. w)‘. ,. .-.:- W 2‘ ; tab-‘- M vii ”figqlnx 1 ',;.i..’ ‘ I ' .- . n O ; v .41} v v ‘5 (j .' ' 0 'Q'ur'ra 1‘ .... 1 ~n§‘.."" ‘1,.\ pm... n. .C“ ‘ .,._._.. ~15 .1“ ‘. w («3" «r "I x-u- I‘murn :- V v- - fr in.“ vv—w QGX\ gfléfl‘il‘? This is to certify that the dissertation entitled RESPONSES TO OCCUPATIONAL FRAUD: A STUDY IN THE BEHAVIOR OF LAW presented by KRISTY HOLTFRETER has been accepted towards fulfillment of the requirements for the PhD. degree in Criminal Justice / Major Prqifessor’s Signature January 9, 2004 Date MSU is an Affirmative Action/Equal Opportunity Institution ’W Michigan State University PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE 6/01 c:/ClFtC/DateDue.p65-p.15 RESPONSES TO OCCUPATIONAL FRAUD: A STUDY IN THE BEHAVIOR OF LAW BY Kristy Holtfreter A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY School of Criminal Justice 2004 ABSTRACT RESPONSES TO OCCUPATIONAL FRAUD: A STUDY IN THE BEHAVIOR OF LAW By Kristy Holtfreter The research was designed to improve understanding of the individual and organizational characteristics connected to different types of occupational fraud, and to explain the decisions of victim organizations to refer fraud cases for civil and criminal processing. .A key hypothesis was that the characteristics of individuals who commit asset misappropriation, corruption, and fraudulent statements will differ; a related hypothesis was that the characteristics of organizations victimized. by’ asset misappropriation, corruption, and fraudulent statements will differ. To examine the likelihood of criminal and civil referrals, additional hypotheses based on Donald Black's (1976) theory and conflict theory were examined. The analyses included secondary data from surveys of 1,142 Certified Fraud Examiners at two time periods: 1997- 98 and 2001-02. Two sets of analyses were performed. The first analysis compared individual and organizational characteristics for cases of asset misappropriation, corruption, and fraudulent statements. The second analysis examined the likelihood of receiving a criminal or civil referral, compared to no referral, given a series of independent variables suggested fur Black’s (1976) theory, conflict theory, and with the addition of control variables suggested by critiques of Black's theory. The results for the first analysis support the hypotheses that several individual characteristics idiffer for the types of fraud, as do several organizational characteristics. Some individual characteristics cflf the sample, namely the average age of offenders, are similar to previous findings in the white-collar crime literature; and in. particular, one type of fraud, corruption, was more likely in organizations that were similar to the corporate crime literature’s portrayal of setting for white—collar crime. The results for the second analysis produced mixed support for Enack's theory and conflict theory. .At both time periods, the best fitting models were those that included control variables, supporting prior critiques of Black’s theory. Differences 1J1 the frequencies of referrals between time periods suggested that attitudes toward punishment of fraud offenders have become more punitive following recent corporate scandals. Copyright by KRISTY HOLTFRETER 2 0 0 4 Dedicated to my parents, Bob and Judy Holtfreter, whose many accomplishments inspired my own achievements. At an early age, you taught me the value of higher education, and the numerous advantages of University life and community. ACKNOWLEDGEMENTS First and foremost, I would like to thank Michael Dean Reisig for his love, friendship, humor, and advice. You are my career role model, and I could not have undertaken the vast responsibilities of teaching and research without your ongoing guidance and support. I am forever appreciative of the generous instrumental support and advice provided by my dissertation chair and mentor, Merry Morash. Thank you for your ongoing encouragement. I am also indebted to Joe Wells and the Association of Certified Fraud. Examiners, who made this project possible by generously providing their data. The assistance of John Warren, Associate General Counsel, significantly contributed to the success of my research. Thank you to my dissertation committee members Steven Dow, Mahesh Nalla, and Steve Kozlowski for their unique perspectives and critical insights on my work. I would also like to thank my office mate and friend Yan Zhang for her brilliant statistical advice. I appreciate the cmgoing emotional support and humor provided by my siblings Kerry, Kelby and Rob, throughout my vi time in graduate school. You are all geniuses! Finally, a hearty meow goes out to Pippi, Rex, and Rusty for your unconditional love and much needed distractions during the course of this project. vii TABLE OF CONTENTS Page List of Tables xi List of Figures xii Chapter 1 INTRODUCTION 1 White-Collar Crime and Organizations 8 The Behavior of Law and Conflict Theory 11 Differential Application of Law 13 Research Objectives 17 Organization of Dissertation 19 Chapter 2 LITERATURE REVIEW 90 The Origins of White-Collar Crime Research 22 Sutherland's Legacy: Early Studies 99 White-Collar Crime Today: Developments in Theory and Research 36 Organizational Crime 29 Conceptual Issues: Offender and Offense-Based Definitions 42 Individual and Organizational Characteristics Related to Types of Occupational Fraud - 46 Individual Characteristics 47 Organizational Characteristics 59 An Integrated Framework: The Behavior of Law and Conflict Theory 67 Up-and-Down Integration 67 The Behavior of Law 68 Conflict Theory 70 Previous Research: The Behavior of Law and Conflict Theory 72 Summary 88 viii Chapter 3 DATA AND METHODS 95 Data 97 Sample 98 ACFE Survey One Procedures 98 ACFE Survey Two Procedures 100 Study Variables 101 Analysis 1: Individual and Organizational Characteristics for Types of Occupational Fraudmmmmmmm102 Variables 102 Individual Characteristics 102 Organizational Characteristics 103 Analysis 2: The Behavior of Law 107 Dependent Variables 107 Independent Variables 107 Control and Supplementary Variables 113 Data Analyses 113 Descriptive Statistics 114 Bivariate Associations 114 Analysis of Variance 114 Overview of Maximum Likelihood Techniques ......... 117 Logistic Regression 118 Multinomial Logistic Regression 126 Conceptual Models 127 Data Considerations 129 Limitations 129 Generalizability 129 Sample Selection Bias 130 Measurement 132 Secondary Data Analysis 133 Strengths 134 Policy Implications 136 Practice Implications 137 Conclusion 138 Chapter 4 RESULTS 140 Analysis 1: Individual and Organizational Characteristics Related to Types of Occupational Fraud 140 Descriptive Statistics 140 Bivariate Relationships 145 Analysis of Variance 151 Individual Characteristics 148 Organizational Characteristics 157 ix Analysis 2: Behavior of Law and Conflict Theorymmmmmmm162 Descriptive Statistics 162 Bivariate Relationships 166 Multivariate Analyses 170 The Behavior of Law: Multinomial Logistic Regression Models 173 Multinomial Logistic Regression Models: Separate Dimensions of Social Life 175 Separate Dimensions of Social Life, Criminal vs. Civil 183 Theoretical Models and Controls: Full Sample 191 Theoretical Models and Controls: 1997-98 Sample 196 Theoretical Models and Controls: 2001—02 Sample 203 Conclusion 210 Chapter 5 DISCUSSION 215 Study Purpose 215 Analysis 1: Key Findings 716 Analysis 2: Key Findings 219 Potential Limitations 722 Implications for Theory 724 Implications for Policy and Practice 232 REFERENCES 735 Table Table Table Table Table Table Table Table Table Table Table Table Table Table 10: 11: 12: 13: 14: LIST OF TABLES Description of Variables for Comparing Types of Fraud Description of Variables, Behavior of Law ....... 106 112 Descriptive Statistics, Analysis One Correlation Matrix of Individual and Organizational Characteristics 144 150 Occupational Fraud Type by Individual Characteristics 156 Occupational Fraud Type by Organizational Characteristics Descriptive Statistics, Dimensions of Social Life Descriptive Statistics, Behavior of Law 161 164 Correlation Matrix of Dimensions of Social Life and Behavior of Law Multinomial Logistic Regression Models of Behavior of Law for Separate Dimensions of Social Life Multinomial Logistic Regression Models of Behavior of Law for Separate Dimensions of Social Life, Criminal vs. Civil Multinomial Logistic Regression Models for Behavior of Law, Full Sample 166 169 182 190 195 Multinomial Logistic Regression Models for Behavior of Law, 1997—98 Multinomial Logistic Regression Models for Behavior of Law, 2001—02 202 709 xi LIST OF FIGURES FIGURE 1: Conceptual Model Linking Individual and Organizational Characteristics to Types of Fraud 127 Figure 2: Conceptual Model Linking Dimensions of Social Life to Behavior of Law 129 xii CHAPTER ONE : INTRODUCTI ON The primary goal of the research undertaken in this dissertation is to improve understanding of the multi-level characteristics (individual and organizational) connected to different types of occupational fraud (i.e., asset misappropriation, corruption, and fraudulent statements) in organizations. A second goal is to describe and explain the legal responses (e.g., civil referral, criminal referral, or no referral) to occupational fraud in organizations by immegrating Black’s (1976) theory of the behavior of law and conflict theory. The study includes organizations that differ in size (i.e., ranging from less than 100 to more than 10,000 employees) and function (i.e., government agencies, non-profit organizations, private corporations, and publicly traded companies). By focusing (n1 fraud victimization 1J1 organizations, this study addresses prior research neglect of this type of behavior. Recognition of the need to study fraud and other acts described as “white—collar crime" can be traced back to Sutherland (1949; 1940), whose original interest in the topic was rooted in addressing biases associated with official crime statistics, which he rightly argued did not capture the offenses committed by high status, respectable individuals in the course of their legitimate occupations. Sutherland proposed that existing criminological theories emphasizing structural conditions such as poverty (e.g., social disorganization theory) were incapable of explaining the crimes of the upper class. The topic of white-collar crime was and still is quite novel in that it seems counterintuitive that accomplished individuals who are otherwise law-abiding would commit criminal offenses in the course of their occupational roles. While factors such as poverty clearly provide plausible motives for the crimes of the unemployed, it is more difficult to conceive of reasons for crimes committed by those holding legitimate occupations. Individual greed may be a potential motive for many crimes. As one fictional corporate raider matter- of—factly stated: “Greed, for lack of a better word, is good. Greed is right. Greed works.” --—Gordon Gecko, in the film wall Street (1987). Although individual greed undoubtedly contributes to some cases of occupational fraud, it is likely that the motivations and explanations for such behavior are much more complex, and differ based on the type of fraud that is committed. Inferring causality about the individual’s decision to commit fraud is beyond the scope of the present study. However, it is possible to determine whether the various characteristics of individuals differ for distinct types of fraud victimization. Researchers’ abilities to measure crime have improved dramatically since Sutherland’s time, and recent statistics indicate that offenses collectively referred to as “white- collar crime” result in annual losses to victims in excess of $250 billion, compared to estimates of $17.6 billion for losses accrued due to personal and household crimes (Rosoff, Pontell, and Tillman, 2002). While these differences have long persisted, the primary focus of criminal justice authorities and. many criminologists has been on explaining, preventing and responding to personal and household crimes. Public perception and fear of violent “crime in the streets” has also fueled the attention of researchers and criminal justice agencies. Recently, however, current and ongoing scandals in various corporations, such as Enron and WorldCom, have raised public awareness of the economic, social, and personal harm resulting from “crime in the suites.” The literature on white-collar crime has grown considerably since Sutherland's initial work, but this body of research is quite fragmented due to disagreements over conceptual and operational definitions of relevant variables. For example, one tradition argues that researchers should adhere to a variant of Sutherland’s original “offender—based” definition by focusing on characteristics of .Lndividual jperpetrators (e.g., status, gender, age) believed to promote white-collar crime. The opposing viewpoint emphasizes the importance of employing an “offense—based" definition, geared toward inclusion of cases based on characteristics of the acts themselves (e.g., typically, crimes lconsidered to violate trust). More recent research suggests that characteristics of both offenders and offenses should be included in an operational definition of white—collar crime. The aforementioned perspectives are reviewed and critiqued in subsequent chapters of this dissertation, which demonstrates the benefits of using an offense-based definition for purposes of the present study. In addition to being plagued by definitional disagreements, the .literature CH1 white—collar' crime .also suffers from a lack of attention to victims. Research in the general area of criminal victimization has increased dramatically in recent years (see Davis, Lurigio, and Skogan, 1997), resulting in improved. understanding about victims of violent crime. Moreover, heightened policy interest in crime victims has resulted in the expansion of victim assistance programs and services. Yet, despite the research and policy attention to victims of violent crime, studies of white-collar crime, with the exception of fraud in the savings and loan industry, have neglected victims. Why have victims of white—collar crime been ignored? Perhaps the answer to this question lies in the ongoing definitional issues, or the neglect may be due to the limited access t1) reliable data available tx; white—collar crime researchers. A. more likely reason, discussed. in additional chapters of this dissertation, relates to the complex details of the offenses themselves. Put simply, unlike victims of violent crimes, victims of white—collar offenses ‘may' not even. realize they' have been ‘victimized (Benson and Moore, 1992). This issue also presents difficulties when it comes to white-collar crime prosecution. As Jesilow, Klempner' and. Chiao (1992:150) explained, “...the hidden nature of many white—collar misdeeds prevents victims from uncovering the offenses and entering the complaint process.” By focusing on known victims of white-collar crime, the present study contributes to an area of research that has been largely neglected. As Vaughan (1980:96) observed, “traditional victimology has inhibited understanding of crimes against organizations, and should be broadened to include the study of organizations as victims.” Despite the fact that this statement was made nearly' a quarter of a century ago, relatively little research on organizations as victims has been conducted. Within the white-collar crime framework, the present study focuses specifically on occupational fraud that victimizes organizations. According to Wells (1997:4) “fraud can encompass any crime for gain which uses deception as its principle modus operandi.” It is important to note, however, that all deceptions are not frauds. Under common law, some form of damage (e.g., monetary) must occur to meet the legal definition of fraud. Common law also specifies four elements that must be present for fraud to exist: (1) a material false statement, (2) knowledge that the statement was false when. it was made, (3) reliance on the false statement by the victim, and (4) damages as a result. The legal definition is the same regardless of whether the offense is civil or criminal; the difference is that criminal cases must withstand a greater burden of proof (Wells, 1997). The following definition of occupational fraud is used in this dissertation: “The use of one’s occupation for personal enrichment through the deliberate misuse or misapplication of the employing organization’s resources or assets” (Association of Certified Fraud Examiners, 2002: 2).1 This offense-based definition clearly identifies organizations as victims, encompasses a wide range of misconduct by employees, managers, and executives, and acknowledges that occupational fraud schemes can be as simple as pilferage of company supplies or as complex as financial statement fraud” The types of fraud very, and will be described further in subsequent chapters. Although variation exists, each case of occupational fraud has four elements in common: (1) it is clandestine; (2) it violates 1 This is an operational definition used to classify cases, not a legal definition of “fraud”. Cases that are identified as “fraud” in the present study may or may not be defined as crimes and/or torts according to legal definitions. For additional information on legal definitions of fraud, see Henry Campbell Black, 1979. 7 the perpetrator's fiduciary duties to the victim organization; (3) it is committed for the purpose of direct or indirect financial benefit tx: the perpetrator; and (4) it costs the employing organization assets, revenues, or reserves. As Shapiro (1990) argues, much conceptual ambiguity can be avoided if researchers work to “collar the crime,” by focusing on specific offenses, rather than the characteristics of offenders. The use of an offense-based definition eliminates conceptual ambiguity and allows for an examination of the connection of individual and organizational characteristics to different types of occupational fraud. White-Collar Crime and Organizations Despite the criminological debate surrounding the operationalization and definition of white-collar crime, there is general agreement in the field that many offenses are committed in cmganizations, either by individuals for personal gain, or by individuals or groups on behalf of their employers. Given this acknowledgement, researchers studying crime in organizations have emphasized the need to consider individual-level and organizational-level influences on the type of crime that occurs. The resulting sociological conceptualization of “organizational crime” stresses the importance of structural factors in determining individual behavior, and has been recognized as the paradigm by which white-collar crime theory is most likeLy to advance (Biderman and Reiss, 1968; Braithwaite, 2001; Reiss and Tonry, 1993). It is therefore crucial that white—collar crime studies be informed by research on organizations. It large amount of research on crime in organizations has been conducted outside the field of criminology. Integrating studies from related areas (e.g., business ethics, organizational psychology) will greatly improve criminologists’ understanding of white-collar crime. The focus in the present study bridges the micro- macro connection in white-collar crime theory by including individual-level and organizational-level variables. The recognized value of multi-level influences notwithstanding, the gap in research Ion crime in organizations is still considerable. A crucial issue in the literature is the fact that most studies that have been conducted. on.‘white-collar' crime .have tended. to focus on organizations as perpetrators, but not victims, of fraud. As noted previously, this limitation is due in part to the difficulty of obtaining information (e.g., the lack of available data and sampling problems). The pmesent study addresses this void in the literature by examining organizations that have been victimized by individual employees. Much of the extant literature on crime in organizations also suffers from what is known as the business firm bias, which is characterized by a research emphasis on large, profit—making organizations (Vaughan, 1992). The stereotype of large firms in financial trouble committing offenses prevails in the literature, and the related. amoral calculator" model suggests that executives are driven by profits and thus make rational decisions to commit crimes as a means of maximizing financial gain (Kagan and Scholz, 1984). While this perspective indeed seems plausible when applied to large, profit—making organizations, it is less applicable in other settings, such as government, non-profit, or private agencies. The tendency to focus on large business firms as offenders has resulted in little knowledge of how organizations varying in size and function are victimized (Reiss and Tonry, 1993). The present study partially corrects this bias by including victim organizations that differ in size as well 10 as function. Examining these types of organizational variation makes it possible to address the still unanswered question of which kinds of organizations generate greater amounts and different types of white-collar crime. The Behavior of Law and Conflict Theory In his study of employee theft, Robin (1970:137) stated: “Perhaps more than. any’ other civil individual, the employer or his corporate representatives are in a position where they must decide whether or not to report a known, apprehended offender to the law enforcement officials, a responsibility and power which has been largely ignored in the sociology of law.” The organization's decision whether or not to refer individual employees’ crimes for formal legal processing remains neglected by criminologists, in spite of the fact that Donald Black’s (1976) theory (M? the behavior of law provides a valuable framework for explaining such decisions. In addition to its potential for explaining the 11 initial decision to refer an occupational fraud case, Black’s theory can also be applied to subsequent decisions in the legal process (e.g., the range of punishment and sanction options). The theory states that law, like other institutions, “behaves.” In other words, it acts in certain patterns, and is not random in its existence or application. Black outlines an explanatory model that can be used to analyze the behavior of law by its relation to five dimensions of social life: stratification, morphology, culture, organization, and. social contrcfl” .According ‘to Black, law is one form of social control, governmental social control. The theory states that law is a quantitative variable that increases or decreases across social settings. Black also describes law as a qualitative variable, arguing that it varies by style and includes four styles: penal, compensatory, therapeutic and conciliatory. Black’s theory of the behavior of law has received a great deal of attention in the field of criminology, and has been subject to many empirical tests, the results of which can generally be described as providing mixed support“ Some studies have fully' supported, the theory, some have supported a limited number of its propositions, and others have contradicted the theory. Previous tests of 12 Black’s theory are discussed in more detail in Chapter Two, Literature Review. Conflict theory also provides an explanation for the legal responses to crime (Vold, Bernard, and Snipes, 1998). According to this perspective, law is not representative of the general public's interests, but is based on the values of powerful groups. Conflict theory suggests that those individuals with less power would be more likely to be referred for criminal processing. In criminology, conflict theory has been applied in a number of different ways. For example, it has been used to explain crime itself as a reaction by minority power groups to the dominant social order, and also to explain the passing of legislation as reflective of powerful groups’ values. Differential Application of Law Several studies not directly’ testing' Black’s theory have addressed the issue of differential application of law for white-collar offenders. These studies indirectly support the theory' of the behavior of law, as well as conflict theory. Past research focused specifically on white-collar crime has examined the effect of socioeconomic status (SES) on choice of sanctions. Work by Wheeler, 13 Weisburd, and Bode (1982) for example, examined the effect of social class on sentencing for eight federal offenses designated as “white-collar crime." Results of this study indicated that there were no significant class differences in sentencing, which counters findings from research on differential application of law based on offender status (Jesilow, Pontell, and Geis, 1986). The work of Wheeler and colleagues is limited by the fact that their sample of defendants appearing in federal court may not represent the larger distribution of white- collar offenders; namely, those whose status and position may prevent criminal or civil sanctioning in the first place. Given data limitations (i.e., primarily samples obtained from court records), and the contradictory findings of existing studies, a closer examination of the relationship of perpetrator and organizational characteristics, to legal responses is warranted. Black’s (1976) theory suggests that a greater quantity of law (e.g., harsher sanctions) is expected for lower class and lower-ranking employees, but this proposition has not yet been tested. To address these issues, the present study will directly test Black’s theory. Availability of a unique data set that includes case outcomes (e.g., the 14 decision to refer a perpetrator for additional criminal or civil processing) makes it possible to increase understanding of differential application of sanctions. An additional limitation of previous research is the inclination toward treating violations of administrative regulations as violations of criminal law. Failure to distinguish. between these two targets of social control translates into treating all violators in organizations as “white-collar criminals” and all violations in organizations as “white-collar crime.” This tendency has resulted in a conflation of criminal laW’ and. administrative regulations, which. is problematic for the consideration of social control strategies. While regulatory agencies use compliance mechanisms, the criminal justice system tends to rely on principles of deterrence (Ayres and Braithwaite, 1992; Braithwaite, 1993). What is more, strategies vary by industry, suggesting that it is crucial to consider the type of organization when drawing conclusions about the justness and effectiveness of various approaches. The present study improves upon past research by clearly distinguishing between different responses (i.e., civil, criminal, and no referral). 