The initial version of this post was published in the spring of 2024 and proposed a simple model to link officially reported sentence numbers to officially reported prosecution ones. While the original model was successful, additional review of the primary sources and the statistics therein revealed a number of conflicts in the data that we considered necessary to address. As a result, we have now done a full rewrite of the post, including a detailed review of the original statistics, a more robust description of reported sentence numbers, and a modified approach to estimating the sentence numbers for more recent years.
Update on September 24, 2024: The analysis has been updated with new data for 2021, and a new source for 2018. The results remain unchanged, however, as the new data point of 2021 fits the previous estimate almost perfectly.
Update on October 7, 2024: There was an error in Method 4, with 0.75 being used (incorrectly) instead of 0.25 in computing numbers of individuals sentenced from individuals whose sentences went into effect. While important, this has not affected the results significantly.
Update on December 18, 2024: Statistics on the lengths of appeal procedures have been updated. However, this has not affected the results.
Judicial statistics from official sources in Xinjiang have played a vital role in establishing the scale of the incarcerations in the region, with the usual skepticism towards official government data being offset by the data accurately reflecting the overall detention trends of the past decade. Specifically,
- the increases in detentions and sentences that followed the start of the “People’s War on Terror” in 2014,
- the unprecedented spike of incarcerations in 2017-2018, and
- the subsequent fall in more recent years
have all been consistently reflected in the statistics, making for a relatively rare phenomenon where a source describing the incarcerations in numbers on the provincial level has been both official and accurate. This has provided an important instrument for advocates and other concerned parties in emphasizing the gravity of the situation, which may incidentally be the reason for why the Chinese authorities essentially cut access to these data – court reports stopped being posted in 2019, with yearbooks publishing only limited statistics, while procuratorate reports stopped being posted in 2021. Of paramount interest have been the statistics on the numbers of criminal sentences, since these serve as a direct reflection of the people interned in Xinjiang through the official court system (camp internments excluded).
Despite their apparent accuracy, official sources for the numbers or magnitudes of individuals sentenced have suffered from a lack of precision, and have at times provided conflicting figures that make detailed analysis difficult. In the case of conflicting data, it is often hard to decide which specific numbers to use, or if particular numbers should be discarded altogether (either by virtue of incomplete reporting or, more bluntly, potential typos or errors).
To address this issue systematically, we have compiled a variety of methods (described in Appendix A) for obtaining ranges on the sentence numbers from the different statistics available, notably:
- numbers of criminal cases concluded (both first-instance and total),
- numbers of individuals sentenced in the same year as they were indicted (by non-Bingtuan procuratorates),
- numbers of individuals whose sentences went into effect,
- numbers of individuals sentenced in specific categories of cases.
For most methods, we have sought to obtain reliable ranges on the numbers of individuals sentenced by building empirical relations (e.g., inferring how many were sentenced when all we know is the number of cases), which we constructed using the available empirical data for the relevant judicial elements, such as the Bingtuan vs. non-Bingtuan rate, the appeal rate, the rate between first-instance and other cases, and the average length of an appeal process.
In the analysis that follows, we apply the various methods to the available statistics to obtain corresponding ranges on the numbers of individuals sentenced in each year. We then check if the ranges are consistent, addressing any conflicts on a case-by-case basis and rejecting those sources or statistics that conflict with the majority. Following the removal of any/all inconsistent sources, we then take the largest consistent range for the data in the sources that remain, thereby obtaining lower and upper bounds/estimates on the individuals sentenced for each year.
To conclude, we then address the lack of reporting since 2019, and propose a method that uses prosecution/indictment numbers as a proxy for estimating the numbers sentenced in 2019, 2020, and 2022 (the number for 2021 is available in an official source).
Review of available sentencing data (2007-2018, 2021)
· 2007 ·
Source (only): 2008 Xinjiang Yearbook
Method 1: The yearbook reports a total of 17698 first-instance criminal cases concluded, which by Method 1 gives a range of 17698 to 35396 individuals sentenced.
Method 3: The yearbook reports a total of 17542 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan). Method 3 suggests this to be the lower bound on the number sentenced, with no upper bound available for this year. Consequently, Method 3 suggests at least 17542 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 17698 individuals (this is identical to the number of first-instance cases and may be a copy-paste typo, but we assume coincidence for now). Method 4 requires an upper bound on the sentences from the previous year to calculate the lower bound for the year in question, which is not available. Consequently, we can only calculate an upper bound, with Method 4 suggesting no more than 18388 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2007. A ✔️ indicates that the given method-source pair is consistent (note: a method-source will always be consistent with itself).
| Method 1 17698 ≤ S ≤ 35396 | Method 3 17542 ≤ S | Method 4 S ≤ 18388 | |
| Method 1 17698 ≤ S ≤ 35396 | ✔️ | ✔️ | ✔️ |
| Method 3 17542 ≤ S | ✔️ | ✔️ | ✔️ |
| Method 4 S ≤ 18388 | ✔️ | ✔️ | ✔️ |
Conclusion: No conflicts are present, with the largest range that satisfies all statistics and corresponding ranges being:
17698 ≤ people sentenced in 2007 ≤ 18388
· 2008 ·
Source (only): 2009 Xinjiang Yearbook
Method 1: The yearbook reports a total of 17593 first-instance criminal cases concluded, which by Method 1 gives a range of 17593 to 35186 individuals sentenced.
Method 3: The yearbook reports a total of 16848 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 22763 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 21459 to 25246 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 16807 individuals, which by Method 4, coupled with the upper bound of 18388 for the previous year, gives a range of 16117 to 17462 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2008. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency.
| Method 1 17593 ≤ S ≤ 35186 | Method 3 21459 ≤ S ≤ 25246 | Method 4 16117 ≤ S ≤ 17462 | |
| Method 1 17593 ≤ S ≤ 35186 | ✔️ | ✔️ | ✘ |
| Method 3 21459 ≤ S ≤ 25246 | ✔️ | ✔️ | ✘ |
| Method 4 16117 ≤ S ≤ 17462 | ✘ | ✘ | ✔️ |
Conclusion: The range given by Method 4 conflicts with the ranges of both Method 1 and 3, and is significantly lower. Appealing to the majority, we exclude Method 4, for a resulting largest consistent range of:
21459 ≤ people sentenced in 2008 ≤ 25246
· 2009 ·
Source 1: 2009 court work report, reported by China Daily
Method 2: The report gives 20770 as the total number of criminal cases concluded, which by Method 2 gives the range of 12462 to 37386 individuals sentenced.
Source 2: 2010 Xinjiang Yearbook
Method 1: The yearbook reports a total of 17516 first-instance criminal cases concluded, which by Method 1 gives a range of 17516 to 35032 individuals sentenced.
Method 3: The yearbook reports a total of 17066 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 22687 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 22981 to 27037 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 10613 individuals, which by Method 4, coupled with the upper bound of 25246 for the previous year, gives a range of 9666 to 11027 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2009. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency.
| Method 1 Source 2 17516 ≤ S ≤ 35032 | Method 2 Source 1 12462 ≤ S ≤ 37386 | Method 3 Source 2 22981 ≤ S ≤ 27037 | Method 4 Source 2 9666 ≤ S ≤ 11027 | |
| Method 1 Source 2 17516 ≤ S ≤ 35032 | ✔️ | ✔️ | ✔️ | ✘ |
| Method 2 Source 1 12462 ≤ S ≤ 37386 | ✔️ | ✔️ | ✔️ | ✘ |
| Method 3 Source 2 22981 ≤ S ≤ 27037 | ✔️ | ✔️ | ✔️ | ✘ |
| Method 4 Source 2 9666 ≤ S ≤ 11027 | ✘ | ✘ | ✘ | ✔️ |
Conclusion: The range derived by Method 4 (effective sentences) is significantly lower than that of the other ranges and conflicts with all of them, while Methods 1-3 are consistent with one another. A closer look at the breakdown of effective sentences (into terms of 5 years or longer and terms shorter than 5 years) suggests an error in the number of shorter terms, since this number is drastically smaller than in previous years (despite the number of longer terms remaining approximately the same).
