The Case of Enron a Dissertation Submitted To

NETWORK STRUCTURES OF CORRUPT INNOVATIONS: THE CASE OF ENRON

A DISSERTATION SUBMITTED TO THE DEPARTMENT OF SOCIOLOGY AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

Brandy Aven May 2010

This work is licensed under a Creative Commons Attribution- Noncommercial 3.0 United States License. http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/mc724tx1889

ii I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Walter Powell, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Karen Cook

I certify that I have read this dissertation and that, in my opinion, it is fully adequate in scope and quality as a dissertation for the degree of Doctor of Philosophy.

Mark Granovetter

Approved for the Stanford University Committee on Graduate Studies. Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file in University Archives.

iii

Abstract

The focus of this dissertation is the interplay of information and social network structure. I investigate how different types of information influence how individuals choose to communicate, and the implications of these choices on the entire social system. In particular, I look at how both overt and covert information is shared across an organization’s communication network over a four-year period. The contrast between overt and covert activities affords the opportunity to examine whether different content types change the way individuals mobilize to accomplish goals within an organization.

First, I analyze how different social network positions allow individuals to obtain different types of information within an organization’s network. I demonstrate that individuals are more likely to participate in an overt communication network when they have large but sparse networks of informal ties through which to acquire information. This finding is contrasted with covert network involvement, where greater closeness centrality or shorter distances to other organizational members predicts participation.

Second, the examination of individuals who participated in both covert and overt information networks highlights the divergent behaviors associated with the two activities. When sharing overt information, individuals are more likely to encourage network cohesion, practice structural egalitarianism, and engage in reciprocal

iv communication. In contrast, when actors share covert information, they attempt to reduce network cohesion and reciprocity and foster a power structure.

Finally, I present results that show the aggregate network effects of individual strategies for sharing either overt or covert information. I find that communication within overt networks is far more reciprocal than communication in covert networks, where information flow is asymmetrical.

This study uses longitudinal email data taken from Enron Corporation between

1998 and 2002; these data provide a unique opportunity to study the communication network of the firm's employees and analyze the messages shared between employees.

In contrast to previous studies, which assume transmission of information along network ties, this dataset allows for actual observation of information transfer between organizational members. I couple social network analysis and qualitative coding to explore how the content of information affects both communication patterns and the spread of information.

Acknowledgements

I would like to thank my committee for all their support of my research. It is to

Walter W. Powell, my advisor, that I owe my greatest intellectual debt. I could not have completed my dissertation or degree without his unyielding support and excellent advice. Thanks to Karen S. Cook who was not only immensely influential in my work but also a wonderful role-model for a scholar. Mark Granovetter also provided the perfect balance of encouragement and criticism at every stage of my research. I am also grateful for the guidance and friendship of Henning Hillmann.

I am so very thankful for all those who provided me with personal and academic support in both the Sociology Department and Graduate School of Business at Stanford. The faculty generously provided their time and insights, which played a critical role in my growth as a scholar. Special thanks to Shelley Correll, Paula

England, Susan Olzak, Paolo Parigi, Cecilia Ridgeway, Nancy Tuma, and, of course,

Morris Zelditch. I also have to thank the department staff who helped me undertake both the mundane and life-altering challenges of my career at Stanford.

I am immensely appreciative of all my colleagues and friends for reading drafts, giving support, and providing astute comments and cheer: Anastazia Older

Aguilar, Kristen Backor, Sara Bloch, Curtiss Cobb, Lynn Chin, Jonathan Haynes,

Jung-eun Lee, Elizabeth McClintock, Amanda Sharkey, Alicia Simmons, and Kaisa

Snellman.

Thank you to all at the Social Science Resource Center, especially my boss,

Patricia Box, who became a wonderful friend. I learned a great deal from my Social

Science and Data Service colleagues and cannot imagine my time at Stanford without them.

A big thank you to my dearest friends outside of academia for all your encouragement and grounding humor about this endeavor: Dylan Kendall and Larissa

Weingart.

I would like to thank my parents, Leo and Freny Berkenbile, for all of their understanding and uncompromising support. I am so grateful for my sister, Tiana

Russell, who provided love and a willingness to listen to my sociological jargon. A big hug to my brothers, Bryan and Eric Berkenbile. My cousin, Jennifer Jiries, gave me her reliable sounding board and keen editing eye. I am very grateful for my lovely aunt, Dinaz Shroff, who along the way gave me so much love and too many care- packages.

Finally, and most importantly, I would like to thank my husband, William

Peter Aven. Pete, you never wavered in your dedication to my goals and dreams. I could not have done any of this without your strength and love. You have provided me with so much, and all of my accomplishments are also yours. It is to him and our amazing daughter, Parker Trevi Heera Aven, that I dedicate this dissertation. Trevi, you are an unending supply of inspiration.

Stanford California, Summer 2010 Brandy Lee Aven

vii

Table of Contents

Abstract...... iv Acknowledgments...... vi List of Tables...... xi List of Figures...... xii Chapter 1 – Introduction ...... 1 Organizational Setting ...... 3 Objective of the Dissertation ...... 5 Social Networks and Organizations ...... 7 Social Networks and Information Content ...... 8 Outline of the Dissertation ...... 12 Chapter 2- Theory and Hypotheses ...... 15 Information Typologies: Overt and Covert ...... 18 Organizational Legitimacy ...... 22 Innovations& Diffusion ...... 24 Information Attainment ...... 27 Social Influence ...... 29 Strategic Interactions ...... 32 Study 1: Diffusion and Recruitment ...... 34 Study 2: Strategic Interaction ...... 38 Study 3: Aggregate Network Structures ...... 42 Overt Innovation Networks ...... 42 Covert Innovation Networks ...... 44 Conclusion ...... 45 Chapter 3 – Research Setting and Methodology ...... 46 The Setting ...... 47 Corrupt Innovations ...... 50 Legitimate Innovations ...... 53 The Email Corpus ...... 54

viii

Organizational Member Attributes ...... 59 Innovations ...... 61 Methodology ...... 62 Research Design ...... 66 Social Network Analysis ...... 67 Qualitative Coding ...... 68 Panel Data ...... 69 Recruitment Models ...... 69 Simulations ...... 69 Limitations of the Data ...... 70 Chapter 4 - Covert and Overt Network Participation ...... 72 Introduction ...... 72 Network Positions: Brokers & Magnets ...... 73 Measures ...... 75 Data ...... 77 Methodology ...... 80 Results ...... 81 Conclusion ...... 88 Chapter 5 – Individual Behavior in Covert and Overt Networks ...... 89 Introduction ...... 89 Data and Methodology ...... 96 Results ...... 100 Qualitative Coding of Email Messages ...... 101 Conclusion ...... 107 Chapter 6 - Covert and Overt Content Networks ...... 109 Introduction ...... 109 Proposed Network Structures ...... 111 Overt Innovation Networks ...... 112 Covert Innovation Networks ...... 113 Data and Methodology ...... 116

Results ...... 117 Conclusion ...... 123 Chapter 7 – Networks, Gender and Whistle-blowers ...... 125 Organizations, Social Networks and Gender ...... 125 Whistle-blowers ...... 127 Conclusions ...... 133 Chapter 8 - Summary of Findings and Conclusion ...... 135 Conclusions ...... 135 Future Research ...... 139

List of Tables

Table 3.a. Individual Email Accounts 56

Table 3.b. Enron Employee Corpus Seed Sample Characteristics 57

Table 3.c. Enron Employee Innovation Participants Descriptives by Content Type 59

Table 4.a. Enron Adoption Descriptives 78

Table 4.b. Measures by Year 84

Table 4.c. Correlations among Selected Independent Variables for Corrupt Adoption 85

Table 4.d. Correlations among Selected Independent Variables for Legitimate Adoption 85

Table 4.e. Conditional Logit Estimates of Legitimate and Corrupt Innovation Adoption by Enron Employees, 1998-2002 86

Table 5.a. Number of Participates in Both Legitimate and Corrupt Innovations 96

Table 5.b. Correlations among Innovation Adoption: Enron Email Networks 1998-2002 98

Table 5.c. t-test for Dual Adopters Ego Communication Networks 100

Table 5.e. Confidential Messages by Innovation Network 104

Table 6.a. Innovation Network Descriptives I 118

Table 6.b. Innovation Network Descriptives II 119

Table 7.a. t-test for Gender Differences Network Characteristics 126

Table 7.b. Measures for Sherron Watkins 132

Table 7.c. Measures for Vince Kaminski 132

List of Figures

Figure 1.a. Dynamic Relationships of Studies Presented 11

Figure 2.a. Network Representation of Broker and Magnet 36

Figure 2.b. Simmelian Triad 38

Figure 2.c. Divide et Impera 39

Figure 2.d. Hierarchy 40

Figure 2.e. Least Upper Boundednesss 41

Figure 2.f. Connectedness 42

Figure 3.a. Enron’s Use of Special Purpose Entities 48

Figure 3.b. Directed and Undirected Network Based on an Email Message 64

Figure 4.a. Percent of Alters who Adopt by Innovation (1998-2002) 76

Figure 5.a. Individuals who Participated in Both Legitimate and 96 Corrupt Innovations

xii

Chapter 1 – Introduction

Organizational communication networks are powerful determinants of information distribution (Kogut and Zander 1992; Obstfeld 2005; Tushman 1977).

Within an organization, certain individuals may be better positioned within the network of relations than others to receive information. As a result, a person’s relationships and associations within the organization determine, at least to some degree, the information she receives or has access to. Location is not the only factor, however, that determines to whom information spreads. Content is central as well. In other words, if we consider the social structure of relations between individuals to be channels through which information passes, then the content is what flows through those channels. The particular characteristics of the information that flows through the network then influence where it will travel. By comparing content-specific communication networks, I demonstrate that it is not simply the nature of the relations that influence network formation but also the information transmitted between individuals.

Most network theorists recognize that different types of relationships (e.g., a friend, an advisor) -- technically termed tie content, influence individual actions differently (Gibbons 2004; Ibarra and Andrews 1993; Podolny and Baron 1997;

Umphress, Labianca, Brass, Kass, and Scholten 2003). Nevertheless, researchers have not systematically compared the structural properties of networks based on the type of information. Further, little is known about the mechanisms that lead to structural differences in content-specific sub-groups or about the factors that influence people to

take part in these sub-groups. Earlier studies have suggested that certain kinds of information lead individuals to be strategic in information transmission and acquisition. For example, in Lee’s (1969) The Search for an Abortionist , abortions were illegal, and neither the doctors who performed abortions nor the women who sought them could do so overtly. Similarly, when psychotherapy was much less accepted and ubiquitous, people searched for therapists through intimate social circles

(Kadushin 1966). In both cases, the sensitivity of the topics required individuals to be selective and careful about sharing and soliciting information. Adding the nature of information communicated between individuals as a determinant of network form brings a new dimension to social network analysis.

The communication processes described here are dynamic in that the set of social relations between individuals affects the message transmission, and message transmission, in turn, affects the set of relations shared between people. As an example, an individual may choose to share personal information with certain existing relations, which reinforces those pre-existing ties. In contrast, neglecting to share personal information with relations can lead to the dissolution of ties. Actors construct and reconstruct their networks through their communication and individual behaviors.

I assert that the specific information transmitted through network structures drives different behaviors that serve to reinforce and restructure particular network configurations. In this dissertation, I explain how the content of what is being communicated and shared among employees plays a large role in shaping networks of interpersonal relations. I extend the concept of tie content to include the type of

information shared between individuals and examine the communications between the organizational members to determine the impact on social structure, looking specifically at the dichotomy between overt and covert information. The particular focus here is how the interaction between content types and the observed social structure determines the formation of and participation in content-specific sub- structures.

Organizational Setting

Although it may be argued that the organization of interest in this dissertation,

Enron Corporation, was pathological in that its members were committing fraud, I argue that it was not the entire organization but a few rogue individuals who were able to coordinate the efforts of some key members to commit fraud. The employees at

Enron were not markedly different from other US employees, but the pressure Enron put on its members may have been more extreme.

Enron had been the darling of Wall Street for 10 years, and by the late 1990s, it was feeling the strain of maintaining that legacy. Additionally, historical changes were being made in the regulation of the energy sector and accounting practices. These changes presented opportunities for Enron to hide debt and inflate profit reports. For this reason, I argue that the corruption that sprang from Enron was in fact an innovation. 1 I do not intend to condone fraudulent behavior, but rather to highlight the

1 This is Merton’s (1968) definition of innovation, where the individuals use non- sanctioned methods to gain socially prescribed goals.

fact that the pressure for organizational success may have driven individual members to exploit certain opportunities, albeit unethical ones, to meet that goal.

Enron Corporation began in 1985 as a small company in Nebraska, but within ten short years it emerged to become an energy giant based in Houston, Texas. In the late 1990s, it rapidly expanded and altered its goals to include buying and developing assets such as pipelines and power plants. This model posed some challenges for the company. In order to grow along these lines, Enron needed large initial capital investments that would not generate earnings or cash flow in the short term. This, in turn, placed immediate pressure on Enron’s balance sheet in terms of performance. In addition, Enron already had a substantial pre-existing debt load. This debt made it difficult for Enron to maintain high credit ratings at investment grade, which was crucial to their energy trading business. The solution that emerged for the energy company was to find outside investors to enter into the partnerships, largely in the form of Special Purpose Entities (SPEs). The SPEs allowed Enron to include joint ventures as assets and shift debt liabilities off of its own balance sheet.

The accounting literature at the time provided very limited guidance as to when

SPEs should be consolidated into the balance sheet of the “parent” company. Treating these investments as “off-balance-sheet” and not consolidating them was preferable for Enron because it enabled the company to present itself more attractively to Wall

Street analysts and rating agencies. This provided the company with a gray area to exploit and allowed them to succeed with their current business model, but it also made it easy to commit fraud. There was very little federal or state oversight of such

financial vehicles and outside auditing firms, such as the firm formerly known as

Arthur Anderson, are used to ensure corporations stay in alignment with accounting principles and laws. In late 2001, it was revealed that Enron had hidden massive amounts of debt illegally using transactions involving these SPEs. In addition, major conflicts of interest were created by the partnerships between these SPEs and Enron, and several of Enron’s top executives and their family members were generously compensated for “managing” these partnerships.

Objective of the Dissertation

This dissertation does not aim to defend the illegal and fraudulent actions of those individuals involved in the Enron scandal, but rather seeks to understand how these corrupt endeavors were carried out. This dissertation also does not seek to explain why corruption emerged at Enron, but how the decisions of the individuals involved in that corruption created variation in the social structure.

I examine longitudinal email data plus supplemental documentation such as meeting minutes and project plans taken from Enron between the years 1998 and

2002. The network information and correspondence was drawn from the Enron Email

Corpus (EEC), a collection of emails subpoenaed and made public record by the

Federal Energy Regulation Commission (FERC). I use sources like Enron’s annual reports, individual interviews, public statements, and testimonies from the Securities and Exchange Commission (SEC) and the Department of Justice (DOJ) to enrich these

EEC data.

Two factors in particular make this dataset valuable and appropriate for the systematic study of information and social structure. First, the dataset provides an opportunity not only to study the social network structure of the organization’s employees, but also to analyze the messages shared between them. In contrast to previous network studies that have assumed the transmission of information along network ties, my dataset allows for the actual observation of information transfer between organizational members (Burt 1992; Podolny and Baron 1997; Wasserman and Galaskiewicz 1994). The EEC provides a unique opportunity to investigate and compare the spread of different information types, as employee emails include both professional and personal messages. Second, the EEC provides an ideal setting to compare covert and overt information because the time span includes the inception of

Enron’s fraudulent partnerships and accounting practices as well as Enron’s legal endeavors and innovations. This setting permits the systematic comparison of different types of information and their effects on network structure.

With these features of my dataset in mind, I first investigate the mechanisms that trigger the diffusion of information to organizational members. I propose that overt information diffuses more readily to individuals in brokerage positions in the organization, where individuals with many close connections in the organization are more likely to gain access to covert information. Second, I argue that individual communication patterns vary based on the type of information they are sharing and discussing. These individual choices and behaviors aggregate to have structural implications, leading to different graph-level outcomes for the network.

Social Networks and Organizations

Organizations generate social networks by providing individuals with a setting in which to develop contacts and ties (McPherson and Smith-Lovin 1982; 1987).

Organizational communication networks are comprised of the interactions shared between individuals and reveal the underlying social structure within an organization.

The social network ties act as conduits for communication that can channel social processes and information; the type of information communicated by an organization’s members can range from task-specific information to descriptions of weekend activities. However, why individuals choose to share certain information with certain individuals is not well understood.

Within an organization’s network, particular groups of actors may be organized within different content-specific sub-structures, and the features of these sub-structures will influence the patterns of information flow. For example, individuals may generate a cohesive communication network about a hobby within the organization, such as a company bowling team, and the high connectivity this creates between actors can improve access to other types of information, such as upcoming promotions.

The organizational members themselves may also attempt to manipulate the transfer of information within a communication network in order to influence outcomes and others’ behavior. Even though members of the organization may be advocates for the spread of an idea, individuals might not want to cooperate to ensure that all parties within the organization have equal access to the information for a

number of reasons. In some settings, individuals or groups may try to hide information regarding an idea to secure a competitive advantage, even within the same organization. Secrecy may also be beneficial when the perception of risk surrounding the adoption of an innovation is too high for most in the organization to condone, or when the innovation is deemed unethical, such as with corrupt innovations.

Individuals may discriminate with regard to who they share information with, seeking to inform only particular friends or strategic partners. In these cases, to implement an innovation, new adopters will have to be recruited carefully so as not to risk jeopardizing those involved and making them vulnerable to condemnation. Clearly, within an organization the information shared between members will travel through the social structure differently depending on the information’s content.

Social Networks and Information Content

When studying how a new message (in the form of information or behavior) spreads from one person to the next through social systems, past research has largely treated different forms of message content as homogeneous, focusing mainly on the network structure that surrounds an individual (adopter) to determine what makes the message spread. In contrast, epidemiological diffusion research has been advanced by taking into account particular variations in disease transmission. For example, early epidemiological models assumed that all contaminations were random events without structural predictors of contagion. These early models of contagion begin with a slow start, followed by an exponential increase and then end with a decline in transmission, resulting in a classic S-shaped diffusion curve (Bailey 1975). Such models remain somewhat applicable for viruses that are highly contagious and airborne. However,

more recent research on sexually transmitted diseases has benefited from taking into account particular characteristics of the disease, such as the means by which it is transmitted, when predicting contagion spread, generating new models with greater predictive power (Bearman, Moody, and Stovel 2004; Dezso and Barabasi 2002;

Moody, Morris, Adams, and Handcock 2003). In other words, the characteristics of the disease and how it is transmitted can act as determinants of the diffusion pattern.

Similarly, understanding variation in information types and underlying mechanisms of adoption of behaviors may help improve models of social influence.

The fact that network literature has, for the most part, ignored the type of information is due in part to the difficulty of acquiring messages or information shared between individuals (Damanpour 1991; Hage 1999), often due to technical reasons and/or privacy issues. While the advent of new communication media such as email and instant messaging provides a means to capture complete messages shared between two or more individuals and map individual, group, and organizational social networks, privacy restrictions on these types of data make them difficult to attain for study. Thus, scholars have had little opportunity to study actual content effects on the social influence processes and patterns of diffusion. This dissertation diverges from most social network studies because the data I use permits the study of messages shared between individuals.

I distinguish between overt and covert information using the instantiations of legitimate and corrupt innovations within an organization and argue that the communication strategies that individuals employ will lead to variations in the larger

network structure. There are many different typologies that one might use to demonstrate the effects of content on structure. I have chosen overt and covert information because the very nature of the content makes it clear why individuals would use different strategies for sharing the information. With overt information the content is public and not constrained and the opposite is true for covert information, which is private and secret. When information is meant to be kept secret, the process by which people would share or discuss the information requires different behaviors and strategies than if the information is public. In the case of overt content, sub- structures are likely to be innocuous or even beneficial to the organization. On the other hand, covert content sub-structures may undermine the health of the organization not only because of their inappropriate objectives, but also through their effects on aggregate network shape. Individuals involved in the corrupt innovation may not only be acting illegally, but also creating fissures within the informal network that limit information flow throughout the organization.

I start with individual involvement, focusing on the individuals who, given their position in the overall network structure, are most likely to participate in a communication network or gain access to either type of information. The idea is that since the character of the information is different, the way in which individuals discover it will also vary. Next, I explore the different communication strategies used by individuals within their egocentric networks when they share corrupt or legitimate information. I compare both quantitatively and qualitatively the behaviors individuals use to share overt or covert information in their local communication structures.

Finally, I look at the aggregate effects of these micro-level decisions in shaping network sub-structures.

This dissertation addresses three specific questions, listed below:

Study 1: What are the structural factors that lead an individual to participate in a legitimate or corrupt network?

Study 2 : How does content change the way individuals communicate within their networks?

Study 3: What are the similarities/differences between legitimate and corrupt networks?

The diagram below displays the three central studies and their relationships graphically (see figure 1.a). It presents an overview of the proposed differences between the two content types and how these differences lead to different outcomes at the level of the individual and the group.

Overt Information Covert Information

Individual Individuals with large Individuals with Acquisition sparse networks (i.e., increased proximity to brokers). others in the network plus access to someone “in the know”.

Individual Relations will be Relations will be Behavior in symmetrical, cohesive, asymmetrical and sparse Ego- and egalitarian. with a hierarchical Networks power order.

Social Highly connected Poorly connected Structure structure with a majority structure with a majority of reciprocated ties. of asymmetrical ties.

Figure 1.a. Dynamic Relationships of Studies Presented

Outline of the Dissertation

This dissertation is organized as follows. First, I examine work on communication networks, information attainment, and social influence and link it to the broader literature on social network structure. I follow with a description of the research setting, Enron Corp., and discuss my methodology. In the first study, I contrast participation in overt and covert communication sub-structures. I then examine individual differences in communication decisions based on overt or covert information. I discuss the effects of information type on social networks within organizations and demonstrate that network form varies based on whether the

information is overt or covert. Specifically, I investigate how information influences individual behavior within the social structure. In the subsequent section, I describe the aggregated graph-level characteristics for the different content types. I provide a brief conclusion concerning the implications of my findings in the final section.

Chapter 2 – In this chapter, I detail the theoretical underpinning of the main hypotheses that I put forward in the dissertation. Several areas of research are addressed and then interwoven to create theoretical arguments that address my research questions.

Chapter 3 – Here I describe my research setting, Enron, and discuss the methodological framework I employed to analyze the data and test the hypotheses I developed based on my main research questions.

Chapter 4 – In this chapter, I explore the structural characteristics that lead to the acquisition of either overt information or covert information within the organization. The focus is on the network positions of individuals before they participate in either a covert or overt communication network.