15 The mere existence of white-collar crime in organizations reflects the failure of existing mechanisms of control and undermines public trust in governmental institutions assigned. to (enforce corporate behavior. Speaking about the scandals associated with energy giant Enron, Business week reporter Bruce Nussbaum recently wrote: “There are business scandals that are so vast and so penetrating that they profoundly shock our most deeply held beliefs about the honesty and integrity of our corporate culture" (Jan 28, 2002). Despite the enormous economic and societal losses due to such offenses, the most effective sanctioning strategies are yet to be discovered. New legislation is promising, but like rmnfli research, focuses (n1 profit—oriented firms and top executives. A.uniform response to fraud, such as the recent passing of the Public Company Accounting Reform and Investor Protection Act of 2002, targets large, publicly traded companies, but offers limited. provisions for smaller, non-public corporations or those operating in 16 alternative capacities. A comparison of existing sanctions is needed to suggest ways in which current policies should be adapted, or as Guebosky (1997) proposed, combined with other regulatory strategies. Research Objectives This dissertation will provide a theoretically informed understanding of the individual perpetrators and organizational processes that are associated with different types of occupational fraud, and of the legal responses to these offenses. After determining what types of occupational fraud are most prevalent, this study will discuss the effectiveness of existing procedures, such as employee background checks, anonymous reporting systems, internal and external audits) in preventing and detecting certain forms of fraud in different types of organizations. Specifically, this research reveals the types of fraud that occur regardless of the internal controls that are in place, the consistency of sanctions in response to particular types of offenses, and the degree to which sanctions are influenced by stratification, morphology, culture, organization, and social control (Black, 1976). 17 The present study contributes to theory, policy, and practice by accomplishing the following objectives: (1) providing a comprehensive description of perpetrators of occupational fraud who victimize organizations; (2) comparing offender characteristics (i.e., age, gender, education, position in organization, criminal history) and organizational characteristics (i.e., size, type, existing internal controls, revenue) for each type of fraud; (3) specifying and testing multinomial logistic regression models predicting the organization’s decision to refer fraud cases based on the five dimensions of social life identified In! Black; and. (4) describing fraud investigators’ perceptions of the response to specific types of detected fraud. Findings will be disseminated to criminal justice researchers and practitioners. Recommendations for future research and suggestions for preventative techniques and sanctioning approaches will be provided. No previous studies have simultaneously included ‘victim organizations of varying size and type, comprehensive perpetrator information, legal responses, perceptions of fraud investigators, and data from two time periods. Additionally, Black's theory has never been tested across multiple organizations. 18 Organization of Dissertation This dissertation is divided into five chapters. Chapter Two, “Literature Review,” presents the conceptual framework for this study, and provides an overview of previous research on individual characteristics, organizational characteristics, and. the behavior‘ of law. In this review, the strengths and limitations of the previous research are discussed at length. The chapter also demonstrates how the propositions of conflict theory can be incorporated within the theory of the behavior of law using up—and-down theoretical integration. Chapter Three, “Data and. Methods,” introduces the research propositions, describes the research design and procedures, and justifies operationalization of the independent and dependent 'variables. Chapter' Four, “Results," describes the major research findings from all analyses. Finally, Chapter Five, “Discussion,” revisits the issues introduced in the previous chapters, and examines the implications of this study for theory, policy, and practice. 19 CHAPTER TWO: LITERATURE REVIEW This literature review is arranged in the following manner: First, the origins of white-collar crime research, beginning with Edwin Sutherland's (1939) well-known introduction of the concept and his interest in differential implementation of law, is reviewed. Next, Sutherland’s legacy, including the applicability of his differential association theory tx: white-collar crime, is traced through the research of Cressey (1953); Geis (1992; 1967; 1962); Clinard and Yeager (1979) and Quinney (1964) among other scholars. Following this overview, more recent developments in white—collar crime research are presented. In this section, conceptual issues related to the distinction between offender-based and offense-based definitions are discussed. Next, an organizational framework is presented, to suggest how white-collar crime research should incorporate: individual. and. Black, (1976:17), stratification is 107 described as “the vertical aspect of social life,” and thus exists when the material things in life are unevenly distributed. Stratification is measured with the variables age and gender. Age is the perpetrator’s age at the time of the offense, and is measured in years. Participants’ gender is coded dichotomously (1 = male; 0 = female). It is expected that more law will be used against younger employees and against female employees. Morphology Morphology refers to the horizontal relationships in social life, and is described by Black (1976: 37) as “the distribution of people in relation to one another, including their division of labor.” Morphology is neasured by the perpetrator’s position in the organizational hierarchy, and is a dummy variable reflecting whether the perpetrator is an executive or manager (coded = 1); or an employee (coded = 0). It is expected that more law will be used against employees than managers or executives. Culture Black (1976:61) defines culture as “the symbolic aspect of social life, including expressions of what is 108 true, good, and beautiful.” Individuals’ culture reflects the extent of culture in their lives as evidenced by factors such as literacy and education. Those individuals with more culture (i.e., those possessing higher educational levels) are assumed to be more aware of their rights and options, and are therefore more likely to be successful in influencing legal outcomes (Avakame, Fyfe and McCoy, 1999). Consistent with. Black’s ideas, culture is measured by the perpetrator’s level of education, and is an ordinal variable reflecting tine perpetrator’s highest formal degree: 1. = high school diploma or less; 2 = bachelor’s degree; 3 = graduate degree. It is expected that more law will be used against less cultured (lower education)employees. Organization Organization, according to Black (1976:85), represents the capacity for collective action, and is defined as “the corporate aspect of social life.” With regard to corporations, those companies that have more employees or those that can be characterized as centralized, such as government agencies, would be considered more organized. Additionally, Black (1976:95) states that “organization 109 increases with size...the more members a group has, the more organized it is.” Black (1976: 97) further argues that “deviant behavior by an individual against an organization is the most serious.” Black also specifically addresses the organization of different types of businesses, and notes the theory of law predicts that a crime against a government agency (which is highly organized) would be considered more serious than a crime against a private business (which is less organized). Black notes that a crime against a “large supermarket” (more organized) is more serious than a crime against a “small grocery store” (less organized). Consistent with Black’s formulations, organization is measured by two variables: type of firm and size of firm. Type of firm indicates whether the organization is a government agency (= 4); publicly traded company (= 3); privately held company (= 2); or non-profit agency (= 1). Size of organization indicates the number of employees and is coded as an ordinal variable: 4 = 10,000 or more; 3 = 1000-9999; 2 = 100-999; 1 = 1-99. It is expected that more law will be used against employees who victimize centralized organizations (government) than less centralized organizations. It is expected that more law 110 (quantity and quality) will be used against employees who victimize large organizations than small organizations.. Social Control .According tx: Black. (1976), social. control ‘represents the normative aspect of social lifEu Several forms of social control exist, and law is merely one form (Black, 1976). Other forms include bureaucracy, custom, and ethics. The more often one form of social control is used, the less often other forms will be used, In the present study, social control is measured by the existing non-legal forms of social control that are available to organizations. Existing internal social control mechanisms (background checks (1 == yes; 0 == no); anonymous fraud reporting (1 = yes; 0 = no); internal audits (1 = yes; 0 = no) or external audits (1 = yes; 0 = no)) were summed to indicate the degree of existing social control, with the new variable ranging from 0-4. It is expected that more laW' will be used in organizations with fewer available mechanisms of social control. 111 Table 2: Description of Variables, Behavior of Law Variable Name Variable Definition Variable Coding Dependent Variable Quantity and Quality of Law Type of initial legal response by the Criminal Referral/Penal victim organization law = 2 Civil Referral/Compensatory law = 1 No referral/No law = O Independent Variables Stratification Vertical aspect of social life Perpetrator’s Age: years Perpetrator’s Gender: 1 = male; 0 = female Morphology Horizontal relationships in social life Perpetrator’s Position in Organization: 1 = Executive or Manager; 0 = employee Culture Symbolic aspect of social life Perpetrator’s Education: 1 = high school diploma 2 = bachelor’s degree 3 = postgraduate degree Organization Corporate aspect of social life Type of Firm: Government agency 4; publicly traded company = 3; private company = 2; non-profit agency = 1 Size of Firm: 1 = l-99;2 = 100-999; 3 = 1000- 9999;4 = 10,000 or more Social Control Normative aspect of social life Sum of Existing Social Control Mechanisms: Background checks + Anonymous reporting + internal audits + external audits Supplementary Variables Time Time Period of Survey Data 1 = 1997-1998; 2 = 2001-2002 Crime Seriousness Loss due to fraud case Dollar loss (sq root) Type of Fraud Classification of occupational fraud Asset misappropriation = 1; Corruption = 2; Fraudulent statements = 3 112 Control and Supplementary Variables Several other variables not used in the primary analyses were considered as control or supplementary variables. The first of these was time, and reflects whether the case occurred in 1997-98 (coded 1) or 2001-2002 (coded 2). The additional supplementary variables included crime seriousness (measured as the dollar loss due to the fraud case) and the type of fraud (asset misappropriation, corruption, or fraudulent statements). These variables were selected because Black argues that his theory is so general it can apply at all times, and that the type of crime or harm caused by the crime does not determine the behavior of law. Data Analyses To examine the research. questions presented. in the beginning of this chapter, several statistical procedures were warranted. These included descriptive statistics, bivariate associations, analysis of variance, and multinomial logistic regression. 113 Descriptive Statistics The first step in the analyses was a generation of descriptive statistics for all variables of interest. This step was performed for each time period, and for the full sample containing data from both time periods. Bivariate Associations After examining the distributions of all variables and comparing the two time periods, the second data analytic step consisted of an assessment of bivariate relationships (e.g., correlations) between perpetrator and organizational characteristics to types of fraud for both time periods, following by a similar assessment to legal responses. The directions of the relationships were examined, and the appropriate measure of strength of association (e.g., chi- square, gamma, t-test) was used. Analysis of Variance Following initial statistical techniques, the first analysis included a comparison of individual and organizational characteristics for the three nmtually exclusive types of occupational fraud. Since all of the 114 cases (n = 1142) were classified as one of the distinct types of fraud, this variable was used to group the cases for the comparisons of individual and organizational characteristics. The purpose of analysis of variance (ANOVA) is to test for significant differences between two or more means (Tabachnick and Fiddell, 2001). This procedure is widely used in the social sciences, and provided the most appropriate way of examining the relationships for the study’s first set of exploratory analyses. There are several assumptions that generally must be met before using ANOVA. These include: independence, normality, and equality of variance. The F-test is remarkably robust to deviations from normality, (Bachman and Paternoster, 1997; Lindman, 1992) so the effect of violating the assumptions was not a serious concern in this study. Additionally, the large sample size offset many of the issues associated with violating these assumptions. Under the null hypothesis in ANOVA, there are no significant mean differences. The alternative, or research hypotheses, states that there are significant differences between means. ANOVA partitions the total variance into the component that is due to true random error (i.e., 115 within groups sums of squares) and the components that are due to differences between means. The two estimates of variance: are then. compared. with. the F—test, which. tests whether the ratio of the two variance estimates is significantly greater than 1. If the F-test is significant for an overall ANOVA model, one can conclude that significant mean differences exist. When the means on certain characteristics of multiple groups are examined, such as the three types of occupational fraud used in the present study, more detailed comparisons can determine exactly where the significant differences occur. This can be accomplished using a post hoc test. For example, the age of individuals who commit asset misappropriation may be significantly different than those who commit corruption, but not fraudulent statements. Following the series of one-way ANOVAs, more detailed comparisons were addressed by using the Bonferroni multiple comparison test. Although Multivariate Analysis of Variance (MANOVA) could also be used to compare means when there are several dependent variables, this procedure assumes that the dependent variables are highly and significantly correlated. In: the present study, this was not the case. 116 Additionally, MANOVA has been discussed as a more powerful statistical technique than a series of separately performed ANOVAs, because it controls for intercorrelations between dependent measures (Bray and Maxwell, 1985) . However, in the present study, the sample size (n = 1142) was sufficiently large, so conservation of power was not a concern . Overview of Maximum Likelihood Techniques The multivariate techniques used to examine the relationships hypothesized based on Black’s (1976) theory of the behavior of law were based on Maximum Likelihood Estimation (MLE). Because previous tests of the theory have used different approaches, modeling techniques used here were performed in manner that allowed comparisons to prior findings. The primary technique utilized was multinomial logistic regression. Unlike traditional multiple regression models, which rely on least squares estimation procedures, logistic regression (including multinomial logistic) procedures use an MLE procedure (Long, 1997; Menard, 1995). MLE approaches try to find estimates of parameters that make the data actually observed "most likely." MLE begins by 117 assuming the general form of the distribution. Next, the initial values of the estimated parameters are used and the likelihood that the sample came from a jpopulation. with those parameters is computed. Finally, the values of the estimated parameters are adjusted iteratively until the maximum likelihood value for the estimated parameters is obtained (Long, 1997). Logistic Regression Many previous studies testing Black’s theory have measured law as a dichotomous dependent variable whereby 1/yes = some form of law occurred and 0/no = no law. As a result, the most appropriate and common approach to modeling the data in prior tests has been logistic regression (see Copes et al., 2001). In this dissertation, several possible choices to model law existed: as a yes/no conceptualization indicating whether fraud cases were referred (including lxnfll civil and criminal referrals) or not (no referrals); as an ordinal variable whereby criminal referrals represented the most law, followed by civil referrals, and then no referrals; and finally, as a reflective of law’s qualitative dimension, with three categories that are not ordered sequentially. Based on 118 these choices, multinomial logistic regression represented the most appropriate technique. Logistic regression. makes it possible to jpredict a discrete outcome from a group of variables that may be continuous, discrete, dichotomous, or a combination (Tabachnick and Fiddell, 2001). For the analyses of the behavior of law, law was operationalized as a nominal variable to indicate whether cases were referred for either civil or criminal processing, compared to cases that were not referred. Since the outcomes were not continuous or normally distributed, other techniques commonly used in the social sciences, such as simple linear regression, would have been inappropriate. Similar to previous tests of Black’s theory, a form of logistic regression provided the opportunity to examine the effects of several independent variables (i.e., measures of the five dimensions of social life identified by Black) on a nominal dependent variable. The multinomial logistic regression. model required. the following assumptions: (1) the model is correctly specified (i.e., the true conditional probabilities are a logistic function of the independent variables), no important variables are omitted, no extraneous variables are included and the independent 119 variables are measured without error; (2) the observations are independent; (3) the independent variables are not linear combinations of each other, since perfect multicolineariby makes estimation impossible, while strong multicolinearity' renders estimates imprecise (Aldrich..and Nelson, 1984). Logistic regression produces information about the odds of an outcome event occurring, which is expressed as an odds ratio associated with each predictor value. The odds of an event are calculated by dividing the probability of the outcome event occurring by the probability of the event not occurring. The odds ratio for a predictor variable indicates the relative amount by which the odds of the outcome increase (odds ratio greater than 1.0) or decrease (odds ratio less than 1.0) when the value of the predictor value is increased knraa one-unit change (Hosmer and Lemeshow, 1989). The outcome variable, i is the probability of