Discussion

An Integrative Theory Addressing Cyberharassment in the Light of Technology-Based Opportunism

PAUL BENJAMIN LOWRY, JUN ZHANG, GREGORY D. MOODY, SUTIRTHA CHATTERJEE, CHUANG WANG, AND TAILAI WU

PAUL BENJAMIN LOWRY ([email protected]) is the Suzanne Parker Thornhill Chair Professor and Eminent Scholar in Business Information Technology at the Pamplin College of Business at Virginia Tech. He received his Ph.D. in Management Information Systems from the University of Arizona. His research interests include organizational and behavioral security and privacy; online deviance, online harassment, and computer ethics; human-computer interaction, social media, and gamification; and business analytics, decision sciences, innovation, and supply chains. Dr. Lowry has published over 120 journal articles in Journal of Management Information Systems (JMIS), MIS Quarterly, Information Systems Research, Journal of the AIS, and other journals. He is a member of the Editorial Board of JMIS, department editor at Decision Sciences Journal, and senior or associate editor of several other journals. He has also served multiple times as track co-chair at the International Conference on Information Systems, European Conference on Information Systems, and Pacific Asia Conference on Information Systems.

JUN ZHANG ([email protected]; corresponding author) is an assistant professor in MIS at the International Institute of Finance, School of Management, University of Science and Technology of China. He holds a Ph.D. in Information Systems from City University of Hong Kong. His research centers on online deviant behaviors, informa- tion privacy and security, and IT-enabled health behavior change. His work has been published in such journals as Journal of Management Information Systems, Information Systems Research, and Computers in Human Behavior.

GREGORY D. MOODY ([email protected]) is a Lee Professor of Information Systems in the Lee Business School at the University of Nevada, Las Vegas and Director of the Graduate MIS program. He holds a Ph.D. from University of Pittsburgh and a Ph.D., from University of Oulu, Finland. His interests include IS security and privacy, e-business (electronic markets, trust) and human–computer interaction (Web site browsing, entertainment). Dr. Moody has published in Journal of Management Information Systems, MIS Quarterly, Information Systems Research, Journal of the AIS and other journals. He is an associate editor of Information Systems Journal and associate editor of AIS Transactions on Human-Computer Interaction.

SUTIRTHA CHATTERJEE ([email protected]) is an associate professor at the University of Nevada, Las Vegas. His research interests are ethical issues in

Journal of Management Information Systems / 2019, Vol. 36, No. 4, pp. 1142–1178.

Copyright © Taylor & Francis Group, LLC

ISSN 0742–1222 (print) / ISSN 1557–928X (online)

DOI: https://doi.org/10.1080/07421222.2019.1661090

IS, IT-enabled innovation, mobile work, and e-commerce. His research has been published in such journals as Journal of Management Information Systems, MIS Quarterly, Journal of the AIS (JAIS), and others. Dr. Chatterjee serves as a senior editor of JAIS and an associate editor of Information Systems Journal. He has chaired/co-chaired tracks or mini-tracks at the Hawaii International Conference on System Sciences and other IS conferences.

CHUANG WANG ([email protected]) is an associate professor at School of Business Administration, South China University of Technology. Her research focuses on the challenge and negative issues of information technology, and on social media, social networks, and mobile commerce. She has published in such journals Journal of Management Information Systems, Information Systems Research, Journal of the AIS, and others.

TAILAI WU ([email protected]) is a lecturer at the School of Medicine and Health Management in Tongji Medical College, Huazhong University of Science and Technology, China. His research interests are in medical informatics and human-computer interaction. His work has been published in several journals including Journal of Management Information Systems, Information Development, Journal of Medical Internet Research, and International Journal of Medical Informatics, among others, and appears in several preeminent IS conference proceedings.

ABSTRACT: Scholars are increasingly calling for a deeper understanding of cyber- harassment (CH) with the goal of devising policies, procedures, and technologies to mitigate it. Accordingly, we conducted CH research that (1) integrated social learning theory (SLT) and self-control theory (SCT); (2) empirically studied this model with two contrasting samples, experienced cyberharassers and less experi- enced cyberharassers; and (3) conducted post hoc tests to tease out the differ- ences between the two samples. We show that for less experienced cyberharassers, CH is largely a social-psychological-technological phenomenon; whereas, for experienced cyberharassers, CH is primarily a psychological- technological phenomenon. Our study makes a threefold contribution: (1) it shows the value of integrating two theories in a holistic and parsimonious manner to explain CH; (2) it shows that SCT alone is a more relevant framework for experienced cyberharassers, whereas a combination of SCT and SLT better explains less experienced cyberharassers; and (3) it reveals that the role of technology in fostering CH is crucial, regardless of the sample. The differential, yet consistent, findings demonstrate that addressing CH is contingent upon not only identifying theoretical approaches but also identifying the particular sam- ples to which these theoretical approaches will be more suitable. Of several implications for practice, the most important may be that anonymity, asynchro- nicity, and lack of monitoring are the technology choices that foster CH, and thus these should be mitigated in designing social media and other communication technologies

KEY WORDS AND PHRASES: cyberharassment, deviance, self-control theory, social learning theory, technology-based opportunism, social-psychological-technological (S-P-T) phenomenon, online harassment.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1143

The first federal prosecution of cyber harassment in the United State was in June 2004 when 38-year-old James Robert Murphy … pleaded guilty to two counts of the use of a Telecommunications Device (the Internet) with Intent to Annoy, Abuse, Threaten or Harass. … Murphy was sending anonymous and uninvited emails to Seattle resident Joelle Ligon and to her co-workers … Murphy also had special email programs in order to hide his identity and he created the “Anti Joelle Fan Club” (AJFC) and repeatedly sent threatening emails from this alleged group … [The judge] noted that he was surprised that Murphy “made no effort to indicate your remorse to the victim, to indicate you were sorry.” … [The] increas[ing number] of people managing several aspects of their lives online, both at work and in their personal lives … has created a vulnerability that attracts criminals, including cyber- stalkers, webcam blackmailers and identity thieves … A quarter of the American population has been bullied, harassed or threatened online and that number almost doubles for those under the age of 35 … [81].

Introduction

As the opening vignette illustrates, technology has provided access to a sociotechnical environment fraught with risks, which often provide new avenues for deviant behaviors [26, 99]. Cyberharassment (CH) is a prominent category of such deviant behaviors [33], which can be defined as:

Engaging in an act or behavior that torments, annoys, terrorizes, offends, or threatens an individual via email, instant messages, or other means [e.g., social media] with the intention of [physically or mentally] harming that person [45, p. 157].

Basically, CH is behaving offensively toward another via online communication technologies with the intention of causing harm [67]. The seeming ubiquity of CH and “its disproportionate toll on vulnerable popula-

tions (e.g., children and sexual minorities), the link with [extreme outcomes such as] suicidality, and the expected continued rise in Internet penetrance and connec- tivity make confronting it an urgent matter” [1, p. 10]. Thus, there is an increasing need to design principles, technologies, and policies to combat CH [72]. Constructing these preventive measures necessitates that researchers devote greater attention to multiple aspects of CH, including gaining a better understanding of (1) the nature of the phenomenon, (2) its antecedents and causes, and (3) methods to mitigate it [99]. These observations have inspired the core objective of this study to better understand CH as a sociotechnical phenomenon in which both human and technical elements co-create the phenomenon. Our motivation stems from a fundamental concern that CH has not been investigated using a compelling sociotechnical theory [35, 124]. Our extensive review of CH-related

1144 LOWRY ET AL.

interdisciplinary studies (see Online Supplemental Appendix C) highlights three fundamental objectives that motivate our research objectives. The first is the need for a stronger framework to understand CH. Following

our review presented in Online Supplemental Appendix C, factors related to CH can be grouped into four main categories: social factors (S) (e.g., peer association, social influence, and social pressure), psychological factors (P) (e.g., moral beliefs, motivation, and attitude), technological factors (T) (e.g., anonymity, online disin- hibition, and technology CB prevention capability), and other individual character- istics and demographic variables (e.g., gender, Internet skills, time spent on Internet, empathy, and experience). Studies mainly look at either S (majority), P (some), and T (rarely), or at most two from this subset (mostly S and P). Not one study has investigated a holistic S-P-T framework of CH. However, our open- ing vignette reveals that social, psychological, and technological considerations are all salient to investigate CH. The second is the need for the inclusion of a compelling set of technological

factors. Considering the unique threats and vulnerabilities pervading cyberspace, the sociotechnical approach does not merely treat CH as an extension of offline harassment, thereby imparting much-needed technology-based theorization into CH research [35, 124]. Based on our literature review in Online Supplemental Appendix C, most studies have simply applied existing criminological theories to a CH context [18] and have typically omitted the unique sociotechnical mechanisms of CH [74]. Many have not incorporated technological factors into their theoretical framework; in studies that have, technology is often nominally addressed, often not as an integral part of the core artifact of the investigation [84]. Surprisingly, few studies have investigated how technology and human factors come together to co- create CH [72]. We argue that the sociotechnical perspective allows for the clear identification of the role of technology in facilitating CH behaviors, considering both human and technological factors of CH, a perspective critically needed to advance CH scholarship [124].1 A better sociotechnical understanding should inspire researchers not only to design policies and procedures aimed at improving deviant behaviors but also to design technologies that inhibit such behaviors [14]. The third need is the investigation of possible nonlinearities in CH behavior.

Deviant behaviors are often nonlinear phenomena [72]; however, CH studies have generally ignored this possibility. Our research goes beyond that to theorize and test the nonlinear influence of the key variable of low self-control (LSC) on CH. Despite the fact that the relationship between self-control on deviant behaviors is nonlinear in criminology research [79], studies on cyberdeviance have largely failed to consider the potential nonlinear effects in the online context. In summary, we address these three research objectives by establishing the first

holistic, social-psychological-technological (S-P-T) theoretical model of CH. SLT and self-control theory (SCT)2 — two leading criminological theories — are integrated into our S-P-T framework to establish a foundational understanding of CH. We posit that integrating these theories provides a more nuanced theoretical framework for

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1145

understanding CH, particularly in light of the mixed results of studies that have leveraged only a single theory to understand this phenomenon [121].

Conceptualizing Cyberharassment

Here, we highlight the nature of CH and differentiates it from other terms that have often been used in research, most notably, cyberbullying. Although this term has often been used in research, in our study, CH was chosen for a more specific focus and to gain more empirical control. This is because cyberbullying is a broader phenomenon that encompasses different types of deviant behaviors, whereas CH is a specific type of cyberbullying. Thus, focusing on CH instead of cyberbullying ensures greater precision in investigating the phenomenon in question. Cyberbullying can be defined as follows:

[It] involves the use of information and communication technologies, such as e-mail, cell phone and pager text messages, instant messaging, defamatory personal Web sites, and defamatory online personal polling Web sites, to support deliberate, repeated, and hostile behavior by an individual or group that is intended to harm others [66, p. 436].

According to a taxonomy of cyberbullying [66], there are seven subcategories of cyberbullying behaviors, including flaming, CH, cyberstalking, denigration, mas- querading, outing, and exclusion. That study defined CH as behaving offensively toward another via online communication tools, whereas cyberbullying has a broader meaning that does not necessarily require bullies to directly interact with victims. Researchers [10] have proposed a similar typology that explained CH as

a subtype of cyberbullying behaviors. According to this typology [10], bullying can be categorized as direct or indirect. The former refers to “harassing others through either direct physical contact or verbal attack” (i.e., harassment) [10, p. 254], and the latter refers to covert aggressive behaviors intended to damage the victim’s social relations (e.g., rumor spreading or excluding someone from an online social group). It has been argued that cyberbullying involves not only CH but also cyberstalking [74] and that CH can be viewed as “a more severe form” of cyberbullying [65, p. 160, 67]. Compared to other types of cyberbullying, such as cyberstalking and exclusion, CH can often cause more direct and emotional damage to victims. It can thus be inferred that, among cyberbullying behaviors, the need to combat — and thus first understand — CH is most crucial.

Theoretical Background

Here, we introduce the theoretical background. First, the basic assumptions of SLT, its major components, and the interrelationships among these components are discussed. SCT is then introduced. A discussion is subsequently provided on the way certain characteristics of cyberspace afford the opportunity for crime and

1146 LOWRY ET AL.

facilitate the influence of LSC. Technology is included within the structure of SCT to address the absence of technological factors in studies.

