzhou Development and Reform Commission, 2023). Yongzhou and Changde provide good examples for a comparison between cities in the same province (Hunan). We analyze the enterprise credit rating systems primarily from four perspectives: score scopes, rating settings, indicators and corresponding weights, and transparency. Supplementary Table A1 presents an overview of the comparative results.

First, in each enterprise credit scoring system, only one credit score is generated, reflecting a high degree of interventionism along the factor scoring. However, the scoring ranges are configured differently across various systems, typically spanning from 100 to 1000. Also, some cities set the minimum score above 0, such as Guangzhou.

Second, these scores are transferred to credit ratings that vary in granularity and criteria. Some cities, such as Yongzhou, offer four ratings, while others, such as Guangzhou, provide a more detailed scale with nine ratings. While Zhejiang and Changde share the same score scope of 1000 and both employ 5 ratings, they differ in their rating criteria. In Zhejiang, an enterprise must attain a score of 850 or higher to achieve an “Excellent” rating, whereas in Yongzhou, a score of 800 suffices for the same rating. Therefore, enterprises’ scores or ratings are not fully comparable between different regions. Both enterprise credit scores and ratings serve as conclusive assessments of an enterprise’s creditworthiness, informing decisions or judgments by governments and/or third parties. However, these scores and ratings undergo periodic revisions based on updated data. In this case, the interventionism degree is regarded as a medium for the application factor. While the scores and ratings can serve market purposes, such as facilitating business transactions, the local governments bear the responsibility for ensuring enterprise compliance with the system’s rules and regulations. They take decisive action against any violations or noncompliance with the system’s requirements. Take Zhejiang province, for example. Administrative enforcement and regulatory departments have utilized the enterprise credit risk model to conduct 220,787 inspections. They further categorize enterprises into three risk levels: highrisk (186,000 enterprises), medium-risk (238,000 enterprises), and low-risk (2,230,000 enterprises) (see Credit Zhejiang at https://credit.zj.gov.cn/art/2022/2/14/art_1229636059_1828.html). For this reason, the interventionism degree is high from the enforcement perspective.

Third, all five cities/provinces select a five-dimension model for the enterprises’ public credit score calculation, although the content of these dimensions varies slightly (see Supplementary Table A1). Zhejiang and Yongzhou share the same dimensions. “Compliance” and “social responsibility” are included in all scoring models but are defined in slightly different ways across cities. The “compliance” indicator generally evaluates the performance regarding administrative management and judicial records. In Changde (Changde Development and Reform Commission, 2022b), “compliance” also sheds light on records of the SCS redlists, which, however, is included in the “social responsibility” dimension in Zhejiang (Zhejiang Development and Reform Commission, 2020) and Yongzhou (Yongzhou Municipal Government Office, 2020). Regarding “social responsibility,” Zhejiang focuses more on charity and honor records (Zhejiang Development and Reform Commission, 2020); Guangzhou stresses the payment of social security and tax (Guangzhou Development and Reform Commission, 2023); and Changde and Yongzhou cover both of these perspectives (Changde Development and Reform Commission, 2022b; Yongzhou Municipal Government Office, 2020). Other dimensions cover basic information, finance and taxation, honors and awards/qualifications, supervision risk, governance, and conducting business with integrity. These five dimensions are referred to as the first-level indicators, which are further broken down into second- and third-level subsets, providing detailed interpretations and revealing differences between indicators. Cities have adopted diverse methodologies to assign weights to indicators, including big data technology, the analytic hierarchy process, and the Delphi method (Guangzhou Development and Reform Commission, 2023). Thus, the same indicator may receive different weights across cities. For instance, the indicator of “social responsibility” plays a more important role in the public credit score calculation in Zhejiang (185 out of 1000) than in Changde (120 out of 1000). Despite differences in content, “compliance” always receives the largest weight (over 40%), showcasing its importance for public credit scores.

Enterprises’ public credit scores are then translated into ratings, which, in turn, dictate the level of supervision they undergo. While the method for translating scores into ratings is explicitly outlined, the specifics of the supervision measures are not always disclosed to the public. This lack of transparency may pose challenges for enterprises, for instance, in Zhejiang and Guangzhou, to fully comprehend the supervision measures imposed on them. In contrast, the other two cities, Changde and Yongzhou, provided detailed specifications. In Changde, A-level enterprises will be prioritized, under equal conditions, in fiscal funding support and commendation rewards selection, while E-level enterprises are restricted from applying for projects backed by government financial funds and from participating in public resource trading activities (Changde Development and Reform Commission, 2022b). Our case study suggests that the enterprise public credit system in Guangzhou exhibits the lowest level of transparency among the four sample cases. Determining an overall transparency level for the local enterprise public credit scoring systems is challenging due to the considerable variation between cities. In this case, we labeled the interventionism degree as low/medium in the perspective of the algorithm.