producing one outcome or another based on a nonlinear function of the best linear combination of predictors, with two outcomes: ) m ~4 1+e" 120 A where )7 is the estimated probability that the ith case (I = 1,...,n) is in one of the categories and u is the usual linear regression equation: U: A-i- Ble + B2X2 + ...Bka+ 6 with constant A, coefficients BJ, and predictors, X]- for k predictors (j = 1,2, ...,k). The linear regression equation produces the logit or log of the odds: 1n1 f=A+BjXU+e The significance of the estimated coefficients in logistic models can be examined to consider the individual effects of each predictor variable on the dependent variable. There are several possible models in multinomial logistic regression: a constant (intercept) only model that includes no predictors, an incomplete model that includes the constant plus some predictors, a full model that includes the constant plus all predictors (which might also contain some interactions and. power terms) and a hypothetical model that would provide an exact fit of 121 expected frequencies to observed frequencies if the correct set of predictors were measured. Given the variety of available models, there are several potential comparisons: between the constant-only model and the full model, between two incomplete models, between a chosen model and the hypothetical model, or a variety of other potential contrasts. Each proposed model may be critiqued individually based on goodness-of—fit statistics. Additionally, the potential, separate models can be compared in this manner to determine which model provides the best fit to the data. A number of measures may be used to assess goodness-of-fit, and are available in SPSS 11.0, the statistical software program that was used in all analyses. The first of these is a classification table, which will show how many of the observations have been predicted correctly after a given cut-off value, c, (usually .5) has been chosen. For each observation in the sample, the outcome variable (e.g., asset misappropriation) is predicted as “1” (i.e., yes) if the fitted probability of committing this type of occupational fraud is greater than c, otherwise it is predicted as O (i.e., no). The table produces the number of correct predictions, which can be expressed as a 122 percentage. A higher overall percentage of correct predictions will indicate a better fitting model. However, a problem with this basic goodness-of—fit measure is that there is no formal test to determine whether a certain percentage of correct predictions is sufficient. Additionally, it is also possible to create a situation where the specified logistic model is the correct model and therefore will fit the data, but the classification will be poor. In addition. to the classification table, SPSS will also produce two statistics that are roughly equivalent to the R2 in linear regression (Tabachnick and Fiddell, 2001). These are the Cox and Snell R2 and the Nagelkerke R4 (also known as adjusted R2). The Cox and Snell R2 has the disadvantage that for discrete models such as logistic regression, it may not achieve the maximum value of 1.0, even when the model predicts all of the outcomes perfectly. The Nagelkerke R‘2 is an.improvement over the Cox and Snell R2 in that it can attain a value of one when the model predicts the data perfectly. The Likelihood Ratio (LR) Test is an option that may be used to formally test whether a variable is significant in explaining some of the variability in the response. This test is appropriate when 123 two models are being evaluated as follows: Model 1: logit(n)= BO + BLX1+ e Model 2: logit(n)= BC + BIXI 82X; + e In this situation, Model 1 is nested within Model 2, since all of the independent variables contained in Model 1 are also included in Model 2. The comparison will indicate whether the additional variables included in the second model are necessary for explanation, or whether the simpler model fits the data as well as the full model. SPSS will produce the —2 Log Likelihood statistic as a measure of discrepancy between each models observed and fitted values. The value of this statistic for the simpler model will be greater than for the full model. The LR test statistic is the difference in the value of the -2 Log Likelihood between Model 1 and Model 2. An alternative to the LR Test is the Wald Test, which is also produced in SPSS. In most cases, this statistic would lead to the same conclusions as the LR Test, but the LR test is considered more robust because, unlike the Wald Test, it is not affected by large standard errors (Menard, 1995). 124 The Hosmer-Lemeshow Goodness-of-Fit Test may also be used to examine model fit (Hosmer and Lemeshow, 1989). With this test, cases are divided into a number of approximately equal groups based on their predicted probabilities of an event occurring (e.g., the “yes” or “1” outcome). The predicted probabilities are then summed to calculate the differences between observed and expected values. This test is also fairly robust, but does have the requirement that there should be a large sample size so expected numbers in the groups will be greater than five and none of the groups will have expected values that are less than one. All of the aforementioned goodness—of-fit statistics have a Chi-Square distribution: they compare observed and expected outcome values. As such, the Chi- Square values produced with each statistical test can be compared tx: the Chi-Square distribution tx: determine whether the values are significant, or in the case of model fit and comparison, not significantly different from each other. 125 Multinomial Logistic Regression The Multinomial Logistic Regression Model is equivalent to running a series of binary logits. The MNLM uses maximum likelihood to derive parameter estimates. Long (1997: 151) suggests that the MNLM is superior to a series of binary logits because: “...all of the logits are estimated simultaneously, which enforces the logical relationship among parameters and uses the data more efficiently.” Following the initial data analysis steps outlined in this chapter (i.e., generation of all descriptive statistics and examination of bivariate associations) for the second set of hypotheses, the multivariate analyses were performed using MNLM. Results were compared to determine whether previous conceptualizations of law (i.e., as a dichotomous variable) were confirmed to be appropriate, or whether alternative approaches (i.e., as qualitative) provide a: better fit to these data. This approach was advantageous given the possibilities of multiple comparisons (i.e., civil and criminal referrals each compared to no referrals, and civil and criminal referrals compared to each other). 126 Conceptual Models The following two figures provide a graphical depicticul of the jproposed. conceptual relationships among the variables used in this study. In Figure One, the comparison of individual and organizational characteristics for the types of occupational fraud are examined. The second set (NE analyses examines the relationships between the five dimensions of social life and the behavior of law. Figure 1: Conceptual Model Linking Individual and Organizational Characteristics to Types of Fraud Individual Characteristics -Asset Misappropriation -Corruption -Fraudulent Statements Organizational Characteristics 127 Figure Two depicts the conceptual relationships between the five dimensions of social life and the quantity and quality of law as described by Black (1976). The second set of analyses consisted of a series of multinomial logistic regression models, in which each group of independent variables (stratification, morphology, culture, organization and social control) were used to predict each type of legal response. Consistent with previous research (Copes et al., 2001), relationships were first assessed for each separate dimension of social life, and then combined in four models: a reduced ‘model including only Black’s theoretical categories; a reduced. model adding only the type of fraud; a reduced model adding only the dollar loss of fraud (crime seriousness); and a full model including all of the theoretical predictors and both control variables. Models were specified for the comparison between civil and criminal referrals to In) referrals, and comparing civil to criminal referrals only. All of the above models were estimated at each time period, and for the full sample. The goodness—of-fit of each respective model was examined and associated statistical significance tests (e.g., chi square) were compared to determine which type of model provided the best fit to the data. 128 Figure 2: Conceptual Model Linking Dimensions of Social Life to the Behavior of Law Snafificauon Morphology Criminal Referral/Penal Culture Law CWH Referral/Compensatory Law Organization No Legal Referral Social Control Data Considerations Limitations Generalizability A potential limitation of these data, common in much of criminal justice research, particularly studies of white—collar crime, is that the cases used in the analyses contained information on individual employees who were caught for committing fraud. The study design specified that CFEs supply information on completed fraud investigations in their employing organizations. As a result, the sample is restricted to those offenders whose criminal activities were detected. Generalizability to the larger, unknown population of offenders who may commit fraud against their employing organizations is therefore a concern, This limitation. makes it impossible to fully determine the deterrent effects of existing internal and external mechanisms of social control. Researchers in criminal justice have long recognized that it is impossible to measure the effects of general deterrence, but samples of detected offenders do allow for speculation about the efficacy of specific deterrence for those offenders (Clear and Cole, 2003). For example, if offenders in the sample have previous fraud convictions, this may suggest that prior punishments were insufficient in deterring the current fraud act. Sample Selection Bias A concern related to generalizability is that of sample selection bias, which is a common limitation in criminal justice research that considers more than one stage of legal processing. In the present study, sample selection bias is not a concern at the point of the initial 130 decision made by the victim organization. This is because all cases in the sample have a known probability of being either referred for criminal processing, civil processing, or not referred. When subsequent analyses predicting legal responses beyond initial referral are performed, sample selection bias occurs because data for each additional stage (e.g., conviction, sentencing, sentence length) is contingent upon whether cases were initially referred for legal processing. This issue can lead to biased inferences about social processes (Winship and Mare, 1992). To account for the sample selection bias, Heckman’s (1979) correction can be used in any additional models that incorporate additional legal processing stages. The correction uses the following two step process: 1) logistic regression is used to estimate the likelihood of an offender progressing to the subsequent stage of legal processing. For each case in this model, the predicted probability of being excluded from subsequent stages, referred to as the hazard rate, is calculated. 2) The hazard rate is then used as an independent variable into additional regression models predicting the subsequent stage of processing. For example, to examine the likelihood of a fraud conviction occurring, sample 131 selection bias would be addressed by using the two steps outlined by Heckman. The procedure would control for the probability of each case not receiving a fraud conviction. At the time of any subsequent data analyses, Heckman’s correction will be incorporated as needed, based upon the final specification of multivariate models. Measurement An additional potential limitation is; that the measurement of occupational fraud is also based on those offenders who have been detected. This issue has persisted in the literature on white-collar crime, resulting in researchers wondering whether we are simply studying the behavior of the most unlucky criminals. Like the broader concern of generalizability, the imperfect measure of white—collar crime has also been accepted. Given that the behavior being studied is illegal, it is unlikely that any superior method of obtaining information could be developed. For several reasons, the data used in the present study are superior to measures such as self-report surveys, which have also been used to measure crime and deviance in organizations (Greenberg, 1993; 1990). For example, unless employees who respond to a self-report 132 survey have also had involvement with the criminal justice system (e.g., arrests, conviction), there would be no way for researchers to validate the survey information. For a related reason, a self-report survey also would not make it possible to study the range of legal responses to types of fraud, which is a goal of the present research. As the previous literature review' demonstrated, the ‘majority' of existing studies on white—collar crime in general and fraud in. particular can. be critiqued. based on their research designs and related methodological issues. Secondary Data Analysis A final potential limitation is the fact that these data were not obtained directly by the researcher, but were originally collected.kn/ an outside source. .Although this issue should be acknowledged, it should also be noted that this is unlikely to be a serious concern. Unlike traditional sources of secondary data, (e.g., the National Archive on Criminal Justice Data), the researcher was in contact with the .ACFE throughout the dissertation study period, and this communication provided detailed information about the survey' design and subsequent data collection issues. 133 Despite the potential limitations, the nature of these data allow for a variety of advanced comparisons that have not been considered. in jprevious research. The following section discusses the study's strengths, which greatly outweigh the aforementioned potential limitations. Strengths Although the sample may be restricted to those perpetrators whose offenses were detected, there are still many advantages associated with these data, For example, the sample size is larger than many previous studies, and data were obtained at two different time periods. Additionally, the information provided by CFEs was obtained from several sources, including the perpetrators themselves, organizational files, and various criminal justice agencies (e.g., courts, police, and criminal records). What is more, multiple units of analysis are available, which. is atypical in studies of white-collar crime in organizations (Vaughan, 1992). As prominent general systems theorists have stressed through the use of a biological metaphor to describe organizations, “the whole is greater than the sum of its parts.” In the study of organizations, particularly those whose employees are 134 involved in crime and/or deviance, it is unlikely that such phenomenon will be fully understood without considering individual and organizational factors. Failing to consider factors at both of these levels limits comprehension of the multiple causal factors that may influence different types of fraud (Reed and Yeager, 1996). When white-collar crime and the legal response to different types of crime are only explained in terms of perpetrator characteristics, such as low self-control (Gottfredson and Hirschi, 1990) conclusions can be reductionist. The present study addresses this disadvantage by examining characteristics of individuals as well as organizations. The inclusion of multiple level factors improves on previous studies, and will contribute to research, policy, and practice on white- collar crime in organizations. As the literature review demonstrated, studies of organizations as victims are extremely rare. Ima previous studies of white-collar crime in organizations, specifically fraud, have concurrently included victim organizations varying in size and function, perpetrator characteristics, legal responses to fraud, perceptions of CFEs, and data from two time periods. Additionally, studies testing Black’s (1976) theory of the behavior of 135 law have not been conducted on multiple organizations as victims at two different time periods. Also, no previous empirical tests of the theory of the behavior of law have operationalized legal responses in terms of quantity and quality. Policy Implications Many' cases of “white-collar crime” in. organizations reflect the failure of existing forms of internal controls (Vaughan, 1992). Despite the enormous economic and societal losses due to different forms of fraud, the most effective sanctioning strategies are yet tx: be discovered. The present study will compare legal responses to different types of occupational fraud, and. will determine whether differential implementation of law, by offender characteristics, such as age and employee position, occurs. Recent anti—fraud legislation is promising, but, like much existing research, focuses on large, 1 profit-making organizations and their top executives who may engage in corporate, Inn: occupational, white-collar' crime. Policy implications t1) be discussed 1J1 this dissertation include the value and apparent justice of deterrence and compliance strategies for different types of fraud against 136 organizations of varying size and function. Based upon the multivariate analyses as well as the perceptual, qualitative data obtained from CFEs, suggestions for changes in existing regulations and sanctions will be made. Practice Implications After determining what types of occupational fraud are most prevalent at each time period, and whether the characteristics of perpetrators and organizations differ based on the type of fraud, this study will discuss the effectiveness of existing processes (i.e., background checks, anonymous reporting systems, internal and external audits) in preventing and detecting specific types of fraud in different types of organizations. The initial analyses will compare individual and cmganizational characteristics by the types of fraud, and will show whether any predictable patterns occur. This analysis will determine whether certain organizations are more vulnerable to certain types of fraud, which will contribute to improvements in prevention and detection of future fraud cases. Specifically, this study will reveal the types of fraud that occur regardless of internal control approaches that are in place in organizations, the consistency of 137 civil and criminal sanctions in response to particular types of offenses, and the degree to which sanctions are driven by organizational and perpetrator characteristics. Recommendations for changes in current organizational systems will be provided, along with suggestions for additional measures (e.g., employee hotlines, training, and education). Perceptual information provided by CFEs will be used to supplement the multivariate findings and to speculate about the effectiveness of existing sanctions. To disseminate knowledge to practitioners in the field of fraud prevention, a detailed summary of the research findings from this dissertation will be made available to the 30,000+ ACFE members. Conclusion To summarize, this chapter outlined the research questions of the study, discussed the data and sample, and operationalized the dependent, independent, and control variables that were used in the analyses. An cwerview of the statistical techniques employed was also provided. The chapter concluded with a discussion of the study’s potential limitations, and benefits. In the next chapter of this dissertation, Results, the findings from all 138 statistical analyses are presented and interpreted, This dissertation will conclude with a discussion of the results and their implications for criminological theory and future research, as well as for the practice side of fraud prevention in organizations. 139 CHAPTER FOUR: RESULTS As described in Chapter 1, the purposes of the research were to: l) examine differences in individual and organizational characteristics for the three types of occupational fraud and 2) describe and explain the legal responses to occupational fraud cases using Black's theory of the behavior of law and conflict theory. The analysis of data addressed these objectives using several statistical techniques: descriptive statistics, bivariate analysis, analysis of 'variance, and. multinomial logistic regression. This chapter presents the results of these analyses. Analysis One: Individual and Organizational Characteristics Related to Specific Types of Fraud Descriptive Statistics A total of 1,142 offenders from the two surveys comprised tflm: full sample. Table 23 provides the descriptive statistics for the overall sample as well as for the individual time periods, and demonstrates that 140 pooling the two datasets is an appropriate technique. The statistics are presented for the full sample. When notable time differences occur, they are also discussed. For the type of occupational fraud, the overwhelming majority of the sample (84% or n = 959) committed asset misappropriation, while 8.7%' 01:: 99) committed corruption and the remaining 7.4% (n = 84) were involved in fraudulent statements. Turning to individual characteristics, the average age of the sample was 40.9 years. Approximately 53.2% (n = 574) were males while 46.8% (n = 505) were females. The majority of the sample (56% or n = 640) had a high school or less education, while around one-third (33.8% or n = 386) had bachelor's degrees and 10.2% (n = 116) had graduate degrees. Over half of the offenders (60.2% or n = 688) held lower level, employee positions and the remaining 39.8% (n = 454) were managers or executives. For organizational characteristics, the greatest percentage of organizations (36.2% or n = 377) had 1-99 employees, followed by 25.8% (n = 265) with 1000-9999 employees, 20.3% or r1== 208) with 100—999 employees, and 17.2% (n = 177) with 10,000 or more employees. The types of organizations represented in.tflua sample included 32.1% 141 (n.== 367) that were privately held, 30.8% hi = 352) that were publicly traded, 24.8% (n = 283) that were government agencies, and 11.2% (n = 128) that were non-profit agencies . For the four existing control mechanisms , one point to be noted is that anonymous reporting was a fairly new technique, and was not asked about during the 1997-98 survey. The data for this variable represent 2001-02 only, and show that 35.6% (n = 202) of organizations had anonymous reporting in place in the year before the fraud case. Over half of the organizations (59.7% or n = 587) used. background. checks, while 50.7% hi =: 528) practiced external audits and 62.4% (n = 661) conducted internal audits. The final organizational characteristic, mean revenue (square root transformation)7, was relevant only to publicly traded companies. For the pooled sample, the mean revenue was $29,874.40. However, a notable difference between the two time periods is that mean revenue of $39,913.80 in 1997-98 was greater than the mean revenue of $24,814.10 during 2001—02. The explanation for this difference can be attributed to the status of the overall 7 Due to extreme values, the square root transformation was used. 142 United States economy. In 1997 alone, combined revenues of the 25 largest U.S. corporations equaled the revenue of the U.S. federal government. Comparatively, the U.S. economy was in a recession for most of 2001, and did not begin to recover ‘until early' 2002 (Associated. Press, 2003). The difference in revenues for each time period was expected. 