Chapter 5 – For this chapter, I compare the different strategies individuals employ when communicating overt or covert information. I demonstrate both quantitatively and qualitatively how content differences lead to variations in an individual’s communication behavior.

Chapter 6 - In this study, I develop an argument and analyze the differences between the network topologies of overt and covert networks.

Chapter 7 - In this chapter, I look at individuals who chose to share covert information in the hopes of stopping it – Enron’s whistle-blowers.

Chapter 8 – Finally, I summarize my main findings and discuss the broader implications of my research.

Chapter 2- Theory and Hypotheses

In this chapter, I review the pertinent literature from several different areas of research and set-out my theoretical arguments. I draw on various distinct research areas to motivate my hypotheses. First, I rely on the organization literature to understand the context for my research setting, Enron Corporation, and the motivations and constraints of the organizational members. Next, I turn to the social psychology literature to provide mechanisms for conceptualizing decisions and behaviors of the organization’s members. I then incorporate findings from social network research to understand the implications of individual action for the network of organizational members and vice versa. Finally, I refer to research on innovations because it sheds light on the particular content of interest in this study. Although these are separate lines of research, I believe that they intersect and synthesize well to provide a robust view of communication networks within organizations. Nonetheless, each stops short of fully explicating the link between the individual’s decisions and the network of social relations that emerges. I contend that information type is the critical piece that simultaneously refines and connects these theories.

Network theorists recognize that different types of relationships or tie content, such as friendship or advisor relations, influence individual actions differently

(Gibbons 2004; Ibarra and Andrews 1993; Podolny and Baron 1997; Umphress,

Labianca, Brass, Kass, and Scholten 2003). I extend this line of research by comparing content-specific sub-structures and argue that it is not simply the nature of the relations that influences behavior, but also the information transmitted between

individuals. 2 Moreover, I argue that how individuals communicate certain types of information leads to structural differences in content-specific sub-groups. By including the information communicated between individuals as a determinant of the network form, my model departs from traditional social network analysis in which the relations or ties themselves are the primary, if not exclusive, focus of investigation.

This paper presents a theory for the micro-strategies that organizational members use to share information that then have aggregate effects on the overall network structure.

Information type coupled with network position affects information attainment, individual participation, and group relations. These differences emerge from the choices members make when sharing information with other members.

The social network and qualitative analysis I use here allow for the actual tracing of specific patterns of information through the system. This analysis permits an understanding of the social network via these processes of communication. Pre- existing relations act as channels that determine information flow. The arrangement of relations among different individuals influences the course that a message will take within the system. The deliberate sharing or withholding of particular types of information can qualitatively alter the relation, as well as its strength. For example, the exclusive sharing of information can strengthen and reinforce a relation, while

2 Content-specific sub-groups or sub-structures are the communication networks about a particular topic. For example, when individuals share information about annual bonuses, the specific people that discuss bonuses and the messages shared comprise the network.

withholding information or neglecting to communicate information can weaken it or lead to its dissolution.

It is difficult to understand communications or social networks without considering information. Information is a fundamental concept for studying relations between individuals. Individuals communicate to create, share, alter, and validate information and these processes help individuals to reach a mutual understanding.

Since social networks are comprised of relational connections along which individuals communicate, understanding information is critical to conceptualizing social networks. The central goal of this dissertation is to investigate the interplay between content and networks of social relations. Specifically, this research analyzes how particular types of information interact with the networks of relations within an organizational setting. The two central questions I ask are: how do networks of communication affect the attainment of different types of information, and, conversely, how does the transmission of certain types of information affect the communication network?

A communication network is comprised of interconnected individuals linked by the sharing of information. Individuals choose what information to share and with whom to share it. These decisions have consequences for both the ego-networks, the individual’s set of relationships, and the aggregated network form, the collective set of relationships. One of the benefits of communication network analysis is that it allows us to see both the interpersonal processes of influence and the effects of the social network structure, such as the diffusion of ideas among the members of a system. The

analysis of communication networks provides a means to understand relationships between people and their social positions and characteristics of the community and its interconnectedness.

In the next section, I discuss the two distinct types of information investigated in this dissertation and explain the implications for social networks.

Information Typologies: Overt and Covert

I distinguish two types of information, overt and covert , which highlight the role of information within social networks. Overt information can be conceived of as information that is public and unrestricted to any interested individual. 3 Within and between groups there are no restrictions on who is allowed to access the information, whereas covert information is secret and only intended to be known by a select few.

Covert information is not permitted to pass freely to all the individuals in the system.

In many cases, covert information also includes concealing the fact that the information exists.

In this study, I compare legitimate organizational endeavors to corrupt activities as the instantiations of overt and covert information, respectively. I propose that both network form and the conditions that lead individuals to participate in a content-specific network vary based on the degree to which the content is perceived to be legitimate for the organization. Using the contrast between legitimate content, or content that can be viewed as pertaining to the organization’s core norms and accepted

3 This is not to say that it is accessible to every member because individuals may not be aware of the information or from whom to obtain it.

operations, and corrupt or illegitimate content, which are organizational activities that can be deemed fraudulent or run in contrast to the organization’s procedures and norms, I ask whether the content type changes the way individuals organize to accomplish a goal within an organization. Corruption and illegal networks have been of interest to many scholars, but previous research has been remiss in evaluating it in terms of other types of networks of interaction (Baker and Faulkner 1993; Simmel

1950).

The interplay of content and network characteristics can result in various outcomes for both the organization and its members. If members share content that is legitimate or overt, it can have positive implications for the organization and the individual, such as engendering greater interconnectedness between organizational members; evidence suggests that the more connected a social group, the more willing they are to trust one another (Coleman 1988; Hardin and Cook 2001). This, in turn, can improve organizational outcomes in the face of a crisis (Krackhardt and Stern

1988).

At the individual level, if we assume that the likelihood of transmission increases when individuals share a strong tie to each other, then information is more likely to diffuse and spread more rapidly within a cohesive communication network, where there is a high degree of interconnectedness between individuals. Cohesive communication structures also have implementation benefits, such as helping organizational members coordinate task interdependencies and encouraging cooperation between individuals. In Coleman’s (1988) view, cohesive or closed

networks promote willingness to cooperate and mutual trust either through group solidarity or monitoring. 4 Closed networks are comprised of many strong ties; because of this, network members are assumed to share higher degrees of trust with each other

(Hardin and Cook 2001). Within closed networks, communication flows easily and swiftly; therefore, every member is likely to share similar information (Granovetter

1973). For the success of the innovation implementation, this shared level of understanding is beneficial for coordination efforts and adaptation.

On the other hand, if we consider corrupt or covert content, individuals can act in a way that does not reflect the organization’s goals and may undermine its effectiveness and survival. Members who share corrupt information may strive to keep it secret and hidden from the surrounding members. For example, Simmel (1950) claimed that the goal of secret societies is to conceal themselves and their members from detection. In order to maintain secrecy, such groups may act in a manner that weakens the organization, such as deliberately hindering information flow within the organization. Covert content networks may also lead to the fractioning of communication links between groups. These cleavages can severely handicap the organization in its day-to-day operations because they create information road-blocks that can stymie work processes.

For covert endeavors, individuals may be clandestine and strategic in sharing the information. When uncertainty and risk are high, control between individuals

4 In terms of networks, closed and cohesive are often used synonymously. I use cohesive here to mean interconnectedness. I refer to closed when the network is interconnected with strong ties.

becomes increasingly important. In such high-risk situations, individuals may only seek out partners with whom they share a high level of trust or that they have the ability to control and monitor. Baker and Faulkner (1993) investigated conspiracies in heavy electrical equipment industries and revealed that centralized communication is critical for organizational conspiracies to be carried out. Evidence from the Maxi

Trials in the mid-1980s suggested that central to the Sicilian Mafia was the so-called

“Cupola” at the highest level of the organization. The “Cupola” consisted of a cohesive core of leading mafia members who held large amounts of information about the organization itself and enforced the policies needed for the organization to function (Stoler, Galling, Suro, and Kalb 1984).

Even terrorists are aware of the importance of network structure in ensuring secrecy. For example, Mousab al Suri (aka Mustafa Nasar the Syrian, an alleged Al

Qaeda affiliate) in a video lecture captured after the fall of Afghanistan in 2001 discusses the structure that a covert organization should adopt (Cruickshank and Ali

2007). Emphasizing network forms by drawing a diagram, he argued for a disconnected communication network to protect the movement. Otherwise, as he states, “In case you are caught, they are all caught” (Cruickshank and Ali 2007: 8).

Covert information, however, is the not the sole domain of criminals and terrorists.

Law enforcement and military procedures must often take secrecy into account, as in undercover investigations or counter-intelligence activities.

Next, I discuss the concept of organizational legitimacy as a means to distinguish between overt and covert information types.

Organizational Legitimacy

As indicated earlier, I distinguish between legitimate and corrupt innovations in an organization as examples of overt and covert information. Legitimacy may be thought of as the view or assumption that a particular behavior or opinion meets some minimum constraint. Although legitimate administrative innovations may take a number of forms, they must fall within the accepted practices of the organization and be considered acceptable by a majority of appropriate members. Moreover, legitimate innovations or practices should be perceived as mapping to the organization’s goals and objectives. Organizational legitimacy permits members to be overt with information regarding their legitimate activities. 5

These practices need not be formalized, but should be held by a majority of individuals and be considered “taken for granted” (DiMaggio and Powell 1983).

Similarly, legitimacy of a practice comes from the prevalence of acceptance among those who occupy similar positions (Suchman 1995). Deciding to adopt an innovation entails a risk in which the benefits and costs are unknown. When individuals are uncertain about the proper response to information, they draw upon others to define a socially acceptable interpretation of risk.

Like overt information, covert information takes various forms; covert information includes activities that are not deemed appropriate for the organization’s

5 There is a rich and complex literature on the topic of legitimacy that I fail to fully explicate here. Instead I simply use the term to mean acceptable to the majority audience of organizational members.

sector or that exceed the level of risk normally tolerated by the firm. Of particular interest here are those covert innovations that may be deemed corrupt. Granovetter

(forthcoming) suggests that for a behavior to be deemed corrupt, a larger audience

(beyond those involved) must view the action as inappropriate by some standard.

These standards usually take into account the responsibilities of those involved as well as who stands to gain or lose from the action. Given that corrupt innovations can summarily be deemed covert, they serve as an excellent test case for understanding variations in networks and individual behavior based on information type. In this study, I use the designation of corruption by a government agency to qualify innovations as corrupt. 6

Powell (1996) suggests that the administrative structure of an organization reflects the varied environmental constraints and opportunities that it confronts. He argues that in complex and uncertain environments, organizations have more opportunities to innovate. He further notes that organizations vary in their responsiveness, either by compliance or interpretation, to their legal environments.

Given these variations in the interpretation of and compliance with the laws, some organizational settings may be more vulnerable to “creative” interpretations of law and thus, the adoption of corrupt innovations, than others.

Additionally, highly competitive organizations that put great emphasis on success, regardless of means, may create settings in which individuals feel that acting

6 Although I use the finding of corruption after the fact, there is evidence that the individuals were fully aware that their endeavors were fraudulent. I discuss this in greater detail in subsequent chapters.

fraudulently in pursuit of success is acceptable. 7 In the case of Enron, tensions arose between the organization’s goals and the legitimate means to achieving those goals.

Enron’s existing business partnerships and innovations were not producing the profits that the corporation and Wall Street analysts had come to expect. In addition, accounting practices and laws regarding these new administrative processes were vague at best.

In the section that follows, I describe why innovations and their diffusions provide a compelling case to understand how different types of content interact with patterns of social relations.

Innovations& Diffusion

The diffusion of innovations provides an opportunity to understand how content and social networks interact within an organization. Innovations or new practices are, by their nature, not integrated into the existing practices of the organization (Rogers 1962). Organizational members require access to information in order to learn about a particular innovation and to evaluate whether they should adopt that innovation. Innovation adoption requires that the individual first be aware of the innovation; the individual then must overcome some degree of uncertainty and possibly resistance. When individuals find it difficult to make a decision, such as whether or not to take part in an innovation, they are more likely to look to others who are proximate or share a similar role (Burt 1992; Coleman, Katz, and Menzel 1966;

7 This parallels Merton’s concept of innovation. The innovator accepts the socially prescribed goals of the organization but rejects the legitimate methods for achieving them; this, in Merton’s view, is what leads to criminal behavior.

Festinger 1954). Thus, observing individual decisions regarding involvement in an innovation presents an excellent opportunity to study the interplay of social networks and content.

Particular communication structures are more conducive than others to collaboration and solving non-routinized problems that are common to the implementation of an innovation. Bavelas (1950) discussed how different forms of communication facilitated the information flow needed to complete a task. He identified a number of small group communication configurations, including chain, circle, wheel, and “comcon” (completely connected network). Following Bevels’ studies, Leavitt’s (1951) also studied information processing for groups. One important finding of their work and the great deal of research that followed it was that decentralized communication structures perform better when the task requires collaborative problem-solving or creativity, such as in the case of an innovation.

Although most social diffusion research to date has analyzed technological innovations, an innovation may also simply be a new idea or practice (Rogers 1962).

Distinguishing between types of organizational innovations is advantageous for understanding adoption patterns in organizations. Past research has advanced many typologies for distinguishing innovation types; due to the nature of the innovations addressed in this study, the distinction most applicable is that made between technical and administrative innovations . Administrative innovations pertain to organizational structure or administrative processes and are related more directly to the organization’s management than to its basic organizational activities (Damanpour

1991; Downs and Mohr 1976; Knight 1967). Administrative innovations are similar to

Schumpeter’s (1934) description of organizational innovations, which includes new organizational forms with new internal structures where activities and processes take place (Have and Toivonen 2007). New organizational accounting practices or employee incentive programs can be considered administrative innovations.

Administrative innovations require organizational resources and cannot be implemented outside the organization. In contrast, technical innovations involve the creation and production of new products (Damanpour and Evan 1984; Damanpour

1991; Kimberly and Evanisko 1981; Knight 1967). The focus of this study is on administrative innovations because novel partnership arrangements and new financial vehicles were commonplace innovations at Enron.

Organizational innovation can be defined as the adoption of an idea or practice new to the organization (Damanpour and Evan 1984; Damanpour 1991; Hage 1999).

This study focuses predominantly on administrative innovations that require more than one individual for implementation. Organizational innovations require a team effort rather than the involvement of just one individual because they are improvements to organizational routines that usually span across sub-organizations and individual member expertise. An entire organizational process is rarely, if ever, contained within one employee’s role. Generally, such processes require the coordination of several individuals and roles.

Next, I propose a process of diffusion at the individual level and how this process may be affected by both patterns of relations and content of the information.

Information Attainment

Information type and network position can be conceptualized as affecting behavior in two sequential ways. First, the individual’s knowledge of or lack of information obviously affects her decisions. Second, the individual’s position and relationships with others can influence her decision in a particular direction (Friedkin

1990). For example, consider the purchase of a firearm: in one instance, consider the purchase of a legal hunting rifle, and in the other, the purchase of an illegal handgun.

If an individual were on the market for the legal rifle, she could readily acquire information about the gun from hunting goods salesmen and/or friends who hunt. On the other hand, if a person were looking to purchase an illegal gun, the information would be constrained and she would have to be careful and discreet in acquiring it.

She would have to inquire with certain individuals she trusted, and only individuals who trusted her in return would be likely to share that information. The decision to purchase a gun is also tightly coupled with information type and network position. An individual is more likely to purchase an illegal gun if she knows that many of her close friends also own one (McPherson and Smith-Lovin 1987). On the other hand, having only one distant acquaintance who owns an illegal gun would not likely influence the individual’s decision.

First, I assert that type of information will affect to whom the message spreads and certain network positions are more likely to receive particular types of information. The next step in the micro-mechanisms of diffusion, after information

attainment, is the adoption of a behavior or acceptance of a practice. This can happen in the head of the actor, such as with the acceptance of certain practices, or it can take the form of engagement in a new behavior or action. The transfer of information along interpersonal ties and the spread of behavior and attitudes can be conceptualized as a form of social influence.

The extant models of network diffusion assume that all potential adopters have access to equal information within the social structure, with proximity acting as the only determinant of spread. Social network researchers have generally assumed that where there is a tie, information is transferred uniformly. This assumption seems highly unlikely in most organizational settings. Complete information by all organizational stakeholders is implausible, if not impossible. In taking access to complete information for granted, the existing diffusion literature focuses on the problem of making decisions regarding the information rather than how and what information is actually acquired or transmitted. In his 1962 work Diffusion of

Innovations , Everett Rogers defines diffusion as "the process by which an innovation is communicated through certain channels over time among the members of a social system" (p. 83). In line with this claim, I contend that information attainment is the first critical step toward adoption within a social structure. In other words, in order for an individual to make a decision, they must be aware of the choice set.

The notion that information and structural access are interrelated is not new.

Network theorists have long contended that individuals who belong to close, cohesive groups will share similar information and be aware of the same resources (Granovetter

1973). In such groups, information will also diffuse more rapidly because the members share many direct connections to one another. In contrast, individuals with weak ties beyond their local group or individuals who have large sparse networks will have greater access to novel information (Burt 1992, Granovetter 1973). A contact that is weak or outside of the local network will provide new information that is not available to the local sub-group.

In the following section, I develop an argument describing why the second step in diffusion, social influence, also varies by content type.

Social Influence

In their seminal article, Salancik and Pfeffer (1978) argue that an individual’s attitudes arise from the social context in which they are formed. Their Social-

Information-Processing (SIP) theory stipulates that individuals develop attitudes as a function of the information they have access to via their social relationships. While

SIP theory has generated a great deal of research, it is not without its critics, who argue that the theory lacks a mechanism by which information moves between individuals. Addressing these concerns, researchers Ibarra and Andrews (1993) applied social network theory and methods to better understand the mechanisms that underlie SIP and gain purchase on the means by which information and network structure affect individual attitudes and behaviors.

Within an organization, members will communicate with other members in hopes of altering their behavior or attitudes (Durkheim 1976; Simon 1965). Influence, as defined by the work of March (1955) and Simon (1957), is simply a particular

instance of causality – the modification of one person’s responses as a result of the actions of another. Social influence occurs when one actor is able to alter the behavior or attitude of another actor through social relationships. Social influence connects the structure of social relations to the behaviors and attitudes of the individuals who comprise the group or network. Understanding social influence is critical to understanding processes of diffusion and demonstrating the explanatory power of network effects on individual action (Laumann 1979).

Past research on social influence has focused predominantly on structural factors as determinants of diffusion. For example, network theorists contend that actors who are more central in the network structure acquire information more readily.

Yet, little is known about how the content of the information being shared through social relations interacts with and possibly modifies behavior. Extant research suggests that the position of an individual in an intra-organizational network acts in part to determine behaviors (Burt 1987; Ibarra and Andrews 1993; Salancik and Pfeffer

1978). The network of relationships facilitates the transmission of organizational resources and information, which in turn affects social influence between actors and patterns of diffusion within organizations.

For network analysts, social influence is determined by the structural concept of social proximity. Increased social proximity of two individuals in a social network is associated with greater interpersonal influence between the two individuals

(Cartwright and Zander 1968; Friedkin 1990). Social proximity has been characterized

in one of two ways throughout the diffusion literature: as social cohesion or as structural equivalence.

One camp argues that social cohesion is the necessary structural element for predicting innovation adoption (Coleman, Katz, and Menzel 1966; Valente 1995).

When making a decision in an ambiguous setting, individuals will seek out the opinions of those with whom they share strong ties (Festinger, Schachter, and Beck

1950; Friedkin 1990a; Friedkin 1990b). In the networks literature, cohesion generally implies networks that are dense with strong ties among members (Festinger 1950).

The cohesion model posits that within a cohesive group, the stronger the tie between ego and alter, the more likely adoption on the part of ego will trigger alter’s adoption.

In the other camp, theorists argue that structural equivalence acts as the primary determinant of social influence between individuals. Two individuals are structurally equivalent if they occupy the same position in the social structure and share the same patterns of interactions with other positions (Burt 1987).8 Structurally equivalent actors need not share a tie or be aware of each other. However, Friedkin

(1982; 1983) posits that for structural equivalence to affect interpersonal influence, ego must be aware not only of alter, but also alter’s opinions and behaviors. In this case, ego looks to the structurally equivalent alter’s adoption behavior when deciding

8 Across scholars and research areas, structural equivalence has various specifications. In the strictest sense, it refers to when two nodes share the same set of relationships to all other nodes in the graph. However, other definitions include “regular” equivalence, where two nodes share the same profile of relations in a graph. These are largely meant to capture similar roles of nodes in the network.

to behave a particular way (such as adopting an innovation) because of their similar roles within the structure (Wasserman and Faust 1994). This process of influence resembles DiMaggio and Powell’s (1983) concept of “mimetic isomorphism” for organizations, where a firm imitates another firm’s practices in the simple belief that it will be beneficial. 9 Since empirical evidence exists for both models, I argue that it is information type that largely determines the mechanism of influence.

Next, I describe how individuals may be strategic in communicating certain types of information and how this behavior has implications for the social network.

Strategic Interactions

Organizational members may themselves attempt to manipulate the transfer of information within a communication network. In the case of a legitimate project, if we assume that the team’s primary goal is efficient implementation of the innovation, then efficient communication between participants is also a priority. This may take several forms at the individual level. First, actors may attempt to introduce their alters so that the alters may share information, thus generating new ties (Obstfeld 2005). Second, individuals may try to remove any bottlenecks of communication, where information is funneled into one individual from several others and therefore altering tie patterns.

Finally, actors may encourage reciprocal exchanges. For example, actors may promote open dialogue about the project, leading to reciprocal communications.

9 I invoke this comparison because DiMaggio and Powell’s (1983) concept of mimetic isomorphism maps well to social influence in networks via structural equivalence. They distinguish mimetic isomorphic change from coercive and normative types of change, which is the mechanisms by which structural equivalence operate in a network.

To implement a corrupt innovation, members might seek to control patterns of communication. When the information is corrupt, communication channels may be limited or altered due to the fact that those involved must protect themselves from detection. Individuals faced with covert information must consider the tradeoff between efficiency and secrecy. In contrast to legitimate innovations, corrupt innovation communication networks will attempt to maximize concealment rather than efficiency (Baker and Faulkner 1993). In an endeavor where the task is legitimate, individuals need not concern themselves with who might overhear the conversation or being detected by non-participants. On the other hand, secrecy is a primary concern for covert networks. Every communication can potentially undermine the activity and those involved. Therefore, secrecy is of greater importance than efficiency of implementation. Beyond limiting information transmission to alters, actors may also attempt to hinder tie formation between their alters.