Numbers of individuals with short (less than 5 years) and long (5 years and longer) effective sentences for the years 2007-2017, as reported in the Xinjiang Yearbook. The fall in short sentences in 2009-2010, despite no changes in the number of long sentences or other similar proxies/statistics, suggests reporting error in this statistic.
For these reasons, we exclude Method 4 and retain the ranges of Methods 1-3, resulting in the largest consistent range of:
22981 ≤ people sentenced in 2009 ≤ 27037
· 2010 ·
Source 1: 2011 court work report
Method 1: The work report gives 17097 as the number of first-instance criminal cases concluded in 2011, noting that this is a 2.02% increase from the previous year, implying approximately 17097/1.0202 = 16758 first-instance criminal cases concluded in 2010. By Method 1, this gives a range of 16758 to 33516 individuals sentenced.
Method 5: The report also notes that 11433 people were sentenced for crimes related to the infringement of others’ rights and property rights in 2011, and that this was 1.39% lower than in the previous year, implying that around 11433/0.9861 = 11594 people were sentenced for these crimes in 2010. As this is only one category of crimes, it implies a lower bound of at least 11594 individuals sentenced.
Source 2: 2014 court work report
Method 2: The work report mentions that the total number of criminal cases concluded in 2010 was 19785, which by Method 2 gives a range of 11871 to 35613 individuals sentenced.
Source 3: 2011 Xinjiang Yearbook
Method 1: The yearbook reports a total of 16391 first-instance criminal cases concluded, which by Method 1 gives a range of 16391 to 32782 individuals sentenced.
Method 3: The yearbook reports a total of 16077 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 20677 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 21698 to 25528 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 9963 individuals, which by Method 4, coupled with the upper bound of 27037 for the previous year, gives a range of 8949 to 10352 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2010. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency.
| Method 1 Source 1 16758 ≤ S ≤ 33516 | Method 1 Source 3 16391 ≤ S ≤ 32782 | Method 2 Source 2 11871 ≤ S ≤ 35613 | Method 3 Source 3 21698 ≤ S ≤ 25528 | Method 4 Source 3 8949 ≤ S ≤ 10352 | Method 5 Source 1 11594 ≤ S | |
| Method 1 Source 1 16758 ≤ S ≤ 33516 | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 1 Source 3 16391 ≤ S ≤ 32782 | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 2 Source 2 11871 ≤ S ≤ 35613 | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 3 Source 3 21698 ≤ S ≤ 25528 | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 4 Source 3 8949 ≤ S ≤ 10352 | ✘ | ✘ | ✘ | ✘ | ✔️ | ✘ |
| Method 5 Source 1 11594 ≤ S | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
Conclusion: All of the method-source pairs except for Method 4 are consistent with one another, while Method 4 provides a range that is much lower than the others. As with the previous year, it is likely that there is an error in the number of short sentences reported (see the relevant figure in the conclusion for the previous year). For this reason, we exclude Method 4 and take the largest consistent range as:
21698 ≤ people sentenced in 2010 ≤ 25528
· 2011 ·
Source 1: 2011 court work report
Method 1: According to the work report, a total of 17097 first-instance criminal cases were concluded, which by Method 1 gives 17097 to 34194 individuals sentenced.
Method 2: The work report notes that a total of 20118 criminal cases were concluded, which by Method 2 gives 12070 to 36213 individuals sentenced.
Method 5: The report notes that (a) 11433 people were sentenced for crimes related to the infringement of others’ rights and property rights, (b) 482 were sentenced for bribery, corruption, and dereliction of duty, and (c) 1624 were sentenced for drug-related crimes. Summing these implies at least 13539 individuals sentenced.
Source 2: 2012 court work report
Method 0: The 2012 work report gives the number of people sentenced in 2012 as 26062, and notes that this is a 5.44% increase from the previous year, which implies 26062/1.0544 = 24717 individuals sentenced.
Method 1: The same report gives the number of first-instance criminal cases concluded in 2012 as 18708, and notes that this is an 8.61% increase from the previous year, implying approximately 18708/1.0861 = 17225 first-instance criminal cases concluded in 2011. By Method 1, this implies 17225 to 34450 individuals sentenced.
Method 2: The report also gives the total number of criminal cases concluded in 2012 as 22106, and notes that this is an 8.99% increase from the year before. This implies a total of 22106/1.0899 = 20283 total cases concluded in 2011, which by Method 2 gives 12169 to 36510 individuals sentenced.
Source 3: 2012 Xinjiang Yearbook
Method 1: Identical to Method 1 for Source 1 (above).
Method 3: The yearbook reports a total of 16830 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 20738 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 21430 to 25212 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 15601 individuals, which by Method 4, coupled with the upper bound of 25528 for the previous year, gives a range of 14643 to 16209 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2011. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double marks are used when the same statistic is reported in two different sources.
| Method 0 Source 2 S = 24717 | Method 1 Sources 1/3 17097 ≤ S ≤ 34194 | Method 1 Source 2 17225 ≤ S ≤ 34450 | Method 2 Source 1 12070 ≤ S ≤ 36213 | Method 2 Source 2 12169 ≤ S ≤ 36510 | Method 3 Source 3 21430 ≤ S ≤ 25212 | Method 4 Source 3 14643 ≤ S ≤ 16209 | Method 5 Source 1 13539 ≤ S | |
| Method 0 Source 2 S = 24717 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 1 Sources 1/3 17097 ≤ S ≤ 34194 | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✘✘ | ✔️✔️ |
| Method 1 Source 2 17225 ≤ S ≤ 34450 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 2 Source 1 12070 ≤ S ≤ 36213 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
| Method 2 Source 2 12169 ≤ S ≤ 36510 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
| Method 3 Source 3 21430 ≤ S ≤ 25212 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 4 Source 3 14643 ≤ S ≤ 16209 | ✘ | ✘✘ | ✘ | ✔️ | ✔️ | ✘ | ✔️ | ✔️ |
| Method 5 Source 1 13539 ≤ S | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
Conclusion: As in previous years, the only inconsistencies are due to Method 4, which gives lower estimates than those of the other methods/proxies. Excluding Method 4 resolves all inconsistencies, and is done here. For this year, the exact number of people sentenced is reported, and is consistent with the remaining methods, leading us to state:
people sentenced in 2011 = 24717
· 2012 ·
Source 1: 2013 China Law Yearbook
Method 0: The yearbook reports the exact number as 26062 individuals sentenced.
Method 1: The yearbook also reports the number of first-instance criminal cases concluded in 2012 as 18708, which by Method 1 gives 18708 to 37416 individuals sentenced.
Method 2: The yearbook also gives the number of total criminal cases concluded as 22106, which by Method 2 gives 13263 to 39791 individuals sentenced.
Source 2: 2012 court work report
Methods 0/1/2: The numbers given are identical to those in Source 1.
Source 3: 2013 Xinjiang Yearbook
Method 1: Identical to Method 1 for Sources 1/2 (above).