143 Table 3. Descriptive Statistics, Analysis One 1997-1998 Survey 2001-2002 Survey Pooled Sample % (n) % (n) % (n) Type of Occupational Fraud Asset misappropriation 81.8 (392) 85.5 (567) 84.0 (959) Corruption 12.7 (61) 5.7 (38) 8.7 (99) Fraudulent statements 5.4 (26) 8.7 (58) 7.4 (84) Individual Ch_aracteristics Mean age in years 40.7 (479) 41.1 (587) 40.9 (1,066) Gender Male 52.8 (253) 53.5 (321) 53.2 (574) Female 47.2 (226) 46.5 (279) 46.8 (505) Education High school or less 54.9 (263) 56.9 (377) 56.0 (640) Bachelor’s degree 35.3 (169) 32.7 (217) 33.8 (386) Graduate degree 9.8 (47) 10.4 (69) 10.2 (116) Position Employee 62.0 (297) 59.0 (391) 60.2 (688) Manager or executive 38.0 (182) 41.0 (272) 39.8 (454) Organizational Characteristics Size 1-99 employees 34.0 (163) 39.1 (214) 36.7 (377) 100-999 employees 19.6 (94) 20.8 (114) 20.3 (208) 1000-9999 employees 28.4 (136) 23.5 (129) 25.8 (265) 10,000+ employees 18.0 (86) 16.6 (91) 17.2 (177) Type Government agency 25.3 (121) 24.4 (162) 24.8 (283) Non-Profit agency 8.1 (39) 13.4 (89) 11.2 (128) Privately held company 33.2 (159) 31.4 (208) 32.1 (367) Publicly traded company 33.4 (160) 29.0 (192) 30.8 (352) Existing control mechanisms Anonymous reporting (% yes)* --- 35.6 (202) 35.6 (202) Background checks (% yes) 71.4 (342) 48.5 (245) 59.7 (587) External audits (% yes) 25.3 (121) 72.3 (407) 50.7 (528) Internal audits (% yes) 68.9 (330) 57.0 (331) 62.4 (661) Mean revenue (square root)“ 39,9138 (160) 24,814.1 (192) 29,8474 (352) * 2001-02 sample only (11 = 663) ** Publicly traded companies only (n = 352) Bivariate Relationships The strength and direction of relationships between study variables were examined with bivariate correlations. Table 4 reports the correlations between the individual and organizational characteristics used in Analysis One. Although some of the correlations between the individual characteristics are statistically significant, most are weak. There were weak, positive correlations between age and gender (r= .18; p 5 .01); age and education(r= .22; p 5 .01); as well as age and position (r= .18; p 5 .01). The correlations were also weak and positive for gender and education (r= .15; p 5 .01) and for gender and position (r= .12; p 5 .01.) These relationships were in the expected direction. Somewhat unexpectedly, the correlation between education and position was also weak (r= .31; p _<_ .01) but still statistically significant and in the expected direction. Age was also related to several organizational characteristics. The correlation between age and size of organization, although statistically significant, was negative and weak (r= -.10; p 5 .01). Age was positively and weakly correlated with working in a government agency (r= .10; p 5 .01) as well as working in a non-profit agency 145 (r= .07; p 5 .01). Age was also negatively and weakly correlated with working in a privately held company (r= - .09; p 5 .01) and working in an organization that conducted internal audits (r= -.11; p 5 .01). Age was not significantly correlated with any of the additional organizational characteristics. Gender was not significantly correlated with any of the organizational characteristics. The correlations between education and the organizational characteristics were very weak, with only two such relationships achieving statistical significance. These included education and size of organization (r= -.O7; p _<_ .01) as well as education and working in a government agency (r= -.O7; p 5 .01). Similar relationships were revealed for position. Being a manager or executive was weakly correlated with size of organization (r= —.07; p 5 .05); working in a government agency (r= -.O7; p 5 .05); and working in an organization that conducted internal audits (r= -.11; p 5 .05). There were several statistically significant correlations among the organizational characteristics. Larger organizations were more likely to be publicly traded companies (r= .38; p _<_ .01), less likely to be government agencies (r= -.33; p 5 .01), and less likely to be non- 146 profit agencies (r= -.14; .01). Larger organizations were also significantly more likely to have anonymous reporting systems (r= .48; p 5 .01); background checks (r= .12; p 5 .01) and internal audits (r= .44; p 5-.CHJ. .As expected, size of organization was also positively related to its annual revenue (r= .38; p 5 .01). The type of organization was correlated with several other organizational characteristics. Being a government agency was negatively correlated with being a publicly traded company (r= -.38; p 5 .01); with being a pmivately held company (r= -.39; p 5_ .01); and with being a non-profit agency (r= -.20; p 5 .01). Government agencies were more likely to use two of the internal control mechanisms: anonymous reporting (r= .16; p 5 .01); and internal audits (r= .13; p 5 .01). It was surprising and unexpected that the correlation between government agencies and background checks was extremely weak and not statistically significant. This suggests that the government agencies where the frauds occurred and/or where the certified fraud examiners worked may not be representative of the larger pepulation of government agencies that typically require background checks as a pre- requisite for employment. For the most part, the relationships between publicly 147 traded companies and existing control mechanisms were positive and statistically significant. Publicly traded companies more often used anonymous reporting (r= .22; p 5 .01); background checks (r= .07; p 5 .05); and internal audits (r= .23; p 5 .01). As expected, publicly traded companies were also positively correlated with annual revenue (r= .25; p 5 .01). There were several significant correlations between privately held companies and other organizational characteristics. Privately held companies less often used anonymous reporting (r= -.25; 1).: .01); background checks (r= -.O9; p 5 .01); internal audits (r= -.30; p 5 .01); and had lower revenues (r= —.22; p 5_.CHJ. The relationships with the existing control mechanisms all suggest that privately held companies, which were also smaller, may not have the needed resources to implement control mechanisms. Similarly, the resources of non-profit agencies may also be limited. Non-profit agencies were negatively correlated with anonymous reporting (r= —.15; p 5 .01); as well as internal audits (r= -.O9; p 5 .01). Several variables ‘were significantly' related. to ‘the use of internal control mechanisms. Anonymous reporting was positively correlated with background checks (r= .38; p 148 .5 .01); internal audits (r= .56; 13:5 .01); external audits (r= .18; p ._<_ .01); and revenue (r= .13; p 5 .01). The correlation between background checks and internal audits was also positive (r= .24; p 5 .01), as was the correlation between internal audits and. revenue (r= .28; p .E .01). These findings imply that organizations with at least one mechanism of internal control are more likely to have additional control mechanisms in place. The bivariate relationships -between these variables only provide a partial picture. For Analysis One, more detailed analyses, reported next, will reveal whether there are significant differences in individuals who commit asset misappropriation, corruption, and fraudulent statements, and in organizations that are victimized by the three types of occupational fraud. 149 82558 a. v a : €83.38 8. v a . 8. :8. S. .2. 8.- 2.8.. .8. 8.- :8. 8.- 8. S.- 3- 8.588881 S. S. :8. 8. 8.- 8. a. 8. 8. 8. 8.- 8.- 882823.: :3: :8. :8.- ..8? :8. :2. :3. ::.- 8.- 8. ::.- assesses :8. 8.- :8.- .8. 8. :2. 8.- 8. 8. 8.- seat 88308.: :2- :8.- :2. :2. :8. 8.- 8. 8. 8.- “1588.83.85? :8: Ln.- ..8- :E- 8.- 8. 8.- ...8. Samson-82a :8.- ..am: :8.- 8. 8.- 8.- :8.- .ooeeisséas :8.- ..8. 8. 8. 8. 8.- .8 SEE 2255.8 8. 8.- 8.- 8. :2. 568 8866.60 .8 .8.- ..8.- 8.- :2.- came :s. :2. :8. :88; .4 :2. :8. Seances :m _. 880 .m ow<._ : 2 N. : S a w a e m I. m N I 28:; mocmtoaofiezu _mcozeNEmeO can .3229: we x322 cocflotoo .v 03.8. 150 Analysis of Variance A series of one-way ANOVA models comparing the three types of cmcupational fraud (i.e., asset misappropriation, corruption, and fraudulent statements) were estimated for individual characteristics and organizational characteristics. The null hypothesis for ANOVA states that there will be no statistically significant differences between means (M1 the individual characteristics and organizational characteristics for each type of fraud. In each one-way ANOVA model, if the overall F-test is significant, the null hypothesis can be rejected and one may conclude that significant mean differences exist. However, to determine which specific means differ significantly from each other, the Bonferroni multiple comparison test is used (p < .05). The test allows for direct comparisons between groups of unequal sample sizes (Mason, Gunst, & Hess, 1989). Individual Characteristics Table 5 jpresents the results of the one-way .ANOVA models comparing occupational fraud types by individual 151 characteristics. For age, gender, education, and position, the Furatios indicated significant variation between asset misappropriation, corruption, and fraudulent statements, which indicates that at least one of the means is significantly different. The Bonferonni multiple comparison test revealed that individuals who commit asset misappropriation are significantly younger than those who commit both corruption (mean difference = -2.73; p < .05) and. fraudulent statements (mean. difference = ~4.03; 19 < .05). However, the age differences between those 'who commit corruption and fraudulent statements are not statistically significant. Where gender is concerned, those who commit asset misappropriation are significantly less likely to be male than those who commit fraudulent statements (mean difference = -.18), but there are no significant gender differences when comparisons are made between asset misappropriation and corruption, or corruption and fraudulent statements. This result differs from Daly’s (1989) finding that men were more likely to commit crimes in groups, which is consistent with corruption. 152 For education, those who commit asset misappropriation have significantly lower levels of education than those who commit both corruption (mean difference = -.26; p < .05) and fraudulent statements (mean difference = —.24; p < .05), but significant educational differences were not detected in the comparison between corruption and fraudulent statements. The final post hoc comparisons for position show that those who commit asset misappropriation are significantly less likely to be managers or executives than those who commit fraudulent statements (mean difference = -.36; p < .05). People who are involved in corruption also hold significantly lower positions in the organization compared. to those who ‘make fraudulent statements (mean difference = .29; p < .05). In sum, the findings support the hypothesis that all of the characteristics of individuals who commit the three forms of occupational fraud (asset ndsappropriation, corruption, or fraudulent statements) differ significantly.2 2Criminal history, operationalized as whether the perpetrator has any prior fraud charges or convictions (1 = yes; 0 = no), was originally considered as a fifth characteristic, but was only available for the 2001-02 data. Less than 5% of the sample had any known criminal history, consistent with prior studies. Preliminary analyses indicated that there were no significant mean differences in the criminal histories of individuals who committed asset misappropriation, corruption, or fraudulent statements, so this variable was dropped. 153 These results suggest that occupational fraud is not a blanket form of white-collar crime perpetrated by individuals who share distinct characteristics such as those commonly assumed in the literature. On the contrary, the characteristics of perpetrators vary based on the specific type of occupational fraud committed. For example, the results suggest that the crime of asset misappropriation does not require any advanced knowledge or skills that would likely be associated with more education and higher-ranking positions. Although. there are significant, individual 'mean differences between the three types of occupational fraud, it is crucial to note that the some of characteristics of this sample as a whole are at least somewhat comparable to the literature’s depiction of the typical white-collar offender. In terms of demographics, recall Wheeler et al.s’ (1988) description of the average white-collar offender as “a white male, age 40.” Consistent with this finding, the average age of the present sample is 40.9. Racial composition of the offenders was not recorded in this study’s surveys, so IK) comparisons t1) the literature on this variable can be made. 154 The averages for other characteristics did not support previous findings. For example, a nearly equal percentage of men and women committed fraud, the majority were lower level employees rather than. managers or executives, and over half possessed a high school diploma while the remaining 44% had four-year or graduate degrees. What can be said about these notable differences? Bear in mind that many previous studies supporting the popular white-collar criminal image were based on samples of convicted offenders. The lack of comparability to the present study suggests what many have long recognized: that those offenders who reach the sentencing stage of the criminal justice system are distinctly different than those who are filtered out at earlier stages, and are therefore unrepresentative of the larger population of offenders (Zatz, 1987). Portrayals of typical white-collar offenders that are based on such limited samples may serve to only perpetuate potential biases in sentencing and related criminal justice policies. 155 8v}. ISL-mm me. me. hm. :oEmom 1.3.2 E. on. on. cosmosom ...mné eo. mm. mm. 5260 A; 8.3. 8.8 8.3 8;: come-k 856688 Eco—seamen scone-SO cocmtooaomflfi 8mm< 8558820 12636:— xo cox-_- ozmi _mcocwozooo .m min-_- 156 Organizational Characteristics Table 6 presents the results of the one-way' ANOVA models comparing occupational fraud type by organizational characteristics. Overall, the F-ratios for size of organization, publicly traded company, privately held company, anonymous reporting, and internal audits all show significant variation between asset misappropriation, corruption, and fraudulent statements. However, the F- ratios for non-profit agency, background checks, external audits, and revenue, do not differ significantly for the three types of occupational fraud. These results partially confirm the hypothesis that the characteristics of organizations victimized by different types of fraud would differ. For the organizational characteristics that were statistically significant overall, the Bonferroni multiple comparison tests reveal more detailed differences between types of fraud. Organizations victimized by asset misappropriation are significantly smaller than organizations victimized by corruption (mean difference = - .42; 53 < .05). .Additionally, organizations victimized by corruption are significantly larger than organizations 157 victimized by fraudulent statements (mean difference = .46; p < .05). When they are victimized, publicly traded companies have significantly more corruption than asset misappropriation (mean difference = .16; p < .05). In the victimized privately held companies, there is significantly more asset misappropriation than corruption (mean difference = .12; p < .05). For non-profit agencies, there is significantly more asset misappropriation than corruption (mean difference = .10; p < .05) as well as significantly more asset misappropriation than fraudulent statements (mean difference = .11; p < .05). When the separate, existing control mechanisms are compared for organizations that experienced the cases of occupational fraud, results show that victimized organizations with anonymous reporting systems are significantly more likely to find corruption than either asset misappropriation (mean difference = .39; g>i< .06) or fraudulent statements (mean difference = .28; p < .05). This suggests that anonymous reporting systems may provide security for those individuals who “rat" on fellow employees, by eliminating potential negative consequences and stigma of whistleblowing. The only other type of control mechanism to attain statistical significance is 158 internal audits. Organizations that conduct internal audits are more likely to find corruption than asset misappropriation (mean difference = -.14; p < .05). How do these results compare to the literature’s image of the large, publicly traded company as a frequent setting for white-collar crime? Compared to organizations that fall victim to asset misappropriation and fraudulent statements, organizations victimized by corruption are significantly larger, and more likely to be publicly traded. In this sample of victimized organizations, corruption is rare in both non—profit agencies and privately held companies; this may be due to the fact that these types of organizations are also smaller, and collusion. with outsiders may' be nearly impossible given potentially close associations between employees. Although corruption victimizes the organization rather than benefits it, this particular form of fraud tends to take places in a setting that is consistent with previous studies of corporate crime. On a related note, the corporate crime literature identifies profit maximization. as a key motivating factor 1J1 white-collar' crime. Although individual offender motivation is beyond the scope of the present study, the findings reported here show that annual 159 revenue of publicly traded companies is not related to any form of occupational fraud. This result highlights key differences between occupational crime, committed against the organization, and corporate crime, committed on behalf of the organization, supporting this typological distinction (Clinard & Yeager, 1980). 160 8v}. mw.m No.wm~.m_ ~13de moSv—dm 308 8553 Begum *Nw.m mm. mm. _o. 26.5 BEBE 3.. on. mu. _m. 8:26 35me _NN we. mo. mm. 8.85 oesoLwo—omm *3 .2 we. mm. mm. wEtooE msoExcoc< LEN mm. 3. mm. Emofioo Eon 36535 e _ Wm mm. 3. om. End—=8 coon: 3335 1.38 .8. mo. m _. Snows ESQ-:02 m _ ._ mm. 5. mm. Snows EoEEoEO ... 5.0 3 .N NRN o~.m ohm 058i mEoEBSm Leo—:osfim cocoa-:60 cocetoeoommL—z Lemme: 8558820 _mcoceficmeO :3 one? BEE 3:033:80 .0 flow-_- 161 Analysis Two: The Behavior of Law and Conflict Theory Descriptive Statistics The sample (n = 1,142) used in Analysis One was also used in Analysis Two. The characteristics of the sample, arranged. by' Black’s conceptual categories for each. time period and the pooled data, are presented in Table 7. For stratification, two measures were included: age and gender. The average age of the sample was 40.9 years. Approximately 53.2%' hi = 574) were males while 46.8% (n = 505) were females. Education is the indicator of culture as defined by Black (1976) and in this dissertation. For this variable, the majority of the sample (56% or n = 640) had high school or less educations, while around one-third (33.8% or n = 386) had bachelor’s degrees and 10.2% (n = 116) had graduate degrees. Morphology was measured using one variable, position in the organization. As the table shows, over half of the cifenders (60.2% or r1== 688) held lower level, employee positions and the remaining 39.8% (n = 454) were managers or executives. 162 To reflect Black’s (1976) theoretical category of organization, two variables were used: size of organization, and type of organization, 'The greatest percentage of organizations (36.2% or n = 377) had 1-99 employees, followed by 25.8% (n = 265) with 1000-9999 employees, 20.3% or r1== 208) with 100-999 employees, and 17.2% (n = 177) with 10,000 or more employees. The types of organizations included 32.1% (n = 367) that were privately held, 30.8% (n = 352) that were publicly traded, 24.8% (n = 283) that were government agencies, and 11.2% (n = 128) that were non-profit agencies. Social control was measured using the degree of existing, non-legal controls available in organizations. This variable ranged from 0 (no mechanisms) to 4 (all four mechanisms present). Just over 200 organizations (207; 18.1%) reported that they did not have any existing control mechanisms in the year prior to the current fraud case. Nearly one quarter (279 organizations or 24.1%) had one mechanism in place, while a slight majority (372 or 32.6%) had two mechanisms in place. Very few organizations reported having three mechanisms (181 or 15.8%), and an even smaller number had all four mechanisms of non-legal social control (103 or 9%). 163 Table 7. Descriptive Statistics, Dimensions of Social Life, Analysis Two 1997-1998 Survey 2001-2002 Survey Pooled Sample % (n) % (n) % (n) Dimension of Social Life Stratification Mean age in years 40.7 (479) 41.1 (587) 40.9 (1,066) Gender Male 52.8 (253) 53.5 (321) 53.2 (574) Female 47.2 (226) 46.5 (279) 46.8 (505) Momhology Position Employee 62.0 (297) 59.0 (391) 60.2 (688) Manager or executive 38.0 (182) 41.0 (272) 39.8 (454) Culture Education High school or less 54.9 (263) 56.9 (377) 56.0 (640) Bachelor’s degree 35.3 (169) 32.7 (217) 33.8 (386) Graduate degree 9.8 (47) 10.4 (69) 10.2 (116) Organization Size 1-99 employees 34.0 (163) 39.1 (214) 36.7 (377) 100—999 employees 19.6 (94) 20.8 (114) 20.3 (208) 1000-9999 employees 28.4 (136) 23.5 (129) 25.8 (265) 10,000+ employees 18.0 (86) 16.6 (91) 17.2 (177) Type Government agency 25.3 (121) 24.4 (162) 24.8 (283) Non-Profit agency 8.1 (39) 13.4 (89) 11.2 (128) Privately held company 33.2 (159) 31.4 (208) - 32.1 (367) Publicly traded company 33.4 (160) 29.0 (192) 30.8 (352) Social Control Existing control mechanisms 0 controls 7.1% (34) 26.1% (173) 18.1 (207) 1 control 30.1% (144) 20.4% (135) 24.4 (279) 2 controls 53.0% (254) 17.8% (118) 32.6 (372) 3 controls 9.8% (47) 20.2% (134) 15.8 (181) 4 controls“ ---- 15.5% (103) 9.0 (107) *Not possible for 1997-98 survey. 164 For the dependent variables, the descriptive statistics for the three possible outcomes are presented in Table 8. The responses reflecting quantity and quality of law are presented simultaneously because, for purposes of frequencies, they' are capturing the same outcome (i.e., criminal referral is the same as penal law and civil referral is the same as compensatory law). The frequencies were similar at both time periods. During 1997-98, a total of 124 cases (25.9%) were not referred; 275 (57.4%) were referred for criminal processing and 80 (16.7%) were referred for civil processing. During 2001-02, a total of 177 cases (26.2%) were not referred, similar to the percentage referred in 1997-98. A total of 466 (67.3%) were referred for criminal processing, up slightly from the 1997-98 survey, and the remaining 40 cases (6%) were referred for civil processing, a lower percentage than in 1997-98 . 