The duration of political imprisonment: Evidence from China

The Chinese regime is well known for the large-scale detention of dissidents and ethnic minorities. However, little is known about the fates of Chinese political prisoners. This study investigates determinants of the duration of political imprisonment in China. I argue that the duration of political imprisonment is shaped by (a) the perceived threat of individuals’ actions, and (b) their ethnic and religious identities. Drawing on the Chinese political prisoner database, I investigate predictors of the duration of political imprisonment with survival models. Since preceding actions shape detention times, I hand-code each prisoner's criminalized actions that led to incarceration. The evidence suggests that the Chinese regime conditions the duration of political imprisonment on prisoners’ demands and their collective action potential. The findings further demonstrate that ethnic Uyghurs and Tibetans are imprisoned significantly longer than non-minority political prisoners. Additional analyses demonstrate that ethnic Uyghurs are also significantly more likely to die in prison.


Introduction
In January 2014, Chinese security officials detained seven students of the Uyghur economics professor Ilham Tohti. The students were detained for their alleged contribution to the website Uyghur Online that was founded by Professor Tohti (Olesen, 2014). While all seven students were charged with separatism (CL, art. 103), they received varying prison sentences of eight, seven, five, three and a half, and three years (Tiantian, 2014). 1 The lightest prison sentence was handed down in the case of Luo Yuwei, who is the only non-Uyghur among the seven students (RFA, 2014). Professor Tohti himself, who allegedly "bewitched and coerced" his students into writing articles for Uyghur Online, was also charged with separatism (CL, art. 103) and sentenced to life imprisonment (Olesen, 2014). The US Congressional Executive Commission on China considers Professor Thoti and his seven students as political prisoners (CECC, 2021). This example illustrates the substantial variation in the duration of political imprisonment across Chinese political prisoners and that there is no deterministic relationship between the type of criminal charge and the incarceration length. Which factors determine the duration of political imprisonment in China? This question has important implications, given that the likelihood that political prisoners suffer from enduring mental and physical harm increases as a function of the duration of incarceration (Sarraj et al., 1996). While it is well established that political imprisonment is widespread in China (Franceschini and Nesossi, 2018;Truex, 2019;Xu, 2020), it remains unclear which factors determine how long Chinese political prisoners are deprived of their liberty.
Political prisoners are herein defined as individuals detained for exercising their human rights under international law, following the conceptualization of the US Congressional Executive Commission on China (CECC, 2021). In line with the approach of the United Nations Working Group on Arbitrary Detention (UN WGAD), I apply international human rights law as an external standard to evaluate whether a detention was politically motivated. 2 I focus on political motivations of states-in contrast to political motivations of individuals-to identify whether a detention qualifies as political imprisonment. This state-centered focus implies that the decisive question is not whether individuals were politically motivated but whether the decision of the state to detain these individuals violated their human rights under international law (Steinert, 2021). Hence, both prisoners detained for political actions and prisoners detained for non-political actions, such as practicing their religion, can be defined as political prisoners as long as their actions are protected by international human rights law. In contrast to previous research on sentencing patterns in China that focus on specific criminal charges (Hou and Truex, 2020;Lu and Miethe, 2002), this broader definition covers both formally charged and extra-judicially detained prisoners. This is important because a large number of Chinese political prisoners are extra-judicially detained in "reeducation through labor" (laodong jiaoyang) camps, i.e. they have never been formally charged (Zenz, 2019).
I develop a novel theory of the duration of political imprisonment in autocratic regimes. I argue that the duration of political imprisonment is shaped by (a) the perceived threat of an individual's actions and (b) the ethnic and religious groups an individual belongs to. Autocratic regimes have incentives to incapacitate those individuals for long times that are considered as threatening to their rule. I suggest that individuals are deemed as most threatening if they make maximalist demands (Klein and Regan, 2018) and have high collective action potentials (King et al., 2013). However, long-term incarcerations of individuals with high collective action potentials are especially likely to trigger backlash protests (Buntman and Huang, 2000;Murphy, 2014). In anticipation of this risk, I argue that autocratic regimes accommodate political prisoners who combine high collective action potentials with minimalist demands.
Furthermore, I suggest that autocratic regimes have incentives to condition the duration of political imprisonment on prisoners' ethnic and religious identities. While political imprisonment may spark domestic opposition (Bell et al., 2013;Lawrence, 2017), the level of opposition is diminished if minority groups are targeted. Citizens tend to be less resistant against human rights violations that are directed against out-groups (Harris and Fiske, 2015;Piazza, 2015). By choosing harsher punishment for minority groups, autocratic regimes create a selective deterrence effect without risking a backlash from members of the majority group. I argue that autocratic regimes are especially likely to choose long detention times for minority groups that credibly signalled their mobilization capacities in the past.
I test these arguments with data on detention times of known Chinese political prisoners from the US Congressional Executive Commission on China (see Liu, 2019;Truex, 2019;Xu, 2020). I hand-coded each individual prisoner's demands and collective action potential based on qualitative information from NGO reports about their respective criminalized actions that led to political imprisonment. Drawing on Cox proportional hazard models, I find that ethnic Uyghurs and Tibetans are imprisoned significantly longer than political prisoners from majority groups. This is true despite the fact that the ethnic Uyghurs included in the dataset tend to be detained for actions with lower collective action potential and weaker demands than the average political prisoner. Additional evidence suggests that ethnic Uyghurs are not only longer in detention but also more likely to die in prison. I also find that the scales of justice are tipped against the religious minority of the Falun Gong. In support of the claim that autocratic regimes seek to minimize the risk of a backlash, I find that political prisoners who combine high collective action potential with minimalist demands are released earlier.
To the best of my knowledge, this is the first quantitative study of the duration of political imprisonment. While evidence from a single case cannot provide a definitive test of a general theory, the suggested framework contributes to our theoretical understanding of political imprisonment as a dynamic process (Truex, 2019). Future research may want to investigate how the patterns traced in this study generalize to other political contexts. The findings of this study add to our understanding of political imprisonment in China and shed light on systematic biases in punishment outcomes (Hou and Truex, 2020;Lu and Miethe, 2002;Xu, 2020). While ample evidence demonstrates that ethnic Uyghurs are targeted in disproportionate numbers (Greitens et al., 2020;Zenz, 2019), this study suggests that ethnic Uyghurs are systematically discriminated even in comparison with other illegitimately detained political prisoners.