Method 3: The yearbook reports a total of 18288 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 22751 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 22196 to 26113 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 16897 individuals, which by Method 4, coupled with the upper bound of 24717 for the previous year, gives a range of 15970 to 17556 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2012. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double or triple marks are used when the same statistic is reported in two or three different sources, respectively.
| Method 0 Sources 1/2 S = 26062 | Method 1 Sources 1/2/3 18708 ≤ S ≤ 37416 | Method 2 Sources 1/2 13263 ≤ S ≤ 39791 | Method 3 Source 3 22196 ≤ S ≤ 26113 | Method 4 Source 3 15970 ≤ S ≤ 17556 | |
| Method 0 Sources 1/2 S = 26062 | ✔️✔️ | ✔️✔️✔️ | ✔️✔️ | ✔️✔️ | ✘✘ |
| Method 1 Sources 1/2/3 18708 ≤ S ≤ 37416 | ✔️✔️✔️ | ✔️✔️✔️ | ✔️✔️✔️ | ✔️✔️✔️ | ✘✘✘ |
| Method 2 Sources 1/2 13263 ≤ S ≤ 39791 | ✔️✔️ | ✔️✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ |
| Method 3 Source 3 22196 ≤ S ≤ 26113 | ✔️✔️ | ✔️✔️✔️ | ✔️✔️ | ✔️ | ✘ |
| Method 4 Source 3 15970 ≤ S ≤ 17556 | ✘✘ | ✘✘✘ | ✔️✔️ | ✘ | ✔️ |
Conclusion: As in previous years, Method 4 gives a range that is much lower than the others, and is inconsistent with six of the other sources. For this reason, we exclude Method 4 as the only source of inconsistency. The exact number of people sentenced is reported for this year, and is consistent with the remaining methods, leading us to state:
people sentenced in 2012 = 26062
· 2013 ·
Source 1: 2013 court work report
Method 0: The work report gives the total as 15830 individuals sentenced.
Method 2: The report notes 21061 total criminal cases concluded in 2013, which by Method 2 gives 12636 to 37910 individuals sentenced.
Method 5: The report also notes that 4567 people were sentenced in murder, bombing, and rape cases, 9398 in robbery, theft, and snatching cases, 2515 in drunk driving or other transportation-related cases, 1715 in drug-related cases, 341 in counterfeit and fraud cases, 48 in cases related to food and drug safety, 479 in bribery and corruption cases, and 61 in dereliction of duty cases. This implies at least 19124 individuals sentenced.
Source 2: 2014 court work report
Method 2: The number reported is identical to that of Source 1 (above).
Source 3: 2014 Xinjiang Yearbook
Method 1: The yearbook reports 18233 first-instance criminal cases being concluded, which by Method 1 gives 18233 to 36466 individuals sentenced.
Method 3: The yearbook reports a total of 14993 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 21853 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 19456 to 22890 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 16438 individuals, which by Method 4, coupled with the upper bound of 26062 for the previous year, gives a range of 15460 to 17079 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2013. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double marks are used when the same statistic is reported in two different sources.
| Method 0 Source 1 S = 15830 | Method 1 Source 3 18233 ≤ S ≤ 36466 | Method 2 Sources 1/2 12636 ≤ S ≤ 37910 | Method 3 Source 3 19456 ≤ S ≤ 22890 | Method 4 Source 3 15460 ≤ S ≤ 17079 | Method 5 Source 1 19124 ≤ S | |
| Method 0 Source 1 S = 15830 | ✔️ | ✘ | ✔️✔️ | ✘ | ✔️ | ✘ |
| Method 1 Source 3 18233 ≤ S ≤ 36466 | ✘ | ✔️ | ✔️✔️ | ✔️ | ✘ | ✔️ |
| Method 2 Sources 1/2 12636 ≤ S ≤ 37910 | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ |
| Method 3 Source 3 19456 ≤ S ≤ 22890 | ✘ | ✔️ | ✔️✔️ | ✔️ | ✘ | ✔️ |
| Method 4 Source 3 15460 ≤ S ≤ 17079 | ✔️ | ✘ | ✔️✔️ | ✘ | ✔️ | ✘ |
| Method 5 Source 1 19124 ≤ S | ✘ | ✔️ | ✔️✔️ | ✔️ | ✘ | ✔️ |
Conclusion: Method 2 gives a very wide range and does not conflict with any of the other methods, prompting us to remove it from consideration as it adds nothing of value when analyzing for inconsistencies. Among the remaining methods, the conflict is between Methods 0/4 and Methods 1/3/5, with Methods 0/4 giving much smaller numbers. A closer inspection suggests a similar pathology in the Method 0 and Method 4 sources, as the 15830 from Method 0 is actually the number that Method 4 gives as those with effective sentences minus those spared from criminal punishment that year (16438 – 608). Given this apparent dependency between the two method-source pairs, given Method 4 often being in conflict for other years, and giving preference to the majority (3 sources vs. 1 or 2), we choose to exclude the numbers reported in Methods 0 and 4, resulting in the largest consistent range of:
19456 ≤ people sentenced in 2013 ≤ 22890
· 2014 ·
Source 1: 2016 China Law Yearbook
Method 2: The yearbook reports a total of 39979 criminal cases being concluded in 2015, and notes that this is a 35.5% increase from the previous year, implying 39979/1.355 = 29505 criminal cases concluded in 2014. By Method 2, this gives 17703 to 53109 individuals sentenced.
Source 2: 2014 court work report
Method 2: The court work report notes a total of 29511 criminal cases concluded, which by Method 2 gives 17706 to 53120 individuals sentenced.
Method 5: The work report also notes that 6150 people were sentenced in cases related to murder, assault, robbery, rape, kidnapping, trafficking of women and children, drug crimes, and picking quarrels and provoking trouble, 963 were sentenced in internet-related crimes, 267 were sentenced in cases related to food and drug safety, agricultural product quality safety, and production safety, and 624 were sentenced in cases related to bribery, corruption, or dereliction of duty. This implies at least 8004 individuals sentenced.
Source 3: 2015 court work report
Method 2: The numbers reported are identical to those of Source 1 (above).
Source 4: 2015 Xinjiang Yearbook
Method 1: The yearbook reports 25714 first-instance criminal cases being concluded, which by Method 1 gives 25714 to 51428 individuals sentenced.
Method 3: The yearbook reports a total of 20829 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 32518 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 27689 to 32576 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 21562 individuals, which by Method 4, coupled with the upper bound of 22890 for the previous year, gives a range of 20703 to 22403 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2014. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double marks are used when the same statistic is reported in two different sources.
| Method 1 Source 4 25714 ≤ S ≤ 51428 | Method 2 Sources 1/3 17703 ≤ S ≤ 53109 | Method 2 Source 2 17706 ≤ S ≤ 53120 | Method 3 Source 4 27689 ≤ S ≤ 32576 | Method 4 Source 4 20703 ≤ S ≤ 22403 | Method 5 Source 2 8004 ≤ S | |
| Method 1 Source 4 25714 ≤ S ≤ 51428 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 2 Sources 1/3 17703 ≤ S ≤ 53109 | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ |
| Method 2 Source 2 17706 ≤ S ≤ 53120 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
| Method 3 Source 4 27689 ≤ S ≤ 32576 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 4 Source 4 20703 ≤ S ≤ 22403 | ✘ | ✔️✔️ | ✔️ | ✘ | ✔️ | ✔️ |
| Method 5 Source 2 8004 ≤ S | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
Conclusion: As in previous years, Method 4 yields a significantly lower range than the others, and is the only source of inconsistency. Additionally, the number of sentences of 5 years or longer going into effect in 2014 is given as 5093 in this source (see the chart in the Conclusion section of 2009), and is notably lower than the number of individuals sentenced to 5 or more years in 2014 as documented by the Xinjiang Victims Database, which is currently at 6472. As this is only the number documented and is unlikely to be comprehensive, the fact that it exceeds the number in the source further suggests underreporting and/or error in the latter.
Consequently, we exclude the range of Method 4, resulting in a largest consistent range of:
27689 ≤ people sentenced in 2014 ≤ 32576
· 2015 ·
Source 1: 2016 China Law Yearbook
Method 2: The yearbook reports 39979 criminal cases being concluded, which by Method 2 yields 23987 to 71963 individuals sentenced.
