Authority, Incentives and Performance: Theory and Evidence from a Chinese Newspaper
Yanhui Wu London School of Economics Job Market Paper February 2011

Abstract Authority de…nes the formal structure of an organization, and is essential for the allocation of resources inside the …rm. This paper develops a theory of authority in a multiple layer hierarchy, in which the distribution of authority alleviates incentive incompatibilities. To examine the theory, I collect monthly personnel data from about 200 journalists over three years in a Chinese newspaper, and provide evidence on their incentives and performance under two basic organizational forms — centralization and decentralization. Relying on an unexpected organizational reform from decentralizing to centralizing editorial power in some divisions of the newspaper, I …nd three main results: 1) centralization improves the quality of the journalists’performance, in terms of the newspaper’ inters nal assessment and the external measures of news content; 2) centralization reduces the journalists’activities for private gain; 3) centralization decreases the editorial activities conducted by managing editors. These results are in line with the theory: a more centralized hierarchy achieves better control over workers’ opportunistic behaviour, at the cost of depressing middle managers’initiative. Key Words: Authority, Organizational Structure, Incentives, Information, Action Distortion, Decision Bias, Media Bias JEL Classi…cations: D2 J5 L2 M5

Economics Department and Centre for Economic Performance, London School of Economics, Houghton Street, London WC2A 2AE, UK. Email: y.wu8@lse.ac.uk. I am extremely grateful to Oriana Bandiera, Luis Garicano, Stephen Redding and Daniel Sturm for their encouragement and support throughout this project. I would also like to thank Marianne Bertrand, Jordi Blanes i Vidal, Robin Burgess, Ben Farber, Robert Gibbons, Claudia Goldin, Chang-Tai Hsieh, Alexander Lembcke, Gerard Padro I Miquel, Ignacio Palacios-Huerta, Steve Pischke, Gerard Roland, Yona Rubinstein, Antoinette Schoar, Matthew Skellern, John Van Reenen, Ph.D. students of the organizational economics reading group at the LSE, and seminar participants at the China Summer Institute, the Econometric Society European Winter Meeting (2010), LSE, and the Organizational Economics Lunch at MIT for valuable comments and suggestions. I thank all those involved in data collection. This paper has been screened to ensure that no con…dential information is revealed.

1

Introduction

Authority, the power of a superior to select actions or decisions for her subordinates, is the core of hierarchy. Authority de…nes the boundary of the …rm (Coase 1937), the nature of an employment contract (Simon 1951), and the structure of an organization (Weber 1922[1968], Simon 1947, Arrow 1974). Recent empirical research strongly supports the relevance of the internal allocation of authority to economic performance at the …rm level, echoing the insights from the history of industrial enterprise (Chandler 1962) and from modern business strategies (Roberts 2004, Besanko et al 2010).1 Despite the burgeoning interest in studying authority, we lack a rigorous empirical understanding of how its allocation a¤ects workers’performance. In this paper, I collect monthly personnel data from about 200 reporters from 2004 to 2006 in a leading commercial Chinese newspaper (the Newspaper hereafter) that is funded by advertising revenues. I estimate causal e¤ects of the distribution of authority on individual performance, relying on an unexpected organizational reform from decentralizing to centralizing editorial power in some divisions of the Newspaper. In line with the view that regards authority as a device to moderate incentive incompatibilities and control opportunistic behavior, the empirical …ndings shed new light on a basic question about hierarchy: how does organizational structure, de…ned by the distribution of authority, a¤ect workers’ incentives to allocate resources, and the resulting economic outcome?
2

The essence of the incentive view of authority lies in the separation of formal authority (nominal control rights) and real authority (e¤ective control). The distribution of formal authority a¤ects agents’ performance by changing their incentives to obtain resources that permit real authority. An in‡ uential theory along this line is that of Aghion and Tirole (1997), who formalize the idea that the distribution of real authority is determined by the information structure, which in turn depends on the contractible arrangement of formal authority. Based on the Aghion and Tirole framework, I build a theoretical model of authority, in which agents have dual agency problems: 1) action distortion caused by the distraction of private activities, and 2) decision bias due to interest misalignment in project selection. Centralization, under which a principal retains both the right to overrule an agent’ decisions and s the right to direct an agent’ actions, exerts two opposite forces on an agent’ incentives: an s s e¤ort-directing e¤ect due to alleviation of his action distortion and an initiative-depressing
For recent empirical studies on organizational structure and performance, see Rajan and Wulf (2006), Acemoglu et al (2007), Bloom and Van Reenen (2007), Csaszar (2008), Bloom, Sadun and Van Reenen (2009), and Guadalupe and Wulf (2010). 2 The incentives view of authority is discussed in Weber (1922[1968]) and Williamson (1985). For recent theoretical inquiries along this line, see Qian (1994), Aghion and Tirole (1997), Baker, Gibbons and Murphy (1999), Rajan and Zingales (1998, 2001), Prendergast (1995, 2002), Zabojnik (2002), Marino and Matsusaka (2005), and Friebel and Raith (2010), among many others. Mookherjee (2006) surveys the literature on organizational structure and incentives from the perspective of mechanism design. Another in‡ uencial view of hierarchy emphasizes authority as a device to coordinate behavior, beliefs and decisions among agents; recent theoretical contributions include Radner (1993), Bolton and Dewatripont (1994), van Zandt (1999), Garicano (2000), Dessein (2002), Alonso, Dessein and Matouschek (2008), Rantakari (2008), and Van den Steen (2010), among others.
1

1

e¤ect due to control of his decision bias. I extend the analysis to a three-layer, principalmanager-worker, hierarchy. A change of organizational structure triggers a chain of responses. More control at the top may depress the manager’ initiative, which in turn promotes the s initiative of her subordinate — the worker. Thus, the impact of organizational structure on the agents’incentives and performance crucially depends on two factors: 1) the nature of the agency problems, and 2) the agents’relative positions along the line of formal authority. For example, when the worker’ action distortion is more serious than his decision bias, and cons versely the manager’ decision bias is more serious than her action distortion, centralization s outperforms decentralization in inducing the worker’ initiative, at the cost of depressing the s manager’ s. The institutional setting of the Newspaper provides a rare opportunity to examine the theory. Making editorial decisions regarding the choice of subjects and the selection of articles is the key task in the production of news content. Information, the essential input, determines the execution of editorial decisions. For example, a reporter, though supervised by an editor, often decides how to select and implement an investigative report, because he has more information than the editor. Therefore, the allocation of formal authority over editorial decisions a¤ects incentives by changing the distribution of information and thus real authority among chief editors, middle managers (division directors and managing editors), and reporters. Moreover, reporters have substantial discretion in their actions, and are likely to divert their e¤orts to pursue private bene…ts such as a "grey income" and business opportunities, which are potentially large in the Chinese media. With the commercialization of the Chinese media, the Newspaper experimented with a decentralized organization, in which editorial power was delegated to middle managers, in early 2000. In September 2005, the Newspaper unexpectedly centralized editorial power in four divisions: Economic and Business, Politics and Law, Education and Health, and General Reports, by creating an editing center headed by chief editors to monitor editorial decisions more closely, while leaving other divisions (Local and Regional News, Entertainment, Consumption-Guide, and Photographing) decentralized. The exogeneity of the reform timing, together with the adoption of di¤erent organizational forms inside the Newspaper, permits me to establish causality using a di¤erence-in-di¤erences estimator, whose validity is supported by the absence of a trend over a long period before the reform. Moreover, the stability of other institutional aspects such as the pay scheme, the evaluation system, and the volume of news content helps to relieve concerns about a number of potentially confounding factors. The empirical analysis draws on rich personnel information and performance measures. Exploiting the internal records from the Newspaper, I match the reporters’ personal characteristics to the monthly observations of their performance in both quantity and quality, which are accurately measured to serve as a basis for their pay. A team of research assistants coded the news content of all the articles written by each reporter based on publicly available

2

archives over the sample years, to directly measure the reporters’initiative and the middle managers’ editorial activities. These external measures are constructed under the close supervision of experts in content analysis and Chinese journalism, and provide a reliable data source. I employ the di¤erence-in-di¤erences approach to estimate the e¤ects of the organizational reform on the reporters’ internal performance measures. Three main …ndings are as follows. First, centralization on average improves the quality measure of the reporters’performance by 20%. Second, heterogeneous treatment e¤ects show that centralization has a larger positive e¤ect on the performance of those reporters who have access to more private bene…ts. For instance, the reporters specializing in economic and business coverage, who have more opportunities to obtain private bene…ts from companies, respond to the reform far more than those who report on public policies. Relative to the other months, the impact of centralization on the reporters’performance is much smaller in the special months of the Chinese New Year and the Mid Autumn Festival, when social norms condone rent seeking behavior. Third, the pattern of individual …xed e¤ects suggests that the reporters who leave the Newspaper after the reform are more likely to have misaligned interests with the Newspaper, relative to those who remain. The last two results strongly support the explanation that centralization redirects the reporters’ incentives from pursuing private bene…ts to desirable journalistic activities. I then examine the e¤ects of centralization on the external measures of news content and editorial activities. Centralization signi…cantly increases the number of investigative reports and feature stories, which require substantial initiative and endeavor from the reporters. Meanwhile, the attainment of private bene…ts, measured by the number of advertising-type articles authored by a reporter, declines drastically after the reform. This result veri…es the e¤ort-directing e¤ect. Moreover, centralization reduces the number of articles originated by , or coauthored with, managing editors, demonstrating initiative substitution between the reporters and the middle managers. Together with the previous results, the empirical …ndings are in line with the theory of authority and incentives that I develop. To my knowledge, this paper is the …rst empirical study that examines the basic theory that organizational structure a¤ects workers’performance through the redistribution of real authority and the resulting changes in their incentives to undertake ex-ante investments (notably, information acquisition). One particular contribution is to spell out that the e¤ects of organizational structure on agents’incentives rest on both the nature of agency problems and the strategic interactions among players at di¤erent positions in a hierarchy. This enriches the existing studies of authority and incentives, in the spirit of the subeconomy view of the …rm (Holmstrom and Milgrom 1994, Holmstrom 1999). My paper is related to the emerging literature that combines rigorous econometric methods with personnel data to achieve a more profound understanding of internal labor markets

3

and resource allocation mechanisms inside organizations.3 The existing studies cover a wide range of topics in labor economics and organizational economics.4 However, the literature provides little evidence on the e¤ects of organizational structure, and is silent on the role of authority. The current paper and my follow-up research are devoted to …lling this gap. The insight from the current research can be generalized to other organizations, as the Newspaper is, to a large extent, a pro…t maximizer. I will discuss the issue of external validity in the conclusion section. Additionally, my research improves the limited economic understanding of media bias. Although this paper does not de…ne and specify media bias, the signi…cant impact of the organizational reform on the composition of news content suggests that organizational structure can be a source of persistent media bias, as suggested by Herman and Chomsky (1998) in their study of journalism. This complements the existing explanations for media bias that focus on ownership (Djankov et al 2003, Besley and Prat 2006), and on consumer demand and market structure (George and Waldfogel 2003, Mullainathan and Shleifer 2005, Gentzkow and Shapiro 2006, 2010). The rest of the paper is organized as follows. The next section presents a simple theoretical model of authority and incentives. Section 3 describes the institutional setting and data. Section 4 explains the empirical strategies. Section 5 presents the main empirical results. Section 6 provides further evidence that sheds light on the mechanism and discriminates between potential alternative explanations. Section 7 concludes. Proofs and extended theoretical analysis, details about data collection, and additional empirical results are provided in a web appendix.5

2

A Theory of Authority and Incentives

This section presents a simple theory of authority in a principal-manager-worker hierarchy, to show the essential function of organizational structure, and to generate empirically testable predictions.6 As in Aghion and Tirole (1997), real authority (e¤ective control) is determined by the structure of information, which in turn depends on the distribution of formal authority (nominal control). In the presence of decision bias in selecting projects due to interest misalignment, a more centralized hierarchy restricts an agent’ real authority to ful…ll her s own preferences, giving rise to a trade-o¤ between better control and depressing initiative. I
The literature is referred to as personnel economics, or sometimes as insider econometrics. Baker et al (1994), Lazear (2000b), Lazear and Oyer (2009), and Ichniowski and Shaw (2009) provide excellent introduction and surveys. 4 For example, Lazear (2000a), Shearer (2004) and Bandiera et al (2007) study pay contracts and productivity; Ichniowski et al (1997) and Gant et al (2002) study human resource management and …rm performance; Liberti and Mian (2009) and Hertzberg et al (2010) study communication and incentives. 5 The appendix is available at http://personal.lse.ac.uk/wuy9/. 6 For this purpose, I abstract away from a number of institutional aspects, such as performance pay, career concerns and team work, which are either stable over the sample period or can be controlled for in the empirical analysis. In the web appendix, I discuss potential interfaces between organizational structure and these aspects.
3

4

depart from the Aghion and Tirole model in two dimensions. First, the person with formal authority has two rights: 1) directing the actions of a subordinate who diverts his e¤ort in information acquisition, and 2) overruling the decisions of a subordinate who selects a project in con‡ of her interest. Second, the introduction of multiple layers in a hierarchy ict generates a chain of responses, highlighting the importance of relative hierarchical positions in the provision of incentives.

2.1

The Model

An organization owned by a principal (chief editor, she) selects one project (a news report) to implement at a time. A manager (managing editor, she) and a worker (reporter, he) are employed to search for projects. The hierarchy is de…ned by the distribution of formal authority: the principal has formal authority over the manager, who in turn has formal authority over the worker. Projects. A variety of projects exist, each generating di¤erent values to each party. For instance, a chief editor, a managing editor and a reporter may have a di¤erent preference ordering of the following three types of reports: an investigative report, a sensational story, and an article about a government o¢ cial. The misalignment of interests can be due to di¤erent valuation of journalism, or due to non-veri…able on-the-job bene…ts.7 Information and authority. The selection of projects …rst of all depends on formal authority. The superior party decides which project to implement and has the right to overrule her subordinate’ decision. However, being able to make proper decisions requires information s about the projects. An uninformed principal will give the decision right to a manager, who then makes decisions if informed, but will delegate the decision right downwards to the worker if uninformed. The worker e¤ectively decides which project to implement whenever he has information advantages over his supervisors. Hence, what the allocation of formal authority de…nes is "the right to the last word," or the sequence of residual claimants of decision right along the hierarchy. Authority also permits a superior to direct the actions of her subordinate within a certain "acceptance area" (Simon 1951). Again, the realization of this aspect of authority requires information. A subordinate can freely allocate his e¤ort between production activities and private activities if his superior is ignorant. Contracts and organizational forms. In the spirit of the theory of incomplete contract a là Grossman and Hart (1986) and Hart and Moore (1990), the input and output of production are assumed to be observable but non-veri…able so that the contractible organizational structure plays a central role in inducing ex ante investments — the acquisition of information in this model. I focus on the choice between two basic organizational forms: decentralization and centralization. Under decentralization, the principal delegates formal authority to
7 The on-the-job bene…ts can be intrinsic motivation (e.g., job satisfaction from decision making), or personal bene…ts such as perks and potential career opportunities.

5

the manager, and commits not to monitor both agents’activities; under centralization, the principal retains formal authority— the right of directing decisions and actions. Timing of the game. At To , the three parties contract on one of the two organizational alternatives and agree on the allocation of formal authority. At T1 , the manager and the worker simultaneously and independently exert e¤orts to acquire information on the projects. At T2 , the agents propose their projects. Under decentralization, the manager selects among the projects. Under centralization, the principal decides the selection of project after she has acquired and reviewed the proposal by an informed manager or the worker’ proposal passed s on by an uninformed manager; alternatively, if an agent proposes no project, the principal can acquire information to direct the agent’ actions. At T3 , the selected project is implemented s without further costs, output of the organization is produced, and all the bene…ts are realized with no uncertainty. Figure 2 depicts the timing of the game. Agency problems. Two types of agency problems may arise in the production process. The …rst type is action distortion at T1 , when the agents divert their e¤orts to private activities. This is the classic moral hazard due to hidden action; it has also been labelled as rent seeking or shirking in the literature of transaction costs economics (Alchian and Demsetz 1972, Williamson 1975). The second type is decision bias at T2 , when the agents, after acquiring information, propose their preferred projects that are in con‡ with the principal’ ict s interest. This is distortion in decision making due to ex post information asymmetry, as highlighted by Aghion and Tirole (1997). to acquire information about the projects, and 1 project proposed by agent i delivers i Payo¤s. Let i 2 fm; wg denote the manager or the worker. Agent i expends e¤orts Ei

Ei to conduct private activities. Ei also

denotes the probability of agent i being informed of the projects. The implementation of a to the agent, and zero to the other agent.8 Thus a higher i 2 (0; 1) to the principal, one unit of on-the-job bene…t i is a congruence parameter measuring

the interest alignment between the principal and agent i in project selection. An agent with is more likely to select a project at the principal’ interest. Alternatively, when s conducting private activities, agent i obtains a non-veri…able bene…t bi 2 (0; 1), referred to as private bene…t. The realization of bi relies on the ignorance of agent i’ superiors, because an s informed superior can direct her subordinate to undertake her selected projects or actions.9 For simplicity, all the parties are assumed to be risk neutral. As performance is not contractible, the principal pays a …xed salary sm to the manager and sw to the worker, regardless of which project is implemented. All cost functions of e¤ort will take a quadratic form. Decentralization. Under decentralization, the principal commits not to intervene. The
I normalize the on-the-job bene…t to one, as what matters is its comparision with the private bene…t that will be discussed. I also assume that the implementation of one agent’ preferred project delivers zero s on-the-job bene…t to the other agent, to sharpen the con‡ between the agents. ict 9 Equivalently, one can think of the monitoring of private activities as protection of the assets of the organization, when the realization of private bene…ts diminishes resources of the organization.
8

6

selection of projects and the resulting payo¤s depend on the allocation of formal authority and the information distribution between the two agents.
D Up = Em m

+ (1

Em )Ew

w

D Um = sm + Em + (1 D Uw = sw + (1

Em )bm

Em )[Ew + (1

sm sw ; 1 2 E ; 2 m 1 2 Ew )bw ] E : 2 w

(1) (2)

With probability Em , the manager is informed and has real authority to select her preferred project, which yields probability 1 m to the principal and one to herself, but zero to the worker. With

Em , the manager is distracted by the private bene…t bm and delegates the w decision right to the worker; then the worker, with probability Ew , will select his preferred project that yields with probability 1 to the principal and one to himself, but zero to the manager, and, Ew , will realize the private bene…t bw .

Centralization. Under centralization, the principal can exercise his formal authority in two ways: she either monitors an agent’ proposed project, or controls the agent’ pursuit of s s private bene…t by directing his or her actions. First, if agent i has acquired information and i proposed a project, the principal, after spending an e¤ort Ep , is informed with probability i Ep , and able to modify the proposed project to obtain one unit of output; with probability i Ep , the principal is uninformed, and will rubber stamp the proposal. Second, if agent

1

i has conducted private activities and proposed no project, the principal can acquire information to direct the agent’ actions and achieve certain value. For simplicity, I assume that s centralization incurs a …xed cost F; and the principal becomes fully aware of agent i’ pris vate activities and is able to obtain a value i activities. In other words, the monitoring of private activities is perfect, and the agents can never realize their private bene…ts under centralization. In the web appendix, I allow con-

2 (0; 1) by directing the agent to production

tinuous and imperfect monitoring of the agents’ private activities under centralization; the main economic insights in the following analysis remain unchanged in this more general case. The payo¤s to the three parties are:
C Up

=

sm +(1

m sw + Em [Ep + (1 w Em )Ew [Ep + (1

m Ep ) w Ep ) w

m

1 m 2 (E ) ] 2 p

C Um C Uw

+ (1 Em )(1 1 2 m = sm + Em (1 Ep ) E ; 2 m 1 2 w = sw + (1 Em )Ew (1 Ep ) E : 2 w m F + (1

Em )

1 w 2 (E ) ] 2 p Ew ) w ; (3) (4)

Note that in the above speci…cation, an agent obtains zero on-the-job bene…t if his or her proposed project is overruled by the principal; an agent does not bene…t from the principal’ s

7

direction of actions. These assumptions can be relaxed easily.

2.2

Analysis

Solve the model by backward induction, and assume interior solutions throughout. Under decentralization, the …rst order conditions of (1) and (2) produce a pair of Nash equilibrium e¤orts of the manager and the worker:
D Em = 1 D bm ; Ew = bm (1

bw ):

Agent i’production initiative is motivated by the on-the-job bene…t, but diverted by the private bene…t bi . The organizational structure that endows the manager with formal authority over the worker depresses the worker’ initiative by a factor bm , which indicates her ignorance s due to the distraction of private bene…t. Under centralization, the principal’ optimal monitoring e¤ort is s i Ep = 1 i:

The principal monitors agent i’ proposed project to counter their selection distortion, the s severity of which is measured by the congruence parameter the Subgame-Perfect-Nash equilibrium:
C Em = m; C Ew = (1 m) w : i.

Anticipating the principal’ s

responses, the agents optimize their allocation of e¤orts according to (3) and (4), leading to

Note that the agents’private bene…t bi does not enter their optimal e¤orts, because the principal’ monitoring of the agents’private activities is assumed to be perfect. s 2.2.1 Trade-o¤ between Control and Initiative

An organizational change from decentralization to centralization yields two opposite e¤ects on each agent’ incentives. On the one hand, the monitoring of private activities controls s the realization of an agent’ private bene…t, and thus directs his or her e¤ort to production s activities. On the other hand, the principal’ monitoring of project selection restricts an s agent’ real authority to choose his or her preferred project, and thus depresses the agent’ s s initiative. Which e¤ect dominates depends on the relative severity of each agency problem. De…nition 1 Agent i is distracted if bi > 1 if 1 i i:

the distraction of the private bene…t is

large, relative to the interest misalignment in project selection; alternatively, agent i is biased > bi : the interest misalignment in project selection is large, relative to the distraction of the private bene…t.

8

The relative position of each agent in the hierarchy generates another trade-o¤: a decline (or an increase) in the manager’ initiative in turn promotes (or depresses) the worker’ s s initiative, resulting in subtle e¤ects on the worker’ incentives. s Proposition 1 (Average Treatment E¤ ect) The e¤ ect of organizational structure on the agents’incentives depends on the nature of their agency problems and their relative positions in the hierarchy, as follows: 1) (Biased Manager and Distracted Worker) Centralization, compared to decentralization, decreases the manager’ initiative, but increases the worker’ initiative. s s 2) (Distracted Manager and Biased Worker) Centralization, compared to decentralization, increases the manager’ initiative, but decreases the worker’ initiative. s s 3) (Biased Manager and Biased Worker) Centralization, compared to decentralization, decreases the manager’ initiative, but has ambiguous impact on the worker’ initiative. s s 4) (Distracted Manager and Distracted Worker) Centralization, compared to decentralization, increases the manager’ initiative, but has ambiguous impact on the worker’ s s initiative.

The impact of organizational structure on the manager’ incentives simply depends on her s type: biased or distracted. This result would also hold in a two-layer hierarchy. However, the impact on the worker’ incentives rests on both the type of manager and the preference match s between the two agents. In the …rst two cases of Proposition 1, the relative severity of agency problems with the manager is opposite to that with the worker. The e¤ect of organizational structure on the worker’ incentives is ampli…ed by the e¤ect on the manager’ and thus is s s, unambiguous. In the last two cases, both agents have the same dominant agency problems; the e¤ect on the worker is no longer clear-cut, because the manager’ initiative substitutes s the worker’ s. Proposition 1 provides guidance for optimal choice of organizational structure. For example, if it is more important to alleviate the manager’ bias in project selection and encourage s the worker to provide e¤ort as in Case 1), then centralization tends to outperform decentralization. But if it is more important to encourage the manager to provide e¤ort, and alleviate the worker’ bias in project selection as in Case 2), decentralization tends to outperform s centralization. More extensive analysis is relegated to the web appendix. 2.2.2 The E¤ort-Directing Mechanism

I have discussed two aspects of control under centralization. One is control of project selection, determined by an agent’ preference s
i.

The other is direction of actions by restricting an

agent’ pursuit of the private bene…t bi . Here, I stress the latter aspect of control, as the s access to private bene…t, to a large extent, depends on the agent’ job assignment and working s environment, and is easier to test empirically. 9

Proposition 2 (Heterogeneous Treatment E¤ ect) Consider the e¤ ect of an organizational change from decentralization to centralization on an agent’ initiative. s 1) The e¤ ect on the manager’ initiative increases in her access to private bene…t; s 2) The e¤ ect on the worker’ initiative increases in his access to private bene…t, and the s increase is enhanced by the manager’ access to private bene…t. s

The e¤ect on the manager only hinges on the interaction between two layers: the principal and the manager. Centralization has a larger e¤ort-directing e¤ect on agents who have more access to private bene…t and who allocate more e¤ort for private activities under decentralization. The second statement — the positive correlation between the e¤ect-directing e¤ect and the worker’ access to private bene…t — is true only if the principal’ monitoring of the s s worker’ private activities is su¢ ciently e¤ective, which is assumed in the current model. In s general, better control of the middle line does not necessarily imply better control at the bottom in a multi-layer hierarchy. For example, if the manager loses initiative under centralization, the worker will obtain greater freedom to allocate his e¤ort; a larger private bene…t may distract him further from production activities, if the monitoring of his private activities is not e¤ective. 2.2.3 Organizational Change and Participation

A change in organizational structure also a¤ects an agent’ willingness to participate in the s organization. For the manager, centralization, compared to decentralization, restricts her freedom to allocate e¤orts and thus decreases her utility. The e¤ect on the worker’ utility s is ambiguous, because more control over the manager’ power may give more power to the s worker. According to Proposition 1, when the manager is biased, centralization gives the worker more freedom to allocate e¤orts between production activities and private activities. The worker faces a trade-o¤ between the on-the-job bene…t and the private bene…t. The principal’ monitoring of private activities causes a loss of private bene…t under centralization. s The gain of the on-the-job bene…t from e¤ort-directing is small for the one who has a strong bias in project selection, and anticipates that the principal will overrule his proposed project. Proposition 3 (Selection E¤ ect) Consider an organizational change from decentralization to centralization, and suppose that the agents’ salary is …xed. 1) The manager always becomes worse o¤ . 2) A worker with more access to private bene…t and/or lower interest alignment with the principal is more likely to leave the organization; conversely, a worker with less access to private bene…t and/or higher interest alignment with the principal is more likely to participate in the organization. The proposition points out a selection e¤ect triggered by an organizational change: centralization tends to hinder the participation of incumbent employees (both managers and 10

workers) and increase turnovers, if the organization does not change its compensation policy. The organization, however, can bene…t from replacing workers whose interests are misaligned with the principal’ with the ones whose interests are better aligned. s,

3
3.1

Institutional Background and Data
Institutional Setting

This section describes the institutional framework, drawing from numerous interviews and the internal documentation of various Chinese newspapers. The Newspaper is an industrial leader at the provincial level in China and represents the current state of Chinese journalism. It employs more than 300 journalists (reporters and editors), and has a daily circulation of about one million in the sample years.10 Although owned by the state, the Newspaper is fully funded by advertising and sale revenues. It operates in a highly competitive local market.11 After paying a …xed fee to the state, the board of the newspaper has a large degree of freedom to distribute the residuals. A large component of the senior managers’income is tied to the Newspaper’ pro…tability. The Newspaper enjoys high autonomy in managerial s practices such as organizational strategies and pay schemes, and in editorial decisions except for reports about major political issues. The content of the Newspaper includes a front section that covers important news, headlines and editorial articles, an Economic and Business section, a Politics and Law section, an Education and Health section, a General Reports section focusing on investigative reports, sudden events and miscellaneous topics, and sections on Regional and Local News, Sports, Entertainment, and Consumption-Guide.12 About 80% of the news content is provided by the employed journalists, the rest by news agencies, freelance writers and other media. 3.1.1 Editorial Power and Production of News

The production of news content involves two major jobs: a reporter (he) covers news and writes reports, and an editor (she) selects and edits articles. Two alternative production procedures prevail. One is editor-oriented: an editor assigns a task to a reporter, who then implements the task according to her instructions. The other is reporter-oriented: a reporter covers news and sends his article to an editor, who then selects and edits the article. The key
10 The circulation number is a rough estimate, since Chinese newspapers may exaggerate their circulation number to attract advertising revenues. 11 The Newspaper competes with 5 other newspapers in the same market, and with more than 20 newspapers that are more specialized in a narrower type of journalism or geography. The Newspaper also faces increasing competition from broadcasters and the internet. 12 The Consumption-Guide is a section on consumer products and service, for example, fashion, housing, luxuary goods etc. There are two supplement sections, one on international news, in which articles are mainly provided by news agencies, and the other on culture and literature, in which articles are provided by freelance writers.

11

distinction between the two alternatives is who has real editorial power: whether an editor or a reporter e¤ectively decides the subject and selects an article for publication. Which procedure is used depends on the nature of tasks and the information obtained by each party. For example, the news coverage of the People’ Congress is usually assigned s to a reporter by an editor, as the event is anticipated and information is largely public. In contrast, a reporter determines the news content of an investigative report, as an editor sitting in an o¢ ce would not have the information. In general, reporters have substantial information advantages over editors in investigative reports, in-depth analysis of industries or government sectors, feature stories, and on-the-scene reports, which require task-speci…c expertise and/or direct contact with news sources. Editors are more e¤ective in making editorial decisions regarding propaganda, regular government activities, anticipated events, publicly accessible information, and columns designed in advance. Reporters and managing editors, supervised and coordinated by division directors, are organized in divisions corresponding to the news sections, except for the front section.13 Chief editors, who supervise their subordinates and approve the …nal publication of news content, may also intervene in editorial decisions. 3.1.2 Incentives and Agency Problems

The quality of news content crucially depends on the information collected by the reporters and the editors. It is essential to incentivize them to collect the desirable information for the Newspaper. Incentives are a particular concern, because journalism is human capital intensive, and the input and output of production are hard to verify. Agency problems occur when a journalist is distracted by private activities, or has misaligned interests with the Newspaper. The agency problem with a reporter is likely to come from the diversion of production activities before an editorial decision is made. Chinese reporters have large rent-seeking opportunities. The "hongbao" phenomenon that people receive money, gifts or other bene…ts from those who request their favors is pervasive in the Chinese media industry.14 Anecdotal evidence suggests that "hongbao" accounts for a signi…cant proportion of their income for some reporters. Moreover, reporters may spend time and e¤ort establishing "guanxi" (social connections in Chinese) to expand career and business opportunities. These bene…ts, referred to as the private bene…t in the theory, detract them from production activities, and may invite them to misuse the resources of their employers. A prevalent example is that a reporter
The front section in the newspaper is the …rst four pages that publish news headlines, important news mostly provided by the o¢ cial news agency (Xinhua), and editorial articles by sta¤ writers who do not cover news. A limited number of articles involve news coverage by the reporters from the specialized divisions. 14 In Chinese culture, "hongbao" is a red envelope with a monetary gift that Chinese people give to their employees, children and relatives on occasions such as new year celebrations, birthdays, weddings etc. It has become popular to give a "hongbao" to request a favor or exchange bene…ts. According to the regulation of Chinese media, journalists receiving "hongbao" from interviewees is considered to be corruption. But unless the amount of money is large and veri…ed, such misbehavior is hardly ever punished.
13

12

submits information in favor of interviewees. Some of this information, such as an advertisingtype report, is particularly detrimental to the Newspaper, as it not only harms news content but may also crowd out advertising revenues. To realize these private bene…ts, a reporter has to endeavor to seek information, establish relations with potential interviewees, and persuade editors. In contrast, an editor has far fewer opportunities to seek rents, as her scope of actions is usually limited to the o¢ ce, and her information source is easy to verify. Thus the editor’ s agency problem mainly occurs at the stage of making editorial decisions, when she evaluates a report di¤erently from the chief editors due to con‡ icts in their valuation of journalism or favoritism towards certain news subjects in exchange of perks. This causes the decision bias that I have discussed in the theory. 3.1.3 Performance Pay and Evaluation

The Newspaper adopts a high-powered payment scheme for the reporters. Besides a …xed base salary accounting for about one third of his wage, a reporter receives a piece-rate type pay directly tied to his monthly performance, which is measured by a score with two components: quantity and quality. The former is a composite measure of the numbers of published articles and words. The latter is assigned by an Evaluation Committee on a daily basis and aggregated up at a monthly level. The evaluation of the quality score is claimed to be "an accurate measure of a reporter’ individual contribution", and "fair to every employee".15 It is based s on well-de…ned rules, and I will show that I am able to imitate this internal evaluation with external measures of news content. When the published articles are authored jointly with other reporters or editors, the scores are adjusted by a sharing rule designed to distinguish between the contribution of each individual reporter. The pay to other employees is relatively low powered. Middle managers (division directors and managing editors) receive a ‡ wage, together with a bonus component based on an at internal assessment of the performance of the whole team in a news section. Chief editors are paid a salary according to their positions in the government/Party hierarchy, and a bonus depending on the pro…tability of the Newspaper in a …nancial year. 3.1.4 Organizational Reform

Centralization and decentralization, depending on the allocation of editorial power, are two basic organizational forms coexisting in the Chinese newspaper industry. From January 2002
15 The Evaluation Commitee is headed by a Chief editor and operated by 9 senior editors and reporters who are not involved with day-to-day news coverage and editing. Every day, the members of the Evaluation Committee select good articles from a list of published articles recommended by managing editors and assign a quality rank (convertable to a score) to each selected article. A list of all the evaluated articles is posted in a public area on the next day. If the editors and reporters do not agree with the evaluation of the committee, they can apply for reevaluation by providing evidence (e.g. readers’ feedback). A chief editor acts as an arbitrator to settle disputes at the end of each month. The …nalized scores aggregated at a monthly level are then converted to money by computer software.

13

to August 2005, the Newspaper employed a decentralized organizational structure by creating pro…t-center type divisions. Under this arrangement, editorial power was formally delegated to the middle managers in a particular division (e.g. Economic and Business News). Chief editors intervened in editorial decisions only in exceptional situations (see Panel A of Figure 1). In September 2005, the Newspaper decided to centralize editorial power in four divisions: Economic and Business, Politics and Law, Education and Health, and General Reports. Twenty-…ve managing editors in these divisions were reallocated to an editing center, headed by two chief editors and several associate editing directors (Panel B of Figure 1). E¤ectively, the chief editors retained the formal authority over editorial decisions, and closely monitored the editing process to clean up low quality or even harmful news content, such as advertisingtype information. The Sports division, however, experienced further decentralization as the sports reporters were formally allowed to make editorial decisions. The other divisions, Regional and Local News, Entertainment, Consumption Guide and Photographing remained unchanged.16 The reform was imposed by the Board, who claimed that centralization would "enhance control", and "improve competency". However, the reform was described as "a surprise" in interviews, as "no obvious problems were perceived". Insider information suggests that the reform was triggered by the appointment of a new general chief editor in June 2005, who was a former government o¢ cial and might have a tendency for centralization.17 The divisions to be reformed were supervised by the retired chief editor, whose job was taken over by the new chief editor. I will show that the data demonstrate no di¤erentiated trends in the performance of the reporters in the reformed divisions and the unchanged divisions before the reform. I restrict attention to the period from 2004 to 2006 despite the availability of data over a longer time, because the operating environment and the internal structure of the Newspaper were very stable during this sample period. There was no signi…cant change in regulation and politics in Chinese newspapers. The Newspaper remained an industry leader in the local market, and there was no entry and exit of competitors. The volume of news content (the number of pages) of the Newspaper was stable throughout the period. The pay schemes and the evaluation system, in terms of both the members of the Evaluation Committee and the evaluation procedure, did not change.
The Newspaper does not have a separate photographing section. But chief editors usually do not intervene the photographers’ work. Therefore the photographing division is regarded as decentralized and will be included in the control group. 17 The new chief editor was selected by the local Party among a list of candidates who had the same position in the governmental hierarchy as the retired chief editor. Unlike the chief editors that were promoted within the newspaper, the external appoinment is to a large extent random depending on the availability of external candidates.
16

14

3.2
3.2.1

Data
Data Collection and Sample Construction

In order to measure the reporters’ incentives and performance as accurately as possible, I construct a unique data set by combining the Newspaper’ internal personnel records and s external measures of news content. The Newspaper provided personal information of all its employees, and monthly performance measures, including the number of articles, the number of words, the quantity score, and the quality score, of all the reporters. A team of Chinese research assistants were hired to classify all the articles collected from the Newspaper’ on-line s archives over the sample period into categories of news content. Together with an experienced journalist, I speci…ed a set of coding rules according to the evaluation system of the Newspaper with reference to the evaluation of the Association of Chinese Journalists. The research assistants were trained to master the basic skills of content analysis in journalism. Then they coded every article by reading its title, authorship, byline, lead paragraph and other information such as formats and pictures. The web appendix explains in detail the data collection and variable coding. In the baseline sample, I exclude the observations of the sports reporters, because they experience a di¤erent organizational reform, and their performance is highly volatile due to exogenous shocks such as the Olympic Games and the World Cup. To reduce potential noise, I also exclude the following observations: 1) new recruits in the …rst three months who are not paid by performance; 2) division directors or editors who cover news occasionally; 3) regular reporters who wrote very little in some unusual situation, for instance, being ill or on holiday. All the excluded observations account for about 15% of the overall observations. The main empirical results presented below are robust in the samples when these observations are included (reported in the web appendix). 3.2.2 Personnel Information

Panel A of Table 1 summarizes the personnel information of 183 reporters in the baseline sample. Among the reporters, 60 percent are men, more than 80 percent have at least a college education, and about half are members of the Chinese Communist Party. The reporters are on average about 33 years old with an 8 year tenure at the Newspaper. Position is an indicator ranking from 1 to 3, representing reporter, chief reporter and senior reporter respectively in the hierarchy of the Newspaper. Quali…cation is a certi…cate authorized by the Association of Chinese Journalists to indicate one’ expertise and experience in journalism, s with 1 referring to assistant journalist, 2 to journalist, and 3 to senior journalist. The average levels of position and quali…cation are both about 1.5. Together with the tenure information, these imply that most reporters are mature enough to understand well the preferences and the evaluation system of the Newspaper, and have the skills and ability to work independently. Panel B of Table 1 reports the summary statistics of 56 managing editors (including a

15

small number of division directors) during the sample period. The gender ratio, education level and fraction of Party members of the managing editors are fairly similar to those of the reporters. The managing editors are on average older and more experienced than the reporters. The means of their positions and quali…cation are about 2.2, both substantially higher than those of the reporters’ . 3.2.3 Internal Measures of Quantity and Quality

I will use the internal quantity and quality scores as baseline outcome variables, because they are accurately measured to serve as a basis for performance pay, and thus good proxies for the reporters’performance. Moreover, these scores are comparable across di¤erent types of journalism given the consistency of evaluation, permitting a di¤erence-in-di¤erences identi…cation strategy. Simple regressions show that the variations in the number of articles and the number of words jointly explain more than 95% of the variation in the quantity score. The R-squared in the regression of the quality score on the quantity score is only about 40%, because the quality score captures the subjects of news content other than the number of articles and words. The quality score has another advantage in that it avoids the concern of article selection, as a high quality article is unlikely to be screened out.18 Therefore, I regard the quality score as a reliable measure of the quality of news content and a reporter’ s production initiative. The basic information on these performance measures is summarized in Panel A of Table 2. In an average month, a reporter writes 32 articles and 18434 words, and earns a quantity score of 2080, and a quality score of 1477. 3.2.4 External Measures of News Content and Editorial Activities

I classify the direct measures of news content into the following mutually exclusive categories: investigative report, feature story, special report19 , advertising20 , propaganda, government o¢ cials, on-the-scene report, sensational/entertaining report, and others. Investigative and feature reports correspond to the common sense of good journalism. Special reports indicate that they are unique or unusual in news subjects, or di¤erent in some important aspects from other newspapers’coverage of similar subjects. I use these three types of articles, particularly the …rst two, as proxies for a reporter’ good journalistic activities and production initiative s (journalistic initiative hereafter), since they require both substantial e¤ort to collect original
18 According to the interviews, on average about 20% articles submited to the editors are rejected. Most rejections are low quality articles. The rejection rate is much higher for junior reporters. A mature reporter is able to anticipate the probability of rejection, and will usually only spend substantial e¤orts on reports that are very likely to get published. 19 An article is coded as special report if it is a long article that contains key words like "special", "unique" and "…rst report", but not identi…ed as an investigative report or a feature story. 20 An article is coded as advertising if it is a promotion of products and/or image of a particular company. Most of the advertising articles are about local …rms. The advertising articles are distinguished from those soft advertisement articles assigned by the Newspaper for business clients, which are provided by the advertising department and not authored by reporters.

16

information and direct contact with news sources. Advertising articles capture the existence and extent of private bene…ts, and are usually regarded as bad journalism. Propaganda is the report of propaganda campaigns originated by the Party. Reports about government o¢ cials indicate the in‡ uence of governments on news content. The input information conveyed by other types of journalism, such as on-the-scene and entertaining/sensational reports, is less clear and will be only brie‡ discussed. y Parallel to the classi…cation of news content, I also categorize articles according to their authorship, which reveals information on editorial activities. For example, an article authored by a reporter jointly with a managing editor indicates that the report is originated and organized by the managing editor. Some articles directly spell out the role of a managing editor as a chief reporter. I classify these articles as "joint with editor". The articles written by reporters but assigned by managing editors to …t columns designed in advance are classi…ed as "column by content". These two types of articles are used to approximate the managing editors’ initiative. The articles that contain the names of external authors, who provide news sources to reporters and may participate in news coverage, also convey information on editorial decisions. There are three sources of external authors: government and public sector, private sector, and freelance writers. Usually the managing editors directly contact the freelance authors, while the reporters work with the other two types. The articles with external authors from the private sector may also indicate a reporter’ opportunities and s intention to establish business relations. Finally, some articles are coauthored with other reporters either within the same division or across divisions. One advantage of these external measures is that they are less sensitive to changes in the quality evaluation of the Newspaper. The major drawback is the incompatibility between di¤erent types of journalism. For instance, it does not make sense to compare business news with entertainment news. Therefore the constructed external measures only apply to the centralization group, in which common measures are plausible. Panel B of Table 2 summarizes the basic statistics of the external measures. A few features are worth pointing out. First, propaganda reports on average account for only about 1% of all the articles written by a reporter in a month, implying that the newspaper is not propaganda driven. Second, a reporter on average only writes 2.5 investigative and feature reports per month, as they require substantial e¤ort. Third, the number of articles "joint with editor" and "column by content" is small, showing that the reporters play a key role in journalistic activities and editorial decisions. In a regression of the quality score on the external measures, the R-squared exceeds 75%, supporting their credibility as reliable measures of the reporters’initiative and e¤ort. The main contributing factors to the quality score are investigative reports, feature stories, special reports, and propaganda articles, con…rming the crucial role of subject selection in determining news quality. As expected, the advertising articles and articles with external authors are negatively correlated with the quality measure. Due to score sharing, the articles

17

with internal coauthors, the "joint with editor" articles, and the "column by content" articles all reduce the quality score.

4
4.1

Empirical Strategies
From Theory to Test

If measures of the agents’ private bene…t are available, I can estimate the model directly, and back out the preference and technology parameters. Unfortunately, such information is usually beyond anyone’ reach. I thus have to rely on indirect evidence to examine the s theoretical mechanism in the model. The three theoretical propositions outlined in Section 2 provide guidelines for such an examination. Since accurate measures of the managing editors’ performance is not available, I will mainly estimate the impact on the worker’ performance. Proposition 1 predicts a reduceds form average treatment e¤ect of centralization on a reporter’ initiative and performance. The s estimation of a causal e¤ect relies on the panel structure of the data and the identi…cation strategy that I will discuss below. However, such an average treatment e¤ect is mute about underlying mechanisms, and can be interpreted in various ways. Thus it is crucial to test Proposition 2 — the heterogeneous treatment e¤ect: with controls of ability, the reporters with larger private bene…ts should respond more to the reform if the e¤ort-directing e¤ect dominates. I will exploit institutional factors such as job assignment and social norms, which reveal information on the reporters’access to private bene…ts, to test this prediction. Proposition 3, the selection e¤ect, also casts light on the e¤ort-directing mechanism, as it is another way to demonstrate heterogeneous treatment: reporters with large private bene…ts or low interest alignment will respond in an extreme manner to select themselves out of their job. I will infer the selection pattern by estimating the individual …xed e¤ects of the entries, stayers and exits. Empirical results that are jointly consistent with these three propositions are in line with the theory, lending support to the mechanism that centralization directs workers’e¤ort from private activities to production activities. To shed further light on the basic trade-o¤ between better control and depressing initiative, and the trade-o¤ between the loss of the manager’ initiative and the promotion of s the worker’ initiative, I will estimate the impact of centralization on the direct measures s of news content and editorial activities. In particular, the e¤ort directing mechanism would result in a negative relation between the e¤ect on the measures of a reporter’ journalistic s initiative and the e¤ect on the measures of his private bene…ts. Furthermore, the initiative substitution mechanism would lead to a negative e¤ect of centralization on the measures of the managing editors’initiative, associated with an increase in the measures of the reporters’ journalistic initiative. Empirical veri…cation of these two hypothetical results favors Case 1 in Proposition 1.

18

4.2

Identi…cation

The organizational reform in the Newspaper creates empirical counterparts of the two organizational forms in the theory: four divisions (Economic and Business News, Politics and Law, Education and Health, and General Reports) experience an organizational change from decentralization to centralization. In other words, a centralization treatment is applied to this group of reporters. Even though the timing of the reform is arguably exogenous, there may be unobservable factors associated with the reform that could cause serious bias. This is of particular concern for the quality measure, which can be sensitive to explicit or implicit changes in editorial and evaluation policy. Fortunately, the remaining decentralized divisions (Regional and Local News, Entertainment, Consumption Guide, and Photographing) can serve as a control group to mitigate potential bias. The identi…cation, therefore, hinges on a valid di¤erence-in-di¤erences (D-I-D hereafter) estimator. Figure 3 plots the average quantity and quality scores in logarithm of the treatment group and that of the control group over time. The time series is fairly volatile due to seasonality and exogenous shocks in the industry. For example, the high performance in March 2005 and March 2006 is driven by the Chinese National People’ Congress. Two features strongly s support the validity of the D-I-D estimator. First, there is no trend in the performance of the treatment group before the reform, con…rming that the reform is exogenous to the reporters’ performance. Second, the performance of the treatment and that of the control groups are very similar in terms of levels and co-movement pattern before the reform, suggesting that the treatment group would behave similarly as the control if there were no treatment. One potential concern is that the e¤ect of the reform would be contaminated if reporters transfer between the treatment and the control after the reform. There are only 6 reporters switching between the two groups over the sample period, and the estimates from the sample that excludes these switchers are virtually the same as from the baseline sample.

4.3

Econometric Speci…cation

The baseline D-I-D regression estimates the following panel speci…cation: Log(Pit ) = t +

i

+ (Ci Rt ) + Xit + "it ;

(5) month level. The de-

where i indicates individual, and t indicates time at the year score or the quality score. Newspaper. i t

pendent variable is the logarithm of a reporter’ performance in terms of either the quantity s is time …xed e¤ects to control for aggregate ‡ uctuations of the is individual …xed e¤ects to control for unobservable individual ability and

preferences, which also helps to overcome the potential selection bias due to the entries and exits of reporters associated with the reform. Ci is a dummy that equals one for the reformed divisions, and zero for the remaining decentralized divisions. Rt is a reform dummy equal to one if a reporter’ performance is observed after the reform. The coe¢ cients of both Ci s 19

and Rt are not identi…able in the presence of individual/division …xed e¤ects and time …xed e¤ects. Ci Rt is the interaction term between the two variables, and its coe¢ cient identi…es the average treatment e¤ect on the treated. Xit is a set of covariables including division …xed e¤ects (some reporters switch across divisions), and time-variant individual characteristics such as age-squared, tenure-squared, position and quali…cation. These covariables help to control for ability, career concerns, and other factors that may a¤ect the reporters’ performance.21 it is the stochastic error term, which may be correlated over time or within certain

clusters in the D-I-D estimation with many periods (Bertrand et al 2004, Angrist and Pischke 2009). I will cluster the standard errors at the individual level to cope with potential time serial correlation. The main results are robust using other clustering strategies.22

5

Main Results

This section presents the main empirical results to investigate the impact of the organizational reform from decentralization to centralization on the internal measures of the reporters’ quantity and quality scores. I start with an investigation of the average treatment e¤ect, then explore the heterogenous treatment e¤ect with regard to the reporters’access to private bene…ts, and …nally analyze the individual …xed e¤ects to examine the treatment on distribution and the selection pattern.

5.1
5.1.1

Average Treatment E¤ects
Descriptive Results

Table 3 displays the reporters’average performance before and after the reform in the treatment (centralization) group and the control (decentralization) group, and the comparison between the two groups. To focus on the impact of centralization on the intensive margin: the change in the average performance of the same reporters before and after the reform, I restrict the sample to a balanced panel that includes 113 reporters who are observed both before and after the reform, and do not switch between treatment and control. Consistent with Figure 3, before the reform, there are no signi…cant di¤erences in either the quantity score or the quality score between the treatment and the control. Panel A shows that the di¤erences in the quantity score under the two organization schemes in both the treatment and control groups are negligible, and the di¤erence-in-di¤erences comparison is small and statistically insigni…cant. These results are not surprising, given that the Newspaper’ vols ume of content is basically …xed and the space to accommodate more articles and words is
The variables age and tenure are not identi…ed due to collinearity in the regression with both individual …xed e¤ects and time …xed e¤ects. 22 The results in the regressions that cluster the standard errors at the division level are considerably less precise because the small number of clusters (9 divisions) substantially in‡ ates the standard errors. But the main results are still signi…cant at the 10% level. The results that cluster the standard errors at the division quarter level are more precise than those that cluster at the individual level.
21

20

limited. However, the comparison of the log quality score (Panel B) suggests that the organizational reform has a strong e¤ect on the reporters’ quality performance. The quality score of the treated reporters is only slightly above that of the control before the reform, but the gap widens dramatically after the reform, amounting to a di¤erence-in-di¤erences comparison of 0.151 in the mean with a standard error of 0.075. It is important to recognize that the result is mainly driven by the negative impact of the reform on the performance of the control, which suggests that there may exist negative common shocks to all the reporters in the Newspaper.23 The lack of response of the reporters’quantity performance rules out the potential spurious relation between the timing of the reform and the expansion of the Newspaper. Rather, the organizational reform is likely to a¤ect a reporter’ journalistic initiative that determines s the quality of news content. 5.1.2 Baseline Estimates

Using the D-I-D approach speci…ed in Equation (5), I estimate the average treatment e¤ects of the reform on the logarithm of the quantity and quality scores. The …ndings in Panel A of Table 4 con…rm the descriptive evidence. The simplest estimation, controlling for only individual …xed e¤ects (Column 1 and 5), shows that the average e¤ect of centralization on the reporters’quantity score is economically small (5.4%) and statistically insigni…cant. But the e¤ect on the quality score is statistically signi…cant at the 1% level and economically large (20.7%), which amounts to a 5% increase in wages. The results hardly change after adding the time dummies (Column 2 and 6), and additional controls including division …xed e¤ects and the time-variant personal characteristics (Column 3 and 7).24 When the individual …xed e¤ects are replaced with controls for time-invariant personal characteristics such as gender, education and Party membership, together with age and tenure (Column 4 and 8), the R-squared is reduced almost half. The estimated e¤ect on the quality score declines dramatically from 19.4% to 6.1% and becomes statistically insigni…cant. The e¤ect on the quantity score becomes negative, though statistically insigni…cant. These results suggest a negative selection associated with the organizational reform, which I will analyze later. 5.1.3 Dynamic E¤ects

Panel B of Table 4 presents the dynamics of the average treatment e¤ects. I replace the interaction term between the treatment dummy and the reform dummy with a set of dummy
One potential common shock is implicitly tighter evaluation of the quality score due to budget constraint. The internal documentation shows that the Newspaper puts more emphasis on shorter articles and better writing after the reform. Evidence suggests that the evaluation may be more strict for the categories of articles that contribute most to the quality score after the reform, as will be shown. 24 The result is robust when division trends are included in the regression.
23

21

variables. "ref ormstart" is a dummy equal to one if a reporter works in the treatment group in the month of the reform (September 2005) and zero otherwise, "August2005" a dummy for a reporter in the treatment in August 2005 (one month before the reform), and "October2005" a dummy for a reporter in the treatment in October (one month after the reform). Similar de…nitions apply to "July2005", "N ovember2005" and "December2005". The regressor "January2006onwards" is a dummy that equals one for a treated reporter from January 2006 and onwards. The dynamic e¤ects of centralization are consistent with the previous …ndings. The insigni…cant estimates of both the quantity and quality scores before the reform con…rm that there is no pre-trend e¤ect. The e¤ects on the quantity score are always insigni…cant. The response of the quality score to centralization is not signi…cant until November 2005 (two months after the reform). The e¤ect becomes larger and more pronounced four months after the reform. The gradually increasing e¤ect rules out the concern that the reformer deliberately increases the quality score to reward (or compensate) the treatment group or to demonstrate the success of the organizational reform, in which case the response would be stronger in the short run. The lack of response in September and October of 2005 may be because these two months are among the special period, in which social norms condone rent seeking behavior and o¤set the e¤ect of the reform. I will examine this argument in the next subsection.

5.2

Heterogenous Treatment E¤ects

To test the heterogenous treatment hypothesis, I estimate the e¤ects of the organizational reform across di¤erent groups of reporters whose task assignment exposes them to di¤erent levels of private bene…ts, and across di¤erent periods, in which the extent of a reporter’ s access to private bene…ts varies. 5.2.1 Access to Private Bene…ts across Task Assignments

It is not unusual that the exposure and access to private bene…ts systematically vary across task assignments within an organization. Well known in the Chinese media industry, economic and …nancial reporters have access to large pecuniary private bene…ts and business opportunities, as they specialize in covering news about companies and products. As in other transitional economies, rent seeking behavior is particularly active in the sectors that experience drastic commercialization and privatization. Education institutions, hospitals and pharmacies in China since 2000 fall into this category.25 The reporters in these two divisions are more likely to divert their e¤orts to pursue private bene…ts. In contrast, the reporters in the Politics and Law division and the General Reports division, who focus on government policies and routines, investigative reports and sudden events, have much more limited access
Corruption in the education industry and the healthcare sector is frequently reported in media and widely debated in public.
25

22

to private bene…ts.26 These conjectures are supported by the distribution of the number of advertising articles across news divisions in the sample: 1145 in Economic and Business, 72 in Education and Health, but only 28 in Politics and Law, and 11 in General Reports. A natural proxy for the extent of the reporters’access to private bene…ts is their allocation to divisions, which are based on task assignment.27 I extend the D-I-D estimation of the e¤ects of centralization on the scores to incorporate the heterogeneous treatment across reporters in the four treated divisions: Economic and Business, Education and Health, Politics and Law, and General Reports, with the control group unchanged. Table 5 presents the results. As expected, the Economic and Business reporters improve their performance substantially, about 20% in quantity and 35% in quality after the reform. The Education and Health reporters improve their quantity score by more than 12% and the quality scores by more than 28%, although the e¤ect on quantity is insigni…cant. On the contrary, the reporters in Politics and Law respond negatively to centralization, although the e¤ect is not statistically signi…cant in the presence of individual …xed e¤ects. The e¤ect on the General reporters’quality score is positive, but economically modest and statistically insigni…cant; the notable decline in their quantity score may result from the increases in the quantity score of their colleagues in Economic and Business and Education and Health, whose increased publications crowd out the General reporters’ Note . that the pattern of negative selection found in the average treatment e¤ect is also present within each division except for General Reports, and most pronounced in Politics and Law, which experiences the largest exits and entries. 5.2.2 Private Bene…ts Condoned by Social Norms

In China, the Spring Festival (the Chinese New Year) and the Mid-Autumn Festival (also the mid …nancial year for companies) are two special time periods, in which Chinese people conventionally seize opportunities to exchange "hongbao", establish social connections, and expand business networks. Therefore, social norms condone rent seeking behavior in these periods. Anecdotal evidence suggests that the restriction of reporters’ private activities is much more relaxed than usual, and some editors may also be involved in the pursuit of private bene…ts. Moreover, the chief editors are usually overloaded as they are engaged in numerous external activities in the local Party and local government, in addition to the management of internal activities. As a result, one should expect little impact of centralization in these periods, if e¤ort directing is the mechanism underlying the reporters’response.
The task assignment of the General reporters is fairly similar to the Politics and Law reporters, except that the former focuses more on exceptional events. It might be possible that reporters receive private bene…ts from governments or from interviewees who are involved in scandals. But these activities are regarded as serious journalism corruption and are risky for a reporter to undertake. Anecdotal evidence suggests that such misbehavior is unusual in leading Chinese newspapers, though it may be more common among reporters working for lower quality newspapers. 27 The task assignment to a reporter usually stabilizes after a two or three year tenure in the Newspaper. For most reporters, their tasks are assigned before the sample year.
26

23

The Spring Festival is often in late January and sometimes in early February, and the Mid-Autumn Festival is usually in September and occasionally in early October.28 Private activities are likely to take place a few weeks before the festivals. Therefore I construct a "special months" dummy equal to one for January and September, and zero for all the other months. Table 6 reports the regressions, in which I add to the baseline estimation speci…ed in (5) an additional interaction between reform_treatment and the dummy of "special months". This triple di¤erence estimation shows that the e¤ect of centralization on the quality score is a 16.5% reduction in the special months, relative to the e¤ect in the normal months, which is a 22% increase. The F-test cannot reject that the sum of these two coe¢ cients equals to zero, and supports that the e¤ect of centralization in the special months is negligible. The impact on the quantity score is insigni…cant in either the special or the normal months, suggesting that the result is more likely to be driven by the reporter’ adjustment of e¤orts, instead of s changes in the volume of the Newspaper and editorial policies during these special periods. The above results are robust if February and October are included in the "special months" to consider the lasting in‡ uence of social norms.

5.3

Estimates of Individual Fixed E¤ects

To complement the above evidence, I estimate the e¤ects of centralization for each individual reporter using the following panel data speci…cation, Log(Pit ) = t +

X i Di [

bef ore (1 i

Rt ) +

af ter Rt ] i

+ Xit + "it ;

(6)

where Di equals one for worker i, and zero otherwise, and all the other variables are de…ned as in equation (5). bef ore i

and

af ter i

are estimates of the …xed e¤ects for each

individual before and after the reform respectively. I refer to the individual …xed e¤ects from the regression of the log quantity score as quantity residuals, and the ones from the regression of the log quality score as quality residuals. Since the regressions control for variables that measure time-variant experience and expertise, these residuals, to some extent, capture the unobservable individual incentives. 5.3.1 E¤ect of Treatment on Distribution

To show the impact of centralization on the distribution of the reporters’ response, I plot the kernel density of the estimated individual …xed e¤ects in Figure 4, using the balanced panel that only includes those reporters who appear during the whole sample period and do not switch (66 stayers in the treatment and 47 in the control). Panel A shows that in the treatment group, the distribution of quality residuals after the reform shift to the right of that before the reform, and the p-value of the Kolmogorov-Smirnov test for the null of equality of
The Spring Festivals in 2004, 2005 and 2006 are 22nd January, 9th February and 29th January respectively, and the Mid-Autumn Festivals are 28th September, 18th September and 6th October respectively.
28

24

distributions is 0.001. However, such a pattern is not observed in the control group, in which the Kolmogorov-Smirnov test does not reject that the two distributions of quality residuals are equal. Given that the stayers are mostly experienced reporters even before the sample period, the changes in the quality residuals are more likely to re‡ improvement of the ect reporters’ production incentives instead of their ability. Moreover, the distribution of the quality residuals in the treatment becomes more concentrated around a higher value after the reform. This is consistent with the intuition that centralization restricts the reporters’ pursuit of private bene…ts and thus homogenizes their incentives. Panel B shows that the distributions of the quantity residuals before and after the reform are statistically di¤erent in the treatment group, but not so in the control group. Interestingly, in the treatment group, the distribution of the quantity residuals shifts to the left after the reform, as opposed to the change in the distribution of the quality residuals. This contrasting result suggests that a reporter’ quality-enhancing e¤orts may substitute his s quantity-enhancing e¤orts. Overall, the results of the stayers’individual …xed e¤ects are in line with the previous estimates of the average treatment e¤ects of centralization, and con…rm that the organizational reform improves the reporters’production initiative. 5.3.2 Selection Pattern

As noted, the e¤ect of centralization on the quality measure decreases from about 20% to less than 7% when the individual …xed e¤ects are excluded, probably because of the exits and entries of reporters in both the treatment and the control. To examine the selection pattern, Table 7 compares the estimated individual …xed e¤ects of the exits, the stayers and the entries. In Panel A, the after-before reform di¤erence in the quality residuals of the stayers in the treatment group is signi…cantly greater than that in the control group. However, the di¤erence in the quality residuals of the entries and the exits in the treatment group is much smaller than that in the control group, and thus o¤sets the positive e¤ect of centralization on the stayers, causing the negative selection in the regression results in Table 3. Panel B …nds a similar pattern in the comparison of quantity residuals. Since the individual …xed e¤ects, particularly the quality residuals, are highly correlated before and after the reform, I compare the individual e¤ects between the exits and the stayers before the reform to infer their di¤erences in unobservable individual characteristics such as incentives and ability. Table 7 shows that both the quantity and quality residuals of the exits are remarkably lower than those of the stayers in the treatment group.29 In contrast, in the control group, the average quantity residual of the exits is larger than that of the stayers before the reform, and the di¤erence in the quality residuals is small. These results suggest that the preferences of the exits may be less aligned with the Newspaper’ interests s
29 In the regression of the quantity residuals on a dummy that equals one for exits and zero for stayers in the treatment group before the reform, the coe¢ cient is -.562 with a bootstrapped standard error .291. In the similar regression of the quality residuals, the coe¢ cient is -.544 with a bootstrapped standard error .403.

25

than those of the stayers. Then, I compare the entries and the stayers after the reform. Not surprisingly, the quality residuals of the entries in the treatment group are of similar magnitude to their counterparts in the control group, because the new recruits usually rotate their task assignment in a few divisions in the two years. In the treatment group, the entries’ quality residuals are very similar to the stayers’ while in the control group, the entries’quality , residuals are substantially larger than the stayers’ The quantity residuals also display the . same pattern. These results suggest that the entries have higher interest alignment with the Newspaper than the exits. To summarize, the …ndings in Table 7 support Proposition 3, which implies that centralization, relative to decentralization, hinders the participation of the reporters with larger private bene…ts and/or stronger decision bias, but facilitates the participation of the ones whose interests are more aligned with the Newspaper’ preferences. s

6

Mechanism

According to the theory, centralization may improve a reporter’ initiative and performance s through two channels: the e¤ort-directing e¤ect through better control of the reporters’ pursuit of private bene…ts, and the initiative substitution between the middle managers and the reporters. The results presented above have indicated the e¤ort-directing channel. This section presents evidence to strengthen this argument and examine the other channel, using the external performance measures, which capture more directly the reporters’ incentives and convey information on the manager’ initiative. I will also provide further evidence to s discriminate between a number of alternative explanations.

6.1

E¤ects on External Performance Measures

As the external performance measures only apply to the treatment group due to the incompatibility in measuring di¤erent journalism between the treatment and the control, I will estimate the following speci…cation: EPit = m +

y

+

i

+ Rt + Xit + "it :

(7)

The dependent variable EPit is an external measure of monthly individual performance without taking logarithm. Since a set of year dummy, I only include the month dummies mies y month dummies are collinear with the reform m to control for seasonality, and the year dumi

to control for business cycles over years.

is individual …xed e¤ects, Rt the reform

dummy, and Xit the time-variant covariables, all de…ned as before. The absence of a control group is less of a concern than when the dependent variables are the internal measures, because the external measures mainly capture the subjects of news content and are less sensitive to changes in evaluation. Table 8 reports the estimates.

26

6.1.1

Trade-o¤ between Production Initiative and Private Bene…ts

Panel A of Table 8 presents the impact of the organizational reform on news content. The e¤ects on the measures of journalistic activities are positive and statistically signi…cant. In particular, the number of investigative reports increases by .325 standard deviations, and the number of feature stories increases by .247 standard deviations, both at the 1% signi…cance level. At the same time, the organizational reform reduces the number of advertising articles by .411 standard deviations, and the result is statistically signi…cant at the 1% level. Consistently, the number of articles with the external authors from private sectors, which may indicate a reporter’ opportunities to attain private bene…ts, decreases after the reform. s These …ndings demonstrate a substitution between the reporters’journalistic initiative and their attainment of private bene…ts, con…rming the e¤ort-directing e¤ect. The e¤ects of centralization on the number of propaganda articles and the number of reports about government o¢ cials are positive, but small and statistically insigni…cant. These …ndings rule out the potential confounding factor that the Party and governments, for the purpose of ideological control, in‡ uence the Newspaper to increase these two types of articles. The e¤ects on the other measures of news content are negligible and insigni…cant. 6.1.2 Initiative of Managing Editors

Panel B of Table 8 reports the estimates of the reform on the external measures of editorial decisions. Centralization reduces the number of articles "joint with editor", the most robust proxy for the initiative of the managing editors, by 0.162 standard deviations and in a statistically signi…cant way. The e¤ects on the number of "column by content" articles and the number of articles jointly with freelance external authors, whom the managing editors contact directly, are all negative, though statistically insigni…cant. This evidence, together with the results in Panel A that centralization improves the reporters’journalistic initiative, indicates the existence of initiative substitution between the reporters and the middle managers. 6.1.3 Heterogenous Treatment E¤ects

Panel C of Table 8 reports selective results from the regressions that split the treatment group into the four divisions as before. The trade-o¤ between the reporters’journalistic initiative (measured by the number of investigative reports and feature stories) and their attainment of private bene…ts (measured by advertising articles) only appears in two divisions: Economic and Business and Education and Health, in which the quality score increases substantially after the reform, as shown in Table 5. The pattern is most pronounced for the Economic and Business reporters, who also experiences the largest improvement in their performance. With regard to the e¤ect on the managing editors’initiative, centralization reduces the number of "joint with editor" articles in Education and Health, suggesting that the improvement in the reporters’ production initiative under centralization is partially driven by the

27

depression of the managing reporters’ initiative, which ampli…es the e¤ort-directing e¤ect. However, such an initiative substitution e¤ect is muted in the Economic and Business division, possibly because the managing editors in this division may also have notable access to private bene…ts, and centralization directs their e¤orts to production initiative as well. This result is consistent with Proposition 2, which posits that the e¤ort-directing e¤ect on the worker is reinforced by that on the manager. The e¤ects of centralization on the General reporters are qualitatively similar to those on the Education and Health reporters, but most estimates are statistically insigni…cant. Interestingly, the estimates in the Politics and Law division are contrary to those in other divisions: centralization reduces the number of investigative reports but increases the number of advertising articles, though statistically insigni…cant. Moreover, centralization substantially reduces the number of "joint with editor" articles authored by the Politics and Law reporters. These …ndings suggest that action distortion is not a major concern for the Politics and Law journalists; the depression of the managing editors’initiative does not cause a su¢ ciently large response from the reporters. The results are consistent with the previous argument that the Politics and Law reporters have much more limited access to private bene…ts. They are suggestive evidence against Case 1 (biased manager and distracted worker), but in favor of Case 3 (biased manager and biased worker), in Proposition 1. I also examine the e¤ects on the external measures in the special months, in which social norms condone the attainment of private bene…ts. (Results are reported in the webappendix.) Consistent with the previous …ndings, the increase in the number of articles that represent journalistic initiative in these special months is substantially smaller than in other months, whereas the e¤ect on the number of advertising articles is positive. Notably, the negative e¤ect on the number of articles, initiated by the managing editors (the sum of "joint with editor" and "column by content" articles), is signi…cantly alleviated in the special months, con…rming initiative substitution between the managing editors and the reporters.

6.2

Alternative Explanations

This subsection examines a number of alternative explanations. All the related empirical results are collected in the web appendix. 6.2.1 Manipulation of the Evaluation System

The positive e¤ects of centralization on the reporters’quality performance could be spurious if the chief editors manipulate the evaluation system to in‡ ate the quality score for the reporters in the treatment group. I examine this possibility by testing the stability of the correlation between the quality measure and the external measures of news content before and after the reform. Speci…cally, I regress the quality score on the measures of news content and their interactions with the reform dummy, controlling for measures of editorial decisions that a¤ect the assignment of scores. None of the interactions between news content and the 28

reform dummy is statistically signi…cant. This result strongly supports the stability of the Newspaper’ evaluation system over the sample period, and rules out the possibility that the s increase in the quality score is caused by a relaxation in the evaluation. In fact, evaluation for some types of journalism may become tighter after the reform, as the coe¢ cients of the interaction terms between reform and the number of investigative, feature and propaganda reports are negative, though statistically insigni…cant. A tighter evaluation explains why the external measures of the reporters’journalistic initiative increase substantially after the reform, but their quality scores do not in the absence of the control group. 6.2.2 Changes in Editorial Policies

One potential concern is that the Newspaper may change its editorial policies after the reform, either because the reform is in part intended to or because the Newspaper takes the opportunity to grow certain sections (e.g. Economic and Business, and Education and Health), or to encourage certain types of reports (e.g. investigative reports and feature stories). This explanation is unlikely for the following reasons. First, the data show that there is no di¤erentiated trend in the performance between each treated division and the control group before the reform. Second, a change in editorial policies is likely to be associated with a change in the quality evaluation system, which serves as the basis for pay and promotion for reporters. As I have shown, the Newspaper’ evaluation of quality score remains stable s before and after the organizational reform. Third, an e¤ective change in editorial policies in favor of certain divisions often requires reallocation of resources, in particular human capital. However, the distribution of the number of managing editors and reporters across divisions hardly changes over time.30 The number of supporting sta¤ for technical issues such as copy-editing and proofreading remains the same for each division. 6.2.3 The Role of the New Chief Editor
31

Another potential concern is that if the appointment of the new chief editor causes systematic bias towards the treatment group after the reform, then the D-I-D estimate will capture this bias. To evaluate this potential confounding factor, I conduct a placebo test, using the fact that the appointment was in June 2005, and the reform took place in September 2005. Speci…cally I include in the baseline regression (5) a dummy that equals one if a reporter is observed after the appointment of the new chief editor and in the treatment group. As shown in Table A8, the di¤erence-in-di¤erences estimation of the impact of the appointment
30 Interestingly, I …nd a slight decline after the reform in the number of reporters allocated to the Education and Health division, whose reporters improve their quality score signi…cantly, but a slight increase in the number of reporters allocated to the General Report division, whose quality performance experiences little change. 31 In terms of other resources such as information technology and …nancial budget, every division in the Newspaper used the same information and computer technology; in the conversations with many reporters, none of them complained that …nancial budget was a major concern for their journalistic activities.

29

on the treated reporters is negligible in both the quantity and quality scores, conditional on the impact of the true reform. Consistent with this result, several institutional features suggest that the new chief editor is unlikely to have leadership qualities or missions to reform the newspaper. He is appointed by the government, which needs to …ll in the position, rather than being selected by the Newspaper from the market. He lacks journalism expertise and is regarded as "a transitional guy".32 Moreover, he is the only replacement among nine chief editors, among whom the chairman has the largest in‡ uence on the Newspaper. 6.2.4 Contributions of Middle Managers

I have focused on the project selection function of the middle managers. But some middle managers, particularly the senior ones, may also play other roles, such as instruction and supervision of how to implement projects. If the organizational reform systematically a¤ects the middle managers in these dimensions, the previous estimates are potentially biased. Changes in the composition of the middle managers. Even though the middle managers’initiative is reduced after the reform, a more able team of managing editors may improve the reporters’ performance through better instruction and editing, which may not be purged from the quality score. An examination of the composition of the managing editors limits the possibility of this explanation. First, the division directors, who among the middle managers potentially have the largest in‡ uence on the reporters’ performance, remain the same people. Second, there are 18 turnovers (including exits and entries) among 56 managing editors during the sample period, with 12 in the treatment and 6 in the control. However, the turnovers mostly take place among junior editors who have a limited impact on the reporters, and their personnel characteristics, including education, working experience and expertise, are fairly similar. Third, the Economic and Business division, on which the organizational reform has a largest impact, experiences the smallest changes in managing editors. Improvement in implementing projects. When the managing editors lose their initiative to acquire information, they may divert their attention to implementing projects. For instance, a managing editor may spend more e¤ort revising and editing a reporter’ s articles to improve their readability and style, which may contribute to the reporter’ quality s score. Insider insights from the Newspaper suggest that the managing editors’input in the reporters’performance is far more important for junior reporters who have yet to accumulate su¢ cient …rm speci…c expertise. Therefore, I estimate the e¤ects of centralization on the junior reporters who have working experience equal to or fewer than 3 years in the Newspaper (Table A7).33 Relative to the impact on the more senior, the e¤ects of centralization on
32 The personal pro…le of this new chief editor shows that he had no working experience in journalism at all before he joined the Newspaper, and he stepped down from the Newspaper after two years. 33 Most interviewees agree that it takes usually 2 to 3 years to acquire the newspaper speci…c expertise to cover news e¢ ciently and write well.

30

the junior’ quantity and quality scores are signi…cantly negative in the treatment group, s whereas such di¤erences in the control group are not obvious. Moreover the negative e¤ect of centralization on the number of articles "joint with editors" and "column by content" is particularly strong for the junior reporters. These …ndings show that the reporters may learn more slowly or receive less support from the editors after the reform — a result that goes against the explanation that centralization improves implementation. 6.2.5 Improvement in Coordination

One important aspect of authority is to provide coordination. Theory suggests that the e¤ect of centralizing authority on coordination is in general ambiguous (see Footnote 2 for reference). In the context of the Newspaper, centralization may allow for more concentrated information processing, and improve the coordination between the managing editors and the reporters or the coordination between the reporters. For example, a chief editor, after gathering relevant informationcan, can better coordinate reporters across di¤erent divisions to cover an event. However, empirical evidence does not support the hypothesis that coordination is improved in the Newspaper after the organizational reform. The measure of the managing editors’initiative, to some extent, captures their coordination activities, because managing editors are most likely to publish their names together with reporters when they organize the coverage of a news event that requires more than one reporter and take the responsibility of coordination. I have shown that this measure signi…cantly declines after the organizational reform. Another measure of coordination is internal coauthorship. Since the sharing rule reduces the performance pay for a report with internal coauthors, the reporters have a tendency to avoid coauthorship if their journalistic activities are not complementary to each other and not coordinated by their superiors. Coordination is particularly important for cooperation across divisions. The last column in Panel B of Table 8 shows that the number of articles with internal coauthorship across divisions declines, though statistically insigni…cant. Interestingly, centralization signi…cantly increases the number of articles that reporters coauthor within the same division. This is mainly driven by the Politics and Law division, in which the initiative of the managing editors declines dramatically after the reform and the reporters may cooperate to compensate the depressed initiative and loss of coordination from their superiors. 6.2.6 Improvement in Soft Environment

Research on social psychology of organization suggests that redistributing authority among managers may change the working environment, such as corporate culture, morale, workers’ intrinsic motivation, peer pressure and other soft factors. It is not evident that the organizational reform a¤ects the working environment in the treated divisions in a systematically di¤erent way than in the control divisions within the same newspaper. Moreover, this expla31

nation cannot rationalize the heterogenous treatment e¤ects in the various dimensions such as across divisions, across months (when social norms condone rent seeking behavior), and along seniority. Finally, this argument cannot explain the opposite impact of the organizational reform on the reporters and on the middle managers.

7

Conclusions

This paper has presented coherent evidence of the impact of organizational structure, determined by the distribution of formal authority, on workers’incentives and performance, drawing on the institutional setting combined with detailed personnel information in a Chinese newspaper. The research has probed two fundamental questions in organizational economics: what is the source of authority? How does the distribution of authority moderate incentive incompatibility among people in a hierarchy? The incentive view of authority is derived from the premise that authority cannot be completely contractible and fully enforceable. The role of organizational structure hinges on its impact on workers’attainment and control of the resources that generate real authority. My paper has contributed to our empirical understanding of this fundamental view of authority. The empirical …ndings are in line with the theory, in which information is the source of real authority, and agents optimize their resource allocation and compete for real authority, in response to a given distribution of formal authority. Moreover, this research casts light on the mechanism of authority at work in a hierarchy. Authority has two aspects: 1) direction of actions and 2) direction of decisions. The former mitigates the agency problem of hidden action, and the latter mitigates the agency problem of decision bias. Therefore, the e¤ect of the distribution of authority on agents’ incentives crucially depends on the nature of agency problems and which aspect of authority plays a larger role. A number of consistent results point to the trade-o¤ between better control and depressing initiative: a centralized hierarchy alleviates an agent’ action distortion and decis sion bias, at the cost of killing his or her initiative for information acquisition. Furthermore, this paper shows that the agents’relative positions in the hierarchy matter: the type of the middle manager is important to direct the impact of authority structure on the worker. Several pieces of evidence have shown that depressing the initiative of agents at a higher layer in turn promotes the initiative of those subject to their authority. In terms of bene…t and cost analysis, the adoption of the centralized organizational structure in the Chinese newspaper improves the quality of its news content, which may lead to a larger circulation and more advertising revenues. With personnel data from a single …rm, the welfare analysis is limited. I am collecting data from a number of Chinese newspapers, and attempting to investigate the impact of organizational strategies on …rm performance in greater detail. The current research is based on personnel data from a speci…c Chinese newspaper, which

32

certainly limits its external validity. However, the Newspaper is fully funded by advertising and sale revenues, and its adoption of organizational strategies is, to a large extent, driven by pro…tability concerns. How to structure the organization of newspapers in response to market competition has been widely debated since the commercialization of Chinese media. The economic questions addressed in my paper are common in the Chinese newspaper industry, and echo the research on organizational structure in Western media, for example, the studies by DuBick (1978) and Carter (1984). Furthermore, the dialogue between the theory and the empirics in my paper suggests that the insight from the study of the Newspaper can be generalized to other organizations, in which the production involves intensive human capital and information collection, and the input from workers is hard to verify and monitor. In other settings and with di¤erent focuses, some recent research points to the same direction: the allocation of authority in‡ uences workers’ incentives and decisions. For example, Argyres and Silverman (2004) provide evidence of the correlation between a …rm’ choice to s operate a centralized or decentralized R&D structure and the type of innovation undertaken. Csaszar (2008) shows that the decision-making structure in mutual funds a¤ects the number of their initiatives and errors in their decisions. Liberti and Mian (2009) present evidence that hierarchical distance in‡ uences the use of information in the decision making process in a multinational bank.

References
[1] Acemoglu, Daron, Philippe Aghion, Claire Lelarge, Fabrizzio Zilibotti and John Van Reenen. (2007). "Technology, Information and the Decentralization of the Firm." Quarterly Journal of Economics, 122 (4): 1759-1799. [2] Aghion, Philippe, and Jean Tirole. (1997) "Formal and Real Authority in Organizations." Journal of Political Economy, 105: 1-29. [3] Alchian, Armen A., and Harold Demsetz. (1972). "Production, Information Costs, and Economic Organization." American Economic Review, 62: 777-795. [4] Alonso, Ricardo, Wouter Dessein, and Niko Matouschek. (2008). "When does Coordination require Centralization." American Economic Review, 98(1): 145-179. [5] Angrist, Joshua D., and Jorn-Ste¤en Pischke. (2009). Mostly Harmless Econometrics: An Empiricist’ Companion. Princeton and Oxford: Princeton University Press. s [6] Argyres, Nicholas S., and Brian S. Silverman. (2004). " R&D, Organization Structure, and the Development of Corporate Technological Knowledge." Strategic Management Journal, 25: 929-958. [7] Arrow, Kenneth J.. (1974). The limits of Organization. New York: Norton.

33

[8] Baker, George, Robert Gibbons, and Kevin. J. Murphy. (1999). "Informal Authority in Organizations." Journal of Law, Economics, and Organization, 15: 56-73. [9] Baker, George, Michael Gibbs, and Bengt Holmstrom. (1994). "The Internal Economics of the Firm: Evidence from Personnel Data." Quarterly Journal of Economics, 109: 881-919. [10] Bandiera, Oriana, Iwan Barankey, and Imran Rasul. (2007). "Incentives for Managers and Inequality among Workers: Evidence from a Firm Level Experiment." Quarterly Journal of Economics, 122(5): 729-773. [11] Bertrand, Marianne, Esther Du‡ and Sendhil Mullainathan. (2004). "How Much o, Should We Trust Di¤erences-in-Di¤erences Estimates?" Quarterly Journal of Economics, 119: 249-275. [12] Besanko, David, David Dranove, Mark Shanley and Scott Schaefer. (2010). Economics of Strategy. John Wiley & Sons. [13] Besley, Timothy, and Andrea Prat. (2006). “Handcu¤s for the Grabbing Hand? Media Capture and Government Accountability.”American Economic Review, 96(3): 720– 736. [14] Bloom, Nick, and John Van Reenen. (2007). "Measuring and Explaining Management Practices across Firms and Countries." Quarterly Journal of Economics, 122(4): 13511408. [15] Bloom, Nick, Ra¤aella Sadun, and John Van Reenen. (2009). "The Organization of Firms across Countries." CEP Discussion Paper No. 937. London School of Economics. [16] Bolton, Patrick, and Mathias Dewatripont. (1994). "The …rm as a communication network." Quarterly Journal of Economics, 109: 809– 839. [17] Carter, Nancy M.. (1984). " Computerization as a Predominate Technology: Its In‡ uence on the Structure of Newspaper Organizations." Academy of Management Journal, 27(2): 247-270. [18] Chandler, Alfred D.. (1962). Strategy and Structure: Chapters in the History of the Industrial Enterprise. Cambridge, MA: MIT Press. [19] Coase, Ronald. (1937). "The Nature of the Firm." Economica, 4: 386-405. [20] Csaszar, Felipe A.. (2008). "Organizational Structure as a Determinant of Performance: Evidence From Mutual Funds." Working Paper. The Wharton School, University of Pennsylvania. [21] Dessein, Wouter. (2002). "Authority and Communications in Organizations." Review of Economic Studies, 69: 811-38. 34

[22] Djankov, Simon, Caralee McLiesh, Tatiana Nenova, and Andrei Shleifer. (2003). "Who Owns the Media?” Journal of Law and Economics, 46(2): 341– 81. [23] DuBick, Michael A.. (1978). " The Organizational Structure of Newspapers in Relation to Their Metropolitan Environments." Administrative Science Quarterly, 23: 418-433. [24] Gant, Jon, Casey Ichniowski, and Kathryn Shaw. (2002). "Social Capital and Organizational Change in High-Involvement and Traditional Work Organizations." Journal of Economics and Management Strategy, 11(2): 289-328. [25] Friebel, Guido, and Michael Raith. (2010). "Resource Allocation and Organizational Form." American Economic Journal: Microeconomics, forthcoming. [26] Garicano, Luis. (2000). "Hierarchies and the Organization of Knowledge in Production." Journal of Political Economy, 108: 874-904. [27] Gentzkow, Matthew, and Jesse M. Shapiro. (2006). "Media Bias and Reputation." Journal of Political Economy, 114(2): 280– 316. [28] Gentzkow, Matthew, and Jesse M. Shapiro. (2010). “What Drives Media Slant? Evidence from U.S. Daily Newspapers.” Econometrica, 78(1): 35-71. [29] George, Lisa, and Joel Waldfogel. (2003). “Who A¤ects Whom in Daily Newspaper Markets?” Journal of Political Economy, 111, 765– 784. [30] Grossman, Sanford J., and Oliver Hart. (1986). "The Costs and Bene…ts of Ownership: A Theory of Vertical and Lateral Integration." Journal of Political Economy. 94: 691-719. [31] Guadalupe, Maria, and Julie Wulf. (2010). "The Flattening Firm and Product Market Competition: the E¤ect of Trade Liberalization on Corporate Hierarchies." American Economic Journal: Applied Economics, forthcoming. [32] Hart, Oliver, and John Moore. (1990). "Property Rights and the Nature of the Firm." Journal of Political Economy, 98: 1119-58. [33] Herman, Edward, and Noam Chomsky. (1998). Manufacturing Consent: The Political Economy of the Mass Media. Second Edition. Pantheon Press. [34] Hertzberg Andrew, Jose Maria Liberti, and Daniel Paravisini. (2010). "Information and Incentives Inside the Firm: Evidence from Loan O¢ ce Rotation." Journal of Finance, 65(3). [35] Holmstrom, Bengt. (1999). "The Firm as a Subeconomy." Journal of Law, Economics and Organization, 15:74-102. [36] Holmstrom, Bengt, and Paul Milgrom. (1994). "The Firm as an Incentive System." American Economic Review, 84: 972-91. 35

[37] Ichniowski, Casey, and Kathryn Shaw. (2009). "Insider Econometrics." forthcoming in Handbook of Organizational Economics, eds. Robert Gibbons and John Roberts. Princeton: Princeton University Press. [38] Ichniowski, Casey, Kathryn Shaw, and Giovanna Prennushi. (1997). "The E¤ects of Human Resource Management Practices on Productivity: A Study of Steel Finishing Lines." American Economic Review, 87(3): 291-313. [39] Lazear, Edward. (2000a). "Performance Pay and Productivity," American Economic Review 90: 5: 1346-1361. [40] Lazear, Edward. (2000b). "The Future of Personnel Economics." Economic Journal, 110: F611-39. [41] Lazear, Edward, and Paul Oyer. (2009). "Personnel Economics." forthcoming in Handbook of Organizational Economics, eds. Robert Gibbons and John Roberts. Princeton: Princeton University Press. [42] Liberti Jose M., and Atif Mian. (2009). " Estimating the E¤ect of Hierarchy on Information Use." Review of Financial Studies, 22 (October): 4057-4090. [43] Marino, M. Anthony, and John Matsusaka. (2005). "Decision Process, Agency Problems, and Information: An Economic Analysis of Capital Budgeting Procedures." Review of Financial Studies, Spring 2005. [44] Mookherjee, Dilip. (2006). "Decentralization, Hierarchies, and Incentives: A Mechanism Design Perspective." Journal of Economic Literature, Vol. XLIV: 367-390. [45] Mullainathan, Sendhil, and Andrei Shleifer. (2005). "The Market for News." American Economic Review, 95(4): 1031– 53. [46] Prendergast, Canice. (1995). "A Theory of Responsibility in Organizations." Journal of Labor Economics, 13(3): 387-401. [47] Prendergast, Canice. (2002). "The Tenuous Trade-o¤ between Risk and Incentives." Journal of Political Economy, 110(5): 1071-1102. [48] Qian, Yingyi. (1994). "Incentives and Loss of Control in an Optimal Hierarchy." Review of Economic Studies, 61, 527-44. [49] Radner, Roy. (1993). The organization of decentralized information processing. Econometrica 62: 1109– 1146. [50] Rajan,Raghuram, G., and Julie Wulf. (2006) "The ‡ attening …rm: evidence from panel data on the changing nature of corporate hierarchies." Review of Economics and Statistics. 36

[51] Rajan, Raghuram, G., and Luigi Zingales. (1998). "Power in a Theory of the Firm." Quarterly Journal of Economics, 113: 387-432. [52] Rantakari, Heikki. (2008). "Governing Adaptation." The Review of Economic Studies, 75: 1257– 1285. [53] Roberts, John. (2004). The Modern Firm: Organizational Design for Performance and Growth. New York and Oxford: Oxford University Press. [54] Shearer, Bruce. (2004). "Piece Rates, Fixed Wages and Incentives: Evidence from a Field Experiment." Review of Economic Studies, 71(2): 513-534. [55] Simon, Herbert. (1947). Administrative Behavior: A Study of Decision-Making Processes in Administrative Organizations, 4th ed. in 1997. Free Press. [56] Simon, Herbert. (1951). "A Formal Theory of the Employment Relationship." Econometrica, 19: 293-305. [57] Van den Steen, Eric. (2010). "Interpersonal Authority in a Theory of the Firm." American Economic Review 100(1): 466-490. [58] van Zandt, Timothy. (1999). "Real-time Decentralized Information Processing as a Model of Organizations with Bounded Rational Agents." Review of Economic Studies, 66: 633-658. [59] Weber, Max. (1968). Economy and Society: An Outline of Interpretive Sociology. New York: Bedminster Press. Translated from the German version …rst published in 1922. [60] Williamson, Oliver. (1975). Market and Hierarchies. New York: The Free Press. [61] Williamson, Oliver. (1985). The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting. New York: The Free Press. [62] Zabojnik, Jan. (2002). "Centralized and Decentralized Decision Making in Organizations." Journal of Labor Economics, 22.

37

Figure 1: Organizational Structure and Allocation of Formal Authority
Panel A: Decentralization Chief Editors Panel B: Centralization Chief Editors Editing Directors Middle Managers (Division Directors and Managing Editors) Reporters Middle Managers (Division Directors and Managing Editors) Reporters

Notes: The arrow line indicates the direction of formal authority. Under decentralization, the formal authority over editorial decision is delegated to the middle managers. Each division (e.g., Economic and Business) works like an independent business unit. Under centralization, the chief editors retain formal authority, and a layer of editing directors headed by chief editors is created to monitor the editorial process more closely.

Figure 2: Timing of the Game

T0 Contract on organizatioal structure

T1 Agents acquire information.

T2 T3 Agents propose projects. Implement Under decentralization, selected projects the manager selects. Under centralization, the principal monitors and selects.

38

Figure 3: Comparison in Performance between Treatment and Control
Panel A: Log Quantity Score
7.8 .2
Ja n. 2 Ja 00 4 n. 2 Ja 00 5 n. 2 37 00 6

7.2

-.2
Ja n. 2

Difference in Log quantity -.1 0 .1

Log quantity 7.4 7.6

treatment

control

00 4

Ja n. 2

00 5

Ja n. 2

37 00 6

Panel B: Log Quality Score
7.6 .4
n. 2 00 4 n. 2 00 5 n. 2 00 6

7.4

6.8

-.1
Ja n. 2

Ja

Ja

Ja

37

Difference in Log quality 0 .1 .2 .3

7

Log quality 7.2

treatment

control

00 4

Ja n. 2

00 5

Ja n. 2

37 00 6

Notes: Panel A (and B) plots the average of the logarithm of the monthly quantity (and quality) score for the treatment and the control respectively from January 2004 to December 2006 (left panels) and the difference of the log quantity (quality) score between these two groups (right panels). The vertical dotted line indicates the timing of reform: September 2005. Treatment is the reporters from the divisions that experienced a centralization reform: Economic and Business, Politics and Law, Education and Health and General Reports. Control is the reporters from the divisions that remained decentralized: Regional and Local News, Entertainment, Consumption-Guide and Photographing.

39

Figure 4: Kernel Density of Estimated Individual Fixed Effects under the Two Organizational Forms
Panel A: Quality Residuals
.25

Quality Residuals in Treatment Group
.5

Quality Residuals in Control Group

.4

.2

0

0
-2

.05

.1

.1

.15

.3

.2

-2

0 2 4 6 individual fixed effects from lnquality regression before reform after reform

8

0 2 4 6 individual fixed effects from lnquality regression before reform after reform

8

P-value of K-S Test: 0.001

P-value of K-S Test: 0.177

Panel B: Quantity Residuals
Quantity Residuals in Treatment Group
.6
.2

Quantity Residuals in Control Group

.4

.2

0

0

2 4 6 individual fixed effects from lnquantity regression before reform after reform

8

0
-2

.05

.1

.15

0 2 4 6 individual fixed effects from lnquantity regression before reform after reform

8

P-value of K-S Test: 0.001

P-value of K-S Test: 0.595

Notes: The sample used is a balanced panel, including only the reporters observed before and after the reform and excluding the 6 reporters who switch between the treatment and the control. Reform is the timing dummy equal to one after (including) September 2005. Treatment is the reporters from the divisions that experienced a centralization reform: Economic and Business, Politics and Law, Education and Health and General Reports. Control is the reporters from the divisions that remained decentralized: Regional and Local News, Entertainment, Consumption-Guide and Photographing. The individual fixed effects are retrieved from running a regression of the log quantity score or the log quality score on the individual dummies and their interactions with the reform dummy, together with a bunch of controls including time dummies, division fixed effects, age-squared, tenure-squared, positions and qualifications as in the baseline regression. The kernel density uses the Epanechnikov kernel. The P-values of K-S test are the corrected P-values of the combined Kolmogorov–Smirnov tests of equality of distributions reported in Stata.

40

Table 1: Summary Statistics of Personnel Data
Panel A: Reporters variables gender (male) mean 0.60 min 0.00 max 1.00

education (college) 0.83 0.00 1.00

Party member 0.47 0.00 1.00

age 32.80 22.00 57.00

tenure 8.20 1.00 27.00

position (1-2-3) 1.50 1.00 3.00

qualification (1-2-3) 1.47 1.00 3.00

Panel B: Managing Editors variables gender education (male) (college) mean 0.57 0.73 min 0.00 0.00 max 1.00 1.00

Party member 0.49 0.00 1.00

age 38.30 25.00 54.00

tenure 13.30 2.00 27.00

position (1-2-3) 2.20 1.00 3.00

qualification (1-2-3) 2.20 1.00 3.00

Notes: These tables summarize personnel information of 183 reporters and 56 managing editors in the sample from January of 2004 to December of 2006. The means of the reporters’personal characteristics are weighted by monthly observations; the means of the managing editors’personal characteristics are weighted by yearly observations. Party_member is a dummy indicating the membership of the Chinese Communist Party. Tenure is the number of years of working experience in the Newspaper. Position is an indicator ranking from 1 to 3, representing reporter, chief reporter and senior reporter respectively in the hierarchy of the Newspaper. Qualification is a certificate authorized by the Association of Chinese Journalists to indicate the expertise and experience in journalism, with 1 referring to assistant journalist, 2 to journalist and 3 to senior journalist.

41

Table 2: Summary Statistics of Individual Performance Measures
Panel A: Internal Measures variables #articles #words quantity score quality score mean 32.60 18,434 2,080 1,477 std dev 21.50 13,223 1,273 1,097 median 28.00 16,188 1805 1,200 min 2.00 230 140 0 max 241.00 144,280 14,850 12,300

number of reporters: 183; number of observations: 4,461 Panel B: External Outcome Measures variables measures of news content # investigative reports # feature stories # special reports # propaganda articles # reports on government officials # advertising articles # sensational/entertaining # on-the-scene reports measures of editorial activities # articles joint with editor # articles column by content # external author (government) # external author (private sector) # external author (freelance) # coauthor (within division) # coauthor (across division) 0.98 1.27 8.84 0.53 0.49 3.75 0.68 1.96 3.01 9.63 1.82 1.34 9.16 6.20 0.00 0.00 6.00 0.00 0.00 2.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 27.00 29.00 79.00 23.00 15.00 164.00 160.00 1.42 1.00 4.88 0.32 3.89 0.51 1.14 0.71 1.62 1.35 8.19 0.90 5.08 1.16 2.60 1.41 1.00 1.00 3.00 0.00 2.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 19.00 11.00 136.00 14.00 33.00 11.00 24.00 10.00 mean std dev median min max

number of reporters: 103; number of observations: 2,446
Notes: Observations are at the individual-month level. Observations in Panel A include the reporters in all the divisions. Observations in Panel B only include the reporters in the reformed divisions, namely, Economic and Business, Politics and Law, Education and Health, and General Reports.

42

Table 3: Reporter Performance in Balanced Panel by Treatment and Reform
Panel A: Average Log Quantity Score treatment group before reform after reform difference (after-before) 7.504 (0.508) 7.513 (0.556) 0.009 (0.047) control group 7.524 (0.549) 7.516 (0.481) -0.008 (0.053) difference (treatment-control) -0.020 (0.076) -0.003 (0.077) 0.017 (0.070)

Panel B: Average Log Quality Score treatment group before reform after reform difference (after-before) 7.199 (0.598) 7.235 (0.610) 0.036 (0.043) control group 7.155 (0.647) 7.040 (0.727) -0.114* (0.062) difference (treatment-control) 0.044 (0.077) 0.195** (0.092) 0.151** (0.075)

Notes: The tables report the mean and standard deviations (in parentheses) of the reporters’ performance in terms of the logarithm of the quantity and quality scores at the individual-month level in the constructed balanced panel, which includes only the reporters who are observed both before and after the reform and excludes 6 reporters who switch between treatment and control. Reform is the timing of the organizational change from decentralization to centralization. The treatment group is the reporters from the reformed divisions: Economic and Business, Politics and Law, Education and Health, and General Reports; the control group is the reporters from the remaining decentralized divisions: Regional and Local News, Entertainment, Consumption-guide, and Photographing. The standard errors on the difference and the difference-in-differences are estimated from running the corresponding OLS regression, clustering the standard errors by individual. *** denotes significance at 1%, ** at 5%, and * at 10%.

43

Table 4: D-I-D Estimates of Average Treatment Effects of Centralization on Internal Performance Measures
Panel A: Baseline Results [1] 0.054 (0.074) -0.040 (0.052) yes log quantity score [2] [3] 0.052 0.053 (0.074) (0.067) [4] -0.066 (0.066) log quality score [5] [6] [7] 0.207*** 0.205*** 0.194** (0.072) (0.073) (0.076) -0.152 (0.057) yes yes yes 4,461 0.278 4,442 0.372 4,442 0.402 yes yes yes yes yes 4,442 0.404 yes yes 4,442 0.206 [8] 0.061 (0.070)

reform× treatment reform individual fixed effects time fixed effects time-variant covariates #observations adj-R²

yes yes

yes yes yes

4,461 0.489

4,461 0.518

4,461 0.542

Panel B: Dynamics July August 2005 2005 log 0.057 -0.006 quantity (0.066) (0.080) log quality 0.013 (0.980) -0.068 (0.110)

Reform Start -0.017 (0.074) 0.050 (0.114)

October November December Jan-06 #obs adj-R2 2005 2005 2005 onwards -0.031 0.075 -0.053 0.080 4,461 0.543 (0.078) (0.084) (0.087) (0.072) -0.023 (0.108) 0.224* (0.124) 0.103 (0.115) 0.229*** 4,442 (0.082) 0.405

Notes: Reform is the timing of the organizational change from decentralization to centralization in September 2005. Treatment is a dummy for the reporters from the reformed divisions: Economic and Business, Education and Health, Politics and Law, and General Reports. The time-variant covariates include age-squared, tenuresquared, position, qualification and division fixed effects. When a regression excludes individual fixed effects (Column [4] and [8]), time-invariant personal characteristics such as gender, education and Party membership and the factors that are collinear with individual and time fixed effects, such as age and tenure, are now included. The regressions in Panel B are based on the D-I-D specification including individual fixed effects, time fixed effects and time-variant personal characteristics, with the reform×treatment dummy replaced by a series of interactions between the timing dummies and the treatment dummy. In particular, “Reformstart” is a dummy for a reporter in the treatment group in the month of the reform (September 2005), “August2005” a dummy for a reporter in the treatment in August 2005, and “October2005” a dummy for a reporter in the treatment in October 2005. Similar definitions apply to “July2005”, “November2005” and “December2005”. “Jan-2006onwards” is a dummy for a treated reporter working from January 2006 and onwards. Standard errors (in parentheses) are clustered by individual. ***denotes significance at 1%, **at 5% and * at 10%.

44

Table 5: D-I-D Estimates of Heterogeneous Treatment Effects of Centralization on Internal Performance Measures across Task Assignment
[1] reform× 0.191*** Economic and Business (0.069) reform× 0.142 Education and Health (0.121) reform× -0.062 Politics and Law (0.107) reform× -0.299** General Reports (0.140) individual fixed effects yes time fixed effects time-variant covariates #observations adj- R² 4,461 0.515 log quantity score [2] [3] [4] 0.192*** 0.191*** 0.120 (0.069) (0.069) (0.081) 0.139 0.120 0.005 (0.122) (0.125) (0.117) -0.065 -0.031 -0.224** (0.107) (0.102) (0.098) -0.307** -0.322** -0.316*** (0.140) (0.134) (0.085) yes yes yes yes yes yes yes 4,461 0.545 4,461 0.552 4,461 0.293 [5] 0.350*** (0.081) 0.293** (0.129) -0.082 (0.082) 0.056 (0.120) yes log quality score [6] [7] 0.349*** 0.345*** (0.081) (0.083) 0.290** 0.278** (0.129) (0.134) -0.087 -0.088 (0.083) (0.091) 0.062 0.038 (0.120) (0.122) yes yes yes yes yes 4,442 0.410 4,442 0.411 [8] 0.218** (0.089) 0.171 (0.118) -0.277** (0.117) 0.086 (0.082) yes yes 4,442 0.218

4,442 0.380

Notes: The reported independent variables are interaction terms between division dummies and the reform dummy. The control group is the remaining decentralized divisions. Time-variant covariates include age-squared, tenuresquared, position, qualification and division fixed effects. When a regression excludes individual fixed effects (Column 4 and 8), time-invariant personal characteristics such as gender, education and Party membership and the factors that are collinear with individual and time fixed effects such as age and tenure are now included. Standard errors (in parentheses) are clustered by individual. ***denotes significance at 1%, **at 5% and * at 10%.

Table 6: Impact of Social Norms on the Effects of Centralization reform × treatment reform × treatment × special months (January and September) p-value of F-test on zero sum of two coefficients log (quantity score) 0.062 (0.068) -0.058 (0.054) 0.956 log (quality score) 0.220*** (0.079) -0.165** (0.080) 0.575

Covariates include individual fixed effects, time fixed effects and the time-variant individual characteristics. #observations 4,466 4,447 adj-R² 0.543 0.405
Notes: Reform is the timing of the organizational change from decentralization to centralization in September 2005. Special_months is a dummy for January and September, in which social norms condone rent seeking behavior. The standard errors (in parentheses) are clustered by individual. ***denotes significance at 1%, **at 5% and * at 10%.

45

Table 7: Comparison of Individual Fixed Effects: Exits, Stayers and Entries
Panel A: Quality Residuals treatment group exits stayers entries before reform 4.067 4.611 (1.467) after reform (1.452) 5.245 (1.360) 5.120 (1.049) control group exits stayers entries 3.442 (0.860) 3.550 (1.815) 4.033 (1.937) 5.082 (1.461)

Panel B: Quantity Residuals treatment group exits stayers entries before reform 3.353 3.915 (1.129) (0.831) after reform 3.357 (0.853) 3.231 (0.679)

control group exits stayers entries 3.004 2.760 (1.166) (1.506) 2.248 (1.495) 3.465 (1.480)

Notes: In the statistics of all the variables, the first line reports the mean values, and the second line reports the standard errors (in parentheses). Reform is the timing dummy equal to one after (including) September 2005. Treatment is the reporters from the divisions that experienced a centralization reform: Economic and Business, Politics and Law, Education and Health and General Reports. Control is the reporters from the divisions that remained decentralized: Regional and Local News, Entertainment, Consumption-Guide and Photographing. The “quantity residuals” are individual fixed effects retrieved by running a regression of the log quantity score on the individual dummies and their interactions with the reform dummy, together with a bunch of controls including time dummies, division fixed effects, age-squared, tenure-squared, positions and qualifications in the unbalanced panel as in the baseline regression. The “quality residuals” are retrieved from a similar regression with the log quality score, instead of the log quantity score, as the dependent variable. In the regressions, the standard errors are clustered by individual. The stayers exclude 6 reporters who switch between the treatment and the control. The results are qualitatively similar when these observations are included.

46

Table 8: Effects of Centralization on News Content and Editorial Activities
Panel A: News Content reporters' initiative #investigative #feature reports stories reform 0.528*** 0.332*** (0.186) (0.127) #obs adj-R² 2,446 0.224 2,446 0.238

private benefits government influence #advertising #external authors #propaganda #government articles (private sector) officials -0.479*** -0.189* 0.021 0.465 (0.139) (0.109) (0.086) (0.375) 2,446 0.535 2,446 0.707 2,446 0.191 2,446 0.655

Panel B: Editorial Activities initiative of managing editors #joint #column with editors by content reform -0.318* -0.294 (0.185) (0.279) #obs 2,446 2,446 adj-R² 0.145 0.637

external authors #freelance #public writers sector -0.065 0.605 (0.176) (0.666) 2,446 2,446 0.283 0.640

internal co-authors #within #across division divisions 2.180*** -0.569 (0.633) (0.558) 2,446 2,446 0.252 0.037

Panel C: Heterogenous Treatment Effects by Task Assignment #investigative reports #feature stories #advertising articles #joint with editors #observations Economic and Business Education and Health Politics and Law General Reports 0.847*** 0.890** -0.580 0.952 (0.258) (0.352) (0.432) (0.676) 0.213 (0.208) -1.050*** (0.289) 0.025 (0.241) 1,019 1.020*** (0.276) -0.206 (0.240) -0.354 (0.755) 345 0.175 (0.265) 0.035 (0.052) -0.777* (0.398) 628 0.468 (0.471) -0.068 (0.043) -0.197 (0.220) 454

Notes: All the regressions include individual fixed effects, time (month and year separately) fixed effects, timevariant individual characteristics defined as before. The standard errors (in parentheses) are clustered by individual. ***denotes significance at 1%, **at 5% and * at 10%.

47