Political imprisonment in China
"The freedom of the person of citizens of the People's Republic of China is inviolable. Unlawful detention or deprivation or restriction of citizens' freedom of the person by other means is prohibited." While these principles mirror the tone of international human rights declarations, they are in fact drawn from Article 37 of the Constitution of the People's Republic of China (NPC, 2021). In stark contrast to those stipulations, the right to freedom of person is violated on a massive scale in China, described as "the global leader in political prisoners" (Puddington, 2018). Political imprisonment has been a key tool of repression in China ever since the crackdown on the pro-democracy protests on the Tiananmen Square, but the number of political prisoners soared under the leadership of Xi Jinping (Chen and Cohen, 2022). Among the most salient examples were the "709 crackdown", targeting hundreds of rights activists in summer 2015, and the ongoing mass detention campaign in Xinjiang of up to 3 million minority citizens (Greitens et al., 2020;Zenz, 2019).
The discrepancy between law and practice reflects a substantial "implementation gap" characterizing the Chinese legal system (Delisle, 2019: 259). This implementation gap results from the deliberate subordination of the law to the goal of political stability (Ng and He, 2017). The performance of courts is evaluated in light of their ability to deliver stability (Weiwen) instead of their adherence to legal principles (Benney, 2016). Chinese judges are encouraged to find ways to resolve cases without using the law, either through mediation or by resorting to other extra-legal means. Many cases do not even reach the judiciary but extra-judicial actors implement administrative detentions (Chen and Cohen, 2022). The decision to imprison a citizen frequently results not from the consistent application of legal principles but from efforts to promote social harmony and stability (Li, 2018). Mirroring these objectives, the Chinese Communist Party (CCP) uses the term "being harmonized" to refer to someone who is detained or arrested (Carter and Carter, 2020).
While there is evidence on the targets (Franceschini and Nesossi, 2018;Trevaskes, 2011), the predictors (Truex, 2019;Xu, 2020), and the objectives (Li, 2018;Zenz, 2019) of political imprisonment in China, we lack a systematic understanding of the question of how long Chinese political prisoners remain in detention. The absence of empirical evidence on the duration of political imprisonment is problematic given that the likelihood of enduring psychological and physical harm increases as a function of the time of incarceration (Sarraj et al., 1996). Longer times of political imprisonment may decrease the prospects of societal integration owing to an inverse relation to the time for education and professional development. Systematic differences in incarceration lengths may also contain valuable insights into the repressive strategies of the CCP.
As of yet, the only empirical evidence on the duration of political imprisonment in China comes from Truex's (2019) study, showing that dissidents arrested before focal events tend to have shorter imprisonment periods. While this "catch-and-release" dynamic around salient dates such as the Tiananmen anniversary adds an important piece to our understanding of the duration of political imprisonment, the explanation remains incomplete. Given that Truex's (2019) study treats the duration of political imprisonment as a dichotomous variable, its explanatory power for variation in continuous detention times is limited.
Another related body of research sheds light on the length of criminal prison sentences in China (Hou and Truex, 2020;Lu and Miethe, 2002). Hou and Truex's (2020) study demonstrates that defendants from ethnic minorities receive significantly longer prison sentences than Han defendants for similar drug crimes. However, this evidence pertains only to prisoners who have formally entered the criminal justice system. Political prisoners are frequently extra-judicially detained and consequently they may never receive a formal prison sentence (Liu, 2019). Even if political prisoners are formally tried, actual times of detention may deviate from formal prison sentences (Chen and Cohen, 2022). Hence, we need a separate investigation of the duration of political imprisonment that focuses on the actual time political prisoners remain deprived of their liberty and that includes extra-judicially detained prisoners.
The absence of empirical evidence on the duration of political imprisonment goes beyond the case of China. Several studies from different countries find evidence for racial, gender or class biases in the length of criminal prison sentences (Mitchell, 2005;Park, 2017). However, no study investigates determinants of the duration of incarceration in the specific context of political imprisonment. To fill this gap, I first develop a general-abstract theory on the duration of political imprisonment focusing on strategic motivations of autocratic regimes. Subsequently, I apply this general theory to the Chinese context and test it with empirical data on detention times of Chinese political prisoners between May 1981 and June 2020.