165 Table 8. Descriptive Statistics, Behavior of Law, Analysis Two 1997—1998 Survey 2001-2002 Survey Pooled Sample 96(n) 96(n) 96(n) Action Taken No Referral 25.9 (124) 26.7 (177) 26.1 (301) Criminal Referral/Penal 57.4 (275) 67.3 (446) 63.1 (721) Civil Referral/Compensatory 16.7 (80) 6.0% (40) 10.5 (120) Bivariate Relationships Table 9 presents a correlation matrix of the bivariate relationships between the dimensions of social life and the resulting legal outcomes. For stratification, age was significantly correlated with gender (r = .18; p _<_ . 01) , the second indicator of stratification. Age was also significantly correlated with position, the indicator of morphology (r = .33; p 5 .01); education, the indicator of culture (r = .22; p 5 .01); size, an indicator of organization (r = -.10; p 5 .01); and social control (r = - .09; p 5 .01). Gender, the other indicator of stratification, was weakly and significantly correlated with morphology (r = .12; p 5 .01) and with culture (r = .15; p 5 .01). Neither stratification indicator was correlated with quality and quantity of the legal response. 166 For morphology, which was measured using position in the organization, there were several significant correlations. In addition to its relationship with both stratification indicators, morphology was also moderately and significantly correlated with culture (r = .31; p 5 .01). Weak, negative but significant correlations were found between morphology and size of organization, one indicator of organization (r = -.O7; p _<_ .05); also with social control (r = -.O6; p 5 .01). Morphology was not correlated with the legal response. Culture, measured with one indicator, education, was weakly but significantly correlated with size, an indicator of organization (r = -.O7; p 5 .05). With the exception of the aforementioned bivariate relationships with age, gender, and morphology, culture was not significantly correlated with any additional dimensions of social life, nor was it correlated with legal outcomes. Organization was measured using two indicators: size and type. Recall that size of organization was correlated with age, morphology, and culture. Type of organization was significantly correlated with social control (r = .36; p _<_ .01). The two indicators of organization were significantly correlated with each other (r = .28; p 5 167 .01). Neither indicator' was correlated with the legal responses. The last dimension of social life examined. in the bivariate analysis was social control, measured with one indicator reflecting the degree of existing control mechanisms in the organization. As discussed previously, social control was significantly correlated with age, morphology, size of organization, and type of organization. Social control was not correlated with the legal responses. The correlation between age and gender was expected given that Black (1976) states that they are both measures of stratification. Because they provide separate indicators of the construct, stratification, it is not likely for age and gender to be correlated. This was also expected for the two indicators of organization: type and size. However, the correlations between different dimensions of social life are inconsistent with the theory of the behavior of law, but are consistent with criticisms of that theory that claim that the five dimensions are not as conceptually distinct as Black argues. The multivariate analyses will further untangle these relationships, to determine whether the hypotheses about the behavior of law can be supported for these cases of occupational fraud. 168 sues-05v 8. v a : 825-03: 8. v a .. 8 8. 8. 8.- 8. 8.- 8.- 33 8 8.80 a 8. 8 8. 8.- 8. 8.- 8.- 33 8 5:80 .8 :8. :8. 2 .- ...8.- 8.- :8- 3250 38m .8 :8. 8.- 8.- 8. 8. an»: 888880 a .8.- ..8- 8.- :2 .- one 88.8880 .m :8. :2. :8. 23.5 .4 :N _. :8. 822862 .8 :- 1w _. C359 coraoccgm .N ems 8:85.28 ._ w a e m 4 m N _ 288$ 33 .3 3336mm 2: can £5 32% be 508555 .3 x532 coca—250 .o 293. 169 Multivariate Analyses for Behavior of Law Recall from Chapter Three that there were several possible approaches to modeling the data for the multivariate analyses of the behavior of law 'variables. The available options were based on pmevious tests of the theory, the majority of which have used a single, dichotomous variable to measure law (i.e., Did you call the police?; 1 = yes; 0 = no). Previous studies predicting calls to police as well as referrals have used logistic regression, which has generally been the most appropriate technique given the fact that available data rarely include both civil and criminal referrals. In the present study, the majority of cases were referred (73.6%), with 63.1% referred for criminal processing and the remaining 10.5% referred for civil processing. Given these distributions, the choice to combine the two referrals and compare all referrals against cases that were not referred. was an option” However, Black's theory itself clearly specifies that there are important differences in civil and criminal responses. Thus, combining the two referrals into one category could obscure distinctions between civil and criminal responses. A regression model that can incorporate these differences 170 is the most appropriate choice. In the interest of specifying the most parsimonious models that best fit these data, the choice of regression models was determined using steps outlined by Long (1997: 162-164). Specifically, the test of whether two outcome variables (i.e., civil and criminal referrals) could be combined was performed. First, cases with civil and criminal referrals were selected. A.bdnary logit was then estimated (n1 this sample U1 == 841 cases). Independent variables representing the dimensions of social life were included in the model. The Likelihood Ratio (LR) was used to test whether all of the slope coefficients in the binary logit are simultaneously zero. .Although this is essentially a hypothesis test, it can provide support for using a particular approach to model the data. If the null hypothesis were accepted, the independent variables would not differentiate between civil and criminal referrals, so logistic regression would be the appropriate choice. However, if at least one of the slope coefficients differed, support for an alternative technique (e.g., multinomial logit) would be demonstrated. Results from this preliminary test indicated that as a group overall, the parameters were not significantly 171 different from zero (-2 Log Likelihood = 575.910; sig = .219). In this situation, Long (1997: 164) stresses the importance of examining the significance of each individual parameter before deciding to revise the model, as one of the independent variables may still have an important effect differentiating between the outcome categories. Following this recommendation, an examination of the individual parameters in the model indicated that one independent variable, culture (measured I»! level of education) had a significant effect (-2 Log Likelihood = 583.625; sig' = .021). .Since the results of this test provided partial support for both choices of models, the research questions were carefully considered. Given the theory’s emphasis on differences in civil and criminal responses, the most appropriate model is the multinomial logistic regression model. This is essentially equivalent to running three binary logistic regressions, but the simultaneous model is more efficient (Long, 1997). This choice resulted in three types of comparisons: civil referrals vs. no referrals, criminal referrals vs. no referrals, and civil referrals vs. criminal referrals. 172 The Behavior of Law: Multinomial Logistic Regression Models To examine the relationships between dimensions of social life and the behavior of law, several multinomial and binary logistic regression models were specified. The initial relationships predicted by Black’s theory were examined using the three categories of law for the dependent variable. Following Copes et al., 2001, a series of models were estimated separately (with each dimension of social life as the predictor) and then simultaneously, with all of the five dimensions of social life, and finally, with the addition of the control variables. The following dimensions of social life (with their indicators listed) were used as independent variables: Stratification (measured by age and gender); Morphology (measured by position in the organization); Culture (measured by level of education); Organization (measured by size and type of business) and Social Control (measured by the degree of existing internal control mechanisms present in the organization). Given the existing critiques of Black's theory, control variables were also considered. For example, Black argues that the type of crime should not matter, because the relationships with the dimensions of 173 social life are more important in determining the behavior of law. Similarly, traditional indicators of the seriousness of a crime (e.g., dollar loss) should not matter. It is in these two areas that many critics have challenged Black; specifically beginning with Gottfredson and Hindelang (1979), who found that the seriousness and type of crime were significant predictors of the legal response. To determine whether these criticisms hold in the present study, additional multinomial logistic regression models were estimated with the inclusion of the type of fraud as well as the dollar loss (square root transformation) due to fraud as an indicator of crime seriousness.3 According to Black, the theory of the behavior of law applies equally at all times. To examine this statement, separate models .were estimated for the 1997-98 sample, the 2001-02 sample, and the full sample, for each of the previously described relationships. The results of the initial multinomial logistic and regression models predicting the behavior of law based on the separate, five dimensions of social life for the full 3 The square root transformation was used to reduce skewing. 174 sample and each sub-sample are presented in Table 10. This comparison uses no referral as the reference category. Multinomial Logistic Regression Models: Separate Dimensions of Social Life Table 10 presents the series of five separate multinomial logistic regressions examining the dimensions of social life at both time periods and for the full sample. For all three samples, Model 1 assesses the relationship between the stratification variables and the log odds of receiving a civil referral vs. no referral and a criminal referral vs. no referral, respectivelyu In 1997-98, the model chi—square (8.87; df = 4) is significant, indicating the presence of a statistically significant relationship between the combination of age and gender and the corresponding legal response. For criminal vs. no referral, the relationship between age and the legal response is significant using the criterion of p 5 .10, indicating that for each additional one year increase in age, there is a significant decrease in the probability of a criminal referral, which is consistent with Black’s theory as well as conflict theory. However, the relationship does not hold. at more stringent levels of 175 significance. Additionally the odds ratio of .98 is close to 1.0, reflecting that the odds of receiving a criminal referral are nearly equal to the odds of no referral. For gender, there is a significant relationship between one’s gender and the probability of receiving a civil referral vs. no referral. The odds ratio of 1.75 shows that compared to males, females are approximately 75% more likely to receive a civil referral (p 5 .10), also consistent with both Black's theory and conflict theory. There are no gender differences between criminal referrals and no referrals. In Model 1 for the 2001-02 sample, the model chi- square is large and statistically significant at p _<_ .01, indicating a relationship between the stratification variables and the outcome. For each one year increase in age, the probability of a case receiving a civil referral increases significantly, but only by 6%. This is in the opposite direction predicted by Black’s theory and conflict theory. There are no age differences for criminal referrals. For gender, there are no differences for civil referrals, but the probability of receiving a criminal referral increases by 55% for females, supporting the theory of the behavior of law as well as conflict theory. 176 In the full sample, the model chi-square of 8.20 shows a statistically significant relationship between the stratification variables and the outcome. However, the only significant relationship between the individual parameters and the outcome is for gender, which indicates the odds of a criminal referral are increased by 28% for females (p 5 .10), again consistent with Black’s theory and conflict theory. Model Two shows the relationship between morphology and the legal outcome. There are no significant relationships in the 1997-98 sample, indicating no support for this dimension of Black’s theory. For 2001-02, the model chi-square of 8.22 is statistically significant at (p 5 .05), showing a relationship between one’s position in the organization and the likelihood of a case being referred. The individual parameters indicate that this relationship exists only for the comparison between civil and no referrals. Compared to managers and executives, the odds ratio of .42 shows that the odds of receiving a civil referral decrease by 58% for employees, which is in the opposite direction predicted by the theory. Neither the model nor the individual parameters are statistically significant for the full sample, leading to the conclusion 177 that there is no support for Black's theory on this dimension of social life. Model Three examines the relationship between culture and the likelihood of receiving a referral. In 1997-98, the model is not significant. In 2001-02, the model chi-square of 15.33 is statistically significant at (p 5 .01), indicating that there is a strong relationship between education and the outcome variable. The relationship does not hold for the civil vs. no referral comparisons, but shows that an increase in education corresponds with a 32% decrease in the likelihood of receiving a criminal referral, which is in the direction predicted by Black's theory and conflict theory. Similar findings are revealed for the full sample: no differences in the comparison of civil vs. no referral, but a statistically significant decrease of 20% in the odds of receiving a criminal referral. Also, the model chi-square of 9.36 for 2001-02 is statistically significant at (p 5 .05), but smaller than the chi-square in the 2001-02 sample. Model Four examines the relationship between the organization variables (size and type) and the corresponding legal response. For 1997-98, the model chi- square of 8.41 is small but statistically significant at p 178 .10, indicating a relationship between the combination of IA size and type of organization to the legal outcome. The individual parameters for this model indicate no differences in civil vs. no referrals for either size or type. However, both variables exert a: significant influence on the likelihood of receiving a criminal referral. Specifically, as the size of a business increases, the odds of receiving a criminal referral increase by 25%, which is what Black’s theory predicts. For type, lower levels of organization correspond with a 20% decrease in the likelihood of a case receiving a criminal referral, which is in the opposite direction than predicted by the theory. At 2001-02, the model chi-square of 15.93 is also statistically significant, and larger than in 1997-98. However, there are no significant findings for the individual parameters of size. For type, the only significant relationship is in the criminal vs. no referral comparison, which shows that, as a business’ level of organization increases, there is a 37% increase in the likelihood of a case receiving a criminal referral, which is consistent with the theory. When these relationships are 179 examined for the full sample, neither the full model nor either individual parameter is statistically significant. Model Five examines the relationship between social control and the outcome. In 1997—98, the model chi-square of 5.17 is statistically significant (p _<_ .10). There is not a relationship between degree of social control and the likelihood of receiving a civil referral; however, there is a relationship for criminal referrals. Specifically, for each one unit increase in the degree of internal social control, there is a 23% decrease in the likelihood of a case receiving a criminal referral. This finding is consistent with Black’s theory. In 2001-02, opposite findings emerge, but the findings also fail to support the theory. The model chi-square of 22.10 is large and statistically significant (p 5 .01). The odds ratio indicates that, for every one-unit increase in the amount of internal social control available to a business, the likelihood of receiving a criminal referral increases by 31%. There is a similar relationship in the full sample. The model chi-square of 11.65 is again statistically significant (p < . 01) . However, like in 2001-02, for the full sample, increased internal social 180 control results in a 21% increase in the likelihood of receiving a criminal referral. Thus far, the separate multinomial regression analyses provide mixed support for the theory. This is consistent with the findings of Copes et al., (2001), who found limited support when the dimensions of social life were assessed separately. 181 8.wm:. 8. wm: 8.. w... 8.8.85 5 828 mp5 .38 8:85:28 5 Echo BEES: £56508 5895585. 08 mom-Em. .mPOZ 182 .8... .8... .8... .8: E: .8: .225 $.58... .8::8.. .8: 8. 3.28.8 .8338. 5:8.- 382 .2: :8- 888. ..o 885 .05on 3.5% .n .8: .88.. .8: .8: 8.... 8.... .8: .2: 8.- 2.: :8. .8: 8. .2: .8- 8.: 8.- 25 S 8.: .. I. .8. 8:88. .8... .8. S .. :8 .8: .8. : .8: 8.. ...: 8.- 8.: 8. .2: 8... 5.7.8. .3: 8. 9% 5.3.8::th .v 8.... .2... .8: 8...: .8: .2... 8:88 8.: :8; .2: 2. 8.2.8.2 .2: :3.- .8: 8. S8. 8.: m... ..N: 2. 8.88m. 9.3:er .M .8: E: 8.: .8: 8.... .8: $8.. .2: 8.- .8: 8.- 3:8... 8.: 8. .8: :8: A5:... .8: 8.- 88: 8.- 888.. $32.30: .N .8: .8... .8... .8: .8... .8: .2: .8. .8: 8. 8m: :8. .8: 8.- .8: 8. 88.78. .880 8&8... .8. .8. : 8:88. _ N 88. : .8: S .8... .8: .8: .8: 8.- .8: 8. .8: .8. .8:..:8. .8: .8.- .8: 8.- 8< 29.39%.th .— 83... .885 =5 5... .885 ..5 s. «x .885 ..5 .louozi 29:8 :3. 8-88 8-88. 8.50:0”. 358...“: d. :20 .m> Btu-.3. 02: 0:... Ecom :0 50.8585 5.. 3.... :o 532%. ..o £0.52 5.82%”. 0.8.5.. 3:55:32 .0. 0...“..- Separate Dimensions of Social Life, Criminal vs. Civil Table 11 presents the results of five multinomial logistic regression models examining the relationships between the five dimensions of social life and the behavior of law. In this table, the comparisons are between criminal and civil referrals only. This permits both the quantitative and qualitative comparisons suggested by the theory: civil and criminal referrals represent two distinct styles of law (compensatory and penal, respectively) but also represent a quantitative distinction in that the theory suggests that a criminal referral is “more law” than a civil referral. Note that, as the model chi-squares represent overall model fit, the chi-square values in Table 11 are the same as those in Table 10, and thus have already been presented. The main interest is in the differences in the odds for the two types of referrals. In Model One for 1997—98, turning to the individual stratification variables, there are no differences for age, but the likelihood of receiving a criminal referral decreases by 42% for females, which is inconsistent with both Black's theory and conflict theory. For 2001—02, for every one—year decrease in age, there is 51 6% decrease in the odds of receiving a criminal referral, which is 183 inconsistent with Black’s theory and conflict theory. For gender, the likelihood of receiving a criminal referral increases by 92% for females, which is in the direction that both Black’s theory and conflict theory predict. In the full sample, only age is significant (p 5 .01), but the relationship is weak and does not hold at more stringent levels of significance. Moreover, the odds ratio is close to 1.0, indicating almost no difference in the odds of a criminal or civil referral for a one year decrease in age. Model Two examines the relationships between morphology and the likelihood of a criminal referral vs. a civil referral. There are no significant findings in 1997- 98. For 2001-02, employees are significantly more likely than managers or executives to receive a criminal referral. Specifically, the odds ratio of 2.92 shows an increase of 192%. This finding supports Black's theory, showing that those at lower levels of a hierarchy are dealt with more harshly by the legal system. The finding does not persist in the full sample, as there are no significant differences. Model Three examines the relationship between culture and the likelihood of a criminal referral vs. a civil referral. For 1997-8, there are no significant findings. 184 In 2001—02, for every one-unit increase in education, there is a 51% decrease in the odds of receiving a criminal referral, which is in the direction predicted by Black’s theory. Similar findings emerge for the full sample, where a one unit decrease in education corresponds to a 30% decrease in the odds of receiving a criminal referral. Model Four examines the relationship between organization (measure by size and type) and the type of referral. In 1997-98, an increase in the size of a business is associated with a 22% increase in the odds of a criminal referral, supporting Black's theory. No significant findings for the individual parameter of type are found. Ihi 2001-02, as well as the full sample, there are .no significant findings for‘ either‘ measure: of organization. Model Five examines the relationship lbetween. social control and the type of referral. The findings for 1997-98 show that a one unit increase in the degree of internal social control corresponds to a 27% decrease in the likelihood of a criminal referral, which is consistent with the theory’s jpredictions. Similarly; although. the direction of the relationship differs in 2001-02, there are statistically significant but contradictory findings, 185 indicating that a one unit increase in the degree of internal social control results in a 37% increase in the likelihood of a criminal referral. For the full sample, the findings are comparable to 2001-02 albeit somewhat weaker, showing that a one unit increase in the degree of internal social control corresponds to a 15% increase in the likelihood of a criminal referral. To obtain a more complete understanding of the relationships presented in Tables 10 and 11, a series of additional multinomial logistic regression models were estimated. For the full sample and for each time period, four models were estimated. Because these analyses included the addition of variables (i.e., controls) at subsequent stages, the model fit statistics can be compared to determine which model best fits the data. Additionally, the pseudo R? (Nagelkerke R2) change can also be examined to show the effect of additional variables on the level of explained variance. As Long (1997) cautions, given the nominal nature of the dependent variable, the level of explained variance is not as important as the classification table in assessing model fit and accuracy of prediction. The results from the classification tables will be discussed in text, but are not presented in the 186 MNLM tables. In situations where the distribution of the dependent variable is skewed with the majority of cases in one category, such as in criminal referrals, it is also useful to evaluate classification of cases against the classification that would be expected by chance. In each model, the most accurate predictions would be expected for the cases with the greatest frequency for the dependent variable (i.e., criminal referrals). The benchmark standard for evaluating model classification versus by chance classification is 21 25% improvement (Tabachnick and Fiddell, 2001). To assess classification accuracy, the classification standard will be examined for each model. Classification by chance, and the subsequent 25% improvement over chance, will be calculated as follows: Percentage of cases in category 1 (no referrals) squared + percentage of cases in category 2 (criminal referrals) squared + percentage of cases in category 3 (civil referrals) squared x 1.25. These results are presented below. For the full sample: .2612 + .6312 + .1052 = .47.7% by chance classification accuracy; x 1.25 = 56.9% minimum 187 For 2001-02: .2672 + .6732 + .0602 = 52.7% by chance classification accuracy; x 1.25 = 65.9% minimum For 1997-98: .2592 + .5742 + .1672 = 42.8% by chance classification accuracy; x 1.25 = 53.4% minimum To summarize, the minimum classification accuracy that would be expected by chance alone in the full sample is 47.7%. An improvement of 25% better than chance is found by multiplying 47.7% by 1.25, which yields a value of 56.9%; To determine whether* the additional multinomial logistic regression models classify cases at a rate better than 25%, the classification accuracy rate must be greater than 56.9%. For 2001-02, the minimum classification accuracy that would be expected by chance alone is 52.7%. An improvement of 25% better than chance is found by multiplying 52.7% by 1.25, which produces a value of 65.9%. To determine whether the additional multinomial logistic regression models classify cases at a rate better than 25%, the classification accuracy rate must be greater than 65.9%. 