Social Learning Theory

SLT is primarily a theory of cognitive psychology based upon the works of noted scholars such as Bandura. SLT suggests that cognitive processing of information from one’s social surroundings is salient to learning [133]. SLT emphasizes the role of the association-reinforcement learning process in “specific forms of deviant behavior” [7, p. 637]. A core SLT assumption is that social influence3 drives the causal mechanisms of learning and modeling deviant behavior. The attitudes, techniques, and motives that support criminal or deviant behaviors are learned through association with criminal or deviant peers [23]. We expand prior adaptations of SLT [e.g., 23] to criminological contexts in our study.4

According to this adaptation, criminal behavior is learned through vicarious positive and negative reinforcements, personal reinforcement, instruction, and observation. SLT identifies four causal mechanisms related to the social-influence-driven learning process: differential association, differential reinforcement, imitation, and definition. Notably, SLT’s mechanisms are sequentially related, each preceding the next. Among the four key mechanisms of SL, differential association occurs first [7].

Differential association is “the process by which individuals directly and indirectly interact and identify” with social groups [74, p. 966]. Through association with deviant peers in social groups, individuals are exposed to the “normative defini- tions, values, and attitudes favorable or unfavorable to a particular behavior” and to the gains and losses associated with certain behaviors [6, p. 106]. Differential association is followed by differential reinforcement. This is the

process by which people form a general perception of the benefits and punishments associated with a certain behavior through the observation of the consequences of similar behaviors performed by others. That is, it involves “the perceived, experi- enced, or anticipated reward and punishment for behavior” [6, p. 108]. Deviant behavioral patterns are reinforced when people observe that certain behaviors have desirable consequences (positive reinforcement) or do not entail punishments (negative reinforcement). Deviant behavioral patterns are weakened when people observe that certain behaviors lack desirable consequences (negative reinforcement) or entail strong punishments (positive reinforcement) [7]. Research has added definitions as well as the concept of “imitation” to SLT as

extensions of the differential association-reinforcement process [7]. Imitation refers to “engagement in behavior after the observation of similar behavior in others” [5, p. 196]. For reinforcement through vicarious experience, if individuals perceive in others’ experiences a high degree of benefits and a low degree of costs, they are more likely to imitate the behavior of the role model. However, researchers [7] empirically tested the influence of imitation on delin-

quency behaviors and found that it explained less than 0.1 percent of several types of

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1147

deviant behaviors (e.g., substance abuse). Thus, omitting imitation is consistent with empirical criminology studies using SLT: compared with the other three SLT compo- nents, imitation is seldom used to explain criminal and deviant behaviors [8, 132]. Definitions refer to “values, orientations, and attitudes toward” criminal or deviant

acts, which are learned through the association with peers or by observing the positive and negative consequences of certain behaviors [6, p. 106]. When individuals form a positive definition (i.e., perceive the behavior to be good) or neutralizing definition (i.e., perceive the behavior to be justifiable) toward a certain behavior, they are more likely to engage in it. Conversely, when individuals hold a negative definition (i.e., perceive the behavior to be undesirable), they are less likely to perform it [7]. A key contribution of our research is its investigation of the relationships among

SLT components. Although SLT “orders and specifies the interrelationships” among its variables [7, p. 638], related empirical studies have seldom scrutinized these interrelationships. However, a careful analysis of SLT reveals the presence of certain sequential mechanisms that should be recognized, particularly the relation- ships between differential association, differential reinforcement, and definitions. Per [7], differential association plays a key role in social learning because it serves

as the prerequisite for the subsequent SLT components. Differential association creates the context in which the subsequent components occur [6]. SLT uniquely emphasizes that the benefits and costs of deviant behavior are learned from social interactions with intimate deviant associates who facilitate both the imitation of behaviors and the formation of definitions regarding the behaviors. Because these deviant associates are important to the individual, their continued support reinforces the individual’s deviant behavior. That is, these differential associations reinforce and provide the normative influence for engaging in deviance [7]. Definitions regarding a behavior are thus formed through vicarious experiences

during both the differential association and reinforcement processes. That is, favor- able or unfavorable definitions can be formed through association and subsequent reinforcement [6]. Given the powerful roles of neutralization and shame in deviant behavior, they are used as the most proximate predictors of CH among all social- learning components. We propose an extended SLT that explicitly considers the causal relationships among the key learning components: differential association → differential reinforcement → definitions.

Self-Control Theory

SCT is also known as the general theory of crime [36] and was originally proposed to explain “all crime, at all times” [39, p.117]. Here, crime is defined as “acts of force or fraud undertaken in pursuit of self-interest” [39, p.15]. SCT is highly generalizable and has proven effective in explaining a variety of criminal behaviors and deviance, including fraud [53], date violence [105], and theft [104]. In technology research, SCT has been used to understand deviant behavior such as digital piracy [49], Internet fraud [41], and computer hacking [18]. SCT is especially useful in explaining major

1148 LOWRY ET AL.

forms of deviant behaviors [90]. Given that CH is the most severe form of cyberbully- ing behavior, it is thus logical to adapt SCT to a CH context. A unique virtue of SCT is that it proposes a more holistic view of crime,

considering both stable personal factor(s) and changing situational factors. Accordingly, it is a solid theory for explaining deviant acts. SCT posits that (1) LSC, a markedly stable personality trait, as well as (2) its interaction with criminal opportunity, are the major causes of criminal and deviant behaviors. In SCT, LSC is defined multidimensionally as “the enduring ‘criminality’ or ‘criminal propensity’ that increases the likelihood that individuals cannot resist the easy, immediate gratification that crime and analogous behaviors seductively, and almost ubiqui- tously, present in everyday life” [95, p. 932]. The second contention of SCT involves the salience of crime opportunity. The concept

of crime opportunity is extended to explain how similar assumptions can be extended for “CH opportunity” and how certain factors can enable individuals to engage in deviant behaviors [26]. Crime opportunity is pivotal. Only the presence of criminal opportunity transforms LSC into deviant behaviors [95]. Under this condition, criminal opportunities moderate the relationship between propensity and criminal events. Per [40], crime opportunity has three characteristics. First, crime opportunity

makes possible a crime that produces immediate rather than delayed gratification; second, it makes criminal behavior easier both “mentally and physi- cally” [39, p.12]; and third, it makes deviance appear less subject to detection and resistance. These three factors become more salient when the role of technology- based opportunism (TBO) in CH is conceptualized. In summary, the interaction effect between LSC and crime opportunity in foster-

ing crime and deviance is well-established in empirical studies [39, 95]. Because similar assumptions should hold for CH opportunity, the opportunity for crime in CH can be studied in terms of technology-related factors.

Foundational Framework for Cyberharassment

In this section, we explain why SLT and SCT serve as complementary theories to the understanding of CH, and why combining these results in a powerful causal model.

Value of Integrating the Complementary Theories

SLT and SCT are complementary theories in the sense that they focus on two different aspects of crime. SLT focuses on crime as a learned behavior through social association, whereas SCT focuses on crime as primarily an individual characteristic, catalyzed by opportunities for enacting the crime. The key difference between SLT and SCT is that SLT predicts that crime is swayed largely by social- learning factors like motives and skills, and definitions for criminal acts through social interactions with deviant peers; conversely, SCT argues that individual factors interact with opportunity to produce deviant behaviors. Thus, SCT embraces

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1149

an interactionist perspective, according to which a stable pattern of criminal behaviors—such as LSC in this context—interacts with opportunities to generate criminal behavior [39]. Despite their differences, SLT and SCT can be combined to establish a comprehensive understanding of deviant acts; they are “complementary” approaches [13, p. 406]. There is value in integrating competing theories [107], often because they provide

complementary viewpoints [46], which allow researchers to expand beyond myopic views to develop a richer understanding of a phenomenon. In fact, incorporating competing theories is one way to provide a strong research contribution [127]. Weber [125] echoes this sentiment, calling for researchers to become reflexive by integrating alternative viewpoints:

Reflexive researchers juxtapose the different perspectives of some phenomena provided by alternative, sometimes competing theories. They compare … contrast … assimilate. They are knowledgeable, … flexible users of theories (p. viii, emphasis added).

Prominent theoreticians have recommended that theory-building should focus on “analogical correspondence and conceptual blending” [85, p. 328]. They understand conceptual blending as the process of bringing together constructs from two domains to produce new insights; they caution that simply borrowing—just apply- ing an existing theory to a new domain—can be particularly problematic [85]. Inspired by these theoretical observations, we propose that SCT and SLT can be compared and synthesized into a compelling model, precisely because they provide competing, and thus complementary perspectives.

Prevalence in Existing Literature of Integrating Our Two Theories

Given our discussion, it is not surprising that many researchers have intuitively integrated SLT and SCT due to their complementary and compelling nature. In fact, their differing assumptions provide the very reason to integrate them, such that, taken together, they represent a complete view of the phenomenon. This contention is consistent with a series of articles [47-51] to study digital/software piracy, where SLT variables and LSC have been integrated as two complementary sets of vari- ables to explain piracy [49, 77]. Other studies also support our contention that the integration of SLT and SCT is possible, and in fact, desirable [4, 58, 59] . For example, studies [95, 96] conducted two meta-analyses on SCT [95] and SLT [96]. In [95] they concluded that “studies including SLT variables, in conjunction with self-control variables [SCT], explain 15.3 percent more variation in crime than do studies that do not control for social learning variables” (p. 949). Incidentally, integrating SLT and SCT increases variance explained by about the same amount when compared to the variance explained by either one [95]. A meta-analysis SLT study [96] had consistent findings with [95] integrating both SLT and SCT variables to explain crime. This study also revealed that the effect size of SLT variables is

1150 LOWRY ET AL.

comparable in magnitude to SCT variables in explaining crime. Thus, our study follows these studies in the literature, which have fruitfully integrated both theories to explain various criminal behaviors. These studies have argued that SLT and SCT can provide comparable insights that increase when both are considered. The appropriateness of integrating SLT and SCT is further elaborated in Online Supplemental Appendix D.

Theoretical Model and Hypotheses

Our theoretical model integrates SLT and SCT and contextualizes the combined model to CH, as depicted in Figure 1. The causal mechanisms of SLT and LSC are first mapped in a CH-specific manner, and then TBO is introduced. To develop the theoretical model, the SLT and SCT constructs were adapted to the CH context following [54, p. 113], who recommended creating “context-sensitive versions” of the explanatory variables when adapting a theoretical model to a phenomenon.

Recontextualizing the Causal Mechanisms to Explain Our Phenomenon

For SLT’s differential association element, negative social influence represents various sources of differential association by including all major reference groups with whom social actors associate in the CH context, including cyberfriends, opinion leaders on social networking sites, and strangers on the Internet [87], as well as situations in which individuals can learn CH. Differential reinforcement is conceptualized as the benefits- and costs-related perceptions formed through obser- ving or experiencing CH behaviors. For the definition elements, shame is proposed to reflect the positive definition for CH, and neutralization is proposed to reflect the neutralizing definition for CH. Multiple CH-specific neutralization techniques have been proposed based on the literature and interviews with online harassers. Developing these context-sensitive interpretations of social-learning elements con- tributes to SLT by specifying how factors can be adapted to the specific context that “affects the accepted relationships between the variables” [128, p. 492]. Shame is a unique form of definition in CH, and its key role warrants further

explanation. Shame has specifically been regarded as an important inhibitor of various forms of offline and online deviant behaviors, including offline harassment and CH [2, 97]. However, the role of shame in inhibiting CH has not been theoretically or empirically established. The model theoretically integrates the concept of shame, which is a mental state in which one feels disgrace or inadequacy that occurs when he or she violates social or cultural values; the feelings result from others’ real or imagined negative evaluations [56]. For theoretical precision, in our model, shame refers specifically to situational shame, which is indignity projected onto or associated with a socially negative incident [56].