Theory of the duration of political imprisonment
Time is a key factor of punishment that shapes its impact on both the culprit and society (Foucault, 1979: 138). This applies particularly to prison sentences whose lengths tend to be a function of the severity of crimes. The more severe a crime is perceived to be according to the criminal code of a state, the longer the prison sentence for the offender. The duration of political imprisonment likewise reflects the perceived severity of individual's actions. However, political imprisonment differs from criminal imprisonment in important respects. First, numerous political prisoners are extrajudicially detained, implying that the severity of individual's actions is not evaluated in a regular trial. 3 Second, even if political imprisonment results from sentences in regular trials, it is problematic to classify the actions of political prisoners as "crimes" given that prison sentences are politically motivated. 4 Consequently, the concept of crime severity is unsuitable to explain the duration of political imprisonment. We need a different theoretical framework to explain how long political prisoners are deprived of their liberty.
In the following section, I develop a novel theory arguing that the duration of political imprisonment is determined by (a) the perceived threat of an individual's actions and (b) the ethnic and religious groups an individual belongs to.

The perceived threat of an individual's actions
Political imprisonment is a repressive strategy that aims to deter and neutralize individuals through their incarceration and seeks to disrupt dissident networks (Kenney, 2017;Murphy, 2014). While the decision to imprison is dichotomous, the extent to which political imprisonment fulfils its goals is a function of its length. The longer political prisoners are detained, the stronger is the deterrent effect for both the individual prisoner and potential future suspects. Longer incarcerations also imply that dissident networks are profoundly disrupted and they increase the likelihood that political prisoners experience long-term mental and physical harm (Sarraj et al., 1996). Hence, autocratic regimes might optimize for unlimited times of political imprisonment.
However, the costs of political imprisonment also rise as a function of its duration. Political imprisonment is costly in the sense that the presence of political prisoners casts doubt on the legitimacy of the regime (Buntman and Huang, 2000). Political imprisonment represents a salient indicator of arbitrary rule, creating "embodied grievances" in the population (Murphy, 2014: 58). As a consequence, autocratic regimes that imprison individuals for politically motivated reasons face domestic opposition (Bell et al., 2013;Lawrence, 2017) and international reputation costs (Trevizo, 2014). In particular, long-term incarcerations of political prisoners may be deemed as inhumane and irreconcilable with the veneer of legitimacy that autocrats seek to cultivate (Guriev and Treisman, 2020). Hence, autocrats have incentives to release political prisoners, making it unlikely that unlimited prison terms represent their default choice.
Instead, it is plausible that autocrats selectively decide for long-term incarceration if they value its benefits higher than its costs. Autocratic rule is inherently fragile and dictators are driven by an existential fear for survival (Greitens, 2016;Svolik, 2012). Since dictators lack popular legitimacy, they are confronted with the threat of being violently overthrown. Hence, it is likely that they value long-term incarceration highest if it minimizes the level of threat to their survival. The length of political imprisonment may be tailored in a case-by-case decision to the perceived level of threat posed by a citizen. 5 Autocratic regimes lack complete information on the capabilities of individual citizen to threaten their rule. Per definition, the regime has not been overthrown yet when deciding on the incarceration lengths of political prisoners, giving the decision a predictive character based on factors that are ex ante unknown. In the absence of complete information about a citizen's true type, punishment tends to be conditioned on observed behavior. The duration of criminal imprisonment is a function of the perceived severity of the observed crime, leading to incarceration. Similarly, the duration of political imprisonment may be a function of the perceived threat of the observed criminalized action, leading to incarceration. Which factors determine to what extent an action is deemed threatening?
First, I argue that an action's perceived level of threat is shaped by its collective action potential. Autocratic regimes are concerned with individuals that initiate collective action, given that coordinated group actions may threaten their survival (King et al., 2013). Collective action potentials (CAPs) can be defined as the ability of individuals to mobilize others and to organize group-based action. Expressed in social network terms, CAPs are highest for those nodes with high values of betweenness centrality, i.e. citizens who are important in spanning between other citizens in the network. While CAPs are unobservable, regimes may infer CAPs from criminalized actions. For instance, if citizens are caught in founding or leading dissident organizations, there is prima facie evidence for high CAPs. In contrast, autocratic regimes may ascribe low CAPs to citizens who individually express dissatisfaction. Autocratic regimes probably consider those citizens most threatening who are deemed to have high CAPs.
Second, I suggest that an individual's perceived level of threat is shaped by the nature of her demands. Criminal prison sentences reflect criminal actions and their underlying motives. Likewise, it is plausible that political imprisonment factors in substantive motivations. In a hypothetical example, citizen A is arrested for organizing a protest march against a specific policy, whereas citizen B is arrested for organizing a protest march against the regime. Despite the fact that both citizens are punished for the same type of action, it is plausible that the regime prefers harsher punishment for citizen B. Individuals are likely to be perceived as less threatening if they have minimalist demands aiming for limited policy changes within the scope of the existing political system. Individuals may be perceived as more threatening if they have maximalist demands targeting the core of autocratic rule and seeking to alter the loci of authority (Klein and Regan, 2018). The nature of individual's demands can only be relevant to the extent that it is observable for the regime. Autocratic regimes may deem citizens as most threatening if they uncover maximalist demands as drivers of criminalized actions.
In summary, I suggest that the perceived threat of an individual is shaped by (a) her CAP and (b) by the nature of her demands. These two dimensions are probably evaluated in their interplay, as illustrated in Table 1. Individuals may be seen as most threatening if they have high CAPs and make maximalist demands.
The discussion suggests that autocratic regimes have incentives to choose long-term political imprisonment for individuals with maximalist demands and high CAPs. However, autocratic regimes are expected to also factor in the costs of long-term political imprisonment. Political imprisonment tends to create grievances and spark opposition (Lawrence, 2017), which is especially likely if political prisoners have high CAPs. Individuals with high CAPs are well connected and publicly known, which makes it likely that their detention aggrieves a large number of citizens. Hence, the detention of publicly known individuals may increase the risk of a backlash (Esberg, 2021). 6 While autocratic regimes have incentives to detain individuals with high CAPs, their long-term detention tends to be especially costly. Owing to these two opposing effects, autocratic regimes may factor in additional information on prisoners' demands. Political prisoners who combine high CAPs with maximalist demands are particularly dangerous for autocrats, given that they unite the capacity to mobilize others with goals that fundamentally threaten the regime. In contrast, political prisoners who combine high CAPs with minimalist demands may be regarded as less dangerous. The risk of a backlash could outweigh the benefits of long-term imprisonment if individuals aim for reforms that do not threaten autocratic survival. Hence, autocratic regimes may have incentives to release such political prisoners earlier. In light of these arguments, I derive the following hypotheses: H1: Political prisoners with maximalist demands are imprisoned for longer time periods than political prisoners with minimalist demands.
H2: Political prisoners with high CAPs are imprisoned for shorter time periods than those with low CAPs if they refrain from maximalist demands.