I argue that in the covert sub-structures, actors discourage or hinder their alters from forming a relation, which provides the actors themselves with more control and power over the separated alters (Cook and Emerson 1978). This places them in a brokerage position between two other members. Burt (1992) originally developed the concept of brokerage from Simmel’s (1955) tertius gaudens , which translates to “the one who benefits.” Burt’s view of networks conceptualizes ties in terms of access to information and resources; in organizational networks, brokers who span “structural holes” have greater access to diverse information and resources. Structural holes are the social gaps between connected sub-groups. In Burt’s concept, the position is not

necessarily achieved strategically by the actor’s manipulation of her relations but instead the position occupied can simply be due to chance.

In addition to the benefits structural holes provide an individual (positive job evaluations, increased career mobility, higher salaries; see (Burt 2002) for a review), structural holes also provide opportunities for malfeasance and corporate mis- governance (Burt 2004; Mitchell 2003). A network full of structural holes provides people with the ability to play others off against each other (Flap and Volker 2001;

Podolny and Baron 1997).

I extend Simmel’s concept of Divide et Impera , where the third actor strategically keeps two actors separate to maintain some degree of power over them

(Simmel 1950: 145). Here the broker is not simply enjoying the benefits of a structural position, she is actively trying to maintain an advantage by keeping her alters from creating a tie. This concept is slightly different from the original conception of brokerage for the simple fact that the broker intends to maintain and reproduce her structurally determined advantage.

In the following three sections, I present the main arguments and the hypotheses for each of the three studies.

Study 1: Diffusion and Recruitment

Given that the information regarding legitimate innovations will be less restricted, my predictions follow Granovetter’s (1973) influential weak-tie argument, in which individuals are more likely to encounter novel information via weak ties.

Since legitimate information need not be concealed from non-participants, members

with non-redundant ties to multiple sub-groups are more likely to learn about and participate in legitimate endeavors. For example, when an actor is involved in a legitimate endeavor within his organization, and he can describe and explain the project to colleagues with whom he shares only a weak tie and who are not involved in the project. 10 These colleagues may in turn share the information with their local sub- groups. In this case, individuals who have ties to this group are more likely to hear of the project and therefore to participate. It follows that members who share ties to multiple sub-groups increase their access to overt information.

For legitimate innovation information spread, I propose that individuals with brokerage positions in the communication network are more likely to encounter overt information innovations. These individuals act as brokers of information and resources by connecting multiple sub-groups in the larger network. For a visual representation of the network position of a broker see figure 2.a. Assuming that overt information will flow unrestrained through all connections, even weak ties, these individuals are structurally advantaged to hear novel information. Evidence for this argument can be found in multiple settings including managers and firms (Fernandez and Gould 1994;

Mehra, Kilduff, and Brass 2001; Mizruchi and Stearns 2001; Stuart and Podolny

1999). For this study, I use Burt’s (1992) measure of network constraint as a

10 By sharing this information, an individual not only reinforces communication channels with others, but also may receive valuable advice or insights from his co- workers. Moreover, sharing the information may help to recruit additional employees who may be useful to the implementation.

determinant of legitimate information attainment. Burt’s measure captures the number of non-redundant ties to other actors within the network.

As mentioned earlier in regard to information type, secrecy is central to corrupt network operations. New participants will have to be recruited carefully so as not to risk jeopardizing those involved and making them vulnerable to condemnation. Direct communication links are ideal for communicating corrupt information because they reduce the chance of the “wrong” individual learning of the malfeasance. Using an intermediary or a non-participant to convey covert information puts the project and the participants at risk. I argue that for corrupt information, the message is tightly coupled to the messenger (the corrupt participant), and corrupt information will more likely be spread through a strong tie. Allowing non-participating intermediaries to have knowledge of the corrupt innovation and its associated members makes those involved vulnerable to discovery. Legitimate innovations, on the other hand, reflect the goals of the organization, which means that non-participating members may be aware of the innovation and able to share information regarding it. In this instance, the message can be decoupled from the adopter and individuals may learn of the innovation from non- adopters. Moreover, as noted by Granovetter (1992), the opportunities for organizational crimes such as embezzlement require the trust of strong direct relationships.

Figure 2.a. Network Representation of Broker an d Magnet

High closeness centrality indicates that an actor has the shortest path to all other actors in the network. This is an excellent position to access others and their information in the network quickly because the actor has fewer steps to reach othe r individuals in the network. Individuals with high closeness centrality are more likely to be exposed to a corrupt innovation adopter because they have multiple direct ties throughout the organization. Through his relations, this individual is deeply enme shed in the organization’s communication network. I use the term “magnet” to describe these individuals because their pattern of relations allows them to draw information from throughout the organization. I provide a graphical example of a magnet in figure

2.a . Therefore, high closeness centrality or being a magnet will increase the likelihood of corrupt innovation adoption.

Below are the predictions for recruitment into and participation in overt and covert networks:

Hypothesis 1: Brokers (individuals with large sparse networks or multiple weak ties) are more likely to participate in a legitimate innovation.

Hypothesis 2a: Magnets (individuals with the shortest path to other members) are more likely to participate in a corrupt innovation.

Hypothesis 2b: Individuals who have an alter that previously adopted a corrupt innovation are more likely to participate in a corrupt innovation.

Study 2: Strategic Interaction

In this section, I propose that individuals who are involved in both types of endeavors, legitimate and corrupt, create very different network structures through their communications. In other words, an individual’s communication structure for implementing a legitimate project will be very different from the structure for the implementation of a corrupt one. In order to understand how individuals can alter structure we must consider triads. The triadic relation is foundational for understanding strategic interaction.

Simmel provided one of the first and most comprehensive theoretical arguments for the importance of triads. Krackhardt defines a “Simmelian Tie” based on Simmel’s theoretical foundations for triads as three individuals strongly and reciprocally tied; for conceptual clarity, I will refer to this as a “Simmelian triad” (see figure 2.b). The dynamics of a triad, Simmel (1950) argued, are fundamentally different than those of a dyad (beyond simply the number of participants). First, there can be no majority in a dyad. An individual in a triad can be “out-voted” by the group, reducing individual power. Additionally, a third party can act to moderate conflict by

arbitrating two opposing parties. When collaboration is central to the goals of the individuals, Simmelian triads will be prominent because this relationship bolsters cooperation (Krackh ardt 1998; Simmel 1950).

Figure 2.b. Simmelian Triad

Assuming that individuals have multiple responsibilities in a large firm, they will need to ration their work efforts. In these cases, if individuals want to share information efficiently, encourage c ooperation and reduce conflict, they should introduce and encourage communications between their alters. For example, once ego shares reciprocal relations with alter B and alter C, ego will allow and/or encourage alter B and alter C to also share a recipro cal tie. This consequently increases network closure both locally and globally.

In the context of corruption, I adopt Simmel’s (1950) term Divide et Impera , where a third actor strategically keeps two actors separate to maintain some degree of power over them. Individual power and rewards can be derived from a brokerage

position between two actors (Cook and Emerson 1978). In this instance, the broker is instrumental in keeping individuals from forming a relation. Ego communicates with alter B and alter C, but she discourages alter B and alter C from forming a relation or sharing information between them (see figure 2.c ). This leads to fewer individuals who are fully aware of the participants and endeavor, which in turn means fewer people to implicate ego. D iscouraging communication between alters also reduces the amount of communication that could possibly be detected. Therefore, corrupt sub - structures will have fewer Simmelian triads than networks in which the information is legitimate.

Figure 2.c. Divide et Impera

When individuals are communicating about corrupt endeavors, their networks will be more hierarchical , which means that there will be more asymmetry in their communications. Asymmetry within a communication network indicates that information flo ws towards certain privileged individuals, which reflects a status and/or power order ( Bonacich 1987 ). In a network with high amounts of asymmetry, the

majority of the nodes act as informants to certain central actors, where the information aggregates (see figure 2.d ).

Figure 2.d. Hierarchy

Finally, corrupt networks will also contain more structural “tie -breakers”, individuals to whom multiple individuals report. Krackhardt (1994) terms these “tie - breakers” upper-bounds. If you conceptualize the networ k as a classical reporting structure, the upper-bounds would be the “boss”. The least upper bound is the closest

“boss” or direct authority. This form serves to funnel information into a few central actors rather than throughout the network. Having multipl e individuals communicate to a “boss” privileges that individual with information and puts her in a position of power. Similar to formal hierarchical organizations like the military, the individuals at the top of the hierarchy sit in the most powerful and influential positions, while the individuals at the bottom are least powerful and influential (see figure 2.e ).

Figure 2.e. Least Upper Boundedness

Below are the three proposed hypothesis of strategic interactions with content types:

Hypothesis 3: Indiv idual ego-networks will have more closure when communicating about legitimate information.

Hypothesis 4: Individual ego -networks will be more centralized when communicating about corrupt information.

Hypothesis 5: Individual ego -networks will have more asy mmetrical relations when communicating about corrupt information.

Study 3: Aggregate Network Structures

Overt Innovation Networks

In an ideal case, legitimate innovations mirror the objectives of the organization and are endorsed by organizational members . Individuals are free to communicate in a manner that best serves the implementation of the innovation. For such legitimate innovations, a cohesive network form will improve the ability to acquire information or assistance. Connectedness is important for innovations and non-routine practices in an organization, particularly because the cohesion allows

members to consult one another through established communication ties and relations

(Krackhardt 1994). When an innovation is introduced, the individuals invo lved need to communicate with each other to acquire pertinent information and learn how to implement the innovation. A connected structure also improves the ability to keep more individuals “in the loop” regarding the innovation. In well -connected groups, members are more likely than non -members to share similar information. Information will also diffuse more rapidly in a well -connected group because members are more likely to share ties (see figure 2.f).

Figure 2.f. Connectedness

Reciprocal communicatio n is also particularly important for innovations because the introduction of a new practice introduces uncertainties that individuals must address. A reciprocal relation between two actors allows the mutual exchange of information. This implies that inform ation flows back and forth between individuals rather than in just one direction. These exchanges permit a discourse between members to help them clarify or refine ideas.

Covert Innovation Networks

Within an organization, corrupt innovation participants may seek to protect themselves through graduated divisions of labor, which serve as organizational buffers, sealing off groups of members from each other (Simmel 1950; Goffman

1970). A division or absence of communication between groups reflects a lack of connectedness in the structure. In view of these findings, the expectation is that corrupt networks will be fractured and mostly comprised of clusters.

In addition, previous research shows illegal activities tend to yield centralized, hierarchical networks (Baker and Faulkner 1993; Simmel 1950). Baker and Faulkner

(1993) investigated conspiracies in the heavy electrical equipment industry and found that centralized and hierarchical communication was required for organizational conspiracies to be carried out. Hierarchical network structures or networks with a large amount of directed asymmetrical ties are more favorable for maximizing access to and control over information. A structure that contains a high number of directed asymmetrical ties also reduces member discourse about the corruption. Directed asymmetrical ties imply differences in power or status within the network, with information flowing to the most powerful or highest status members reinforcing hierarchy (Bonacich 1987; Brass, Butterfield, and Skaggs 1998). In the case of corrupt networks or networks that prize secrecy, a hierarchical form will reinforce the power of some individuals to manage information and thus, other network members.

Below are the remaining two hypotheses for this dissertation:

Hypothesis 6: Corrupt communication networks will have more asymmetrical ties than legitimate networks.

Hypothesis 7: Legitimate communication networks will be more connected than corrupt networks.

Conclusion

In this chapter, I set out the theoretical underpinnings of my hypothesis for each study. I address each study individually in the subsequent chapters. The central claim of my dissertation is that message content alters the way individuals interact and share information and that, in turn, affects both the local ego-networks and the aggregate measures for the entire social system. These system-level differences then influence who is recruited into the content sub-structures. Understanding this dynamic process will provide greater insight into the mechanisms that change network form.

In the next section, I describe the methods I employ and Enron Corporation, the organizational site that I use for my analysis. In chapter 4, I demonstrate that the individual’s structural attributes are different for recruitment into content-specific networks. In the chapter that follows, I provide evidence that individuals make strategic choices about their relations based on content, which in turn creates differences in the egocentric networks. In chapter 6, I demonstrate the aggregate differences in content-specific network structures. Chapter 7 explores those that spoke out against the fraud at Enron from a structural perspective. I present my conclusions in chapter 8.

Chapter 3 – Research Setting and Methodology

This chapter describes the research setting, Enron Corporation, and the methodology used to study the company. I look at six innovations at Enron— three of which were legitimate and the others deemed corrupt. This chapter has three sections.

The organizational setting is presented first along with a detailed description of the innovations. The second section presents an overview of the data and operationalization of the concepts. I conclude with a discussion of the multi-level analysis employed to test the hypothesis.

Enron’s reputation for pioneering new products and markets makes it an excellent setting for studying innovation. Enron was largely recognized for its organizational innovations, such as partnership arrangements and financial vehicles.

Enron was also instrumental in developing new commodities markets and accounting practices. Fortune Magazine named Enron “America’s Most Innovative Company” for six consecutive years (1996-2001), which underscores that Enron was perceived as ground-breaking both within its sector and throughout the business community.

The analysis is based primarily on Enron emails that were aggregated to represent the company’s communications networks. These data provide a unique opportunity to conduct a detailed study of communication networks for both overt and covert innovation activities. At the time the data were collected, email was the

accepted means of conducting business. Previously, email at that time was considered private and was rarely used as evidence in trials.11

The Setting

Enron was based in Houston, Texas and was one of the world’s leaders in electricity, natural gas, and communication. Originally founded as an energy company in 1985 in Omaha, Nebraska, Enron quickly expanded into other sectors. After many years of dramatic success, record-making profits, and a favored investment for Wall

Street analysts, Enron was brought under investigation by both the Security and

Exchange Commission (SEC) and the Federal Energy and Regulatory Commission

(FERC). Enron was believed to be “cooking” its books, based on a memo by Sherron

Watkins that was leaked to the press. On October 22, 2001, the SEC announced that it would be exploring several suspicious Enron deals. In December 2001, the Houston- based energy trading company filed for bankruptcy – making it then the largest bankruptcy of its time.

Following investigations, it was revealed that Enron hid massive amounts of debt using “creative accounting practices” and “off-balance sheet” transactions involving Special Purpose Entities (SPEs). In addition, major conflicts of interest were created by the partnerships of these SPEs and Enron. Enron’s Chief Financial Officer,

Andrew Fastow, was generously compensated for authorizing and managing these partnerships. These activities not only precipitated the investigation of Enron’s

11 New York v. Microsoft in 2001 was the first publicized trial that used email as evidence and occurred after the corrupt email exchange at Enron Co.

accounting and management practices, but also led to the enactment of federal laws such as the Sarbanes-Oxley Act of 2002, designed to compel boards of directors to be more vigilant in their fiduciary duties and oversight of companies. This act sought to prevent future corporate accounting scandals like those perpetrated by Enron, Tyco

International, and WorldCom.

Enron rapidly expanded in the late 1990s. In accordance with its business plan, it was moving to buy and develop assets such as pipelines and power plants. Enron would then expand the asset by building a wholesale or retail business around it. This business model posed some challenges for the company. First, in order to grow along these lines, Enron needed large initial capital investments, which were not expected to generate earnings or cash flow in the short term. This placed immediate pressure on

Enron’s balance sheet in terms of performance. Second, Enron already had a substantial debt load. Although maintaining high credit ratings at investment grade was crucial to their energy trading business, this debt placed pressure on Enron’s credit ratings. The only feasible solution for the energy giant was to find outside investors to enter into the agreements, largely in the form of SPEs. Through SPEs,

Enron could treat the joint ventures as investments and not include all of the entities’ assets and liabilities on its balance sheet (Cruver 2003; Fox 2002; Mclean and Elkind

2003; Swartz and Watkins 2004).

Treating these investments as “off-balance-sheet” was preferable because it enabled the company to present itself more attractively to Wall Street analysts and rating agencies. Enron used such SPEs in many businesses: financial asset sales,

where Enron w ould simply sell the debt or equity interest to the SPE; lease transactions, where Enron would sell the asset to the SPE and then the SPE would lease it back to Enron (which is what they did for their headquarters in Houston); and hedging of SPE stock agai nst Enron stock. Hundreds of US companies have used

SPEs to keep trillions of dollars off their balance sheets through partnerships and assorted obligations. At the time, the accounting literature provided only limited guidance on when SPEs should be conso lidated. At the same time, accounting firms were putting pressure on both the SEC and the Federal Accounting Standards Board to endorse certain common place SPE practices. Below is a graphic that details a typical

SPE arrangement with Enron (see figure 3.a).

Figure 3.a . Enron’s Use of Special Purpose Entities

Corrupt Innovations

The three corrupt innovations that I investigate are the SPEs: Joint Energy

Development Investment (JEDI), Chewco, and Talon. The creation of these SPEs and all of their transactions with Enron violated existing accounting regulations. Under applicable accounting rules, an SPE could receive off-balance-sheet treatment only if independent third party investors contributed a minimum of 3 percent of the SPE’s capital and the third party investments were genuinely independent. If these criteria were not met, then Enron was required to consolidate the SPEs onto its balance sheets

(Section 10(b) of the Exchange Act). This caused Enron to file “materially false and misleading” annual and quarterly reports (SEC filing complaint 17692). With these three vehicles, Enron’s management was able to present itself more attractively to investors. Moreover, the individuals involved with these SPEs used them not only to inflate Enron’s earnings, but also to generate enormous private profits for themselves.

The first corrupt innovation, JEDI, was originally a joint investment partnership between Enron and the California Public Employees’ Retirement System

(CalPERS). CalPERS was a limited partner in JEDI and contributed $250 million in cash. Chewco, the second corrupt innovation (named for the Star Wars character

Chewbacca), was part of a group of partnerships that executed financing deals.

Chewco was formed to buy out CalPERS’s 50 percent interest in Enron via JEDI.

Enron, under a secret side-deal, put up the financial collateral in order to provide the 3 percent outside equity Chewco required (Ackman 2002; McLean and Elkind 2003); this meant that Chewco did not qualify as a third-party SPE. In fact, Chewco did not warrant SPE status because it was owned by the domestic partner of Michael Kopper,

an executive at Enron and Andrew Fastow’s top deputy, who was in fact controlling and partly funding Chewco. Both Chewco and JEDI should have been included in

Enron’s accounting books, but Enron preferred to treat the SPEs as “investments” not under Enron’s control and did not consolidate JEDI or Chewco onto its balance sheets or include their debts in its financial statements. For example, Enron included its contractual share of gains and losses from JEDI on its income statement but would not show JEDI’s debt on its balance sheets. This kept up to $600 million in debt off

Enron’s books (McLean and Elkind 2003).

Although Chewco was formed as an SPE to purchase JEDI, Enron was unable to locate an outside investor and instead financed the purchase almost entirely with its own debt. Following the sale of Chewco, Kopper transferred his Chewco ownership interest to William D. Dodson, Kopper’s domestic partner. During this time, Chewco paid Enron $17.4 million as a subordinate loan fee. Meanwhile, JEDI was the source of these payments to Enron. In March of 1998, Enron recorded a 28 million dollar asset.

In 2001, after coming under scrutiny by the press, Anderson Accounting reviewed the SPE status of both Jedi and Chewco and decided that they did not satisfy the minimum SPE accounting rules. Enron and Anderson concluded that these SPEs lacked the sufficient outside equity to qualify for non-consolidation. The retroactive consolidation decreased Enron’s reported net income of $893 million by $95 million in 1999 and by $8 million (of $979 million) in 2000. After the review, Enron decided to consolidate both SPEs into their financial statements. This increased Enron’s

reported debt by $711 million in 1997, $561 million in 1998, $685 million in 1999, and $628 million in 2000.

In addition to grossly misrepresenting profit and debt, the SPEs allowed the

Enron employees who were involved to gain sizable financial rewards either through fees or interest payments. A partnership called “Southampton Place” allowed investors such as Fastow to invest small amounts of money but receive enormous profits in return. For example, Fastow invested $25,000 and received $4.5 million in a two- month period. Two other employees invested $5,800 each and received $1 million over the same time period. Kopper, who was acting as manager of Chewco, received

$2 million in compensation and other management fees. 12

Talon, the third corrupt innovation, was designed to offset Enron’s significant potential mark-to-market losses on certain investments. Talon was funded mainly through Enron; the remaining “outside” 3 percent or $30 million was provided by

LJM2. LJM2 was another SPE of Enron and had just secured $41 million from Enron prior to investing in the creation of Talon. According to the SEC, Talon failed to meet the minimum equity test as required by accounting rules, and LJM2 lacked substantive control over Talon. Thus, similar to Chewco and Jedi, Talon was under the substantive control of Enron (SEC Filing complaint 18335). In addition, the SEC charged that

Talon, like Jedi and Chewco, did not have an actual business purpose that warranted

SPE status.

12 Unlike Fastow, Kopper was not a senior officer at Enron, so he was not required to complete a proxy statement disclosure to Enron’s Board of Directors, thus keeping them in the dark about the conflict of interest.

Legitimate Innovations

I compare JEDI, Chewco, and Talon to three legitimate innovations that were in progress at Enron during the same time. These legitimate innovations include Enron

Online, the Dabhol India Power Plant Project, and the Blockbuster Venture. These three endeavours were comparable to the corrupt partnerships in terms of capital expenditure and accounting arrangements, with the notable exception that these partners were truly outside investors. Enron Online was the first online commodity trading system that allowed global transactions; this online venture capitalized on world markets and new trading technologies. Launched in 1999, it was the first web- based transactions market. The main commodities were energy related, such as natural gas and oil, but the market traded over 500 different products, including TV commercial ad space.

The Dabhol India Power Plant Project, referred to here as Dabhol, was a joint venture with General Electric and Bechtel Corporation for the creation of an Indian power company in the state of Maharashtra. With Enron managing and operating the plant, GE provided the turbines while Bechtel constructed the plant. Starting in 1998,

Enron proposed a 300 million dollar project to build pipelines from Dabhol to Hazira, which would connect Dabhol to the existing gas infrastructure (Rai 2001). Dabhol was an ambitious undertaking for Enron in terms of both size and importance. In addition to being a joint venture, the project required that Enron work closely with India’s government. This project was not without controversy; many Indians felt that the company was excessively expensive and environmentally hazardous. For example, in early 2001, the Maharashtra state utility, which was under contract to buy all of

Dabhol's electrical output, complained that the prices were too high and stopped paying its bills, accruing $64 million in penalty fees before Enron cut off the power and shut the plant down (Rai 2001). Also, the project’s authorities were accused by environmental groups of not conducting a proper environmental assessment of the project which could be harmful to the local wildlife (Allison 2001).