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1151

N eg at iv e

so ci al

in flu en ce

C yb er -

ha ra ss m en t

(C H )

H 4a

S ha m e

C os ts

N eu tr al iz at io n

B en ef its

H 4b (- )

H 3d (- )

H 3a (- )

H 3b

H 3c

H 1( -)

H 2

H 5

LS C

T ec hn ol og y-

ba se d

op po rt un is m

(T B O )

H 6

A no ny m ity

La ck of

m on ito rin g

A sy nc hr on ic ity

P hy si ca l

ac tiv iti es

Im pu ls iv ity

S im pl e

ta sk s

R is k- ta ki ng

S el f-

ce nt er ed

T em pe r

D en ia lo f

in ju ry

C la im of

en tit le m en t D ef en ce of

ne ce ss ity

D en ia lo f

la w

D en ia lo f

re sp on si bi lit y

D iff er en tia l

as so ci at io n

D iff er en tia l

re in fo rc em en t

D ef in iti on (+ /-)

(f ro m as so ci at io n/

re in fo rc em en t)

C H op po rt un ity

af fo rd ed by TB O

F ig u re

1 . In te g ra ti o n o f S L T an d S C T fo r th e C H

C o n te x t

1152 LOWRY ET AL.

Shame plays a salient role in inhibiting deviant behaviors because it forms a part of one’s identity [9]. Shame can be regarded as a form of self-sanction in which “deviant behaviors would generate extreme feelings of shame and directly threaten the core of one’s self-worth” [37, p. 1347]. Per research on accountability and deindividuation [122, 123], when individuals perceive a higher degree of shame toward certain deviant behavior, they feel more conscious of the possible harm to their self- identities, and as a result, they behave in a more conforming manner to maintain positive self-identities. In addition, shame reflects individuals’ moral judgments about certain actions. Individuals feel shame when they perceive CH to be morally unacceptable, which inhibits related behaviors [74]. Therefore, we hypothesize:

Hypothesis 1. Shame has a negative association with cyberharassment.

Neutralization (i.e., neutralizing definition), or the extent to which individuals tend to defend and justify CH, should be a key facilitator of CH. This prediction is based on neutralization theory, which originated in criminological research and has been widely used to explain delinquent behavior. Neutralization is the process of justifying deviant behavior [109]. Namely, it is the process by which “the moral content of an unethical action is masked or overlooked …, [which] allows employees to engage in unethical acts without considering the ethical implications of their actions” [119, p. 622]. Neutralization is especially effective in explaining online deviant acts when the risk the behavior poses to the offender is relatively low [73]. Thus, it has been applied to phenomena such as policy noncompliance [109], software or music piracy [44], and unethical behavior in exchanges with companies [42]. In our research, neutralization is fundamentally a form of moral disengagement from

the victim’s possible (adverse) conditions and a means of rationalizing such conditions as justifiable [74]. For example, neutralization can be used to justify online harassment as “harmless fun” [74]. With a higher degree of neutralization, individuals are more likely to adopt techniques such as the denial of responsibility, denial of injury, defense of necessity, appeal to higher loyalties, and condemnation of condemners to justify their CH behaviors to be morally acceptable [109], each serving to remove the moral restrictions of engaging in CH. Therefore, we hypothesize:

Hypothesis 2. Neutralization is positively associated with cyberharassment.

Per SLT, the general costs of deviant behaviors are usually learned (personally or vicariously) through sanctions and punishments; similarly, the benefits are learned by witnessing or experiencing various positive outcomes of engaging in deviant behaviors. We express costs and benefits more generally, as is often the case in applications of rational choice theory to deviant behaviors [56]. Thus, the participants define what they consider to be costs and benefits for themselves. This assumption is particularly useful for SLT because costs and benefits are learned and not necessarily entirely rational or predictable across all forms of CH. Specifically, the perceived benefit of CH is defined as the set of expected favorable consequences for an individual engaging in CH based on

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1153

their learned experiences, whereas the perceived cost of CH is the set of unfavorable consequences based on their learned experiences. SLT suggests that definitions are to a large extent formed through the differential

reinforcement process [6]. This differential reinforcement process takes place when individuals find CH behaviors to be highly praised and supported by other deviant peers, while the probability of being punished is relatively low. Witnessing other cyberharassers being highly rewarded and seldom punished, individuals form the beliefs that CH is morally correct, and they feel less shame for becoming a cyberharasser. Thus, we hypothesize:

Hypothesis 3. Perceived cyberharassment benefits influence (a) shame nega- tively and (b) neutralization positively. Perceived cyberharassment costs influ- ence (c) shame positively and (d) neutralization negatively.

A key SLT assumption is that cost–benefit constructs are typically learned through actual experience and social influence rather than “calculated” in real time for a certain situation. Considering that CH was virtually unknown until the rapid proliferation of information and communication technologies, a fundamental assumption of theory development has been that CH arises primarily through differential association. It has been claimed that differential association occurs first in the social learning process and that the other variables largely depend on it [7]. That is, future deviants must have substantial association with the behaviors and outcomes of existing deviants to imitate their behavior. The key SLT concept of social influence, which refers to the tendency of individuals to rely on others’ actions to identify appropriate behaviors [27], was leveraged in this study to represent the differential association process. In this context, negative social influence occurs when people interact with online

peers who engage in CH. This influence often occurs in online deviant groups [98], where multiple cyberharassers and potential cyberharassers form a group to share their direct and indirect experience of CH. In an online environment with higher negative influence, individuals are more frequently exposed to positive rather than negative experiences of CH. As a result, they feel that CH is of greater benefit (e.g., fun, social approval, and demonstrating power) and less cost (e.g., absence of punishment). Thus, we hypothesize:

Hypothesis 4. Negative social influence is (a) positively associated with perceived cyberharassment benefits and (b) negatively associated with per- ceived cyberharassment costs.

Extending the Causal Mechanism of Low Self-Control to Explain Cyberharassment

To explain the link between LSC and CH, we leverage SCT, which argues that individuals with LSC are more likely to commit crimes when presented with the

1154 LOWRY ET AL.

opportunity [39]. Per SCT, people with LSC are less able to control their emotions and behaviors and thus more inclined to seek immediate gratification. LSC makes people more emotionally driven and directly influences the intention to perform delinquent behavior regardless of rational considerations. Thus, such people are more likely to have a higher degree of CH as “an efficient and effective means to satisfy immediate gratification,” whether the gratification consists of pure hedon- ism, revenge, or the wielding of power [19, p. 228]. LSC is an absence of the capacity to self-regulate, and thus it encourages the conversion of deviant intentions into actual behavior [83]. LSC is therefore crucial in the decision to commit a crime [115] by violating norms of conduct [56] and engaging in behavior that stalks, victimizes, and generally harasses individuals in cyberspace [19, 118]. Although these arguments indicate that LSC has a positive influence on deviant

behavior such as CH, the relationship is also theoretically interesting due to the nature of its influence. Unlike other relationships in the model, it has been empiri- cally established that the effects of self-control on deviant behavior (like CH) are nonlinear [79]. Consequently,

“studies are needed to investigate further the possibility of a nonlinear association and to test [it] empirically” [79, p. 447], thus “moving from specification of a parsimonious, linear relationship … to a more complex, nonlinear view of how self-control affects criminal behavior” [113, p. 712].

Theoretically, this nonlinear relationship occurs due to the so-called ‘cybernetic process’ in which self-control and the selection process (i.e., selecting criminal behavior) activate each other [75]. LSC creates negative feedback processes in which the inability to control oneself, and thus to manage temptations to commit deviant acts fosters LSC [15]. Thus, as LSC increases, an increasingly vicious phase begins in which the tendency for crime continues to increase due to the cascading effects of LSC [79]. The uniqueness of self-control in producing nonlinear relationships can be attributed

in part to its distinctiveness from the other constructs in the model in terms of development. Unlike the other constructs, self-control is partly determined by genet- ics. Self-control (or lack of it) develops from a nonlinear distribution of such genetic influences across various possible levels of self-control [106]. Thus, it is not surprising that self-control itself exhibits nonlinear influences. Empirical evidence also supports this contention, because it has also been observed that as self-control increases, prosocial behavior increases at a faster rate [25]. Thus, as LSC increases, CH should also increase at an increasing rate. Formally, we hypothesize:

Hypothesis 5. As low self-control increases, cyberharassment increases at an increasing rate.

According to SCT, opportunity is a salient contingency in the enactment of crime [129]. The general consensus is that opportunity presents a powerful situational contingency, which in turn catalyzes the influence of LSC on actual deviant

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1155

behavior [21]. Thus, the victims of such behavior are often opportunistically selected [28]. To summarize, there is sufficient evidence that deviance is “an interactive function of self-control and crime opportunity” [40, p. 10, 95]. In formal terms, opportunity for crime moderates the strength of the factors influencing deviance. Accordingly, we raise a key artifact question: How does technology promote or

inhibit this opportunity for crime? Research generally acknowledges the crucial role of technology in both creating and blocking opportunities for criminal or deviant acts [26]. In fact, it has been empirically demonstrated that certain technologies are greater enablers of such acts than others [57]. Due to the pervasiveness of comput- ing technology, opportunities for deviant behavior frequently arise [30]. Conversely, such opportunities can be diminished by thoughtfully designed technologies, such as value-sensitive and participatory social-media artifacts [20]. Thus, there is fertile ground for research into technology-related opportunities for

CH and the way these opportunities catalyze the effects of other factors on CH. SCT is extended to the technological environment by identifying the role of technology as the opportunistic enabler of CH. We propose that TBO acts as a key moderator in the relationship between CH and its antecedent, LSC. In the context of this study, TBO can be defined as the extent to which technology can provide opportunities — through characteristics or vulnerabilities existing in cyber- space — to engage in CH [100]. TBO is treated as a multidimensional construct because technology provides different opportunities or vulnerabilities for exploita- tion [26]. We propose three subdimensions of TBO, based on the online disinhibi- tion effect proposed by [112]. Among the technology-based factors listed in [112], three fit well with the definition of crime opportunity in SCT: anonymity, asyn- chronicity, and lack of monitoring.5 Crucially, these three technology-based factors map well with the three characteristics of crime opportunity.6

First, crime opportunity involves the perception of “little risk of detection and little risk of resistance” [39, p. 12]. Accordingly, a technology-enabled anonymous environment presents copious opportunities for crime. Technology-enabled anon- ymity is a technological factor unique to CH — it does not apply to offline harassment [94]. Anonymity reduces the amount of accountability perceived in the environment because behaviors become more difficult to link to an individual [92, 93]. Anonymity in cyberspace can be used to avert responsibility, overcome restrictions, and suspend moral cognitive processes, and it can lead people to engage in behaviors online they would never consider offline [112]. Second, it has been suggested that crime opportunity exists in situations char-

acterized by immediate gratification and delayed cost [39]. Thus, the asynchronicity of interactions in cyberspace makes individuals perceive a higher crime opportunity online. Specifically, asynchronicity provides “more time to reflect on the appro- priate response for their role” [38, p. 100]. This means that messages can be carefully edited to “optimize their self-presentation and self-disclosure” [22, p. 4].

1156 LOWRY ET AL.

Moreover, asynchronicity provides time for perpetrators to consider and research alternatives before taking action [131]. Third, crime opportunity exists in situations where crime or deviance is “mentally

and physically easy” [39, p. 12]. The lack of monitoring of cybercommunication facilitates the ease of engaging in online crime. It is widely known that people experience less mental stress when not monitored by others [89]. The perceived lack of monitoring is relevant in the CH context because a user can commit CH virtually anywhere on an online social platform and can engage in multiple acts of CH at virtually the same time. Thus, law enforcement agents face considerable difficulty in monitoring and identifying online delinquents [114]. Except under authoritarian regimes, in most online contexts, service providers do not track or record individuals’ online activities for real-time monitoring and surveillance. We propose that TBO moderates the LSC→CH relationship through its various

dimensions. This is similar to the treatment of other complex constructs in the literature. For example, [88] investigated the moderating role of psychological contract violation, which is a multidimensional construct like TBO. Likewise, [61] discussed the moderating role of ERP delivery, another multidimensional construct. It has been argued that lower dimensions should be combined into a higher-order construct for parsimony and meaningfulness [60]. The combination of the three TBO dimensions into the compelling TBO moderator follows this recommendation, which can be construed to be a more meaningful and holistic contribution to the CH literature. Technology provides the opportunity for crime and thus interacts with LSC to produce

deviant behavior, captured by TBO. LSC captures the individual propensity to engage in deviant behaviors such as CH, but this propensity requires an appropriate opportunity to be expressed as CH behavior. In cyberspace, TBO manifests through one or more of its dimensions. First, the sense of anonymity weakens the perception of sanction threats pertaining to cyberdeviance by removing the potential for social evaluation [31]. Per [70], if the online environment affords anonymity, online deviant behaviors are less likely to be punished, and social disapproval regarding deviant behaviors tends to be weaker. Thus, the TBO dimension of anonymity provides adequate opportunity to engage in deviant behaviors [44]. The second dimension of TBO, asynchronicity, also provides opportunities for CH. Asynchronicity frees people from “having to cope with someone’s immediate reaction” [112, p. 322-323], thus strengthening the pattern of immediate gratification. Asynchronicity provides individuals with greater crime oppor- tunity because reactions are delayed, providing time to carefully plan responses [112]. The third dimension of TBO, the lack of monitoring, captures another aspect of opportunity. The lack of monitoring implies that activities are not tracked [122, 123]. Lack of monitoring in cyberspace allows users to feel less accountable for actions and consequently creates less pressure to behave in a socially-acceptable manner [120]. We conclude that each dimension of TBO facilitates deviant behavior. Thus, TBO

effectively captures the various opportunities for CH in an online environment. Given that SCT (in a CH context) argues that the LSC→CH relationship is

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1157

moderated by opportunities to engage in CH, TBO can be positioned as the key moderator in this relationship:

Hypothesis 6. Technology-based opportunism positively moderates the influ- ence of LSC on CH.

