Authoritarian Evolution: Agency and Institutional Change in the Controlled Chinese Press

What drives gradual change in authoritarian regimes? This paper argues the institutional turn in comparative authoritarianism benefits from more robust engagement with theories of endogenous change that give greater weight to the role of cognition, ideas, and agency. While structural accounts of elite bargaining at critical junctures are key to understanding authoritarian change they are indeterminate with respect to long-term outcomes. Empirically, this article addresses unexplained puzzles, such as the emergence of a relatively professional print media in the otherwise controlled environment of the Chinese press. This analysis thus highlights political factionalism, principle-agent problems, and horizontal selection of policy ideas as important additional mechanisms of institutional change in authoritarian systems.

the Headlines, and BHow Institutions Evolve: Evolutionary Theory and Institutional Change,^(with Sven Steinmo) in Polity.
What drives gradual change in authoritarian regimes? Despite emerging literatures on both comparative authoritarianism and endogenous institutional change, scholars remain puzzled by questions of authoritarian stability and change. Lacking formal feedback mechanisms and officially prescribed roles for non-elite actors, scholars often assume autocrats largely respond to internal threats from other elites or external threats of large-scale collective action. Seminal work on authoritarian durability by Brownlee [6] placed Hosni Mubarak on the cover as an example of skillful manipulation of factions within a single party regime. A few years later, changes wrought in part by new media and satellite television had facilitated the regime's fall.
Why did scholars miss the endogenous changes that facilitated the Arab Spring? While partially stemming from an analytic bias toward institutional stability, I argue existing research on comparative authoritarianism also overlooks the space available for lower level actors to subtly influence policy implementation, gradually converting institutions endogenously over time. In other words, autocracies can evolve, and understanding the mechanisms by which this occurs provides a more robust ontology of authoritarian systems. In this paper, I argue that the standard structuralist account is a necessary but insufficient framework for explaining authoritarian institutional change. As a compliment, I outline an evolutionary theory of institutional change that explains how multiple mechanisms of change can exist under conditions of political control and stability.
Recent scholarship has sought to explain how authoritarian institutions are built and adapted to maintain durability [4,7]. Adopting a largely top-down perspective, this literature has clarified the elite calculus of co-optation that underlies dictators' expansion of governing coalitions and parties, and the emergence of new institutions, such as legislatures or rigged elections [6,11,12]. In this framework elites choose institutional change as a mechanism to stay in power.
While implicit threats from below make these Bendogenous^models of change, most approaches are essentially structuralist in nature. The locus of change lies mainly in intra-elite bargaining, and official decisions are assumed in large part to shape behavior throughout the rest of the system. On the surface this makes sense, as societal actors do not have a formal role in the policy process and face extreme power asymmetries. It is easy therefore to assume few societal-driven reasons why authoritarian regimes should change at all, so long as their hold on power is secure. However, this literature suffers from two key blind spots: first, scholars are primarily focused on Bfounding moments^and the question of institutional creation, largely ignoring endogenous change during periods of normal politics. Secondly, as models are largely static, few scholars investigate the timing and sequencing of major institutional changes, so it is difficult to determine whether such actions are proactive efforts to preempt societal challenges or reactions to changes that have already taken place informally. An evolutionary approach focuses attention on mechanisms of gradual change and allows researchers to avoid reductionism by evaluating multiple change mechanisms.
The principal-agent framework illuminates a number of reasons why authoritarian institutions evolve beyond the basic goals precipitating their creation. Political preferences amongst elites often vary more than scholars assume, and agents have discretion in the interpretation and implementation of abstract rules. Previous research has shown there is considerable space for ideational entrepreneurs to create new models of practice in autocracies. As other agents replicate successful institutional models, these modes of practice proliferate, becoming what Tsai [40] terms Binformal adaptive institutions.T he creation, selection, and replication of new practices are shaped by authoritative power relations, but sometimes also transgress official intent. Informal practices and institutions with and shape official policy endogenously.
This article argues learning and horizontal selection are important supplements to theories of authoritarian institutions which benefit from explicit engagement with general theories of endogenous change that give greater weight to the role of agency-in particular the adaptive strategies of rule-takers [28]. Ontologically, a more cognitive ontology of institutions recognizes the importance of mental schemas and ideas as the basic unit of institutional creation and replication [33]. Theoretically, a modified version of general evolutionary theory put forward by Lewis and Steinmo [22], provides a useful theoretical lens for understanding change in complex bureaucratic authoritarian regimes. While I provide a skeletal discussion of this framework, the primary goal of this article is empirical-to show how this framework adds insight into unexplained empirical puzzles, such as the degree of change in a stable authoritarian system such as China. Empirically, I explain changes in Chinese journalistic practices based on the combined effects of top-down state regulation and the bottom-up agency of news organizations in re-shaping journalistic norms.

Cognition and Agency in Theories of Endogenous Institutional Change
Traditionally the new institutionalisms have been overly structural, with agents often framed as prisoners of the institutions they inhabit [38] or universally self-interested based on the structural logics of their conditions [1,19]. In arguing for a more dynamic theory of change, Mahoney and Thelen [28] have described this problem clearly: BIf institutions are changed not just in response to exogenous shocks or shifts, then their basic properties must be defined in ways that provide some dynamic element that permits such change.^To address this challenge, scholars have made two key ontological moves. First, they developed a more explicit theory of agency, and secondly, they have posited more Binteractive^theories of change focused on the constant endogenous interaction of Brule-makers^and Brule-takers^(see also, [15,32]).
This ontology provides an important addendum to the literature on authoritarian change by positing a greater role for bottom-up agency. Even if institutions are created or adapted for rational reasons of control or stability, there may be multiple ideas on how to achieve that base preference, and leaders cannot predict all behavioral outcomes once they choose policy. Even the most strategic institutional rule change is likely to have Bunintended consequences.^This occurs because rule-takers respond differentially to abstract rules.( [28]:8) highlight institutions are Bdistributional instruments laden with power implications…any given set of rulesformal or informalthat patterns action will have unequal implications for resource allocation [italics in original].^Ruletakers vary considerably in the extent to which new rules meet their perceived interests or personal preferences and consequently may alter implementation. Abstract rules inherently contain ambiguity as to how they apply to specific contexts, providing space for agency.
The literature specifies multiple mechanisms by which rule-takers might alter the functioning of institutions. Hacker [16] argues those that seek to change an existing policy framework must ask two key questions: how easily can they achieve their goals within existing institutions, and how costly would it be to replace existing institutions with new ones? In authoritarian contexts rule-makers face fewer costs in replacing existing rules, a fact that explains the overriding focus of the existing literature on topdown decisions of Breplacement^-the creation of entirely new institutions-or Blayering^-the addition of new rules on top of existing institutions.
In contrast, rule-takers face prohibitive costs to creating new rules, so they must work within existing constraints, leaving them with two strategies: do nothing or take engage in subtle agency. BDrift^refers to a strategy of doing nothing until shifts in socioeconomic conditions render an institution irrelevant to current conditions. BConversion^occurs when rule-takers actively seek to shift the basic functions, goals and operation of existing rules. Hacker [16] argues Badaptation through conversion reflects the reality that most institutions or policies allow actors working within their constraints to pursue multiple ends.^Thus actors tasked with policy implementation may seek to re-interpret rules in order to suit their own goals and objectives within the discretion available for adaptation.
Opportunities for conversion increase as autocracies become more complex, decentralized, and multi-layered. Additional layers facilitate conversion by creating alternative, sometimes conflicting, institutional logics. Tsai [40] argues Bmultilayered institutional environments offer transformative possibilities; in particular, formal institutional contexts with overlapping jurisdiction and inconsistent or unrealistic mandates provide opportunities for actors to adjust, ignore, or evade discrete portions of formal institutions.^Geddes [14] highlights how autocracies generally expand their power base over time and become more pluralistic politically and institutionally. Thus the scope for the agency of rule-takers improves under mature, institutionally complex authoritarian regimes.
Research on the mind, cognition, and decision-making provide useful conceptual supplements for understanding how agents respond and adapt to environmental stimuli. North [33] has argued that attention to cognition and decision-making is one of the key missing pieces of institutional scholarship. In this vein, an increasing number of scholars are taking Bideas^seriously as the basic building blocks of institutions and change [3,34]. North ([33]: 383) argues that institutional evolution is based on the construction and reconstruction of Bmental models^that Bexplain and interpret the environment typically in ways relevant to some goal.^Thus institutions function to the extent that cognitive models are deeply ingrained and consistently acted upon. Bell ([3]: 884) frames the challenge thusly: Bhow to describe and explain contingent degree of agency-centered discretion (arguably the ultimate propellant of institutional change) within the constraint, conditioning and empowerment associated with institutionally embedded agents [italics in original].D ue in part to these theoretical challenges, Schmidt [34] points out that all major branches of new institutionalism-rational, historical, and sociological-have turned to ideas and discourse to explain gradual change. She argues that these common points of emphasis constitute a new category of Bdiscursive institutionalism^or Bconstructivistî nstitutionalism, while Bell [3] argues for Bsuitably tailored historical institutionalismb ased on Bbounded discretion.^Instead of new paradigms, Lewis and Steinmo [22] argue the evolutionary algorithm of adaptation, selection, and replication provides an encompassing theory of endogenous change-a meta-theoretical framework-that synthesizes all new institutional scholarship on gradual change. Tailored to account for the intentionality of human beings and cognitive abilities for communication and learning, this framework integrates micro-level concerns for agent variation and decision-making with macro-structural processes related to political and other environmental selection. This approach thus provides an explanation for the role of agency within otherwise stable institutional arrangements.
Ontologically, this paper takes Bschemas^and Bideas^as the basic mechanisms of endogenous institutional change. Schemas are defined as cognitive rules that tell individuals how to behave in certain contexts. Institutions are thus defined as the widespread, deeply-ingrained, and Binter-subjectively understood schemas about the formal and informal rules of behavior^( [22]: 333). Cognitive schemas are the Bgenesô f institutions, facilitating the replication of behaviors across actors and contexts, while Bideas^-defined as creative solutions to collective action problems-are Bmutations.T his clarifies the extent to which institutions are imperfectly replicated, the space for agency, and the opportunities for institutional conversion, even in authoritarian settings.

An Evolutionary Framework for Gradual Institutional Change
Instead of assuming agent homogeneity, research in genetics and cognitive psychology supports baseline assumptions of agent heterogeneity. Research based on evolutionary biology has found inherent variability in individual political preferences [9]. Weyland [41] argues positing preferences based on findings from cognitive psychology has Bthe advantage of resting on a wealth of well-corroborated empirical findings about human decision making.Â n evolutionary framework explains why preferences and behaviors change, even when structural conditions remain constant. Prospect theory highlights how behaviors shift with proximate information [18], and continued neuroscientific advances have supported research on how people alter decision-making in ways that are Becologically rational^ [29]. Thus, preferences shift in response to context, because they are more varied, more complex and more multi-layered than is often posited in reductionist accounts. Agents often value both individual and group-oriented outcomes simultaneously and the extent to which they balance these competing pressures becomes key to understanding behavior at any point in time. This explains why human behavior often diverges from the material individualism of early game-theoretic models.
Additionally, institutional research must account for the fact that human beings are uniquely innovative in their ability to imagine new problem solutions. Ideas are mutations that slightly alter and convert the behavior of agents. The study of ideas does not constitute a separate form of institutionalism but in fact informs and specifies the mechanisms of change that are central to the new institutionalisms. Ideas explain why agents often alter their behavior in response to new information, why agents shift the implementation of rules they perceive as being at variance with their interests or beliefs, and how agents develop new practices in response to abstract mandates. In sum, cognition, schemas, and ideas provide the theoretical foundation for elaborating communication-based mechanisms of endogenous change.
In evolutionary terms, ideas constantly interact with and are shaped by Bselection mechanisms,^which are defined as processes that narrow but do not eliminate the scope of variation within a system. Selection is central to determining which mutations are able to survive and replicate and which are constrained or eliminated. In political systems, scholars have outlined a variety of mechanisms by which policy ideas are selected, principally by institutions and authoritative power structures. Indeed, institutions are one of the distinguishing features of human societies, because they represent Bartificial^selection mechanisms designed to channel behavior in particular directions.
Here too, an evolutionary approach avoids the problems of reductionism. For example, multi-level selection theory (MLS) argues selection occurs at multiple levels of analysis. Theorists highlight how selection is often Bmyopic,^allowing multiple traits to survive and proliferate locally, while still being constrained in their ability to proliferate more broadly. This argues for a view of institutions as complex adaptive systems and an epistemology that disaggregates political regimes into multiple mechanisms of institutional stasis and change-a taxonomy.
From a political science standpoint, this paper builds out the concept by differentiating selection based on the direction of power relationships between selectors-those that make decisions that impact the implementation of any particular policy-and the object of selection-policy ideas. Where the selectors' choice threatens the continued existence of a policy idea, a mechanism is considered to be vertical in nature. The direction of selection denotes whether the process is top-down in terms of regime repression and control, or bottom-up in the form of the collective choices of citizens, whose tacit support institutions rest upon. Horizontal selection occurs between selectors that do not threaten the continued viability of a policy idea. For example, local leaders do not hold power over one another, but nevertheless Bselect^certain policy models, thereby causing them to diffuse.
Comparative authoritarianism focuses mostly on vertical mechanisms of power structures, punishment, and repression, while largely neglecting mechanisms of horizontal communication, learning, and adaptation. This article provides a useful supplement by illustrating how horizontal selection also drives institutional evolution even under stable power structures. In societies with institutions for horizontal communication and learning, and some degree of local discretion, rule-takers learn from events elsewhere and consciously adapt them to their own circumstances. Unlike Bblindê volution, human selection is often targeted-as agents choose policies based on what is relatively more successful elsewhere. Horizontal selection helps to account for sociological processes and mechanisms that facilitate the creation and replication of new informal institutions that transgress stated policy.
In sum, authoritative structures clearly shape the choices of rule-takers, but do not overly determine them. This occurs because friction between top-down and bottom-up selection, principle-agent problems, and the existence of horizontal mechanisms of learning and change. The evolutionary framework outlined here allows researchers to more fully explicate the mechanisms of endogenous change in autocracies-providing a more agency-centered compliment to structural accounts. Even under repressive conditions, autocrats must rely on schemas to shape behavior throughout a complex society. Schemas are imperfectly replicated across individuals-generating political fragmentation, political opportunity, new variations upon which selection operates, and potential shifts in agent preferences. The literature on authoritarian institutions thus benefits from broader engagement with theories of endogenous institutional change rooted in an evolutionary approach. The following section shows how this framework yields explanations to important empirical puzzles, in particular the Chinese government's maintenance of control over a dynamic information environment.

The Evolution of the Press in China: Change under Continuity
The People's Republic of China (PRC) meets many of the scope conditions outlined above for analyzing gradual change in authoritarian regimes. As one of the most durable authoritarian regimes in the modern era, the party-state is defined by multilayered institutional complexities. Secondly, the policy process in China is explicitly interactive and adaptive. The reform process has involved decentralization, local policy experimentation, and a policy process characterized by Bcrossing the river by grouping for stepping stones.^This interactive policy process is deeply embedded in the governing structure [17], and the practice of replicating successful models can be described as explicitly evolutionary in that policy is driven by experiential knowledge and learning.
This process of interactive and adaptive policy reform has also played itself out in information industries, yet scholarship has largely reduced the question of press reform to vertical state control in an era of commercialization and technological change. Following Tiananmen, the Color Revolutions, and the Arab Spring, the Chinese Communist Party (CCP) has consistently viewed information control as central to its durability [36]. The literature highlights state-driven views of the press defined by efforts to fix the propaganda system [5], keep control over a commercialized press [39], and bolster Internet censorship. As with much of the literature on Chinese authoritarianism, the view that emerges is one of a durable regime overseeing dramatic socioeconomic change.
I support the general understanding of top-down political selection as the dominant factor shaping behavioral outcomes in the Chinese press, but argue an elite-centric and structural approach to understanding media misses important additional mechanisms of change that impact on the diversity and quality of public discourse in the public sphere. Political factionalism, principle-agent problems, and horizontal selection have all driven important changes to the kind of information that citizens have access to in print. The iterative interaction of top-down selection with adaptive strategies amongst rule-takers has led to the emergence of multiple models of journalism-a situation I describe as Bpluralized authoritarianism.^Additionally horizontal selection amongst rule-takers has at times transgressed the conservative intent of the state and converted institutions to gradually include forms of more professional discourse. This explains a key empirical puzzle articulated by a top veteran editor, who noted the period since 2008 has been characterized by some of the most open reporting seen in China in the modern era and a much higher number of conflicts with the authorities-Bthis seems contradictory but is real.^1 In this regard, Chinese censorship is not always a proactive attempt to limit discourse but sometimes a reaction to change that has already occurred. Much is often made of the current round of repression, but such analysis omits alternative mechanisms of change, and cannot account for why repressive episodes generally occur in repeated cycles.
Empirically, the historical analytic narrative developed below focuses on Bchanges in journalistic norms and practices^as the dependent variable. I modeled these changes over time, using a mixed-method approach that combines content analysis with more than 100 in-depth fieldwork interviews spanning the period of 2006-2016 (Table 1). In order to grapple with the complexities of iterated interactions over time, and the interaction effects of both top-down and bottom-up variables, I reduce the narrative of media change to a multistage model-conceptually rooted in the basic Darwinian algorithm for gradual change-variation, selection, and replication. This model allows us to think about the interaction of agent variation with broader structural Bselection mechanisms^such as the regulatory state. The analysis and evaluation of the means by which journalistic practices replicate is inherently designed to evaluate the relative importance a different structural selection mechanisms, such as censorship, economic profit, and journalistic norms. While I evaluate the relative waiting a different explanatory variables, the nature of the empirical evidence, means that I can only say that multiple independent variables are significant, but cannot provide a specific coefficient in terms of how much. A more precise evaluation of the relative weighting of these factors must be done with more precise forms of quantitative data. Nevertheless, the overall picture that emerges, is one of contested, interactive, and iterative gradual change-one where the power of the state is still dominant but not hegemonic-and state actions are but a heavily weighted variable amongst many that determine the behavior of journalists over time.
Before developing the evolutionary narratives below, it should be noted that I focus on the traditional print press rather than broadcasting or new media for theoretical reasons. In terms of political control, the print press occupies a modal position between the tightly controlled broadcast media and the less tightly controlled new media. In terms of political influence, the print press is not as important as broadcast, but generally more widely influential than new media. Important enough to warrant direct political oversight, the print media meets the ontological conditions of layering and variation that are suitable for studying decisions that could have systemic institutional impact. Unlike broadcast, its commercial nature represents the friction between political control, economic reform, and social selection within authoritative institutions that is a good test of alternative mechanisms of change. As a user-driven medium, I view new media as qualitatively different than the traditional press and thus outside the scope of this study. Most of the literature indicates that traditional media it's still the primary agenda setter, and one could also argue that the variation in traditional media practices are in part driven by be growth, competition with, and increased agenda setting power of online media. Analysis of how many of the arguments highlighted here Btravel^to explain new media content are obviously areas of increased scholarly attention and further study.

Ontological Prerequisites: Institutional Layering and Friction
In the post-1978 reform era the party-state has faced a dilemma: how to build an information-driven market economy while maintaining control of public opinion. Consistent with the overall process of economic reform, the primary solution has been  Blayering.^The state layered market-driven mandates on top of existing censorshiporiented institutions. The propaganda department continues to issue press directives at all levels of government, although there is no pre-publication censorship. Control is delegated to news organizations, which are required to have state-affiliated ownership, and the key agents-chief editors-are all party members. In these aspects, the institutions of press control have remained constant. At the same time, starting in the late 1980s, the government began to emphasize Bcomplete commercialization,^and eliminated most direct state subsidies to all but a few key firms. 2 The literature on the reform era press has discussed how commercialization has generated a high degree of Bfriction^within the system, as chief editors now must cope with Btwo masters^-the regime and the market [23].
From an evolutionary perspective, layering and friction have important implications for institutional change, as friction generated structural incentives for innovation. This increased the degree of agent variation within the system as news organizations sought to develop commercially successful news content acceptable to the regime. This variation is a key condition for accelerating institutional evolution, as there are more policy ideas-or journalistic models-on which selection mechanisms operate.

Stage I: Agent Variation: The Emergence of Multiple Ideas
Evolutionary biology assumes agent heterogeneity invariably generates new ideas, but the extent to which mutations have an evolutionary impact depends in part on an agent's position within the system. Conversion occurs most often amongst rule-takers tasked with policy implementation. Even the most powerful dictators must rely on principled agents within the main levers of control-police, military, judiciary, and media. China's multilayered and decentralized control structure means there is Bdiscretion^for local agents to subtly convert institutions. In the Chinese politics literature, this type of opportunity has long been conceptualized as Bfragmented authoritarianism^ [25]. Fragmented authoritarianism remains a durable heuristic, because it Basserts that policy made at the center becomes increasingly malleable to the parochial organizational and political goals of various vertical agencies and spatial regions charged with enforcing that policy^( [30]: 996). . Friction between conflicting mandates in the press created a variety of ideas and experimental models. Key areas of agent variation lie in the ownership of media organizations and the role of chief-editor, tasked with placating multiple constituencies-the state, firm employees, and the market. Generally editors vary on the relative weighting they give to political career, group profits, or broader conceptions of national institutional reform. These preferences have an organizational impact throughout the company. In general conservatives prefer to think of the press as a mechanism for state control, propaganda, and profit. The goals are essentially top-down in seeking to reinforce state control through Bpublic opinion channeling.^In contrast, liberal reformists tend to frame the press as an accountability mechanism-the Bthird branch^of government. They adopt a more bottom-up perspective that views media's roles as reflecting the interests of the population and providing a mechanism of Bpublic opinion supervision.Ŵ hile these dichotomous labels demonstrate fundamental differences in Bideas^that motivates organizations, they only begin to capture the variation in interpretations and strategies. Based on more than 100 fieldwork interviews (see online appendix Table A for complete list), and an original content analysis of news editorials, I have inductively identified five different Bideal type^models of press behavior that vary in the degree to which they seek to convert press institutions. These models are summarized in Fig. 1.
On the conservative end, mouthpiece newspapers, or Bparty papers^represent the most traditional form of authoritarian journalism, providing propaganda directly from official speeches and talking points. So orchestrated that many will run the exact same picture and headline on a given day, evidence indicates that these papers are not profitable, relying on forced subscriptions, financial transfers from commercially successful publications within the same conglomerate, or state subsidies. This traditional model is designed to promote risk-averse and politically orchestrated behavior. As the editor of one national party paper stated, their publication was not successful in appealing to its audience, in part because the chief editor cared mostly about his position within the party. 3 By all accounts, the party paper model is increasingly irrelevant to the modern public sphere as consumers now have more credible choices.
The popular failure of traditional propaganda journalism has prompted strategies designed to Bfix^the propaganda system. This has lead to at least two distinct conservative models: Bparty minimalism^and Brisk-averse sensationalism.^Party minimalism describes a model that does the absolute minimum to placate authorities, while shifting rhetorical strategies away from the stultified, esoteric and formulaic language of the CCP. One of the goals is to make political coverage more understandable to the general public. As the chief-editor of one local party paper noted, he does the minimum amount possible to stay in compliance with the guidelines of the propaganda department. BOf course if the propaganda authorities call me directly with specific instructions, I listen,^he said, Bbut otherwise I focus on trying to improve the appeal of the paper to average citizens.^. 4 Risk-averse sensationalism is the most conservative model of journalism that tries to significantly update and gradually convert state-controlled media organizations toward a more popular orientation. Essentially eschewing political risk, this model combines four elements: avoidance of politically sensitive topics, greater professionalism in terms of area specific knowledge, tabloid sensationalism, and Bservice^news. In many ways this model illustrates the most pervasive account of the Chinese press outlined by Stockmann [41]-politically reliable and commercially successful journalism.
On the more liberal side of the spectrum, there are at least two distinct models seeking to gradually convert media institutions toward a greater public accountability.
The first is what I call Bshielded professionalism.^This strategy seeks to find a middle ground between political sensitivities and a desire for more professional reporting in the western sense of critical public deliberation on political institutions. Concurrently, these publications generally avoid political confrontation, often using the regime's rhetoricor that of selected leaders-as a rhetorical shield and framing their arguments as consistent the state's goals as they urge change. They represent a nuanced form of critical public deliberation that looks for the spaces within the regime within which serious issues can be discussed, even if in a tangential or implied fashion.
The final and most important form of liberal journalism is Btransgressive professionalism.^This model more vigorously pursues institutional conversion toward an independent press. Indeed, many of the editors associated with the groundbreaking Southern Media Group (SMG) conceived of their activities as trying to create a new form of journalism in the PRC. According to New York Times journalist Howard French [10], BChinese journalists say that more than any other publications, Southern Metropolitan Daily and its sister magazine, Southern Weekend. . .have been responsible for reinventing the Chinese press, by creating a journalism that even within the constraints of state ownership and censorship pushes for the truth.^This model is characterized by an overall effort to introduce the concept of societal feedback into reporting. Commentator Mo Zhixu reduces the model of the Southern Media Group to two dimensions: the promotion of political reform and responsiveness to the rights demands of the emergent middle class [42]. These organizations are less shielded, more Bbold^in their coverage of sensitive topics, willing to directly critique existing policy, and pioneers in investigative reporting. The comments of one veteran editor conceptualized the riskacceptance of these publications clearly: propaganda authorities do not say Byes^or Bno,^but instead communicate vague parameters of Bmaybe yes^and Bmaybe no.Ŵ here many publications self-censor to avoid sensitive topics, journalists here are riskacceptant in assuming the government said Bmaybe yes^until they are told explicitly not to publish a particular story. 5 These practices illustrate the discretion of transgressive publications to interpret formal rules in ways that facilitate conversion.
To provide evidence of systematic variation across news organizations, I draw on an original content analysis of news editorials from five commercially successful newspapers in three of the largest media markets: Beijing, Guangdong, and Jiangsu provinces. The coding system for this project was developed from 2006 to 2008 with graduate students in the PRC, then finalized and implemented by Chinese graduate students at the University of Colorado. Throughout I employed multiple coders as recommended by Krippendorff [20], so that results could be compared to ensure intercoder reliability, calculated as Cohen's Kappa. This analysis measures editorial content in two ways: first, by identifying the topic that is being discussed in the editorial. A list of these topics can be found in Table 2.
Secondly, I counted the number of rhetorical devices throughout each editorial. A list of the rhetorical devices and their operational definitions can be found in Table 3. Throughout the study, I employed multiple coders, so that results could be compared to ensure inter-coder reliability, calculated as Cohen's Kappa in the last column of Table 3.
All results reported are from a random sample of editorials from each paper from the first half of 2008. A list of the newspapers, organizational background, and sample size are outlined in Table 4.
I operationalize the concept of political risk in media coverage based on topic. Topics in the PRC are generally considered to be Bsensitive^-meaning highly 5 Interview with veteran editor of national publication, Beijng, June 18, 2013.

Party Propaganda
Risk-averse Sensationalism Transgressive Professionalism Party Minimalism Shielded Professionalism CONTROL Fig. 1 Spectrum of media models in the PRC

56
Risk Averse Sensationalism censored-and Bnon-sensitive^-indicating general openness for media discretion. Fieldwork interviews in these markets generated prior expectations for each newspaper's likely outcome. At the provincial regulatory level, interviewees considered Jiangsu to the most politically conservative media environment, followed by Beijing and Guangdong. Figure 2 presents the proportion of coverage on politically sensitive topics from the five newspapers. Consistent with expectations, the Jiangsu paper in this study, Modern Express (ME), displays the lowest proportion of risky editorials, followed by the Guangzhou Daily (GD), a party paper associated with the Guangzhou city government. Unexpectedly, Beijing Youth Daily (BY), considered to be an overly commercial outlet, displayed the highest proportion of risky editorials by a wide margin in the total sample. Despite firm-level variation, overall political risk acceptance varies in ways generally consistent with prior expectations based on local political environment and organizational reputation. Finally, I operationalize Btransgressive professionalism^as an index of four rhetorical strategies: editorial discussion of important policy topics, balanced evidence and support, a critical tone, and a strong push for policy change. One should expect to find this model in more open political environments, or where there are particularly riskacceptant preferences at a publication. Thus, I expect papers in Guangdong to have a higher proportion of transggressive content due to more open political environment, and The South Metro Daily (SM) to have the highest proportion based on the liberal preferences of the SMG group outlined above.
Variation amongst organizations with respect to transggressive editorials is reported in Fig. 3. Overall, as expected, Beijing and Guangdong papers have a higher percentage of transggressive editorials than the Jiangsu paper, ME, which displays the lowest proportion. While region explains broad trends in the data, there is also clear variation across organizations based on their ownership's preferences. For example, relatively liberal SM displays a significantly larger proportion than the more conservative Guangzhou Daily (GD). Similar differences exist within the Beijing market, with BY publishing a surprisingly high proportion of transggressive content, while BN-considered to be the most politically liberal paper-unexpectedly displaying a lower proportion, albeit the third highest proportion in the sample. These findings demonstrate two important points. Empirically, there is significant variation amongst news organizations in China, due to regulatory fragmentation and agency at the firm level. Both local regulators and publications have agency in how they interpret and implement abstract rules. Secondly, journalistic models vary systematically in ways that cannot be reduced to structural causes alone. Neither overarching regime censorship, nor the variation in provincial regulation of the press can fully predict outcomes. Agent preferences at the firm level matter. These results illustrate that agent variation is real and has important implications on how the overall press system evolves. However, in order to understand how agency drives institutional evolution we must look at its interaction with mechanisms of selection and replication.

Stage II: Political and Economic Selection Mechanisms
The layering of the market on top of political control institutions has led to multiple selection mechanisms impacting the viability and replication of different media models. While much of the discourse of Chinese politics pundits focuses on the ebbs and flows the Bpolitical environment^at any particular time (i.e., degree of repression), the untold story is that of broader structural factors driving selection and the evolution of the press over the longer term. The most important of these are commercialization and horizontal learning amongst media professionals. Unless the state decides the rollback the commercial incentives of news organizations and a general mandate for professionalization, these mechanisms will continue to drive long-term gradual change in the press environment.
Vertically, one can distinguish between top-down selection pressures from the state, and the bottom-up selection pressures of the market. Most standard accounts place primary importance on political selection. The propaganda department is both secretive and explicitly designed to censor content. Additionally, there are up to 14 regulatory institutions that enforce media management. While political reform overall has been characterized by government downsizing, Bthe same period has seen a steady increase in the number of government departments and bureaus in the communication field( While it is tempting to reduce media evolution in China to shifts in party-state policy, there are informal constraints on state repression. These stem from three sources: factionalism within the CCP, friction in the state's policy goals, and societal backlash from repression. Thus, while the regime has the coercive capacity to shutter news organizations, this is an increasingly rare tactic. The last major closure for political transgressions occurred in 2003, when a new SMG paper, The twenty-first Century World Herald, was shut down after it published calls for political reforms and information about internal CCP decisions that had not been publically announced. 6 In an information age, the history of overarching repression of media organizations has generally not reflected well on the regime, often creating social backlash. Consequently, repression has become more discriminant, focused on managing the editorships or personnel of organizations. This achieves the signaling purposes of repression, while avoiding negative public backlash.
Indiscriminant repression is constrained in part by elite factionalism within the party. As detailed below, many transgressive publications have implicit or explicit backing from elites with preferences for more transparent information. Thus factionalism at the elite level filters throughout the system and helps to create some of the structural spaces for agent discretion. Friction in policy objectives can have a similar effect on repression, when the state also values economic revenues, employment, and social stability. Liberal publications are often prominent brand names, profitable revenue streams, and employing large numbers of professionals. This further explains why the regime replaces the editor but preserves the brand. For example, Southern Weekend has seen numerous editors fired over the years, but has nevertheless become one of the bestknown brands in the country. As one chief editor argued, B[the authorities] might get rid of me, but they would not close down the entire organization, because it would put too many people out of work.^7 Employment matters for both its revenue and stabilitygenerating impact.
Social stability also impacts the regime's media management via restive online discourse and real-world social mobilization. Regime repression of the press in the past has caused significant societal Bbackfire.^For example, when liberal editor Li Datong was pushed out of his position at Freezing Point, a popular supplement to the official China Youth Daily, the public attention and outcry on social media seemed to encourage authorities to rethink their actions and re-open the paper under new leadership [24]. These challenges were apparent during protests stemming from censorship of SMG's Southern Weekend. Newly installed propaganda authorities in Guangdong province censored their traditional new year greeting, which called for the Bdream of constitutionalism.^As Gao [13] recounts: Bcrowds of protesters gather[ed] in front of the headquarters of Southern Weekend.. . carried posters calling for freedom of the press.. . Soon, the supporters were engaged in a battle of words and symbols with a group of counter-protesters,. .. [who] cursed their opponents as 'traitors to the nation.'T he threat of social unrest is not a Bhard^constraint on the ability of the state to repress, but it does impact decision-making, as the state cannot completely block these actions from entering the online public sphere. Taken together these factors explain why repression has tactically become more discriminant and subtle over time.
To be clear, top-down regime selection is clearly of primary importance and I take no issue with a general emphasis on this. In cases of high profile censorship of media leadership in organizations, such as Freezing Point and Caijing, state policy clearly has an impact on the journalistic norms and behaviors of those organizations. The widespread Bchill^that has come over public discourse and media freedoms in China under the Xi Jinping administration clearly has an impact in altering journalist calculus about what is acceptable. The broader analytical point from a perspective on evolutionary institutional change however is that political selection is often viewed as overlydeterminative, at the expense of other factors that are also significant and driving change over the long term. In other words the entire system cannot be reduced to the preferences of propaganda elites for political control. Other mechanisms also drive the selection and replication of ideas of media practice, such as economic competition, and the culture and norms of journalists themselves. Any evidence that the censorship regime is re-acting to stories-as we can see with numerous Internet driven case studies-or the regime is re-imposing constraints on media organizations, as I highlight below, should be seen as prima facie evidence that other structural factors in addition to the state are shifting how journalists behave.
Particularly salient in this regard are market selection and the horizontal selection of norms of media professionalism. Operating vertically in a bottom-up fashion, the marketplace influences which models are politically and socially valuable from a revenue and employment perspective. Commercial success determines how financially viable a particular model is and provides a degree of protection from repression, as there are vested political interests behind each successful publication. From a societal perspective, the past decades of market reform in China reveals that society prefers Bsensationalism^and Bprofessionalism^over propaganda. Indeed this is a familiar pattern throughout market driven media systems worldwide. In China, the Btransggressive^Southern Metro Daily is also characterized by a high degree of sensational content. 8 On the more liberal end of the spectrum, the economic profitability of publications such as Southern Weekend, Economic Observer, Caijing-formerly BThe Economist of China^-and now Caixin, demonstrates that a significant portion educated elites and urban middle classes demand professional news that conforms more closely to ideals of investigative information, supervision by public opinion, and rational-critical discourse.
While economic selection has been emphasized in the literature, a key, often overlooked, mechanism of authoritarian change is the role of horizontal selection by professional communities. In the PRC, this form of learning and selection takes place at multiple levels: ranging from global to local interactions. Media globalization and new media provide professionals with direct exposure to international models of journalism. Similar to the mandate of commercial success, the regime's mandate for professionalization has contributed to earnest engagement and learning from foreign models. Officials overseeing curriculum in journalism schools indicate that about half of their training focuses on western theories of the press and media practice. 9 Additionally, local learning and selection have also had a broad evolutionary impact. The emergence of transggressive professionalism in the Guangdong press provides the best example, where horizontal selection has worked in conjunction with a uniquely open regulatory environment. Provincial authorities in Guangdong have been instrumental in altering political selection pressures and providing a Bprotective umbrella^to the provincial media-defending their brands against political repression from higher authorities. This structural opportunity has allowed news organizations to learn from their liberal neighbors in Hong Kong and pioneer models of transggressive professionalism. 10 Easy access to Hong Kong media and inter-personal interaction with Hong Kong journalists during press conferences provided an alternative, more open, model from which mainland journalists learned directly. Finally, broad public access to Hong Kong news sources creates an additional bottom-up societal demand for more open coverage in the Guangdong press in order to avoid a public credibility gap. 11 In all of these ways, horizontal learning has contributed to the selection of more transggressive forms of journalism. Indeed the Guangdong papers in the sample, SM and GD, both display the lowest proportion of Bshielded^editorials, indicating that the provincial political selection, vested economic interests, and horizontal learning has driven qualitatively different media outcomes. In order to fully understand how transggressive journalism has emerged to have a broad systemic-indeed evolutionary-impact on the public sphere, we must understand the means by which this policy idea has replicated and how political, economic, and social selection all play a role in shaping its eventual trajectory over time.

Stage III: Replication: Why Political Selection is a Necessary but Insufficient Explanation
The history of the transggressive model of journalism illustrates how ideas shape institutional replication and the process of media evolution in China. More than any other, this model emphasizes political risk-acceptance, institutional conversion, and challenges the conservative preferences of the regime. Thus its survival and replication represents a Bhard test case^for institutional evolution. The systemic proliferation of this model over recent decades highlights two findings about media evolution in China: first, political selection does prevent the horizontal spread of ideas. There is a key disjuncture between the principal Bobject of selection^of autocrats-people-and the mechanisms that drive institutional replication and change-schemas and ideas. While the state can remove key individuals it cannot completely control how other professionals in the firm with common preferences think. 12 There is also a temporal dynamic to repression in that it must be forceful and consistent to be effective. Even if it is effective in discouraging certain behaviors at time t, latent preferences may re-emerge at time t + 1 when environmental conditions have shifted. This is why Bcrackdowns^in China often come in cycles. Repeated attempts to roll back aggressive professionalism have not prevented its continued development and replication. If anything, recent interviews with top editors find that the idea has continued to proliferate and exert influence in many publications, perhaps more so that at any time previously. 13 The chill brought about by President Xi Jinping's anti-corruption drive since 2013 inevitability discourages many from acting on latent preferences in the short term, but has not fundamentally altered the factors driving a more professional press environment in the long run.
The history of transggressive professionalism highlights why a view focused on political selection is overly deterministic. Despite the fact that central propaganda authorities were constantly dissatisfied with southern papers, 14 models developed locally in Guangdong and within a few key firms came to have national prominence and profitability, encouraging their adoption elsewhere in places such as Beijing. The Beijing News established in the mid-2000s represented the hybrid of the ethos of the Southern Media Group and state control, as a joint venture between SMG and the national Guangming Daily. The expansion of the idea was fraught, as the paper repeatedly ran into conflicts with the authorities and had numerous editors fired. Nevertheless it continues to exist, albeit in a more conservative form.
Caijing (财经) also illustrates this interaction of vertical and horizontal selection pressures and how they drive the replication of ideas. Known as The Economist of China, Caijing built a brand based on in-depth investigative reporting, professional economic news, and rational-critical discourse. As the kept pushing the boundariesfor example reporting on school collapses during the 2008 Sichuan earthquakepropaganda authorities became increasingly concerned. The tipping point came when the magazine engaged in sensitive coverage of ethnic riots in western Xinjiang province in 2009. Informal constraints on political repression were readily apparent: from a market standpoint, Caijing was very profitable, accounting for around 50% of the total revenue of its parent company [2]. Interviews with employees indicated that up to 20% of their readership were government officials, elites who also had a preference for more objective news. 15 It had the backing of Wang Qishan, now the top leader responsible for the deepest anti-corruption campaign in recent years. Additionally, chief editor Hu Shuli was internationally famous and named one of Time magazine's B10 most important women in Asia.N onetheless, political pressure built on the parent company, the Stock Exchange Executive Council, which had been instrumental in protecting the magazine's editorial freedom and risk acceptance. Instead of shuttering the magazine or firing Hu, the state created backdoor pressure, prompting Hu and her entire senior staff resign en masse. They immediately announced their intentions to start a new publication under the operating umbrella of the Zhejiang provincial government, Caixin, which continues to promote ideas of transgressive journalism today [2]. Caijing still maintains a reputation for professional economics reporting, while Hu effectively replicated her model within a new publication. The story of Caixinindeed its very existence-does not make sense within a purely top-down account of elite political selection. If anything it represents the proliferation of the transggressive professional model rather than a reversal of it. One can certainly argue that government censorship had an impact in this case, and that Caixin is not the same as Caijing used to be. Indeed interviews with journalist at similar organizations, such as Freezing Point, indicate that there is indeed an important impact that such censorship has. The censorship of effects maybe durable for sometime, but they are not necessarily determinative, and they do not account for that broader impact that norms of transgressive journalism or repression might have on others. It is very difficult to measure this sort of thing, without more comprehensive and systematic survey data. Again, the basic point is that repression matters, but unless repression is able to prevent the replication of journalistic norms, or outweigh the significance other competitive pressures. Despite repeated cycles of repression and censorship against transgressive journalist, the size and significance of this sector of the Chinese media landscape has continue to exist proliferate over time-something in direct opposition to state preferences. Thus censorship cannot be a determinative factor, so understanding the interaction and combined effects of political, economic, and social selection is the only way to explain changes in journalistic practices over this time period.
This narrative thus highlights three mechanisms by which professional ideas have proliferated under conditions of political control. First, ideas are replicated within organizations and continue to exist even if the leadership is removed. Economists talk of corporate culture, and media companies are often no different. This explains why many transggressive publications in China been punished repeatedly. Secondly, economically successful models-even of the liberal variety-are likely to be copied elsewhere in some form by profit seekers. The story of The Beijing News and Caixin reflects this pattern. Political decentralization provides numerous opportunities for horizontal selection and replication, as there will likely always be relatively risk-acceptant government agencies sympathetic to preferences for critical discourse and willing to take a chance for the sake of profit. Thirdly, even if key leaders are repressed, their ideas can remain and continue to replicate virally. Recent interviews support the finding that despite recent government repression of the press, the idea of transggressive professionalism continues to proliferate and motivate new journalistic behavior. Thus political, economic, and social selection, converge to explain why the period since 2008 has been characterized by both more open reporting and more frequent conflicts with the regulatory authorities. 16 An evolutionary perspective on institutional change thus provides the basis for understanding unexplained mechanisms of change that add to standard top-down models of authoritarian change. Clearly, power matters for encouraging the replication of ideas and discouraging others, but unless a regime goes to totalitarian ends to control people and ideas, there is often some space for agent discretion. China's fragmented decentralized system of control premised on principled agents creates space for discretion at a cognitive and behavioral level. The existence, survival, and proliferation of transggressive professionalism in the context of repression highlight how change can still occur under continuity in political control. Political control is not the same as political stasis.

Conclusions: Agency and Authoritarian Evolution
Overall, the evolution of the news media in China over the past twenty-five years provides a number of insights into the mechanisms of endogenous institutional change in authoritarian regimes. First and foremost, a cognitive ontology illustrates the extent to which there is space for agency even under conditions of authoritarianism. Existing research on authoritarian institutional change and media in China are largely stories of top-down political change at the elite level. This analysis agrees with this basic story, but highlights how it is also overly reductionist. Authoritarian institutions can be more complex, particularly in mature, institutionally layered regimes. This paper illustrates that targeted horizontal selection of ideas across professional organizations can have an important evolutionary impact even when institutions of control and stability seem constant. Instead of reducing institutional change to a small group of elites, this approach explains how fragmentation, agent variation, innovation, land earning generate multiple mechanisms of change at the micro-level.
In China censorship policy is rightly placed as the key indicator of institutional change in the press. Instead viewing this as the product of pro-active elite control, I argue it is often difficult to tell whether policy is the locomotive or caboose of institutional change (see also, [21]). An adequate explanation of institutional change requires additional mechanisms and more precise understandings of event sequencing. Whether a regime is proactive in promoting policy change, reactive in trying to slow it down, or selective in codifying existing informal behaviors matters greatly for understanding the extent to which elite bargaining is the primary cause of change.
To be clear, I do not argue that we are witnessing anything approximating a liberal press in the western sense. Political selection matters more than anything else and this is not in dispute. The regime is very successful in controlling vast swaths of the public sphere in China, and enjoys a high degree of legitimacy due in part to information control. At the same time, political control is not the same as political uniformity or political stasis, and it does not preclude the emergence and evolution of more professional modes of journalism. This is the key analytic take-away-endogenous change can occur even when it appears that overarching institutions are stable. Elite bargaining sets the agenda, but that agenda is influence by bottom-up and horizontal dynamics as well. Moreover in institutionally decentralized and complex regimes, the elite agenda can diffuse through the system in a variety of ways. In China, professional journalistic models have emerged and survived because there are elite allies protecting them and ideas supporting their replication. This highlights how political factionalism might have more long-term consequences than bargaining results at any particular critical juncture.
While one might assume that authoritarian rollback is always possible [8], this process is far more difficult in an information age than is often assumed. Gorbachev tried to rollback glasnost, but was unable to once key elites and citizens had come to expect greater press freedoms [35]. This article argues that-similar to a reformist USSR-regime control in the PRC is constrained and shaped by factional and ildeational differences, as well as potential societal backlash. Bottom-up selection of the market and horizontal selection of ideas across organizations, all interact with political selection in an information age and provide important supplements to our understanding of what drives change in China.
Unless the regime is willing to rollback the fundamental institutional reforms that drive change, institutional evolution will continue to ebb and flow. Relatively riskacceptant models of journalism have replicated despite repeated cycles of repression over the past two decades, which explains why authorities have come in greater conflict with journalists. As one journalist noted, even Bif they close [SM] it wouldn't matter. . . The . . . example has already been absorbed by journalists all over China, and their goal is not just to copy it, but to do even better^ [10]. From an evolutionary perspective the existence of ideational variation and expansion of communications increases the interactivity of the system and the likelihood of new mutations. For example, some feel that competitive pressures have changed practices even in heavily controlled organizations. As one veteran journalist noted, even journalists in the heavily statecontrolled Xinhua News Agency want to be known as doing good Bprofessional^work, not propaganda. 17 While critics may question the qualitative evidence offered herein, new public opinion research supports the view that the media landscape is more pluralized, and the diversification of the media sector is an important explanation for why Chinese society is more ideologically factionalized. In a recent survey of Chinese netizens, Ma [27] finds that one's media consumption is significantly correlated with political ideology-with those consuming the more liberal sources more likely to support democracy and a less hawkish foreign policy. Thus triangulating between interviews, content analysis, and survey evidence gives one greater confidence that the narrative told herein is valid.
This research has some important findings for comparative authoritarianism. As with institutional analysis more broadly, authoritarian institutions may not be as Bsticky^as they appear on the surface. The general disciplinary bias toward stasis is particularly acute with authoritarian systems, as change may be more difficult to perceive and study. There is an inherent bias toward elite power, which is understandable, but also indeterminate. Overall stability may mask important changes under the surface. Secondly, this article has argued for greater attention to the interactive dynamics between principles and agents. As with the literatures on institutionalism or contentious politics, comparative authoritarianism can benefit from more interactive and dynamic theoretical frameworks that better capture state-society relations, their diversity, and the alternative mechanisms of stability and change. Evolutionary theories on cognition and change provide one avenue, but much more can be done to theorize agent decisionmaking under conditions of power asymmetry.