Lastly, in 2000, Enron Broadband partnered with Blockbuster, Inc. to provide on-demand movies over the Internet. The goal of the Blockbuster partnership was to provide live or on-demand media broadcasting. The project depended not only on the partnership with Blockbuster, but also on the development of broadband content delivery applications, which was supported by another Enron venture. The Enron/

Blockbuster partnership’s success hinged largely on Enron’s ability to provide broadband access, which was tenuous at best.

In the next section, I describe the data used to analyze the communication networks for these six enterprises within Enron.

The Email Corpus

This study examines longitudinal email data taken from Enron Corporation between 1998 and 2002. Four factors in particular make this dataset valuable and appropriate for the systematic study of various types of content networks. First, the dataset includes all of the email correspondence for the sample of employees, both professional and personal discourse. Second, the dataset presents an opportunity to study not only the social network structure of the organization’s employees, but also to analyze the messages shared between them. In contrast to previous studies on social

influence that have assumed the transmission of information along network ties, the data allow for actual observation of information transfer between organizational members. This permits comparison of the spread of various types of content. Third, the data provide a unique opportunity to investigate and compare the spread of corruption, as employee emails include messages that the employees may otherwise be unwilling to share. Finally, email exchanges closely parallel friendship networks and work networks (Marmaros and Sacerdote 2006).

The network information and correspondence are drawn from the Enron Email

Corpus (EEC), a collection of emails subpoenaed and made public by the Federal

Energy Regulation Commission (FERC). 13 In addition, I use various other sources, such as annual reports, individual interviews, public statements, and testimonies from the Securities and Exchange Commission (SEC) and the Department of Justice, to enrich the EEC data. The EEC dataset is comprised of professional and personal email messages sent by individuals over the five-year period.

The seed sample consisted of 129 individuals from which a snowball sample of email recipients and senders was created. Each email includes the following information: sender; recipients; recipient form: to, carbon copy (CC), and blind carbon copy (BCC); transmission form: original, reply, and forward; date; folder title; subject; and message content. Within the EEC, there are over twenty-seven thousand unique senders and recipients after the data were normalized to remove redundant emails (one

13 The dataset used here is composed of both the FERC online database and the dataset made available by Shetty and Adibi (2008) in MySQL format.

individual may have had several email accounts throughout their tenure at Enron), group emails, and distribution lists. 14

The nature of the data unfortunately results in a limited sample in the initial years. This is a common issue for panel data (although usually in the reverse, with larger populations that dwindle in subsequent periods) and can be addressed with statistical techniques discussed later in this paper. The sample is based on egocentric data, which over time is similar to a snowball sample. As individuals communicate with more individuals over time and the individuals that they communicate with also communicate more, the sample grows. This results in the eventual observed network of 14,272 employees, which was 65% of Enron’s 22,000 total population for 2001 (see table 3.a) (Standard & Poor’s Compstat 2009). Even though only Enron employee email addresses are used in the analysis, the EEC includes emails from those outside

Enron. 15

14 Email addresses have been normalized to represent actual Enron employees in a variety of ways. First list-serve or group email accounts were removed, such as “[email protected]”. In addition, emails accounts were combined when they belonged to a single individual. For example, Kenneth Lay had both “ken.lay@enron” and “kenneth.lay@enron”.

15 Individuals outside of Enron are included in the analysis to calculate aggregate network measures for Enron employees because having relationships outside the organization may change the effects of the relationships with individuals from within the company. For example, individuals with many ties outside the organization may receive different or less information than those with many ties within the organization and therefore may be less likely to participate in either type of innovation.

Table 3.a. Individual Email Accounts Total Enron Only Size of Enron Percent of Year Network Network (appr.) Population

1998 422 357 17,800 2.01% 1999 1,791 1,298 17,900 7.25% 2000 12,136 6,528 20,600 31.69% 2001 25,786 13,391 22,000 60.87% 2002 27,827 14,272 22,000 64.87%

In addition to the EEC, I have compiled organizational and demographic information for a portion of the employees in the dataset from various sources. This information includes gender, occupational title, department and sub-organization in which the individual worked, and to whom the employee reports. From the SEC filings (1996-2005), I acquired Enron’s annual reports and internal documentation to ascertain titles and reporting hierarchies. I also referred to documentation from the

United States Department of Justice (2006) trials that included board meeting agendas to understand the corporate reporting structures and activities.

Even though I do not have information for Enron’s entire population, the sample contains a representative cross-section of the organization by both sub-division and professional title. The members included in this study span six different sub- organizations under the Enron Corporation. Sixteen different titles are also represented in the dataset, which includes top management as well as traders (see table 3.b).

Finally, materials such as affidavits, executive calendars, and internal documents were used to ensure that a complete population of adopters was identified.

Table 3.b. Enron Employee Corpus Seed Sample Characteristics ( N=129) Variable Mean Gender (Male) 71.32%

Corporation: Enron Americas 3.10% Enron Corporation 6.20% Enron Global Markets 1.55% Enron North America 63.57% Enron Transportation Services 0.78% Enron Wholesale Services 1.55% Unknown 23.26%

Title : Analyst 5.40% Associate 3.10% Chief Officer 7.00% Director 3.90% Director of Trading 10.90% Employee 12.40% Executive 1.60% Executive Vice President 2.30% Legal Counsel 12.40% Manager 3.90% Manager of Trading 10.90% Managing Director 2.30% Specialist 7.00% Trader 3.10% Vice President 8.50% Vice President of Trading 5.40%

Organizational Member Attributes

Although I do not have the formal organization reporting chart for Enron, the richness of the informal data plus additional resources allowed me to identify the titles for a small sample of individuals. Table 3.c shows the percentage of seed sample by gender, corporation, and title. This was largely based on either hand-coding the titles from email messages or extracting titles from internal memos and court documents.

For example, if an email included “We would like to congratulate Joseph X on his new role as Northern California Accounts Trader”, I would then code Joseph X with that title. Additionally, internal memos often listed individuals and their related titles at Enron.

Email aliases within Enron predominantly followed a pattern of the first name, a period, and the last name of the employee. For example, Kenneth Lay would be

“[email protected]”. This pattern allowed me to systematically parse the email aliases to acquire the first and last names of most of the individuals. I then created a list from the US Social Security’s database of the 1,000 most common names and the associated gender for each year between 1940 and 1960. Using a simple matching program, I compared the employee’s first name to the Social Security name list, which provided me with employee gender based on the first name.

Table 3.c. Enron Employee Innovation Participants Descriptives by Content Type Legitimate Corrupt Variable Participants Participants (N=1,391) (N=182) Gender (Male) 63.12% 59.89%

Corporation: (N= 1,960) (N=240) Enron Americas 0.15% 1.25% Enron Corporation 0.31% 1.67% Enron Global Markets 0.10% 0.00% Enron North America 3.04% 6.67% Enron Transportation Services 0.05% 0.00% Enron Wholesale Services 0.10% 0.42% Unknown 96.24% 90.00%

Title : Analyst 0.26% 0.83% Associate 0.10% 0.00% Chief Officer 0.46% 2.50% Director 0.10% 0.42% Director of Trading 0.56% 0.00% Employee 0.46% 0.42% Executive 0.10% 0.42% Executive Vice President 0.15% 0.83% Legal Counsel 0.66% 4.58% Manager 0.20% 0.00% Manager of Trading 0.56% 0.00% Managing Director 0.15% 0.42% Specialist 0.05% 0.00% Trader 0.15% 0.00% Vice President 0.46% 1.25% Vice President of Trading 0.36% 0.83% Unknown 95.20% 87.50%

Innovations

For this study, I focus on administrative or organizational innovations.

Administrative innovations pertain to organizational structure or processes and are related more directly to the organization’s procedures and operations than technical innovation, which relate to what the organization produces (Damanpour and Evan

1984; Damanpour 1991; Kimberly and Evanisko 1981; Knight 1967). Schumpeter’s

(1934) description of organizational innovations also includes organizational forms with new internal structures where activities and processes take place.

I make the further distinction that the adoption of an innovation may take one of two possible forms. In the first, individual adoption, innovations require implementation of the adoption by only one individual. This is similar to the Coleman,

Katz, and Menzel study (1966) of physicians in which an individual doctor chose whether or not to adopt a drug for prescription. Innovation diffusion in this instance simply means the aggregate of individuals making individual choices to adopt an innovation. The second form of innovation adoption, team adoption, requires a team of individuals to carry out the implementation of the adoption; this is more common in complex organizational settings. For example, a sub-group may develop a new methodology for completing service orders and work in concert to bring about its implementation throughout the organization. In this case, members are recruited because they are instrumental to the innovation’s successful implementation. It is important to note that the team need not be a formal group. In the case of individual adoption, early adopters have no instrumental reason to advocate others to adopt.

Individuals may champion the innovation, but if others fail to adopt it, it fails only in

that one case. In the case of team adoption, if the individuals critical to implementation of the innovation do not adopt it, the adoption may fail in its entirety.

As is the case with many types of innovations, the delineation of when the individual comes to “adopt” the practice is difficult to discern. Rogers (1962) lists five stages to the innovation adoption process: knowledge, persuasion, decision, implementation, and confirmation. As he aptly noted, it is particularly difficult to acquire empirical evidence for the decision stage as it happens within the head of the individual. For that reason, I treat participation in the implementation as the point of adoption. The process described here includes the sharing of pertinent and instrumental information with the assumption that there is no need to share such information with those whom are not involved.

Methodology

I employ a multi-theoretical, multilevel framework proposed by Monge and

Contractor (2003) for studying communication networks. This framework parallels

Wasserman and Faust’s (1994) five levels of network analysis. It is important to note that although each level represents a different unit of analysis, each level is interrelated. I conduct analysis at three of the proposed five levels. The first, the global level, investigates the entire Enron email network. Here I look at the individual’s global network properties and their propensity to become involved in a content- specific network. At this level, we can see emergent properties of the content network typologies. At the second, the individual actor level , I drill down to the individual ego- networks. Here, the focus is on behaviours of the actor, such as position in the network

or participation in an innovation and the behaviors. For the third level, clusters are analyzed based on content. I disaggregate the network by content and then derive measures based upon the subsequent sub-structures.

The analysis comprises three steps. First, I address the dynamic interaction between the content and aggregated networks to understand individual outcomes. In other words, I use the location of an individual within the entire network of relations to predict participation in content-specific networks. Second, ego-networks are used to understand how individuals may employ different strategies of interactions when faced with certain types of information. Finally, I compare communication networks of legitimate innovations and corrupt innovations. In this comparison, z-scores are used to examine the substantive differences between the networks.

In order to understand the structure of communications between individuals and the effects of their individual positions on subsequent behavior, I employ social network analysis. This analytic technique allows me to represent the connection between interrelated units, such as the employees at Enron, and explore the properties of those connections. Typically, the actors or entities are referred to as nodes within the network. The connections or relations, such as “communicates with” or “provides advice to” are represented by ties or links between the nodes. Nodes may be assigned certain attributes, such as organizational title and gender. The ties themselves may also be assigned certain properties that are attributes of the relation. Generally, network research has focused on tie direction, which captures the origin of the tie and receiver,

and strength, which is a means to capture the magnitude of the relation. I add the type of information as a tie attribute as a means to distinguish the content of the connection.

Constructing an appropriate network from emails poses a challenge. In general, emails are sent directly to recipients; these exchanges can be used to construct a one- mode directed network. These one-mode directed networks are critical to this analysis in order to determine the symmetry of information flow between organizational members. Simply constructing a one-mode network, however, loses the ties between those who are included in the recipient list. These are instances when multiple people are included to receive a message. To handle this, I treat all recipients of an email as sharing a tie. For this reason, I also construct a two-mode or bipartite network of emails and individuals, which includes the senders and receivers. In a two-mode network, senders and receivers are associated through an email message; two-mode networks allow the inclusion of the relations between the recipients of an email.

To capture as much structural information as possible, I assemble not only the one-mode directed and two-mode undirected graphs for the emails, but also a hybrid of the two types. This multiplex network contains both the directed and undirected ties.

I convert the two-mode network into a one-mode network and then merge the networks together. In all three networks (one-mode, two-mode, and multiplex), tie

strength is based on the number of email exchanges or co -occurrences on a recipient list. 16

Figure 3.b. Directed and Undirected Network Based on an Email Message

I also employ qualitative analysis of the ema ils. This method was used in two distinct ways to enrich my understanding of the network properties. First, I used qualitative coding to identify the content -specific substructure. This structure was comprised of all individuals who participated in the imp lementation of a particular innovation. The communications between individuals, namely emails, were also labeled as containing information pertinent to a certain innovation. Next, after reading through all the emails regarding an innovation, I conducted wo rd counts for the comparison of certain message features. Finally, I included emails that demonstrate the strategic interactions at the individual level and provide insight into the employees’ rationales that guided their behavior.

16 For the multiplex networks, senders w ere excluded from the bipartite network prior to merging the networks to avoid double counting of ties and thus increasing the tie weight.

The email messages provide the means for identifying behaviors such as innovation adoption. Individuals who were the senders and individuals who were the intended recipients of emails pertinent to an innovation implementation were coded as adopters. The communications or ties regarding the innovation were also coded with the respective innovation labels.

By analyzing the content of the messages as well, I can determine patterns of communication within the network. The messages in this study were coded by a two- stage qualitative process, using both publicly available software programs and programs that I have modified. First, I identified term patterns in the entire email corpus. Second, I used repeated iterations of the text searches to allow me to improve and refine the dictionary of terms and patterns to search for throughout the text.

Finally, once I could identify the set of email messages regarding a particular topic, I applied traditional techniques to code the messages qualitatively.

Research Design

To accomplish each level of analysis, various methods and tools were employed. The technical steps are listed here in chronological order of application; however, the process of preparing and analyzing the data was iterative. First, the email text files were parsed using a Perl program I wrote and imported into the database

MySQL. The email header and content were separated into separate variables to create the variables message id, sender, receiver, date, subject, and message. Python scripts were used to query the database and create edge-lists by content, date and individual.

Most of the social network measures were calculated with R version 2.7.2. Finally, I

used Python scripts to initially parse the emails with particular content for qualitative coding.

Social Network Analysis

The network can be treated as either a two-mode or one-mode network. In the case of the one-mode, the relations or ties were coded for directionality, such as

“sends email to” or “receives email from.” This provides the originating and the receiving nodes. In the case of emails, there can be multiple receivers but only one sender. Individual email messages were coded by particular projects.

Each node in the network represents an individual member of the organization.

Although emails to those outside Enron occur in the dataset, they were extracted for this analysis. Nodes were coded as participating within a content-specific network.

I incorporate Krackhardt’s (1994) four measures of informal organizations, which speak to the critical dimensions of innovation implementation within an organization. He originally devised the measures as a way to investigate Simon’s

(1981) claim that informal structures will evolve toward a hierarchical form. His first measure is connectedness, which he associates with the ease with which the group can implement or deal with change. Graph hierarchy is the second measure and reflects the amount of asymmetry in a network structure. This measure demonstrates the degree to which communication is not reciprocated. His efficiency index provides the minimum edges required in a network to connect the existing components. Finally, least-upper- boundedness (LUBness) is associated with positional power in the structure to resolve underling conflict.

Qualitative Coding

Since the entire corpus of email messages was too large for traditional qualitative coding techniques, I first queried the data for certain terms (for example, I looked at all email messages that contained the term Jedi). This step allowed me to identify a subset of the emails based on certain term patterns. This process not only gave the communication concerning the SPEs but also the email with similar key words, such as emails about Jedis from the movie Star Wars that were obviously not related to the SPE named Jedi. To distinguish between these cases, once I had identified the set of email messages on a particular topic, I applied traditional techniques to code the messages qualitatively. This allowed me to exclude non- relevant email messages.

Panel Data

The individual-level data were treated as panel data with annual snapshots of network measures and employee attributes. This allowed me to incorporate time- variant attributes, such as closeness centrality, for the individuals in the network. The main advantage of panel data models is that they can account for unobserved heterogeneity across individuals through time. Unobserved heterogeneity may be due to omitted variables. Longitudinal data provide a much better chance to control for individual heterogeneity than cross-sectional data (for a more detailed review, see

Hsiao 1985).

Recruitment Models

In order to analyze the determinants of recruitment into a communication network, I use the first acquisition of instrumental information as the dependent variable. I employ a conditional logit model, which is similar to conditional fixed- effects logistic regression, to predict adoption of either of the innovation types, legitimate or corrupt.

Simulations

Since a standardized means of making comparisons among network measures does not currently exist, I produce simulated networks to calculate z-scores for the networks. This helps to avoid the issues associated with comparing networks of different sizes (Pattison, Wasserman, Robins, and Kanfer 2000). To obtain a z-score for each network, the properties of the networks were compared against sets of 100 random network simulations conditioned on size and density.

Limitations of the Data

Despite significant advantages, the data used here impose some limitations on the study of content and social network form. The first and most important limitation is related to the problem of network sampling. A method for reliable sampling of social networks has not yet been established. I employ a snowball method based on a seed set of emails. The initial seed sample was 129, which eventually allowed me to capture 14,272 unique email accounts. Although the data supplying agency claims that

92% of the accounts are present, my estimates are far more conservative at 65% (see table 3.a). 17

Another shortcoming of the data relates to message censoring. Only messages that were available on the servers were used in the analysis. Messages could potentially have been permanently deleted from the servers. For a message to be lost, both the sender and all receivers would have to permanently delete the email from their accounts. This poses two issues: I do not know the total volume of emails, and certain messages may have been intentionally deleted, causing a bias in the sample.

However, since the sender and all recipients would have to delete the email for the message to be lost completely, I am confident that this was a rare occurrence.

The method of capturing titles by either emails or documents also has limitations. First, it provides titles for only a small percent of the organization. Second,

17 FERC originally obtained the data for its investigation ,; the data were then maintained by Lockheed Martin and its subsidiary Aspen.

it is not possible in most cases to learn when the individual acquired or changed titles or when the person left Enron. To address this in future analysis, I plan to collect resumes via popular professional networking sites for the Enron employees in my dataset. This may provide me with their titles and the duration of the titles.

One final limitation is the question of the study’s generalizability. Even though

I study one organization, I stress that the use of email and the communication patterns found here are not unique to this firm. Further, since there is evidence to support my claims across multiple individuals and innovation networks, I am confident that the findings are robust and could be applied to not only other organization but also new settings, such as markets.

Chapter 4 - Covert and Overt Network Participation

Introduction

Chapters 4, 5, and 6 comprise the empirical core of this dissertation; in these chapters, I investigate the relationship between information content and social structure. This chapter addresses the central question and hypotheses of the first study.

In order for Enron to continue to be recognized as a company at the forefront of innovation, it needed to produce new and creative ventures. In most cases, Enron pioneered lucrative administrative innovations, such as partnerships with other

Fortune 500 companies that were both legal and respected in the energy sector and on

Wall Street. Certain individuals, however, were not confident that Enron could maintain its rapid growth through only legitimate deals and collaborations. These individuals took it upon themselves to exploit accounting loop-holes that would not only improve Enron’s balance sheets, but would also generate profits for them individually.

The characteristics of the individuals who participated in either a legitimate or corrupt activity are the focus of this chapter. Chapter 2 presented arguments for the effect of information on recruitment into certain networks, and described how information spreads differently through a social system based on content. In this chapter, I identify both legitimate and corrupt innovation activities within Enron to explore the structural determinants of covert or overt information spread. The aim is to understand the conditions that permit access to information that would lead to participation in content-specific structure.

Network Positions: Brokers & Magnets

Within an organization, various structural positions provide differing amounts of exposure to information and opportunities and thus act as determinants of who participates in an organizational innovation. In other words, position within the network influences individual participation in an innovation. Several structural indices have been touted as conducive to the spread of innovations, including high network centrality, which increases access to information in general and low network constraint, which improves access to non-redundant information (Burt 1992; Ibarra

1993).

Most of the extant network models of diffusion assume that all potential adopters have access to equivalent information within the social structure, but this assumption seems highly unlikely in most organizational settings. Complete information by all organizational stakeholders is implausible, if not impossible. In taking access to complete information for granted, the existing diffusion literature focuses on the problem of making decisions regarding the information, rather than how and what information is actually acquired or transmitted.

Information attainment is the first critical step to adoption within a social structure (Rogers 1962). For example, in order for an individual to make a decision, they must be aware of the choices available to them. If the choice set is not known, it will for obvious reasons influence the outcome. In general, diffusion studies have assumed that all ties within the network have the same potential to transfer information regardless of the information.

Network theorists have long contended that individuals who belong to close, cohesive groups will each share similar information (Granovetter 1973). Information also diffuses more rapidly in such groups because actors share many direct connections to one another (Knoke and Yang 2008). In contrast, individuals with weak ties beyond their local group or individuals who have large sparse networks will have greater access to novel information (Burt 1992; Granovetter 1973). These models treat individuals as non-discriminating conduits of content.

I argue that for corrupt information, the message is tightly coupled to the messenger, the corrupt adopter. Allowing non-participating intermediaries knowledge of the corrupt innovation and its associated members makes those involved vulnerable to discovery. In contrast, since legitimate innovations reflect the goals of the organization, non- participating members may be aware of the innovation and able to share information regarding it. In this instance, the message can be decoupled from the adopter and individuals may learn of the innovation from non-adopters.

Since secrecy is central to corrupt network operations, new participants will have to be recruited carefully so as not to risk jeopardizing those involved. Direct communication links are ideal for corrupt network members because they reduce the chance of the “wrong” individual learning of the malfeasance. As Granovetter (1992) noted, opportunities for organizational crimes such as embezzlement, require strong direct relationships.

As mentioned earlier, magnets are individuals who on average have a shorter path length to others in the organization because of their ability to draw information

towards themselves from throughout the network. A magnet is also more likely to have a direct tie to a corrupt participant and participate in corruption himself. On the other hand, legitimate innovation information is less restricted in terms of who can know and have access to it. In view of the fact that short direct ties are not necessary for legitimate innovation information spread, a brokerage position will be the best predictor of legitimate innovation participation.

Below I revisit the hypotheses stated earlier in Chapter 2:

Hypothesis 1: Brokers (individuals with large sparse networks or multiple weak ties) are more likely to participate in a legitimate innovation.

Hypothesis 2a: Magnets (individuals with the shortest path to other members) are more likely to participate in a corrupt innovation.

Hypothesis 2b: Individuals who have an alter that previously adopted a corrupt innovation are more likely to participate in a corrupt innovation.

Measures

Freeman’s (1977) closeness centrality operationalizes the extent to which an individual is proximate or “close” to others within the social group. With regard to closeness centrality, centrality decreases as the communication path (number of intermediaries required to reach another actor) increases in length. High closeness centrality allows for independent and efficient communication between individuals.