Demographically targeted punishment
The Roman goddess Iustitia is frequently depicted blindfolded to symbolize the principle of "equality before the law", i.e. the idea that defendants are treated equally irrespective of their ethnic, religious, racial or sexual identities. This key principle of criminal imprisonment does not apply to political imprisonment. Instead, I suggest that the duration of political imprisonment is demographically targeted. Repression is demographically targeted if individuals from specific ethnic, religious, racial or other externally identifiable groups are subjected to harsher repression than others (Rozenas, 2020). Demographically targeted repression implies that even if two individuals behave identically, they are treated differently by the judiciary if they come from different demographic groups. Per definition, all political prisoners are victims of discriminatory treatment owing to political rationales of the state. Discriminatory treatment may result from two different logics: it is either connected to specific types of behavior or to specific types of identities (or both). In the former case, individuals whose actions are deemed as politically threatening are discriminated vis-à-vis individuals whose actions are regarded as politically neutral. In the latter case, individuals from specific demographic groups are discriminated vis-à-vis individuals from other demographic groups. Why is it rational for autocracies to condition the duration of political imprisonment on demographic characteristics?
Conditioning punishment on ascriptive group-level characteristics allows autocratic regimes to exploit out-group discrimination in the population. As stated above, autocracies face incentives to release political prisoners to accommodate domestic opposition. The level of domestic opposition against human rights violations is diminished if a smaller faction of society is affected by them. Individuals tend to be more acceptant of human rights violations directed at "the other" owing to out-group biases (Harris and Fiske, 2015;Piazza, 2015). Politicians who "dirty their hands" when confronting members of marginalized groups even tend to gain public support (Conrad et al., 2018). Beyond such strategic rationales, subconscious biases may also contribute to the harsher punishment of minority groups.
Out-group biases are especially likely to emerge if demographic minorities differ on dimensions that are fundamental to the identities of in-groups. Religious and ethnic identities fulfil this condition as they provide a feeling of belonging connected to shared traditions and values. In contrast to fluid demographic characteristics such as age, they also tend to be sticky, transmitting a life-long sense of allegiance (Denny and Walter, 2014). While the salience of ethnicity and religion is socially constructed and differs over time and across individuals, it is unquestionable that religious and ethnic affiliations lie at the heart of the identities of many individuals. As a consequence, it is common that out-group biases emerge along ethnic or religious lines (e.g. Mamdani, 2020).
Against this backdrop, I expect that autocratic regimes have incentives to capitalize on out-group biases against ethnic and religious minorities in the choice of incarceration lengths. Autocratic regimes may be especially likely to opt for this strategy if minority groups are deemed as threatening to their survival. Shared ethnic and religious identities represent an important mechanism of collective mobilization as they facilitate coordinating a movement and maintaining its cohesion (Denny and Walter, 2014). However, ethnic and religious minority groups differ in their capacities and willingness to mobilize against the regime. Autocratic regimes might deem those minority groups as especially threatening that credibly signalled their mobilization capacities in the past (Roessler and Ohls, 2018). By choosing harsher punishment for such minority groups, they create a selective deterrence effect without risking a backlash from the majority. Hence, conditioning punishment on ethnic and religious group characteristics may allow autocracies to increase the costs of identity-based political mobilization.
Drawing the different arguments together, I suggest that the duration of political imprisonment is shaped by (a) the perceived threat of individuals' actions and (b) the ethnic and religious groups they belong to. I argue that these two dimensions independently shape the duration of political imprisonment. Holding the perceived threat of actions constant, individuals from ethnic and religious minorities are expected to be longer detained than other political prisoners. Hence, I derive the following hypotheses: H3a: Political prisoners from ethnic minorities are imprisoned for longer time periods than those from ethnic majority groups.
H3b: Political prisoners from religious minorities are imprisoned for longer time periods than those from religious majority groups.

Empirical case: political imprisonment in China
I test the theoretical claims with the Political Prisoner Database (PPD) provided by the US Congressional-Executive Commission on China. 7 I follow several recent studies that employ the PPD to conduct quantitative analyses related to political imprisonment in China (e.g. Gruffydd-Jones, 2021;Liu, 2019;Truex, 2019;Xu, 2020).
The PPD stores information on known individuals that have been detained in China for exercising their human rights under international law. 8 The information on individual detainees includes, for instance, basic demographic details, current detention status, detention and release dates, information on legal processes, and the location of detention. The PPD draws on publicly available Chinese and English sources and on data collected by international non-governmental organizations (NGOs) that collaborate with local human rights activists. These organizations include the Dui Hua Foundation, Human Rights in China, the Network of Chinese Human Rights Defenders, and Human Rights Watch. The PPD dates back to the beginning of 1981 and the dataset is continuously updated. As of July 23, 2020-the date the dataset was pulled-10,271 political prisoners were stored in the dataset.

