Decision Support Systems xxx (2013) xxx–xxx

DECSUP-12392; No of Pages 9

Contents lists available at ScienceDirect

Decision Support Systems

j ourna l homepage: www.e lsev ie r .com/ locate /dss

Social structural behavior of deception in computer-mediated communication

Jinie Pak ⁎, Lina ZhouDepartment of Information Systems, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, MD 21250, USA

⁎ Corresponding author.E-mail addresses: jpak3@umbc.edu (J. Pak), zhoul@um

0167-9236/$ – see front matter © 2013 Published by Elsehttp://dx.doi.org/10.1016/j.dss.2013.08.010

Please cite this article as: J. Pak, L. Zhou, Socia(2013), http://dx.doi.org/10.1016/j.dss.2013

a b s t r a c t

a r t i c l e i n f o

Available online xxxx

Keywords:DeceptionSocial structural approachDeception behaviorSocial network analysisComputer-mediated communication

Deception essentially takes place in social interaction.While deception has been studied from the perspective ofinterpersonal interaction, little is known about social structural characteristics of deceptive communication. Tofill the knowledge gap, this research investigates deception behavior in computer mediated communication(CMC) via the lens of social structure by answering the questions of how one deceiver socially interacts withmultiple receivers and what structural characteristics can be used to delineate deception in CMC. To this end,wefirst conceptualize deception in terms of social structure by drawing on the interpersonal deception and socialnetwork theories. We then propose a model of structural behaviors of deception in CMC that consists of threecomponents: centrality, cohesion, and similarity, followed by an empirical evaluation of the model with real-world data collected from a game website. The findings of this study provide new evidence that deception is astrategic activity where the deceiver juggles between the dual goals of promoting his or her deceptive agendaand avoiding detection.

© 2013 Published by Elsevier B.V.

1. Introduction

Computer mediated communication (CMC) provides social benefitsfor individuals and organizations to create, enhance, and re-discoversocial ties through interactive and transparent forms of communicatingand collaboratingwith others [1]. As the CMC technologies continuouslyevolve; however, online deception has become a growing threat to oursociety particularly due to the prevalence of online social networks,where awealth of sensitive information could be harvested and exploitedfor cyber-attacks on a large number of receivers (e.g., [23,50]). In addition,deception negatively impacts group decision making process throughhampering the decision making ability of others [34]. Therefore, there isan emerging need to understand deception behavior in the context ofmultiple receivers, which is instrumental to detecting deceptive informa-tion and uncovering malicious senders.

Deception behavior has traditionally been grouped into twomaincategories: verbal (e.g., negative affect) and non-verbal behaviors(e.g., facial expression) [13,22,84]. Verbal behavior is directly relatedto the spoken or written content and language [37,79], whereas nonver-bal behavior “focuses on accessory features that are exhibited while aperson is producing content” [78]. Given that text is the primarymodalityavailable in CMC, verbal behavior has been the focus of extant onlinedeception research. Despite the availability of nonverbal behavior inCMC, it has largely been under explored in the study of online deception.There has been very limited but promising evidence for the efficacyof non-verbal behavior in online deception detection [82]. In thestudy, Zhou and Zhang explored and empirically confirmed keyboard,

bc.edu (L. Zhou).

vier B.V.

l structural behavior of decep.08.010

participatory, and sequential behaviors being new channels of nonverbalcues to online deception, and called for research into new sources of non-verbal behavior of deception in CMC. To answer the call, the current studyexamines deception behavior via the lens of social structure.

This research looks into the context of CMC that involves one deceiverand multiple receivers. Deceptive communication encompasses back-and-forth interaction between a deceiver and receivers [51]. Accordingly,deception can be viewed as a social phenomenon where individuals areconnected through interactions and embedded in a structure of suchrelationships. The structure of ongoing social relations can be describedby social structural behavior [8], and thus are related to deceptive com-munication and may serve as a new source of online deception behavior.There are two fundamental questions that must be answered when ana-lyzing online deception from the social structure perspective. First, howcanwe conceptualize deceptive interactions as a social structure? Second,if social structure is a channel of deception behavior display,what kinds ofsocial behaviors can be used to discriminate deceptive from truthfulcommunication?

To address the above questions, we first conceptualized deceptivecommunication as social relationships between deceivers and receivers,and then proposed a research model of social structural behaviors ofdeception in CMCby drawing on the underpinnings of the interpersonaldeception theory and social network paradigms [6]. The model predictsthat deception has impact on the sender's centrality, cohesion, andsimilarity in a social structure. We further operationalized the selectedsocial structural behaviors with social network measures [64], and em-pirically validated the research model using real-world data collectedfrom a game website. Results support most of the hypothesized effectsof deception on social structural behaviors in CMC. Hereafter deceptionin CMC and online deception are used interchangeably.

tion in computer-mediated communication, Decision Support Systems

2 J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

The rest of this paper is organized as follows. In Section 2, we buildtheoretical foundation for a social structural approach to deceptive inter-action. Subsequently, we propose a research model of social structuralbehavior of deception. In Section 4, we introduce method design indetail, followed by data analyses and results reported in Section 5. InSection 6, we discuss the findings, implications and limitations of theresearch. Finally, we conclude the paper in Section 7.

2. A social structural approach to explaining deception

In this section, we argue that deception is essentially a type of socialinteraction, which lays the theoretical foundation for a social structuralapproach to explaining deception.

2.1. Deception as a type of social interaction

Social interaction is defined as a situation where an individual'sbehaviors are continuously reorganized by, and influence anotherindividual's behaviors, and vice versa [61]. As a series of processes, socialinteraction can be broken down into three processes: motivational,interactional, and structuring processes [61]. Specifically, motivationalprocesses indicate that individuals are compelled and driven to interactwith others; interactional processes involve actual influence on eachother's behaviors, signaling a course of behavior as well as interpretingboth one's own behavioral signals and those of others; and structuringprocesses denote that social interactions are repeatedly occurring acrosstime as well as organized (structured) in a physical space. Deceptionis a type of complex social interaction [58] that takes place betweenone or more senders and one or more receivers, and accordinglydeceptive communication involves the same set of processes. More-over, the motivational, interactional, and/or structuring componentsof deception have been well established in the deception literature[10,12,22].