Central actors need to rely on fewer actors as intermediaries through which to communicate. On the other hand, peripheral actors must depend on others as intermediaries. Thus, these central actors can access information more rapidly throughout a network than peripheral actors, who must depend on others as

intermediaries to communicate information. In essence, the measure captures the degree of structural embeddedness of an actor (Granovetter 1985). Therefore, the magnet can both pull information toward himself but is also constrained by the reciprocal force of his own magnetism.

Closeness centrality measures to whom an actor is connected through both direct and indirect links (Bonacich 1987). Although all relationships with equal proximity contribute equally to the individual’s score, the edge weights are considered when determining the shortest paths. The closeness of a vertex i is the reciprocal of the sum of geodesic distances to all other vertices of i:

−  n  1   ∑ d i j C =  i=1  , i ≠ j j  n −1      where d ij is the length of the shortest path from member j to actor i in the network.

On the other hand, information and behavior are more homogenous within groups. Information from proximate others may be more trustworthy, but the information tends to also be redundant. Therefore, people who share ties across groups have better access to more diverse information sources and are more likely to be aware of novel information in the organization. Burt’s (1992) view of networks conceptualizes network ties in terms of the information and resources that individuals can acquire, and he describes the benefits of brokerage , where an individual is situated between groups of individuals and has the ability to span structural holes (i.e., when one member is connected to many other members who themselves are not connected

to each other). Individuals with networks rich in structural holes have access to many distinct information flows.

The concept of brokerage extends the original ideal type of Simmel’s (1955) tertius gaudens , which translates to “the one who benefits”. The measure of network constraint indicates the extent to which an individual is limited to information within sub-groups of individuals (Burt 1992). Hence, brokers have lower network constraint measures and thus greater access to novel information from different sub-groups.

Network constraint on actor i is calculated by aggregating all intermediaries (third parties) q that i must pass through to contact j:

2   C =  p + p p  ,q ≠ i, j p = z / z ij  ij ∑ iq qj  ij ij ∑q ij  q 

where p ij is the proportion of i's network invested to contact j and z ij is the strength of the relationship between i and j. Squaring defines constraint as a measure of the lack of primary structural holes around j.

Data

In this analysis, I look at six different innovation networks of which three are legitimate and three are corrupt. Observing each type of network throughout its tenure, as depicted in figure 4.a, we see that both sets of networks, legitimate and corrupt, were actively acquiring new members. Each line in figure 4.a represents one of the six individual networks over time. The y-axis is the total number of individuals who joined the network over the total possible number of individuals who could have

joined. The total possible number of employees who could have joined or the risk-set includes all of the alters of the individuals who were involved in the prior year. It is evident from figure 4.a that both legitimate and corrupt networks grew at similar rates and neither network was limited to only its original members. This raises the question: which individuals become involved in either type of network? Were the people who participated different in terms of their positions in the informal network?

0.025 Blockbuster (N=17,613) 0.02 Dabhol (N= 22,628) 0.015 EnronOnline (N=67,554)

0.01 Chewco (N=5,919) Jedi 0.005 (N=16,875) Percent of Alters who Adopt who Alters of Percent Talon 0 (N=4,311) 1999 2000 2001 2002

Figure 4.a. Percent of Alters who Adopt by Innovation (1998-2002)

Table 4.a provides the descriptives for the mean number of employees who participated in each of the individual activities. Not surprisingly, more individuals participate in legitimate innovations. What is surprising is that a higher number of alter adopters are linked to ego with an out-going tie; their relationship is comprised of ego sending emails to alter. It is important to note that the time of adoption between

ego and alter is not taken into account in this table; therefore, we cannot know if alter is influencing ego or vice versa.

Table 4.a. Enron Adoption Descriptives Variable Mean SD Egos who adopt: Corrupt 0.009 0.108 Chewco 0.002 0.042 Jedi 0.006 0.076 Talon 0.001 0.033 Legitimate 0.109 0.405 Blockbuster 0.008 0.090 Dabhol 0.008 0.087 Enron Online 0.049 0.215 Alters who Adopt Chewco 0.006 0.078 Jedi 0.027 0.161 Talon 0.005 0.072 Blockbuster 0.029 0.168 Dabhol 0.033 0.179 Enron Online 0.098 0.297 Alters who Adopt (In) Chewco 0.002 0.048 Jedi 0.014 0.118 Talon 0.002 0.040 Blockbuster 0.014 0.117 Dabhol 0.016 0.125 Enron Online 0.062 0.241 Alters who Adopt (Out) Chewco 0.010 0.152 Jedi 0.027 0.170 Talon 0.009 0.099 Blockbuster 0.033 0.205 Dabhol 0.032 0.175 Enron Online 0.105 0.429

Methodology

In the following analysis, I test the likelihood of an individual participating in an innovation network with regards to structural position in the network. To put this differently, this study examines whether an individual’s position in the entire communication network can determine participation in a legitimate or corrupt sub- group. The network measures are based on all of the individual’s preceding communications and include all content types. I look at the overall structural characteristics of the individual prior to participation within the specific intraorganization network. The networks of interest again include the corrupt innovations (Chewco, Jedi, and Talon) and the legitimate innovations (the Blockbuster

Initiative, Enron Online, and the Dabhol project).

I employ Burt’s (1992) network constraint as an index to measure the extent to which a person’s contacts are within the same group. The lower an individual’s network constraint index, the easier it will be for her to receive information beyond her immediate group and learn about a legitimate innovation. In this case, I expect that lower measures of network constraint will increase the adoption of legitimate innovations. This is due to the expectation that information about legitimate innovations need not be conveyed by a legitimate innovation adopter and that legitimate information will be shared more freely throughout the informal network.

Next, I use closeness centrality to determine the individual’s access or proximity to others within the informal network. The index of closeness centrality has an inverse relationship to average path length. The shorter the average distance between an actor and other actors, the more direct and efficient her communication to others is because

she has fewer intermediaries involved in her communications. High closeness centrality increases the odds of an individual learning about the corrupt innovation directly from a previous adopter. Hence, I predict that closeness centrality will increase the likelihood of adopting a corrupt innovation but will have no effect for legitimate adoption.

Results

In table 4.b, I provide the means and standard deviations for closeness centrality, and network constraint, as well as the number of egos and their alters involved in either legitimate or corrupt innovations. Although both closeness centrality and network constraint vary by year, there is neither an increasing nor decreasing trend for the measures. This indicates that even though the sample grows over the observed time periods, the measures are not biased by the overall network’s size. In tables 4.c and 4.d, I present the means, standard deviations, and correlations for the independent variables used in the analysis for both the legitimate and corrupt innovation adopters.

The dependent variable is involvement in either a legitimate innovation or a corrupt innovation within a particular year. Enron employees were coded as participating in an innovation in one of two ways, which were not mutually exclusive.

First, if an employee was either the sender or intended recipient of an email instrumental to the innovation for that year, she was coded as participating. Second, if the qualitative analysis of the emails and complementary documents provided evidence that the employee was involved, the individual was also coded as a

participant. The dependent variable of innovation participation is not a fixed-state measure, meaning that an individual may be counted as participating in an innovation in one year but not in the following year, if she ceased to be involved. This permits a more dynamic understanding of the structure’s influence on individual behavior.

These data can be treated as panel data with repeated network measures of individuals over time. I employ a conditional logit model, which is the similar to the fixed-effects logit for panel data, to predict adoption of either of the innovation types, legitimate or corrupt. This model controls for unmeasured person-specific effects and unobserved heterogeneity while examining the influence of the covariates on adoption

(Allison 2005). Conditional logit models measure within-person variation. Whereas between-person estimates indicate who will adopt within the sample, within-person estimates tell us under what conditions an individual will adopt. Between-person variation is very likely to be contaminated by unmeasured personal characteristics that are correlated with adoption. Restricting the analysis to within-person variation eliminates the possibility of contamination and is much more likely to get unbiased estimates. Examining within-person variation also accounts for possible biases in the sample. Unfortunately, discarding the between-person variation can yield standard errors that are considerably higher than those produced by methods that utilize both within- and between-person variation. For that reason, the magnitude of the effect is of lesser importance (Allison 2005).

The models address binary decisions in which the individuals have two distinct and mutually exclusive choices: to adopt the innovation or to reject it. For both legitimate and corrupt adoption, the following measures were included: Burt’s (1992)

network constraint, Freeman’s (1977) closeness centrality, variables for the number of alters who have adopted either a legitimate or corrupt innovation prior to ego (this is a lagged variable), and an interaction term crossing an alter who adopted prior with closeness centrality. Given that within-person models can amplify the effect size

(Allison 2005), I focus on the direction and significance of the findings; however, the coefficients have been standardized.

In table 4.e, I present the conditional logit estimates for both legitimate and corrupt innovation adoption. In the case of legitimate innovation adoption for individuals, network centrality is the only significant measure. Network constraint remains significantly and inversely related to legitimate adoption in all three models, -

2.570 (p < .01) for the complete model. This result supports the first hypothesis that low network constraint increases the likelihood of participation in a legitimate innovation. Additionally, across models 1 through 3, closeness centrality does not have a significant effect on legitimate innovation adoption. These findings refine

Burt’s (1992) claim by specifying that for legitimate innovations, individuals who span more groups are more likely to participate because they have greater access to information and opportunities throughout the organization.

For corrupt innovation adoption, the story is a bit more complex. In model 4, where we do not control for alters who adopted prior to ego, network constraint behaves similar to its behavior in the legitimate innovation model in that network constraint is the only relevant variable and negatively affects adoption at -8.47 (p <

.10). On the other hand, if the number of alters who adopted prior to ego is introduced into the model (see models 5 and 6), network constraint loses its significance and

closeness centrality becomes statistically significant at 20.88 (p < .05). Low network constraint is the only significant coefficient when we do not control for proximity to a previous corrupt adopter. However, the effect of network constraint is attenuated when previously adopting alters are controlled for. These findings provide evidence in support of hypotheses 2a and 2b.

To conceptualize it another way, when we include a measure for the number of friends or colleagues who have adopted a corrupt innovation in the previous time period, having short paths to others becomes the relevant measure and access to non- redundant information is no longer significant. Greater closeness centrality increases the odds of adoption when someone involved in a corrupt innovation is proximate.

However, controlling for an alter who has previously adopted a legitimate innovation does not change the significance of network constraint. It is only low network constraint or increased access to other groups of individuals within the aggregated network (via your own connections) that raises the likelihood of adopting a legitimate innovation. This finding reinforces the overall claim that network mechanisms that lead to network participation differ by content type.

Table 4.b. Measures by Year 1998 1999 2000 2001 2002 Variable Mean SD Mean SD Mean SD Mean SD Mean SD Ego Measures Network Constraint 0.92 0.23 0.73 0.35 0.62 0.39 0.63 0.39 0.64 0.39 Closeness Centrality 0.36 0.33 0.29 0.27 0.24 0.25 0.23 0.25 0.23 0.25 Legitimate Adoption 0.01 0.10 0.01 0.09 0.09 0.29 0.09 0.28 0.08 0.27 Corrupt Adoption 0.00 0.00 0.00 0.00 0.01 0.08 0.01 0.09 0.01 0.09 Alter Measures Number of Alters who Adopt Legitimate 0.00 0.07 0.24 0.75 3.91 11.93 6.56 22.84 6.66 23.64 Number of Alters who Adopt Corrupt 0.00 0.00 0.00 0.00 0.58 2.24 1.02 4.13 0.89 3.71 Observations 422 1,791 12,136 25,786 27,827

Table 4.c. Correlations among Selected Independent Variables for Corrupt Adoption

Variable Mean (SD) (1) (2) (3) (1) Network Constraint -0.933 (0.843) 1.000 (2) Closeness Centrality 0.202 (0.738) -0.642*** 1.000 (3) Number of Legitimate Alters (lagged) 33.058 (56.264) -0.464*** 0.332*** 1.000

(4) Number of Corrupt Alters (lagged) 4.32 (9.366) -0.472*** 0.385*** 0.791***

* p < .10; ** p < .05; *** p < .01

Table 4.d. Correlations among Selected Independent Variables for Legitimate Adoption

Variable Mean (SD) (1) (2) (3) (1) Network Constraint -1.331 (0.57) 1.000 (2) Closeness Centrality 0.303 (0.502) -0.341*** 1.000 (3) Number of Legitimate Alters (lagged) 75.928(104.238) -0.327*** 0.162*** 1.000 (4) Number of Corrupt Alters (lagged) 17.65 (18.885) -0.397*** 0.203*** 0.806***

* p < .10; ** p < .05; *** p < .01

Table 4.e. Conditional Logit Estimates of Legitimate and Corrupt Innovation Adoption by Enron Employees, 1998-2002 Legitimate Innovation Adoption Corrupt Innovation Adoption (1) (2) (3) (4) (5) (6) Network Constraint (square root) -1.462*** -2.558*** -2.570*** -8.466* -3.356 -3.334 (0.220) (0.599) (0.588) (4.334) (5.485) (5.503) Closeness Centrality 0.084 0.143 0.265 -1.135** 20.400** 20.880** (0.150) (0.273) (0.273) (0.447) (8.302) (8.792) Legitimate Alters (lagged) -0.002 -0.097** (0.002) (0.041) Corrupt Alters (lagged) 0.172*** 0.151 (0.0518) (0.102) Legitimate Alter X Network Constraint -0.051** (0.022) Corrupt Alter X Closeness Centrality 0.035 (0.153)

Individuals 1,913 1,851 1,851 195 147 147 Observations 7,999 5,884 5,884 898 559 559 Pseudo-R2 (within individual cases) 0.890 0.892 0.894 0.763 0.792 0.803

* p < .10; ** p < .05; *** p < .01

Conclusion

In this section, I demonstrate that the individual’s position with the network influences the likelihood of him receiving information about a particular innovation. I show that content type alters the mechanisms by which organizational member are recruited into communication sub-structures. This study specifies that brokerage positions are optimal for acquiring overt information, and the magnet position increases access to covert content in organizations.

When certain individuals with particular network characteristics are recruited, their relations and position alter the existing sub-structure both at the time of the recruitment. This has implications for the topology of the network. At the time of participation, member A’s structural characteristics, such as number of alters or frequency of communication with those involved in the activity, will have implications for the communication structure she has joined. When subsequent members then join the network, it is from the existing member’s alters that the recruits are drawn.

In the next chapter, I investigate communication patterns for individuals who are recruited into both corrupt and legitimate networks.

Chapter 5 – Individual Behavior in Covert and Overt Networks

Introduction

In the previous section, I have shown that content-specific networks vary on dimensions that have repercussions for accessing information. In this chapter, I demonstrate that individual communication patterns for the same person will differ between overt and covert information. A central argument of this dissertation is that the micro-communication strategies of organizational members lead to different ego- network forms, which have global effects on the entire social structure. These differences, I contend, emerge from the decisions members make when sharing information with others, such as what to share and with whom. This chapter explores both quantitatively and qualitatively the different behaviors of individuals who participate in overt and covert types of innovation. I first explore the patterns of content-specific communication ties for individuals who participate in both legitimate and corrupt endeavors. Next, I provide qualitative evidence to support my claim that type of content drives actors to improve or hinder access to information in their local structure.

Previous models of diffusion and communication networks have focused largely on the structure of relations rather than the information transmitted between parties (see Valente 2005 for review). With few exceptions (e.g., Marsden 1986), micro-level models of interaction, such as Coleman’s (1990) model of purposive action, assume unrestricted access to information for all existing relationships. In other words, regardless of content, these models assume information has the potential to

flow uniformly where a tie exists. I expand upon this line of research and contend that the type of information transmitted between individuals leads to variation in behavior and network structure. My model differs from traditional social network analysis in which the relations are the primary focus by treating ties between actors as channels of information. In my model, the information that is conveyed between individuals informs the relations and is the underlying mechanism for network dynamics such as tie formation and dissolution.

In this section, I propose that individuals who are involved in both legitimate and corrupt endeavors create characteristically different network forms through their communication decisions. When individuals are faced with different types of content, specifically information linked to overt and covert activities, they must make trade- offs between operational efficiency and secrecy (Baker and Faulkner 1993). By examining individuals’ entire social networks of relations and extracting communications for both legitimate and corrupt projects, I show that actors interact with their alters strategically to maximize either information openness or control. In addition, actors will attempt to manipulate the ties between the alters that surround them.

For models of interaction, I draw from the theoretical foundations of social exchange theory (e.g., Homans 1974, Blau 1964, and Emerson 1976), which combine behaviorism (Homans, Emerson) with concepts and principles adapted from microeconomics (Blau). For the purposes of this study, I assume that an actor is

“purposive” (to borrow Coleman’s (1990) terminology) but socially constrained by

her organizational and institutional setting. I also assume that the actor has the capacity to act on her preferences so as to maximize her outcome. The key to understanding how individuals affect their local structure is the triad, which is foundational for examining strategic interaction.

As mentioned in Chapter 2, Simmel (1950) provided one of the first and most comprehensive theoretical arguments for the importance of triads. More than simply the number of participants, the dynamics of a triad are fundamentally different than those of a dyad. As an initial case, let us consider a balanced triad or what I term a

Simmelian triad. In this micro-structure, ego A shares a strong tie with both of her alters B and C, and alters B and C also share a strong tie (see figure 2.b). 18 Unlike in a dyadic relationship, the presence of a third-party introduces the possibility of a majority. This means that within a triad, an individual’s power is somewhat diminished because she can be “out-voted” by a majority of two. In addition, the presence of a third party can serve to temper disagreements and conflicts by arbitrating disputes between the alters. This structure helps to reinforce itself by bolstering cooperation and facilitating cohesion. Additionally, these structures are found to be more common and more stable over time (e.g., Heider's (1958) Balance Theory).

Conceptually, the Simmelian triad is ideal for sharing information effectively, encouraging cooperation, and reducing conflict. If we consider a shared task where cooperation is essential and there are no restrictions on access to information, as is the

18 This term is adapted from Krackhardt’s “Simmelian Tie”; however, the focus here is on the role of the triadic relationship.

case with a legitimate innovation, we can assume that a Simmelian Triad is the preferred arrangement for participants; individuals may therefore encourage open reciprocal ties between their alters. This increases network closure for both the individual’s network and the aggregate structure.

Alternatively, when the priority is to be covert, ego may intentionally obstruct a relation between alters B and C (what Simmel (1950) terms a divide et impera,

“divide and conquer”) to obtain an advantage for herself. This concept is similar to

Burt’s (1987) broker concept with one notable and important exception: in this case, ego actively tries to attain or maintain this arrangement.19 She communicates with both alter B and alter C, but she discourages alter B and alter C from sharing a relation or information between them (see figure 2.c). Social exchange theory provides rich evidence for the benefits of a brokerage position within micro-structures (Bonacich

1998; Cook and Emerson 1978; Markovsky, Willer, Patton 1988). In this arrangement, there is the possibility that alters may not even be aware of their position in relation to others. Hence, with a covert network individuals would be less aware of all of the participants and few would be fully informed about the associated activities than in an overt setting. Lacking an accurate understanding of the network can stymie an alter since knowledge of the structure is highly correlated with power within that structure

(Krackhardt 1990). Maintaining distance between alters also creates network cleavages, rendering it difficult to implicate the entire group. In a covert endeavor, this

19 Another minor difference here is the focus on the micro-structure or the ego-centric network, while Burt’s conception takes into account the complete social structure to determine brokerage potential.

is advantageous because if one actor is compromised or found out, he will not be able to name all of the participants. Discouraging communication exchange between alters also reduces the amount of communication traffic that occurs, minimizing possible detection.

Two additional network characteristics pertain to the structural means of information control: LUBness and hierarchy. The first measure, LUBness, captures the proportion of upper-bounds or “tie-breakers”, which are individuals to whom multiple individuals in the network report. LUBness is one of four measures suggested by

Krackhardt (1994) for summarizing informal structures. This position can be conceptualized as the “boss” in a formal reporting structure. The least-upper-bound is the closest “boss” or direct authority. When the network has a high LUBness score, the form serves to funnel information into a few central actors rather than throughout the network (see figure 2.e). Having multiple individuals communicate to a “boss” privileges that individual with information and puts her in a position of power. Similar to formal organizations like the military, the individuals at the top of the system sit in the most influential positions and the individuals at the bottom are in the least influential positions.

In a directed network, two actors i and j may be said to share an upper bound if there exists some actor k that either i or j may “appeal” to via their ties. Actor k is known as a least upper bound for i and j because she is the closest node that both parties can petition for information or a decision. Deviation from this condition can be measured by counting the numbers of pairs of nodes that do not share an upper bound

relative to the number of pairs that can based on the number of actors and the span of the control upper bound.

The LUBness of V is given by

 V  1−   max( V )

Where V = the number of pairs that have no LUB, summed over all components, and:

− − = (Nn )(1 Nn )2 Max (Vn ) 2

When all vertex pairs have a least upper bound, LUBness is equal to 1.

The third measure, network hierarchy, presents the extent to which people are asymmetrically reachable, or the proportion of non-reciprocated ties (see figure 2.d)

(Krackhardt 1994). Baker and Faulkner (1993) investigated conspiracies in the heavy electrical equipment industry and revealed that hierarchical communication was required for organizational conspiracies to be carried out. A hierarchical structure that contains a high number of directed ties reflects differences in power or status within the network with information flowing to the most powerful or highest status members

(Brass, Butterfield, and Skaggs 1998).

A completely hierarchical network is one in which all communication flows in one direction. By counting the number of pairs that have reciprocated ties relative to the number of observed pairs of ties, we can assess the degree to which a structure deviates from a complete hierarchy. In other words, a structure is less hierarchical when there are a higher proportion of reciprocated ties. A hierarchy index equal to 1

means that none of the pairs share a reciprocated tie; an index equal to 0 means that all ties are reciprocated. Provided below is the formula for hierarchy:

 V  1−   max( V )

Where V is the number of symmetrical pairs and max ( V) is the total number of pairs.

If we consider that individuals are acting purposively towards certain ends, yet are confined by social and organizational constraints, then we can expect them to employ different strategies to communicate corrupt and legitimate information.

Therefore, when communicating corrupt information, egos will divide et impera, leading to few Simmelian triads in the communication structure as compared to the networks in which the information is legitimate. In corrupt networks, the majority of the nodes will be informants to centralized actors. Therefore, corrupt structures will have a higher degree of LUBness. Individuals will also maintain more asymmetry in their communication, causing their ego-networks to be more hierarchical. This means that the information in corrupt communication networks will flow towards certain highly privileged individuals, reflecting a status and/or power order (Bonacich

1987). 20

20 It is important to highlight the distinction that Cook et al. (1983) make between positive and negative exchange systems. Communication systems are akin to positive exchange networks where an actor’s power is a positive function of the powers of those one has power over.

Below I revisit the three proposed hypotheses for strategic interactions based on content types:

Hypothesis 3: Individual ego-networks will have more closure when communicating about legitimate information.

Hypothesis 4: Individual ego-networks will be more centralized when communicating about corrupt information.