Methodology

General Methodological Approach

Because CH involves behavior considered socially unacceptable in most cultures and thus could raise issues of social desirability bias during an empirical investiga- tion, the established approach for this type of research is using an anonymous self- reported survey. This choice increased the assurance that honest responses could be obtained from subjects while maintaining anonymity. Studies of deviant behaviors have used cross-sectional methodologies with accurate results and strong predictive validity [52]. Moreover, most CH research has focused on self-reported cross- sectional surveys to achieve the most accurate behavioral results [80, 111]. SLT was first established using cross-sectional data and has since been the leading approach in the literature [47, 50, 132]. Moreover, surveys are especially suitable for theory development in relatively mature areas of research [76]. Surveys also capture fine-grained user data, including individual and contextual variables that may be of interest (e.g., demographic data) [103].

Data Collection, Sample Selection, and Measures

Data were collected via the Qualtrics™ online panel service. Recruiting participants via market research firms has been fruitfully employed by research [34, 70]. It allows for the efficient administration of online surveys and the recruitment of voluntary, motivated, and willing research participants who diligently complete surveys for suitable compensation [34]. Such sources can also provide an effective way to gather generalizable, anonymous responses [91, 122]. Using a Qualtrics-based online survey approach also provided a wide range of respondents who had experience in committing CH, which allowed for accurate model testing.7

US adults were selected because US federal laws governing Internet use and freedom of expression are among the least restrictive in the world, thus providing a better scope for the investigation of CH. Several established data screening procedures were performed to enhance data quality and reduce social desirability bias.8 To ensure that the participants understood the survey and that the questions were applicable to them, all participants were required to be experienced cyberharassers, that is, to have at least “occasionally” engaged in one of the 21 CH behaviors.9

The sample had 214 female (35.5 percent) and 388 male (64.5 percent) subjects with an average age of 40.3 years (standard deviation [SD] of 13.91 years). They had an average

1158 LOWRY ET AL.

of 13.72 years of Internet-use experience (SD of 5.06 years) and an average of 17.2 years of computer-use experience (SD of 7.31). All measures were designed for experienced cyberharassers, based on established

measures, and modified to the context as necessary. For concision, all measurement details are reported in Online Supplemental Appendix A.

Study 1 Analysis and Results

Reliability and Validity

The covariance-based SEM tool STATA (version STATA/SE 14.2) was used for the analysis. Assessing factorial validity with SEM is different from the approach commonly used with components-based methods, such as partial least squares regressions [43, 71]. Well-established approaches were used to obtain the factorial validity and reliability of the data, to establish a lack of common-methods bias, and to test mediation and moderation relationships.10

Structural Model and Hypotheses Testing

Table 1 summarizes the model testing results, which provide support for most of the predicted relationships in the model except for H1 and H3a. In short, shame did not negatively impact CH as predicted; its effect was insignificant. H3 was mostly supported, with only one exception: benefits did not have a significant impact on shame. Two key findings can be discerned from Table 1. First, the negative social influence→(benefits, costs)→neutralization→CH mechanism is a powerful causal model for predicting CH. Comparatively, shame does not play a major role in this causal chain, ostensibly because the sample consisted of experienced cyberharas- sers and thus experience reduces shame’s salience. The second finding is that LSC plays an equally prominent role in predicting CH; the relation is strong and increasingly positive (i.e., nonlinear), and in fact greater than the effect of neutra- lization on CH (see Figure 2). The linear component of the LSC→CH relationship is also positively moderated by TBO, showing technology creates CH opportunity.

Post Hoc Analyses

Because two competing yet complementary theories were integrated, it was neces- sary to conduct separate post hoc tests to compare the integrated model with SLT- only and SCT-only models. The results are tabulated in Supplemental Tables B.7 and B.8, respectively. The trend of the results in terms of the coefficients held well against the separate post hoc studies conducted by applying only SLT and only SCT to the sample. The only slight difference was that the shame→CH relation was significant (p = 0.044) in the SLT-only analysis. Statistically, this is because the removal of LSC (resulting in an SLT-only model) makes shame marginally more

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1159

salient. Overall, there was no major difference in terms of the strength of relation- ships in the SLT components in the integrated model and those in the SLT-only model. The results of the SCT-only model also corroborated the integrated model well. Both LSC and the moderating influence of TBO retained a strong salience. Surprisingly, the CH variance explained by the SLT-only post hoc test was marginal

(3.3 percent). Moreover, the variance explained by the integrated model was about the same as that of the SCT-only model: the SLT components did not significantly contribute to CH. This unexpected result may have been an artifact of the sample of experienced cyberharassers. Due to this result and the lack of support for the two hypotheses related to shame, we conducted follow-up research through Study 2.

Table 1. Summary of Model Test Results (Study 1)

Hypothesis/Relationship β St. Err. z p Low CI

High CI

H1. Shame → Cyberharassment −.048 .046 −1.04 .300 −.139 .043 H2. Neutralization → Cyberharassment .148 .046 3.04 .001 .043 .238 H3a. Benefits → Shame −.034 .042 −0.83 .408 −.116 .048 H3b. Benefits → Neutralization .224 .042 4.53 .000 .115 .365 H3c. Costs → Shame .448 .038 12.04 .000 .375 .521 H3d. Costs → Neutralization −.324 .043 −7.48 .000 −.410 −.240 H4a. Negative social influence →

Benefits .342 .041 8.26 .000 .260 .423

H4b. Negative social influence → Costs −.258 .045 −11.29 .000 −.347 −.148 H5. LSC → Cyberharassment .329 .048 6.95 .000 .237 .422 H6. TBO → CH (baseline to test

moderation) .183 .043 4.29 .000 .099 .266

H6. TBO x LSC → Cyberharassment .141 .037 3.12 .000 .095 .285 H7. LSC2 → Cyberharassment .281 .045 6.30 .000 .237 .422 Control Variables Age → Cyberharassment −.193 .054 −3.56 .000 −.299 −.086 Computer experience →

Cyberharassment −.078 .054 −1.45 .148 −.183 .028

Computer usage → Cyberharassment .016 .037 0.43 .664 −.056 .088 Education level → Cyberharassment .075 .038 1.95 .051 −.004 .149 Gender → Cyberharassment −.107 .038 −2.82 .005 −.182 −.033 Internet experience →

Cyberharassment −.029 .051 −0.58 .564 −.129 .071

Situational moral beliefs → Cyberharassment

.048 .046 1.05 .294 −.133 .052

Work experience → Cyberharassment .075 .056 1.33 .185 −.036 .186

Notes: Model fit: χ2: 4,247.410 (df = 1,162); RMSEA: 0.066; CFI: 0.937; TLI: 0.928; SRMR: 0.099; CD: 0.999. R2s: Benefits = .112; Costs = .113; Neutralization = .204; CH = .285; Shame = .207.

1160 LOWRY ET AL.

Study 2: Follow-Up Study

Although the integrated model received reasonable support from the data of Study 1, the lack of support for the two hypotheses related to shame and the low variance explained by the SLT-only post hoc test indicated that these findings could be artifacts of the sample (i.e., experience). For experienced cyberharassers, shame may be less salient than it is for less experienced or new cyberharassers, causing the variance explained to be marginal. These participants would likely have experienced shame the first time they considered CH; however, their continued engagement in CH suggests that shame and neutralization over time had become less important factors. Thus, it was investigated whether this finding was generalizable beyond experienced cyberharassers. Although single-study investigations can be useful, they can potentially be

skewed or biased due to the nature of the sample or the source of data collection. Study 2 provided a means of improving confidence in the generalizability of the theory. Although the existence of methods bias was explicitly tested, testing the model in a completely different context decreased its likelihood and of biases, including social desirability bias [63, 64].

Relatively Contrasting Sample (Study 2)

The second sample was a relatively contrasting sample with individuals who were less experienced cyberharassers. This sample had multiple advantages. First, the

Figure 2. Nonlinear Effect of Low Self-Control on Cyberharassment (Study 1) Note: The y-axis represents cyberharassment; the x-axis represents low self-control.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1161

contrasting sample helped us determine whether the model held true for both experienced and less experienced cyberharassers. If the model held for the con- trasting sample, then it would show that the integrated approach is indeed a compelling means of investigating CH, but mainly for less experienced cyberharassers. Second, analyzing a contrasting sample is often a powerful empiri- cal technique. In the literature, for instance, contrasting samples have been used to provide comparative insights for a more granular understanding and to reduce methodological limitations, such as systematic variance within the samples [108]. Generally, scholars have often recommended testing a theory separately across

multiple samples to determine its robustness and to offer interesting comparative insights [63, 64]. Thus, a second vendor, Amazon’s MTurk, was used to recruit a purposeful, contrasting sample that consisted of participants who had less experi- ence engaging in CH. Except for the use of a different data panel and different sample selection criteria for the contrasting sample, the same methods, techniques, and procedures were employed to directly compare the two studies. To ensure that the respondents in Study 2 were less experienced harassers, we applied the follow- ing sample selection criteria: for each of the 21 CH behaviors, we required every respondent to have at most “occasionally” engaged in it (for all 21 measurement items of CH, it was required that the ratings were 3 or lower out of 5). The same instrument (Online Supplemental Appendix A) was used, except that some prompts were redesigned for less experienced cyberharassers. The sample consisted of 124 females (42.6 percent), 166 males (57.0 percent),

and 1 not identified (0.3 percent) subjects. Age averaged 38.3 years (SD of 12.73 years). The subjects averaged 16.21 years of Internet-use experience (SD of 4.01 years) and averaged 20.35 years of computer-use experience (SD of 6.54 years).

Study 2 Results and Analysis

The same procedures were followed to analyze this dataset. Online Supplemental Appendix B presents the results of Study 2, including summary statistics and reliability (Supplemental Table B.2), validity (Supplemental Table B.12), mediation and moderation (Supplemental Tables B.5, B.6), method bias, and multicollinearity. Table 2 shows the testing results, which largely mirror the findings of Study 1. The fit statistics were acceptable as per the standards discussed. Because two theories were integrated, separate post hoc tests were conducted to

compare the integrated model with SLT-only and SCT-only models. See results in Supplemental Tables B.9 and B.10.