Duration of political imprisonment in China
The key variable of interest-the duration of political imprisonment-is created by subtracting the "date of detention" from the "actual date of release" for each individual political prisoner. 9 If only information on the date of detention is available and political prisoners are currently detained as indicated by the variable "detention status", I treat these cases as right-censored taking the date when the dataset was pulled as the censoring date. I focus on the actual duration of political imprisonment instead of the length of formal prison sentences for two reasons: first, only 53.7% of the political prisoners covered in the PPD have been formally charged. The remaining 46.3% are held in police detention or they are serving an administrative term of reeducation through labor. Second, individuals who are formally charged tend to spend significant amounts of time in pre-trial detention (Xiong and Wei, 2017). Even the formal arrest date is no valid indicator for the start date of political imprisonment. In the Chinese legal system, the term "arrest" (daibu) does not refer to the initial deprivation of a person's freedom but to the formal approval of a detention that may have taken place earlier (Chen and Cohen, 2022). Hence, I focus on the date of the detention capturing the day when public security officials took a person into custody. Arguably the duration of political imprisonment should cover the entire time period that an individual was deprived of their liberty for political motives. Figure 1 illustrates the duration of political imprisonment for known Chinese political prisoners as a survival function. Survival refers in this context to the time that individuals remain detained. The detention times range from 1 day to 244 months with an average time of 29.6 months. Figure 2 shows that the average length of political imprisonment varies substantially across Chinese provinces. 10 The longest average time of political imprisonment is reported for the Jilin province. This could reflect the fact that the Falun Gong movement was founded in the Jilin province, which has been subject to a massive eradication campaign by the CCP (Cheung et al., 2018).

Criminalized actions of Chinese political prisoners
The PPD provides no information on political prisoners' CAPs and on their demands. Therefore, we manually coded both dimensions with regard to the "criminalized action" a political prisoner was accused of. 11 We studied the case description of each individual political prisoner in the "short summary" column in the PPD and consulted further information from NGO reports in the provided links. We hand-coded a prisoner's "criminalized action" based on descriptions of the type of action s/he was caught in or accused of, leading to imprisonment. Instead of focusing on formal prison charges, we searched for the substantive actions that individuals were criminalized for. The advantage of our approach is that this variable can also be coded for political prisoners who were extrajudicially detained and have not been formally charged. 12 We classified criminalized actions along the dimensions of (a) CAPs and (b) the nature of demands. CAPs capture the extent to which an action has a collective character and the capacity to mobilize others. The coders were instructed to search for pieces of evidence suggesting that a prisoner was actively involved in group foundation or leadership activities. CAPs were coded as high if individuals were caught founding opposition groups or leading group-based protest against the regime. We further coded CAPs as high if prisoners are publicly known or are symbolic figures. 13 Otherwise, CAPs were coded as low.
The dimension of demands refers to the objectives underlying a criminalized action. While true objectives are unobservable, the description of the accusations against individual prisoners provides useful insights into their objectives as perceived by the regime. We coded demands as maximalist if individuals demand fundamental political changes and their opposition is directed against the current political system or the rule of the CCP. The coders were instructed to search for pieces of evidence hinting toward goals such as democratization, leadership change or separatism. If no such evidence was traced, demands were coded as minimalist.
Each political prisoner was assigned to one category within the two dichotomized dimensions of CAPs and demands. 14 This resulted in four mutually exclusive and collectively exhaustive categories of criminalized actions. 15 I label individuals with maximalist demands and high CAPs as mobilizing dissidents and those with maximalist demands and low CAPs as non-mobilizing dissidents. Individuals with minimalist demands and high CAPs are named mobilizing reformers, while individuals that combine minimalist demands with low CAPs are called non-mobilizing reformers.
To illustrate the category of non-mobilizing reformers (upper-left panel in Table 1), consider the case of Jin Guangming, who was detained in July 2008. After being forcibly evicted from his home, Guangming petitioned against his eviction and the inadequate compensation he received. In response, Guangming was detained and forced to sign an "eviction agreement". Guangming's demands can be classified as minimalist since he only demanded compensation in relation to his eviction. We classified his CAP as low given that he individually protested by means of petitioning (xinfang).
To illustrate the category of mobilizing reformers (lower-left panel in Table 1), consider the case of Ou Quanjiang. He was accused of organizing a large protest against the planned construction of a waste incineration plant in Ningxiang. His demands can be classified as minimalist as his protest was solely directed against a specific policy within the limits of the current political system. Quanjiang's CAP can be classified as high given that he organized a large-scale collective protest.
The category of non-mobilizing dissidents (upper-right panel in Table 1) can be illustrated with the case of the Tibetan nun Phuntsog Nyidron, who was imprisoned in November 1989. She participated in a protest against Chinese occupation by secretly recording songs on a cassette with proseparatism propaganda. Her demands are classified as maximalist as she was involved in the Tibetan separatist movement. Her CAP is low given that she did not organize but only participate in group-based protest.
As an example for the category of mobilizing dissidents (lower-right panel in Table 1), consider the case of Ding Zilin, who was detained in March 2004. Zilin was accused of being the founder and leader of the Tiananmen Mothers, a group of more than 100 women whose relatives were killed during the Tiananmen crackdown. Zilin's demands can be classified as maximalist since she calls for far-reaching reforms such as democratization and the release of all political prisoners. In light of her leadership role and the initiation of international campaigns, her CAP can be classified as high. Figure 3 illustrates the distribution of these categories across the newly coded categorical "criminalized action" variable. 16 Based on this classification, most known political prisoners have low CAPs and minimalist demands (47.7%). Only a small group of the political prisoners has high CAPs and maximalist demands (2.7%).
To demonstrate how the different types of criminalized actions relate to the duration of political imprisonment, I mapped the average detention lengths across each subgroup as shown in Figure 4. 17 In line with the theoretical expectation, maximalist demands are associated with longer political imprisonment than minimalist demands. CAPs are positively related to detention length if demands are maximalist, but negatively related if demands are minimalist. This suggests that the regime may choose to release mobilizing reformers in anticipation of a potential backlash.