Hypothesis 5: Individual ego-networks will have more asymmetrical relations when communicating about corrupt information.

Data and Methodology

In order to analyze different communication patterns for the same person, I first identified 227 Enron employees who participated in both corrupt and legitimate networks. 21 In table 5.a, I present the number of individuals by the content-specific structures. For example, of the 66 Enron employees who were involved in the legitimate project, Dabhol, 37 also participated in the corrupt project, Chewco.

21 When studying corrupt versus legitimate activities, there is the question of whether the individuals had some unobserved heterogeneity, such as belonging to certain sub- organizations in the structure. Analyzing the group that participates in both types of networks mitigates this concern since the same individual can be compared across the two content types.

Table 5.a. Number of Participates in Both Legitimate and Corrupt Innovations (N=227) Chewco Jedi Talon Adoption Adoption Adoption Variable (N=85) (N=75) (N=75) Blockbuster Adoption ( N=75) 45 41 3 Dabhol Adoption ( N=66) 37 41 3 Enron Online Adoption ( N=191) 63 151 19

In the figure that follows, I graphically display the number of individuals who participated in multiple sub-structures by legitimate and corrupt. The first bar along the top ( N= 160) depicts all of the individuals who participated in only one legitimate innovation. The color blocks within the bar display the number of people who participated in one, two, or three corrupt networks. For example, there were only nine people who participated in only one legitimate innovation and also in three corrupt networks. Interestingly, CEO Kenneth Lay participated in five of the six networks studied here.

Individuals in Corrupt Networks 1 (N=191) 2 (N=26) 3 (N=10)

1 (N=160) 138 13 9

2 (N=29) 25 3 1

Kenneth Lay participated in 5 of the 6 networks. 3 (N=38) 28 10 Individuals in Legitimate Networks in Legitimate Individuals (Total N=227)

Figure 5.a Individuals who Participated in Both Legitimate and Corrupt Innovations

The correlations across projects can be seen in table 5.b. There is a positive and significant correlation between participation across all of the endeavors. This could be an indication that this particular sample of individuals had a tendency to be active in any new project at Enron. Also, note that the correlation of participation was generally higher for the corrupt endeavors. Perhaps this suggests that once these individuals rationalized their participation in one corrupt network, it was easier to recruit them into another corrupt activity. Nevertheless, it was not the same group of corrupt participants who were involved in all three networks.

Table 5.b. Correlations among Innovation Adoption: Enron Email Networks 1998-2002 Variable (1) (2) (3) (4) (5) (1) Chewco Adoption (2) Jedi Adoption 0.23*** (3) Talon Adoption 0.34*** 0.15*** (4) Blockbuster Adoption 0.10*** 0.10*** 0.04*** (5) Dabhol Adoption 0.15*** 0.09*** 0.02*** 0.57*** (6) Enron Online Adoption 0.11*** 0.12*** 0.04*** 0.15*** 0.15***

* p < .10; ** p < .05; *** p < .01

After identifying the individuals, I created ego-centric networks of communication for all the individuals 2-degrees out from ego. Ego-centric networks or ego networks are the set of relations defined by an individual and her contacts with others. I limited the ego network to only 2-degrees out for the following reason: individual egos can know and, to some degree, shape what their alters and their alters’ alters know, but after 2-degrees it would be difficult to alter the amount of information that others may receive or influence their behavior. In other words, beyond 2-degrees is outside ego’s “sphere of influence”.

Within these networks, I coded email message content based on the topics and extracted the particular communication sub-graphs, such as all of ego’s and his alters’ communication about Talon. This permits analysis of the different types of content and how they lead to different patterns of relations. Next, I computed hierarchy and

LUBness measures for these networks. 22 Counts of the number of Simmelian triads are calculated by identifying triads. This measure captures how many of ego’s friends are also friends, creating a reciprocal triadic relationship.

Results

In table 5.c, I present t-tests between the networks of individuals for corrupt and legitimate networks. Included in the table are the mean hierarchy and LUBness scores and the Simmelian triad counts. All three measures are statistically different between the two different types of content. As predicted in hypotheses 4 and 5, both the hierarchy and LUBness means are significantly higher for the legitimate networks.

Providing support for hypothesis 3 is the significantly high count for Simmelian triads in the legitimate networks. This finding suggests that there is far more triad closure within legitimate communication networks.

22 Unfortunately, Krackhardt’s other measures of efficiency and connectedness are not appropriate for ego-networks. Ego-centric networks are completely connected sub- graphs and there is no possibility of isolates, which makes connectedness score equal to 1. The concept of efficiency is also not a useful index for a communication subset of the graph.

100

Table 5.c. t-test for Dual Adopters Ego Communication Networks Legitimate Network Corrupt Network (N= 149) (N= 129) Mean (SD) Mean(SD) t Hierarchy 0.0439(0.0048) 0.1598(0.0156) -7.5025***

LUBness 0.2244(0.0079) 0.3559(0.0251) -5.2912***

Closed Triads (Simmelian Ties) 3,007.356(235.4151) 5.1550(0.4753) 11.8632***

* p < .10; ** p < .05; *** p < .01

Qualitative Coding of Email Messages

To further illustrate the argument that individuals are strategically sharing or withholding information and attempting to manipulate proximate relations, I provide qualitative evidence and specific examples that demonstrate that individuals are actively trying to maintain secrecy and protect themselves. First, in addition to the networks’ structure varying based on content type, the content of the emails diverged.

The most striking difference is the emphasis on confidentiality when the email communicated information about a corrupt innovation. These emails stress that only those that are sanctioned by the group are permitted to read the emails. This is not altogether surprising but does add weight to my claim that individuals were maximizing secrecy and information control. As evidence that confidentiality was more commonly stressed in the corrupt email transmissions, I provide counts for the number of times the word “confidential” or a variant such as “confidentiality” appears

101

in an email messages. In this section, I also include selected quotes pertaining to confidentiality to highlight how it was conveyed.

The number of times the word “confidential” or a variant appeared in an email is presented in table 5.e. The first column lists the different communication networks by project. The second column is the total number of messages sent by project and type and is followed by the number of unique messages. To count the number of unique messages, I removed messages that were sent to multiple people or forwarded to additional recipients. In the next two columns, I provide the number of times the word “confidential” appears in total and in the unique messages. In the final column is the percentage of unique messages where “confidential” appears in the body or subject of the email. For corrupt innovation networks, “confidential” is used almost 18% of the time, where it is only used in 11% of the emails pertaining to legitimate innovations. Moreover, how the term is used is also quite different. In the case of

Chewco, Jedi and Talon, the term is used as an explicit direction to protect the information contained in the email. For example, below is a clause that appeared at the end of an email about Talon:

Talon Email Subject: Project Raptor III (2 of 2 emails) Date: 08/29/2000 08:14 AM

“++++++++++++++>CONFIDENTIALITY NOTICE>++++++++++++>

The information in this email may be confidential and/or privileged. This email is intended to be reviewed by only the individual or organization named above. If you are not the intended recipient or an authorized representative of the intended recipient, you are hereby

102

notified that any review, dissemination or copying of this email and its attachments, if any, or the information contained herein is prohibited. If you have received this email in error, please immediately notify the sender by return email and delete this email from your system.

Thank You”

The quote states in very explicit terms that the information is private and only intended for certain audiences. The email message explains that if an individual does receive the email by mistake, the individual is to notify the sender and delete the email. However, for legitimate endeavors such as Blockbuster, the term “confidential” was mentioned either in a forwarded news article or as a casual directive. For example, the message below simply had “confidential” in the subject line: 23

Blockbuster Email: To: EMPLOYEE 1/NA/Enron@ENRON Subject: Confidential Date: 12/13/2000 04:22 PM

“EMPLOYEE 1,

As part of my strategy, I want to glean information. Hopefully it will give me an understanding of their process and goals, so that I may be more effective in my discussions with their media group. Following are a few questions I will ask Daphne (I don t expect Daphne to have all the answers):1) Help me understand how Universal wants to save money on advertising? 2) Who at Universal is driving that process? (Names, Titles, etc.)3) What are the saving targets? (% or $ amount)4) Where do they want to save in regards to advertising? (Spot TV, Network TV, Cable, Radio, NP, etc.)5) What is their annual advertising

23 I have removed the actual names and replaced each with “Employee Number”.

103

budget? (Per film? Metric? Per media, etc.)6) How much does she understand about media buying? (GRP s, CPM s, Demos, Dayparts, Reach & Frequency, etc.)7) What is her role in this process? I m sure I will have a few more questions, as we speak with each other.

Regards, EMPLOYEE 2 Director Enron Media Services Global Media Risk Management”

104

Table 5.e. Confidential Messages by Innovation Network Percent of Total Total Unique Unique Total Total Messages Messages Messages Unique Network Messages with with with Messages "Confidential" "Confidential" "Confidential"

Legitimate Innovations Blockbuster 1,687.0000 272.0000 307.0000 35.0000 0.1287 Dabhol 4,290.0000 729.0000 480.0000 81.0000 0.1111 Enron Online 7,252.0000 739.0000 363.0000 60.0000 0.0812 Mean 4,409.6667 580.0000 383.3333 58.6667 0.1070 Corrupt Innovations Chewco 485.0000 79.0000 32.0000 20.0000 0.2532 Jedi 1,194.0000 151.0000 76.0000 20.0000 0.1325 Talon 764.0000 91.0000 70.0000 13.0000 0.1429 Mean 814.3333 107.0000 59.3333 17.6667 0.1762

105

Another striking characteristic of the corrupt email messages was that they commonly listed all the involved parties. In the example below, the names of particular employees involved in the activities are listed beside each of the vehicles, including JEDI:

Jedi Email Subject: Las Vegas Cogeneration II, L.L.C.- Tolling Agreement Confidentiality Obligation Date: 5/9/2001 5:32:00 AM

“… A significant number of our merchant investments are owned in whole or in part through various structured finance vehicles. Amendments, restructurings, dispositions, follow-on investments and other significant transactions related to those assets may be prohibited or may require internal or external consents. As the commercial teams may not always be familiar with these vehicles or the special considerations that may apply to these vehicles, you may want to confirm with the attorney/legal assistant that is most familiar with the particular vehicle that (i) the appropriate people are "in the loop" with what you are doing with the asset, and (ii) that there won’t be any unpleasant surprises (like a required consent that has not been obtained) five minutes before you are supposed to close whatever transaction you are working on. The contacts for the most active vehicles are: JEDI – Employee 1 JEDI II -- Employee 2, Employee 3 Raptor and Trutta -- Employee 4, Employee 5, Employee 6 Merlin – Employee 7 Rawhide – Employee 8,

106

List implicating employees Employee 9 Hawaii – Employee 10. … Again, if you have any questions, please contact Employee 11. Employee 12 Enron North America Corp .…” 24

There are two possible explanations for listing individuals. First, it may simply

have been intended to delineate the appropriate parties to contact, thus reinforcing the

existing ties within the structure and avoiding “unpleasant surprises.” Although this

may seem to run contrary to my earlier claim that corrupt participants will want to

limit knowledge of other participants, I believe it highlights the tension between

secrecy and efficiency for covert innovations. These individuals cannot implement the

innovations without sharing some relationships with each other. In this email whom to

communicate with is very explicit as opposed to in the case of legitimate innovations.

Second, the effect of listing individuals also seems to act as a social pact. In essence, it

signals that if ego is caught, his alters will also be implicated with him. Thus, the list

creates a social contract in which there may not be complete trust between members.

Conclusion

In this chapter, I provide evidence that individuals will channel legitimate and

corrupt information quite differently through their ego-networks. Both the quantitative

and qualitative analyses provide support for my claim that the strategic

communication of information shapes both the individual’s local network and the

surrounding social structure. Individual decisions about what information to share and

with whom to share it can have dramatic implications for the social structure, such as

24 I have added the italicized section in the text to underscore the claim that corrupt participants were in fact attempting to control and manipulate information within the network.

107

leading to cleavages or increasing reciprocal communications. The overall communication network that arises also has implications for the organization. For example, the increased connectivity may improve the firm’s ability to adapt or lead to more efficient implementations of innovations.

Extant research on both network positions and social exchange theory has demonstrated the benefits of a brokerage position; however, little of this research has investigated when actors may be strategic in achieving such positions within the social system. For accomplishing a legitimate goal, it may be advantageous for individuals to increase their local cohesion and therefore reduce their brokerage opportunities. On the other hand, when the enterprise is corrupt, maintaining disconnected alters provides greater control over information and the alters.

108

Chapter 6 - Covert and Overt Content Networks

Introduction

To review, in the previous two chapters I demonstrated that how individuals acquire information and how they communicate will differ based on the type of content, focusing specifically on legitimate and corrupt information. In this section, I propose that the structures vary for the legitimate and corrupt innovations’ content- specific networks. These differences are driven by individual choices at the dyadic and triadic level discussed in Chapter 5.

Covert networks face the dilemma of maintaining secrecy and ensuring the necessary coordination between their members. Baker and Faulkner (1993) in their analysis of conspiracies note that the constraint of secrecy distinguishes the covert organization from the overt organization. In their view, corrupt organizations must solve a fundamental dilemma of how to stay secret and at the same time ensure the necessary coordination and control of their members. This trade-off hinges on access to information for the network members. On the one hand, to maximize organizational efficiency in the context of an innovation implementation, members must have open access to one another to find necessary information. Or on the other hand, to maximize secrecy, limited information is shared throughout the network and communication about the activity is restricted. Covert information and the network members’ communications must be tightly constrained in order to limit exposure and possible detection. I analyze two sets of content-specific communication structures where the organizational outcome is the same – to implement an administrative innovation; however, one set is legitimate and the other is corrupt.

109

The extant literature that addresses covert organizations characterizes them as structures that lie somewhere between hierarchical and completely decentralized

(Mishal and Rosenthal 2005). McAllister (2004) notes that the internal communication and coordination of covert networks are organized according to an amorphous structure. Two works that have considered the information structures of covert information provide evidence for the trade-off between organizational efficiency and secrecy based on simulation of data. In Lindelauf, Borm and Hamers’s (2009) simulation study, communication structures for several different secrecy and information scenarios were analyzed to provide optimal network forms. 25 Baccara and

Bar-Isaac (2006) analyzed the tradeoff between diffusing information widely through the organization at the cost of leaving the organization more vulnerable to external threats. Although some current studies have applied social network analysis to the study of covert networks (Koschade 2006; Magouirk et al. 2008; Sparrow 1991;

Kinsella 2008) with the aim of detection and disruption, the research does not make comparisons to networks with legitimate aims. A systematic comparison, such as the one I conduct here, empirically highlights the distinguishing features of covert networks as compared to overt networks for the same setting.

Empirical observations suggest that communication structures will affect group performance (see, for instance, Bavelas et al. (1950) and Guetzkow and Simon

(1955)). In the 1950s, Bavelas and his colleagues experimented with communication

25 Lindelauf, Borm and Hamers’s (2009) definition of an optimal network is one in which there cannot be another network that has both a higher secrecy measure and information measure.

110

patterns among teams and examined how these patterns influenced team effectiveness

(Bavelas 1950; Bavelas et al., 1951; Leavitt 1951). This research focused on the interplay of information complexity and structure and found that only when the information is complex and the tasks required creative problem-solving did decentralized communication configurations perform better than centralized ones. In a subsequent study, Baldwin et al. (1997) found evidence that more ties within a team enhanced team performance as well.

Proposed Network Structures

Within a covert network, conflicting objectives simultaneously inform behavior. First, communication should be minimized to reduce the danger of exposure; second, sufficient channels of information should exist so as not to undermine coordination. Provided that legitimate networks prioritize organizational efficiency and coordination, these networks are likely to be connected structures with reciprocated communications, which provide the benefits of access to information throughout the structure and the ability to orchestrate communication between actors efficiently. In contrast, corrupt networks will be faced with maintaining secrecy in addition to implementing the project; this constrains the amount of connections in the network. The ties between individuals in a corrupt structure will tend to be asymmetrical so information will move in particular directions within the group, privileging certain members. Asymmetry reinforces a power structure that serves to control both information and individuals.

111

Overt Innovation Networks

When an innovation is introduced, the involved individuals need to communicate with each other to acquire pertinent information and to learn how to implement the innovation. In the case of legitimate innovations, the project’s goals mirror the objectives of the organization. For such legitimate innovation structures, the network form is likely to be highly connected because this improves the ability to acquire information or assistance. Connectedness is important for innovations and non-routine practices in an organization, particularly because the connected substructure allows members to consult one another through established communication ties and relations (Krackhardt 1994). A connected sub-group also improves the ability to problem-solve creatively (Bavelas 1950). In well-connected groups, members are more likely than non-members to share similar information (Granovetter 1973).

Information will also diffuse more rapidly in a well-connected group because members are more likely to share ties.

Reciprocal communication is particularly important for innovations or implementation of a new practice because it introduces uncertainties that individuals must address. A reciprocal relation between two actors reflects the mutual exchange of information; this implies that information flows back and forth between individuals rather than in just one direction. These exchanges permit discourse between members to help them clarify information or refine ideas. Since efficiency is central to most legitimate innovation implementations, communication between members is likely to be reciprocated.

112

Covert Innovation Networks

To implement a corrupt innovation, members might seek to control patterns of communication. In contrast to legitimate innovations, corrupt innovation communication networks will attempt to maximize concealment and secrecy rather than efficiency (Baker and Faulkner 1993). Within an organization, corrupt innovation participants may seek to protect themselves through graduated divisions of labor that serve as organizational buffers, sealing off groups of members from each other

(Simmel 1950; Goffman 1970). A division or absence of communication between groups reduces the connectedness of the structure (Krackhardt 1994). In view of these findings, the expectation is that corrupt networks will be less connected, leading to a fractured network comprised of clusters and isolates.

In addition, previous research shows that illegal activities tend to yield centralized, hierarchical networks (Baker and Faulkner 1993; Simmel 1950). Baker and Faulkner (1993) investigated conspiracies in the heavy electrical equipment industry and revealed that centralized or hierarchical communication was required for organizational conspiracies to be carried out. Hierarchical network structures are more favorable for maximizing control over information. A structure that contains a high number of directed ties reduces member discourse about the corruption. Directed ties also imply differences in power or status within the network, with information flowing to the most powerful or highest status members (Brass, Butterfield, and Skaggs 1998).

In the case of corrupt networks or networks that prize secrecy, a hierarchical form will reinforce the power of some individuals to manage information and thus, other network members.

113

Below are the final two hypotheses for content-specific network forms:

Hypothesis 6: Corrupt communication networks will have more asymmetrical ties than legitimate networks.

Hypothesis 7: Legitimate communication networks will be more connected than corrupt networks.

Network Measures

Krackhardt (1994) provides four elegant definitions and measures for understanding informal structures, two of which I apply in this chapter. First, connectedness measures the underlying proportion of the graph that is a single component. A component is a sub-group that is completely connected: all individuals share at least one tie to the group. A connectedness score of 1 means that a graph is connected into a single component and all actors are embedded in the social structure.

Conversely, a connectedness score of 0 means that all members are isolated from one another and do not share ties. When a social network is comprised of multiple components (un-connected sub-groups), the proportion of unreachable actors can be high. Below is Krackhardt’s (1994) formula for connectedness:

 V  1−    N(N − 2/)1 

where V is the number of pairs that are not mutually reachable, and N(N-1)/2 is the total number of pairs.

In the case of legitimate networks, participants have little need for secrecy and therefore can communicate via the means that are most efficient for the group. On the

114

other hand, for corrupt networks, there are advantages to creating organizational buffers, so that if one individual or group is found out, the others involved are not implicated (Goffman 1970). For this reason, I argue that in corrupt networks, actors will conspire to isolate individuals, reducing access to the group and, therefore, access to information. Subsequently, I expect the connectedness index for legitimate networks to be high, while the corrupt networks should have low connectedness scores.

The second measure, hierarchy, illustrates the extent to which people are asymmetrically reachable, or the proportion of non-reciprocated ties (Krackhardt

1994). A completely hierarchical network is one in which all communication flows in one direction. Asymmetrical or hierarchical structures reflect status and/or power differences within the group. Certain privileged individuals will receive information as hubs, but will not reciprocate. By counting the number of pairs that have reciprocated ties relative to the number of observed pairs of ties, we can assess the degree to which a structure deviates from a complete hierarchy. In other words, a structure is less hierarchical when there are a higher proportion of reciprocated ties. A hierarchy index equal to 1 means that none of the pairs share a reciprocated tie; an index equal to 0 means that all ties are reciprocated. Provided below is the formula for hierarchy:

 V  1−   max( V )

Where V is the number of symmetrical pairs and max( V) is the total number of pairs.

115

Legitimate innovation networks are likely to have fewer asymmetrical relations. A less hierarchical form increases individual ability to orchestrate communication efficiently and share information between individuals. In contrast, corrupt innovations contain many asymmetrical relations between members. This hierarchical structure limits access to information and allows certain individuals to control that information.

Data and Methodology

I am interested in showing differences between the sets of content-specific structures, but unfortunately comparing global measures of networks such as connectedness and hierarchy can prove difficult given network variation in size and density. 26 All of the networks here have different numbers of nodes and edges, which makes direct comparison of graph statistics meaningless. The application of a classic method of data normalization, z-score transformation, provides a way of standardizing data across a range of networks and allows for comparison of network data independent of the original network size and density (Robins and Alexander 2004).

To obtain a z-score for each network, the properties of the networks were compared against sets of 100 random network simulations conditioned on size and density. Comparing the network’s z-scores based on conditioned simulated networks avoids the related issues of comparing networks of different sizes (Pattison,

Wasserman, Robins, and Kanfer 2000). I expect that when grouped by content type, the observed networks will differ in similar ways from the simulated graph

26 Network density is the proportion of observed ties over the total number of possible ties (Wasserman and Faust 1994).

116

distributions. For instance, since legitimate content permits employees to organize relations in certain ways, the same process will be at work for all three networks; thus they will all differ from the conditioned simulations in a similar way. Conversely, the content-specific network’s z-scores are expected to differ between the two sets of legitimate and corrupt innovation networks because of different underlying social processes.