Finally, at the city level, there is typically no distinct oversight body solely dedicated to monitoring both the design and operation of the system. Instead, responsibility falls to the local branch of the NDRC, which leads, coordinates, and monitors the development of the SCS, indicating that it is not functioning as a separate oversight body. Meanwhile, the system incorporates a mechanism for “objection applications,” allowing companies to address inaccuracies or omissions in their data. Consequently, we consider the interventionism degree for accountability as the medium.

In parallel with the enterprise public credit scoring systems, there are also credit scoring systems for individuals at the local level. In the next section, we investigate these personal credit scoring systems across Chinese cities.

6. Personal credit scoring systems across cities

The implementation of the SCS in China has accelerated the development of personal credit scoring systems across cities (Li and Kostka, Reference Li and Kostka2022). Aligned with the national Guidelines on the Construction of a Personal Credit System (State Council, 2016), cities have introduced unique personal credit scoring systems targeting individuals above 18 years old, who live and work in a given city. Thus, the interventionism degree is categorized as high for both drafter and user factors. These systems extend beyond traditional financial credit assessments, encompassing a wide spectrum of social, economic, and moral aspects, purportedly seeking to create a comprehensive portrayal of the individuals’ creditworthiness to facilitate economic and social activities and, thus, fostering a more trustworthy and reliable society. For this reason, it signifies a moderate level of interventionism from the perspective of the aim.

We have identified 56 such systems across China (see Supplementary Table A4), encompassing 55 systems from cities at various administrative levels, such as provincial (e.g., Tianjin), sub-provincial (e.g., Jinan and Xiamen), prefecture (e.g., Wenzhou and Wuxi), and county (e.g., Xintai and Yiwu) levels, along with one system from Hainan province. These systems were launched between 2017 and 2022. As shown in Figure 2, they are particularly prevalent in the eastern part of China. In particular, Zhejiang and Shandong provinces have the most (12 and 10, respectively) personal credit scoring systems, followed by Jiangsu (7) and Henan (5) provinces. In contrast, cities in the western areas usually do not (yet) have a quantified scheme. This uneven geographic distribution broadly aligns with the varying economic development levels across different regions of China, resulting in a higher concentration of SCS personal credit scoring systems in the more economically developed areas.

Figure 2.

Chinese cities with personal credit scoring systems. (The red spots on the map indicate the presence of personal credit scoring systems in cities, while the shades of green represent the number of such systems within specific provinces: the darker the shade, the higher the number of systems in that province. Hainan province is assigned a different shade of green, as there is one single personal credit scoring system at the provincial level for all cities. Red spots with blue outlines refer to cities that are analyzed in detail. Additionally, Zhejiang and Hainan provinces are also analyzed in detail in this paper. This map is developed based on data that were collected between February and May 2024.)

Cities that have implemented personal scoring systems exhibit varying degrees of development. Advanced systems boast well-defined scoring and rating scales, diverse application scenarios for rewards, and a significant user base, as exemplified by Hangzhou’s Qianjiang Score, which boasts over five million registered users. Conversely, some systems are in their initial stages, lacking a clear structure and scoring method, as seen in Hohhot, Anqing, and Xianning. In Supplementary Table A2, we selected four examples to present and compare the current status of personal credit scoring systems across Chinese cities. These cities and provinces span different administrative levels and initiated their personal credit scoring systems in various years. Hainan presents a unique case by constructing a single credit scoring system for all cities in the province. In addition, Hangzhou and Yiwu, both hailing from Zhejiang province, are notable as model cities for the implementation of the SCS. Similar to the analysis of enterprise credit scoring systems, we compared personal credit scoring systems from four perspectives.

First, cities have adopted very different score scopes. The personal credit scoring systems usually establish a default score as the initial baseline (see Supplementary Table A2), adjusting it based on individual behavior. However, for certain systems, the score scope is not clearly disclosed, as in the case of Hainan province.