According to the Interpersonal Deception Theory (IDT), deception isdefined as “a message knowingly transmitted by a sender to fostera false belief or conclusion by the receiver” [10]. As implied in thedefinition, interactivity and strategic communications are the two keyelements of deception in terms of the continuous influence of a sender'sbehavior on a receiver and vice versa. First, to varying degrees and indiverse ways, deceivers and receivers are motivated and mobilized tointeract with each other. Second, the two parties mutually influenceeach other's behavior with or without strategic moves [58] and suchstrategic moves are utilized in information, behavior, and image man-agement. During interaction processes, deceivers and receivers maysignal or leak unexpected words, nonverbal leakage, strategic thinking,and emotional stress [10]. While deceivers adjust their verbal and non-verbal behaviors or deception tactics based on the receiver's feedback orresponse [13], receivers' perceived credibility of deceivers and theirdetection accuracy have continuous effect on truth bias, context interac-tivity, and deceivers' communication skills [11,15,81]. Substantiallydeception involves a series of interactive process of monitoring andadjusting communication behavior based on mutual responses or feed-backs between deceivers and receivers, and their interaction patternsare structured by such repeated interactive processes. Thus, deceptionexemplifies the key processes of social interaction, and social interactionis essential to deception.

Extended from IDT, deceptive communication is not just a type ofsocial interaction but more of strategic interaction driven by deceptiveintent, leading to unique patterns of deceptive social interaction. Forinstance, deceivers are more likely to use control attempts when theyare negotiating their outcomeswith their partner orwhen they perceivethat their partner is questioning their decision [34]. IDT was originallyproposed from the interpersonal context, and accordingly many studieshave focused on deceptive communication in dyads. The theory has re-cently been extended to explain deception when the deceiver interactswith twoormore receivers [51,81]. This research expands thedeception

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

literature by focusing on deception that involves amulti-way communi-cation between one sender (i.e. deceiver) and multiple receivers.

2.2. Examining deception via the lens of social structure

Social structure has been one of the central concepts in social theoryand analysis. Despite the fact that social structure has been used insophisticated theoretical propositions or frameworks [4], there is nogenerally agreed upon definition of social structure; and the conceptof social structure is often implicitly assumed. Fundamentally social struc-ture is a network (structure) with a set of relations among actors in thatnetwork, and a structural approach studies social structure comprehen-sively by examining the patterns of embeddedness and connectednessof actors [27]. In social science, the social structural approach often refersto social network analysis (SNA) with emphasis on structural patterningin social networks (structures) [27,33,38]. SNAoffers analytical and statis-ticalmethods formeasuring patterns and structures of interaction amongsocial actors at different levels of the network such as ego networks andwhole networks [56].

An actor in a network can be defined or categorized by patterns ofhis/her relations with other actors, namely social role. To be specific, asocial role is a combination of particular sets of behavioral, meaningful,and structural attributes of a social actor [67], and the patterns of thoseattributes of social actors are relatively stable. SNA is used to uncoverthe power and influence of actors and to identify subgroups and theirsocial roles such as leaders, gatekeepers, and brokers [38,42,66]. Asdiscussed in Section 2.1, deception involves social interaction betweenindividuals. The social structural approachhelps us to identify deceptionbehavior by looking into theways that deceivers interactwith, influenceand are influenced by receivers, either directly or indirectly. In terms ofa social structural view of deception, deceivers and receivers are treatedas social actors and the chain of their strategic activities is interpreted asrelations in a social network. In other words, the ongoing iterative pro-cesses of cognitive and behavioral adjustments between deceivers andreceivers are interpreted as links (relationships) between them.

This study aims to examine deception in group communication fromthe social structural perspective based on two principles: 1) the deceiverand receivers can be modeled as social actors who are motivated bygoals, intentions, interests, or tasks; and 2) most deceptive interactionsconsist of the exchange of valued items (e.g. information or materials).Given that online deception research has been narrowly focused on ver-bal behavior with few exceptions (i.e., [76]), the social structural ap-proach to deception not only provides sociological explanations foronline deception but also provides a newavenue for deception detection.

3. The research model and hypotheses development

According to the social structural approach to deception, deceiversand truth-tellers are treated as distinct social roles, and their interac-tions, behavioral expectations, and structural relations are expected tobe different. Specifically, we hypothesize that the deceiver's deceptiveintent has an impact on three structural characteristics of his or her net-work: centrality, cohesion, and similarity. The research model ispresented in Fig. 1.

3.1. Centrality

Centrality is an important structural attribute of a social network thatsignals potential importance, influence, and prominence of an actor inthe network [28]. In other words, an actor's centrality in a network im-plies whether or not the actor has power or influence over other actorsin the network [44]. Centrality can be manifested in the embeddednessor connectedness of an actor in a network, which allows the actor to im-pose constraints on or provide opportunities for social interaction [38].Substantial evidence has shown that centrality measures can capturebehavioral tactics in organizations [62,8,7]. For example, assertive

tion in computer-mediated communication, Decision Support Systems

Deceptive Intent

CentralityIn-degreeOut-degreeBetweennessCloseness

Cohesion

Similarity

H1

H2

H3

Fig. 1. The research model.

3J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

behavior implies power, whereas upward appeal and ingratiation imply alack of resource or dependence on others.

The possible impact of deceptive intent on power or influence of adeceiver in a network is not straightforward, because in terms ofstrategic moves and control attempts in interaction, deceptive commu-nication is expected to be influenced by two types of strategies: persua-sive and protective strategies [10]. During deception, a deceiver engagesin greater strategic activity that is designed to manage information, be-havior, and image as the interaction unfolds. On the one hand, a deceiveraims to persuade or mislead others into a false belief about a certaintopic; on the other hand, the deceiver has to pretend to be innocent ortrustworthy in order to protect himself/herself from being caught in de-ception. These two dichotomous strategies in deceptive interaction fun-damentally affect behaviors of deceivers [10], including structuralbehaviors. In social interaction, a deceiver should juggle betweenbeing in a dominant position to deceive others (persuasive) and beingpassive and submissive to avoid detection [77]. Negotiation researchhas also shown that bargaining tactics typically deal with gaining infor-mation fromone's opponentwhile concealing information about oneself[39,43]. Therefore, we expect that the impact of deceptive intent on cen-trality vary with specific centrality metrics that have differing implica-tions for power or influence.

The simplest and intuitively obvious concept of centrality is based onthe degree of an actor, the number of nodes towhom an actor is directlytied. According to Freeman [28], degree centrality implies the visibilityor the potential for activity in communication. In directed networks,degree centrality can be divided into the degree count for incomingties (in-degree or prestige) and for outgoing ties (out-degree). Anactor with high in-degree centrality influences other actors as far ashis or her opinions are taken into account, and an actor with high out-degree centrality is in an advantaged position by making more choicesfor satisfying needs and being less dependent on other actors [16].Depending on the type of ties (e.g. supervisory relation and seekingadvice), either in-degree or out-degree measures may be interpretedas prominence or influence [60,75].

An actor, who intends to deceive others in an interactive contextwhere actors are ‘engaged’ with one another on a voluntary basis,should take a more influential role than a truth-telling actor. In such in-formal group CMC, however, an individual can take charge of whom heresponds to, but have little control over who responds to him. In searchengines, pages with higher in-degrees are more likely to be authorities[45]. Moreover, according to preferential attachment on directed graphs[18], nodes receive new links in proportion to their in-degree. Thus, in-degree centrality is a better indicator of prominence in a network ofsocial interaction than out-degree centrality is. Accordingly, the influence

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

of a deceiver can be manifested in his high in-degree, and the firsthypothesis is proposed as follows:

H1-a. Deceivers have a higher level of in-degree centrality thantruth-tellers.

Out-degree as well as in-degree centrality is eventually useful toexplaining the level of communication activity. Given that deceivers areless-forthcoming than truth-tellers [22], we argue that deceptive intentalso has effect on out-degree. In CMC, individuals who normally do nothave the ability to form a social relationship are free to share information[65], and the sharing takes place in informal communication flow andbarter of the interactions [21]. A deceiver is less likely to engage in infor-mation sharing because of his or her hidden deceptive agenda, but he orshe is more likely to suppress own communication activity or visibilityto protect himself/herself from suspicion of receivers [78,84]. Conse-quently, a deceiver speaks or writes less and in less detail. The deceiversattempt of holding back also helps him or her to minimize possible dis-crepancies resulting from his/her inability to control all aspects of behav-ior equally. Thus, we propose that:

H1-b. Deceivers have a lower level of out-degree centrality thantruth-tellers.

Betweenness centrality focuses on the extent to which actors arestrategically located on the shortest paths between other pairs of actors[26]. An actor can influence a group bywithholding or distorting informa-tion in transmission [57,7,35]. Such a positionwould offer the deceiver anadvantage in manipulating information, which is crucial to the success ofdeception. Nomatter whether themanipulation of information is accom-plished through falsification (deliberately creating false information),concealment (hiding relevant information), or equivocation (skirting is-sues by changing the subject or offering indirect responses) [10,24,2],the deceiver is expected to occupy a high-betweenness position in orderto control communication and coordinate group processes by conveyingfalse, incomplete, or vague information.

H1-c. Deceivers have a higher level of betweenness centrality thantruth-tellers.

Closeness centrality is measured as a function of geodesic dis-tance (shortest path) [28], namely the inverse of the average geode-sic distance between an actor and all other actors in the network. Thiscentrality suggests two intuitions about the role of an actor: control ofcommunication and efficiency of information flow [7]. First, the controlof communication represents a different dimension from other centrality

tion in computer-mediated communication, Decision Support Systems

4 J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

measures [75], which determines structural interdependence of an actorby the extent to which an actor can avoid the potential control of others[35]. For instance, an actor taking a non-central position in a network interms of closeness would depend on others as an intermediary of mes-sages. Second, the efficiency of information flow is positively associatedwith the notion of closeness in terms of fewer message transmissions,shorter time and lower costs in spreading information to other actors ina network. In general, deception requires more cognitively complex pro-cesses than stating the truth because a deceiver needs to not only comeup with deceptive messages that are different fromwhat he or she reallybelieves but also sift through his or her mind to find any hint of a seem-ingly plausible reason to justify the message in case of being questioned[10,76,77]. Deception is often described as an adversarial game, whichinvolves unexpected words, nonverbal leakage, strategic thinking, andemotion. Depending on the others' feedback or response, a deceiver hasto adjust his or her behavior and strategicmoves. As a result, the deceiveris more deliberate in self-regulating his or her behavior, thoughts andfeeling, and is more likely to usurp mental resources than a truth-teller[22], leaving him or her with less cognitive capacity available formaintaining a close relationship with the rest of the group. In addition,a study of terrorist networks has shown that their members are likely toprioritize security over efficiency over the execution of any single attack[54]. In a similar vein, a deceiver who interacts with multiple receiversis likely to prefer the protective strategy when faced with the tradeoffbetween efficiency and security [80].

H1-d. Deceivers have a lower level of closeness centrality thantruth-tellers.

3.2. Cohesion

Structural cohesion is an important feature of the relational dimen-sion of social solidarity and subgroups. Structural cohesion refers tothe psychological identification of members within a collectivity and arelational component, which shows the patterns of multiple connec-tions within the group [53]. The members of larger groups are rarelyconnected directly to all other members, and consequently subgroupsare likely to form [25]. For example, if kinship or friendship existsbetween actors in a group, they are connected to one another withstronger interpersonal ties, and accordingly this subgroup ismore cohe-sive, compared with other actors. The identification of such subgroupprovides an understanding of information and resource flow, access toinformation, and allocation of power in a social structure [29,30]. Tosupport subgroup identification, it is useful to understand how a cohe-sive group of actors is embedded in a network.

In a group that involves a single deceiver, the deceiver is the mi-nority. When trying to influence other group members, the minoritytends to teamupwithpeoplewhoare similar to self because it is easier forone to influence others when he or she is a part of in-group [49]. In addi-tion, the overall success of deception often hinges on building ‘trust’relationship with other members through initial displays. In order toestablish the trust relationship, an individual with deceptive intenthas to act like collaborating with others. Moreover, the establishmentof trust relationship at an early stage would help the deceiver to pushhis or her own agenda to his or her supporting group later. Studies ofcriminal networks have shown that the networks often can bepartitionedinto subgroups consisting of individuals who closely interact with oneanother [20,41,72]. Therefore, we propose the following hypothesis:

H2. Deceivers have a higher level of cohesion than truth-tellers.

3.3. Similarity

Structural similarities can be measured at the levels of individualactors, subgroups within the entire network, and the entire network[56]. Given the focus of this study on deception behavior, structural

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

similarity was measured at the level of individual actors (i.e., ego-network) as were other structural constructs. Structural similarity isfundamentally related to structural equivalence, which measures thedegree to which actors are similar. The construct has served to identifysets of actors who are very similar to one another, and distinct from ac-tors in other sets [38].

Friedkin [29] suggests that structural similarity is based on inter-personal solidity and identification. The more similar the two actors'structural positions, the more similar their initial orientations onissues are likely to be. Conversely, the more discrepant between actors'initial orientations on issues, the more dissimilar are their structuralpositions. In contrast with truth-tellers, deceivers approach the grouptask or the target issue with distinctively different intentions andmotives from other receivers. This sets deceivers' strategic and/or non-strategic choices in terms of with whom to interact and how often to in-teract apart from those of truth-tellers, which in turn lead to differentstructural embeddedness. Therefore, we propose the last hypothesis asfollows:

H3. Deceivers have a lower level of similarity to other receivers thantruth-tellers.

4. Research methods

In this section, we introduce research methods in detail, includingdata collection, social network construction, and operationalization ofsocial structure constructs.

4.1. Data collection

The data was collected from a mafia game website [63]. This gamecontains a deceiving role, who is motivated to deceive in order to winthe game. Additionally, the game proceeds by rounds, which allowsdeception to unfold over time. The simplest version of the game wasselected in this study to minimize possible confounding effects ofother factors. This game setup consists of seven players divided intotwo groups: mafia sided (i.e. mafia and hooker) and non-mafia sidedplayers (i.e. cop, watcher, and three villagers). The objective of themafia game is to eliminate the opponent group. We randomly selected72 games, and their outcomes are equally divided between the twoopposing groups.

The game is played via online chat rooms. Each round of the gameconsists of two stages: day and night. During the day, all playersdiscuss and make collective decisions about who the mafia might bevia voting. The player who gets the most votes would be eliminatedfrom the game. During the night, the mafia kills (eliminates) one ofnon-mafia sided players, and the cop selects one of the other survivingplayers for investigation and accumulates the identity of that player.The game continues to the next round until only one of the twoopposing groups has surviving members, and the surviving side wouldwin the game [9]. The game has set rules against disruptive acts suchas self-voting and inactive players. In such cases, players would receivewarning and suspension from the game. Further, a player who doesnot participate actively is more likely to become a target of early elimi-nation because players tend to vote randomly on the first day of thegame.

All the roles are randomly assigned. Each player is only aware of hisor her own role in the game except for the mafia sided players who areinformed of the identities of othermembers in the same group. In orderto win, the mafia player intends to deceive about his/her own andothers' true identity, while the cop is expected to share what he/shetruly knows about other players during a game. Therefore, the mafiaplayer was treated as the deceiver and the cop as the truth-teller inthe current study.

tion in computer-mediated communication, Decision Support Systems

5J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

4.2. Social network construction

To construct social networks from online chat sessions, we needto identify two types of elements: nodes and links. The nodes wereidentified by extracting unique identifiers of game players fromchat logs. The identifier of each player is automatically attachedto the beginning of messages sent by the player.

The identification of links between nodes, which is more complexthan node extraction, was implemented by inferring interactionalcoherence between chat messages. In this study, interactionalcoherence was analyzed with heuristic rules that infer ‘reply-to’relationships between exchanged messages. The rules were initiallydeveloped based on unique features of CMC such as disruptedturn-taking and ambiguity of sequential coherence [3, 40] and thenadapted and expanded for inferring underlying communicationrelationships among nodes [47,55]. Specifically, four types ofheuristic rules were used to infer ‘reply-to’ relationships, includingdirect addressing of users, temporal proximity, temporal density,and monitoring coherence of messages [45,49–51, 59].

In order to resolve the ambiguity of conversational and textualcoherence in group communication, message coherence was ana-lyzed based on both linguistic cohesion and conversational structure[31] in this study. Linguistic cohesion indicates semantic relationsbetween text elements in a discourse [32], which was assessedusing the following five cues [36]: reference (words used to referback to previously mentioned subjects), substitution (words usedinstead of another), ellipsis (connection becomes clear through theexclusion of certain words), conjunction (segments linked throughspecific linking words), and lexical cohesion (links created throughlexical repetition). Conversational structure, on the other hand, reflectsthe role that given sentences, phrases, or utterance play in a conversa-tion. More detail about interactional coherence analysis can be foundin Appendix A.

The procedure of social network construction is illustrated with thefollowing sample chat messages, where pi (i = 1,2,3) denotes identi-fiers of three different players.

p1: p2, are you mafia though?p2: Nop3: mafia herep3: btw

First, three unique identifiers of players were extracted fromthe chat messages, which make three nodes in the network. Sec-ond, an analysis of interactional coherence between chat mes-sages shows that the first message directly addresses p2, whichis immediately followed by a message from p2, and thus thereexists a ‘reply-to’ relationship between the second message andthe first message. Similarly, based on an analysis of coherenceof messages, the third message references and is in proximityto the first message, so a ‘reply-to’ relationship also exists be-tween the two messages. Finally, a social network is constructedthat consists of three nodes (p1, p2, p3) and two links (p2–p1,p3–p1).

Following the above procedure, we constructed social networksfor each of the selected games through a manual analysis of chatmessages. At the beginning, a subset of 14 games was randomlyselected and analyzed by two coders independently. An analysisof the inter-rater reliability of the two coding results showed thatthe kappa statistics was 0.64, which was moderately satisfactory.The inconsistent results were resolved through face-to-face dis-cussion, which led to refinement of some heuristic rules and chat-ting acts, as introduced above. The revised heuristic rules andchatting acts were then used to analyze the remaining games aswell as re-analyze the 14 games.

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

4.3. Operationalization of independent and dependent variables

Independent variable:

• Deceptive Intent: It is a binary variable with two possible values:deceiving (i.e., themafia player) and truth-telling (i.e., the copplayer).

Dependent Variables:

• Centrality: It was measured with four variables [5]:o In-degree: the number of edges that point toward the node of

interest.o Out-degree: the number of edges that point from the node of

interest.o Closeness: an inverse of the sum of the shortest distances from

the focal node to all other nodes.o Betweenness: the number of times a node occurs on a geodesic

path between a pair of other nodes.• Cohesion: Structural cohesion was operationalized as a dimension ofsocial embeddedness based on communication connectivity in termsof clustering coefficient [53], which measures the degree to whichactors in a network tend to cluster together. Clustering coefficient isdefined as the number of edges connecting a node's neighbors dividedby the total number of possible edges between the node's neighbors[5].

• Similarity: Structural similaritywas calculated based on the pattern ofrelations of the target actor to a group of other referencing actors [38],and the group of villagers was selected as the referencing actors in thisstudy. The similar wasmeasured as the Pearson correlation coefficient[69], where +1.00 indicates a perfect structural equivalence betweentwo nodes, and −1.00 an exact opposite.

5. Data analyses and results

A total of 72 seven-node networks were constructed, the values ofsocial structural variables were computedwith UCINET [5]. The descrip-tive statistics is reported in Table 1.

A paired-sample t-test was performed to test the hypotheses, andthe results are reported in Table 2. The results show that deceivershave a higher level of in-degree centrality (p b .05), betweenness cen-trality (p b .01), and clustering coefficient (p b .001), and a lower levelof closeness centrality (p b .01) and structural similarity (p b .01), incomparison with truth-tellers. However, the analysis on outdegree didnot yield significant result. Thus, hypotheses H2 andH3were supportedand hypothesis H1-a partly supported.

6. Discussion

6.1. Major findings and alternative explanations

The study examined deception behavior in CMC via the lens of socialstructure. Our proposed research model of social structural behavior ofonline deception was largely supported by the results of an empiricalstudy. The findings of this study show that deceivers find themselvesin different types of social structures characterized by centrality, cohe-sion, and similarity from truth-tellers.

As predicted, deception has an influence on most of the centralitymeasures. Specifically, deceptive intent has a positive impact onbetweenness centrality and a negative impact on in-degree and close-ness centralities. However, deception was not found to have an effecton out-degree centrality. We can provide the following alternativeexplanations. First, the out-degree centrality is used to measure theability of an individual to interact directly with many others, or tomake many others aware of his or her views [38]. Deception is drivenby intent, which is usually hidden or not easily observable to enhancethe chance of deception success [48]. Thus, in CMC where participantshave the control over the frequency and the targets of their interactions,

tion in computer-mediated communication, Decision Support Systems

Table 1Descriptive statistics of the social structural behaviors.

Social network measures Min Max Mean Std. dev. Std. error

T D T D T D T D T D

In-degree 0 0 7 7 4.01 4.62 1.674 1.706 .197 .201Out-degree 0 1 7 7 3.98 4.04 1.756 1.336 .207 .158Betweenness 0 0 13.33 6 1.35 2.62 2.167 2.008 .255 .267Closeness .09 0 .20 .20 .853 .056 .031 2.112 .004 .247Clustering coefficient 0 0 1.00 1.51 .669 .842 .236 .260 .028 .031Similarity − .237 − .276 0.424 0.340 .427 .131 .600 .705 .071 − .083

T: truth-teller; D: deceiver.

6 J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

a deceiver would shun away from disclosing his/her viewpoint in orderto protect himself/herself. Second, given that out-degree centralityreflects a level of productivity [46], the lack of effect of out-degreecentrality in this study confirms the finding of an earlier study of onlinedeception behavior [76]. Third, as suggested by previous research ondeception and adaptive interaction, deceivers tend to control andmain-tain appropriate conversational involvement depending on receivers'reactions and other contextual factors [68]. For instance, deceiversshownon-significant increase in response to increased receiver involve-ment, while compensating decreased receiver involvement. Thus, thecompensation and reciprocity relationship between deceivers andtruth-tellers in synchronous CMC would reduce the difference in theirlevels of involvement.

As expected, deceivers show a higher level of cohesion than truth-tellers, which is also consistent with the findings from the analyses ofterrorist networks [70,71,73]. Additionally, the finding that deceiversare less similar to other receivers than truth-tellers confirms our predic-tion, which provides fundamental evidence that deception leads to adifferent social structure.

6.2. Research implications

The findings of this study extend extant theories and research onexplaining the impact of deceptive intent on online behavior inmultipleaspects. First, to the best of our knowledge, this is the first study thatinvestigates deception behavior in CMC via the lens of social structure.To address one major limitation of previous studies of online deception[83], this research extends the scope of online deception behavior byexploring non-verbal behavior with respect to social structure. Second,our findings with regard to centrality, cohesion, and similarity of socialstructure of deceptive interactions provide empirical evidence in

Table 2Paired-sample t-test results.

Hypothesis Socialstructuralbehavior

Measures Meandifference(D–T)

Std.error

p-Value

H1-a In-degree .603 .259 .023⁎

H1-b Centrality Out-degree .041 .239 .864H1-c Closeness − .793 .247 .002⁎⁎

H1-d Betweenness 1.227 .371 .001⁎⁎

H2 Cohesion Clusteringcoefficient

.177 .038 .000⁎⁎⁎

H3 Similarity Correlationcoefficient

− .296 .105 .003⁎⁎

⁎ p b .05.⁎⁎ p b .01.⁎⁎⁎ p b .001.

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

support of the proposed social structural approach to deception ingroup communication. In addition, this research creates new knowl-edge about deception behavior concerning the social interaction be-tween the deceiver and receivers. Third, the differing impacts ofdeceptive intent on various centrality measures highlight the impor-tance of studying the centrality of deceptive communication in mul-tiple dimensions. Fourth, this research develops heuristic rules andchat acts based on interaction coherence analysis to identify ‘reply-to’ relations in CMC in social network construction. Last but not theleast, our findings provide empirical support for the generalizabilityof tradition deception theories [10,14,52,74] to CMC. The theoriespropose that deception is a strategic activity where the deceiverjuggles between the dual goals of promoting his deceptive agendato and preventing himself from detection and even suspicion bythe receivers.

Our findings that deception leads to higher in-degree and be-tweenness centralities and lower closeness centrality have two im-portant implications: 1) different centrality measures have differentimplications for the power and influence of deceptive communication;and 2) deceivers strategically position themselves in a social networkwhile leaking non-strategic behavior via social structure. Specifically,the finding on in-degree centrality of deceivers suggests that deceiversattempt to take a prominent position by engaging receivers in socialinteraction to gain trust and cooperation of the receivers; the findingon betweenness centrality implies that deceivers seek to allow, with-hold, or distort incoming and outgoing communication of receivers byexerting control over information transfer and diffusion in a network;and the finding on closeness centrality indicates that, while deceiversare more distant from and thus less efficient in accessing receivers incommunication networks than truth-tellers, deceivers are more inde-pendent of others' control or less reliant on others to get access to infor-mation. Consistent with the view that deception is a cognitively taxingprocess that engages both strategic and non-strategic behaviors duringinteractions, the findings on centrality as a whole imply that thedeceiver holds a favorable position as a mediator or a broker in anadversarial network that provides opportunities for him to persuadeand influence others through exclusive access to resources (e.g., informa-tion [17,35,38]) while simultaneously limiting his effort in reaching outto many receivers to protect his deceptive intent from getting noticedor arousing suspicion.

The finding that deceivers are more likely to belong to tightly knitgroups than truth-tellers suggests that deceivers form alliance withreceivers in order to gain credibility and get access to information. It ispart of a deceiver's strategic move to build a sense of trust and rapportwith receivers in the same group. Forming an alliance with receivers isa strategic move on the deceiver's side, which may help him/her topush for his/her agenda or false information.

Deceivers are structurally less similar to receivers than truth-tellersin terms of ties to receivers. This finding suggests that deceivers areembedded in a different type of social structure from truth-tellers.Moreover, it provides strong evidence for the validity of the social

tion in computer-mediated communication, Decision Support Systems

7J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

structural approach to studying online deception, and shed light onstructural behaviors of deceivers in CMC groups.

6.3. Practical implications

This research provides several practical implications. First, thestructural deception behavior identified in this study can be usedto enhance human users' ability to detect online deception and todevelop automatic deception detection solution. For instance, thefindings of the current study can be used to enrich trainingmaterial on deception behavior and increase awareness of onlinedeception for both professionals and laypersons in deception detection.Second, advancing our knowledge about online deception behavior orcues to online deception can aid promoting a more secure andtrustworthy cyberspace, where social media users may develop trustrelationships, online businesses may identify credible and influentialcustomers and gain knowledge about particular products or marketsfor viralmarketing, and government agenciesmay uncover and preventonline deception, fraud, and crime. Third, the proposedmethods for an-alyzing social interaction in CMC expand the application of social net-work analysis techniques to addressing broad online behaviors.

6.4. Limitations and future directions

This study exposes several limitations and raises importantresearch issues for further inquiry on deception behavior in CMC.First, in addition to the three types of social structural behaviorsexamined here, other characteristics of social structure such asembeddedness are worthy of exploration for online deception. Sec-ond, we did not consider the dynamics of social structural behaviorsdespite of the panel data collected in this study. A follow-up analysisof temporal patterns of structural behavior of deceptive communica-tion is warranted, using techniques such as dynamic network analy-sis [19]. A third limitation is that the social structural behaviors inour research model were operationalized on unweighted networks,which overlook the intensity of ties or relationships between actors.Fourth, this study is solely focused on social structural behaviorsof deceptive communication, and future studies should combine

Rules for inferring relationships among nodes.

Category Description

Linguistic feature Direct addressing If a message directly addresses another useis established between the sender and the

Reference If a message contains ‘you’ to reference somthe words or phrases mentioned in an earlis established between the senders of the m

Substitution If a message contains words as substitutesmentioned in an earlier messages, then a rebetween the senders of the messages.

Conjunction If two messages are connected with a conjuis established between the senders of the m

Conversational Feature Question–Answer If one message contains conversational initdeclarative question, and yes/no question),conversational response to the previous meand disagreement), then a relationship is eof the messages.

Statement/Opinion If one message contains a statement or ancontains a response (e.g., acknowledgmentrelationship is established between the sen

Appendix A

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

the new insights gained on structural behavior with prior findingson verbal behavior of online deception to improve the performanceof deception detection. Last but not the least, although our findingsextends the dimension of online deception behavior to social structure,we did not sample broadly across communication media. Futureresearch to replicate our findings in CMC settings with varying levelsof synchronicity should be conducted.

7. Conclusion

Deception hampers the effective use of CMC for both personaland business purposes. In this article, building on the notion thatdeception is a type of social interaction, we proposed a social struc-tural approach to investigating online deception behavior. Drawingfrom deception and social network theories, we developed aresearch model of structural behaviors of deception. The resultsof an empirical study largely supported the model, showing thatdeception influences three types of structural constructs — centrality,cohesion, and similarity. In addition, the impacts of deception on cen-trality varied depending on the type of centrality measures. Our find-ings provide new insights into deception behavior by looking intohow deceivers position themselves in a network structure establishedthrough social interaction. Specifically, deceivers gain power over othersin a network not by reaching out or getting publicity, but rather by themeans of controlling resource accessible to others, gaining prominence,and forming supporting groups through trust building. These findingshighlight that deception is a strategic activity where the deceiversmanage the trade-off between persuasive and protective strategies ininteraction with receivers. This research has important implicationsfor both research and practice as online deception continues to evolvein the networked economy.

Acknowledgments

This research is supported in part by UMBC DRIF, the National ScienceFoundation (IIS-1250395), and a Department of Education GAANNFellowship (P200A120100). Any opinions, findings or recommendationsexpressed here are those of the authors and are not necessarily those ofthe sponsors of this research.

Example

r's ID, then a relationshipaddressee.

A: B, are you mafia though?B: No

eone, or if a message repeatsier message; then a relationshipessages.

A: lynch this grammar naziB: I'm not a grammar nazi.C: I am a grammar nazi you know.

for something or someonelationship is established

A: Da is afk.B: hes not there

nction, then a relationshipessages.

A: I'll be on youB: and then the persons nameA: You be on me

iation (e.g., wh-question,and another message containsssages (e.g., answer, agreement,stablished between the senders

A: Do you have to have any special training?B: No, I haven'tA: Well, how old are you?B: 18 yrsA: So you're taking a government course?B: Not really?!?

opinion and a following message, accept, and reject), then aders of the messages.

A: He's probably, oh, a good two years old, big, old,B: Nah, He's about five months old Well,A: rabbits are darling.B: I think it would be kind of stressful.

tion in computer-mediated communication, Decision Support Systems

8 J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

References

[1] J.Q. Anderson, L. Rainie, The Future of Social Relations, Pew Internet & American LifeProject, http://www.pewinternet.org/Reports/2010/The-future-of-social-relations.aspxJuly 10 2010[accessed on May 10, 2012].

[2] L. Anolli, M. Balconi, R. Ciceri, Deceptive Miscommunication Theory (DeMiT): A NewModel for the Analysis of Deceptive Communication, in: R.C., G.R., L. Anolli (Eds.),Say Not to Say: New Perspectives on Miscommunication, IOS Press, 2001.

[3] T.Ö. Berglund,Disrupted turn adjacency and coherencemaintenance in instantmessag-ing conversations, Language @ Internet 6 (2009).

[4] P.M. Blau, Exchange and Power in Social Life, Transaction Publishers, 1964.[5] S.P. Borgatti, M.G. Everett, L.C. Freeman, UCINET 6 for Windows Software for Social

Network Analysis, Analytic Technologies, Harvard, MA, 2002.[6] S.P. Borgatti, P.C. Foster, The network paradigm in organizational research: a review

and typology, Journal of Management 29 (2003) 991–1013.[7] D.J. Brass, M.E. Burkhardt, Potential power and power use: an investigation of structure

and behavior, Academy of Management Journal 36 (1993) 441–470.[8] D.J. Brass, K.D. Butterfield, B.C. Skaggs, Relationship and unethical behavior: a social

network perspective, Academy of Management Review 23 (2012) 14–31.[9] M. Braverman, O. Etesami, E. Mossel, Mafia: a theoretical study of players and coa-

litions in a partial information environment, The Annals of Applied Probability 18(2008) 825–846.

[10] D.B. Buller, J.K. Burgoon, Interpersonal deception theory, Communication Theory 6(1996) 203–242.

[11] J.K. Burgoon, J.A. Bonito, B. Bengtsson, C. Cederberg, M. Lundeberg, L. Allspach, Inter-activity in human–computer interaction: a study of credibility, understanding, andinfluence, Computers in Human Behavior 16 (2000) 553–574.

[12] J.K. Burgoon, J.A. Bonito, A. Ramirez, N.E. Dunbar, K. Kam, J. Fischer, Testing the inter-activity principle: effects of mediation, propinquity, and verbal and nonverbal mo-dalities in interpersonal interaction, The Journal of Communication 52 (2002)657–677.

[13] J.K. Burgoon, D.B. Buller, Interpersonal deception: III. Effects of deceit on perceivedcommunication and nonverbal behavior dynamics, Journal of Nonverbal Behavior18 (1994).

[14] J.K. Burgoon, D.B. Buller, L. Guerrero, W. Afifi, C. Feldman, Interpersonal deception:XII. Information management dimensions underlying deceptive and truthful mes-sages, Communication Monographs 63 (1996) 50–69.

[15] J.K. Burgoon, T. Qin, The dynamic nature of deceptive verbal communication, Journalof Language and Social Psychology 25 (2006) 76–96.

[16] M.E. Burkhardt, D.J. Brass, Changing patterns or patterns of change: the effects of achange in technology on social network structure and power, Administrative ScienceQuarterly 35 (1990) 104.

[17] R.S. Burt, Thenetwork structure of social capital, Research inOrganizational Behavior 22(2000) 345–423.

[18] A. Capocci, V.D.P. Servedio, F. Colaiori, L.S. Buriol, D. Donato, S. Leonardi, et al.,Preferential attachment in the growth of social networks: the internet encyclopediaWikipedia, Physics Review E 74 (2006).

[19] K.M. Carley, Dynamic networks: what is a network? Knowledge Creation DiffusionUtilization, 2012. 1–63.

[20] H. Chen, W. Chung, J.J. Xu, G. Wang, Y. Qin, M. Chau, Crime data mining: a generalframework and some examples, Computer 37 (2004) 50–56.

[21] C.U. Ciborra, R. Andreu, Sharing knowledge across boundaries, Journal of Informa-tion Technology 16 (2001) 73–81.

[22] B.M. DePaulo, J.J. Lindsay, B.E. Malone, L. Muhlenbruck, K. Charlton, H. Cooper, Cuesto deception, Psychological Bulletin 129 (2003) 74–118.

[23] H. Du, S.J. Yang, Discovering collaborative cyber attack, SBP2011, Springer, 2011,pp. 129–136.

[24] P. Ekman, Lying and deception, in: N.L. Stein, P.A. Ornstein, B. Tversky, C. Brainerd(Eds.), Memory for Everyday and Emotional Events, Lawrence Erlbaum Associates,Mahwah, New Jersey, 1997.

[25] D.R. Forsyth, Group Dynamic, Wadsworth, Cengage Learning, Belmont, CA, 2010.[26] L.C. Freeman, Centrality in social networks conceptual clarification, Social Networks

1 (1979) 215–239.[27] L.C. Freeman, The Development of Social Network Analysis: A Study in the Sociology

of Science, Empirical Press, 2004.[28] L.C. Freeman, D. Roeder, R. Mulholland, Centrality in social networks: II, Experimental

Results 2 (1979) 119–141.[29] N.E. Friedkin, Structural bases of interpersonal influence in groups: a longitudinal

case study, American Sociological Review 58 (1993) 861–872.[30] N.E. Friedkin, Social cohesion, Annual Review of Sociology 30 (2004) 409–425.[31] T. Fu, A. Abbasi, H. Chen, A hybrid approach toweb forum interactional coherence anal-

ysis, Journal of the American Society for Information Science 59 (2008) 1195–1209.[32] T. Fu, H. Chen, Analysis of Cyberactivism: Analyzing Content Development and A

Case Study of Online Free Tibet Activities Visualizing Social Interactions in WebForum, Artificial Intelligence (2008) 0–5.

[33] M.M. Fuller, A. Wagner, B.J. Enquist, Using network analysis to characterize foreststructure, Natural Resource Modeling 21 (2008) 2–4.

[34] G.A. Giordano, J.S. Stoner, R.L. Brouer, J.F. George, The influences of deception andcomputer-mediationondyadic negotiations, Journal of Computer-MediatedCommuni-cation 12 (2007) 362–383.

[35] E.M. Hafner-burton, A.H. Montgomery, Centrality in Politics: How Networks ConferInfluence, Human Rights, , 2010. 1–21.

[36] M.A.K. Halliday, R. Hasan, Cohesion in English, Longman, London, 1976.[37] J.T. Hancock, L. Curry, S. Goorha, M. Woodworth, On lying and being lied to: a

linguistic analysis of deception in computer-mediated communication, DiscourseProcesses 45 (2008) 1–23.

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

[38] R.A. Hanneman, M. Riddle, Introduction to Social Network Methods, PublishedUniversity of California, Riverside, CA, 2005.

[39] K. Hausken, Game-theoretic and Behavioral Negotiation Theory, Group Decision andNegotiation 6 (1997) 511–528.

[40] S.C. Herring, Interactional coherence in CMC, Proceedings of the 32nd AnnualHawaii International Conference on Systems Sciences, IEEE, 1999, p. 13.

[41] C.E. Hutchins, M. Benham-Hutchins, Hiding in plain sight: criminal networkanalysis, Computational and Mathematical Organization Theory 16 (2009)89–111.

[42] C. Kadushin, Introduction to Social Network Theory: Some Basic Network Conceptsand Propositions, 2004.

[43] P.H. Kim, Power dynamics in negotiation, Academy of Management Review 30(2005) 799–822.

[44] C. Kiss, M. Bichler, Identification of influencers — measuring influence in customernetworks, Decision Support Systems 46 (2008) 233–253.

[45] J. Kleinberg, Authoritative sources in a hyperlinked environment, Journal of the ACM46 (1999) 604–632.

[46] Z. Kozareva, E. Riloff, E. Hovy, Semantic class learning from the web with hyponympattern linkage graphs, Proceeding of ACL-08:HLT, 2008, pp. 1048–1056.

[47] I. Lakatos, The Methodology of Scientific Research Programmes: PhilosophicalPapers, Cambridge University Press, 1978.

[48] D. Li, E. Santos, Deception detection in human reasoning, IEEE International Conferenceon Systems Man and, Cybernetics, 2011, pp. 165–172.

[49] A. Maass, R.D. Clark, Hidden impact of minorities: fifteen years of minority influenceresearch, Psychological Bulletin 95 (1984) 428–450.

[50] M. Maccomascaigh, T. Bell, J. Murphy, Magic Quadrant for Web Content Manage-ment, Management, 2010. 1–24.

[51] L.K. Marett, J.F. George, Deception in the case of one sender and multiple receivers,Group Decision and Negotiation 13 (2004) 29–44.

[52] S. McCornack, Information manipulation theory, Communication Monographs 59(1992) 1–16.

[53] J.Moody, D.R.White, Structural cohesion and embeddedness: a hierarchical conceptionof social groups, American Sociological Review 68 (2003).

[54] C. Morselli, C. Giguere, K. Petit, The efficiency/security trade-off in criminalnetworks, Social Networks 29 (2007) 143–153.

[55] P. Mutton, Inferring and visualizing social networks on internet relay chat,Proceedings. 8th International Conference on Information Visualisation, IEEE,2004, pp. 35–43.

[56] J. Scott, Social network analysis: a handbook, Contemporary Sociology 3 (2000)208.

[57] M.E. Shaw, Group structure and the behavior of individuals in small groups, TheJournal of Psychology 38 (1954) 139–149.

[58] K.E. Sip, J.C. Skewes, J.L. Marchant, W.B. McGregor, A. Roepstorff, C.D. Frith, What if Iget busted? Deception, choice, and decision-making in social interaction, Frontiersin Neuroscience 6 (2012) 58.

[59] A. Stolcke, K. Ries, N. Coccaro, E. Shriberg, R. Bates, D. Jurafsky, et al., Dialogue actmodeling for automatic tagging and recognition of conversational speech, Computa-tional Linguistics 26 (2000) 339–373.

[60] M.T. Torfason, J.A. Kitts, Prominence, Encyclopedia of Social Networks, Sage Publica-tions, George Bar, New York, 2010.

[61] J.H. Turner, A Theory of Social Interaction, Stanford University Press, Stanford,California, 1988.

[62] R. Vempati, V.R. Krishnan, Power and influence strategies: an analysis across depart-ments, NMIMS Management Review 12 (2000) 1–10.

[63] P. Wang, Epic Mafia, http://wiki.epicmafia.com/index.php?title=Main_Page2012.[64] S. Wasserman, K. Faust, Social network analysis: methods and applications, Social

Networks 8 (1994) 825.[65] B. Wellman, Designing the Internet for a networked society, Communications of the

ACM 45 (2002) 91–96.[66] B. Wellman, Network analysis: some basic principles, Network 1 (2007) 155–200.[67] H.T. Welser, Visualizing the signatures of social roles in online discussion groups

finding social roles in online discussion, 8 (2007) 1–32.[68] J.K. White, C.H. Burgoon, Adaptation and communicative design: patterns of interac-

tion in truthful and deceptive conversations, Human Communication Research 27(2001) 9–37.

[69] S. Williams, Pearson's correlation coefficient, The New Zealand Medical Journal 109(1996) 38.

[70] J.J. Xu, H. Chen, CrimeNet Explorer: A Framework for Criminal NetworkKnowledge Discovery, ACM Transactions on Information Systems 23 (2005)201–226.

[71] J.J. Xu, B. Marshall, S. Kaza, H. Chen, Analyzing and visualizing criminal networkdynamics: a case study, ISI 2004 LNCS 3073, Springer-Verlag, 2004, pp. 359–377.

[72] C.C. Yang, N. Liu,M. Sageman, Analyzing the terrorist social networkswith visualizationtools, Intelligence and Security Informatics, 2006, pp. 331–342.

[73] C.C. Yang, T.D. Ng, Terrorism and crime related weblog social network: link, contentanalysis and information visualization, 2007 IEEE Intelligence and Security Informatics,IEEE, 2007, pp. 55–58.

[74] L.N.T. Yeung, T.R. Levine, Information manipulation theory and perceptions ofdeception in Hong Kong, Communication Reports 12 (1999).

[75] B. Zemljič, V. Hlebec, Reliability of measures of centrality and prominence, SocialNetworks 27 (2005) 73–88.

[76] L. Zhou, An empirical investigation of deception behavior in instant messaging, IEEETransactions on Professional Communication 48 (2005) 147–160.

[77] L. Zhou, J.K. Burgoon, J.F. Nunamaker, D. Twitchell, Automating linguistics-basedcues for detecting deception in text-based asynchronous computer-mediated com-munications, Group Decision and Negotiation 13 (2004) 81–106.

tion in computer-mediated communication, Decision Support Systems

9J. Pak, L. Zhou / Decision Support Systems xxx (2013) xxx–xxx

[78] L. Zhou, J.K. Burgoon, D.P. Twitchell, T. Qin, J.F. Nunamaker, A comparison of classi-fication methods for predicting deception in computer-mediated communication,Information Systems Management (2004) 139–165.

[79] L. Zhou, J.K. Burgoon, D. Zhang, J.F. Nunamaker, Language dominance in interperson-al deception in computer-mediated communication, Computers in Human Behavior20 (2004) 381–402.

[80] L. Zhou, Y. Sung, D. Zhang, Deception performance in online group negotiation anddecision making: the effects of deception experience and deception skill, GroupDecision and Negotiation 22 (2013) 153–172.

[81] L. Zhou, D. Zhang, A comparison of deception behavior in dyad and triadic groupdecision making in synchronous computer-mediated communication, Small GroupResearch 37 (2006) 140–164.

[82] L. Zhou, D. Zhang, Typing or messaging? Modality effect on deception detectionin computer-mediated communication, Decision Support Systems 44 (2007)188–201.

Please cite this article as: J. Pak, L. Zhou, Social structural behavior of decep(2013), http://dx.doi.org/10.1016/j.dss.2013.08.010

[83] L. Zhou, D. Zhang, Automatic deception detection in computer-mediated communi-cation, IEEE Intelligent Systems (2012) 73–75.

[84] M. Zuckerman, B.M. DePaulo, R. Rosenthal, Verbal and nonverbal communication ofdeception, in: L. Berkowitz (Ed.), L. Berk (Ed.), Advances in Experimental SocialPsychologyAcademic Press, New York, 1981, pp. 1–59.

Jinie Pak is a Ph.D doctoral candidate student of Information Systems at the University ofMaryland, Baltimore County, USA. She received hermasters degree in Information Systemsfrom the University of Maryland, Baltimore County. Her research interests include onlinedeception detection, social network analysis, dynamic network analysis, and social media.

Lina Zhou is an Associate Professor of Information Systems, University of Maryland,Baltimore County, USA. Her current research interests include deception detection,ontology, social network analysis, and mobile web.

tion in computer-mediated communication, Decision Support Systems