Summary Results and Discussion: Structural Model and Post Hoc Tests

Study 2 largely mirrored the findings of Study 1 in terms of the strengths of relationships, with only two exceptions. First, in Study 2, benefits→shame was

1162 LOWRY ET AL.

significant (p = 0.017), as was shame→CH (p = 0.006). This implied that shame was more salient for less experienced cyberharassers. Second, the variance explained for CH, especially by the SLT-only post hoc test, was much more substantial (14.8 percent compared to 3.3 percent). Study 2 thus confirmed the existence of the negative social influence→(benefit, costs)→ (neutralization, shame)→CH mechanism and reinforced the contention that combining SCT’s LSC and TBO with SLT improves the understanding of CH. That is, as LSC increased, CH increased at a faster rate (see Figure 3). Consequently, Study 2 strongly corroborates the value of integrating SLT and

SCT into a unified theory. The variance explained in CH (the only factor common to the two separate post hoc tests) revealed that the SLT-only analyses explained about 14.8 percent of the variance in CH and that the SCT-only study explained

Table 2. Summary of Model Test Results: Study 2 (MTurk)

Hypothesis/Relationship β St. Err. z p Low CI

High CI

H1. Shame → Cyberharassment −.211 .074 −2.51 .006 −.258 −.033 H2. Neutralization → Cyberharassment .155 .072 2.15 .031 .014 .296 H3a. Benefits → Shame −.151 .063 −2.39 .017 −.274 −.027 H3b. Benefits → Neutralization .240 .076 3.15 .002 .091 .389 H3c. Costs → Shame .467 .060 7.78 .000 .349 .585 H3d. Costs → Neutralization −.178 .079 −1.99 .023 −.324 −.077 H4a. Negative social influence →

Benefits .295 .057 5.23 .000 .184 .405

H4b. Negative social influence → Costs −.267 .058 −4.59 .000 −.381 −.153 H5. Low self−control →

Cyberharassment .134 .054 2.17 .015 .037 .291

H6. TBO → CH (baseline for moderation test)

.163 .068 2.62 .004 .036 .262

H6. TBO x LSC → Cyberharassment .111 .041 2.10 .018 .012 .235 H7. LSC2 → Cyberharassment .209 .090 2.08 .019 .090 .358 Control Variables Age → Cyberharassment −.121 .135 −0.90 .368 −.386 .143 Computer experience →

Cyberharassment −.082 .085 −0.96 .337 −.249 .085

Computer usage → Cyberharassment −.050 .065 −0.77 .443 −.178 .078 Education level → Cyberharassment .003 .065 0.05 .962 −.125 .131 Gender → Cyberharassment −.056 .068 −0.83 .407 −.188 .076 Internet experience → Cyberharassment .093 .080 1.16 .247 −.064 .251 Situational moral beliefs →

Cyberharassment −.107 .072 −1.49 .135 −.248 .033

Work experience → Cyberharassment −.008 .131 −0.06 .950 −.265 .248

Model fit: χ2: 2,624.929 (df = 1,162); RMSEA: 0.066; CFI: 0.949; TLI: 0.939; SRMR: 0.098; CD: 1.000. R2s: Benefits = .082; Costs = .116; Neutralization = .246; CH = .280; Shame = .310.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1163

about 16.4 percent of the variance in CH. In contrast, the combined model explained about 28 percent of the variance in CH. This is a valuable finding because it reveals that integrating the theories contributes to a better understanding of CH—in contrast to Study 1, in which the SCT-only analysis explained almost as much variance as the integrated model. This difference could be explained by the fact that shame was a much stronger predictor of CH in Study 2 than in Study 1, which in turn was likely determined by the fact that for the experienced cyberhar- assers in Study 1, shame had less salience.

Discussion: Integrating Findings from the Two Studies

In view of the consistent findings across the two studies, the three most salient CH factors were identified: neutralization, LSC, and TBO, which is truly an S-P-T cluster of factors that provide powerful insights into the nature of CH. Thus, the research objective of identifying a causal, sociotechnical theory to better understand CH was fulfilled. This reveals that CH is a justificatory phenomenon based on situational factors, especially for less experienced cyberharassers, which is further augmented by stable individual traits and TBO. There are highly salient differences between Study 1 and Study 2. In Study 1, the

SLT component of the theory does not account for much CH variance; almost all variance is explained by the SCT component. In contrast, in Study 2, the variances

Figure 3. Nonlinear Effect of Low Self-Control on Cyberharassment (Study 2) Note: The y-axis represents Cyberharassment; the x-axis represents low self-control.

1164 LOWRY ET AL.

explained by the SLT and SCT components are comparable. The integrated model in Study 2 also explains almost the same variance of CH as obtained by adding the variances from the SLT-only and SCT-only analyses. Thus, it appears that the model operates in a differentiated manner: for less experienced cyberharassers, social, technological, and individual factors for CH should be considered. For experienced cyberharassers, the social component is not applicable, and the psychological and the technological components predominate. Upon further reflection, this finding appears to be internally consistent with both

SCT and SLT. SLT argues that criminal behavior, like CH, is socially learned. For experienced cyberharassers, this learning mechanism has been completed in the past; thus, it takes on less relevance in a current context. That is, when individuals have committed CH over a long period of time, social influence has less salience, and individual and technological factors prevail. In contrast to SLT, SCT frames CH as an outcome of LSC, an individual factor. Thus, the effect of LSC on CH is much more prominent than SLT components for experienced cyberharassers. Arguably, although both SLT and SCT continue to be relevant for experienced cyberharassers, the role of SLT decreases, and thus SCT becomes a more powerful explanation of CH for experienced cyberharassers.

Contributions to Research and Practice

Broad Research Contributions: A Compelling Sociotechnical Theory of Cyberharassment

Our study contributes to research on various levels, from broadly pushing the boundaries of CH research to providing specific contributions in terms of empirical findings. Our first contribution is the proposal of a clear sociotechnical framework for understanding CH. In conducting a comprehensive literature review, as shown in Online Supplemental Appendix C, we found that research on CH and related cyberdeviance (e.g., cyberbullying, cyberstalking, cyberaggression) has focused mainly on social factors; and, in some cases, psychological factors that facilitate cyberdeviance. Compelling technological factors are rarely studied in the literature, with perhaps one exception [74]; but even this study was concerned primarily with the social aspect of cyberdeviance. This lack of a clear sociotechnical approach to CH is perhaps one of the major

reasons that CH studies have often reported mixed and inconclusive evidence [121]. Because CH involves the use of technology and because “it is imperative that we gain a better understanding of why this phenomenon occurs” [130, p. 136], we argue that technology should not be excluded from the equation [72]. People take advantage of vulnerabilities provided by technology to self-disclose and behave unethically, and the technology environment is often underestimated by naïve users who fail to exercise the necessary caution, thus making them easy targets for predatory behaviors [33, 74]. The relevance of technology to CH and the gaps in

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1165

the literature underscore the necessity of developing a holistic S-P-T framework of CH (a true sociotechnical framework), which our study is the first to provide, to the best of our knowledge. The second broad contribution of our study is evidenced by the assertion that

reflexively incorporating competing theories yields a strong research contribution [125, 127]. Our study is among the first to provide a clear synthesis of two theories from criminology to explain CH. The two components of our integrated theory — SLT and SCT — explain CH in a complementary fashion, and each can predict different variances in this construct, demonstrating the efficacy of the overall model. The theoretical contributions of our study are further evaluated in Online Supplemental Appendix D, following recommendations by leading theorists [29, 102, 126].

Research Contributions

The contributions of our study follow largely from the theoretical integration of SLT and SCT. In the introduction, we argued that the integration of SLT and SCT into a comprehensive, yet parsimonious sociotechnical theory yields four funda- mental insights that go beyond the existing literature. These insights clarify (1) the role of LSC in CH, (2) the role of technology in CH, (3) the curvilinear nature of the influence of LSC, and (4) the ability to recognize the need for contingency- specific theorization. First, both studies showed that LSC is a strong predictor of CH. This highlights

the importance of individuals’ monitoring of their own behaviors. Most information security research has focused on how managerial-controlled policies, procedures, and associated negative costs change security-related behaviors. However, the present study investigated why people engage in outright deviant behavior based on individual reasons. It cannot be maintained that a cyberharasser simply did not know that CH was improper, harmful, or contrary to accepted norms. Thus, factors that not only deter these behaviors but encourage them should also be addressed. LSC, which our study highlights, are important. Second, the scope of SCT was extended by including technology as a factor,

which has long been proposed but rarely explicitly conceptualized and empirically tested in media and CH studies. Notably, the original formulation of SCT does not specifically address technology. In our study, TBO, a new construct, captures the salience of technology in CH. This adaptation of SCT to the technological environ- ment is a particularly important contribution. Not only does TBO have an influence on CH, but it also has an augmented influence when combined with LSC. Moreover, TBO only influences the linear component of the LSC→CH relation- ship. These results demonstrate that online social media platforms are more likely to encourage CH due to their lack of monitoring, asynchronous communication mechanisms, and heightened anonymity. Such a clear yet parsimonious delineation

1166 LOWRY ET AL.

of technological factors brought together under a compelling second-order construct (TBO) related to CH, represents a unique contribution. Third, the theorized and empirically demonstrated a nonlinear relationship

between LSC and CH is perhaps one of the most interesting empirical findings of our study. Crucially, “the omission of nonlinear relationships … [can be] potentially misleading, and therefore … a possible limitation” [116, p. 842, emphases added], and thus IS scholars have strongly advocated nonlinear modeling for more accurate understanding of IS phenomena [82]. Our analysis revealed new relationships in which the LSC→CH relationship had linear and quadratic components, and TBO moderated only the linear component. Thus, there is a substantial interaction effect and a main quadratic effect. In cases of complex interactions and influences, assuming simple linearity can lead to theoretical and empirical inaccuracy and thus to an inaccurate understanding of the phenomenon [79]. Fourth, a fundamental contribution of the study is that it highlights the need to

engage in context-specific theorization based on the samples we investigate. This contribution stems from testing our S-P-T model of CH with two contrasting samples: experienced and less experienced cyberharassers. We find that for less experienced cyberharassers, CH is an S-P-T phenomenon in which the influences of SLT and SCT are comparable. Namely, for less experienced cyberharassers, our integrated framework accounts for the social and technological factors that con- tribute to CH. By contrast, for experienced cyberharassers, CH is revealed to be more of a P-T phenomenon dominated particularly by the SCT component of our integrated theory. This finding provides fertile ground for future research to con- sider contingent theorization of CH, notably for future work on contingent pre- ventive mechanisms for CH. For example, when investigating experienced cyberharassers and attempting to prevent their behavior, researchers should focus primarily on SCT-based implications. This implies that the focus should be on LSC and TBO. Conversely, for less experienced cyberharassers, the integrated model is more appropriate, and the focus should be on LSC and TBO as well as social influences. Moreover, our study contributes to SLT and SCT individually. Both are well-

established criminological theories. This research provides additional insights into the nature of each theory and thus represents an important contribution. For example, in our study, SLT was not only extended from the traditional offline context to the CH context, but the interrelationships among four social-leaning components were also explored to obtain a complete picture of SLT to demonstrate that this understanding is more relevant for less experienced cyberharassers. These theorized interrelationships were, in fact, crucial to revealing that SLT is perhaps not as efficacious for under- standing experienced cyberharassers. Not theorizing these relationships between the SLTcomponents would likely have suppressed this important finding, which ultimately prompted us to conduct a second empirical study. Also, by adapting SLT differently from the literature [e.g., 74], we leverage the

malleability of SLT to expand the understanding of CH. This extension allows us to

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1167

develop further granular insights in terms of two particular components in the SLT model: neutralization and negative social influence. We find that in the context of neutralization, costs of CH behaviors have more

salience relative to benefits. Information security research has highlighted the importance of deterrence mechanisms and fear-appeal mechanisms in either redu- cing the benefits of insecure behaviors or increasing the perceived costs of such behaviors. The deterrence theory and protection motivation theory have been predominant in this literature stream; however, our SLT adaptation shows how negative social influence can alter perceived benefits and costs, which are largely outside of managerial influence and are highly salient to CH (but may also be salient to security research). Learning how and why significant others behave in deviant ways provides a powerful influence for engaging in CH. If either costs or the likelihood of being caught is observed to be low when others engage in these behaviors, people learn that their behaviors are likely to incur no costs. Thus, costs do in fact seem to deter those who lack this motivation to engage in CH, and when perceived costs have no impact, people may be motivated to engage in CH. The novel construct of negative social influence, as part of the adapted SLT, is

introduced to the CH literature stream to explain deviant behaviors. Research has used social influence to explain how individuals are influenced by peers when deciding how they will act with respect to a security policy (e.g., D’Arcy et al. [32] and Liang and Xue [68]); but, negative social influence is quite different, because it is the central tenet of SLT and explains not only how individuals are influenced by coworkers but also that influential others teach and model behaviors that will be imitated and potentially become habitual. Our study shows that negative influence has increased relevance for less experienced cyberharassers. Again, this finding was revealed only because we adapted SLT and also specified the antecedents and outcomes of each SLT component.

Contributions to Practice

First, our study has three direct implications for social media platforms and other communication technologies; these implications are related especially to TBO and its three dimensions of anonymity, asynchronicity, and lack of monitoring. Because TBO is enabled by the technical aspects of the platforms, it is possible that social media platforms can reduce some of these effects. Regarding anon- ymity, perhaps at least one verifiable identifier per account should be required so that truly anonymous accounts can no longer be created, making it possible to identify an account through a verified profile attribute (e.g., address, name, or phone number) as opposed to other attributes that can easily be created or altered (e.g., email address, other online profile, or user name). By removing the perception of complete anonymity, individuals would be less likely to perceive TBO and engage in CH.