Second, rating scales vary from city to city, commonly using five or six ratings. Even among the systems adopting the same score scope, credit ratings might be defined differently, as seen in the contrast between Hangzhou and Changde. Further, similar to enterprise credit scoring systems, individual credit scoring systems in various cities produce a single score as a definitive measure of an individual’s trustworthiness. This score is updated continuously and directly utilized in decision-making processes by both government authorities and market participants. For some cities, such as Xiamen, even real-time measurements to offer live and dynamic scores are conducted (see https://port.xm.gov.cn/zmhd/cjwt/202301/t20230113_2714953.html). Local governments are tasked with the enforcement of these scoring systems. Therefore, the interventionism degrees for the factors of scoring, application, and enforcement are high, medium, and high, respectively.

Third, individual credit scores are primarily calculated based on five core dimensions (see Supplementary Table A2). Identical to indicators for enterprise public credit score, “compliance” and “prosocial behavior” are integral components present in all personal credit scoring systems across various cities, albeit the latter might be labeled differently (e.g., public morality, social welfare, and social contribution). Basic and identity information encompasses an individual’s educational background, marital status, and profession, among other details. Business credit information includes financial credit records and the fulfillment of economic contracts, while social governance/credit entails administrative records and social welfare information. In this way, the systems endeavor to provide a comprehensive portrayal of people’s trustworthiness by incorporating various dimensions beyond financial considerations.

Fourth, the analysis of transparency in personal credit scoring systems reveals a similar structure as that used for enterprises. Among the four sample cities in Supplementary Table A2, they all offered at least some explanation for the first-level indicators, but only half of them disclosed fine-grained indicators as well as corresponding responsible departments. Weights assigned to the indicators remain undisclosed across all sampled cities, rendering it challenging for residents to assess the significance of these indicators. The algorithms used in personal credit systems vary significantly from city to city. Our understanding of the algorithm in Hainan is limited due to scant information disclosure. Among the other three sample cities, two distinct algorithmic models have been identified. Yiwu and Changde employ a straightforward calculation model, providing detailed scoring criteria that specify how points are added or deducted for various actions. For instance, in Changde, donating to charitable causes earns 20 points, while violating public smoking bans results in a 30-point deduction (Changde Development and Reform Commission, 2022a). In contrast, Hangzhou is presumed to use a more complex algorithmic model for its personal credit score. To support this, the city has established a big data laboratory to continuously test and refine the model. However, the specifics of Hangzhou’s algorithm remain undisclosed.

In contrast to enterprise public credit scoring systems, which emphasize both punishments and rewards, personal credit scoring systems across cities predominantly focus on rewards. Penalty measures associated with low scores for individuals are often not mentioned. The rewarding measures for personal credit scores are disclosed but in a rather general manner. Typically, they are illustrated by providing a few example scenarios that encompass both public and commercial services. Public services usually cover health care (treatment first and payment afterward), parking (discounts), and library usage (deposit-free), while commercial services may include travel (discounts for admission tickets to scenic spots) and job hunting (system signals verified education and profession background) (see, for example, Qianjiang Score http://qianjiangfen.cn/index.html). Currently, across all cities, efforts are being made to introduce additional score application scenarios to enhance the value of the scoring systems (Liang and Chen, Reference Liang and Chen2022). As shown in Supplementary Table A2, personal credit scoring systems in some cities (e.g., Yiwu and Changde) present a much higher transparency level than those from other cities (e.g., Hangzhou and Hainan). Given the variations in algorithmic transparency between cities, we consider the interventionism degree for the algorithm as low/medium and for the factor accountability as medium.

With the growing prevalence of personal credit scoring systems, local governments are actively exploring the feasibility of establishing a mutual recognition mechanism through information sharing. Under this mechanism, personal credit scores from different cities would be reciprocally acknowledged, allowing residents with high scores to access benefits and privileges across various cities. In 2022, 10 cities signed a strategic cooperation agreement to promote the establishment of such a mechanism (see https://credit.fzgg.tj.gov.cn/detail.do?contentId=70da9420bdc04f45b13184d873e5db80). This trend implies that the future of personal credit scoring systems may continue to evolve in a fragmented manner, albeit with the introduction of common or different mutual recognition mechanisms. The development of a common mechanism would suggest that the objective of implementing algorithmic regulation for social control, particularly focusing on individuals at the national level, could be attained without the necessity of a single unified SCS. While the implementation of such a mechanism aims to enhance the overall efficacy of the SCS, it also faces substantial challenges, particularly in the areas of regulatory frameworks and technical aspects within the realm of algorithmic regulation.