188 Finally, the calculations on the previous page showed that, for 1997-98, the minimum classification accuracy that would be expected by chance alone is 42.8%. An improvement of 25% better than chance is found by multiplying 42.8% by 1.25, which produces a value of 53.4%. To determine whether the additional multinomial logistic regression models classify cases at a rate better than 25%, the classification accuracy rate must be greater than 53.4%. 189 O.WQoo§no.W&:..c_.WQ§ 900.08.. a. mote $.00 .28 £085.88: 5 88.0 88.0.88 .8=0_0E0o0 805.08.65.82 08 8.5.5 ”mu-OZ 5...8... .2: .8: .3. A3.8.8 .8: .~.:...8. 5.28 5.. .2:..m.- 885.328.. .9589 3.5m. .m .8: .2: 8. 8m: .8: 8. .8. .2: 8.- 25 s 8.8 s :8... 5.8... .2: .8: E. .2: .8: 8.. .8: .~.:.o~. 88 =o.:c~.§uw.0 .v 5:8... 8: 83:8...- e...8.2 8.... .8: :8.- 52. .8: .2: 8.- 8825 0.5.396 s an. .8: 88 mm. 5:8... .8: .8: :8. c 2.. .2... .8: 2.- 888.. ..MoBauonN .88.. ..N: o..- .8: .8: .8. .8: 88.8.- 585 S ..om.w Soo..~ 81.x... .8. .8: .8.- .8: .8:...8.- .8. .8: 8. 8.. :o..30.\.:ah.m. .. . .0808. 5 m. ..5 .8 5 N. =5 .8 8 fix ..5 .8 .885 .885 .885 2.88 8... 8- 88 8-88. 3:5 :20 .m> 858:0: 0.2-. .Eoom ..o 82208.0 0.88% 8.. 33 :o ..o.>m..0m ..o 20.52 5.80.5: 0.8.3.. 3:85:22 .2 03......- 190 Multinomial Logistic Regression: Theoretical Models and Controls, Full Sample Table 12 presents the results of a series of four multinomial logistic regressions for the full sample. In these models, not referred is the reference category. Model One includes all five dimensions of social life as independent variables. Although some of the individual parameters are significant, the model chi-square is small and not statistically significant, showing poor model fit for the five dimensions of social life. Moreover, the Nagelkerke R2 indicates that a model representing Black’s theoretical categories explains only 2% of the variation in the outcome. The classification table statistics show that this model Ihas 51 63.5% overall accuracy' rate, which is greater than the minimum 25% improvement over chance of 59.6% that was calculated and presented on pages 182-183. The criteria for classification accuracy is satisfied in this model. In Model Two, type of fraud, a control variable, is added to the predictive model. The addition results in a larger‘ and statistically' significant. model chi-square «of 28.49 (p 5 .10) . However, Nagelkerke R2 remains the same, 191 indicating no improvement in the amount of explained variance with the addition of the type of fraud. The overall classification for the model is nearly equal at 63.4% which is greater than the minimum 25% improvement over chance of 59.6%. The criteria for classification accuracy is also satisfied in this model (see pages 182- 183) . Model Three includes all five dimensions of social life, but adds a different control variable: crime seriousness, measured by the dollar loss (square root transformation) due to the fraud case. Like the type of fraud, Black (1976) argues that indicators of crime seriousness should not matter in determining the behavior of law. Gender is the only individual parameter that is statistically significant, and the overall fit of model three fails to attain statistical significance. The chi- square of 22.53 is smaller than the chi-square for model two, although larger than model one. However, the Nagelkerke R2 increases to 3% in this model. As the theory predicts, the addition of crime seriousness alone does not significantly improve the fit of the model. Compared to the previous two models, the classification of 63.8% in model 192 three is higher than model one and model two, and also satisfies the criteria for classification accuracy. In Model Four, the five dimensions of social life are entered as independent variables, with both controls also included. Compared to the previous three models, the chi- square of 32.99 is the largest, and statistically significant at p 5 .05. Moreover, the Nagelkerke R2 of 5% is the highest of the four models. The overall rate of classification for the full model is 63.5%, which is greater than the 59.6% standard and again, satisfies the criteria for classification accuracy. Concerning the individual parameters, stratification is the only dimension of social life to show statistical significance. One variable, gender, is statistically significant for both civil vs. no referrals and criminal vs. no referrals. Holding all other variables constant, as Black’s theory specifies, the results show that the odds of receiving a criminal referral increase by 60% for females, while the odds of receiving a civil referral increase by 36% for females. These relationships are in the hypothesized direction of Black’s theory, as well as conflict theory, showing that those of lower stratification ranks are dealt with more severely by the legal system. In model four, the 193 onlqr other independent variables to attain statistical significance are the control variables. For type of fraud, these comparisons use fraudulent statements as the reference category, which is compared to asset ‘misappropriation and corruption, respectively. Compared to cases of fraudulent statements, the odds ratio of 3.33 shows that corruption cases are 233% more likely to receive a civil referral. Additionally, the odds of receiving a criminal referral are 122% greater for cases of asset misappropriation compared.tx> fraudulent statements. Crime seriousness, the second control variable, is statistically significant at the least strict criterion of p 5 .10 for civil vs. no referral and criminal vs. no referral; however the odds ratios indicate that the odds of either referral vs. no referral remain even. These findings for the control variables in model four are inconsistent with Black’s theory, but support previous critiques that offense specific factors such as the type of crime and the dollar loss of crime play a role in the legal response (Gottfredson and Hindelang, 1979). 194 .o. w as... ”.no. w 2 3 .o. . w R 8 00:80.3. u 320.2 .02 8.08.08... 5 82.8 8200 .85. 8080558.. 5 Echo c.8988 .8.=0_0E0ou “802.08.85.85. 08 805.5 ”ME-OZ 195 88. 88. N8. 88. w. 80.8....882 : 8888.88 .8 88.88 ..8 88.88 S. 88.2 8 N8. .882 88.82.. 8.82.. 88.888. 88.28. 888.3... 88.. - 88: .88... .88... .88... .88.: :88. .88: .88. .8388. .88: ...88. 8888888 .88: .28. .2: 3.. .88: :88... 883.5 .882 .8: .88... .88... .8:...88. .88: 88. .88: 2.. .88: 88. 88.2.82 .8... .88... .8... .88... .8: .88.. .8: .88.. .885 .88: 8.. .2: 88.- .88: 88. .2: 8.- .8: .8. .2: 8.- .8: .8. .2: 8.- 88.88.. .2... .88.. .8... .88.. .88... .88.. .88... .88... .88.: ... .2: 8.- .88: 8.. .2: 8.- .88: 88. .2: 8.- .88: 88. .2: 88.- 25 .88... .88.. .8... .88.. .8... .88... .88: .88.. .88: .8. .2: 88.- .2: 2.- .2: 8.- .88: 88. .2: 88.- .8: 88. .2: .8.- 8.8 .88.. .2... .8: .2: .88.. .2... .8... .2... .2:....- .8.: 8.. .2: 2.- .8.: 8.. .2: 2.- .2: 2. .8388.- .2: 2. 88888 .88.. .2: .88.. .2... .88.. .2... .88.. .2: .8.: 8. .88: 8.. .8... 88. .88: 8.. .2: 88.- .88: 2. .2: 8.- .88: 8.. 88.88.. .88... .88... .8... .88... .R: .88... .8... .88: .2: .88. .88: 8... .2:...88. .88: .8... .8.: 88. .88: .8... .8... 88. .8: .8... .885 .88.. .88... .88.. .88... .88... .88: .88.. .88... .8: 8..- .8: 88. .8: 8.- .8: 88. .8: 8.- .8: .8. .8: 8.- .88: 8.- 88< .885 .80 .885 .35 .885 .35 .885 .85 8 .88.). 8 .882 N .88.). . .882 0.95% :2“. .38... ..o .o.>8:0m ..o. 8.053. 5880.318. 0_.m.wo-. 8.50:...32 .N. 038-8 Multinomial Logistic Regression: Theoretical Models and Controls, 1997—98 Table 13 presents the results of a series of four multinomial logistic regressions for the 1997-98 sample. In these models, not referred is the reference category. Model One includes all five dimensions of social life as independent variables. The model chi-square of 25.97 is statistically significant (p 5 .05), and the Nagelkerke R2 indicates that approximately 6% of the variation in the type of legal response is explained by the combination of Black's five theoretical categories. Similar to findings for the full sample, classification for this model is 58.2%, which is greater than the 25% improvement over chance rate of 53.4%, satisfying the criteria for classification accuracy. Turning to the individual parameters, there are noteworthy findings for both measures of stratification. For age, results show that for each one—year increase in age there is a 2% decrease in the odds of receiving a criminal referral, with all other variables in the model held constant. For gender, the results show that females have a 75% greater chance of receiving a civil referral than males. The findings for age and gender are consistent 196 with both Black’s theory and conflict theory. In Model One, there are no significant parameters in the categories of morphology and culture. For organization, results indicate that an increase in the size of a business corresponds to a 24% increase in the odds of a criminal referral, consistent with the predictions of Black’s theory. For type of business, the second measure in the organization category, an increase in the direction of more organization is associated with a 20% decrease in the odds of a criminal referral, which is inconsistent with Black’s theory. Complimentary results emerge for the social control category: for each one unit increase in available internal social control, there is 23% decrease in the odds of a criminal referral. Model Two includes the five theoretical categories, but adds type of fraud as an additional variable. The model chi—square of 29.44 (p _<_ .05), is larger than this statistic for Model One, indicating that the addition of the type of fraud control variable significantly improves the fit of the model. Additionally, the level of explained variance increases in this model (Nagelkerke R2 = .07). The overall classification of the model is 58.9%, which is almost the same as Model One’s rate of 58.3% and also 197 greater than the 25% improvement over chance classification of 53.4% calculated on pages 182-183. Like Model One, however, the most accurate predictions of Model Two are for criminal referrals. For stratification, the results are similar to Model One. A one-year increase in age is associated with a 2% decrease in the odds of a criminal referral. Gender continues to exert a significant influence on the likelihood of a civil referral:- the odds of a civil referral increase by 80% for females, controlling for all other variables in the model. In other words, regardless of the type of fraud committed, this relationship persists, which supports Black's theory and conflict theory. In Model Two, the categories of morphology and culture fail to achieve statistical significance. - For organization, the type parameter is significant, and shows that less organized businesses correspond to a 19% decrease in the odds of a criminal referral, consistent with the theory’s predictions. For social control, a one-unit increase in available internal social control is associated with a 23% decrease in the odds of a criminal referral, which is in the direction predicted by the theory. Finally, although the addition of the type of fraud 198 variable significantly improved the fit of the model, none of the individual parameters for this variable are statistically significant . Model Three reports the results of the multinomial logistic regression with the addition of crime seriousness, the second control variable, to the five theoretical categories. The model chi-square of 24.40 is statistically significant, but is lower than Model One. The level of explained variance increases to 8% when crime seriousness is added to the model. The overall classification of 55.6% accuracy for this model is less than the previous two models, but still greater than the 25% improvement over chance rate of 53.4%. In Model Three, the stratification category includes significant findings for both age and gender. Like the previous models, an increase in age is associated with a slight decrease in the odds of a criminal referral (2%). For gender, females have an 87% greater chance of receiving a civil referral than males, all other variables held constant. In this model, no other parameters are statistically significant, leading to the conclusion that Black's theory performs well for only both measures of stratification. 199 l EFL. I Model Four presents a full model that adds both control variables to the five theoretical categories. Much like the results for the full sample, this model provides the best fit to the data, with a chi-square of 31.58 (p 5 .05) and a Nagelkerke R2 of .10. For the individual parameters consistent with Black's theory, stratification is the only category to include significant results. Specifically, as age increases by one year, the odds of a criminal referral increase slightly (by 2%). In this model, the parameters for gender are the strongest, showing that females have a 91% increase in the odds of a civil referral, all other variables held constant. The relationship holds regardless of the type of fraud committed or the dollar loss from the fraud. The only other variable to achieve statistical significance in this model is type of fraud, one of the controls. Recall that the comparison group is fraudulent statements. The findings show that the odds ‘of receiving a civil referral increase by 131% for cases of asset misappropriation, compared to cases of fraudulent statements. Additionally, for cases of corruption, there is a 110% increase in the odds of receiving a civil referral compared to cases of fraudulent statements. These 200 findings are inconsistent with Black’s theory, which would argue that the type of crime does not influence the behavior of law. For the full model, the overall classification rate is 57%, which is greater than the 25% improvement over chance classification accuracy of 53.4%. Like the previous three models, Model Four has the highest rate of prediction accuracy for criminal referrals, which was expected given the distribution of the dependent variable. Interestingly, this model also has the highest accuracy rates in predicting both civil referrals and no referrals (8.6% and 7.2%, respectively). This suggests that the addition of both control variables plays a role in improving the ability to predict civil referrals and no referrals. 201 _o. w 3.: ”.no. w a 2. 5.. w & _._ 60:20.32 n autumn.— 82 98.3.5 5 8:8 ago new 6805:22— E Echo Eat—Ea £56508 woumvcaccfimfi. 8a mum-saw ”uh-OZ 202 2. 3. S. 3. "a 3.83032 s. 32.: .333 t. 33.3 a. St 3.2 s.» .822 3.3% 2.32 835 N33 385.3: ms: - 83: 83.: 83.: 83.: 83 3. 83 3. 83 8. 83 8. 3892.3 $2 :3: :3: .8: G3 3.- 3:32.33 £3 3. s3 2: 839:8 .32 5...: E: :3: £3 2.. 33:33 :3 mm. :3 m: $2282 :.2 5.: :2 8:: :3 83.: :2 .8: .858 :3 :- :3 3. :3 2.. :3 2. 83 Km.- 83 8. $3 LN.- 83 8. 38.on :.2 :2 :.2 :2 :2 :2 82 :2 :3 2.- a: 2.- $3 2.- £3 8.. 5313 s: 2.- £3 .2- 83 a..- 25 83: :2 83: 82 :3: :3 :3: :2 a3 2. $3 3.- 33 2. $3 :.- :3 _N. :3 3.- :3:_~. :3 8.- 9% 5.: 5.: 5.: E: :3: :2: :32 5.: :3 2. G3 2. :3 2. G3 2. $3 3. £3 2. $3 3.- s3 3. 8:83”: $2 $2 32 a2 82 :2 :2 :8: 83 a..- s3 2.- 83 G.- :3 :.- $3 on- :3 8.- :3 2.. G3 3. Susan. $2 :3: 82 SW: :2 83.: :32 5.: :3 3.- :3 .8. :3 3.- $3 .3. :3 3.- :3 .2. s3 :.- 83 .3. .380 :3 :3 :3 $2 :3 :32 :3: :3: :3 .8.- 33 8.- :3 .3.- 33 3.- :3 2.3.- :3 8.- :3 2.3.- 33 8.- a? 3555 :5 335 :5 .235 :5 3:55 :5 v .032 m .032 N 33.2 _ 332 3-32 .33 .30 83205 to: 2252 coamflwom otflmoq 3:82:32 .2 2:3 Multinomial Logistic Reggession: Theoretical Models and Controls, 2001-02 Table 14 presents the results of a series of four multinomial logistic regressions for the 2001-02 sample. In these models, not referred is the reference category. Model One includes all five dimensions of social life as independent variables. The model chi-square of 37.19 is statistically significant (p 5 .01), and the Nagelkerke R2 indicates that approximately 9% of the variation in the type of legal response is explained by the combination of Black's five theoretical categories. For this model, the overall classification accuracy is 69.3%, which is greater than the 25% improvement over chance classification accuracy rate of 65.9%. For the individual parameters, the stratification category shows significant relationships for age. For every one year increase in age, there is a 6% increase in the odds of a civil referral, which is opposite of the theory’s predictions. Unlike the 1997-98 sample and the full sample, gender does not have a significant influence on the outcome variable in the 2001-02 sample. Neither morphology nor social control are statistically 203 significant. For culture, measured by education, results show that a one-unit increase in education (i.e., from a high school or less degree to a bachelor's degree or from a bachelor's degree to graduate degree) corresponds to a 31% decrease in the odds of a criminal referral. This relationship is in the direction predicted by Black’s theory and conflict theory. Only one other theoretical category includes a significant parameter. Within the category of organization, the results show that as the type of business becomes more organized, there is a 33% increase in the odds of a criminal referral, which supports the theory's predictions. Model Two adds the type of fraud to the five theoretical categories. The model chi-square of 44.50 is statistically significant and larger than this statistic for Model One, indicating that the fit of the model improves with the addition of the control variable, fraud type. Additionally, the amount of explained variance also increases, to 11%, in this model. Like the previous model, Model Two shows that age, a measure of stratification, exerts a significant influence on the outcome. For every one—year increase in age, the odds of a civil referral increase by 6%, all else constant. Culture is also 204 significant, indicating that a one unit increase in the level of education results in a 30% decrease in the odds of a criminal referral, consistent with Black's theory and conflict theory. For the category of organization, one measure, type, shows that an increase in the organization of a business is associated with a 36% increase in the odds of a criminal referral, which is consistent with the theory’s predictions. For social control, the results show that an increase in available internal social control results in a 16% increase in the odds of a criminal referral, which is contradictory to Black’s theory. Although the addition of the control variable type of fraud significantly improves the fit of the model, none of the individual parameters for this variable are statistically significant, which can be interpreted as partial ‘support for Black’s theory. The overall classification accuracy of this model is 69.5%, which is greater than the 25% improvement over chance classification accuracy rate of 65.9%. Model Three presents the results of the multinomial logistic regression adding the second control variable, crime seriousness. The model chi-square of 45.81 (p 5 .01) is slightly larger compared to both previous models. The 205 Nagelkerke R2 also increases to .12 in this model. Additionally, the overall correct classification of the model is 71.1%, which is greater than the 25% improvement over chance classification accuracy rate of 65.9%. For the individual parameters, results are fairly consistent compared to previous models. A one year increase in age results in a 6% increase in the odds of a civil referral, again in the opposite direction of the theory’s predictions. In this model, however, the category of morphology is significant, which differs from the two prior' models. Compared. to) 'managers and. executives, employees have 46% greater odds of receiving a criminal referral, which is consistent with Black’s theory. In this model, the significance of culture (education) disappears. For organization, type is significant, showing that there is a 37% increase in the odds of a criminal referral if the business is more organized, which supports Black’s theory. Like the previous model, the individual parameters for crime seriousness are not statistically significant although the addition of this information provides a better fit of the model to the data. The final, full model shown in Table 14, Model Four, adds both control variables to the reduced theoretical 206 model. The model chi-square of 54.60 (p _<_ .01) is larger than any previous model. Moreover, compared to the reduced theoretical model (Model One), the level of explained variance increases from 9% to 14%. The overall classification of the full model is 70.5%, slightly lower than the model adding only crime seriousness. Like all previous multinomial logistic regressions, the full model most accurately predicts criminal referrals, and poorly predicts non—referrals and civil referrals. Additionally, the overall classification accuracy is greater than the 25% improvement over chance classification accuracy rate of 65.9%. For the individual parameters, the findings generally resemble those in the previous models. For stratification, a one year increase in age results in a 6% increase in the odds of a civil referral, and there are no gender differences. Morphology is not significant in the full model, but a one unit increase in culture (education) corresponds to a 28% decrease in the odds of a criminal referral, which is in the opposite direction predicted by Black’s theory. For organization, the results show that greater organization of a business results in a 21% increase in the odds of a criminal referral, which is 207 consistent with Black's theory. However, for the category of social control, increased internal social control results in a 40% increase in the odds of a criminal referral, which is opposite of the theory's predictions. Finally, one other parameter in the model , asset misappropriation under type of fraud, shows that compared to cases of fraudulent statements, asset misappropriation cases have a 147% increase in the odds of a criminal referral. This finding does not support Black’s theory, because the type of fraud should not make a difference in the legal response. 