1168 LOWRY ET AL.

Second, individuals are less likely to perceive that they are being monitored when their behaviors are typically not observed by others in the immediate vicinity. CH involves individuals and their devices, and others rarely observe the actual act of CH. Even though all behaviors are performed in a public space, little monitoring is actually perceived by individuals due to the vast size of these platforms. Individuals thus only see themselves and their devices, and with no physical presence nearby, monitoring of their behaviors is fairly low. Because this is a form of perception, it can potentially be manipulated by artifacts (e.g., warnings and smart agents). Third, perceived asynchronicity of communication tends to disinhibit individuals

and increase CH. Because the immediate consequences for and the reaction of the victim are deferred and physically removed from the cyberharasser, these costs are simply not perceived to be as high as they should be. The ability to instantly respond to someone, but to avoid bearing the immediate costs, weakens the perceived cost of engaging in CH and further disinhibits potential cyberharassers. If actions on these platforms occurred on a more real-time basis so that harassers could see the direct effects of their behaviors, the ability to become disinhibited, which encourages CH in the future, could be mitigated.

Limitations and Future Research

This study has limitations that point to compelling research opportunities. First, although the use of cross-sectional surveys improves generalizability, it limits the ability to establish causality empirically; however, given that two relatively well- known, causal theories were adapted, issues of causality are likely mitigated from a theoretical standpoint, although further empirical validation is still needed. A longitudinal examination of the CH phenomenon is specifically called for because this is an especially useful, valid manner in which to gather causal empirical evidence in a CH context. Second, the study was conducted only in a US context. Although this is appro-

priate given the stringent US laws on privacy and security, it is important to investigate how well the integrated CH theory holds in other countries/cultures, especially non-Western ones (e.g., China and India). Cross-cultural differences have been surfaced in such contexts in terms of other forms of social media use [69]. Likewise, research in countries with vastly different cyber laws around CH and privacy would be particularly relevant. Third, given the importance of LSC in this model, future studies should investi-

gate constructs that alter LSC in certain situations. Can situational factors alter self- control or at least increase its salience in situations in which deviant behaviors may occur? The concepts of control and monitoring may be important for this type of research because this literature stream emphasizes that the purpose of control is to bring behaviors or outcomes into conformity with policies or norms [86]. Control literature has long found that managerial oversight can alter behaviors, and the same concept is likely to apply to security-related behaviors. Likewise, the theory

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1169

of interpersonal behavior [117] proposes that facilitating conditions represent key impacts on intentions. Thus, managerial control and oversight may be able to either increase or decrease the ability to engage in deviance, making LSC less salient for situations concerning organizational security-related behaviors. Given the high salience of LSC, a final avenue for future research is to identify

a way to reduce it. For example, can LSC be addressed through primary and secondary socialization in individuals? Can schools implement programs to teach individuals the importance of self-control? Because many cyberharassers are ado- lescents and teens, such research could provide insights into ways to encourage individuals not to engage in CH, regardless of TBO.

Conclusion

In this paper, CH is explained using a sociotechnical theory by integrating SCT and SLT. The integrated theory was tested across two different studies involving experienced and less experienced cyberharassers. The results demonstrated the general applicability of the model and its high salience for less experienced cyberharassers. Specifically, neutralization, LSC, and technology were the strongest factors related to CH. It is clear that CH is a growing menace, and the findings will hopefully inspire researchers and practitioners to identify ways to combat it through the appropriate design of policies, programs, and technologies. Thus, multiple future studies are required to retest or extend the findings.

Acknowledgement: The authors acknowledge support from the National Natural Science Foundation of China (Grant No. 71801205, 71921001, 71871095, 71601080, 71801100, and 71801217), and the Fundamental Research Funds for the Central Universities (Grant No. WK2040160028).

NOTES 1. Following [124], the three key foci for producing leading CH research are as follows.

(1) Ground CH phenomena in a strong theoretical basis using new and insightful theoretical perspectives, including factoring in the role of technology, in CH. (2) Use powerful research methods to address CH phenomena, specifically from a sociotechnical angle. Given the rapid technological advances in cyberspace, there is not only a need to infuse CH research with sophisticated causal theory but also to “consider emerging methods and strategies that are relevant to new and emerging media, online behaviors, and the online spaces in which … people congregate” [110, pp. 197-198]. (3) Engage in causal modeling to unearth and determine the key human and technological factors associated with CH to mitigate it by developing powerful interventions.

2. Crucially, we use self-control theory, not social control theory (also known as social bond theory) or social cognitive theory. These are three distinct theories that use the same acronym but are not related.

3. Social influence refers to the tendency of individuals to rely on others’ actions to identify and model acceptable behaviors [27]. In this context, positive social influence refers to social examples of socially appropriate behavior; negative social influence, the focus of this study, is the opposite.

1170 LOWRY ET AL.

4. SLT is a broad theory of learning originally proposed by [12] to explain the critical role played by social context in general learning; it is related to Bandura’s social cognitive theory. Learning is a cognitive process that occurs in social contexts and thus involves not only direct instruction but also modeling and observation of others’ behaviors, including the costs and benefits thereof [12]. For concision, the broader theory of SLT is not explained; rather, the focus is a contextualized subset created by Akers and Burgess along with other research- ers [3, 7, 23], which is particularly useful for explaining deviant and aggressive behaviors.

5. Lack of monitoring is developed mainly on the basis of [112]’s concept of invisibility. The lack of monitoring refers to the degree to which individuals perceive their CH behaviors as physically visible to and monitored by others.

6. Again, these include: (1) immediate rather than delayed gratification; (2) relative ease, both mentally and physically; and (3) the perception that deviant acts are less subject to detection and resistance.

7. At the beginning of both Study 1 and Study 2, participants were given a list of 14 major CH behaviors compiled from [80] and [111]. Study 1 participants were asked if they had recently committed one or more of these acts. If so, they were “experienced cyberhar- assers” and allowed to continue; otherwise, they were excluded. In Study 2, to qualify as a “less-experienced cyberharasser” and to be allowed to continue, participants had to answer “never” or “rarely” to all questions about CH behaviors.

8. We required all respondents to be employed full-time. Respondents were fluent in English, were at least 18 years old, had at least five years of computer and Internet experience, and had reported recently committing at least one act of CH. To eliminate “professional” survey takers, participants who had participated in more than a handful of such surveys were blocked. Participants were dropped if they did not meet the screening criteria or if their surveys were incomplete. Due to the length and sensitive nature of the survey, and to decrease mono-method bias, the best practice of using attention-trap questions was followed to determine whether the respondents were reading all questions fully and answering honestly or were succumbing to social desirability bias [70, 73]. Such participants were dropped before they could continue with the survey. Moreover, working with data providers, the pilot test determined the average amount of time it took participants to complete the survey. In the final study, any response that took one-third of the average time or less was marked for deletion, because the respondent was likely not paying full attention to all questions (100 percent of these had also failed the attention traps). Items were also randomized within the instrument so participants would be less apt to detect underlying constructs; measures with different scaling and anchors were used; reverse-coded items were used; and extensive warnings and instructions to participants to maintain their focus on the survey were provided. To further demonstrate that common-methods bias was not an issue, per [101], an organizational commitment measure was gathered to use as a marker variable. As the analyses revealed, we conclude common-methods bias was not an issue. Additional measures were taken to reduce social desirability bias. First, strong assurances of anonymity were provided to participants. To ensure anonymity, the best practice of using truly anon- ymous research panels was followed by working with a third party [11, 73, 91]. Panel data better guarantees anonymity, because the respondents never interact with the researcher and the researcher never has access to their contact information. This allowed for gathering respondents from a wide range of industries and positions, who would have been virtually impossible to reach otherwise, which strengthened the generalizability of the results.

9. We specifically required that the rating of at least one of the 21 CH behaviors was 3 or above out of 5. 10. Convergent and discriminant validities were assessed by STATA’s confirmatory factor

analysis. Considering standards in the literature, such as CFI > 0.9, SRMR < 0.1, RMSEA < 0.08 [24, 55, 62], TLI > 0.9 [17], and CD > 0.45 [16], the model fit was good (see Table 1). Convergent validity was supported by large and standardized loadings for all constructs (p < .001) and t-values that exceeded statistical significance. Convergent validity was also supported by calculating the ratio of factor loadings to their respective standard errors that exceeded |10.0| (p < .001) [43, 71]. Cronbach’s alpha and summary statistics for the constructs are shown in Table B.1 in Online Supplemental Appendix B. Discriminant validity

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1171

was tested by showing that the measurement model had a significantly better model fit than a competing model with a single latent construct and was better than all other competing models in which pairs of latent constructs were joined. The χ2 differences between the competing models (omitted for brevity) were significantly larger than that of the original model, as also suggested by factor loadings, modification indices, and residuals [78]. The correlation matrix and average variance extracted (Online Supplemental Appendix B, Table B.11) also strongly supported the claims of discriminant validity. In summary, these tests confirmed convergent and discriminant validity. Moreover, tests were conducted for com- mon-methods bias, mediation, and moderation (see Online Supplemental Appendix B).

Supplemental Material

Supplemental data for this article can be accessed on the publisher's website

ORCID

Paul Benjamin Lowry http://orcid.org/0000-0002-0187-5808 Jun Zhang http://orcid.org/0000-0001-6275-9387 Gregory D. Moody http://orcid.org/0000-0001-7287-7336 Sutirtha Chatterjee http://orcid.org/0000-0001-8956-220X Chuang Wang http://orcid.org/0000-0002-1981-5352 Tailai Wu http://orcid.org/0000-0002-2025-3123

REFERENCES 1. Aboujaoude, E.; Savage, M.W.; Starcevic, V.; and Salame, W.O. Cyberbullying:

Review of an old problem gone viral. Journal of Adolescent Health, 57, 1 (2015), 10–18. 2. Ahmed, E.; Harris, N.; and Braithwaite, V. Forgiveness, reconciliation, and shame:

Three key variables in reducing school bullying. Journal of Social Issues, 62, 2 (2006), 347–370.

3. Akers, R.L. Rational choice, deterrence, and social learning theory in criminology: The path not taken. The Journal of Criminal Law and Criminology, 81, 3 (1990), 653–676.

4. Akers, R.L. Self-control as a general theory of crime. Journal of Quantitative Criminology, 7, 2 (1991), 201–211.

5. Akers, R.L. Social learning theory. In R. Paternoster, and R. Bachman (eds.), Explaining Criminals and Crime: Essays in Contemporary Criminological Theory. Los Angeles, CA: Roxbury, 2001, pp. 192–210.

6. Akers, R.L.; and Jennings, W.G. The social learning theory of crime and deviance. In M.D. Krohn, A.J. Lizotte, and G.P. Hall (eds.), Handbook on Crime and Deviance. New York, NY: Springer, 2009, pp. 103–120.

7. Akers, R.L.; Krohn, M.D.; Lanza-Kaduce, L.; and Radosevich, M. Social learning and deviant behavior: A specific test of a general theory. American Sociological Review, 44, 4 (1979), 636–655.

8. Akers, R.L.; and Lee, G. A longitudinal test of social learning theory: Adolescent smoking. Journal of Drug Issues, 26, 2 (1996), 317–343.

9. Arbore, A.; Soscia, I.; and Bagozzi, R.P. The role of signaling identity in the adoption of personal technologies. Journal of the Association for Information Systems, 15, 2 (2014), 86–110. 10. Aricak, T.; Siyahhan, S.; Uzunhasanoglu, A.; Saribeyoglu, S.; Ciplak, S.; Yilmaz, N.;

and Memmedov, C. Cyberbullying among Turkish adolescents. CyberPsychology & Behavior, 11, 3 (2008), 253–261.

1172 LOWRY ET AL.

11. Awad, N.F.; and Ragowsky, A. Establishing trust in electronic commerce through online word of mouth: An examination across genders. Journal of Management Information Systems, 24, 4 (2008), 101–121. 12. Bandura, A. Social Learning Theory. Englewood Cliffs, NJ: Prentice-Hall, 1977. 13. Baron, S.W. Self-control, social consequences, and criminal behavior: Street youth and

the general theory of crime. Journal of Research in Crime and Delinquency, 40, 4 (2003), 403–425. 14. Bauman, S.; and Yoon, J. This issue: Theories of bullying and cyberbullying. Theory

Into Practice, 53, 4 (2014), 253–256. 15. Baumeister, R.F.; Masicampo, E.; and Vohs, K.D. Do conscious thoughts cause

behavior? Annual Review of Psychology, 62, (2011), 331–361. 16. Bollen, K.A. Structural equations with latent variables. New York: John Wiley, 1989. 17. Boss, S.R.; Galletta, D.F.; Lowry, P.B.; Moody, G.D.; and Polak, P. What do systems

users have to fear? Using fear appeals to engender threats and fear that motivate protective security behaviors. MIS Quarterly, 39, 4 (2015), 837–864. 18. Bossler, A.M.; and Burruss, G.W. The general theory of crime and computer hacking:

Low self-control hackers. In T.J. Holt, and B.H. Schell (eds.), Corporate Hacking and Technology-Driven Crime: Social Dynamics and Implications. New York, NY: Information Science Reference, 2011, pp. 38–67. 19. Bossler, A.M.; and Holt, T.J. The effect of self-control on victimization in the

cyberworld. Journal of Criminal Justice, 38, 3 (2010), 227–236. 20. Bowler, L.; Knobel, C.; and Mattern, E. From cyberbullying to well-being:

A narrative-based participatory approach to values-oriented design for social media. Journal of the Association for Information Science and Technology, 66, 6 (2015), 1274–1293. 21. Braga, A.A.; and Clarke, R.V. Explaining high-risk concentrations of crime in the city:

Social disorganization, crime opportunities, and important next steps. Journal of Research in Crime and Delinquency, 51, 4 (2014), 480–498. 22. Brink, A. Affective experiences with sex and sexual satisfaction among Dutch adoles-

cents: The consequences of online and offline communication about sexuality with friends. In Faculty of Social and Behavioural Sciences, Master: Utrecht University, 2014. 23. Burgess, R.L.; and Akers, R.L. A differential association-reinforcement theory of

criminal behavior. Social Problems, 14, 2 (1966), 128–147. 24. Byrne, B.M. Structural equation modeling with EQS: Basic concepts, applications,

and programming. Mahwah, NJ: Lawrence Erlbaum, 2006. 25. Carlo, G.; Crockett, L.J.; Wolff, J.M.; and Beal, S.J. The role of emotional reactivity,

self-regulation, and puberty in adolescents’ prosocial behaviors. Social Development, 21, 4 (2012), 667–685. 26. Chatterjee, S.; Sarker, S.; and Valacich, J.S. The behavioral roots of information

systems security: Exploring key factors related to unethical IT use. Journal of Management Information Systems, 31, 4 (2015), 49–87. 27. Cialdini, R.B. Influence: Science and Practice. Boston: MA: Pearson Education, Inc.,

2009. 28. Cockbain, E.; and Wortley, R. Everyday atrocities: does internal (domestic) sex

trafficking of British children satisfy the expectations of opportunity theories of crime? Crime Science, 4, 35 (2015), 1–12. 29. Corley, KG and Gioia, DA. Building theory about theory building: What constitutes

a theoretical contribution? Academy of Management Review, 36, 1 (2011), 12–32. 30. Cramer, M.; and Hayes, G. Acceptable use of technology in schools: Risks, policies,

and promises. IEEE Pervasive Computing, 9, 3 (2010), 37–44. 31. D’Arcy, J.; and Herath, T. A review and analysis of deterrence theory in the IS security

literature: Making sense of the disparate findings. European Journal of Information Systems, 20, 6 (2011), 643–658. 32. D’Arcy, J.; Hovav, A.; and Galletta, D. User awareness of security countermeasures

and its impact on information systems misuse: A deterrence approach. Information Systems Research, 20, 1 (2009), 79–98.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1173

33. Dempsey, A.G.; Sulkowski, M.L.; Dempsey, J.; and Storch, E.A. Has cyber technology produced a new group of peer aggressors? CyberPsychology, Behavior & Social Networking, 14, 5 (2011), 297–302. 34. Dumas, T.L.; Phillips, K.W.; and Rothbard, N.P. Getting closer at the company party:

Integration experiences, racial dissimilarity, and workplace relationships. Organization Science, 24, 5 (2013), 1377–1401. 35. Espelage, D.L.; Rao, M.A.; and Craven, R.G. Theories of cyberbullying. In S. Bauman

(ed.), Principles of Cyberbullying Research: Definitions, Measures, and Methodology. New York, NY: Routledge, 2013, pp. 78–97. 36. Evans, T.D.; Cullen, F.T.; Burton, V.S.; Dunaway, R.G.; and Benson, M.L. The social

consequences of self-control: Testing the general theory of crime. Criminology, 35, 3 (1997), 475–504. 37. Ferris, D.L.; Brown, D.J.; Lian, H.; and Keeping, L.M. When does self-esteem relate to

deviant behavior? The role of contingencies of self-worth. Journal of Applied Psychology, 94, 5 (2009), 1345–1353. 38. Freeman, M.A.; and Capper, J.M. Exploiting the web for education: An anonymous asyn-

chronous role simulation. Australasian Journal of Educational Technology, 15, 1 (1999), 95–116. 39. Gottfredson, M.R.; and Hirschi, T. A General Theory of Crime. Stanford, CA: Stanford

University Press, 1990. 40. Grasmick, H.G.; Tittle, C.R.; Bursik, R.J.; and Arneklev, B.J. Testing the core empiri-

cal implications of Gottfredson and Hirschi’s general theory of crime. Journal of Research in Crime and Delinquency, 30, 1 (1993), 5–29. 41. Gregory, E.E.; and Grace, A.A. Psychodemographic factors predicting Internet fraud

tendency among youths in Southwestern, Nigeria. Journal of Educational and Social Research, 5, 2 (2015), 159–164. 42. Gruber, V.; and Schlegelmilch, B. How techniques of neutralization legitimize norm- and

attitude-inconsistent consumer behavior. Journal of Business Ethics, 121, 1 (2013), 29–45. 43. Hair, J.F.; Tatham, R.L.; Anderson, R.E.; and Black, W. Multivariate data analysis. 6:

Pearson Prentice Hall Upper Saddle River, NJ, 2006. 44. Harris, L.C.; and Dumas, A. Online consumer misbehaviour: An application of

neutralization theory. Marketing Theory, 9, 4 (2009), 379–402. 45. Hazelwood, S.D.; and Koon-Magnin, S. Cyber stalking and cyber harassment legisla-

tion in the United States: A qualitative analysis. International Journal of Cyber Criminology, 7, 2 (2013), 155–168. 46. Henderson, A.D.; and Fredrickson, J.W. Top management team coordination needs and

the CEO pay gap: A competitive test of economic and behavioral views. Academy of Management Journal, 44, 1 (2001), 96–117. 47. Higgins, G.E. Gender differences in software piracy: The mediating roles of

self-control theory and social learning theory. Journal of Economic Crime Management, 4, 1 (2006), 1–30. 48. Higgins, G.E.; Fell, B.D.; and Wilson, A.L. Digital piracy: Assessing the contributions

of an integrated self-control theory and social learning theory using structural equation modeling. Criminal Justice Studies, 19, 1 (2006), 3–22. 49. Higgins, G.E.; Fell, B.D.; and Wilson, A.L. Low self-control and social learning in

understanding students’ intentions to pirate movies in the United States. Social Science Computer Review, 25, 3 (2007), 339–357. 50. Higgins, G.E.; and Makin, D.A. Does social learning theory condition the effects of

low self-control on college students’ software piracy. Journal of Economic Crime Management, 2, 2 (2004), 1–22. 51. Higgins, G.E.; and Wilson, A.L. Low self-control, moral beliefs, and social learning

theory in university students’ intentions to pirate software. Security Journal, 19, 2 (2006), 75–92. 52. Higgins, G.E.; Wolfe, S.E.; and Marcum, C.D. Music piracy and neutralization:

A preliminary trajectory analysis from short-term longitudinal data. International Journal of Cyber Criminology, 2, 2 (2008), 324–336.

1174 LOWRY ET AL.

53. Holtfreter, K.; Reisig, M.D.; and Pratt, T.C. Low self-control, routine activities, and fraud victimization. Criminology, 46, 1 (2008), 189–220. 54. Hong, W.; Chan, F.K.; Thong, J.Y.; Chasalow, L.C.; and Dhillon, G. A framework and

guidelines for context-specific theorizing in information systems research. Information Systems Research, 25, 1 (2013), 111–136. 55. Hu, L.T.; and Bentler, P.M. Cutoff criteria for fit indexes in covariance structure

analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6, 1 (1999), 1–55. 56. Hu, Q.; Xu, Z.; Dinev, T.; and Ling, H. Does deterrence work in reducing informa-

tion security policy abuse by employees? Communications of the ACM, 54, 6 (2011), 54–60. 57. Huang, Y.-y.; and Chou, C. An analysis of multiple factors of cyberbullying among

junior high school students in Taiwan. Computers in Human Behavior, 26, 6 (2010), 1581–1590. 58. Jennings, W.G.; Higgins, G.E.; Akers, R.L.; Khey, D.N.; and Dobrow, J. Examining

the Influence of delinquent peer association on the stability of self-control in late childhood and early adolescence: Toward an integrated theoretical model. Deviant Behavior, 34, 5 (2013), 407–422. 59. Jennings, W.G.; Park, M.; Tomsich, E.A.; Gover, A.R.; and Akers, R.L. Assessing the

overlap in dating violence perpetration and victimization among South Korean college students: The influence of social learning and self-control. American Journal of Criminal Justice, 36, 2 (2011), 188–206. 60. Johnson, R.E.; Rosen, C.C.; and Chang, C.-H. To aggregate or not to aggregate: Steps

for developing and validating higher-order multidimensional constructs. Journal of Business and Psychology, 26, 3 (2011), 241–248. 61. Karimi, J.; Somers, T.M.; and Bhattacherjee, A. The impact of ERP implementation on

business process outcomes: A factor-based study. Journal of Management Information Systems, 24, 1 (2007), 101–134. 62. Kline, R.B. Principles and practice of structural equation modeling. New York:

Guilford Press, 2005. 63. Lee, C.; and Bobko, P. Self-efficacy beliefs: Comparison of five measures. Journal of

Applied Psychology, 79, 3 (1994), 364–369. 64. Leung, K.; Friedman, R.; and Chen, C.C. Special issue on leveraging

phenomenon-based research in China for theory advancement. Organizational Behavior and Human Decision Processes, 122, 2 (2013), 305–306. 65. Li, Q. Cyberbullying in schools: A research of gender differences. School Psychology

International, 27, 2 (2006), 157–170. 66. Li, Q. Bullying in the new playground: Research into cyberbullying and cyber

victimisation. Australasian Journal of Educational Technology, 23, 4 (2007), 435–454. 67. Li, Q. New bottle but old wine: A research of cyberbullying in schools. Computers in

Human Behavior, 23, 4 (2007), 1777-1791. 68. Liang, H.; and Xue, Y. Avoidance of information technology threats: A theoretical

perspective. MIS Quarterly, 33, 1 (2009), 71–90. 69. Lowry, P.B.; Cao, J.; and Everard, A. Privacy concerns versus desire for interpersonal

awareness in driving the use of self-disclosure technologies: The case of instant messaging in two cultures. Journal of Management Information Systems, 27, 4 (2011), 163–200. 70. Lowry, P.B.; D’Arcy, J.; Hammer, B.; and Moody, G.D. Cargo Cult’ science in

traditional organization and information systems survey research: A case for using nontradi- tional methods of data collection, including Mechanical Turk and online panels. Journal of Strategic Information Systems, 25, 3 (2016), 232–240. 71. Lowry, P.B.; and Gaskin, J. Partial least squares (PLS) structural equation modeling

(SEM) for building and testing behavioral causal theory: When to choose it and how to use it. IEEE Transactions on Professional Communication, 57, 2 (2014), 123–146. 72. Lowry, P.B.; Moody, G.D.; and Chatterjee, S. Using IT design to prevent

cyberbullying. Journal of Management Information Systems, 34, 3 (2017), 863–901.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1175

73. Lowry, P.B.; Moody, G.D.; Galletta, D.F.; and Vance, A. The drivers in the use of online whistle-blowing reporting systems. Journal of Management Information Systems, 30, 1 (2013), 153–189. 74. Lowry, P.B.; Zhang, J.; Wang, C.L.; and Siponen, M. Why do adults engage in

cyberbullying on social media? An integration of online disinhibition and deindividuation effects with the social structure and social learning (SSSL) model. Information Systems Research, 27, 4 (2016), 962–986. 75. Magen, E.; and Gross, J.J. Getting our act together: Toward a general model of self-

control. In R. Hassin, K. Ochsner, and Y. Trope (eds.), Self Control in Society, Mind and Brain. Oxford, UK: Oxford University Press, 2010, pp. 335–353. 76. Malhotra, M.K.; and Grover, V. An assessment of survey research in POM: From

constructs to theory. Journal of Operations Management, 16, 4 (1998), 407–425. 77. Malin, J.; and Fowers, B.J. Adolescent self-control and music and movie piracy.

Computers in Human Behavior, 25, 3 (2009), 718–722. 78. Marsh, H.W.; and Hocevar, D. Application of confirmatory factor analysis to the study

of self-concept: First- and higher order factor models and their invariance across groups. Psychological Bulletin, 97, 3 (1985), 562–582. 79. Mears, D.P.; Cochran, J.C.; and Beaver, K.M. Self-control theory and nonlinear effects

on offending. Journal of Quantitative Criminology, 29, 3 (2013), 447–476. 80. Menesini, E.; Nocentini, A.; Palladino, B.E.; Frisén, A.; Berne, S.; Ortega-Ruiz, R.;

Calmaestra, J.; Scheithauer, H.; Schultze-Krumbholz, A.; and Luik, P. Cyberbullying defini- tion among adolescents: A comparison across six European countries. Cyberpsychology, Behavior, and Social Networking, 15, 9 (2012), 455–463. 81. Montaldo, C. Cyberstalking and Internet Harassment - Then and Now. 2018.

ThoughtCo: ThoughtCo, 2017. 82. Moody, G.D.; Lowry, P.B.; and Galletta, D. It’s complicated: Explaining the relation-

ship between trust, distrust, and ambivalence in online transaction relationships using poly- nomial regression analysis and response surface analysis. European Journal of Information Systems, 26, 4 (2017), 379–413. 83. Murray, K.T.; and Kochanska, G. Effortful control: Factor structure and relation to

externalizing and internalizing behaviors. Journal of Abnormal Child Psychology, 30, 5 (2002), 503–514. 84. Orlikowski, W.J.; and Iacono, C.S. Research commentary: Desperately seeking the

“IT” in IT research—A call to theorizing the IT artifact. Information Systems Research, 12, 2 (2001), 121–134. 85. Oswick, C.; Fleming, P.; and Hanlon, G. From borrowing to blending: Rethinking the

processes of organizational theory building. Academy of Management Review, 36, 2 (2011), 318–337. 86. Ouchi, W.G. A conceptual framework for the design of organizational control mechan-

isms. Readings in Accounting for Management Control. US: Springer, 1979, pp. 63–82. 87. Pauwels, L.; and Schils, N. Differential online exposure to extremist content and

political violence: Testing the relative strength of social learning and competing perspectives. Terrorism and Political Violence, 28, 1 (2016), 1–29. 88. Pavlou, P.A.; and Gefen, D. Psychological contract violation in online marketplaces:

Antecedents, consequences, and moderating role. Information Systems Research, 16, 4 (2005), 372–399. 89. Piccoli, G.; and Ives, B. Trust and the unintended effects of behavior control in virtual

teams. MIS Quarterly, 27, 3 (2003), 365–395. 90. Polakowski, M. Linking self-and social control with deviance: Illuminating the struc-

ture underlying a general theory of crime and its relation to deviant activity. Journal of Quantitative Criminology, 10, 1 (1994), 41–78. 91. Posey, C.; Roberts, T.; Lowry, P.; Bennett, B.; and Courtney, J. Insiders’ protection of

organizational information assets: Development of a systematics-based taxonomy and theory of diversity for protection-motivated behaviors. MIS Quarterly, 37, 4 (2013), 1189–1210. 92. Postmes, T.; and Spears, R. Deindividuation and antinormative behavior: A

meta-analysis. Psychological Bulletin, 123, 3 (1998), 238–259.

1176 LOWRY ET AL.

93. Postmes, T.; Spears, R.; and Lea, M. Breaching or building social boundaries? SIDE-effects of computer-mediated communication. Communication Research, 25, 6 (1998), 689–715. 94. Postmes, T.; Spears, R.; Sakhel, K.; and De Groot, D. Social influence in

computer-mediated communication: The effects of anonymity on group behavior. Personality and Social Psychology Bulletin, 27, 10 (2001), 1243–1254. 95. Pratt, T.C.; and Cullen, F.T. The empirical status of Gottfredson and Hirschi’s general

theory of crime: A meta-analysis. Criminology, 38, 3 (2000), 931–964. 96. Pratt, T.C.; Cullen, F.T.; Sellers, C.S.; Thomas Winfree Jr, L.; Madensen, T.D.; Daigle,

L.E.; Fearn, N.E.; and Gau, J.M. The empirical status of social learning theory: A meta- analysis. Justice Quarterly, 27, 6 (2010), 765–802. 97. Putnam, D.E. Initiation and maintenance of online sexual compulsivity: Implications

for assessment and treatment. CyberPsychology & Behavior, 3, 4 (2000), 553–563. 98. Pyrooz, D.C.; Decker, S.H.; and Moule Jr, R.K. Criminal and routine activities in

online settings: Gangs, offenders, and the Internet. Justice Quarterly, 32, 3 (2015), 471–499. 99. Ransbotham, S.; Fichman, R.G.; Gopal, R.; and Gupta, A. Special section introduction

—Ubiquitous IT and digital vulnerabilities. Information Systems Research, 27, 4 (2016), 834–847. 100. Ransbotham, S.; and Mitra, S. Choice and chance: A conceptual model of paths to information security compromise. Information Systems Research, 20, 1 (2009), 121–139. 101. Richardson, H.A.; Simmering, M.J.; and Sturman, M.C. A tale of three perspectives: Examining post hoc statistical techniques for detection and correction of common method variance. Organizational Research Methods, 12, 4 (2009), 762–800. 102. Rivard, S. Editor’s comments: The ions of theory construction. MIS Quarterly, 38, 2 (2014), iii–xiv. 103. Roach, M.; and Sauermann, H. Founder or joiner? The role of preferences and context in shaping different entrepreneurial interests. Management Science, 61, 9 (2015), 2160–2184. 104. Schreck, C.J. Criminal victimization and low self-control: An extension and test of a general theory of crime. Justice Quarterly, 16, 3 (1999), 633–654. 105. Schreck, C.J.; Stewart, E.A.; and Osgood, D.W. A reappraisal of the overlap of violent offenders and victims. Criminology, 46, 4 (2008), 871–906. 106. Schwartz, J.A.; Connolly, E.J.; Nedelec, J.L.; and Beaver, K.M. An investigation of genetic and environmental influences across the distribution of self-control. Criminal Justice and Behavior, 44, 9 (2017), 1163–1182. 107. Seibert, S.E.; Kraimer, M.L.; and Liden, R.C. A social capital theory of career success. Academy of Management Journal, 44, 2 (2001), 219–237. 108. Sharma, R.; Yetton, P.; and Crawford, J. Estimating the effect of common method variance: The method-method pair technique with an illustration from TAM research. MIS Quarterly, 33, 3 (2009), 473–490. 109. Siponen, M.; and Vance, A. Neutralization: New insights into the problem of employee information systems security policy violations. MIS Quarterly, 34, 3 (2010), 487–502. 110. Spears, BA and Zeederberg, M. Emerging methodological strategies to address cyber- bullying: Online social marketing and young people as co-researchers. In S. Bauman, D. Cross, and J. Walker (eds.), Principles of Cyberbullying Research, Definitions, Measures, and Methodology. New York, NY: Routledge, 2013, pp. 196–209. 111. Strawhun, J; Adams, N; and Huss, MT. The assessment of cyberstalking: An expanded

examination including social networking, attachment, jealousy, and anger in relation to violence and abuse. Violence and Victims, 28, 4 (2013), 715–730. 112. Suler, J. The online disinhibition effect. CyberPsychology & Behavior, 7, 3 (2004), 321–326. 113. Sullivan, C.J.; and Loughran, T. Investigating the functional form of the self- control–delinquency relationship in a sample of serious young offenders. Journal of Quantitative Criminology, 30, 4 (2014), 709–730. 114. Tavani, HT. Defining the boundaries of computer crime: Piracy, break-ins, and sabo- tage in cyberspace. ACM SIGCAS Computers and Society, 30, 3 (2000), 3–9.

JOURNAL OF MANAGEMENT INFORMATION SYSTEMS 1177

115. Tibbetts, SG and Gibson, CL. Individual propensities and rational decision-making: Recent findings and promising approaches. In A.R. Piquero, and S.G. Tibbetts (eds.), Rational Choice and Criminal Behavior Recent Research and Future Challenges. New York, NY: Routledge, 2002, pp. 3–24. 116. Titah, R.; and Barki, H. Nonlinearities between attitude and subjective norms in information technology acceptance: A negative synergy? MIS Quarterly, 33, 4 (2009), 827–844. 117. Triandis, H.C. Interpersonal Behavior. US: Brooks/Cole Publishing Company, 1977. 118. Turanovic, J.J.; and Pratt, T.C. “Can’t stop, won’t stop”: Self-control, risky lifestyles, and repeat victimization. Journal of Quantitative Criminology, 30, 1 (2014), 29–56. 119. Umphress, E.E.; and Bingham, J.B. When employees do bad things for good reasons: Examining unethical pro-organizational behaviors. Organization Science, 22, 3 (2011), 621–640. 120. Utz, S.; and Beukeboom, C.J. The role of social network sites in romantic relation- ships: Effects on jealousy and relationship happiness. Journal of Computer-Mediated Communication, 16, 4 (2011), 511–527. 121. Vakhitova, Z.I.; Reynald, D.M.; and Townsley, M. Toward the adaptation of routine activity and lifestyle exposure theories to account for cyber abuse victimization. Journal of Contemporary Criminal Justice, 32, 2 (2015), 169–188. 122. Vance, A.; Lowry, P.B.; and Eggett, D. Using accountability to reduce access policy violations in information systems. Journal of Management Information Systems, 29, 4 (2013), 263–290. 123. Vance, A.; Lowry, P.B.; and Eggett, D. Increasing accountability through user-interface design artifacts: A new approach to addressing the problem of access-policy violations. MIS Quarterly, 39, 2 (2015), 345–366. 124. Walker, J.; Craven, R.G.; and Tokunaga, R.S. Introduction. In S. Bauman, D. Cross, and J. Walker (eds.), Principles of Cyberbullying Research: Definitions, Measures, and Methodology. New York, NY: Routledge, 2013, pp. 32–34. 125. Weber, R. Editor’s comments. MIS Quarterly, 27, 3 (2003), iii–xii. 126. Weber, R. Evaluating and developing theories in the information systems discipline. Journal of the Association for Information Systems, 13, 1 (2012), 1–30. 127. Webster, J.; and Watson, R.T. Analyzing the past to prepare for the future: Writing a literature review. MIS Quarterly, 26, 2 (2002), xiii–xxiii. 128. Whetten, D.A. What constitutes a theoretical contribution? Academy of Management Review, 14, 4 (1989), 490–495. 129. Wilcox, P.; Gialopsos, B.M.; and Land, K.C. Multilevel criminal opportunity. In F. T. Cullen, and P. Wilcox (eds.), The Oxford Handbook of Criminological Theory. Oxford, UK: Oxford University Press, 2013, pp. 579–606. 130. Williams, R.; Elliott, I.A.; and Beech, A.R. Identifying sexual grooming themes used by Internet sex offenders. Deviant Behavior, 34, 2 (2012), 135–152. 131. Wills, S.; and McDougall, A. Reusability of online role play: Learning objects or learning designs? In L. Lockyer, S. Bennett, S. Agostinho, and B. Harper (eds.), Handbook of Research on Learning Design and Learning Objects: Issues, Applications and Technologies. Hershey, PA: IGI Global, 2009, pp. 761–776. 132. Winfree, L.T.; Bäckström, T.V.; and Mays, G.L. Social learning theory, self-reported delinquency, and youth gangs: A new twist on a general theory of crime and delinquency. Youth & Society, 26, 2 (1994), 147–177. 133. Wood, R.; and Bandura, A. Social cognitive theory of organizational management. Academy of Management Review, 14, 3 (1989), 361–384.

1178 LOWRY ET AL.

Copyright of Journal of Management Information Systems is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.