Sections 5 and 6 delved into the establishment and execution of two pivotal components of the SCS across cities. At this stage, there is no clear, unified trend that all cities are following in constructing their SCSs. Instead, cities have developed distinct regulations and established their own unique models. As cities assess the trustworthiness and creditworthiness of their residents and businesses, they similarly undergo scoring and ranking within the framework based on the progress of their SCS development. In the subsequent section, our focus shifts to the city credit scoring system.

7. City credit scoring system

City rankings have evolved into a tool for policymakers to assess urban performance and formulate strategic policies for future development. The Chinese city credit scoring system was introduced under the SCS framework, aligning with the broader pattern observed in numerous national and international city rankings that assess cities based on criteria, such as creativity, innovation, sustainability, and smart city initiatives (Dobbs et al., Reference Dobbs, Smit, Remes, Manyika, Roxburgh and Restrepo2011; Kontokosta and Malik, Reference Kontokosta and Malik2018; Cities of Opportunity, 2016). The city credit scoring system provides a condensed overview of the overall progress in the construction and implementation of the SCS in each city, offering a comparative perspective. This is reflected in the Composite Index of City Credit (CICC), which is issued by the State Information Center and serves as the city’s credit score. Despite its significance, the city credit scoring system is largely overlooked in the existing literature on the SCS.

From an implementation perspective, the NCPCI, commissioned by the NDRC, is responsible for evaluating and assessing the performance of Chinese cities in constructing their SCSs. Thus, the state’s role as the drafter of this system results in a high level of interventionism in the drafting process. Unlike the SCS model cities construction project, which is based on voluntary participation, the city credit scoring system is mandatory for all cities, indicating a high level of interventionism from the perspective of the user.

The CICC has a score scope between 0 and 100. Unlike personal and enterprise credit scoring systems discussed in the previous sections, the city credit scores are not converted into credit ratings (e.g., A, B, C, D, or excellent, good, average, and fair) as is commonly done in enterprise and personal credit scoring systems. Instead, cities are ranked monthly based on their numeric scores. Therefore, we consider the degree of interventionism for scoring to be high and for application to be medium.

The CICC adopts a three-level indicator system. At the first level, there are 8 indicators (see Supplementary Table A3), which are further broken down into 20 second-level and 35 third-level indicators. Based on explanations provided by the City Credit Monitoring and Early Warning Indicators (hereafter City Credit Indicators for short) (National Center for Public Credit Information, 2022), the eight first-level indicators can be categorized into three groups, focusing on the SCS construction progress, the implementation and application of the SCS results, and creditworthiness/trustworthiness propaganda and public opinion. These indicators denote not only the components of city credit but also what the Chinese government cares about and expects regarding SCS implementation.

Indicators 1, 2, and 8 measure the progress of the SCS construction (see Supplementary Table A3). They examine cities’ SCS policy implementation, SCS information infrastructure, and the innovation of SCS construction. In particular, Indicator 2, which is about the collection of different types of credit-related data, the accuracy of the data, and data security, receives the highest number of maximum points as an individual indicator, amounting to 28 out of 100. This highlights the substantial reliance of the system on big data. Indicator 8 pertains to the promotion of typical (innovative) experiences at the national level, reflecting a programmatic attitude of exploration in the face of new developments in China’s reform progress. Notably, the construction of enterprise and personal credit scoring systems, as discussed in Sections 5 and 6, is not distinctly emphasized within the framework of these indicators that assess the progress of SCS construction across cities.

Indicators 3–5 and 7 evaluate the implementation of the SCS and the application of the SCS results. Together, they constitute the largest share of the CICC, with a maximum number of points of 53 (out of 100). All these four indicators focus on the enterprise SCS. Indicators 3–5 address the utilization of credit reports in administration and public services, public credit-based financing (especially for micro, small, and medium enterprises), and the ratios of trustworthy and untrustworthy enterprises. These three indicators are also reflected in the enterprise public credit scoring system (see Section 5). Indicator 7 evaluates the creditworthiness of local government departments and state-owned enterprises (SOEs), as well as the local government’s adherence to implementing the SCS blacklists for market access. Government trustworthiness and creditworthiness are emphasized as one of the four pillars established in the construction of the SCS, as outlined in the Planning Outline (State Council, 2014). According to the City Credit Indicators, this is assessed through metrics such as the frequency of listings on the SCS blacklist and the occurrence of overdue debts among SOEs.