208 3. w as: ".8. w a t 2. w a . .8652 u 85%. 32 .3323 5 8:2 33 23 3856.3 5 265 235.... £86208 3:33:22: 2.. 33m ”$02 :. 2. :. 3. Na 3.2.332 538.3 :33»? 338.3 a... 3.2.: as”: .302 2:3 2:3 8:: 3:: 32:93 :3 - 83: :3: :3: :3: :3 3. :3 8. :3 8. 83 8. 3058:»: :n: :3: :3: :3: :3 :N. :3 2. :3 3. :3 mm. 33.68 :t: :2 :5: :2 :3 .3. :3 .3 :3 mm. :3 8.- 22st :3: :2 :3: :2 .2: :2 :3: :2 .828 :3 .2. :3 :.- :3 :- :3 3.- 83.2. :3 2.- 83 z. :3 2.- 38:8 8...: :3: :n: 3.: 8:: :3: :n: :3: :3 :mm. :3 8. :3 :2. :3 3. :3 3:. :3 3. :3 :2. :3 3. 25 :2 :2 :2 :3: :2 :2 :2 :3: :3 2.- :3 3.- :3 2.- :3 8. :3 3.- :3 3.- :3 3.- :3 3. a: :2 :2 :2 :3: :2 :2 :2 :3: :3 .2. :3 8. :3 2...- :3 8. :3 :2.- :3 3. :3:::.- :3 8. 8:82: :3: :0: Gt: :3: 8:: :3: :1: :3: :3 3.. :3 3:. :3 ..mm. :3 3. :3 :. :3 3. :3 3. :3 cm. 828.: :22: :2 :3: :2 :3: :2 :0: :2 :3 mm. :3 3.- :3 3. :3 :.- :3 :N. :3 3.- :3 :N. :3 2.- .350 :3: :5: :3: :3: :3: $3.: :3: :3: :3 8. :3 :3. :3 8. :3 :3. :3 3. :3 :3. :3 3. :3 53. om< _EEco EC .235 :5 .235 :5 3555 :5 v 332 m .032 N .232 _ .322 No-58 .33 ..o to_>m:om .o: 2252 :23“:me ocflwoq 3:52:32 .3 Ban-_- 209 Conclusion The purpose of this study was twofold: to examine differences in individual and organizational characteristics for the three types of fraud, and to describe and explain the legal responses to occupational fraud cases by integrating Black’s theory of the behavior of law and conflict theory. For Analysis One , the hypothesis that the characteristics of individuals who commit asset misappropriation, corruption, and fraudulent statements will differ was supported. Moreover, the characteristics of the sample as a whole are similar to previous findings in the white-collar crime literature: namely, the average age is comparable to many previous studies relying on official sources of data (e.g., criminal records). The hypothesis that organizational characteristics associated with each type of occupational fraud would differ was also supported. One type of occupational fraud, corruption , tended to occur in organizations with characteristics that were similar to the corporate crime literature’s portrayal of a large, publicly traded business as the common setting for white-collar crime; however, an important difference is that prior studies depict this 210 setting as it relates to crime that benefits the organization, not victimization against it. For .Analysis Two, the results were somewhat mixed, supporting both theories for some models, and contradicting both theories in others. Like many previous studies, the individual theoretical categories were fairly weak predictors of the behavior of law when examined separately. This general finding persisted in comparisons to non- referrals, as well as in the comparisons between civil and criminal referrals. In the full sample, the reduced theoretical model had a poor, non-significant fit to the data. The addition of fraud type slightly improved model fit, but gender was the only variable to have a significant effect on the outcome, consistent with the theory's predictions. The addition of crime seriousness also improved model fit compared to the reduced theoretical model, although. not as much as the model adding type of fraud. The significance of gender persisted in the direction predicted by the theory. For the full sample, the best fitting model was the full model adding both type of fraud and crime seriousness, which is contrary to the theory. Like the prior models, females continued to have significantly increased odds of receiving 211 both criminal and civil referrals, all else constant, and regardless of both control variables. In the full model, the only other variable that significantly influenced the outcome was type of fraud, one of the controls. In sum, the findings for gender support both conflict theory and the theory of the behavior of law, although none of the other theoretical categories are significant. In 1997-98, overall model fits were improved compared to the full sample. The fit of the reduced theoretical model was significant, and, in addition to age, gender, and social control, size of business influenced the outcome in the direction predicted, although the significant findings for type of business were in the opposite directions. When fraud type was included, model fit was slightly improved, and the significance of gender persisted. Type of organization and social control were significant, but not in the expected direction. The addition of crime seriousness to the reduced theoretical model improved model fit, but not as much as the addition of only fraud type. In this model , both stratification measures were significant, but age was in the opposite direction. Only the findings for gender continued to support Black’s theory and conflict theory. In 1997-98, like the full sample, the 212 best fitting model included the addition of both control variables. In this model, however, only age and gender supported both theories. Additionally, type of fraud was significant in the full model, contrary to theory. In 2001-02, the best fitting' models were revealed. All of the reduced and full models significantly fit the data, but the poorest fit was the reduced theoretical model. However, across all models, age was significant, ibut in the opposite direction predicted by the theory. In 2001—02, however, the findings for gender exhibited in the previous samples (full sample and 1997-98) did not persist. Findings for education, the measure of culture, were in the expected direction across all models. The only other category to influence the outcome in the direction predicted by Black’s theory was type of business. Moreover, one of the control variables, type of fraud, specifically asset misappropriation compared to fraudulent statements, significantly influenced the outcomes, contrary to the theory’s predictions. The final chapter, Discussion, provides a short summary of the important findings. Implications for theory, policy, and future directions of research on 213 occupational fraud and the behavior of law are also discussed. 214 CHAPTER FIVE: DISCUSSION The last chapter has several objectives. Before delving into discussion for each separate analysis, a brief overview of the study's objectives is provided. Next, the results from this study are presented and discussed for Analysis One (comparison of individual and organizational characteristics for three types of fraud) and Analysis Two, (prediction of the response to fraud) respectively. Following the coverage of results from both analyses, the limitations of the study itself and of particular findings are examined. Chapter Five concludes with a consideration of this study’s implications for theory, policy, and practice. Study Purpose This study had two main objectives. The first goal was to improve the understanding of individual and organizational characteristics connected to the three types of occupational fraud (i.e., asset misappropriation, corruption, and fraudulent statements). A second goal of the research was to use an integration of Black’s (1976) 215 theory of the behavior of law and conflict theory to provide a description and explanation of the legal responses to occupational fraud. Analysis One: Key Findings The first aim of Analysis One was to determine whether individual characteristics differ significantly for the three types of occupational fraud. The second purpose of Analysis One was to determine whether organizational characteristics differ significantly for the three types of occupational fraud. The research established the extent to which this sample of individuals within organizations reflects existing, well-known images of white-collar crime. The findings from Analysis One confirm that the characteristics of individuals who commit asset misappropriation, corruption, and fraudulent statements differ, as do many of the organizational characteristics associated with these three types of occupational fraud. How do employees who commit occupational fraud differ from the popular portrayal of the typical white-collar offender represented le previous studies? Not all offenders conform to the popular image. The results paint 216 a clear picture of employees who commit fraudulent statements: they are more likely to be older, more educated males who hold managerial or executive positions in their victim organizations. This profile resembles that of past studies (Wheeler et al., 1988), and suggests that previous findings based on samples of convicted offenders are generalizable to broader settings. The other two types of occupational fraud, asset misappropriation and corruption, appear to be committed by individuals who share characteristics that are less representative of the white- collar crime prototype. These offenders are younger on average, equally likely to be male or female, and do not differ significantly from each other in their levels of education. These findings provide partial support for Daly’s (1989) study, which also showed distinct differences in the white-collar crimes of males and females, due in part to characteristics such as occupational roles and education. While asset misappropriation and corruption are equally likely to be committed by males and females, fraudulent statements are distinctly linked to being male, and are also tied to higher positions in organizations. Although 217 gender and position were significantly related, the correlation is weak (r = .12), providing only limited support for the possibility of a glass ceiling effect that keeps female employees in this sample from advancing to managerial or executive level positions. Additional research on gender differences in motivation for offending would facilitate understanding of this discrepancy. Qualitative interviews with both male and female offenders who commit occupational fraud would shed light on whether their motivations differ, as Collins and Collins (1999) found in their study of inmates serving federal prison sentences for white-collar offenses. How do organizations that are victimized by occupational fraud compare to the image of the large, profit—making organization that is frequently described in the literature? The results showed that the three types of fraud also differed with regard to characteristicsiof the organizations where they most often occurred. Asset misappropriation occurs significantly more often in smaller organizations. Corruption takes place in larger organizations, and in particular, publicly traded companies. More corruption is discovered in organizations 218 with anonymous reporting systems, and is also more likely to be revealed when organizations perform internal audits. Finally, fraudulent statements also tend to victimize smaller organizations, as well as privately held companies, and CR) not appear' to affect non-profit agencies, which would be expected given that this particular form of fraud often involves misrepresentation of a company's financial performance. Most remarkable about these findings is that they would not even be revealed if, consistent with prior research, only the large, publicly traded companies were sampled. Future studies of occupational fraud should continue to examine multiple types of organizations. More detailed comparisons, such as between industries, or within different cultural contexts, could also contribute to better understanding of the ways in which macro-level factors (e.g., competition) lend themselves to fraud, in some settings, but not others. Analysis Two: Key Findings The findings from the second set of analyses were consistent with prior studies that found mixed support for Black’s (1979) theory of the behavior of law. In the full sample, the reduced theoretical model (without any control 219 variables) failed to explain the likelihood of receiving criminal or civil referrals. The best fitting model included the type of fraud and crime seriousness, the two control variables. While this contradicted Black's theory, it supported common criticisms of the theory (Gottfredson and Hindelang, 1979). For the full sample, all models revealed that females were more likely to be referred for criminal processing, and also, for civil processing, than not referred. For the full sample, this was the only finding that supported Black’s theory and conflict theory, both of which suggest that females are lower in the stratification system and therefore more likely to be subjected to law. In 1997-98, the best fitting model again was the model that included both control variables. Females had an increased likelihood of receiving a civil referral rather than a non-referral, although this finding did not hold for criminal referrals. Older offenders were less likely to receive criminal referrals, which is in the direction predicted by Black’s theory and conflict theory. In the reduced theoretical model, criminal referrals were more likely in larger businesses, which supported Black’s theory. Criminal referrals were also less likely in more 220 organized businesses, contradicting the theory. These findings persisted when type of fraud, a control variable, was added to the model, but did not persist with only the addition of crime seriousness, or with both control variables added. In 1997-98, an increase in internal social control was associated with a decrease in the likelihood of a criminal referral; this relationship was found in the reduced theoretical model and also in the model including type of fraud. These findings support the theory’s propositions for the category of social control. In 2001-02, the best fitting model again included both control variables. Across all models, older offenders had an increased likelihood of a civil referral, which is in the opposite direction predicted by Black’s theory and conflict theory. However, this relationship did not hold for criminal referrals, supporting the idea that the style of law varies with age. Unlike the findings from 1997-98 and the full sample, in 2001-02 gender did not influence the likelihood of either form of referral. Another difference at this time period was that increases in education significantly decreased the likelihood of a criminal referral. This result is in the direction predicted by Black’s theory as well as conflict theory. 221 Across all 2001-02 models, as the type of business increased in the direction of more organization, so did the likelihood of a criminal referral, which supports Black's theory. The findings for social control were in the opposite direction: in the full model, and in the model adding the type of fraud, increased amounts of internal social control increased the likelihood of a criminal referral, which may reflect a trend toward “100% prosecution policies” in organizations. Additionally, type of fraud, a control variable, significantly affected the likelihood of a criminal referral, making it more likely for asset misappropriation compared to fraudulent statements. This finding also contradicted the theory’s statement that the type of offense should not influence the behavior of law. Potential Study Limitations The results of this study should be interpreted with caution given the limitations of the sample. As noted in Chapter Three, participants were asked to report on their most recent, completed fraud case. There is no guarantee that this suggestion was actually followed, which raises the concern that the responses may be unrepresentative of 222 the larger' population. of occupational fraud cases. However, as the descriptive statistics show, the percentages of each type of fraud are similar at each time period, as are individual and organizational characteristics. The consistency in the frequency of the variables across the two time periods suggests that potential sampling bias is not a serious concern. Despite the possible limitations, the sample used in the present study is an improvement over samples in previous research for several reasons. Much of what criminologists assume to be fact about the typical white- collar offender is based on samples of convicted individuals, which are far less representative of working employees at multiple levels and types of organizations who may commit occupational fraud. Although the sample is restricted. to jperpetrators whose offenses were detected, the sample size is larger than those in many' previous studies, and the information provided by CFEs_was obtained from several sources; both of these factors increase its external validity. 223 Implications for Theory This study contributes to the white-collar crime literature in several ways. First , it provides further insight on organizations as victims, rather than as perpetrators, of fraud. The inclusion of both individual and organizational characteristics responds to previous calls for research advocating the need for a micro-macro connection in studies of white-collar crime (Coleman, 1992; Vaughan, 1992). On a related front, this study also corrects for the common, business firm bias in corporate crime research by including organizations that differ in size as well as function. Future studies should continue tx: consider the characteristics of individuals and organizations not just for the types of fraud considered here, but also for other forms of white-collar crime. The differences between the present study’s results for Analysis One and those from earlier research may be due to the restrictions of the particular offense-based definitions of white-collar crime tluat are employed. Additional comparisons based on siJnilar; offense-based definitions of white-collar crime would contribute to convergent validity across existing and 224 future research. In other words, summarizing and comparing findings would be straightforward and more appropriate when similar operational definitions are employed. A question that remains unanswered in this study is: what causes occupational fraud? Because secondary data used in the current research were obtained from investigators, rather than perpetrators, it was not possible to speculate on the reasons individuals may have for committing asset misappropriation, corruption, and fraudulent statements. The issue of causation is important in white-collar crime research, and a greater, theoretically informed understanding of potential causal factors is warranted. For example, more research on offenders’ motivations needs to be conducted to determine what factors influence the decision to commit fraud. It is likely that existing organizational practices, organizational culture, and life situations unique to individual employees, contribute to victimization against organizations. For instance, feelings of perceived inequity or unfair treatment by employers may contribute to individuals’ belief that occupational fraud is justified (Greenberg, 1993, 1990). Due to the difficulties 225 associated with conducting white—collar crime research, a fuller and theoretically informed understanding of this phenomenon can only' be accomplished. in cooperation. with organizations. 131 particular, qualitative accounts from offenders and fraud investigators would be extremely beneficial in the generation of additional hypotheses and the development of grounded theories of motivation for white—collar offending. More joint ventures between researchers and organizations are vital for both the theoretical and practical sides of white—collar crime. The results from this study also have implications for further tests of Black’s (1976) theory of the behavior of law. Like many' prior studies, this study found. only partial support for the theory. However, also like outcomes from many prior tests, the results may be due to data and measurement issues. The five dimensions of social life may not be as mutually exclusive as Black argues. Additionally, the results also speak to weaknesses of the theory itself, many' of which have been pointed out in previous critiques. All of the indicators of Black’s dimensions of social life were merely proxy measures, and alternative indicators may improve prediction. With regard to data and measurement 226 concerns, it is possible that additional, unmeasured variables may explain the findings. For example, in the category of stratification, other indictors of stratification, such as race, ethnicity, and income, were not recorded in the ACFE’s surveys. As a result, it was not possible to examine whether there are significant differences for these measures. If race and ethnicity data were available, it is quite plausible that the contradictory findings for age in 2001—02 may be explained by these other variables. In 2001-02, an increase in age increased the likelihood of a referral, which is contradictory to Black’s theory as well as conflict theory. It may be that the older offenders have an increased likelihood of referrals because a larger proportion of them are minorities. Given the existing research showing that minorities are processed differently by the criminal justice system (see Zatz, 1987), it is also likely that perpetrators’ race and ethnicity may play a role in employers’ decisions to refer a case for civil or criminal processing. Similarly, it is also likely that offenders’ incomes may play a role in the employer’s decision to refer a case outside of the organization. Black’s theory and conflict 227 theory both suggest that higher income offenders, who enjoy a higher status in the stratification system, may be protected from legal processing. To provide a more complete test of Black’s theory, future studies should incorporate these additional measures of stratification. The category of morphology only included one variable: position in the organization. Additionally, the category of culture only included one variable: level of education. Both of these categories could be improved with the inclusion of additional variables. Rather than using income as a measure of stratification, income could be used as a measure of morphology since it would provide some indicator of vertical distance (e.g., those with lower incomes would be further away from the top of the organization in distance, and more likely to be referred). Likewise, the measurement of culture could be improved by incorporating the number of years an individual has been employed by the victim organization. Black’s theory would suggest that longer-term employees posses a greater amount of culture compared to shorter—term employees; as a result, longer—term employees, who are also trusted more, may be less likely to receive a referral. 