TY - JOUR
AU - Shin, Wonsun
AB - Abstract In dyadic online chats with customers, agents commonly employ scripted responses and converse with several customers simultaneously in order to enhance efficiency. These techniques, however, can affect dimensions of interactivity—conversational contingency and response latency—undermining interpersonal assessments, satisfaction, and organizations’ relationships with customers. This research incorporates aspects of interactivity to the social information processing (SIP) theory of computer-mediated communication, that addresses conversational behaviors that affect interpersonal relations in the absence of nonverbal cues. In a 2 × 2 between-subjects experiment, observers watched one of four versions of a dialogue between a customer and sales support agent, which differed with respect to the agent’s response latency and conversational contingency. Results confirmed deleterious effects of non-contingency on outcomes. Contingency moderated latency effects. Mediation analyses showed indirect effects of contingency via interpersonal judgments on organization/customer relations. Implications for a more comprehensive approach to SIP conclude the study. Real-time computer-mediated communication (CMC) chat systems have become a popular way for organizations to communicate with customers. Industry analysts claim that chats not only lower interaction costs with customers while providing critical purchase- and support-related answers, they also foster positive relationships between organizations and consumers (Cole, 2017). At the same time, there is growing recognition that variations in chat experiences can lead to frustration rather than elation, infuriate users, and turn potential customers from fans to detractors. Experts are aware that customers prefer certain aspects of responsiveness in a chat, including small wait times for a chat to initiate, and “warm, human help” that is not robotic in tone (Levin, 2017). Ironically, some of the very communication strategies that organizations use to make chat efficient may actually incur negative relational outcomes. Organizations often direct their representatives to conduct several CMC conversations by their representatives with multiple customers simultaneously. Such uses may impede agents’ responsiveness to each customer with whom they are engaged, degrading response speed and/or content relevance. Further, in order to mitigate this problem, organizations’ chat agents often use scripted responses or “shortcuts,” i.e., pre-written blocks of text that they can insert into a chat as a response to customers’ questions. Because scripted responses are generic, however, their relevance to a customer’s specific question are dubious, and may be more obscure than a genuine, original answer to a customer’s question. From a communication perspective, these two issues—temporal responsiveness and the relatedness of replies to users’ messages—constitute two subdimensions of interactivity: response latency and conversational contingency. The meta-construct, interactivity, has a substantial history in mediated and non-mediated communication research (Walther, 2017). At the same time, the effects of interactivity on relational processes have been studied less often in connection with specific theoretical frameworks. The present study examines the two particular aspects of interactivity in an online chat—latency and contingency—for their potential to affect relational judgments about both a chat agent and the organization with whom the agent is associated. Response latency refers to the time interval between when one message is sent and the reply it engenders appears (see Walther & Tidwell, 1995, for a review). In the context of an organization-customer online chat, it is the time between the moment a customer sends a message to an organizational representative and the moment the representative’s reply reaches the customer. Conversational contingency refers to the degree to which responses over the course of a conversation implicitly refer and semantically relate to the content of prior statements that elicited responses (Sundar, Bellur, Oh, Jia, & Kim, 2016). It is not the denotative or connotative value of content, but a property of the interrelationship among respective interlocutors’ messages that renders content interpretable; it is the threadedness of a conversation beyond a simple question/answer adjacency pair, in which the meaning of some utterances depend on and implicate the meanings set forth in prior utterances. The social information processing (SIP) theory of CMC (Walther, 1992) provides a framework with which to understand the impact of these dimensions of interactivity on impression development and relational evaluations in CMC. The theory posits that communicators can adapt the content, style, and timing of messages online to convey and to infer interpersonal impressions and relational messages of the nature that face-to-face communicators rely primarily on nonverbal communication to accomplish. Previous research on SIP has demonstrated how online content conveys immediacy and affection, receptivity and involvement, liking, and attraction, among other outcomes. The application of SIP to interactivity and chat requires extensions to the theory that are consistent with its framework, yet previously unarticulated. Specifically, although the theory has been applied to various forms of online conversational content and online chronemics, it has yet to be applied explicitly to non-content, implicit aspects of messages conveyed by conversational qualities such as the contingency dimension of interactivity, or interactions among contingency and latency. Another extension to SIP’s scope has been proposed but remains untested: Pang, Shin, Lew, and Walther (2018) argued that the relational outcomes that derive from CMC encounters with organizations’ agents affect not only evaluations of specific interaction partners, but affect participants’ image of the entire organization with which the agent is associated (see also Dou & Sundar, 2014). This study provides a theory-based explanation for inadvertent problems in a practical setting, and prospective solutions to these problems as well. It also addresses and extends understanding about interactivity in CMC by examining response latency and conversational contingency and their mutual effects on interpersonal attraction, satisfaction, and perceived organizational/customer relationships. Finally, the anticipated and unanticipated interaction effects present challenges for future SIP research. The findings urge greater attention to how numerous code systems operate within and between CMC messages, with the potential of these codes and sub-codes to complement or undermine one another, just as nonverbal messages do in multimodal channels. Interactivity and social information processing The concept of interactivity has generated a considerable amount of research in CMC. The term, interactivity, has numerous meanings across domains of human–computer interaction, website feature design, and human–human interaction via CMC and face-to-face. Generally, interactivity is the responsiveness one experiences from another entity, be it a computer system or person. Research on interactivity has examined the degree of user control and website responses (e.g., McMillan & Hwang, 2002), expectations for audience responses among different social media platforms (Zhao, Lampe, & Ellison, 2016), or the effects of different channels, platforms, and interface characteristics on promoting perceptions of interactivity (see Sundar et al., 2016 for a review). Despite the longstanding association of interactivity with technological systems, Rafaeli (1988) argued that the construct is not tied to CMC: It is a property of language-based conversations. Interactivity is neither guaranteed nor prevented by the communication medium. While real-time chat systems are strongly associated with interactivity (and in Sundar et al.’s (2016) research, adding a real-time chat system to a web interface significantly increased perception of interactivity), Rafaeli (1988) and others point out that interactivity objectively varies based on differences in communicators’ conversational behaviors, whether face-to-face or via other media. Indeed, assessing interactivity’s effects across a variety of communication modalities including CMC and face-to-face interaction, Burgoon et al. (2000) found that lesser interactivity negatively affected perceptions of a conversational partner’s expertise, dependability, trust, and attraction. No matter the media, it appears, differences in interactivity may be associated with a variety of important interpersonal impressions and outcomes. Yet for all the recent research on interactivity in technology, little theoretical connection appears specifying how or why interactivity per se affects these outcomes. Rafaeli and Sudweeks (1998) suggested that greater online interactivity affects acceptance and satisfaction with a conversation, but they offered no explanation or evidence for this assertion. Research in interpersonal communication provides useful benchmarks, however. “Historically,” according to Sundar et al. (2016), “concepts of dialogue, mutual conversation, and feedback have dominated theories (…) of interpersonal communication, with fervent attempts to incorporate these tenets into mediated communication as and when allowed by developments in interactive technologies” (p. 596). Interpersonal communication research has examined more specific aspects or subdimensions than a global interactivity construct, however, and exploring several of these subdimensions offers more precision with which to study interactivity’s effects. Before turning to these specific dimensions, we discuss the consideration of interactivity as a source of social information in CMC. SIP theory (Walther, 1992) explains how communicators exhibit and interpret social information online without the nonverbal cues that generally provide this information in multi-modal media and face-to-face encounters, in order to form impressions and exchange relational communication. It asserts that communicators exchange social information through whatever cue systems are available to them, and if physical nonverbal cues are unavailable (as in text-based chat), they rely primarily on verbal, linguistic, textual (Walther, 1992), and chronemic cues (Walther & Tidwell, 1995). It also asserts that this process is affected by time and rate of communication, which are determined in part by the symbol-carrying capacity of the communication medium. Although text-based cues are more potent in CMC than face-to-face interaction, messages transpire more slowly and with less social information per transmission than in face-to-face interaction where a variety of cue systems can convey a multitude of messages in a single utterance. Early SIP research focused on individuals’ interpersonal perceptions and relational evaluations of online verbal content that mediated the interaction effects of communication medium by time, in response to earlier CMC research that used content analysis as evidence of the main effects of the medium. Its research quickly incorporated discourse functions such as self-disclosure, dominance-seeking, and other characteristics such as chronemic response latencies, language immediacy, and other conversational aspects affecting the socioemotional tone of online messages exchange processes that affect the extent to which CMC achieves (or fails to achieve) relational states comparable to those that may occur face-to-face (see Walther, Van Der Heide, Ramirez, Burgoon, & Peña, 2015, for a review). By subsuming conversational contingency within SIP, the present research extends the scope, explanatory power, and organizing power of the theory to another quality of online discourse. However, the theory is potentially broader than the effects of any single form of discourse, verbal content, or style, as the interaction effect hypotheses and results of this study demonstrate. Although previous SIP research has involved one form of interactivity—chronemic response latencies—that research has been limited. Research using SIP has not considered the other dimension of interactivity on which the present study focuses, conversational contingency. As both of these dimensions are expected to affect relational tone in CMC, they are fitting foci for SIP’s framework, and the inclusion of contingency provides an extension of the theory in terms of a new mechanism through which SIP principles may operate, i.e., how message behaviors available in CMC affect impressions and relations in the absence of face-to-face nonverbal cues. This work also embeds certain dimensions of the interactivity construct in a more elaborate theoretical framework than has tended to be done in the past. This is not to suggest that every quality that has been ascribed to the term, interactivity, should or could be integrated with SIP. For one thing, “there is little consensus among researchers about the definition of interactivity,” according to van Noort, Vliegenthart, and Kruikemeier (2016, p. 353). Other approaches are often more platform-dependent; they may be connected to aspects of website control and customization in addition to or instead of conversational qualities (e.g., Liu & Shrum, 2002). Response latency and contingency have theoretical and empirical roots in interpersonal communication that pre-date, yet inform, CMC research. Chronemic response latency One important feature of interactivity is response latency, i.e., the speed with which responses appear. Walther and Tidwell’s (1995, p. 356) original study of time cues in CMC concluded, “the amount of time between one message and the next (has) great potential to affect the judgments we make of those who initiate or respond to communication” in CMC. Research on response latencies is found in both face-to-face and CMC contexts. In face-to-face encounters, a response lapse over 3 seconds can constitute an interaction failure (McLaughlin, 1984), and speakers who wait 4–10 seconds to reply are perceived as less attractive and competent, with lower social skills, than those who do not (see McLaughlin & Cody, 1982, for a review). Within CMC, Kalman, Scissors, Gill, and Gergle (2010) argued that the lack of other nonverbal cues may increase the potency of time cues on social judgments relative to their effects in offline communication. In real-time CMC group applications, significant pauses between comments led to less trust among participants (Kalman, Scissors, & Gergle, 2013). Similar findings exist in studies of multi-user chats in virtual worlds (e.g., Cherny, 1999). Although some response latency in CMC is excused through attributions about the communicator’s typing skill or system delays in receiving the other’s message, these excuses are more likely to be granted for tardy email responses than delays during live conversations such as real-time chats (Turner & Reinsch, 2010). Most CMC research on response latencies has examined asynchronous CMC, where some delay is expected. Park and Sundar (2015) also examined agents’ responses in customer service interactions using asynchronous CMC with no delay, a one hour delay, or a six hour delay. Faster responses improved perceptions of copresence and service quality. In other studies, the effect of latency was moderated by context or a quality of a message sender. Walther and Tidwell (1995) found that email messages between an organizational supervisor and subordinate that appeared to be separated by either 4 minutes, or by 24 hours and 4 minutes, led to significant differences in the level of affection observers attributed to the exchanges in a task-related context; in a social context, the effect was reversed. Research by Sheldon, Thomas-Hunt, and Proell (2006) and Kalman and Rafaeli (2011) found that evaluations of a job applicant who sent email about an interview after a longer (rather than shorter) period of time were moderated by the individual’s attractiveness for the job: A slow latency was forgiven for a promising applicant, whereas a slow latency was injurious for a less attractive applicant. Response latency in real-time chat applications Practitioners using real-time online chats for customer support seem to be attuned to some aspects of latency, but not others. Chow (2016) and Klimczak (n.d.) describe industry studies showing that customers lose patience and are less satisfied with real-time customer support when a chat agent does not initiate conversation within a minute after customers request it. Beyond initiating chats, however, no attention seems to have been paid to the matter of response latency within chat sessions. Among other causes, latencies can occur when individuals engage in several dialogues at once. Yet many applications of real-time chat for customer support involve the deliberate allocation of a number of simultaneous customer chat sessions to an individual chat agent. This approach is promoted by purveyors of customer chat systems, and “best practices” literature in that industry describes the efficiency of such arrangements: “People working with online visitors using live chat solution are more efficient than sales clerks (…) or call center agents. They are able to handle multiple conversations at once (…)” (Klimczak, n.d., n.p.). Because organizational agents who staff multiple, simultaneous chats must juggle these conversations, they seem prone to committing delayed or extended response latencies. Conducting multiple conversations divides users’ attention and distracts them (Reinsch, Turner, & Tinsley, 2008). Research on multicommunicating—organization members carrying on multiple, simultaneous conversations using different media—suggests that while some multicommunicating may be stimulating and performance-enhancing, attempts to manage many simultaneous conversations online can lead to a “precipitous decline” in communication performance, including periods of silence, that damage impressions others make of the multicommunicating individual (Reinsch et al., 2008, p. 395). When an individual suspects s/he is not the only one getting an agent’s attention, a multicommunication episode is considered unsuccessful (Turner & Reinsch, 2010). In summary, the literature from traditional communication as well as CMC suggests that inordinate response latencies incur negative interpersonal and professional attributions. The current experiment explores the extent to which delays in an agent’s real-time chat responses affect the level of interpersonal attraction observers attribute to an agent, which in turn affects satisfaction with the agent’s performance. H1: A faster response leads to more attractive perceptions of a responder than a slower response. H2: A faster response leads to greater satisfaction than a slower response. Another recent extension to the scope of SIP theory suggests that individuals tend to extrapolate their perceptions of a specific organizational agent’s relational communication to an organization’s relationship with customers and other stakeholders as a whole (Pang et al., 2018). The dialogic characteristics exhibited by an organizational representative can thereby affect an entire organization’s image insofar as individuals are concerned. Therefore, the current study also tests the effect of response latency on perceptions of an organization’s customer relationship: H3: An organization whose agent provides a faster response is perceived to have a better relationship with customers than one whose agent responds with a slower response. Content interactivity: contingency Another quality of interactivity is the level of contingency among message sequences. Rafaeli (1988) defined this aspect of interactivity as the content-level inter-relationship of statements within a series of utterances, such that the meaning of a statement depends on and refers implicitly to ideas that were expressed in utterances prior to those statements. Burgoon et al. (2000) focused on contingency as the most important content-related form of interactivity. Contingency, they said, is the degree to which the meaning of one conversant’s messages depend on the prior messages of a co-conversant (Burgoon et al., 2002).1 Research on contingency in face-to-face communication falls under the label of conversational coherence. Kellermann and Sleight (1989) suggested coherence occurs when messages meet expectations that are established by prior utterances in a manner that “one utterance follows another in a rule-governed manner (and) each succeeding sentence connect with what has already been introduced” (p. 95). The degree to which a communicator exhibits this quality affects receivers’ judgments about that communicator. Early CMC research asked whether and how conversation coherence can be achieved without the nonverbal cues that facilitate conversation management offline (see Herring, 1999, for a review). Some researchers asserted that without nonverbal cues, conversational contingency is impossible in CMC (e.g., McGrath, 1990). Subsequent research demonstrated that CMC users employ a variety of methods to overcome CMC’s limitations and achieve conversational contingency, such as adjacency pairs and threading (Anderson, Beard, & Walther, 2011; Condon & Cech, 2001). Various features of CMC platforms can make contingency among messages perceptually more salient, for instance, when a system displays the conversational history of asynchronous exchanges, or when real-time chat messages persist on-screen for some time (Sundar et al., 2016). Although the perception of contingency that may result from these displays is definitionally different than actual semantic contingency among utterances characterized by conversational coherence, perceived contingency has also been shown to mediate the effect of actual semantic message responsiveness on evaluations of the message source (Sundar et al., 2016). Contingency in real-time chat applications Many applications of real-time chat for customer support include techniques to make the chats more efficient. In addition to promoting agents handling multiple chats simultaneously, chat system developers have devised computational scripting techniques to provide customers with “canned responses” (Klimczak, n.d.), or “shortcuts” (Basu, 2014, p. 1). Executable by typing an abbreviation, shortcuts display messages in a chat as though they were typed by the agent, from simple greetings to complex, detailed descriptions of products or services. One trade publication (Basu, 2014, p. 3) offers this example of a canned response for a travel-related business’s chat system: Our package to London costs $1000 and lasts for 5 days and 4 nights. We will put you up in a gorgeous five star hotel right on the river Thames and near all the attractions. There are undoubtedly advantages to using scripts in online chat, to save time, and to reduce errors (Basu, 2014; LiveChat, n.d.). However, when an agent executes a script that is only partially responsive to a customer’s question, it can lack contingency. A script such as the one above might be executed in answer to a customer’s hypothetical question, “What hotel is included in the London package?” The scripted response is not entirely responsive to the question, and some of its content is superfluous. In our experience, chat agents often provide generic replies in response to relatively more specific questions. They seem more impersonal, disingenuous, and/or incompetent the more frequently they occur within a chat encounter. Whether chat responses from an agent are scripted or original, however, their degree of contingency, like other forms of implicit relational messages, should affect interpersonal impressions of the agent and subsidiary evaluations, from a SIP perspective: H4: Responses that are more contingent create more attractive perceptions of an agent than responses that are less contingent. H5: Responses that are more contingent create greater satisfaction than responses that are less contingent. H6: An organization whose agent provides more contingent responses is perceived to have a better relationship with customers than one whose agent provides responses that are less contingent. The SIP formulation not only specifies main effects of conversational properties on interpersonal evaluations; it is also sensitive to the interactions of temporal factors and conversational behaviors on relational evaluations. In this case, the combination of response latency and contingency may produce additive effects on evaluations. That is, the predicted main effects may simply combine to strengthen (or subvert) one another: A fast, contingent reply would be best, followed by fast scripted replies and slow contingent replies, with slow scripted replies as least desirable. However, the interaction effect of latency by contingency may produce an ironic effect in which the contingency factor moderates the latency effect. An agent may seem to a chat customer equally as genuine if contingent responses appear after a slower latency as after a faster latency. It may seem as if an agent has taken time to examine the question, perhaps look up information that will enrich the answer, and type out original answers that respond to the specifics of the customer’s prior statements. Such a finding would be consistent with some of the literature on chronemics and CMC, in which a slow latency signals greater affection than a short latency when it occurs within sociable interactions (Walther & Tidwell, 1995). In such a case, a slow latency is forgiven since it can be attributed to genuine concern for and responsiveness to the other person, rather than attributed to inattention or incompetence. If this is correct, the combination of a slow latency with a contingent response may be quite desirable, i.e., improve evaluations of the agent’s attractiveness, increase satisfaction, and by extension, improve the assessment of a company’s relationship with customers. A slow latency combined with a less contingent answer, however, should garner the poorest evaluations of an agent and the agent’s organization, in this analysis. A fast latency with a contingent response may be as effective, but may or may not connote as much effort on the part of the agent to address the customer’ specific concerns; it should garner nearly as positive a set of evaluations as the slow latency/contingent reply. A fast latency with a scripted reply should not engender positive evaluations, but should be less disappointing than a slow latency/less contingent response. H7: Response latency and contingency interact such that (a) an agent’s attractiveness, (b) satisfaction, and (c) the organization’s relationship with customers are most positive when contingency is great (regardless of whether responses are fast or slow); but a less contingent response that is fast is poorer; and a less contingent, slow response is poorest. Finally, additional hypotheses were tendered to address Pang et al.’s (2018) proposed extension of SIP theory to organizational relationships with customers. Pang et al. argued that the relational imputations that arise from interpersonal micro-interactions between stakeholders and organizational agents online are likely to generalize, perceptually, to reflect other aspects of an organization and its image (see also Dou & Sundar, 2014). To test this assertion, two mediation hypotheses are as follows: H8: The effects of (a) response latency and (b) contingency on perceptions of an organization’s relationship with customers are mediated by attraction and satisfaction. Method A 2 × 2 between subjects factorial experiment tested the effects of response latency (fast vs. slow) and message contingency (more contingent vs. less contingent) on observers’ perception of an agent’s attractiveness, satisfaction with a chat episode, and perception of the organization’s relationship with customers. Participants (N = 131) comprised volunteers from a pool of subjects managed by Qualtrics, a company that provides services for online questionnaires and data collection. Participants were from the United States, aged 18 or older, who received remuneration in the form of Qualtrics credits for their participation. Participants’ ages ranged from 21 to 79 years old, M = 51.32, SD = 14.3. Most participants were female (59.5%). Eighty-five percent identified as white, 8.4% as black, 3.8% as Asian, and 3.1% as Hispanic/Latino. Stimuli Participants observed one of four simulations of a real-time chat between a customer and a customer support agent. The simulations were created and displayed using hypertext markup language and javascript codes that allowed for precise differences in the verbal content and the timing of the appearances of messages. Each simulation portrayed a text-based online chat between an agent from a fictitious online merchant, shoestore.com, and a potential customer, on the merchant’s website. Participants witnessed the customer start the chat by typing his name (Joe), as prompted, into a space provided in a chat window. The chat system then informed the customer that he is connected to Ro, an agent from shoestore.com. The customer then asked the agent some questions about shoes and whether his size is available in the country where he is currently working. The chat took place from the customer’s perspective: Participants saw the customer type each word letter by letter. Replies by the company’s chat agent appeared as an entire block on the customer’s chat window (i.e., participants could not see the agent typing letter by letter). For added realism, the customer appeared to make spelling errors at two specific points in the chat, which he appeared to correct using the backspace command. The stimuli appear on http://cmcresearch.org/chat/. How the agent responded to the customer’s questions differed in response latency and/or contingency according to the four experimental conditions. In the more contingent version of the chat, the customer service agent’s replies were more responsive, coherent, and directly relevant to the customer’s inquiries and idiosyncratic statements. In the less contingent version, the agent’s responses were more generic: They reflected fewer particulars of the customer’s inquiries, and thereby resembled canned responses that connected to the customer’s queries in a more general, rather than specific, fashion. Two examples illustrate the differences between highly contingent chats and less interactive chats. In the first, the customer, Joe, asks the agent, Ro, if shoestore.com ships to Singapore. In the contingent condition, Ro replies: “we don’t currently offer international shipping to Singapore,” explicitly referring to Joe’s stated location. However, in the less contingent condition, Ro replies: “we do not currently offer international shipping,” without reference to Joe’s location. In a second example, Joe also asks Ro if shoestore.com manufactures sized 10E shoes. In the more contingent condition, Ro replies: “You can contact our Singapore Team at +65 800-8888-6453. They can tell you if there are 10E shoes available,” directly referring to Joe’s question about sized 10E shoes. However, in the less contingent condition, Ro replies: “The Singapore Team can assist you with finding what you need. Try contacting the Singapore Team at +65 800-8888-6453,” with no reference to the specific shoe size Joe mentioned. These examples show that the less contingent chats were not completely unresponsive. Rather, the agent’s responses obliquely addressed the customer’s questions, whereas the more interactive chats reflected responses that were specifically responsive to the questions that prompted them. (The less contingent version was based on an actual online chat with a shoe retailer.) Both versions of the chats had an equal number of words. The other independent variable, chronemic response latency, led to the appearance of the agent’s chat reply either relatively quickly or slowly. In the quick responses condition, an 8 s latency passes between each time the customer finished typing (and appeared to hit the enter key to post his chat message), and the time the agent’s response appeared in the customer’s chat window. In the slow responses condition, the response latency was 40 s instead of 8 s before each time that the agent’s response appeared. Across all conditions, the customer appeared to wait for 4 s after each message from the agent before typing out the first letter of his next message. The specific response latencies (8 s or 40 s) reflected a range of typical response times from informal trials by the researchers with actual online customer support agents. Although responses sometimes exceeded 40 s in these trials, a response latency longer than 40 s seemed unsuitable for the experiment, out of concern that participants may stop paying attention or drop out. After viewing the appropriate chat, participants answered an online questionnaire and were then thanked for their efforts. Procedure This experiment was administered online. Participants who consented to participate were randomly assigned to one of the four experimental conditions to watch one of the experimental chat conversations before completing a questionnaire. A timer in the online research system prevented participants from skipping the entire chat before it ended; it hid the button to proceed to the questionnaire until the chat was complete. Each simulated chat screen displayed about 10 lines of text at any given point, after which the topmost lines scrolled up and out of view; results reported below indicate that this level of message persistence was sufficient for participants to experience the interrelatedness of the chat statements or the lack thereof. The subsequent questionnaire measured all dependent variables. Participants took more time to complete the study when the agent responded slowly (M = 15.46 min, SD = 6.65) than when the agent responded quickly (M = 11.65 min, SD = 6.69). Measures Participants completed a number of established scales after viewing their respective stimuli. The scales measured participants’ attraction toward the customer service agent (Ro), their satisfaction with the chat, and their perception of the organization’s investment in relationships with customers. All items were measured on seven-interval scales anchored by strongly disagree (1) to strongly agree (7). Interpersonal attraction McCroskey and McCain (1974) identified three dimensions of interpersonal attraction, including task attraction and social attraction (as well as physical attraction, which is not relevant to the present study). Participants completed measures of task and social attraction toward the customer service agent. Task attraction measures the desire to work on a task together with another individual. Three out of the original six Likert scale provided optimal reliability, including “You could count on her getting the job done,” “I have confidence in her ability to get the job done,” and “If I wanted to get things done, I could probably depend on her”; Cronbach’s α = .96. Social attraction is the desire for friendship with another individual. The three items producing sufficient reliability for this measure include “I think she could be a friend of mine,” “I would like to have a friendly chat with her,” and “She would be pleasant to be with”; α = .88.2 Satisfaction Flavián, Guinalíu, and Gurrea’s (2006) three-item scale assesses satisfaction with a website. Items were modified to refer to the specific interaction participants had observed, e.g., “The experience that the customer (Joe) had with this web chat has been satisfactory.” Cronbach’s α = .98. Organizational relationship One purpose of this study was to assess whether relational communication by an organization’s agent toward a customer affects the relationship of the customer with the organization as a whole. The study therefore adapted De Wulf, Odekerken-Schröder, and Iacobucci's (2001) three-item scale measuring perceptions of an organization’s relationship investment with its customers.3 Items included “Shoestore.com makes various efforts to improve its tie with regular customers,” and “Shoestore.com really cares about keeping regular customers”; α = .98. The measure does not refer to the specific interaction or individual organization member, but rather, perceptions of an organization’s relational orientation toward customers generally. For this reason the measure is consistent with the relational focus of SIP, yet extends the scope of SIP to encompass secondary relational orientation toward an organization as suggested by Pang et al. (2018). Results Hypotheses for response latency and contingency were analyzed simultaneously using factorial analysis of variance (ANOVA), and the results are summarized in Table 1. The first hypothesis predicted that fast latencies produced more attraction than slow latencies, but H1 was not supported, with no main effect of latency on task attraction, F (1, 127) = .001, p = .97, or social attraction, F (1, 127) = 1.18, p = .281. No interaction effects were found on either task or social attraction. Hypothesis 2 predicted that the provision of a fast latency leads to greater satisfaction than a slow latency. There was a significant interaction effect of latency by contingency, F (1, 127) = 5.76, p = .018, d = .43, that rendered any potential main effect of latency uninterpretable (see Figure 1). Interaction effects are discussed below accompanying analysis of H7, the interaction hypothesis. The main effect test of latency on satisfaction yielded F (1, 127) = .40, p = .53. Table 1 F-scores of All ANOVA Results . Task Attraction . Social Attraction . Satisfaction . Organizational Relationship . Hypotheses H1, H4 H1, H4 H2, H5 H3, H6 Latency 0.00 1.18 0.40 0.49 Contingency 7.29** 13.98** 11.77** 8.57** Latency X Contingency 2.77 3.43 5.76* 5.72* . Task Attraction . Social Attraction . Satisfaction . Organizational Relationship . Hypotheses H1, H4 H1, H4 H2, H5 H3, H6 Latency 0.00 1.18 0.40 0.49 Contingency 7.29** 13.98** 11.77** 8.57** Latency X Contingency 2.77 3.43 5.76* 5.72* Note. *p < .05, **p < .01. Open in new tab Table 1 F-scores of All ANOVA Results . Task Attraction . Social Attraction . Satisfaction . Organizational Relationship . Hypotheses H1, H4 H1, H4 H2, H5 H3, H6 Latency 0.00 1.18 0.40 0.49 Contingency 7.29** 13.98** 11.77** 8.57** Latency X Contingency 2.77 3.43 5.76* 5.72* . Task Attraction . Social Attraction . Satisfaction . Organizational Relationship . Hypotheses H1, H4 H1, H4 H2, H5 H3, H6 Latency 0.00 1.18 0.40 0.49 Contingency 7.29** 13.98** 11.77** 8.57** Latency X Contingency 2.77 3.43 5.76* 5.72* Note. *p < .05, **p < .01. Open in new tab Figure 1 Open in new tabDownload slide Pattern of interaction between contingency (contingent vs. non-contingent) and response latency (fast vs. slow) on chat satisfaction. Figure 1 Open in new tabDownload slide Pattern of interaction between contingency (contingent vs. non-contingent) and response latency (fast vs. slow) on chat satisfaction. Hypothesis 3 predicted that an organization whose agent provides fast response latencies generates a more positive organizational relationship with customers than an organization whose agent responds with slow latencies. Although the main effect of latency was not significant, F (1, 127) = .49, p = .48, results again indicated a significant interaction of latency by contingency on organizational relationship, F (1, 127) = 5.72, p = .018, d = .42, the pattern of which obviated any main effect of latency (see Figure 2 and Table 1). Figure 2 Open in new tabDownload slide Pattern of interaction between contingency (contingent vs. non-contingent) and response latency (fast vs. slow) on organizational relationship with customers. Figure 2 Open in new tabDownload slide Pattern of interaction between contingency (contingent vs. non-contingent) and response latency (fast vs. slow) on organizational relationship with customers. Hypotheses 4–6 concerned the effects of response contingency on the agent’s attractiveness, observers’ satisfaction with the chat, and the organization’s relationship with customers. The interaction effects noted above were ordinal with respect to the contingency factor, and therefore did not compromise interpretation of contingency main effects. Hypothesis 4 was supported. More contingent responses led to greater task attraction (M = 4.78, SD = 2.56) than did less contingent responses (M = 3.62, SD = 2.29), F (1, 127) = 7.29, p = .008, d = .48. More contingency also resulted in greater social attraction (M = 4.67, SD = 2.24) than less contingency (M = 3.30, SD = 1.81), F (1, 127) = 13.98, p < .001, d = .66. Hypothesis 5 was also supported: Greater response contingency resulted in greater satisfaction (M = 4.80, SD = 2.86) than less contingency (M = 3.22, SD = 2.51), F (1, 127) = 11.77, p = .001, d = .61. Additionally, there was an interaction effect between contingency and latency on satisfaction (see Figure 1 and Table 1), which is described following H7, an interaction hypothesis. Hypothesis 6 focused on how a chat agent’s responses affected participants’ impression of the entire organization’s relationship with customers, and it, too, was supported: More contingent responses stimulated a more positive organizational relationship assessment (M = 4.80, SD = 2.88) than did less contingent responses (M = 3.47, SD = 2.45), F (1, 127) = 8.57, p = .004, d = .50. Once again, there was an interaction effect between contingency and latency on organizational relationship (see Figure 2 and Table 1), discussed below with H7, the interaction hypothesis. Because H7 predicted specific difference patterns for the interaction effect of latency by contingency, its analysis employed contrast analysis, which offers a more precise test of directional predictions than does an omnibus ANOVA (Rosenthal & Rosnow, 1985). Contingent responses—with either a fast or slow response latency—were expected to draw the most favorable evaluations, so each of these two conditions was given a contrast weight of +1.5. The fast latency/less contingent response condition was predicted to be poorer, and was assigned a contrast weight of -1. Slow latency/less contingent responses were expected to be the poorest, and received a contrast weight of -2. The contrast analyses were significant for each of the dependent variables: task attraction, t(127) = 2.48, p = .015; social attraction, t(127) = 3.30, p = .001; chat satisfaction, t(127) = 3.23, p = .002; and organizational relationship, t(127) = 2.76, p = .007 (see Table 2 for descriptive statistics). On the surface, these findings suggest that contingent responses from an agent generally had beneficial effects on participants’ perceptions and evaluations, regardless of whether the more contingent responses came either quickly or slowly, compared to less contingent responses that came after a short delay, and that non-contingent responses following a long delay would be the most deleterious. Table 2 Contrasts, Means (and SD) for Effects of Conversational Contingency and Response Latency on Chat Satisfaction, Task and Social Attraction, and Organizational Relationship Response Latency . Contingent . Non-contingent . Fast . Slow . Fast . Slow . n 28 41 31 31 H7 contrast weights 1.5 1.5 −1 −2 Chat satisfaction 5.62a 4.24a,b 2.78b 3.66b (2.44) (3.02) (2.21) (2.75) Task attraction 5.18a 4.50a,b 3.25b 4.00a,b (2.30) (2.72) (2.09) (2.46) Social attraction 4.82a 4.57a 2.76b 3.84a,b (1.98) (2.42) (1.36) (2.06) Organizational 5.51a 4.23a,b 3.05b 3.88a,b relationship (2.12) (3.10) (2.32) (2.55) Response Latency . Contingent . Non-contingent . Fast . Slow . Fast . Slow . n 28 41 31 31 H7 contrast weights 1.5 1.5 −1 −2 Chat satisfaction 5.62a 4.24a,b 2.78b 3.66b (2.44) (3.02) (2.21) (2.75) Task attraction 5.18a 4.50a,b 3.25b 4.00a,b (2.30) (2.72) (2.09) (2.46) Social attraction 4.82a 4.57a 2.76b 3.84a,b (1.98) (2.42) (1.36) (2.06) Organizational 5.51a 4.23a,b 3.05b 3.88a,b relationship (2.12) (3.10) (2.32) (2.55) Note. Different superscripts indicate significant differences using the Scheffe test, p < .05. Open in new tab Table 2 Contrasts, Means (and SD) for Effects of Conversational Contingency and Response Latency on Chat Satisfaction, Task and Social Attraction, and Organizational Relationship Response Latency . Contingent . Non-contingent . Fast . Slow . Fast . Slow . n 28 41 31 31 H7 contrast weights 1.5 1.5 −1 −2 Chat satisfaction 5.62a 4.24a,b 2.78b 3.66b (2.44) (3.02) (2.21) (2.75) Task attraction 5.18a 4.50a,b 3.25b 4.00a,b (2.30) (2.72) (2.09) (2.46) Social attraction 4.82a 4.57a 2.76b 3.84a,b (1.98) (2.42) (1.36) (2.06) Organizational 5.51a 4.23a,b 3.05b 3.88a,b relationship (2.12) (3.10) (2.32) (2.55) Response Latency . Contingent . Non-contingent . Fast . Slow . Fast . Slow . n 28 41 31 31 H7 contrast weights 1.5 1.5 −1 −2 Chat satisfaction 5.62a 4.24a,b 2.78b 3.66b (2.44) (3.02) (2.21) (2.75) Task attraction 5.18a 4.50a,b 3.25b 4.00a,b (2.30) (2.72) (2.09) (2.46) Social attraction 4.82a 4.57a 2.76b 3.84a,b (1.98) (2.42) (1.36) (2.06) Organizational 5.51a 4.23a,b 3.05b 3.88a,b relationship (2.12) (3.10) (2.32) (2.55) Note. Different superscripts indicate significant differences using the Scheffe test, p < .05. Open in new tab Despite the significant contrast tests, the pattern of the means suggest that a different interpretation is possible. Although contingent responses were superior to non-contingent responses on each dependent variable, the pattern of means among non-contingent responses was not consistent with the prediction that slow non-contingent responses are poorer than fast non-contingent responses; the pattern among these cells may actually be the reverse. To examine these anomalous findings, a Scheffe post hoc test examined the pairwise differences among the four means for each dependent variable. The Scheffe results provide a robust accounting for the interaction effects noted in the tests of H1–H3, as well. It appears that the effects of response latency consistently depend on whether replies are contingent or not. Among all four conditions, fast, contingent replies scored most favorably on task attraction, social attraction, satisfaction, and organizational relationship, whereas fast, non-contingent replies (rather than slow, non-contingent replies) generated the lowest scores for each outcome. Consistent with H7, fast/contingent replies led to significantly greater task attraction, social attraction, chat satisfaction, and a more favorable organizational relationship than fast/non-contingent replies. However, contrary to H7, slow/contingent replies were not evaluated more favorably than slow/non-contingent replies for all four dependent variables. In addition, fast/non-contingent replies were not evaluated more favorably than slow/non-contingent replies on all dependent variables. This pattern disconfirms the directional hypothesis described in H7. Finally, two mediation tests assessed H8 to see whether the effects of: (a) response latency, and (b) contingency on organizational relationship were mediated by social attraction, task attraction, and satisfaction.4 Direct and indirect effects were estimated using Model 6 (serial mediation) of the PROCESS macro (Hayes, 2013) for SPSS. Each mediation test was based on 10,000 bootstrap samples and 95% bias-corrected confidence intervals. The use of serial mediation integrates a test of interpersonal (i.e., social and task) attraction as a mechanism in Rafaeli and Sudweeks’ (1998) claim that greater contingency leads to increased satisfaction, as well as a test of interpersonal attraction and satisfaction as mechanisms in Pang et al.’s (2018) claims that relational attributions from interpersonal interactions can extend to perceptions about an organization. For H8a, our analysis did not find a significant indirect effect of response latency on organizational relationship with social attraction, task attraction, and satisfaction as serial mediators, 95% CI = [-.461, .052]. H8a was not supported. The analysis for H8b found a significant indirect effect of contingency on organizational relationship, with social attraction, task attraction, and satisfaction as serial mediators. Compared to individuals who received less contingent responses from an agent, individuals who received more contingent responses experienced more social attraction toward the agent (b = 1.37, SE = .36, t(129) = 3.82, p < .001, d = .67), which led to greater task attraction toward the agent (b = .93, SE = .07, t(128) = 14.34, p < .001, d = 2.53), which in turn led to greater satisfaction with the chat experience (b = .45, SE = .09, t(127) = 4.79, p < .001, d = .85); greater satisfaction with the chat experience led to more positive perceptions of the organization’s relationship with customers (b = .81, SE = .05, t(126) = 16.10, p < .001, d = 2.87). For the indirect effect path described, point estimate = .46, SE = .16, 95% CI = [.21, .87], partially standardized effect size = 1.68. With inclusion of the mediators, the direct effect of contingency on organizational relationship was not significant (b = −.16, SE = .17, t(126) = −.95, p = .34), suggesting a full mediation model. Hypothesis 8b was supported, showing that social attraction, task attraction, and satisfaction mediate the relationship between contingency and an organization’s relationship with customers. Figure 3 shows the unstandardized regression coefficients between various components of the model tested in H8b. Figure 3 Open in new tabDownload slide Path coefficients of mediators between contingency and organizational relationship with customers. Note: p < .01 for all solid lines, ns for all dashed lines. Figure 3 Open in new tabDownload slide Path coefficients of mediators between contingency and organizational relationship with customers. Note: p < .01 for all solid lines, ns for all dashed lines. Discussion This study advances our understanding of CMC’s relational dynamics by extending SIP theory in several important ways. First, it incorporates two forms of interactivity—response latency and conversational contingency—in patterns that extend SIP’s general approach to messages and chronemics that affect impressions and relational evaluations online, in the absence of other nonverbal cues. The current results add to the corpus of SIP research associating empirically observable features of online message behavior with relational judgments, in the absence of physical, face-to-face nonverbal cues.5 Whereas attraction and satisfaction are often based on non-conversational qualities of communication participants in face-to-face interaction, in a task-oriented online exchange, social information is conveyed by the conversational contributions of the participants, from which interpersonal evaluations are derived (e.g., Weisband & Atwater, 1999). The expansion of SIP to include and explain some dimensions of interactivity in CMC is at once a natural progression and at the same time an important extension. Most SIP research has focused on the semantic or functional qualities of verbal strategies through which CMC users develop impressions and manage relational inferences online. The tradition includes verbal content expressing affinity in CMC (e.g., Walther, Loh, & Granka, 2005), language immediacy (O’Sullivan, Hunt, & Lippert, 2004), functional exchanges such as self-disclosures (see Ruppel et al., 2016, for a review), as well as some chronemics cues. A recent study suggested that SIP “mainly focuses on text-based CMC and thus identifies different strategies (…) Such as the use of emoticons and chronemics” (Croes, Antheunis, Schouten, & Krahmer, 2018). Against such a background, the inclusion of subtextual interaction characteristics such as contingency and real-time response latency are important new milestones. They are not content-specific, as previous variables have often been, but rather, they generate meaning through interrelationships of message exchange. They are conversational cues, and therefore within the broad domain of SIP, but a different class of cues than those upon which SIP has generally focused. Although some research on chronemic latencies in CMC found that response delays negatively affect relational judgments (e.g., Kalman et al., 2013; Park & Sundar, 2015), the lack of interpretable main effects for response latency, and the moderation of latency by contingency in the present study, is consistent with a number of studies on latency in the past. However, the interaction effect of message contingency by latency in the present study shows that the effect of latencies differs as a concurrent result of other simultaneously emerging qualities of the conversation rather than simply the a priori qualities of the person (Kalman & Rafaeli, 2011; Sheldon et al., 2006), or the task vs. social context (Walther & Tidwell, 1995), which is a novel finding. It appears that latency in CMC may operate differently than it does in face-to-face communication, or that face-to-face communication research did not sufficiently consider moderators of latency effects in earlier studies. The original interaction hypothesis (H7) predicted that contingency is more important than latency, so that contingent replies would be beneficial whether they were fast or slow. Although contingent replies were found to be more positive than non-contingent replies in all cases, fast latency had a positive effect if replies were contingent, but a negative effect if replies were non-contingent. It appears that contingency is a necessary factor, while a short latency is a desirable factor if and only if replies are contingent. The H7 prediction underestimated the negative effect of a fast, non-contingent reply. Future research may explore various explanations for this effect. One may surmise that a fast, non-contingent reply seems robotic, or that, without a delay signaling that an agent is busy, it is at best inattentive. Whereas previous studies on latencies in CMC generally involved asynchronous interaction, where some delay is expected, this study extends knowledge about latencies to the realm of real-time chat. Whether or not reactions to latencies in real-time chat are more sensitive than asynchronous CMC is with respect to other message features, such as contingency, remains to be seen. Future research may also explore whether there is a threshold response latency that is intolerable, no matter what the degree of contingency is. Message contingency findings uniformly supported hypotheses. Although contingency is often studied as a structural or interface characteristic of CMC (but see Bellur & Sundar, 2017), SIP concerns contingency as a property of conversations and the coherence among messages in sequences. This approach is similar to the foundational definition Rafaeli (1988) offered for the general notion of interactivity. Although Rafaeli originally conjectured that interactivity should