Source 2: 2015 court work report
Method 2: The numbers provided are identical to those in Source 1 (above).
Method 5: The work report notes that 10193 people were sentenced in cases related to murder, assault, robbery, rape, trafficking of women and children, drugs, and picking quarrels and provoking trouble, 2344 were sentenced in internet-related cases, 182 were sentenced in cases related to food and drug safety, agricultural product quality safety, and production safety, and 690 were sentenced in bribery, corruption, and dereliction of duty cases. This implies at least 13409 individuals sentenced.
Source 3: 2016 court work report
Method 2: The work report gives 33445 as the number of total criminal cases concluded in 2016, adding that this is a 16.3% decrease from the previous year, which implies 33445/0.837 = 39958 cases in 2015. By Method 2, this gives a range of 23974 to 71925 individuals sentenced.
Source 4: 2016 Xinjiang Yearbook
Method 1: The yearbook reports 35170 first-instance criminal cases being concluded, which by Method 1 gives 35170 to 70340 individuals sentenced.
Method 3: The yearbook reports a total of 29907 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 49158 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 41596 to 48937 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 35340 individuals, which by Method 4, coupled with the upper bound of 32576 for the previous year, gives a range of 34118 to 36717 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2015. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double marks are used when the same statistic is reported in two different sources.
| Method 1 Source 4 35170 ≤ S ≤ 70340 | Method 2 Sources 1/2 23987 ≤ S ≤ 71963 | Method 2 Source 3 23974 ≤ S ≤ 71925 | Method 3 Source 4 41596 ≤ S ≤ 48937 | Method 4 Source 4 34118 ≤ S ≤ 36717 | Method 5 Source 2 13409 ≤ S | |
| Method 1 Source 4 35170 ≤ S ≤ 70340 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
| Method 2 Sources 1/2 23987 ≤ S ≤ 71963 | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ |
| Method 2 Source 3 23974 ≤ S ≤ 71925 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
| Method 3 Source 4 41596 ≤ S ≤ 48937 | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 4 Source 4 34118 ≤ S ≤ 36717 | ✔️ | ✔️✔️ | ✔️ | ✘ | ✔️ | ✔️ |
| Method 5 Source 2 13409 ≤ S | ✔️ | ✔️✔️ | ✔️ | ✔️ | ✔️ | ✔️ |
Conclusion: As in most other years, the range given by Method 4 is lower than those given by others, with an explicit conflict taking place with Method 3. Given how the entirety of the range of Method 3 is included in those of the other methods, while only a portion of the range of Method 4 is, the former is given priority, Method 4 is excluded, and the largest consistent range that results is:
41596 ≤ people sentenced in 2015 ≤ 48937
· 2016 ·
Source 1: 2017 China Law Yearbook
Method 2: The yearbook reports the total number of criminal cases concluded as 33445, which by Method 2 gives 20067 to 60201 individuals sentenced.
Source 2: 2016 court work report
Method 2: The work report gives identical numbers to those in Source 1 (above).
Section 3: 2017 Xinjiang Yearbook
Method 1: The yearbook reports the number of first-instance criminal cases concluded as 29450, which by Method 1 gives 29450 to 58900 individuals sentenced.
Method 3: The yearbook reports a total of 20697 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 41406 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 39948 to 46998 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 31094 individuals, which by Method 4, coupled with the upper bound of 48937 for the previous year, gives a range of 29258 to 32306 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2016. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency. Double marks are used when the same statistic is reported in two different sources.
| Method 1 Source 3 29450 ≤ S ≤ 58900 | Method 2 Sources 1/2 20067 ≤ S ≤ 60201 | Method 3 Source 3 39948 ≤ S ≤ 46998 | Method 4 Source 3 29258 ≤ S ≤ 32306 | |
| Method 1 Source 3 29450 ≤ S ≤ 58900 | ✔️ | ✔️✔️ | ✔️ | ✔️ |
| Method 2 Sources 1/2 20067 ≤ S ≤ 60201 | ✔️✔️ | ✔️✔️ | ✔️✔️ | ✔️✔️ |
| Method 3 Source 3 39948 ≤ S ≤ 46998 | ✔️ | ✔️✔️ | ✔️ | ✘ |
| Method 4 Source 3 29258 ≤ S ≤ 32306 | ✔️ | ✔️✔️ | ✘ | ✔️ |
Conclusion: As in most other years, the range given by Method 4 is lower than those given by others, with an explicit conflict taking place with Method 3. Given how the entirety of the range of Method 3 is included in those of the other methods, while only a portion of the range of Method 4 is, the former is given priority, Method 4 is excluded, and the largest consistent range that results is:
39948 ≤ people sentenced in 2016 ≤ 46998
· 2017 ·
Source 1: 2018 China Law Yearbook
Method 2: The yearbook reports a total of 147272 criminal cases concluded, which by Method 2 gives 88363 to 265090 individuals sentenced.
Source 2: 2017 court work report
Method 2: The work report gives the total number of criminal cases concluded for the 5-year period of 2013 to 2017 as 297000. Noting the numbers given in previous work reports (33445 in 2016, 39979 in 2015, 29511 in 2014, 21061 in 2013) implies an approximate total of 297000 – 33445 – 39979 – 29511 – 21061 = 173004 criminal cases concluded in 2017. By Method 2, this gives 103802 to 311408 individuals sentenced.
Section 3: 2018 Xinjiang Yearbook
Method 1: The yearbook reports the number of first-instance criminal cases concluded as 127861, which by Method 1 gives 127861 to 255722 individuals sentenced.
Method 3: The yearbook reports a total of 53886 individuals sentenced following indictments from procuratorates that year (presumably excluding Bingtuan), for 216597 indicted individuals, which by Method 3, coupled with the statistics of the year prior, gives a range of 74595 to 87759 individuals sentenced.
Method 4: The yearbook reports sentences going into effect for 99787 individuals, which by Method 4, coupled with the upper bound of 46998 for the previous year, gives a range of 98024 to 103675 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2017. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency.
| Method 1 Source 3 127861 ≤ S ≤ 255722 | Method 2 Source 1 88363 ≤ S ≤ 265090 | Method 2 Source 2 103802 ≤ S ≤ 311408 | Method 3 Source 3 74595 ≤ S ≤ 87759 | Method 4 Source 3 98024 ≤ S ≤ 103675 | |
| Method 1 Source 3 127861 ≤ S ≤ 255722 | ✔️ | ✔️ | ✔️ | ✘ | ✘ |
| Method 2 Source 1 88363 ≤ S ≤ 265090 | ✔️ | ✔️ | ✔️ | ✘ | ✔️ |
| Method 2 Source 2 103802 ≤ S ≤ 311408 | ✔️ | ✔️ | ✔️ | ✘ | ✘ |
| Method 3 Source 3 74595 ≤ S ≤ 87759 | ✘ | ✘ | ✘ | ✔️ | ✘ |
| Method 4 Source 3 98024 ≤ S ≤ 103675 | ✘ | ✔️ | ✘ | ✘ | ✔️ |
Conclusion: For this year, both Methods 3 and 4 give ranges that are significantly lower than, and generally inconsistent with, those given by the other methods (which are based on case numbers). Methods 1 and 2 are given preference, with 3 and 4 excluded, both because of the majority principle (3 sources vs. 2) and because documented victim data is also more consistent with the larger ranges given by Methods 1 and 2. Namely, over 10000 Uyghurs have been reported as sentenced in 2017 in Konasheher County alone, or over 3.58% of the corresponding Uyghur population of Konasheher (278745). Even if representative of Kashgar Prefecture only (with an Uyghur population of 4295195, leaving out such large Uyghur-majority prefectures as Hotan and Aksu), such a rate would imply an approximately 150000 people sentenced in 2017 in Kashgar Prefecture alone, further prompting us to reject the lower ranges of Methods 3 and 4 for this year.