Results

In tables 6.a and 6.b, I present the global network measures for the content- specific sub-structures. I compare the three corrupt networks (JEDI, Chewco, and

Talon) to the three legitimate innovation networks (Enron Online, the Dabhol Project, and the Blockbuster Venture). The innovation networks are grouped by content-type. I include the mean statistics for both sets of content-specific networks. The first column, innovation duration in years, shows that the corrupt and legitimate networks had similar durations; the range was 2-3 years for both sets of networks. The number of participants is found in the second column, network size; as one would expect, significantly more individuals at Enron were involved in legitimate innovations. The mean network size was 676 for legitimate innovation networks and only 118 for corrupt networks (see table 6.a). The next two columns show network diameter and average path length. Network diameter is an important measure, because it quantifies the farthest possible path between two individuals included in the network

(Wasserman and Faust 1994). The diameters of the legitimate networks are larger, on average, than those of the corrupt networks, ranging from 8 for the legitimate networks to 4.67 for the corrupt networks. If we consider transmission of a message

117

between participants on different sides of the network and assume that messages take the short routes to be conveyed, the diameter is the longest path for the message to travel. Average path length is the mean distance between every possible pair in the network; this figure is also slightly larger in the legitimate networks: 2.762 for legitimate versus 1.979 for corrupt networks. Average path length conveys the mean number of intermediaries involved when transmitting a message between pairs of individuals. It is not surprising that both diameter and average path length are larger for legitimate networks, as these networks contain more members. Although diameter and average path length can be informative, it is difficult to compare these measures with certainty given the variation in network size.

118

Table 6.a. Innovation Network Descriptives I Innovation Average Network Network Graph Network Duration Path Size Diameter Density (Years) Length Legitimate Innovations Blockbuster 3 418 8 2.714 0.0042 Dabhol 3 264 7 2.489 0.0069 Enron Online 2 1,346 9 3.084 0.0021 Mean 2.667 676 8 2.762 0.0044 Corrupt Innovations Chewco 3 137 4 1.955 0.0075 Jedi 3 171 6 2.089 0.0069 Talon 2 46 4 1.892 0.0353 Mean 2.667 118 4.667 1.979 0.0166

Note: All of the global network measures are based on multiplex networks with the exception of hierarchy, which is based only on directed networks.

119

Table 6.b. Innovation Network Descriptives II Simulated Simulated Observed Connectedness Observed Hierarchy Network Distribution for Distribution for Connectedness (z-score) Hierarchy (z-score) Connectedness Hierarchy Legitimate Blockbuster 0.8369 0.9924(0.0064) -24.1739 0.0084 0.0135(0.0001) -103.1829 Dabhol 0.8708 0.9983(0.0034) -37.7081 0.0042 0.0273(0.0001) -128.4409 Enron Online 0.9778 1.0000(0.0000) 0.0000 0.0138 0.0101(0.0000) -594.6436 Mean 0.8952 0.9969(0.0033) -20.6273 0.0088 0.0170(0.0001) -275.4225 Corrupt Chewco 0.2637 0.9713(0.0199) -35.5336 0.0149 0.0300(0.0003) -54.7858 Jedi 0.8033 0.9988(0.0037) -52.8658 0.0139 0.0438(0.0004) -80.7115 Talon 0.571 0.9913(0.0185) -22.7211 0.0705 0.1122(0.0034) -12.2875 Mean 0.546 0.9871(0.014) -37.0402 0.0331 0.0620(0.0013) -49.2616

Note: All of the global network measures are based on multiplex networks with the exception of hierarchy, which is based only on directed networks.

120

In table 6.b, I present the statistics from the observed content-specific networks and the means, standard deviations, and z-scores from the simulated distributions. The z-scores were calculated by taking the observed graph indices, such as connectedness, minus the averaged measure across all of the random networks divided by the standard deviation. The z- scores in table 6.b provide standardized measures for comparing indices across the networks despite variation in size or density (Robins and Alexander 2004). From table 6.b, it is at once apparent that both sets of observed content-specific networks differ substantially from the conditional random graph distributions, as well as from each other. Since the observed network structures are clearly different from their conditioned simulations, we can infer that additional social processes are in operation.

By comparing the observed graphs against the distributions of random graphs of appropriate size and density, I am able to draw conclusions about underlying differences in the networks (Robins and Alexander 2004). In table 6.b, the evidence shows that the legitimate networks are not more connected structures than their random counterparts. The mean connectedness z-score for all three legitimate networks is -28.202. The corrupt networks have a much lower connectedness score of -43.546. This indicates that, on average, the legitimate networks resembled a single structure rather than separate factions. On the other hand, on average a greater portion of corrupt innovation participants are not connected to the content-specific network. However, when we consider the range of connectedness for both legitimate and corrupt networks, it becomes evident that the differences between the groups are driven primarily by the very low z-score (0.000) of the Enron Online network.

121

Considering this, the groups may be similarly connected, and the evidence supporting hypothesis 7 is weak. If this is the case, it is still interesting to note that despite being larger in size, the legitimate structures are just as connected as the considerably smaller corrupt structures. Even though the legitimate innovation networks are much larger than the corrupt ones, individuals can reach one another just as readily.

The legitimate innovation networks also appear to be less hierarchical than the simulations and the corrupt networks given their z-scores providing some evidence in support of hypothesis 6. The z-scores for hierarchy in the legitimate networks are extreme, with an average score of -275.423, as compared to the average hierarchy z-score for corrupt networks of -49.261. Pairs of individuals within a legitimate structure are far more likely to share reciprocated ties. This indicates that there is greater exchange of communication in legitimate structures. On the other hand, corrupt networks are more asymmetrical or hierarchical than legitimate networks, directing information to flow in only one direction. Overall, legitimate and corrupt networks do appear to differ, particularly on the dimensions of hierarchy and, to some extent, connectedness. Given these findings, I conclude that corrupt innovations reflect a structure similar to a rimless hub-and-spoke structure, a central clique surrounded by satellite ties.

In a rimless hub-and-spoke structure, a small clique serves as the hub at the center of the network. Within the hub, the group is cohesive and the ties are strong, indicating that trust is high. Surrounding the hub are spokes, ties to individuals who do not share ties with each other. The individual spokes generally have skills important to the implementation of the

122

innovation that those in the hub do not. For example, the corrupt innovations explored here required the skills of legal professionals and accountants to be realized. The spokes also provide the hub with access to information and resources throughout the organization. The hub requires the spokes to implement the innovation but does not share the same level of trust with the spokes as it does with those in the hub. Keeping the spokes isolated helps to disguise the illegitimacy of the innovation and also mitigates the risks of coalitions or possible whistle- blowers. This structure leads to a two-stage adoption process. First, those in the hub who are close to an early adopter will be the most vulnerable to adoption. Next, once the individuals in the hub adopt, the individual satellite adopters or the spokes become vulnerable to adoption.

Thus, spokes permit the hub to span structural holes. Maintaining a network structure full of structural holes not only maximizes access to information, but also allows members of the networks to be played off against each other, permitting malfeasance and corporate mis- governance (Brass, Butterfield, and Skaggs 1998; Burt 2004; Mitchell 2003). This ideal type maximizes secrecy and control, in contrast to the ideal type of legitimate networks, where individuals freely share information for efficiency and effectiveness.

Conclusion

In this final study, I look at the topological implications of covert and overt information within an organization. I find evidence for the endogenous mechanisms that lead to particular network characteristics when communication is parsed by content. For corrupt communications, the structure is sparse with comparably low connectivity between actors.

Communication is unidirectional, which reflects an underlying power structure that carries out

123

implementation and enforcement. Traffic along existing ties is also minimal, which reduces the possibility for discovery.

The analysis in this paper presents a framework for understanding communication networks when content is considered. Due to the scarcity of data in the domain of corrupt networks, this work provides an additional look into clandestine structures. Future research could apply this analysis to other settings such as markets where collusion is detected.

124

Chapter 7 – Networks, Gender and Whistle-blowers

After Enron’s famous demise, which was catalyzed by Sherron Watkin’s letter to then-

CEO Kenneth Lay, some speculated that females are more likely to speak out against organizational corruption. In this final section, I briefly explore possible structural differences based on gender and investigate the network characteristics of two “whistle-blowers” at

Enron. This chapter is only meant to be exploratory and does not provide any conclusive evidence or tests.

Organizations, Social Networks, and Gender

Prior research suggests that women’s social networks differ from men’s in a variety of ways. In general, ego networks show strong homophily, especially on characteristics such as sex, race, ethnicity, age, and education (Marsden 1988). Therefore, male and female networks tend to be gendered. Although men and women have networks of similar sizes, women’s networks generally have fewer non-kin relations (Marsden 1987).

Work settings also differentiate men and women’s opportunity structures. In a study on gendered occupational networks, Campbell (1988) found that women know people in fewer occupational categories than men, and women are less likely to know individuals outside of their own occupational category than men. According to Kanter (1977), when women do find themselves in high-ranking positions within an organization, they are more likely to be in a distinct minority, and their “token” status puts them under additional pressures not felt by their male counterparts. Even when women occupy the same rank or title

125

as men they tend not to have the same network centrality within the informal structure (Kanter

1977; Smith-Lovin and McPherson 1993). This is not to say that the occupational networks of men and women will always be different. For entrepreneurs, for whom social networks are critical for success, there are very few gender differences in network structure (Aldrich,

Reese, and Dubini 1989).

Although Enron was notorious for what some called a “cowboy” culture, women often rose high in the corporate ranks. In the table that follows, I show the different network attributes based on gender at Enron. On the whole, men and women did vary on all of the attributes listed, but the differences were minimal (see table 7.a). The men were slightly more constrained by their network, which indicates a lower degree of tie-redundancy than females.

The males at Enron also had slightly higher closeness centrality than the women. However, women had a higher in and out-degree on average, meaning more incoming and outgoing ties to others in the network. Women also knew more participants on average in both the legitimate and corrupt networks. This means that given my models women would be more likely to participate in corrupt innovations. In future studies, I intend to tease apart the role of gender and content-specific sub-groups.

In the next section, I discuss the pertinent literature on whistle-blowers and examine the characteristics of Enron’s whistle-blowers.

126

Table 7.a. t-test for Gender Differences Network Characteristics Male Female (N= 17,726) (N= 10,976) t Network Constraint 0.4477(0.0029) 0.4107(0.0037) 7.8846***

Closeness Centrality 0.2244(0.0014) 0.2350(0.0018) -4.6210***

In-degree 7.5539(0.1764) 9.5500(0.2833) -6.3127***

Out-degree 7.8576(0.1621) 8.5843(0.1846) -2.8836**

Number of Legitimate Alters 12.1428(0.2841) 14.3932(0.3883) -4.7599***

Number of Corrupt Alters 1.6953(0.0456) 2.0332(0.0689) -4.2584***

p <.05, ** p<.01, *** p<.001. Standard deviations appear in parentheses after means.

Whistle-blowers

Miceli and Near (1985) define whistle-blowers as “organization members who disclose illegal, immoral or illegitimate practices under the control of their employer to persons or organizations who may be able to effect action" (p. 6). Prior studies on whistle- blowing behavior have suggested several important factors, including power processes (Near et al. 1993), feelings of efficacy (Miceli and Near 1984), and prosocial orientation (Dozier and Miceli 1985).

In terms of the power process perspective, whistle-blowing represents an influence process where in order to persuade the dominant coalition to terminate the misconduct being committed by one or more of the organization's members, the whistle-blower attempts to exert

127

power over the organization or some of its members (Greenberger et al. 1987; Near and

Miceli 1985). The dominant coalition may respond in one of two ways. They may accept the power action and stop the wrongdoing or avoid ending the wrongdoing. In the latter scenario, the dominant group may even retaliate against the whistle-blower in an effort to change the power balance.

Other researchers have suggested that whistle-blowers may be motivated by the degree to which they perceive they will be efficacious (Farrell and Petersen 1982; Near and Miceli

1985). If organization members view the situation as potentially under their control, they should be more motivated to blow the whistle. Miceli and Near (1984) compared employees who observed wrongdoing but took no action to whistle-blowers. Their results and subsequent studies indicated that whistle-blowers were more aware of the appropriate channels for blowing the whistle within the organization and also viewed themselves as better performers than did non-whistle-blowers (Miceli and Near 1984; Miceli et al. 1989).

Some scholars have proposed that certain stable personality characteristics of individuals, such as a prosocial orientation, may lead to predictable whistle-blowing behavior.

A prosocial orientation means a concern for the welfare of others and group; however, it is not simply driven by unselfish (altruistic) motives on the part of the individual. Senneker and

Hendrick (1983) found gendered differences in prosocial behavior in that both men and women decrease helping in the presence of same-sex bystanders, but women are less likely to help in the presence of mixed-sex bystanders, whereas men are not (p. 916).

128

Given this previous research, there is no reason to believe that women would be more likely than men to report wrongdoing within their organization. In fact, these studies indicate the opposite – that men would be more predisposed to whistle-blowing.

Most individuals who are familiar with Enron know about Sherron Watkins, the woman who wrote the “whistle-blowing” letter to Enron’s then-CEO Kenneth Lay, but few have heard of Vince Kaminski. Watkin’s memo was leaked outside of Enron, but Kaminski, despite his attempts, was never able to garner the same attention.

In 1999, Kaminski not only suspected corruption at Enron, he also spoke out against it.

In his original position as head of research, his job was to lead his department in complex option modeling and pricing for the traders. When he learned of the first special purpose vehicle, Talon, he told Rick Causey, Enron’s Chief Accounting Officer, that the project “was so stupid only Andy Fastow [Enron’s CFO and mastermind behind Enron’s corrupt SPEs] could have come up with it” (Swartz and Watkins 2003: 170). After the deal went through,

Kaminski was cut off from receiving certain financial information, and his group was moved several floors to the Enron equivalent of Siberia. This did not deter him. In 2001, he even went so far as to send memos about the vehicles to Arthur Anderson employees, who were the official auditors of Enron. Below is one of the many email messages that he sent in an attempt to stop the corruption at Enron:

129

Vince Kaminski Message Date: October 04, 2001 8:28 AM Subject: LJM/Raptor valuations

“Ryan,

In the follow up to the meeting we had on Wednesday I would like to reinforce one point I made. I feel strongly that I cannot support the valuations my group has produced so far for the LJM/Raptor related transactions without examination of all the related legal documents. I feel that we did solid work based on verbal information, but I cannot guarantee the quality of the final product without looking at the contracts. These transactions are too complex and controversial to bypass due diligence requirements that you would expect from any professional.

Vince Kaminski”

Both Watkins and Kaminski had the credentials of many of the top management positions. Vince Kaminski had a doctorate in economics from what is now Warsaw Business

School and an MBA in finance from Fordham University. Sherron Watkins graduated with honors from the University of Texas, where she also received her master’s in accounting.

Each had also excelled in prior positions at Fortune 500 companies.

Neither Watkins nor Kaminski intended to attract outside attention to the activities at

Enron only to stop the misconduct. Sherron Watkin’s memo famously brought both legal and media scrutiny onto the company, but that was not her intention even by her own account. As she stated in an interview after Enron’s collapse, “When a company cooks the books, its best bet is to come clean itself. Rat-finking outside isn't going to correct the problem” (Reingold

2003).

130

Ironically, Vince Kaminski was even more outspoken about the corruption and fraudulent practices at Enron than Sherron Watkins. He, however, was never able to garner the same attention. In October of 2001, Kaminski also took the opportunity to communicate directly with Lay, but unlike Watkins, he did so publicly. Watkins had sent a memo in envelop marked “confidential”, while Kaminski confronted Lay in person surrounded by

Enron’s management. All the managing directors were brought together to review Enron’s third-quarter report and examine the corporate finances in detail. When challenged by other directors about the company’s financial situation, Lay claimed, “Well, we don’t think we did anything wrong, but knowing what we do now, we would never do it again” (Swartz and

Watkins 2003: 310). Kaminski, who was present at the meeting, rose from his chair and approached Lay. He began, “I am in the terrible position of having to disagree with you”

(Swartz and Watkins 2003: 310). He then went on to tell Lay and the directors that what

Fastow did was not only improper but “terribly stupid”.

Since the actions and intentions were similar for both Watkins and Kaminski, it seems less plausible that gender was a determinant. One possible determinant would be network position. At the very least, Watkins and Kaminski would share a tie with alters that participated in any of the corrupt networks. However, there is some evidence to suggest that organizational role may also influence willingness to intercede in the face of corruption. In previous research, individuals were more likely to blow the whistle (particularly to parties within the organization) when they held jobs in which the whistle-blowing was role- prescribed, i.e., as inspectors or auditors (Miceli and Near 1985, Miceli et al. 1989). This

131

would have described Watkin’s new position within Enron. In June 2001, Watkins was assigned the job of reviewing all of the assets Enron was considering for sale and determining the economic implications of a sale (Swartz and Watkins 2003: 268).

Below are Sherron Watkins and Vince Kaminski’s measures of network constraint, closeness centrality, and the number of alters for legitimate and corrupt innovation adoption across their tenure at Enron (see tables 7.b and 7.c). The interesting point about Sherron’s measures is that she maintains low network constraint (after 2000), but also low closeness centrality. After changing positions in the company in mid-2001, two factors change for

Waktins. First, her network centrality decreases dramatically. This may be due in large part to the position change in the organization. Second, in 2001, she associates with one alter who participates in a corrupt endeavor.

On the other hand, Vince’s network constraint and closeness centrality start high and decrease over the years. In the case of Kaminski, the number of alters who adopt either legitimate or corrupt innovations is large as compared to both the overall average and the number for Sherron. Clearly, Vince knew many of the individuals involved. Interestingly,

Kaminski’s network constraint also plummets between 2000 and 2001. In his case, in January of 2001, he was called back into service by Rick Buy, head of Risk Assessment and Control, to analyze some cross-guarantees, promises of one entity to pay obligations of a sister entity, for Talon.

132

Table 7.b. Measures for Sherron Watkins Variable 2000 2001 2002 Network Constraint 1.08 -1.19 -1.19 Closeness Centrality -1.06 -1.06 -1.06 Number of Alters who Adopt Legitimate 0.00 3.00 21.00 Number of Alters who Adopt Corrupt 0.00 1.00 6.00

Table 7.c. Measures for Vince Kaminski Variable 1999 2000 2001 2002 Network Constraint 0.52 0.50 0.01 0.01 Closeness Centrality 0.83 0.72 0.49 0.47 Number of Alters who Adopt Legitimate 0 200 359 376 Number of Alters who Adopt Corrupt 0 14 36 30

The measures in table 7.b and 7.c are simply meant to be descriptive. The challenge with modeling whistle-blowers is, of course, how rarely they occur.

Conclusions

In this section, I have briefly explored some of the differences between the network positions of men and women at Enron. Although the differences between men and women at

Enron appear trivial, the question remains whether these factors, in aggregate, would alter the experiences and choices by gender.

133

I have also looked specifically at the structural characteristics of two whistle-blowers at the organization. With only two individuals, it is difficult (if not impossible) to create a predictive model for whistle-blowing. However, I would contend that regardless if there is a quality inherent in the whistle-blowers, such as a higher prosocial orientation, there may also be structural qualities or positional correlates that increase the likelihood of speaking out against misconduct. The most obvious might be proximity to others who are involved in corruption. Other structural properties may become apparent with subsequent research.

I find little evidence that would show that there was something unique to Sherron

Watkins as a woman that made her a whistle-blower, especially given the fact that Kaminski, a man, also attempted to stop the corruption. However, the right data (social networks of a sample of whistle-blowers) could prove useful in understanding the dynamics that lead individuals to go against the grain. Future studies should explore the ego-networks of both men and women, controlling for organizational position. This would shed light on how gender influences network form and communication patterns.

134

Chapter 8 - Summary of Findings and Conclusion

Conclusions

This study demonstrates that content affects the conditions under which individuals are recruited into networks, the communication patterns of individuals, and the topology of informal networks . Disaggregating social ties by content type provides a richer understanding of both network structure and recruitment into the sub-structures.

In the first study, I shed light on how various forms of information may activate network ties differently. I demonstrated that the individual’s position within the network influences the likelihood of him receiving information about the innovation. These results show that low network constraint or increased access to other groups of individuals within the aggregated network (via your own connections) increases the likelihood of participation in a legitimate innovation, regardless of proximity to a legitimate adopter. However, this is not the case for corrupt innovations, where high closeness centrality and proximity to a prior adopter determine participation. These findings refine the brokerage argument by specifying that for overt information, such as legitimate innovations, individuals who span more groups are more likely to participate because they have greater access to information and opportunities throughout the organization. When the information is covert or secret, it does not circulate in the same manner as overt information. Instead, it is closeness centrality, direct ties to many individuals in the organization, plus the presence of someone who has participated in corruption in the previous time-period that prove to be the strongest determinants of corrupt innovation spread. Considering whether content is legitimate or corrupt changes the predictive

135

models and provides a new way to conceptualize social structures. Thus, taking content into account when investigating networks helps to disentangle the effects of social structure on individual behavior.

In my second study, I used both quantitative and qualitative analyses to provide support for my argument that the strategic communication of information shapes and alters both the individual’s local network and the surrounding social structure. I show that the characteristics of the networks arise endogenously with the objectives of the communication networks. For example, when the goal is legitimate, it may be advantageous for individuals to increase their local cohesion and therefore reduce their brokerage opportunities. On the other hand, when the enterprise is corrupt, maintaining disconnected alters provides greater control over information and alters. Additionally, the choices about communication patterns at the ego-centric level lead to sharing corrupt information asymmetrically and centralizing flows, which creates or reinforces a structural power order. These individual choices led to topological differences in the communication networks as a whole based on content. The overall communication network that emerges from the individual choices has implications for the organization. For example, a lack of connectivity may reduce the firm’s ability to adapt or handle external shocks.

In the third study, I find evidence that the nature of the information communicated changes the network form. First, despite being larger in size, legitimate innovation networks are as connected as corrupt networks. Legitimate networks are also far less hierarchical than their corrupt counterparts are. Generally, as the population increases the connectedness

136

decreases but the objective of the content promotes connectivity of the group. The same mechanism is not at play for covert structures. The lack of communication reciprocity in corrupt structures serves to reduce the opportunities for discovery and foster a power order of the members.

These findings are important to organizational research more generally and to network theory in particular. For organizational research, these results show how legitimate and corrupt content shape informal networks and influence the preconditions for group membership. The results also demonstrate how content shapes the communication patterns within an organization. For network theory, the results support the role of content specification in social network research. Beyond reinforcing the role of informal networks, my results indicate that content-types affect network form and membership. Disaggregating social networks by content shared provides insight into the dynamics of how ties form and how networks are reproduced.Given these findings, it is clear that disaggregating networks by information content presents new opportunities to better understand the link between social structure and individual behavior.

This dissertation extends existing research in at least three ways. First, I present a typology of content to characterize network structures. Based on content, I demonstrate variations in the global configurations of the networks as well as the indices that predict individual participation in sub-groups. Second, by basing my research on longitudinal data, I demonstrate the dynamic effects of network structure on individual adoption. In doing so, I find the mechanisms for participation vary between legitimate and corrupt content types.

137

Third, I develop new methods for using email to represent organizational communication networks. By combining both the one-mode and two-mode networks, I am able to capture ties between organizational members that reflect both interactions and associations. Establishing appropriate methods for email and online communication is increasingly important as communication network data become more prevalent.