Multivariate analyses
I model the determinants of detention lengths with individual political prisoners as the unit of analysis. Explanatory variables such as ethnicity or religion are non-randomly assigned and causally prior to effectively all other variables precluding causal identification (Hou and Truex, 2020;Sen and Wasow, 2016). I use a conditioning approach with the more modest aims of (a) identifying correlations with the duration of political imprisonment and (b) testing whether the correlations hold after accounting for a rich set of covariates.

Explanatory variables and controls
The main explanatory variables capture (a) a prisoner's criminalized action and (b) her ethnic and religious identity. I create three binary variables of criminalized actions differentiating between mobilizing dissidents (HCAP_MAX), mobilizing reformers (HCAP_MIN), and non-mobilizing dissidents (LCAP_MAX). The group of non-mobilizing reformers (LCAP_MIN) is treated as the reference group. To operationalize ethnic identities, I draw on the "ethnic group" variable provided by the PPD. I create binary variables for ethnic Uyghurs, Tibetans, and other ethnic minorities. Ethnic Han are treated as the reference category. To measure religious identities, I use the "religion" variable from the PPD. I create binary variables for Muslim, Christian, Buddhist, Falun Gong, and other religious groups. I treat non-religious prisoners as the reference category since the CCP officially proclaims state atheism and most Chinese do not self-identify with any religious group (Yang, 2011).
I condition on prisoner's sex and age with variables taken from the PPD. I create binary controls for several occupations from the PPD including journalists, workers, lawyers, entrepreneurs, students, professors, and unemployed prisoners. I control for recidivism since detention lengths are likely to be longer if individuals have a criminal record. The PPD includes individual prisoners as new observations for each detention period. To identify repeatedly detained individuals, I created unique "name-residence-occupation" identifiers for each prisoner and coded every repeated occurrence of the same identifier as recidivism. 18 I account for the type of detention with the "legal process" variable from the PPD. I include a binary variable for individuals that have been formally charged, treating extra-judicially detained prisoners as the reference category.
Detention lengths may also vary systematically across time. I control for focal event detentions that occurred at salient dates, such as during Party Congresses of the CCP, since these detentions tend to be short, following a "catch-and-release" dynamic (Truex, 2019). I create a binary indicator coded as 1 if political prisoners were detained during the month of a focal event, building on Truex's (2019) list of focal events. 19 Additionally, I control for mass amnesties since they could reflect collective accommodation strategies that are unrelated to prisoner identities. I measure mass amnesties with a binary variable coded as 1 if at least 15 political prisoners were released on the same day. 20 Finally, I include fixed effects for years of detention and for the provinces where prisoners are detained. I present descriptive summary statistics in Table A.1.

Statistical model
The duration of political imprisonment is operationalized with the time in months that a prisoner remains in detention. Prisoners who are still detained at the date when the dataset is pulled are treated as right-censored cases. I use Cox survival models that assume proportional hazard while being agnostic about the functional form of the hazard. The hazard rate refers to the conditional probability that a prisoner is released in a given month t given that she has not been released before.

Results
The main results are presented in hazard ratios in Table A.3. Hazard ratios above 1 imply that prisoners are more likely to be released, while hazard ratios below 1 indicate a reduced propensity to be released. Table A.3 presents four different model specifications: (1) without fixed effects; (2) including fixed effects for years; (3) including fixed effects for provinces; and (4) including two-way fixed effects. 21 The findings of model (1) are visualized with a coefficient plot in Figure 5.
The most consistent finding across all four model specifications is that ethnic Uyghurs, ethnic Tibetans, and Falun Gong prisoners are discriminated in incarceration lengths. The hazard ratios for these minorities are below 1 and statistically significant in all model specifications. 22 This suggests that these minority groups are less likely to be released than political prisoners from ethnic or religious majority groups. Based on model 4, the probability of being released is, on average, 86% lower for ethnic Tibetans and 93% lower for ethnic Uyghurs compared with ethnic Han. For Falun Gong prisoners, the probability of being released is, on average, 62% lower compared with nonreligious prisoners. In Figure 6, I present adjusted survival curves for ethnic Uyghurs based on model 4. The figure illustrates that ethnic Uyghurs have higher average survival rates in prison than other ethnic groups, implying that they are less likely to be released.
Subsequently, I run a subset analysis including only non-mobilizing reformers, which is the modal category of criminalized actions. Figure A.1 summarizes the findings of this subset analysis. Ethnic Uyghurs, Tibetans, and Falun Gong prisoners are significantly less likely to be released even in comparison with political prisoners who are accused of similar criminalized actions. This suggests that the discrimination effects are not an artifact of heightened propensities for maximalist demands or high CAPs.
To explore whether discrimination occurs despite similar legal procedures, I run subset analyses for charged and extra-judicially detained prisoners. The findings visualized in Figure  demonstrate that ethnic Uyghurs tend to remain longer in detention both after formal prison charges and in the absence of charges. Ethnic Tibetans tend to be longer incarcerated only if they are extrajudicially detained, while Falun Gong prisoners tend to be longer detained only after formal charges.

Robustness and additional analyses
I test the proportional hazard assumption visually with scaled Schoenfeld residuals that indicate no systematic pattern over time. I also investigate proportional hazards for each individual covariate by applying the cox.zph-function of the survival package in R (Therneau, 2020). I stratify those covariates with high test statistics to allow for their non-proportionality. The findings remain robust to covariate stratification as demonstrated in Table A .8. To reduce model dependence, I employ coarsened exact matching (Iacus et al., 2012). The binary indicator for ethnic Uyghurs is matched on the covariates charged, male, and non-mobilizing dissidents (LCAP_MAX) that are unequally distributed between ethnic Uyghurs and other ethnic groups. 23 The unequal covariate distribution is substantially reduced through coarsened exact matching as illustrated in Figure A.3. Subsequently, I re-run the survival analysis with the matched data. Table A.4 demonstrates that the hazard ratio for ethnic Uyghurs remains negative and statistically significant after non-parametric covariate balancing. Figure A.4 illustrates that there are three outlier cases in the distribution of detention times. I exclude the outliers and re-run the survival analysis as shown in Table A.5. The binary variable capturing ethnic Uyghurs remains statistically significant. I also run an additional analysis with imputed missing data using AMELIA (Honaker et al., 2011). The findings shown in Table A.6 suggest that ethnic Uyghurs, Tibetans, and Falun Gong prisoners are significantly less likely to be released, while mobilizing reformers are more likely to be released. The results hold when I code missing values on the ethnicity-variable as 0, assuming that the absence of explicit information on a prisoner's ethnicity implies that they belong to the majority group of the Han. 24 So far, the analysis focused exclusively on detention times including only prisoners who (a) have been released or (b) that were still detained when the dataset was pulled. The PPD also includes 174 political prisoners who died in prison, representing a third outcome type. At first, I study this outcome independently by running a logistic regression model that predicts the binary outcome of dying in prison. I use the covariates from the main analysis and the "age"-variable given that high ages could be linked to natural deaths in prison. 25 The results of the logistic regression model are presented in Figure 7. The coefficient plot shows that ethnic Uyghurs and Falun Gong prisoners are more likely to die in prison even when accounting for their ages.
Finally, I re-estimate the likelihood of being released in a model that treats death in prison as a competing risk. Death in prison represents a competing risk since failure from the event of interest, i.e. being released, is no longer possible if a prisoner has died. In the presence of a competing risk, competing risk analysis can be used to estimate the marginal probability of the event of interest. Competing risk analysis partitions the probability of different events into the probabilities of each type of event via cumulative incidence functions.
I create an "event" variable that is coded as 0 if prisoners are detained at the pulling date, as 1 if they died in prison, and as 2 if they are released. As outcome variable, I use the time in months until an event, i.e. the time from detention to release or to death if a prisoner died in prison. 26 Figure 8 illustrates the cumulative incidence functions comparing between ethnic Uyghurs and all other ethnic groups. The figure illustrates that ethnic Uyghurs are, on average, both longer detained and more likely to die in prison. The results of the competing risk analysis are presented in Table A.7. They show that ethnic Uyghurs are less likely to be released also when adjusting for the competing risk of death in prison.
It is important to clarify that I do not claim to provide causal estimates of the effect of ethnicity (or religion) on the duration of political imprisonment. The relationship between ethnicity and choices of incarceration lengths is mediated by the decision to detain a citizen. This decision is itself affected by a citizen's ethnicity, which introduces a post-treatment bias to the analysis (see Knox et al., 2020). Given that ethnic minorities are imprisoned at higher rates, the threshold for detaining non-minority citizens is higher. 27 This means that the majority citizens observed in the detention data may not be comparable with the observed minority citizens. Having said that, the introduced bias is negative, which makes it less likely to find support for the hypotheses. Hence, it is plausible that the true discrimination effect is even stronger.

Discussion
To what extent does the empirical evidence from China align with the theoretical framework on the duration of political imprisonment? H1 is only partially supported by the empirical results. The multivariate evidence suggests that prisoners with maximalist demands tend to be detained longer than prisoners with minimalist demands but only if they have high CAPs. This suggests that prisoners' demands and actions must be jointly considered to explain variation in the duration of political imprisonment. In line with H2, the empirical evidence suggests that prisoners with high CAPs are imprisoned for shorter time periods than those with low CAPs if they refrain from maximalist demands. This suggests that the CCP may choose to accommodate influential individuals in anticipation of a backlash. However, the effect is not robust to yearly fixed effects and, hence, should be taken with a grain of salt.
H3a and H3b state that political prisoners from ethnic and religious minorities are imprisoned for longer time periods than non-minority prisoners. The empirical evidence provides strong support for both hypotheses. Ethnic Uyghurs, Tibetans, and Falun Gong prisoners are significantly less likely to be released. The discrimination of ethnic Uyghurs is the most robust finding holding across all model specifications. While ethnic Uyghurs and Tibetans are less likely to be released, this does not apply to other ethnic minorities (e.g. Mongols, Kazaks, Manchus). Likewise, while Falun Gong prisoners tend to remain longer in prison, this is not the case for other religious groups such as Muslims or Buddhist.
Qualitative evidence suggests that the CCP deems the minority groups of the Tibetans, Uyghurs, and the Falun Gong as especially threatening. Ethnic Uyghurs and Tibetans have been described as the least acculturated minorities owing to their cultural and linguistic distinctiveness, representing a challenge to China's quest to be a viable nation-state (Clarke, 2013). Scholars emphasize that the CCP faces an "internal security dilemma" (Clarke, 2007: 324) in Xinjiang and Tibet and that it considers transnational Uyghur networks as a vector by which Islamist terrorism could enter China (Greitens et al., 2020). Against this backdrop, ethnic Uyghurs and Tibetans have been linked to the "three evils of separatism, extremism and terrorism" by the CCP (Zenz, 2019). 28 These minority groups also credibly signalled their mobilization capacities in the past. The Tibetan uprising in 1959 showcased the determination of the Tibetan minority to rise against Chinese occupation (Jian, 2006). Likewise, the fear of Uyghur mobilization was fuelled by several violent incidents such as the Urumqi riots in July 2009 (Greitens et al., 2020). Similarly, the Falun Gong demonstrated their mobilization capacity, for instance, in the "Zhongnanhai Incident" in April 1999, when more than 10,000 individuals held a mass prayer in direct proximity of the the central headquarters of the CCP (Cheung et al., 2018). Consequently, the discrimination effect applies in particular to the Chinese "problem minorities" (Hou and Truex, 2020: 2) that are perceived to undermine stability and to threaten China's territorial integrity.
To what extent are these findings generalizable beyond China? The scope conditions of this study are limited given that the evidence comes from a single autocracy that is unique on various dimensions. Hence, the empirical findings must be cautiously interpreted as China-specific. It is possible that the discrimination of minority citizens in incarceration lengths applies especially to autocracies where ethnic conflicts represent a major source of instability. The Chinese regime faces different separatist movements, making China a "most likely case" to find evidence for ethnic biases in punishment outcomes. Another unique feature of the Chinese regime might be its extensive digital surveillance capacities (see Xu, 2020). Other autocratic regimes may have less information about the CAPs and demands of their citizens, which might weaken the relationship between citizens' actions and the duration of political imprisonment. We need empirical evidence from different autocracies to understand to what extent the suggested theoretical framework offers a generalizable explanation of the duration of political imprisonment.

Conclusion
This study set out to explain the duration of political imprisonment in China. The first key finding of this study is that certain ethnic and religious minorities are systematically discriminated in the duration of political imprisonment in China. The discrimination applies especially to ethnic Uyghurs and Tibetans and to the Falun Gong. This supports evidence that the CCP disproportionately targets its political out-groups (Greitens et al., 2020;Zenz, 2019) and suggests that the notion of "Han chauvinism" (Leibold, 2010: 544) extends to punishment outcomes. While there is rich evidence on the discrimination of minorities in criminal imprisonment (Johnson, 2008;Omori and Johnson, 2019), this is the first study to show that discrimination occurs also in the context of political imprisonment. The findings support evidence of ethnic inequalities in judicial sentencing in China (Hou and Truex, 2020), but show that these biases apply also to extra-judicial detentions.
The second key finding of this study is that the CCP tends to release political prisoners earlier who combine high collective action potentials with minimalist demands. Scholars have argued that the CCP chooses relatively lenient punishment for individuals whose demands are not seen as destabilizing (Trevaskes, 2011) and that it attempts to buy off protest leaders . This study suggests that demands and collective actions potentials must be jointly considered to explain repressive outcomes. In analogy to criminal punishment that factors in crimes and motives, political imprisonment likewise may be conditioned both on an individual's actions and on the demands driving these actions.
I acknowledge that the evidence permits no causal claims. Beyond the fact that it is inherently difficult to identify causal effects of ethnicity (Sen and Wasow, 2016), there are also data constraints with regard to potentially measurable factors. For instance, it is plausible that prisoners' behavior in prison may shape incarceration lengths, but we lack such information in the Chinese context. However, this is unlikely to nullify the findings since it is not plausible that minority prisoners are on average more recalcitrant in prison. It also must be considered that the coding procedure of CAPs and demands is error prone given that available information is thin. Future research should work toward more fine-grained and comprehensive data on Chinese political prisoners and investigate to what extent the findings of this study travel to different political contexts.
This study demonstrates that we have to go beyond a legalistic approach to explain punishment outcomes in China. Judicial actions have a "contingent character" (Ng and He, 2017: 15), being shaped by the efforts of the CCP to promote stability. Frequently, the judiciary is entirely circumvented and extra-judicial actors decide on how long individuals are deprived of their liberty. Hence, we should shift our focus from the written law to the empirical realities to explain the duration of political imprisonment in China and in other autocracies.

Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the German Academic Scholarship Foundation and the Graduate School for Economic and Social Sciences at the University of Mannheim.

Supplemental material
All data, replication materials, and instructions regarding analytical materials upon which published claims rely are available online through the SAGE CMPS website: https://journals.sagepub.com/doi/suppl/10.1177/ 07388942221080105.

Notes
1. The abbreviation CL refers to the Criminal Law of the People's Republic of China. 2. It is important to go beyond a state's own domestic laws to identify political imprisonment, given that autocracies may outlaw disliked behavior such as critique of the government. Since it would be problematic to deny the political prisoner status to victims of such laws, the UN WGAD applies the external