Taking the largest consistent range for the remaining methods then yields:
127861 ≤ people sentenced in 2017 ≤ 255722
· 2018 ·
Source 1: 2018 court work report
Method 0: The work report gives the total as 133198 individuals sentenced.
Method 2: The report also gives the total number of concluded criminal cases as 74348, which by Method 2 yields 44608 to 133827 individuals sentenced.
Source 2: 2019 Xinjiang Yearbook
Method 3: The yearbook reports a total of 121354 individuals whose sentences went into effect following indictments from procuratorates that year (presumably excluding Bingtuan), for 135546 indicted individuals. Unlike in previous years, the number given is sentences gone into effect, and should in principle exclude those sentenced whose appeals have not been concluded. Consequently, the 121354 is a lower bound on those sentenced. Method 3, coupled with the statistics of the year prior, can be applied on the lower end to obtain at least 284065 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2018. A ✔️ indicates that the given method-source pair is consistent, while ✘ marks an inconsistency.
| Method 0 Source 1 S = 133198 | Method 2 Source 1 44608 ≤ S ≤ 133827 | Method 3 Source 2 284065 ≤ S | |
| Method 0 Source 1 S = 133198 | ✔️ | ✔️ | ✘ |
| Method 2 Source 1 44608 ≤ S ≤ 133827 | ✔️ | ✔️ | ✘ |
| Method 3 Source 2 284065 ≤ S | ✘ | ✘ | ✔️ |
Conclusion: Method 3 (Source 2) conflicts with both the exact number given in Source 1 and the range inferred from the number of cases given in Source 1, and is significantly higher. This is essentially because of the number of individuals who were indicted but not sentenced in 2017 being reported as extremely high in the yearbook the previous year, which is very possibly an error and is also a reason for why Method 3 was excluded in the analysis for 2017. Excluding it again here, we take the number of Method 0 and state:
people sentenced in 2018 = 133198
· 2021 ·
Source (only): 2022 Xinjiang Yearbook
Method 0: The yearbook gives the (likely rounded) total as 44600 individuals sentenced.
Method 1: The yearbook reports the (likely rounded) number of first-instance criminal cases concluded as 30300, which by Method 1 gives 30300 to 60600 individuals sentenced.
Consistency chart comparing the sentence ranges as derived from the various methods and sources for 2021. A ✔️ indicates that the given method-source pair is consistent.
| Method 0 S = 44600 | Method 1 30300 ≤ S ≤ 60600 | |
| Method 0 S = 44600 | ✔️ | ✔️ |
| Method 1 30300 ≤ S ≤ 60600 | ✔️ | ✔️ |
Conclusion: There are no inconsistencies, with Method 0 giving the (likely rounded) precise number as:
people sentenced in 2021 = 44600
General consistency review and summary
Prior to consolidating the estimates in a single table or graph, we can do a general review of the consistency comparisons to note that the estimates provided by Method 4 (individuals whose sentences went into effect) are almost always significantly lower than those given by other methods, at times conflict with case-by-case victim documentation, and are frequently the main/only source of inconsistency. The reason for this is currently unclear, and may stem from nuances that we are unaware of, either (1) in the precise definition of “effective sentences”, which to the best of our knowledge refers to sentences with appeal procedures concluded, or (2) in potential reporting errors or limitations in this statistic. The fact that the estimates provided are consistently lower from year to year suggests something systematic, and not sporadic, but is nevertheless considered as an unreliable indicator of the actual sentence numbers, given the other statistics being greater in number and providing consistently higher estimates.
For this reason, we remove Method 4 as a valid proxy altogether, and revise the corresponding ranges on the sentence numbers.
Lower and upper bounds on the individuals sentenced in criminal cases in Xinjiang between 2007 and 2018, and 2021, following the consolidation of various sources/statistics and the exclusion of inconsistencies.
| year | number sentenced (lower) | number sentenced (upper) |
|---|---|---|
| 2007 | 17698 | 35396 |
| 2008 | 21459 | 25246 |
| 2009 | 22981 | 27037 |
| 2010 | 21698 | 25528 |
| 2011 | 24717 | |
| 2012 | 26062 | |
| 2013 | 19456 | 22890 |
| 2014 | 27689 | 32576 |
| 2015 | 41596 | 48937 |
| 2016 | 39948 | 46998 |
| 2017 | 127861 | 255722 |
| 2018 | 133198 | |
| 2021 | 44600 | |
| total | 568963 | 748907 |
We can also present this data as a plot.
Lower/upper bounds on individuals sentenced in criminal cases in Xinjiang between 2007 and 2018, and 2021, following consolidation of various sources/statistics.
Estimating sentences for 2019, 2020, and 2022
Because sentencing numbers were not published in 2019-2020 (with reports for 2022 still not available), it has been common to use procuratorate statistics (on indictment numbers) as a proxy to estimate the numbers of individuals sentenced since then. This approach is natural and valid, seeing how an indictment is a prerequisite for a sentence and seeing how the two statistics (indictments and sentences) have generally been well correlated in the past, with the same rises and falls from year to year typically noted for both.
Unlike with the sentence statistics, where some significant inconsistencies have been noted, the reported indictment numbers have generally been more stable and reliable, with available sources – procuratorate reports and yearbooks – usually giving figures that either match exactly or are very close. As suggested by the Xinjiang Yearbook, these do not include prosecutions from the Bingtuan, but are likely still a good proxy since prosecution numbers from Bingtuan units are expected to be small in comparison and most likely relate to the non-Bingtuan number in a linear manner (approximately being a fixed fraction of the non-Bingtuan numbers – see Appendix A, Method 3).
Indictment data for 2007 to 2022, as given in yearbooks and work reports.
To link the indictment numbers to sentencing ones, we require a model that establishes a relationship between the two. Given how most documented sentences have taken place within a year of indictment (see Appendix C), a reasonable model would be the following:
S_Y = c_1 (P_{Y-1} + P^{BT}_{Y-1}) + c_2 (P_{Y} + P^{BT}_{Y})where P_{Y-1} and P_{Y} are the number of people prosecuted in the years Y-1 and Y, respectively, P^{BT}_{Y-1} and P^{BT}_{Y} are their Bingtuan analogues, and c_1, c_2 \in (0,1) are model parameters.
This model basically states that the number of sentences in a given year is the sum of a portion of the prosecutions from the year in question and a portion of the prosecutions from the year prior. Here, c_1 and c_2 represent the speed or efficiency of the sentencing process, with higher c_2 and lower c_1 corresponding to faster processes where more indictments result in sentences in the same year. To add more realism to the model, we can additionally restrict c_1 and c_2 as:
and
\frac{c_1}{c_1 + c_2} \geq 0.1which ensure that:
- essentially all of the prosecutions result in sentences, with the odds of a court rejecting an indictment assumed to be too low to be significant;
- at least 10% of the sentences in year
Yare the result of the prosecutions in yearY-1, which serves to accommodate the time lag observed in the available data (see the statistics of Method 3 in Appendix A and the statistics in Appendix C).
The Bingtuan statistics, P^{BT}, are generally not available. However, given how the relative sizes of the Bingtuan and non-Bingtuan judicial systems have not changed significantly over the years, with the former notably smaller, we can assume the Bingtuan indictments to be a certain fraction of the non-Bingtuan ones:
with \gamma believed to generally remain in the 0 to 0.15 range (Appendix A, Method 3 includes a review of the available data).
Consequently, we can rewrite the model as
S_Y = c_1 (1+\gamma) P_{Y-1} + c_2 (1+\gamma) P_{Y}or, since we assume \gamma to be approximately constant, as
where a = c_1 (1+\gamma) and b = c_2 (1+\gamma), with 1 \leq a + b \leq 1.15 and a \geq 0.1(a+b).
In many applications, empirical models are fit using a least-squares approach, where the model parameters are chosen to give a model that minimizes the squares of deviations from the observed data points. In our case, this standard approach is not applicable since the observed data are not given as points (exact values) but as intervals.
To handle this, we can use a generalization of the least-squares method and fit a model by applying interval regression. Unlike with standard least-squares, which can usually be solved by a number of tools and methods (e.g., Excel), interval regression requires us to formulate our fitting task as an optimization problem, where we minimize the sum of the squares of deviations from the data, while simultaneously allowing the data to take any values that remain within our intervals (\underline S, \overline S, as determined in the previous section):
where we use s_y to represent the possible numbers of individuals sentenced and force them to stay within the prescribed bounds. If we were to know the exact sentencing numbers, we would simply have \underline S_y = s_y = \overline S_y, which would effectively remove these as variables and return us to the standard least-squares formulation. The additional realism constraints on a,b are also included.
In optimization terms, this problem is a linearly constrained quadratic programming (LCQP) problem, with a convex objective, which makes it easy to solve via a variety of available solvers. The optimum obtained is a^* = 0.1, b^* = 0.9, which appears to be sufficient and captures the observed sentencing trends and magnitudes, as seen in the plot below:
We can also use this model to obtain a lower estimate on the individuals sentenced in 2022, as 44603*0.1+37819*0.9 = 38497.
Looking at the cumulative numbers over several key periods, the best-fit model estimates:
- at least 579767 people sentenced since 2017,
- at least 703018 people sentenced since 2014,
- at least 834517 people sentenced since 2008,
which excludes the numbers of those sentenced in 2023 and after.
Appendix A: Proxies/methods for estimating sentence numbers
In most cases, official sources don’t report the exact number of individuals sentenced in Xinjiang for a specific year. However, it is still possible to compute lower and upper bounds on the exact number by using other statistics, and these are reviewed here.
Method 0: Exact number
This is the ideal situation in which the exact number is reported. Some work reports provide this number, but most sources don’t.
Method 1: First-instance cases
Work reports and yearbooks frequently give the number of first-instance criminal cases (刑事一审案件) concluded in a given year, denoted here by C_{1,Y} (after the first instance, the defendant may appeal, leading to a second-instance, and typically final, hearing). It follows from basic logic that the number of people involved in first-instance cases should be equal to or greater than the number of first-instance cases. Given how the conviction rate in China is nearly 100%, it follows that almost all of the people involved end up sentenced, allowing us to state:
Empirically, court statistics in China almost always report case-to-person counts that have a ratio of between 1:1 and 1:2, making it very likely that the number of people sentenced is smaller than double the number of first-instance cases:
S_Y \leq 2 C_{1,Y}
This yields the range:
C_{1,Y} \leq S_Y \leq 2 C_{1,Y}Method 2: Total cases
Sometimes, first-instance case numbers are not reported, with only the total number of cases given (denoted here by C_{T,Y}). To get bounds on sentences from the latter, we can use the limited reported statistics available to estimate the number of first-instance concluded criminal cases (as a fraction of total concluded criminal cases) and then apply Method 1.
Numbers of first-instance criminal cases concluded vs. total criminal cases concluded on the national level, for 2017-2023.
| year | first-instance | total | portion | source |
|---|---|---|---|---|
| 2017 | 1296650 | 1852037 | 70.01% | 2017 SPC report |
| 2018 | 1198383 | 1825835 | 65.63% | 2018 SPC report |
| 2019 | 1297191 | 1975511 | 65.66% | 2019 SPC report |
| 2020 | 1115890 | 1692680 | 65.92% | 2020 SPC report |
| 2021 | 1255671 | 1789952 | 70.15% | 2021 SPC report |
| 2022 | 1038523 | 1513890 | 68.60% | 2022 SPC report |
| 2023 | 1243255 | 1763199 | 70.51% | 2023 SPC report |
From the national-level statistics, we see that first-instance cases make up approximately between 65% and 70% of the total concluded criminal cases each year. To account for potential variations owing to locality (Xinjiang vs. national) and time (older years may have slightly different statistics or reporting standards), we can also consult specific work reports for Xinjiang where both first-instance and total numbers are given.
Numbers of first-instance criminal cases concluded vs. total criminal cases concluded in Xinjiang, for 2011 and 2012.
| year | first-instance | total | portion | source |
|---|---|---|---|---|
| 2011 | 17097 | 20118 | 84.98% | 2011 court work report |
| 2012 | 18708 | 22106 | 84.63% | 2012 court work report |
These numbers suggest a higher portion than the one nationally reported (of around 85%).
Adding an additional safety buffer of around 5% to both sides, we propose the conservative bounds:
0.6C_{T,Y} \leq C_{1,Y} \leq 0.9C_{T,Y}Method 1 can now be applied to obtain:
0.6C_{T,Y} \leq S_Y \leq 1.8C_{T,Y}Method 3: Individuals sentenced following procuratorate indictments
The procuratorate statistics section of the Xinjiang Yearbook provides not only the number of individuals indicted in a given year but also the number of those who have already been tried and sentenced. These numbers do not include Bingtuan procuratorates (unlike the court statistics, which include Bingtuan courts), making them a lower bound on total sentences. Furthermore, from the wording and the context, the number reported appears to only refer to those individuals who were indicted that year, and would not include those people for whom indictments had been filed the year before (or earlier).
To establish the relationship between total sentences and the ones reported in the procuratorate section, let us first break down the sentences for a given year as:
S_Y = P_{Y,Y} + P_{Y-1,Y} + S^{BT}_{Y}where P_{Y,Y} is the number of individuals prosecuted in the year Y and sentenced in the year Y (excluding Bingtuan), P_{Y-1,Y} is the number of individuals prosecuted in the year Y-1 and sentenced in the year Y (excluding Bingtuan), and S^{BT}_Y are the sentences given through the Bingtuan system. A general assumption that we make is that the time between prosecution and sentence is almost never more than a year (this is supported by empirical data, provided in Appendix C), which allows us to ignore P_{Y-2,Y} since we expect this quantity to be too small to be significant.
Sentencing statistics as given in the procuratorate section of the Xinjiang Yearbook, for 2007-2018. The last column is an estimate that assumes that the remainder not sentenced in a given year is sentenced in the year that follows. Asterisks (*) denote that the reported number likely includes not-guilty verdicts, though this is assumed to be too low to be significant (based on non-guilty verdict numbers from later years, it is unlikely that the quantity exceeds 10 or 20). The portion sentenced as reported for 2017 is considered to be questionable and potentially erroneous (marked by an exclamation point), as its relative value to the total indictments is very low in comparison to other years. The portion sentenced as reported for 2018 is marked with (E) to indicate that this is the number of individuals whose sentences went into effect, and in principle should exclude those who were sentenced but appealed, with appeals not yet resolved by the end of the year (believed to be less than 3-4% of the cases).
| year | individuals indicted (excluding Bingtuan) | portion sentenced | remainder | P_{Y,Y}+P_{Y-1,Y} (estimate) |
|---|---|---|---|---|
| 2007 | 22153 | 17542* | 4611 | > 17542 |
| 2008 | 22763 | 16848* | 5915 | 21459 |
| 2009 | 22687 | 17066* | 5621 | 22981 |
| 2010 | 20677 | 16077 | 4600 | 21698 |
| 2011 | 20738 | 16830 | 3908 | 21430 |
| 2012 | 22751 | 18288 | 4463 | 22196 |
| 2013 | 21853 | 14993 | 6860 | 19456 |
| 2014 | 32518 | 20829 | 11689 | 27689 |
| 2015 | 49158 | 29907 | 19251 | 41596 |
| 2016 | 41406 | 20697 | 20709 | 39948 |
| 2017 | 216597 | 53886 (!) | 162711 | 74595 |
| 2018 | 135546 | 121354 (E) | < 14192 | > 284065 |
Consulting the table above, P_{Y,Y} may be taken as the portion sentenced. In principle, P_{Y-1,Y} should be approximately equal to the remainder for the year Y-1, as almost all who were not sentenced in the year of indictment are expected to be sentenced the following year. The last column in the table provides the sum of these, using as a lower bound the portion sentenced when the remainder from the previous year is not available.
While statistics on Bingtuan prosecutions and sentences are very limited, it is still possible to gauge the approximate magnitude of the Bingtuan judicial system from those available:
- The 2015 China Law Yearbook gives the number of people prosecuted by non-Bingtuan procuratorates for 2014 as 34816, while giving the Bingtuan number as 1486 (4.27%). The same yearbook also notes that the Bingtuan number increased by 10.8% from the previous year, implying the number of individuals prosecuted by the Bingtuan procuratorates to be 1341 for 2013. A separate work report for 2013 gives the number of individuals prosecuted as 21853 (likely non-Bingtuan), suggesting a percentage of 1341/21853 = 6.14% for that year.
- The 2016 China Law Yearbook reports the number of concluded criminal cases by the Bingtuan courts in 2015 as 1460, and the total number of concluded criminal cases as 39979 (presumably including Bingtuan, but this is not stated), for an approximate Bingtuan percentage of 1460/(39979-1460) = 3.79%.
- The 2017 China Law Yearbook reports the Bingtuan procuratorates as indicting 1672 people in 2016, while 41305 were indicted by the non-Bingtuan ones (for a percentage of 4.05%). The Bingtuan courts are reported as having concluded 2173 criminal cases, while the total number is given as 33445 (for a percentage of 2173/(33445-2173) = 6.95%).
- The 2018 China Law Yearbook gives the number of people indicted by the Bingtuan procuratorates in 2017 as 1964, while the number for the non-Bingtuan ones as 246597 (for a percentage of 0.80%). For the same year, the Bingtuan courts are reported as having concluded 2167 criminal cases, while the total number (including Bingtuan) is given as 147272, implying the non-Bingtuan number to be 145105 (for a percentage of 2167/145105 = 1.49%).
- The court work report for 2018 notes that concluded court cases for the Bingtuan courts numbered approximately 41100 that year, compared to the total number of approximately 420000 (or approximately 420000-41100 = 378900 non-Bingtuan cases), for a percentage of 41100/378900 = 10.85%. This is for all cases, however, and not just criminal ones.
- The 2019 China Law Yearbook notes that the number of individuals indicted by Bingtuan procuratorates was up by 67.5% from the year before, implying 1964*1.675 = 3290 individuals indicted in 2019. With the number of non-Bingtuan indictments being 135546, as given in the 2018 procuratorate work report, this yields a percentage of 3290/135546 = 2.43%.
- The 2020 China Law Yearbook reports the number of individuals indicted by Bingtuan procuratorates in 2019 as 3472. The 2019 procuratorate work report gives the number of individuals indicted by (presumably) non-Bingtuan procuratorates as 96596, for a percentage of 3472/96596 = 3.59%.
- Analogously, the 2022 China Law Yearbook reports the number of individuals indicted by Bingtuan procuratorates in 2021 as 3419. The 2021 procuratorate work report gives the number of individuals indicted by (presumably) non-Bingtuan procuratorates as 44603, for a percentage of 3419/44603 = 7.67%.
All of these suggest that the Bingtuan judicial system only contributes a small fraction to the total numbers, in most cases not exceeding 10%. For additional safety, we can use 15% as the upper limit, and state that:
0 \leq S^{BT}_{Y} \leq 0.15S_YThis then allows to reformulate the balance as
P_{Y,Y} + P_{Y-1,Y} \leq S_Y \leq P_{Y,Y} + P_{Y-1,Y} + 0.15S_Yor
P_{Y,Y} + P_{Y-1,Y} \leq S_Y \leq \frac{P_{Y,Y} + P_{Y-1,Y}}{0.85}Method 4: Effective sentences
The Xinjiang Yearbook regularly reports the number of criminal sentences that go into effect for a given year (生效判决). Because a sentence is considered effective after the appeal procedures have been completed, this statistic may be corrupted by a time lag, including sentences that were handed down the previous year (but only went into effect in the given year) but not including sentences that were handed down and haven’t gone into effect because they’ve been appealed.
The corresponding balances may be written as:
E_Y = S_{Y-1,Y} + S_{Y,Y} S_Y = S_{Y,Y} + S_{Y,Y+1}where E_Y is the number of people whose sentences went into effect in the year Y, S_{Y-1,Y} is the number of people sentenced in the year Y-1 with sentences going into effect in Y, S_{Y,Y} is the number of people sentenced in the year Y with sentences going into effect in the same year, and S_{Y,Y+1} is the number of people sentenced in the year Y with sentences going into effect in Y+1. The general assumption (see Appendix B) is that it almost never takes longer than a year for a sentence to go into effect, as the appeal procedures generally don’t take that long.
Substituting the second balance into the first yields:
E_Y = S_{Y-1,Y} + S_{Y} - S_{Y,Y+1}or
S_{Y} = E_Y - S_{Y-1,Y} + S_{Y,Y+1}To obtain ranges on S_{Y-1,Y} and S_{Y,Y+1}, we can reference the available appeal statistics (Appendix B), which suggest an appeal rate of under 15% and an average appeal period of no more than 3 months for the cases that are appealed. For S_{Y-1,Y}, this suggests that, of those sentenced in year Y-1, fewer than 15% would appeal, with 75% of those appealing having their cases resolved and sentences going into effect in the year Y-1 (assuming a constant frequency of criminal cases). The portion that wouldn’t have their cases resolved in the same year would then be:
where \overline S_{Y-1} is an upper bound on the individuals sentenced in the year Y-1.
An identical procedure may be done for S_{Y,Y+1}, which yields:
This finally allows us to rewrite the main balance as:
E_Y - 0.0375 \overline S_{Y-1} \leq S_{Y} \leq E_Y + 0.0375 S_{Y}where the right side may be rearranged to get:
E_Y - 0.0375 \overline S_{Y-1} \leq S_{Y} \leq \frac{E_Y}{0.9625}Method 5: Sum of specific components
Certain court reports give statistics for how many people were sentenced in specific types of cases, without necessarily giving the total number. In this case, the numbers may be added to obtain a lower bound on the total.
Appendix B: Appeal statistics
Appeal procedures can influence statistics on sentence numbers because they introduce a lag (an individual may be sentenced, but their sentence doesn’t officially go into effect until the appeal procedure is finished). Typically, people are given a single chance to appeal, leading to a second-instance case hearing, which often results in the court maintaining the original verdict. Knowing the average time frames and appeal rates allows us to compute how different reported statistics regarding different instances can relate to total sentence numbers.
Standard appeal rates
For Xinjiang, the appeal rate following a first-instance verdict is often under 15%, as may be seen in the 2017 court work report, which gives the judgment-acceptance rates for 2013-2017 (figure reproduced below).
The 2013 court work report also specifies 85% as the minimum standard judgment-acceptance rate for the base-level courts (with the standard lowered to 70% for intermediate-level courts).
While these rates apply to all cases, including non-criminal ones, we can do a general nationwide comparison of first-instance and second-instance case numbers for criminal cases specifically, relying on statistics reported in the China Law Yearbook.
| year | first-instance criminal cases concluded | second-instance criminal cases concluded | second-to-first rate |
|---|---|---|---|
| 2006 | 701379 | 94092 | 13.42% |
| 2007 | 720666 | 92364 | 12.82% |
| 2008 | 768130 | 95831 | 12.48% |
| 2009 | 766746 | 100398 | 13.09% |
| 2010 | 779641 | 102370 | 13.13% |
| 2011 | 839973 | 98919 | 11.78% |
| 2012 | 986392 | 108096 | 10.96% |
| 2013 | 953976 | 102991 | 10.80% |
| 2014 | 1023017 | 118915 | 11.62% |
| 2015 | 1099205 | 141155 | 12.84% |
| 2016 | 1115873 | 148441 | 13.30% |
| 2017 | 1296650 | 154277 | 11.90% |
These statistics show that the number of second-instance criminal cases (resulting from appeals) is typically less than 15% of the first-instance ones, corroborating the general judgment-acceptance rates and standards reported for Xinjiang above.
Consequently, if C_{1,Y} and C_{2,Y} are the numbers of first-instance and second-instance criminal cases concluded for year Y in Xinjiang, respectively, we can state the following:
as an empirical rule that we believe to hold with a very high probability.
Standard appeal procedure lengths
To estimate how long a typical appeal of a criminal case in Xinjiang takes, we consider all documented victim cases where these periods are known.
| case | time spent on appeal procedure |
|---|---|
| Abdujelil Helil | 164 days |
| Metabdulla Iminniyaz | 66 days |
| Wang Zengying | 76 days |
| Huseyinjan Jelil | 83 days |
| Jia Zhiyuan | 108 days |
| Karma Samdrub | 13 days |
| Alimjan Hemit | 140 days |
| Erkin Qurban | 17 days |
| Peng Jianbiao, Meng Xiumei, Zhong Jun, Zhong Hua, Xia Zhonglian | 253 days |
| Yang Wanjun, Ma Chengying, Zhang Chengjun | 63 days |
| Zhong Renquan, Ji Siliang, Sun Zhongming, Xiao Yuzhi, Xu Jiwen | 96 days |
| Huang Shike | 88 days |
| Nie Shigang | 123 days |
| Abdulber Omer | 28 days |
| Rusul Haliq | 59 days |
| Luo Chuanmei | 46 days |
| Wu Shengmin | 57 days |
| Wang Xiaoying | 59 days |
| Li Yulan | 75 days |
| Liang Yan | 54 days |
| Ban Guoyun | 26 days |
| Song Kaicai | 92 days |
| Wang Deqiang | 56 days |
| Zhang Haitao | 318 days |
| Average | 90 days |
This suggests that the average appeal procedure takes around 90 days, or ~25% of a year.
Assuming that cases come in at a regular frequency throughout the year, this allows us to put an upper bound on the number of first-instance cases that are appealed in the year Y and the appeals for which are resolved in the same year (with the sentence going into effect):
Appendix C: Time periods between prosecution and sentence for known cases
The data below, based on documented cases for which both indictment and sentence times are known, provide an idea of how long this procedure usually takes in Xinjiang.
| case | indictment filed | date of sentence | days between |
| Nebi Ghoja’ehmet | May 30, 2018 | June 12, 2018 | 13 |
| Nie Shigang | March 22, 2019 | April 1, 2019 | 10 |
| Jin Dehuai | December 18, 2017 | September 15, 2018 | 271 |
| Ismayil Sidiq | May 16, 2018 | May 29, 2018 | 13 |
| Adil Yaqup | August 20, 2018 | October 26, 2018 | 67 |
| Tian Weiguo | January 7, 2016 | March 24, 2016 | 77 |
| Sun Sujuan and others | July 24, 2019 | September 15, 2019 | 53 |
| Ababekri Dayim | May 16, 2018 | May 29, 2018 | 13 |
| Helchem Pazil and others | March 20, 2019 | April 2, 2019 | 13 |
| Karma Samdrub | May 21, 2010 | June 24, 2010 | 34 |
| Nurlan Pioner and Tohti Islam | June 29, 2018 | August 31, 2018 | 63 |
| Sun Wenlian | January 1, 2019 | January 28, 2019 | 27 |
| Pang Qing and others | January 14, 2019 | April 1, 2019 | 77 |
| Du Guobiao | March 31, 2014 | September 22, 2014 | 175 |
| Liu Guangfeng | May 17, 2017 | September 22, 2017 | 128 |
| Asqar Azatbek | September 17, 2018 | December 9, 2018 | 83 |
| Zhang Haitao | December 25, 2015 | January 15, 2016 | 21 |
| Zhang Qingyun | September 7, 2017 | March 19, 2018 | 193 |
| Li Famei | January 18, 2018 | June 27, 2018 | 160 |
| Ouyang Qiuping and others | July 18, 2019 | September 4, 2019 | 48 |
| Serikzhan Adilhan | April 4, 2019 | May 23, 2019 | 49 |
| Cui Jiafang | February 12, 2018 | March 22, 2018 | 38 |
| Song Zhigang | March 4, 2018 | April 21, 2018 | 48 |
| Zhao Haijiang | August 24, 2015 | October 15, 2015 | 52 |
| Jehovah’s Witnesses case | April 24, 2019 | September 30, 2020 | 525 |
| Ablet Barat | March 15, 2004 | July 28, 2004 | 135 |
| Xu Jie | December 1, 2014 | March 26, 2015 | 115 |
| Wang Jinrui and Li Gang | February 11, 2018 | June 11, 2018 | 120 |
| Chen Caihua | July 20, 2018 | September 29, 2018 | 71 |
| Wang Rongrong and others | January 4, 2019 | March 29, 2019 | 84 |
| Jiang Chunhua and Zhang Fengyin | December 26, 2018 | March 20, 2019 | 84 |
| Wang Xiupin and others | June 25, 2019 | September 10, 2019 | 77 |
| Fan Lianjun | September 1, 2016 | September 26, 2016 | 25 |
| Fu Guirong | September 12, 2014 | October 10, 2014 | 28 |
| Zhou Zunhua and others | October 13, 2014 | December 10, 2014 | 58 |
| Wu Shengmin | August 22, 2013 | December 31, 2014 | 496 |
| Xiao Fengying and Yan Zhenhua | February 13, 2015 | March 30, 2015 | 45 |
| Liu Aihua and others | April 30, 2015 | May 20, 2015 | 20 |
| Cui Xinfeng | May 4, 2015 | June 15, 2015 | 42 |
| Liu Chunqiong | March 29, 2016 | June 2, 2016 | 65 |
| Li Jia | February 29, 2016 | May 5, 2016 | 66 |
| Liu Wenbiao and others | September 20, 2017 | October 11, 2017 | 21 |
| Liu Guixin | July 20, 2017 | October 11, 2017 | 83 |
| Tao Hui | March 1, 2017 | June 20, 2017 | 111 |
| Zhang Fuqing | August 5, 2016 | September 8, 2016 | 34 |
| Hou Qingrong | February 11, 2019 | April 29, 2019 | 77 |
| Tian Yiqin and others | April 10, 2019 | June 25, 2019 | 76 |
| Ban Guoyun | August 26, 2014 | October 24, 2014 | 59 |
| Gao Huizhu and Jin Jingzhi | August 8, 2017 | October 8, 2017 | 61 |
| Wang Hongjun | March 25, 2016 | May 12, 2016 | 48 |
| Deng Yonghua | March 30, 2015 | June 1, 2015 | 63 |
| Jiang Yanghua | April 13, 2020 | June 9, 2020 | 57 |
| Cheng Hui and others | September 6, 2019 | December 18, 2019 | 103 |
| Average | 86.3 |
This suggests that the average time between indictment and verdict is less than 3 months.