Finally, Indicator 6 delves into the broader environment of the SCS to assess factors that are more difficult to quantify. It includes factors related to propaganda efforts for creditworthiness, including trustworthiness culture promotion (e.g., see Chen et al., Reference Chen, Engelmann and Grossklags2022), public opinion dynamics, and major incidents of untrustworthiness in the four pillars of SCS development defined in the Planning Outline: government credit, commercial credit, social credit, and judicial credit (State Council, 2014), such as issues related to food security and illegal fundraising. Specifically, big data technology is employed to monitor the specific sub-indicators. In general, these indicators reflect the process of constructing and implementing SCS rather than serving as benchmarks for a superior model.

In addition, eight scenarios will result in severe score deductions (National Center for Public Credit Information, 2022), including falsification of materials, inaccessibility or political compliance issues on the local SCS websites, data security issues, misuse or generalization of credit applications, noncompliance to medium- and long-term coal contracts, unregulated fee collection by industry associations, and bond default by local governments or local SOEs.

Concerning the transparency of the city credit scoring system, detailed explanations, fine-grained indicators, as well as information regarding weights, data source, and scoring details are all provided. Notably, data are sourced not only from government departments, but also from the mainstream digital media platforms, such as official government websites, Tencent News, sina.com, and 163.com. Cities at different administrative levels are scored based on the same criteria but are ranked in two separate groups. One group comprises 36 provincial capitals and cities at the provincial level, while the other includes 261 prefecture-level cities. The scoring and ranking results are presented in the section of “City Credit (城市信用)” on the website of Credit China (www.creditchina.gov.cn) and are updated on a monthly basis, providing free public access. However, only the Top 20 and Top 50 cities from each group are featured, denoting a public praising mechanism. In contrast, cities with low scores and poor rankings are conspicuously absent. Based on the data that have been made available, it appears that scores exceeding 90 are seldom achieved, and scores over 80 indicate a very good performance. Full datasets, encompassing scores for all cities, are currently unavailable, and the statistical distribution of the scores remains undisclosed, making it difficult to interpret the scores. Accordingly, for the city credit scoring system, while the scoring method is highly transparent, the interpretation of the scores lacks transparency. In addition, there are few clues about whether and how these scores are connected to specific rewards or punishments. Therefore, the degree of interventionism is regarded as a medium from the perspective of the algorithm.

The city credit scoring system establishes multilayered, direct, and indirect goals. The above analyzed indicators assess not only the progress and execution of SCS initiatives, but also the effectiveness of cities as centers for economic development, underscoring the critical role of fostering economic growth. This aligns with other city rankings that mostly address one common aspect: attracting wealth generation (Saez et al., Reference Saez, Heras-Saizarbitoria and Rodriguez-Nunez2020). Thus, the city credit scoring system steers Chinese cities into a competition that includes not only SCS construction and implementation, but also attracting wealth generation. In addition, on the “City Credit” website, just below the section for the city credit rankings, there is also a section called “Business Environment (营商环境),” advertising the cities’ efforts in promoting a favorable business environment. We assign a medium interventionism label to the factor aim for the city credit scoring system.

The degree of interventionism for the city credit scoring system is classified as a medium from the perspective of enforcement given the fact that there is not a clear rewarding and sanctioning mechanism associated with city scores. However, there may be a possibility that career prospects of administrators and budgeting are implicitly linked to the scores. In addition, the system is deemed to have a high level of interventionism from the perspective of accountability, due to the absence of a supervisory body responsible for overseeing its operations and the lack of a mechanism for handling “objection applications,” unlike the other two systems.

8. Discussion and concluding remarks

The act of measuring the behavior of various societal participants (i.e., companies, individuals, and cities) through digital technologies exemplifies algorithmic regulation, as implemented by the SCSs, by transforming abstract values such as trustworthiness or compliance into quantifiable metrics. These metrics enable authorities to monitor, evaluate, and influence behaviors systematically, embedding algorithmic logic into governance processes (Elgert, Reference Elgert2018). However, this process also abstracts complex social realities into simplified indicators, which may fail to fully capture the nuanced and context-specific dimensions of human behavior. As such, while effective for achieving measurable outcomes like economic efficiency, it risks marginalizing broader considerations such as social equity. This section concludes by examining the interventionist nature of the SCSs, addressing the transparency dilemma they present, and delving into the fairness paradox inherent in the system.

8.1. Algorithm-enhanced interventionism

Building upon the framework from Síthigh and Siems (see Section 2) (Síthigh and Siems, Reference Síthigh and Siems2019), our analysis unveils a heightened level of interventionism across the enterprise, personal, and city credit scoring systems (see Table 2), which partially stands in contrast to the observations made in previous work (Síthigh and Siems, Reference Síthigh and Siems2019). Our findings align with those presented in Síthigh and Siems (Reference Síthigh and Siems2019) for four of the eight factors—drafter, user, scoring, and application. Specifically, the factors drafter, user, and scoring exhibit pronounced levels of interventionism. This characterization stems from the active participation of central or local governments in drafting policies, the compulsory involvement of businesses and citizens, and the adoption of a singular scoring metric at the respective city level. Most individuals and enterprises will have one score in the city where they are based. However, for those who move around or act in different cities, multiple scores by different cities will be available. From the perspective of cities, each one issues only one score for a given individual or enterprise. Therefore, we refer to this as a singular scoring metric. In addition, interventionism is labeled as a medium from the perspective of application for all three studied systems. This is due to their provision of definitive results in the form of a concrete number and/or rating, coupled with periodic updates.

Table 2.

Degrees of interventionism in the SCS

Conversely, our study reveals disparities across the remaining four factors compared to the findings presented in Síthigh and Siems (Reference Síthigh and Siems2019). First, we contend that beyond credit-related objectives, the SCS is designed and executed to fulfill a broader socioeconomic goal of impacting economic and social order (State Council, 2014). Our findings about indicators of different systems show that they cover a wide range of aspects of social and economic behavior, thereby providing empirical evidence supporting the systems’ pursuit of socioeconomic objectives. Different socioeconomic goals are also highlighted across SCS-related policies. Second, the degree of interventionism regarding enforcement for both enterprise and personal credit scoring systems is high, as the local government institutions take the responsibility to ensure compliance. Third, the transparency of the algorithms within enterprise and personal credit scoring systems varies across the sample cities. In half of the investigated cities, it is highly transparent, while in the other half, it is moderately transparent (refer to Sections 5 and 6), indicating a mixed landscape regarding algorithmic transparency. Hence, we designate it as low/medium for these two systems. The city credit scoring system is highly transparent regarding the scoring method but is vague in terms of the incomplete results disclosure and the corresponding rewards and punishments, and is thus labeled as medium. Fourth, there is a lack of a separate oversight body to monitor the three credit scoring systems. However, both the enterprise and personal credit scoring systems at the city level are overseen by the NDRC, as is the city credit scoring system. Therefore, we regard the level of interventionism from the accountability perspective as medium for the enterprise and personal credit scoring systems, but high for the city credit scoring system.

It is instructive to observe how cities in China assume a dual role within the SCS framework, functioning both as the enforcers of the system and as subjects being evaluated. This duality underscores the SCS’s pervasive approach to governance that extends its reach to encompass all entities. In a more general sense, the city credit scoring system, like other city rankings, corresponds to the logic of “measurementality” that emerges “from privileging scientific techniques for assessing and measuring the environment as a set of standardized units that are further expressed, reified, and sedimented in policy and discourse” (Turnhout et al., Reference Turnhout, Neves and De Lijster2014, p.583). It is a manifestation of the trend toward a “metric society” (Mau, Reference Mau2019; Shore and Wright, Reference Shore and Wright2015).

Within the conceptual framework, the algorithm stands out among the eight factors, exhibiting correlations with factors such as scoring and enforcement. These correlations foster a synergistic effect, amplifying the interventionism degree of the SCS. In fact, different departments, such as the tax bureau, score or evaluate enterprises or individuals independently. With a specific algorithmic model, scores or ratings from various departments can be transformed to one single credit score. In enterprise and personal credit scoring systems, computational algorithms enhance the control loop by automatically associating rewards and punishments with scores, thereby facilitating enforcement. In summary, our research concludes that the SCS represents a government-centric, interventionist approach to governing societies, which aligns with Dai’s assertion (Dai, Reference Dai2018). Furthermore, our analysis indicates differences between the three systems: the city credit scoring system exhibits the highest level of interventionism with the state playing a more dominant role, while the degree of interventionism for enterprise and personal credit scoring systems is slightly lower.