Finally, by developing my account of Enron, I hope to shed light on the dark side of networks. Network theories have been remiss in their accounts of illegitimate or corrupt practices. With few exceptions, the research to date has concentrated on the positive aspects of informal networks. In addition to theoretical implications, this research also has practical applications. Understanding the underlying structures that facilitate the adoption and spread of corruption could have meaningful implications for policy-makers and businesses.

Incorporating the content of what is shared through social networks (whether in organizations, communities, or markets) helps us to better understand the underlying mechanisms that shape networks and lead to various types of participation.

138

Future Research

In regards to this data, two factors were not fully examined in the dissertation due limitations of both the data and the original coding. First, the occupational titles and roles were not available in the data. The formal characteristics of the individuals in the corporation may help to explain particular variations in network position and should be included as controls in future models. Second, the time of adoption was not captured sequentially between individuals. Although the data was coded by year of participation, it does not have the order of individual’s adoptions. Coding the data with more finite time would afford us a more detailed view of social influence between individual members. I intend to address both shortcomings in upcoming studies by acquiring additional data and refining the coding of adoption.

One topic that is not explored in this dissertation but is intrinsically linked is the role of trust. Unfortunately, neither the scope of the dissertation nor the characteristics of the data allowed for a complete examination of trust in the networks. There is plenty of anecdotal evidence that corruption and covert activities require trust between the participants like, for instance, the common idiom “thick as thieves.” In the case of Enron, it is difficult to disentangle trust from structural power. For example, was trust the mechanism that brought about group solidarity in the case of corrupt activities or was it the ability to monitor and sanction individuals who broke the pact of secrecy? Potential studies would examine how trust operates across different type of content-specific sub-groups.

The legitimate and corrupt network typology can also be extended to other settings outside of corporations. For example, this typology could be used in markets to compare

139

cooperation and collusion or in adolescent networks to compare afterschool program enrollment and gang recruitment. Exploring legitimate covert endeavors, such as undercover investigations of drug cartels may also prove the most informative test case. My expectation is that covert law-enforcement communication networks will also resemble the findings presented here. Additionally, future research should not be limited to the typology presented here. Various content types may be consequential for social structure.

Future studies could extend this research by conducting analogous laboratory experiments to specify further the micro-mechanisms of information control through network manipulation. For example, in a laboratory setting, the communication patterns of teams would be observed while subjects participate in either an overt or covert task. A controlled study would mitigate the issue of omitted variables influencing communication patterns. An alternative method would use agent-based modeling to simulate individual preferences for communicating covert and overt information. The agents would be programmed to maximize either efficiency or secrecy and efficiency based on the following parameters: asymmetry, centralization of power, and cohesion. A simulation would control for the volume and density of the communications but allow for the comparison of the structural outcomes between the two sets of agent preferences. Simulations would allow us to observe the system-level ramifications of individual decisions based on content.

140

References

Ackman, Dan. 2002. "Enron the Incredible." Forbes.com .

Aldrich, Howard E., Pat Ray Reese, and Paola Dubini. 1989. "Women on the Verge of a Breakthrough?: Networking Among Entrepreneurs in the United States and Italy." Journal of Entrepreneurship and Regional Development 1:339-356.

Allison, Paul D. 2005. Fixed Effects Regression Methods for Longitudinal Data Using SAS . Cary, NC: SAS Publishing.

Allison, Tony. 2001. " Enron's Eight-Year Power Struggle in India." in Asian Times Online .

Baccara, Mariagiovanna and Bar-Isaac, Heski. 2006. “How to Organize Crime.” Social Science Research Network .

Bailey, N.T.J. 1975. The Mathematical Theory of Infectious Diseases . London: Griffin.

Baker, Wayne E. 1984. "The Social Structure of a National Securities Market." American Journal of Sociology 89:775-811.

Baker, Wayne E. and Robert R. Faulkner. 1993. "The Social Organization of Conspiracy: Illegal Networks in the Heavy Electrical Equipment Industry." American Sociological Review 58:837-860.

Bavelas, A. 1950 “Communication Patterns in Task-oriented Groups.” Journal of the Acoustical Society of America 22: 725-730.

Bearman, Peter S, James Moody, and Katherine Stovel. 2004. "Chains of Affection: The Structure of Adolescent Romantic and Sexual Networks." American Journal of Sociology 110:44-91.

Blau, P. M. Exchange and Power in Social Life . New York: Wiley, 1964.

Bonacich, Phillip. 1987. "Power and Centrality: A Family of Measures." American Journal of Sociology 92:1170-1182.

Brass, Daniel J., Kenneth D. Butterfield, and Bruce C. Skaggs. 1998. "Relationships and Unethical Behavior: A Social Network Perspective." The Academy of Management Review 23:14-31.

Burt, Ronald. 1987. "Social Contagion and Innovation: Cohesion versus Structural Equivalence." American Journal of Sociology 92:1287-1335.

141

—. 1992. Structural Holes: The Social Structure of Competition . Cambridge, MA: Harvard University Press.

—. 2002. “The Social Capital of Structural Holes." Pp. 148-190 in The New Economic Sociology: Developments in an Emerging Field, edited by M. F. Guillén . New York: Russell Sage Foundation.

—. 2004. "Structural Holes and Good Ideas." American Journal of Sociology 110:349-99.

—. 2005. Brokerage and Closure: An Introduction to Social Capital. New York: Oxford University Press, USA.

Campbell, KE. 1988. "Gender Differences in Job-Related Networks." Work and Occupations 15.

Cartwright, Dorwin and Alvin Zander. 1968. Group Dynamics . New York: Harper & Row.

Coleman, James S., E. Katz, and H. Menzel. 1966. Medical Innovation: A Diffusion Study . Indianapolis: Bobbs-Merrill.

—. 1988. "Social Capital in the Creation of Human Capital." American Journal of Sociology 94.

Coleman, J.S. 1990. Foundations of Social Theory. Cambridge, MA: Harvard University Press.

Standard and Poor's. 2009. Compustat.

Cook, Karen S., Richard M. Emerson, and Mary R. Gillmore. 1983. "The Distribution of Power in Exchange Networks: Theory and Experimental Results." American Journal of Sociology 89:275-305.

Cook, Karen S. and Richard M. Emerson. 1978. "Power, Equity and Commitment in Exchange Networks." American Sociological Review 43:721-739.

Cruickshank, Paul and M. H. Ali. 2007. “Abu Musab Al Suri: Architect of the New Al Qaeda Studies.” Conflict & Terrorism 30:1–14.

Cruver, Brian. 2003. Enron: Anatomy of Greed . London: Arrow Books.

Dalton, Melville. 1959. "Relations between Staff and Line." in Men who Manage : John Wiley.

142

Damanpour, F. and W. M Evan. 1984. "Organizational Innovation and Performance: The Problem of Organizational Lag." Administrative Science Quarterly 29:392-409.

Damanpour, Fariborz. 1991. "Organizational Innovation: A Meta-Analysis of Effects of Determinants and Moderators." The Academy of Management Journal 34:555-590.

Dezso, Zoltan and Albert-Laszlo Barabasi. 2002. "Halting Virus in Scale-Free Networks." Physical Review .

Doreian, Patrick. 1981. "Estimating Linear Models with Spatial Distributed Data." Pp. 359- 388 in Sociological Methodology , edited by S. Leinhardt. San Francisco: Jossey-Bass.

Downs, G. W. and L. B. Mohr. 1976. "Conceptual Issues in the Study of Innovation." Administrative Science Quarterly 21:700-714.

Dozier, Janelle Brinker and Marcia P. Miceli. 1985. "Potential Predictors of Whistle-Blowing: A Prosocial Behavior Perspective." The Academy of Management Review 10:823-836.

DiMaggio, Paul J. and Walter W. Powell. 1983. "The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields." American Sociological Review 48:147-160.

Durkheim, Emile. 1976. Elementary Forms of Religious Life . London: George Allen & Unwin.

Emerson, Richard. 1976. “Social Exchange Theory.” Annual Review of Sociology 2:335-362.

Farrell , Dan and James C. Petersen. 1982. “Patterns of Political Behavior in Organizations.” The Academy of Management Review 7: 403-412.

Federal Energy and Regulatory Commission. “Information Released in Enron Investigation.” hosted by Lockheed Martin Corporation (formerly Aspen Corporation) outside of FERC’s website.

Festinger, L. 1950. "Informal Social Communications." Psychology Review 57:271-282.

Festinger, Leon. 1954. "A Theory of Social Comparison." Human Relations 7:117-140.

Festinger, Leon, Stanley Schachter, and Kurt W. Beck. 1950. Social Pressures in Informal Groups . Stanford, CA: Stanford University Press.

Fox, Loren. 2002. Enron: The Rise and Fall . New Jersey: John Wiley & Sons, Inc.

Freeman, Linton C. 1977. "A Set of Measures of Centrality based on Betweenness." Sociometry 40.

143

Friedkin, Noah. 1982. “Information Flow through Strong and Weak Ties in Intraorganizational Social Networks.” Social Networks 3:273-285.

—.1983. “Horizons of Observability and Limits of Informal Control in Organizations.” Social Forces 62:54-77.

—. 1990. "Social Networks in Structural Equation Models." Social Psychology Quarterly 534:316-328.

Friedkin, Noah and K. S. Cook. 1990. "Peer Group Influence." Sociological Methods and Research 19:122-143.

Friedkin, Noah and E. C. Johnsen. 1990. "Social Influence and opinion." Journal of Mathematical Sociology 15:193-205.

Galaskiewicz, Joseph and Ronald S. Burt. 1991. "Interorganization Contagion in Corporate Philanthropy." Administrative Science Quarterly 36:88-105.

Galling, Walter, Roberto Suro/Rome, Barry Kalb/New York, Walter Galling, Roberto Suro/Rome, and Barry Kalb/New York. 1984. "The Sicilian Connection." in Time Magazine.

Gibbons, Deborah E. 2004. "Friendship and Advice Networks in the Context of Changing Professional Values." Administrative Science Quarterly 49:238-262

Goffman, Erving. 1963. Information Control and Personal Identity . New York, NY: Prentice- Hall, Inc.

—. 1970. Strategic Interaction . New York, NY: Blackwell Publishers

Granovetter, Mark. 1973. "The Strength of Weak Ties." American Journal of Sociology 68:1360-1380.

—. 1985. "Economic Action and Social Structure: The Problem of Embeddedness." American Journal of Sociology 91:481-510.

—. 1992. "Problems of explanation in economic sociology." Pp. 25-56 in Networks and organizations: Structure, form, and action , edited by N. N. R. G. Eccles. Boston: Harvard Business School Press.

—. Forthcoming. Society and Economy: The Social Construction of Economic Institutions. Cambridge, MA: Harvard University Press.

Greenberger et al. 1987. “Oppositionists and Group Norms: The reciprocal Influence of Whistle-Blowers and Co-workers.” Journal of Business Ethics 6: 527-542.

144

Guetzkow, H. and H. A. Simon. 1955. "The Impact of Certain Communication Nets upon Organization and Performance in Task-oriented Groups." Management Science 1:233- 250.

Gulati, Ranjay. 1998. "Alliances and Networks." Strategic Management Journal 19:293-317.

Gundlach, Michael J., Scott C. Douglas, and Mark J. Martinko. 2003. "The Decision to Blow the Whistle: A Social Information Processing Framework " The Academy of Management Review 28:107-123.

Hage, J.T. 1999. “Organizational Innovation and Organizational Change.” Annual Review of Sociology 25:597-622.

Hardin, Russell and Karen S. Cook. 2001. "Norms of Cooperativeness and Networks of Trust." Pp. 327-47 in Social Norms , edited by M. H. a. K.-D. Opp. New York: Russell Sage Foundation.

Have, Robert van der and Marja Toivonen. 2007. "The Applicability of Schumpeter's Innovation 'Cases' in the Classification of Service Innovations." in XVIIth International Conference on RESER. Tampere, Finland.

Homans, George. 1950. The Human Group . New York: Harcourt Brace and Company.

—. 1974. Social Behaviour: Its Elementary Forms . New York: Harcourt Brace Jovanovich, Inc.

Hsiao, C. 1986. "Analysis of Panel Data." in Econometric Society Monographs, vol. 11 . New York: Cambridge University Press.

Ibarra, Herminia. 1993. "Network Centrality, Power, and Innovation Involvement: Determinants of Technical and Administrative Roles." The Academy of Management Journal 36:471-501.

Ibarra, Herminia and Steven B. Andrews. 1993. "Power, Social Influence, and Sense Making: Effects of Network Centrality and Proximity on Employee Perceptions." Administrative Science Quarterly 38:277-303.

Kadushin, Charles. 1966. "The Friends and Supporters of Psychotherapy: On Social Circles in Urban Life." American Sociological Review 31:786-802.

Kanter, Rosabeth. 1977. Men and Women of the Corporation . New York: BasicBooks.

Kaufer, D. and K. Carley. 1993. Communication at a Distance: The Influence of Print on Sociocultural Organization and Change . Hillsdale, NJ: Lawrence Erlbaum Associates. Inc.

145

Kimberly, J. R. and M. Evanisko. 1981. "Organizational Innovation: The Influence of Individual, Organizational, and Contextual Factors on Hospital Adoption of Technological and Administrative Innovations." Academy of Management Journal 24:689-713.

Kinsella, D. 2008. “The Black Market in Small Arms: Examining a Social Network.” Contemporary Security Policy 27:100–117.

Knight, K. E. 1967. "A Descriptive Mode of the Intra-Firm Innovation Process." Journal of Business 40:478-496.

Knoke, David and Song Yang. 2008. Social Network Analysis . Thousand Oaks: Sage Publications, Inc.

Kogut, Bruce and Udo Zander. 1992. "Knowledge of the Firm, Combinative Capabilities, and the Replication of Technology." Organization Science 3:383-397.

Krackhardt, David. 1988. "Informal Networks and Organizational Crises: An Experimental Simulation." Social Psychology Quarterly 51:123-140.

—. 1994. "Graph Theoretical Dimensions of Informal Organizations." Pp. 89-111 in Computational Organizational Theory , edited by K. Carley and M. Prietula. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Krackhardt, David and Robert Stern. 1988. "Informal Networks and Organizational Crises: An Experimental Simulation." Social Psychology Quarterly 51:123-140.

Laumann, E.O. 1979. “Network Analysis in Large Social Systems: Some Theoretical and Methodological Problems.” Pp. 379-402 in Perspectives on Social Network Analysis, edited by Paul W. Holland and Samuel Leinhardt New York: Academic Press.

Leavitt, H. 1951. “Some Effects of Certain Communication Patterns on Group Performance.” Journal of Abnormal and Social Psychology 46:38-50.

Lee, Nancy Howell. 1969. The Search for an Abortionist . Chicago: University of Chicago Press.

Lindelauf, Roy, Peter Borm, and Herbert Hamers. 2009. "The Influence of Secrecy on the Communication Structure of Covert Networks." Social Networks 31:126-137.

146

Magouirk, Justin, Scott Atran, and Marc Sageman. 2008. “Connecting Terrorist Networks.” Studies in Conflict & Terrorism 31:1-16.

March , J. G. 1955. “An Introduction to the Theory and Measurement of Influence .” American Political Science Review 49: 431–451.

Markovsky, Barry, David Willer, and Travis Patton. 1988. "Power Relations in Exchange Networks." American Sociological Review 53.

Marsden, Peter V. 1987. “Core Discussion Networks of Americans.” American Sociological Review 52: 122-131.

—. 1988. "Homogeneity in Confiding Relations." Social Networks 101:57-76.

Marsden, Peter V. and Noah Friedkin. 1993. "Network Studies of Social Influence." Sociological Methods and Research 22:127-51.

Marsden, Peter V. and Joel M. Podolny. 1990. "Dynamic Analysis of Network Diffusion Processes." in Social Networks through Time , edited by J. Weesie and H. Flap. Utrecht: University of Utrecht, ISOR.

McAllister, B. 2004. “Al Qaeda and the Innovative Firm: Demythologizing the Network”, Studies in Conflict & Terrorism 27:297–319

McPherson, J. M. and L. Smith-Lovin. 1987. "Homophily in Voluntary Organizations: Status Distance and the Composition of Face to Face Groups." American Sociological Review 52:370-379.

McPherson, J. Miller and Lynn Smith-Lovin. 1982. "Women and Weak Ties: Differences by Sex in the Size of Voluntary Organizations." The American Journal of Sociology 87:883-904.

McLean, Bethany. 2001. “Is Enron Overpriced.” in Fortune Magazine .

McLean, Bethany and Peter Elkind. 2003. Smartest Guys in the Room: The Amazing Rise and Scandalous Fall of Enron : Portfolio Hardcover.

Merton, R. K. 1968. Social Theory and Social Structure . New York: The Free Press.

Miceli, Marcia P. and Janet P. Near. 1984. The Relationships among Beliefs, Organizational Position, and Whistle-Blowing Status: A Discriminant Analysis. The Academy of Management Journal 27:687-705.

Mishal, S. and M. Rosenthal. 2005. “Al Qaeda as a Dune Organization: Toward a Typology of Islamic Terrorist Organizations.” Studies in Conflict & Terrorism 28: 275–293.

147

Mitchell, Lawrence. 2003. "Structural Holes, CEOs, and the Missing Link in Corporate Governance." GWU Law School, Public Law Research Paper No. 77.

Monge, Peter R. and Noshir S. Contractor. 2003. Theories of Communication Networks. New York: Oxford University Press.

Moody, James, Martina Morris, Jimi Adams, and Mark Handcock. 2003. "Epidemic Potential in Low Degree Networks." Manuscript. Ohio State University.

Near, Janet P. and Marcia P. Miceli, 1985. "Organizational Dissidence: the Case of Whistle- blowing." Journal of Business Ethics 4:1-16.

Near, Janet P., Terry Morehead Dworkin, and Marcia P. Miceli. 1993. "Explaining the Whistle-Blowing Process: Suggestions from Power Theory and Justice Theory." Organization Science Focused Issue: The Legalistic Organization 4:393-411.

Obstfeld, David. 2005. "Social Networks the Tertius lungens Orientation, and Involvement in Innovation." Administrative Science Quarterly 50:100-130.

Pattison, P.E., S. Wasserman, G.L. Robins, and A. Kanfer. 2000. "Statistical Evaluation of Algebraic Constraints for Social Networks." Journal of Mathematical Psychology 44:536-568.

Podolny, Joel M. and James N. Baron. 1997. "Resources and Relationships: Social Networks and Mobility in the Workplace." American Sociological Review 62:673-693.

Powell, Walter W. 1996. "Fields of Practice: Connections between Law and Organizations." Law & Society Inquiry 21:959-966.

Rai, Saritha. 2001. "Enron's Dispute with Utility In India Grows More Tangled." in New York Times .

Reingold, Jennifer. 2003. "The Women of Enron : The Best Revenge." in Fast Company .

Robins, Garry and Malcolm Alexander. 2004. "Small Worlds Among Interlocking Directors: Network Structure and Distance in Bipartite Graphs." Computational & Mathematical Organization Theory 10:69-94.

Rogers, Everett M. 1962. Diffusion of Innovations . New York, NY: Free Press.

Sacerdote, Bruce and David Marmaros. 2005 “How Do Friendships Form?” NBER Working Paper Series.

Salancik, Gerald R. and Jeffrey Pfeffer. 1978a. "A Social Information Processing Approach to Job Attitudes and Task Design." Administrative Science Quarterly 23:224-253.

148

—. 1978b. "Uncertainty, Secrecy, and the Choice of Similar Others." Social Psychology 41:246-255.

Schumpeter, J. A. 1934. The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle . Cambridge, MA: Harvard University Press.

Security and Exchange Commission. 1934. Exchange Act.

Security and Exchange Commission. Litigation Release No. 18335, AAER Release No. 1852, Sep. 10, 2003. < http://www.sec.gov/spotlight/enron.htm>

Security and Exchange Commission. Litigation Release No. 17692, AAER Release No. 1617, Aug. 21, 2002. < http://www.sec.gov/spotlight/enron.htm>

Senneker, P., and C. Hendrick. 1983. “Androgyny and Helping Behavior.” Journal of Personality and Social Psychology. 45:916–925.

Shetty, Jitesh and Jafar Adibi. 2008. "Enron Dataset."

Simmel, Georg. 1950. The Sociology of Georg Simmel . Translated by K. H. Wolff. New York: Free Press.

—. 1955. Conflict and the Web of Group Affiliations . New York: Free Press.

Simon, Herbert A. 1957. Administrative Behavior . New York, NY: Macmillan.

—. 1965. "The Architecture of Complexity." Pp. 63-76 in General Systems: Yearbook of the Society for General Systems , vol. 10, edited by L. von Bertalanffy and A. Rapaport.

—. 1981. The Sciences of the Artificial. Cambridge (Second Edition). MA: MIT Press.

Smith-Lovin, Lynn and Miller McPherson. 1993. "You Are Who You Know: A Network Approach to Gender." Pp. 223-251 in Theory on Gender, Feminism on Theory , edited by P. England. New York Aldine de Gruyter.

Sparrow, M. 1991. “The Application of Network Analysis to Criminal Intelligence: an Assessment of the Prospects.” Social Networks 13:251–274

Stoler, Peter, Walter Galling, Roberto Suro (Rome), and Barry Kalb (New York) 1984. "The Sicilian Connection." in Time Magazine , October 15, 1984.

Suchman, Mark C. 1995. "Managing Legitimacy: Strategic and Institutional Approaches." Academy of Management Review 20:571-610.

149

Swartz, Mimi and Sherron Watkins. 2004. Power Failure: The Inside Story of the Collapse of Enron . New York: Doubleday Business.

Tushman, Michael L. 1977. "Special Boundary Roles in the Innovation Process." Administrative Science Quarterly 22:587-605.

Umphress, Elizabeth Eve, Giuseppe (Joe) Labianca, Daniel J. Brass, Edward (Eli) Kass, and Lotte Scholten. 2003. "The Role of Instrumental and Expressive Social Ties in Employees' Perceptions of Organizational Justice." Organization Science 14.

Valente, Thomas W. 1995. Network Models of the Diffusion of Innovation . Creskill, NJ: Hampton Press.

—. 2005. "Models and Methods for Innovation Diffusion." in Models and Methods in Social Network Analysis , edited by P. J. Carrington, J. Scott, and S. Wasserman. Cambridge, UK: Cambridge University Press.

Volker, B., and H. Flap. 2001. “Weak Ties as a Liability: the Case of East Germany.” Rationality and Society 13:397–428.

Wasserman, S. and J. Galaskiewicz. 1994. Advances in Social Network Analysis . Thousand Oaks: Sage Publications.

Wasserman, Stanley and Katherine Faust. 1994. Social Network Analysis: Methods and applications. New York: Cambridge University Press.

150

151