228 The category of organization included two indicators: the size and type of the business. While these measures are consistent with the theory’s formulations, additional measures may also improve prediction. In particular, knowledge of the number of years a business had been operating could provide a better measure of organization. According to Black (1976: 93), more established businesses are more organized, which suggests that a greater likelihood of referrals should occur when a business has been operating for a longer period of time. The category of social control included one variable that reflected the degree of existing internal social control in the organizations used in this sample. However, all 1,142 organizations in this sample also had access to another form of social control given the fact that there was an investigation by a certified fraud examiner. It is therefore not possible to speculate about the behavior of law in organizations without access to this form of non- legal social control. This caveat is a liwdtation of the sample, rather' than. the 'theory’ itself, and. represents .a common generalizability issue in white-collar crime research. 229 To conduct a ‘more complete test of Black’s (1976) theory across organizations, future studies should include additional, improved measures of each theoretical category. Moreover, given the fact that the best fitting models from this study were those that added control variables, future theoretical explanations should be refined to shed light on why control variables significantly influence legal outcomes. The type of fraud made a difference in the likelihood of referrals, as did the dollar loss of the fraud. Much like the findings in studies of violent criminal victimization, the type of crime and the seriousness of the crime both matter, suggesting that the range of explanatory factors in the theory are too limited. Although this finding contradicts Black’s theory, it supports many' previous critiques (Braithwaite and. Biles, 1980; Gottfredson and Hindelang, 1979). With a few exceptions (e.g., Lessan and Sheley, 1992), most prior tests of Black’s theory have been limited to a single time period. In the present study, two time periods were included, and the best fitting models were revealed for the 2001-02 data. A greater percentage of cases (75.1%) were referred for either type of processing in 1997—98 than in 2001-02 (67.9%). However, between 1997-98 230 and 2001-02, the percentage of criminal referrals increased nearly 10% (from 57.4% to 67.3% respectively) while the percentage of civil referrals between the time periods decreased by nearly the same percentage (from 16.7% to 6%). The distribution of referrals in 1997-98 are consistent with Wells’ (1997) finding that organizations may be reluctant to pursue criminal processing, which has a higher burden of proof, in favor of civil processing, which may be more successful. Between the two time periods, increased publicity from scandals in corporations such as Enron and Worldcom may have influenced organizations’ decisions to pursue a criminal prosecution, which would explain the changes in the distribution of referrals. As Black (1976: 3) argues, “the quantity of law varies across the centuries, decades and years, months and days, even the hours of a day. It varies across societies, regions, communities, neighborhoods, families, and relationships of every kind.” The differences in results at each time period support Black’s general statements. Specifically, the quantity of law increased in the direction of criminal law, which also reflects a difference in the quality of law at the two time periods. Given these results, future tests of the theory should consider 231 different time periods , perhaps even from different generations, if at all possible. It would also be beneficial for researchers to incorporate a measure of public attitudes toward the punishment of white-collar crime, to place the time differences in a broader, societal context (Cullen et. al, 1983) Implications for Policy and Practice The results also jprovide important implications for the practical side of white-collar crime prevention. Specifically, organizations that make anonymous reporting mechanisms (e.g., fraud hotlines) available to their employees can increase the likelihood that fraud is detected” Other procedures, snufll as internal audits, are also valuable tools for discovering fraud from within. It should. be noted, however, that the causal direction. of these relationships could. not be determined. in. Analysis One. In future research, it will be important to establish temporal ordering; Such a test would likely require a pre- and post-test design whereby baseline levels of fraud in an organization are established, an anonymous reporting mechanism is implemented, and a subsequent examination of fraud detection is conducted. 232 An unexpected finding 1J1 the analyses of organizational characteristics was tflua fact that, although background checks are in place in organizations, there are no significant effects.4 Albrecht (2003: 43) refers to this problem as a non-control factor. In other words, it is not a lack of control mechanisms that allow frauds to be perpetrated, but the fact that existing controls are either overridden or ignored. They may be in place in organizations, but are not consistently followed. This is somewhat surprising given the bureaucratic, documented rule orientation that characterizes government agencies as well as large organizations (Weber, 1946). Additionally, Albrecht (2003) proposes that background checks may not be successful due to another non-control factor, the failure to discipline fraud perpetrators. Rather than investing in what may be a costly prosecution, and coping with fallout from negative publicity, many organizations currently deal with their criminal employees through dismissal. This strategy is problematic, because 4 Similar to the findings for anonymous reporting mechanisms, the existence of background checks and their causal relationships with fraud could not be assessed. It may be possible that background checks are in place in response to prior cases of fraud; if they were not available, perhaps an even greater amount of fraud would occur given more lenient hiring practices. The pre- and post- test design previously described would also address the issue of causality. 233 it makes it possible for offending employees to build an impressive work history that is not marred by a criminal record, contributing txa additional, more financially devastating frauds at each new job. If these offenders really do have the most to lose, as specific deterrence theory suggests, organizations would clearly benefit from implementing a policy of prosecuting fraud cases to the highest possible extent that the law allows. 234 REFERENCES Albonetti, Celesta A. 1999 The avoidance of punishment: A legal- bureaucratic model of suspended sentences in federal white-collar cases prior to the federal sentencing guidelines. Social Forces 78:303- 329. Albonetti, Celesta A. 1998 Direct and indirect effects of case complexity, guilty pleas, and offender characteristics on sentencing for offenders convicted of a white- collar offense prior to sentencing guidelines. Journal of Quantitative Criminology 14:353-378. Aldrich, John H., and Forrest D. Nelson 1984 Linear Probability, Logit, and Probit Models. Thousand Oaks, CA: Sage. Alexander, Cindy F. 1999 On the nature of the reputational penalty for corporate crime: Evidence. Journal of Law and Economics 42:489-526. Alford, C. Fred 1999 Whistle-blowers: How much we can learn from them depends on how much we can give up. The American Behavioral Scientist 43:264-277. Baucus, Melissa S., and David A. Baucus 1997 Paying the piper: An empirical examination of longer-term financial consequences of illegal corporate behavior. Academy of Management Journal 40:129-151. 235 Baucus, Melissa S., and Janet P. Near 1991 Can illegal corporate behavior be predicted? An event history analysis. Academy of Management Journal 34:9-36. Benson, Michael L., Maakestad 1990 Francis T. Cullen, and William J. Local prosecutors and corporate crime. Crime and Delinquency 36:356-372. Benson, Michael L., Cullen, 1988 William J. Maakestad, Francis T. and Gilbert Geis District attorneys and corporate crime: Surveying the prosecutorial gatekeepers. Criminology 26:505-517. Benson, Michael L., and Elizabeth Moore 1992 Are white-collar and common offenders the same? An empirical and theoretical critique of a recently proposed general theory of crime. Journal of Research in Crime and Delinquency 29:251-272. Benson, Michael L., and Esteban Walker 1988 Sentencing the white-collar offender. American Sociological Review 53:294-302. Bickman, Leonard, and Debra J. Rog (Eds). 1998 Handbook of Applied Social Research Methods. Thousand Oaks, CA: Sage. Biderman, Albert D., and Albert J. Reiss 1968 On exploring the ‘dark figure’ of crime. Annals of the American Academy of Political and Social Science 374:1-15. 236 Black, Donald 1976 The Behavior of Law. San Diego, CA: Academic Press. Black, William Kurt 1998 The Best Way to Rob a Bank is to Own One. Unpublished Dissertation, University of California, Irvine. Braithwaite, John 1982 Enforced self regulation: A new strategy for corporate crime control. Michigan Law Review 80:1466-1507. Braithwaite, John, Brent Fisse, and Gilbert Geis 1987 Covert facilitation and crime: Restoring balance to the entrapment debate. Journal of Social Issues 43:5—41. Calavita, Kitty, and Henry N. Pontell 1990 “Heads I win, tails you lose”: Deregulation, crime, and crisis in the savings and loan industry. Crime and Delinquency 36:309-341. Calavita, Kitty, and Henry N. Pontell 1990 "Other People's Money" revisited: Collective embezzlement in the savings and loan and insurance industries. Social Problems 38:94-112. Cavender, Gray, and Aogan Mulcahy 1998 Trail by fire: Media constructions of corporate deviance. Justice Quarterly 15:697-719. 237 Chirayath, Verghese, Kenneth Eslinger, and Ernest De Zolt 2002 Differential association, multiple normative standards, and the increasing incidence of corporate deviance in an era of globalization. Journal of Business Ethics 41:131-140. Clark, John P., and Richard C. Hollinger 1980 A research report: Theft by employees. Security Management 24:106-110. Clarke, Lee, and James F. Short 1993 Social organization and risk: Some current controversies. Annual Review of Sociology 19:375-399. ‘ Clinard, Marshall, and Peter Yeager 1980 Corporate Crime. New York: The Free Press. Coleman, James W. 1987 Toward an integrated theory of white-collar crime. American Journal of Sociology 93:406-439. Coleman, James W. 1985 The Criminal Elite: The Sociology of White- Collar Crime. New York: St. Martin’s Press. Cressey, Donald R. 1953 Other People's Money: A study in the Social Psychology of Embezzlement. Glencoe, IL: Free Press. Cullen, Francis T., and Bruce G. Link 1980 Crime as an occupation: A new look. Criminology 18:399-410. 238 Cullen, Francis T., Bruce G. Link, and Craig W. Polanzi 1982 The seriousness of crime revisited: Have attitudes toward white-collar crime changed? Criminology 20:83-102. Cullen, Francis T., Richard A. Mathers, Gregory A. Clark, and John B. Cullen 1983 Public support for punishing white-collar crime: Blaming the victim revisited? Journal of Criminal Justice 11:481-493. Dalton, Dan R., and Idalene Kesner 1988 On the dynamics of corporate size and illegal activity: An empirical assessment. Journal of Business Ethics 7:861-870. Davis, Robert C., Arthur J. Lurigio, and Wesley G. Skogan (Eds.) 1997 Victims of Crime.(:2“d Edition). Thousand Oaks, CA: Sage. Eitle, David J. 2000 Regulatory justice: A re-examination of the influence of class position on the punishment of white-collar crime. Justice Quarterly 17:809- 839. Elis, Lori A., 1995 and Sally S. Simpson Informal sanction threats and appeals to morality—additive versus multiplicative models. Journal of Research in Crime and Delinquency 32:399-424. Emerson, Robert M. 1991 Case processing and interorganizational knowledge: Detecting the “real reasons” for referrals. Social Problems 38:198-212. 239 E rmann , M. David, and Richard J. Lundman (Eds). 1996 Corporate and Governmental Deviance: Problems of Organizational Behavior in Contemporary Society. New York: Oxford University Press. Ermann, M. David, and Richard J. Lundman 1978 Deviant acts by complex organizations: Deviance and social-control at organizational level of analysis. Sociological Quarterly 19:55-67. Evans, Sandra 8., and Richard J. Lundman 1983 Newspaper coverage of corporate price—fixing. Criminology 21:529-541. Farberman, Henry 1975 A criminogenic market structure: The automobile industry. Sociological Quarterly 16:438-457. Fisse, Brent 1983 Reconstructing corporate criminal law: Deterence, retribution, fault, and sanctions. Southern California Law Review 56:1141-1246. Fisse, Brent 1981 Community service as a sanction against corporations. Wisconsin Law Review 5:970-1070. Fleckenstein, Marilynn P., and John C. Bowes 2000 When trust is betrayed: Religious institutions and white-collar crime. Journal of Business Ethics 23:111-115. Geis, Gilbert 1982 On White-Collar Crime. Lexington, MA: Lexington Books. 240 Geis, Gilbert, and Joseph F.C. DiMento 2002 Empirical evidence and the legal doctrine of corporate criminal liability. American Journal of Criminal Law 29:341-375. Geis, Gilbert, and Stotland, Ezra(Eds). 1980 White-collar Crime: Theory and Research. Thousand Oaks, CA: Sage. Gottfredson, Michael R., and Travis Hirschi 1990 A general theory of crime. Palo Alto, CA: Stanford University Press. Greenberg, Gerald 1993 Stealing in the name of justice: Informational and interpersonal moderators of theft reactions to underpayment inequity. Organizational Behavior and Human Decision Processes 54:81-103. Greenberg, Gerald 1990 Employee theft as a reaction to underpayment inequity: The hidden costs of paycuts. Journal of Applied Psychology 75:561-568. Greenberg, Gerald 1987 A taxonomy of organizational justice theories. Academy of Management Review 12:9-22. Groves, W. Byron, and Graeme Newman (Eds). 1986 Punishment and Privilege. Albany, NY: Harrow and Heston. Hagan, John, and Fiona Kay 1990 Gender and delinquency in white-collar families: A power-control perspective. Crime and Delinquency 36:391-407. 241 Hagan, John L., and Ilene H. Nagel 1982 White-collar crime, white-collar time: The sentencing of white-collar offenders in the Southern district of New York. American Criminal Law Review 20:258-289. Hawkins, Keith, and John M. Thomas (Eds). 1984 Enforcing Regulation. Law in Social Context Series. Boston, MA: Kluwer-Nijhoff. Heimer, Carol A. 1988 Social structure, psychology, and the estimation of risk. Annual Review of Sociology 15: 491-519. Herbert, Carey, Gary S. Green, and Victor Larragoite 1998 Clarifying the reach of a general theory of crime' for organizational offending: A comment on Reed and Yeager. Criminology 36:867-883. Hollinger, Richard C., and John P. Clark 1983 Deterrence in the workplace: Perceived certainty, perceived severity, and employee theft. Social Forces 62:398-418. Jesilow, Paul 1982 Adam Smith and white-collar crime—some research themes. Criminology 20:319—328. Jesilow, Paul, Gilbert Geis, and Henry Pontell 1991 Fraud by physicians against Medicaid. Journal of the American Medical Association 23:3318-3322. Kahan, Dan M., and Eric A. Posner 1999 Shaming white-collar criminals: A proposal for reform of the federal sentencing guidelines. Journal of Law and Economics 42: 365-391. 242 Kesner, Idalene F., Victor, Bart, and Bruce T. Lamont 1988 Board compensation and the commission of illegal acts: An investigation of Fortune 500 companies. Academy of Management Journal 29:789-799. Klein, Rochelle Lee; Bigley, Gregory A., and Karlene H. Roberts 1995 Organizational culture in high reliability organizations: An extension. Human Relations 48:771-793. Kramer, Ronald C., Raymond J. Michalowski, and David Kauzlarich 2002 The origins and development of the concept and theory of state-corporate crime. Crime and Delinquency 48: 263-282. Krantaz, Matt 2003 Experts say CEOs’ entrepreneur traits may lead to trouble. USA Today, May 12, pg. 18. Lee, M. T., and M. David Ermann 1999 Pinto “madness” as a flawed landmark narrative: An organizational and network analysis. Social Problems 46:30-47. Levi, Michael .1987 Regulating Fraud: White-Collar Crime and the Criminal Process. London: Tavistock. Levi, Michael, and S. Jones 1985 Public and police perceptions of crime seriousness in England and Wales. British Journal of Criminology 25:234-250. 243 Lynch, Michael J., Mahesh K. Nalla, and Keith W. Miller 1986 Cross-cultural perceptions of deviance: The case of Bhopal. Journal of Research in Crime and Delinquency 26:7-35. Metzger, Michael B., and Charles R. Schwenk 1990 Decision making models, devil’s advocacy, and the control of corporate crime. American Business Law Journal 28:323-377. McCarthy, Bill, and John Hagan 2001 When crime pays: Capital, competence and criminal success. Social Forces 79: 1035-1059. Mitchell, Terence R., Denise Daniels, Heidi Hopper, Jane George-Falvy, and Gerald R. Ferris 1996 Perceived correlates of illegal behavior in organizations. Journal of Business Ethics 15:439-455. Moore, Elizabeth, and Michael Mills 1990 The neglected victims and unexamined costs of white-collar crime. Crime and Delinquency 36:408-418. Nagel, Ilene H., and John L. Hagan 1981-2 The sentencing of white-collar criminals in federal courts: A socio-legal exploration of disparity. Michigan Law Review 80:1427-1465. Needleman, Martin L., and Carolyn Needleman 1979 Organizational crime: Two models of criminogenesis. The Sociological Quarterly 20:517-528. O’Donnell, Jayne 2003 Highflying penny-stock promoter now man behind bars. USA Today, May 12, pg. 3B. 244 O’Donnell, Jayne, and Richard Willing 2003 Prison time gets harder for white-collar crooks. USA Today, May 12, pg. 1A. Paternoster, Ray, and Sally 8. Simpson 1993 Sanction threats and appeals to morality: Testing a rational choice model of white-collar crime. Law and Society Review 30:549-83. Perrow, Charles 1984 Normal Accidents: Living with High Risk Technologies. New York: Basic Books. Punch, Maurice. 1999 Tackling business crime within companies. Security Journal: 39-52. Reed, Gary E., and Peter Cleary Yeager 1996 Organizational offending and neoclassical criminology: Challenging the reach of a general theory of crime. Criminology 34:357-381. Reiss, Albert J. 1984 Selecting strategies of social control over organizational life. In Enforcing Regulation, edited by Keith Hawkins and John M. Thomas. Boston: Kluwer-Nijhoff. Schlegel, Kip, and David Weisburd (Eds). 1992 White-Collar Crime Reconsidered. Boston: Northeastern University Press. Shapiro, Susan P. 1990 Collaring the crime, not the criminal: Reconsidering the concept of white-collar crime. American Sociological Review 55:346-65. 245 Shapiro, Susan P. 1987 The social control of interpersonal trust. American Journal of Sociology 93:623-658. Short, James F., and Lee Clarke (Eds). 1992 Organizations, Uncertainties, and Risk. Boulder, CO: Westview Press. Short, James F., and Laura Shill Schrager 1978 Toward a sociology or organizational crime. Social Problems 25: 407-419. Simpson, Sally S. 2002 Corporate Crime, Law and Social Control. Cambridge, UK: Cambridge University Press. Simpson, Sally S. 1987 Cycles of illegality: Antitrust violations in corporate America. Social Forces 65:943-963. Simpson, Sally S. 1986 The decomposition of antitrust: Testing a multilevel, longitudinal model of profit-squeeze. American Sociological Review 51:859-875. Simpson, Sally S., and Christopher S. Koper 1992 Deterring corporate crime. Criminology 30:347- 375. ' Snider, Laureen 1990 Cooperative models and corporate crime: Panacea or cop-out? Crime and Delinquency 36:373-390. 246 Staw, Barry M., and Eugene Swajkoski 1975 The scarcity munificence component of organizational environments and the commission of illegal acts. Administrative Science Quarterly 20:345-354. Stone, Christopher D. 1975 Where the Law Ends: The Social Control of Corporate Behavior. New York: Harper and Row. Sutherland, Edwin H. 1961 White-Collar Crime. Forward by Donald R. Cressey. New York: Holt, Rinehart and Winston. Sutherland, Edwin H. 1949 White—Collar Crime. New York: Dryden Press. Sutherland, Edwin H. 1939 Address to the American Sociological Society. Swajkowski, Eugene 1992 Accounting for organizational misconduct. Journal of Business Ethics 11:401-411. Tillman, Robert, and Michael Indergaard 1999 Field of schemes: Health insurance fraud in the small business sector Social Problems 46:572-590. Tillman Robert, Kitty Calavita, and Henry N. Pontell 1997 Criminalizing white-collar misconduct: Determinants of prosecution in savings and loan fraud cases. Crime, Law and Social Change 26:53- 76. Tillman, Robert, and Henry N. Pontell 1985 Organizations and fraud in the savings and loan industry. Social Forces 73:1439-1463. 247 Tonry, Michael, and Albert J. Reiss, Jr. (Eds). 1993 Beyond the law: Crime in complex organizations. Chicago: University of Chicago Press. Trevino, Linda Klebe, and Bart Victor 1992 Peer reporting of unethical behavior: A social context perspective. Academy of Management Journal 35:38-64. Van Maanen, John, and Edgar H. Schein 1979 Toward a theory of organizational socialization. Pp. 209-264 in Research in organizational behavior. Vaughan, Diane 1999 The dark side of organizations: Mistake, misconduct, and disaster. Annual Review of Sociology 25:271-305. Vaughan, Diane 1998 Rational choice, situation action, and the social control of organizations. Law and Society Review 32:23-61. Vaughan, Diane , 1997 The trickle-down effect: Policy decisions, risky work, and the Challenger tragedy. California Management Review 39:80-103. Vaughan, Diane 1996 The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA. Chicago: University of Chicago Press. 248 Vaughan, Diane 1990 Autonomy, interdependence, and social control: NASA and the space shuttle Challenger. Administrative Science Quarterly 35:225-257. Vaughan, Diane 1983 Controlling Unlawful Organizational Behavior: Social Structure and Corporate Misconduct. Chicago: University of Chicago Press. Walsh, James P., and James K. Seward 1990 On the efficiency if internal and external corporate control mechanisms. Academy of Management Review 15:421-458. Weick, Karl E., and Karlene H. Roberts 1993 Collective mind in organizations: Heedful interrelating on flight decks. Administrative Science Quarterly 38:357-381. Weisburd, David, Elin Waring, and Ellen Chayet 1995 Specific deterrence in a sample of offenders convicted of white-collar crimes. Criminology 33:587-598. Weisburd, David, Elin Waring, and Ellen Chayet 1990 White-collar crime and criminal careers: Some preliminary findings. Crime and Delinquency 36:342-355. Wheeler, Stanton, David Weisburd, and Nancy Bode 1982 Sentencing the white-collar offender: Rhetoric and reality. American Sociological Review 47:641-59. 249 Will, Susan, Henry N. Pontell, and Richard Cheung 1998 Risky business revisited: White-collar crime and the Orange County bankruptcy. Crime and Delinquency 44:367-387. Williams, Robert J., and J. Douglas Barrett 2000 Corporate philanthropy, criminal activity, and firm reputation: Is there a link? Journal of Business Ethics 26: 341-351. Yeager, Peter Cleary, and Gary E. Reed 1998 Of corporate persons and straw men: A reply to Herbert, Green, and Larragoite. Criminology 36:885-897. 250 u((lggmnmwpn((31131: