1 Introduction

Cases do not have any precedential status in socialist legal systems. In the People’s Republic of China, ‘judges do not have law-making power and the courts generally do not follow precedent’ (Zhang 2017). In the last decade, however, the SPC has sought to interpret and unify law by promulgating Guiding Cases (‘zhidaoxing anli’). Guiding Cases must be consulted by all judges in the course of their adjudicatory duties. Guiding Cases have therefore been described as ‘a new source of “judge-made law” in China’ (Liu 2021) and ‘the remarkable terminus a quo’ of China’s trend ‘toward embracing case law’ (Wang 2020). One commentator has even speculated that ‘“judicial activists” might be able to assert greater SPC authority by developing a body of model interpretive approaches’ through Guiding Cases (Jia 2016).

At the same time, a substantial body of research contends that Guiding Cases have not altered the fundamental tenets of socialist legality. The terms ‘Guiding (zhidaoxing)’ and ‘Cases (anli)’ signify that ‘the SPC does not intend to establish a system of binding precedents’, that is, cases that ‘can be directly referred to as a legal basis for a court decision’ (Ahl 2014). Ambiguity about the binding quality of Guiding Cases is also reflected in the SPC’s requirement that lower courts ‘refer to (canzhao)’ these decisions when resolving similar disputes (Deng 2015). Ostensibly, the verb ‘refer to (canzhao)’ implies that the requirement is ‘neither compulsory nor discretionary’ (Peng 2018) and that Guiding Cases occupy an intermediate status between codified law and prior judicial decisions (Lei 2015). These theoretical debates have been reinforced by empirical studies confirming that Guiding Cases are very seldom cited to by lower courts (Gechlik et al. 2016; Guo and Sun 2021; Ma 2019; Cohen 2017; Daum 2017). The paucity of citations has conventionally been interpreted as evidence of the futility of transplanting stare decisis into a socialist legal regime (Wang 2019).

Citations, however, are an imperfect measure of the influence of cases, especially in jurisdictions that do not recognize judicial opinions as a source of law. This article proposes to quantify the impact of Guiding Cases on adjudicatory outcomes by looking at unattributed text reuse. The unattributed reuse of text from Guiding Cases in lower court judgments is detected using computational methods and validated by trained humans. To preview our results, we find more instances of unattributed text reuse than citations for the two Guiding Cases studied here. Guiding Cases seem to have a bearing on judicial outcomes even when they are not invoked by name. Our findings not only illuminate the practical operation of the Chinese legal system but also contribute to a more sophisticated understanding of legal culture and tradition. From a comparative perspective, the apparent reticence of Chinese courts to cite judicial decisions provides a foil to the rich literature analyzing citation networks using computational methods, focused mostly on common law jurisdictions (Frankenreiter and Livermore 2020). That literature has addressed whether and how citations evince better decision-making, for example (Baker et al. 2009; Ash and MacLeod 2021), or how citations manifest the partisan attitudes of judges (Frankenreiter 2017; Gulati and Choi 2008). We demonstrate here one technique for detecting the influence of cases in jurisdictions that reject the notion of stare decisis.

2 Background

In 2011, the SPC formally established the Guiding Case system ‘[i]n order to summarise adjudication experiences, unify the application of law, enhance adjudication quality, and safeguard judicial impartiality.’Footnote 1 Guiding Cases address a wide variety of legal topics ranging from breach of contract to homicide to unfair competition to liability for traffic accidents. Each Guiding Case consists of seven key sections, namely ‘Title’, ‘Keywords’, ‘Main Points of the Adjudication’, ‘Related Legal Rules’, ‘Basic Facts of the Case’, ‘Results of the Adjudication’, and ‘Reasons for the Adjudication’. Beginning in 2015, the ‘Main Points of the Adjudication’ of Guiding Cases (‘Main Points’) must be ‘refer[red] to (canzhao)’ by courts at all levels when adjudicating similar cases.Footnote 2 The Main Points are abstract rules distilled by the adjudication committee of the SPC from the original judgments (Hu 2014).

Take Guiding Case No. 18. This case involves a labour contract dispute resolved by a basic people’s court in Zhejiang Province in December 2011 and designated by the SPC as a Guiding Case on November 8, 2013.Footnote 3 The employee Wang Peng, was assigned a C2 grade in the semi-annual performance assessments for the second half of 2008, the first half of 2009, and the second half of 2010. Under the employer ZTE’s corporate policy, a grade of C2 meant that the employee’s ‘performance need[ed] improvement’ and that the employee would be deemed, in principle, ‘incompetent for the job’. ZTE eventually terminated Wang Peng’s services on the grounds that he was incompetent for the job after the reassignment of his role in January 2009. In 2011, Wang Peng filed for labour arbitration and was awarded compensation for ZTE’s unlawful termination of his labour contract. ZTE then challenged the arbitration decision in a local court based on Article 40 of the Labour Contract Law which permitted an employer to terminate a labour contract if ‘the employee [was] incompetent to his position or remain[ed] so after training or changing his position’.Footnote 4 The Basic People’s Court of Binjiang found that the C2 grade did not establish an employee’s ‘incompeten[ce] for the job’ and that ZTE had also failed to prove the reassignment was due to Wang Peng’s incompetence. It therefore upheld the arbitral award. The SPC selected the original judgment as a Guiding Case, appending to it the following Main Points:

A worker being ranked at the bottom in an employer’s grade assessment is not equivalent to [him] being “incompetent for the job” and does not meet the statutory requirements for unilateral termination of a labour contract. An employer cannot rely on this ground to unilaterally terminate a labour contract.Footnote 5

As illustrated by this example, a Guiding Case narrates the facts of the case and lends the SPC’s imprimatur to the result and to a particular interpretation of the law.

According to the ‘Detailed Rules for the Implementation of the Provisions of the Supreme People’s Court on Case Guidance’, courts must quote in the reasoning sections of their decisions the serial number and Main Points of an applicable Guiding Case.Footnote 6 The Guiding Case system thus represents an attempt to introduce case-based adjudication into the Chinese legal system (Shao and Guo 2022; Yu and Gurgel 2012). Proponents argue that ‘illustrating the law through [Guiding] Cases’ can help judges ‘discern adjudicatory rules’ and ‘provide supporting reasoning for their judgments” (Sun 2021). If successful, Guiding Cases might also become an instrument of judicial policymaking. By carefully picking the judgments to be designated as Guiding Cases and hence, to be referred to by all courts, the SPC could not only fill the interstices of legislation but do so in an advised and deliberate manner.

On the one hand, this development is not a complete break from the past. Since 1985, the SPC Gazette has published exemplary cases for the edification of lower courts.Footnote 7 Compared to legislation and other normative documents, exemplary cases may be released with little procedure or consultation, making them an especially suitable vehicle for developing the law in evolving socio-economic circumstances (Zhang 2017). Yet, exemplary cases were never meant to be binding on lower courts. Guiding Cases, on the other hand, are supposed to be referred to by all courts when adjudicating similar disputes. This aspect of Guiding Cases has provoked controversy because socialist legality denies judicial decisions any precedential status. Some jurists have tried to square the circle by positing an intermediate status for Guiding Cases. Guiding Cases are sometimes characterized as a ‘quasi-legal source’, enjoying a status that is subordinate to enacted laws but superior to other judicial opinions (Lei 2015; Peng 2018). Other jurists are more sceptical of the legal authority of Guiding Cases. For instance, a then-director of the SPC's research office reasoned straightforwardly that since legislative powers are solely vested in the National People’s Congress and its standing committee, any changes to existing law should be made through the legislative process (Zhou 2010). ‘China does not belong to a “separations of powers” country’, he elaborated, and ‘the SPC cannot formulate new legal norms’ (Zhou 2010).

Empirically, researchers have documented the marginal impact of Guiding Cases on judicial practice. Citations to Guiding Cases are sparse, and many Guiding Cases are never cited at all (Daum 2017). One study counted only 181 citations to Guiding Cases in all decisions handed down between 2010 and 2015 (Gechlik et al. 2016). Another counted 2818 citations to Guiding Cases among all judgments published on Chinalawinfo as of December 31, 2020 (Guo and Sun 2021). A search for the term ‘Guiding Cases’ on the official portal China Judgments Online returned 561 hits out of the 8,723,182 files uploaded to the site as of 2016 (Cohen 2017). Using keywords—‘zhidaoxing anli’, ‘zhidao anli’, and ‘anli zhidao zhidu’—to query the same national database, another investigator identified 1545 cases among those decided from December 2011 to May 2017 as having cited Guiding Cases (Zhang 2018). To illustrate the limited relevance of Guiding Cases to the day-to-day practice of the lower courts, a judge in a provincial high court performed a similar analysis and found approximately 620 judgments in 2016 and 810 judgments in 2017 referring to these cases (Ma 2019). As she noted, ‘[c]ompared to around 20 million cases closed by courts nationwide each year, the above figures are almost negligible’ (Ma 2019).

The ‘extremely low’ incidence of citations is taken as symptomatic of ‘the dysfunction of the [G]uiding [C]ase system as a type of case law’ (Wang 2019) and as proof of the incongruity of case-based adjudication in China (Ahl 2014; Zuo and Chen 2015; Finder 2017; Jia 2016). This ‘lack of enthusiasm’ (Zhang 2017) for Guiding Cases has been imputed to legal culture and tradition, including the unfamiliarity of Chinese judges with analogical reasoning (Xiang 2016; Research Team of High People's Court of Sichuan Province 2012).

Citations, however, do not always demonstrate persuasiveness, even in common law jurisdictions. Judges may discuss a case in their opinions only to distinguish it. Indeed, courts have declined to follow Guiding Cases because of differences in case types and material facts (Tan and Shen 2021). In the context of Chinese socialist legality, the doctrine that judicial decisions are not a source of law may inhibit courts from citing Guiding Cases even when they are treated as conclusive (Zhang 2017). Indeed, some judges claim that they avoid citing Guiding Cases for fear that it exposes their judgments to another ground of attack on appeal (Lin 2015).

To overcome these difficulties, a research team at Peking University looked for cases where (1) the parties referred to Guiding Cases, and although (2) the judgments rendered did not cite these Guiding Cases, (3) ‘the judicial outcomes were consistent with the spirit of the Guiding Cases’. The study reported 4196 cases satisfying these criteria (Guo and Sun 2021). It is plausible that reference by the parties to Guiding Cases influenced the judicial outcomes in these disputes. Even so, the reported count—like all methods for quantifying the impact of judicial decisions—could be over- or under-inclusive. It would be over-inclusive if courts were already disposed to rule the same way independently of Guiding Cases. It would be under-inclusive if judges adhered to Guiding Cases even when they were not raised by the litigants. This latter possibility is not remote given the SPC’s commitment to the Guiding Case system. Since 2020, the SPC has required a presiding judge to locate and list similar cases for filing purposes.Footnote 8 Hence, empirical research that relies solely on explicit mentions of Guiding Cases in court or litigation documents may fail to draw a fair conclusion as to the true impact of the Guiding Case system on Chinese adjudicative practice.

3 Method

We bring fresh evidence to the debate on Guiding Cases by looking at the diffusion of their reasoning in the judicial system. The method adopted here searches for instances of text reuse between the Main Points of Adjudication of Guiding Cases and the legal rationales given by lower courts. Our premise is that reuse of text uniquely ascribable to the Main Points of Adjudication is a strong indicium of the influence of Guiding Cases on judicial decision-making. This idea is implemented by applying the Basic Local Alignment Search Tool (or BLAST) algorithm to Chinese legal text.

BLAST was originally developed to identify regions of similarity between genomic or protein sequences (Altschul 2014).Footnote 9 But the algorithm can also be used to locate the replication of text fragments in large document collections. BLAST and its derivatives have been employed by researchers in the digital humanities to analyse the reuse of text in Denis Diderot and Jean d’Alembert’s Encyclopédie (Olsen et al. 2011), in over a hundred and fifty years of Finnish newspapers (Salmi et al. 2021), and in the classic novels of Chinese literature (Vierthaler and Mees 2019). In particular, Vierthaler and Gelein adapt BLAST to quantify and visualise intertextuality in late imperial Chinese prose, finding, among other things, extensive copying from the middle section of Water Margin in the first part of Plum in the Golden Vase (Vierthaler and Mees 2019).

Following Vierthaler and Gelein, we apply the BLAST algorithm to Chinese characters. Intuitively, the algorithm builds the longest shared sequence as measured by common characters while penalising character differences between two sequences. The starting point is the Levenshtein distance between the sequences, given by the minimum number of single-character substitutions, deletions, or additions needed to transform one sequence to the other (Levenshtein 1966). Table 1 shows how the Levenshtein distance is computed for a pair of text snippets. In the example, the Levenshtein distance between the two snippets is four: two substitutions, an insertion and a deletion. The BLAST similarity score between two sequences is defined as \(1-\frac{d}{l}\) where d is the Levenshtein distance and l the length of the sequences. Keeping the same example, the similarity score between the two snippets is \(1-\frac{4}{17}\) which is approximately 0.765. The algorithm runs on a source text and a target text and comprises the following steps:

  • Decompose the two texts to be compared into overlapping n-grams. These n-grams are called seeds.

  • Record the occurrence of each seed in either text in an index.

  • For every seed that occurs in both texts, go to the respective seed locations in the texts. The similarity score at this point is 1 since the seed occurs identically in both texts.

  • While the similarity score between the two sequences is above the threshold, expand the sequence in each text rightwards by one character and compute the similarity score between the two expanded sequences.

    • If the length of the expanded sequences exceeds a maximum number m, then only the last m characters of the sequences are used to compute the similarity score.

  • Once the similarity score between the two sequences falls below the threshold, return the match if the sequences exceed a minimum length. The match consists of the two sequences at the last point where their similarity scores increased.

We deploy this algorithm to generate matches between Guiding Cases and local court decisions.

Table 1 Illustration of Levenshtein distance for Chinese legal text
Full size table

The BLAST algorithm could in principle be applied to all Guiding Cases and all local court decisions. However, to make human validation of the machine-generated matches feasible, we restrict the scope of our analysis along two dimensions. First, we limit our attention to Guiding Cases 24 and 60. Guiding Case 24 concerns a tortfeasor’s liability for a traffic accident when the harm caused is magnified by the victim’s own pre-existing physical condition.Footnote 10 Guiding Case 60 clarifies the circumstances under which the amount of a “special and valuable ingredient,” emphasised in the labels or manuals of a food product, should be disclosed to the consumer.Footnote 11 We select these two Guiding Cases because prior research has identified them as being among the most cited (Zhang 2020; Guo and Sun 2019). This means that there will be a healthy number of overt citations to serve as a baseline for assessing the prevalence of unattributed text reuse.

Second, we look only for text reuse by the Beijing and Shanghai courts. Courts in both municipalities pioneered the consultation of previously decided cases even before a similar case search mandate was implemented nationwide in July 2020. For example, in 2019, the Beijing High People’s Court issued an Opinion instructing judges on how to search for analogous cases and identify inconsistent adjudicatory standards in civil litigation.Footnote 12 The Opinion lists several types of prior judicial decisions to be considered, ranking Guiding Cases as the most persuasive.Footnote 13 Back in 2015, the Beijing Intellectual Property Court had already established a research database of prior judicial decisions and encouraged parties to submit relevant cases in support of their claims (Guo et al. 2017; Yang 2017). In Shanghai, too, judges have been checking internal court databases for similar cases for several years (Yu and Wu 2019). Because Beijing and Shanghai promoted judicial consideration of past cases early on, there is reason to suppose that the proportion of citations to unattributed text reuse will be higher in these two municipalities as compared to other Chinese jurisdictions.

The full texts of the Guiding Cases are publicly available, including those for Guiding Cases 24 and 60. However, only the Main Points of Adjudication are used as source texts.Footnote 14 In addition, names of statutes or regulations are deleted from the Main Points of Adjudication. This step pre-empts candidate matches that are merely common references to legislative and regulatory materials. Most significantly, we excise the first paragraph of the Main Points of Adjudication of Guiding Case 60 because it merely restates a provision of an administrative document promulgated by a national agency. It would be quite impossible to tell whether a local judgment reproducing that language was influenced by the administrative document or the Guiding Case. We therefore elect the more conservative approach of precluding all such references from being matched.

Our text reuse algorithm analyses the texts of Beijing and Shanghai local court decisions. The full texts of these decisions are not publicly available for bulk download. We obtained a dataset under licence from Chinalawinfo. The corpus of decisions from January 1970 to September 2021 comprises approximately 3 million documents. The number of decisions per year has increased markedly over time due to the burgeoning caseload of these courts and also to judicial reforms promoting transparency and disclosure. To ensure that any language reused from the Guiding Cases is integral to the adjudicative outcome, we use only the reasoning sections of decisions as target texts. That section is segmented via regular expressions and then extracted for further, minimal pre-processing.Footnote 15 Special characters are stripped from the decisions. We then execute a complete set of pairwise comparisons between the Main Points of Adjudication of Guiding Cases 24 or 60 and the reasoning sections of decisions issued by local courts in Beijing and Shanghai from January 1970 to September 2021.

As a benchmark, we also conduct string searches for citations to the Guiding Cases by local courts. Queries consist of the terms ‘zhidao anli’, ‘anli zhidao’, or ‘zhidaoxing anli’, preceded or followed by the number of the Guiding Case in question, i.e., ‘24 hao’, ‘di 24 hao’, ‘ershisi hao’, or ‘di ershisi hao’ for Guiding Case 24 and ‘60 hao’, ‘di 60 hao’, ‘liushi hao’, or ‘di liushi hao’ for Guiding Case 60. These string searches can be conducted over the reasoning sections or entire texts of local judgments. The former option risks undercounting the number of decisions which explicitly mention the Guiding Cases. The latter option might produce an overcount since the Guiding Cases could have been discussed but found inapposite—hence their omission from the reasoning section. In the interest of completeness and transparency, we will report both counts.

4 Results

4.1 Unvalidated matches

After some experimentation, the parameters for the algorithm are calibrated at 4 for the length of the n-grams or seeds, 0.5 for the similarity score threshold, 30 for the minimum length of a match, and 30 for m, the maximum number of characters used to compute the similarity score. These parameters are adopted throughout. The BLAST algorithm identifies 488 decision-level matches for Guiding Case 24 and 103 decision-level matches for Guiding Case 60. By comparison, the string searches return 46 decisions citing Guiding Case 24 and 22 decisions citing Guiding Case 60 in their reasoning sections. The counts stand at 78 and 95 respectively if taken over entire decisions.

Some decisions cite the Guiding Cases and also recite their Main Points. These are instances of attributed text reuse. By contrast, some decisions reproduce the Main Points but do not credit the Guiding Cases. These instances of unattributed text reuse are of primary interest here, and their decision-level counts are given in Table 2. For Guiding Case 24, the number of decisions engaging in unattributed text reuse is approximately six to ten times the number of decisions citing the Guiding Case. For Guiding Case 60, the number of decisions engaging in unattributed text reuse is approximately four times the number of decisions citing the Guiding Case in the outcome sections. It is approximately four fifths the number of decisions citing the Guiding Case anywhere in the judgment.Footnote 16

Table 2 Decision-level counts for citations to and unattributed text reuse from Guiding Cases 24 and 60
Full size table

The BLAST algorithm may generate false positives, that is matches that do not actually capture references by local court decisions to the Guiding Cases. For example, the algorithm identifies matches between the Main Points of Adjudication of Guiding Case 24 and local court decisions that predate the issuance of the Guiding Case itself. These decisions could not possibly be applying Guiding Case 24. The textual similarity between the matched sequences can only be coincidental.

4.2 Validated matches

To validate our matches, all sequence pairs are manually inspected and labelled. A match is labelled as a true positive if the Guiding Case appears to have supplied the basis of the local court decision and a false positive otherwise. This is a task calling for legal skill and judgment. All matches are labelled twice by human coders who either hold a Chinese law degree or have passed the Chinese bar examination. These human coders see only the matched sequences and the relevant source and target texts. In particular, they are blinded from the similarity scores computed by the algorithm. For our purposes, a match is labelled as a true positive even if the matched sequence is not the most suggestive nexus between the local court decision and the Guiding Case. All differences between the human coders are resolved by consensus.

4.2.1 Matches for Guiding Case 24

Is a tortfeasor liable for the entirety of a victim’s injuries when those injuries are exacerbated by the victim’s pre-existing condition? In Guiding Case 24,Footnote 17 the Binhu Basic People’s Court reduced the damages for a pedestrian’s injuries by 25% because of a pre-existing medical condition. The defendant had grazed the pedestrian while driving his car and there was no dispute that the former bore full responsibility for the accident. The trial court concluded, however, that the victim’s osteoporosis also contributed to the severity of his fractures and was a factor that must be accounted for in the calculation of damages. The Wuxi Intermediate People’s Court overturned the decision below. Invoking Article 26 of the Tort Liability LawFootnote 18 and Article 76 of the Road Traffic Safety Law,Footnote 19 the appellate court reasoned that a reduction in damages could only be awarded if there were fault on the victim’s part. The physical condition of the victim did not constitute such fault, ‘osteoporosis from old age being an objective element of the accident’s consequences and not having a causative relationship in law’. The Main Points of Adjudication of Guiding Case 24 affirmed that where ‘the victim of a traffic accident is faultless, the effect of his physical condition on the consequences of the harm is not a legal circumstance that can reduce the responsibility of a rights infringer’. In endorsing the decision of the Wuxi Intermediate People’s Court, the SPC articulated a version of the eggshell-skull rule familiar to the common law.

Of the 488 machine-generated matches for Guiding Case 24, 282 are true positives; the remainder are false positives. Most of the false positives feature generic legal phrases contained in the applicable rules, such as ‘the consequences of the harm’ or ‘the responsibility of a rights infringer’. For example, the BLAST algorithm matched the following snippet from a local judgment to a similar one from Guiding Case 24: ‘is also at fault for the consequences of the harm, [it] can reduce the responsibility of a rights infringer. Infringing others causes physical injuries’.Footnote 20 On closer examination of the legal reasoning of that decision, however, it transpires that the local court did and could not have taken the Guiding Case into account because the victim there did not suffer from a pre-existing physical condition nor were the facts somehow analogous.

Of the true positives, many are straightforward instances of local judgments reusing language from the Main Points of Adjudication of Guiding Case 24. In some of these cases, the local courts followed Guiding Case 24 despite potentially distinguishable facts, such as the victim and the tortfeasor both being at fault. Finally, 139 matches are labelled as true positives because the local judgments recited some of the arguments given in the Reasons for the Adjudication section of the Guiding Case. Even though these judgments might not always have repeated the most distinctive or critical phrases of the Main Points, it is evident on inspection that they were retracing the legal logic of Guiding Case 24.

Table 3 gives the decision-level incidence of citations to and validated, unattributed text reuse of Guiding Case 24. Figures 1 and 2 show the counts over time. Overall, the number of decisions that engaged in validated, unattributed text reuse is approximately four to six times the number of decisions citing the Guiding Case. How significant are these numbers in the broader scheme of things? Guiding Case 24 is about pre-existing conditions and was published on 26 January 2014. A string search for the phrases ‘pre-existing condition’ and ‘Tort Liability Law’ turned up 397 decisions rendered in or after January 2014 containing both those terms. On a conservative estimate then,Footnote 21 between 11.6 and 19.6% of decisions that could have applied the Guiding Case cited it and between 67.7 and 70.5% of them reproduced language from the Guiding Case without acknowledging so.

Table 3 Decision-level counts for citations to and validated, unattributed text reuse from Guiding Cases 24
Full size table
Fig. 1
figure 1

Decision-level counts for citations to and validated, unattributed text reuse from Guiding Case 24 over time, citations being counted only if they occur in the reasoning sections of local court decisions

Full size image
Fig. 2
figure 2

Decision-level counts for citations to and validated, unattributed text reuse from Guiding Case 24 over time, citations being counted if they occur anywhere in the local court decisions

Full size image

4.2.2 Matches for Guiding Case 60

Guiding Case 60 concerns a dispute arising under Article 20 of the Food Safety Law which defines national food safety standards to include ‘[r]equirements concerning food safety and nutrition labels, markings and user instructions’.Footnote 22 The General Principles for the Labeling of Prepackaged Foods (‘General Principles’) mandated that ‘[i]f food labels or food manuals particularly emphasise the addition of some type or amount of valuable or special ingredients, then the amount added of such emphasised ingredients should be indicated’.Footnote 23 The General Principles was promulgated by General Administration of Quality Supervision, Inspection, and Quarantine and the Standardization Administration and implemented on October 1, 2005—before the Food Safety Law came into effect. In Guiding Case 60,Footnote 24 the Yancheng Intermediate People’s Court upheld the penalty levied by the Dongtai Administration for Industry and Commerce on the Aokang Company. In packaging its blended oil product, the Aokang Company had prominently displayed the word ‘olive’, accompanied by graphic depictions of ‘olives’, and had also represented on hangtags that the product was made ‘with 100% extra virgin olive oil from Italy’. Olive oil, the appellate court also noted, had a market price and nutritional value that are on average higher than that of other edible oils. The Aokang Company should have but did not state the amount of olive oil added to the blend. This amounted to a violation of food safety standards.

Of the Main Points of Adjudication of Guiding Case 60, the first essentially restates the rule stated in the General Principles and declares violation of the rule to be a violation of the Food Safety Law. The second expounds ‘emphasise’ as encompassing ‘names, differences in colour, font, font size, graphics, order of arrangement, written explanations, identical content that repeatedly appears, different content pointing to the same thing’. ‘Valuable and special ingredients’, on the other hand, ‘refers to special ingredients which are different from general ingredients, have a higher nutritional value for the human body, and have a market price and nutritional components that are often higher than other ingredients’.

Because of the legislative and regulatory background to Guiding Case 60, the validation of its matches presents some unique challenges. As a preliminary matter, recall that we excised that part of the Main Points of Adjudication of Guiding Case 60 which restates Article 4.1.4.1 of the General Principles. Hence, none of the sequences is matched based on language which appears extensively in both the Guiding Case and the General Principles. There is, however, another difficulty which is that GB7718-2011 Implementation Instrument also elucidates the meaning of ‘particularly emphasise (tebie qiangdiao)’ and ‘valuable and special (youjiazhi youtexing)’.Footnote 25 The Instrument provides guidance for interpreting the General Principles and predates the publication of Guiding Case 60. Specifically, the Instrument instructs that ‘particularly emphasise’ encompasses a food producer ‘attracting consumers’ attention to the product, ingredients, or components through the publicity of such ingredients or components’ and ‘highlighting or implying the addition or inclusion of one or more types of ingredients or components in the form of words on labels other than the contents of ingredient lists’.Footnote 26 The Instrument also specifies that ‘valuable and special’ refers to ‘relatively special ingredients or components’.Footnote 27 For ingredient or components to qualify as ‘valuable and special’, ‘the extent to which... [they] are beneficial to human body [has to be] beyond the extent that should be achieved by the food under normal circumstances’.Footnote 28 They also have to be ‘different in nature from normal ingredients or components of the food’.Footnote 29 The Instrument thus addresses the very issues touched on in Guiding Case 60 and employs the same vocabulary and concepts to do so.Footnote 30 There are, however, subtle differences between the explanations furnished in the Instrument and the Guiding Case and our human coders can tell them apart. Of the 103 matches returned by the algorithm, 78 are true positives and 25 are false positives. Most false positives involve repetition of the term ‘valuable and special’. Many of the true positives, on the other hand, reprise the Guiding Case’s explanation of ‘emphasise’ and ‘valuable and special ingredients’.

Table 4 gives the decision-level incidence of citations to and validated, unattributed text reuse of Guiding Case 24. Figures 3 and 4 plot the counts over time. Once again, how significant are these numbers? Guiding Case 60 was published on 20 May 2016 and our analysis here only looked at the second of the two Main Points, i.e. what it means for an ingredient or component to be ‘valuable and special’. A string search for the phrase ‘valuable and special’ across all local decisions yielded 650 opinions rendered in or after May 2016 containing that term. So, on a conservative estimate, between 3.4 and 14.6% of decisions that could have applied the Guiding Case cited it and between 9.5 and 11.7% of them reproduced language from the Guiding Case without acknowledging so. These figures are lower than those for Guiding Case 24, perhaps because of the comparative recency of Guiding Case 60.

Table 4 Decision-level counts for citations to and validated, unattributed text reuse from Guiding Cases 60
Full size table
Fig. 3
figure 3

Decision-level counts for citations to and unattributed text reuse from Guiding Case 60 over time, citations being counted only if they occur in the reasoning sections of local court decisions

Full size image
Fig. 4
figure 4

Decision-level counts for citations to and unattributed text reuse from Guiding Case 60 over time, citations being counted if they occur anywhere in the local court decisions

Full size image

5 Discussion

Do Guiding Cases influence judicial outcomes in China? How do Chinese judges who are reticent to cite Guiding Cases as a legal source nevertheless enforce their authoritative guidance? Our observations here are limited to the text of judicial documents but several patterns emerge from a close reading of all the validated matches.

The first is the practice of statutory ventriloquy. Consider Guiding Case 60. Of the 78 true positives, 76 local judgments quietly adopted the Guiding Case’s gloss of the terms ‘emphasise’ and ‘valuable and special ingredients’ while referencing either the General Principles or the Instrument. As a variation on the theme, 22 of those judgments concatenate phrases from the General Principles or the Instrument and the Guiding Case and attribute the ensuing collage to the former set of materials. For example, in Liu Jia v. Beijing Yonghui Supermarket Co., Ltd., the court, in determining whether a label for tea seeds and olive oil particularly emphasised olive oil as a component, stated that.

‘particularly emphasise’ refers to a [food] producer attracting consumers’ attention to the product, ingredients, or components through the publicity of such ingredients or components and stressing in labels through various forms of demonstration, including names, differences in colour, font, font size, graphics, order of arrangement, written explanations, identical content that repeatedly appears, and different content pointing to the same thing.Footnote 31

Article 4.1.4.1 of the General Principles was cited as the source of these propositions but the elaboration of ‘particularly emphasise’—the very term at issue—comes in part from the Main Points of Guiding Case 60. The reproduction of the Main Points was not gratuitous. The court held that the food product in question did not emphasise olive oil as a component because the fonts and the sizes of ‘tea seeds’ and ‘olive’ in the label were the same. The court also considered that the tagline on the product label—‘Jinhao Tea Oil Eat Less Oil Eat Better Oil’—did not amount to a claim that olive oil was more beneficial to human body than normal food. In so doing, it applied the part of the Main Points it did not recite.

Another example is given by Shoulan Li v. China Resources Vanguard Shop where the questions were whether olive oil was emphasised by a label for edible olive and sunflower oil and whether olive oil constituted a ‘valuable and special’ component.Footnote 32 Invoking the Food Safety Law and the General Principles as legal bases, the court asserted that

[e]mphasise means that food producers attract consumers’ attention to the product, ingredients, or components through the publicity of such ingredients or components and highlight or imply the addition or inclusion of one or more types of ingredients or components in the form of words on labels other than the contents of ingredient lists. Valuable and special ingredients refer to [ingredients which] have a higher nutritional value for the human body and have a market price and nutritional components that are often higher than other ingredients.Footnote 33

The definition of ‘emphasise’ is drawn from the Instrument and the definition of ‘valuable and special ingredients’, from Guiding Case 60. The court proceeded to deem olive oil as a valuable and special ingredient since

[t]he Duoli olive and sunflower edible blended oil Shoulan Li purchased was made of two ingredients olive oil and sunflower seed oil. The nutritional value and market price of olive oil were higher than sunflower seed oil.Footnote 34

More interestingly, the court, despite quoting the language of the Instrument rather than the Guiding Case, also turned to the Main Points to determine whether olive oil was ‘emphasised’ as a component. Answering in the affirmative, the court noted, among other things, that ‘olive oil’ repeatedly featured on the packaging through graphics, font and written explanations.

Besides statutory ventriloquy, judges also frequently retrace the jurisprudential grounds of the Guiding Cases. Rather than quote the Main Points of Adjudication, some local judgments simply rehearsed the underlying reasoning as given by the Guiding Cases. Consider Guiding Case 24. As previously related, courts are supposed to ‘refer to’ the Main Points when adjudicating similar cases and according to the Main Points of Adjudication of Guiding Case 24, if ‘the victim of a traffic accident is faultless, the effect of his physical condition on the consequences of the harm is not a legal circumstance that can reduce the responsibility of a rights infringer’.Footnote 35 But many judges do not reiterate the Main Points, even if they have taken instruction from them. Indeed, they often do not even mention the Guiding Case at all. Instead, they recapitulate the legal-doctrinal arguments advanced in Guiding Case 24, for example that ‘a physical condition does not constitute the victim’s fault prescribed by the Tort Liability Law’, ‘a physical condition is not a legal cause of the consequences of the harm’ and hence that ‘China’s legislation on traffic insurance does not warrant a mitigation of tort liability due to the victim’s physical condition’. Chunhua Zhang v. Ning Bian & China Pacific Property Insurance Co. Ltd. is illustrative.Footnote 36 The plaintiff dislocated her hip in a motor accident and claimed RMB 386,587.72 yuan in tort damages. At trial, the defendant’s insurance company submitted that a femoral neck fracture suffered by the plaintiff fifteen years ago exacerbated her injury and that the defendant’s liability should therefore be reduced. The court replied that

in traffic accidents, when calculating whether disability damages should be reduced, the analysis should be based on whether the victim is at fault for the occurrence or amplification of the injuries. Even if the old injury that plaintiff in this case had fifteen years ago, to some extent, affected the occurrence of the consequences of the harm, this does not constitute fault as prescribed by the Tort Liability Law or other laws. Plaintiff Zhang Chunhua should not, due to [the fact that his] personal physical basic condition had a certain effect on injuries and disabilities caused by the traffic accident, bear corresponding liability.Footnote 37

This response differs in only a few words from the following passage of the Reasons for the Adjudication section of Guiding Case 24:

in traffic accidents, when calculating whether disability damages should be reduced, the analysis should be based on whether the victim is at fault for the occurrence or amplification of the injuries. In this case, although the personal physical condition of plaintiff Rong Baoying, to some extent, affected the occurrence of the consequences of the harm, this does not constitute fault as prescribed by the Tort Liability Law or other laws. Rong Baoying should not, due to [the fact that his] personal physical condition had a certain effect on injuries and disabilities caused by the traffic accident, bear corresponding liability.Footnote 38

But nowhere does the judgment mention the Guiding Case. Chunhua Zhang is not an isolated example. Of the 282 true positives, there are a total of 139 cases like it.Footnote 39

6 Conclusion

Alien to the socialist conception of judicial power, the Guiding Case system allows the SPC to make policy through cases by picking—and, sometimes recasting—the judgments that must be ‘refer[red] to’ by all judges when adjudicating similar disputes. Previous empirical studies find, however, that Guiding Cases have had very little sway on the lower courts. Many of these studies reach this conclusion after counting the number of citations to Guiding Cases. By looking beyond citations to the reuse of text in the judgments of lower courts, we offer another perspective on the influence of Guiding Cases on the Chinese legal system. Our results indicate that formal citations represent only a fraction of judicial decisions relying on Guiding Cases. Guiding Cases are more influential than suggested by past studies.

Why do judges apply Guiding Cases but not invoke them by name? The best explanation lies in the confluence of bureaucratic incentives and legal tradition and culture (Howson 2008). The percentage of first-instance cases appealed, reversed, or petitioned is an important component of assessing judicial performance (Ng and Chan 2021).Footnote 40 Inconsistency between a judicial decision and a Guiding Case will be challenged by litigants and their lawyers at second instance, thereby hurting the evaluation and advancement of the responsible judge (Peng 2017; Liu and Xing 2018; Sun 2018a, b). Chinese judges are therefore professionally—and financially—rewarded for following Guiding Cases.

But there are no rewards for citing Guiding Cases, over and above following them. The prevailing party would certainly not complain about the court’s omission in mentioning a Guiding Case. The unsuccessful party would not be able to criticize the court for departing from a Guiding Case. And a judgment that otherwise conforms to a Guiding Case would not be disapproved for failing to make its adherence express (Wang 2022). Under these circumstances, judges may prefer to keep their reliance on Guiding Cases tacit. This preference could be strategic. As alluded to at the beginning, some judges believe that basing their decisions on Guiding Cases makes them more vulnerable on appeal (Lin 2015). The reticence to cite Guiding Cases might also stem from a lack of confidence in their own analogical reasoning skills (Sun 2022). Silence relieves judges from ‘the burden of reasoning with the Cases’ (Sun 2018a, b). At the same time, legal doctrine and orthodoxy continue to have a hold on the judicial mind. Although some judges openly refer to cases—even foreign ones—in their decisions, it remains that judicial opinions are not a source of law in socialist China. As one judge attested, ‘Guiding Cases are only for reference, no one dares to directly use them as the basis for adjudication in the specific case handling process’ (Zhang 2013). It is enough for judges to hew to the language of Guiding Cases, thereby ‘enhanc[ing] the acceptability of the[ir] judgments’ and ‘prevent[ing their] own judgments from being overturned on appeal’ (Sun 2018a, b).

The SPC advisedly chose the phrase ‘guiding case (zhidaoxing anli)’ over the term ‘precedent (panli)’ to forestall any accusation of establishing a case law system (Deng 2015).Footnote 41 In applying Guiding Cases, local judges engage in statutory ventriloquy or jurisprudential retracing to avoid citing them as a source of legal authority. Both techniques serve to maintain the dogma of legislative supremacy even as statutory law is being supplemented—and sometimes modified—through judicial initiative.Footnote 42 It is frequently observed that superficial convergence in legal rules and terminology may belie substantive divergence in action.Footnote 43 But as the empirical operation of the Guiding Case system demonstrates, performative resistance can also disguise and even facilitate an undercurrent of legal adaptation and assimilation. Guiding Cases may, in the fullness of time, come to be accepted as precedent in socialist China.Footnote 44