produce involvement and satisfaction, neither that essay, nor a demonstration of interactivity in online discussions (Rafaeli & Sudweeks, 1998) empirically linked interactivity with interpersonal outcomes. The present results add credence to the long-held speculation that greater conversational interactivity improves interpersonal assessments, and it embeds the finding within a comprehensive theoretical framework focusing on relational communication. Another extension to SIP is in demonstrating Pang et al.’s (2018) assertion that observers’ extend their relational judgments about an interaction and an organizational agent’s communication-based attractiveness to broader attributions about the organization that the agent represents. Few prior studies on SIP have purposefully extended its findings to particular contexts. Although, as a general theory of online communication, we expect generalizability to organizational and other settings (e.g., Walther, 1995), the inclusion of specific measures to assess the theoretically-consistent effects on contextual variables has been unusual. The main and interaction effects, and particularly the mediation analyses, strongly suggest that the dyadic interpersonal qualities of a CMC encounter affect perceptions of an organization at large. Additional research should explore the durability of these attributions, and how strong a role they play in potential consumers’ search for both value and dependability. A third, and perhaps more important, extension to SIP is in the implication of the interaction of latency by contingency, and the finding that a disruption in one factor (contingency) has more potency and changes the effect of the other factor (latency), in terms of their consequences for relational judgments. Although a basic premise of SIP theory is that individuals use whatever cue systems they have at their disposal in order to manage impressions and relationships, SIP also describes CMC as operating with a minimal set of communicative cues, compared to multimodal communication such as voice, video, and face-to-face interaction. The original articulation of SIP recognized that the variety of nonverbal cue systems in face-to-face interaction do not all just reinforce a single message evinced by one sub-set of cues: “While nonverbal messages may emphasize or be redundant to (other) messages, the relationship of these cue systems includes substitution and contradiction functions” (Walther, 1992, p. 63). Nevertheless, the original SIP framework theorized communication codes in CMC, such as linguistic content or style characteristics and chronemic elements, as mediators of the interaction effects of time and communication channels. Less consideration has been paid to the interaction effects among the qualities of the code systems as well. Our results argue for expanded attention to the relationships among the code systems within language, chronemics, and interactional structures that SIP subsumes. Research in SIP has studied message forms such as self-disclosure and question-asking (Tidwell & Walther, 2002), as well as semantic cues to affinity (Walther et al., 2005). But CMC also supports codes related to the structure of interactions, with specific sub-codes such as conversational contingency, response latency, adjacency pairs, accommodation, style matching, and so on. From the perspective that even plain-text CMC can convey to participants and observers numerous dimensions within semantic, structural, chronemic, and interactional code systems, an expanded view of SIP must entertain the interrelationships between qualities of the multiple codes CMC conveys. How contingency and latency interact, in the present study, provides a paradigmatic lesson for future SIP research to explore other interaction effects and how they guide relational inferences at the local and larger contextual level. One limitation of this research is the use of observers to evaluate the conversations, rather than assessing reactions from actual interactants. This sacrifice was made for the sake of experimental control. The range and variety of questions that actual customers, or even play-acting research participants might tender, and the lack of an authentic database of answers, precluded such an approach in this initial investigation. Researchers investigating other forms of interactivity (Sundar et al., 2016) have devised databases and scripts for limited interaction on recommendations systems, a technique which future research like this might emulate. That said, previous research testing for differences between participants’ and observers’ ratings of relational communication in CMC (Ramirez, Zhang, McGrew, & Lin, 2007) found no differences due to the actor/observer role for short-term CMC episodes. Another limitation is that this experiment examined only one communication context: shoe shopping. Online chat support takes place in a variety of contexts and industries. Future research might inquire whether contingent and fast replies are more important whether one or another of these two factors has greater impact in other contexts. For product categories that involve complicated instructions and guidance (e.g., how to fix problematic software installations), message contingency may be much more important than response latency. For medical emergencies or unlocking a mobile phone, faster may be better. What types and dimensions of tasks call for greater contingency vs. shorter latency remains to be seen. Moreover, if a user knowingly interacts with a conversational robot, will a faster response be more expected, and a slow one less forgivable, or will a response delay be attributed to computer lag? The question of computers’ likeability (see Reeves & Nass, 1996) when they violate rather than emulate human conversational conventions may deserve re-visitation. There are clear practical implications of this theoretically-based study. In traditional settings, individuals’ impressions of an organization are affected by many factors such as reputation, history, advertising, etc. It is a contemporary reality, however, that our first (and potentially last) encounter with an organization may be with its website to which we arrived via a search engine. In that case, information and interaction via the web comprise the extent of one’s impression. Perhaps because of this, according to one purveyor of best practices for online customer service, “3 out of 5 customers take their business to the competitors after poor customer service experience” (Klimczak, n.d., n.p.). Efforts to improve the efficiency of CMC chat seem to have led to dysfunctional solutions from a dialogic communication point of view. When agents use scripted responses to overcome the potential response delays to which multicommunicating make them prone, their efforts may backfire. It may not be the case, as Todman, Rankin, and File (1999, p. 289) suggested regarding the use of stored text in CMC, that, “given the disruptiveness effect of long pauses between turns,” it is “preferable to respond with something off-the-shelf than to delay responding while a novel response is generated.” On the contrary: A slower response that is more contingent on a customer’s input is significantly better than a fast but potentially disingenuous reply. Innovations designed to alleviate the need for personal attention, in the form of scripts or shortcuts, might or might not be useful for the most frequently asked questions, but as far as satisfying individuals, they appear to leave chat users negatively impressed. Acknowledgment The authors extend their gratitude to S. Shyam Sundar, and the editors and reviewers at JCMC for helpful suggestions during the development of this article. Footnotes 1 Although the notion of contingency bears some similarity to message tailoring and the practice of tailoring online content to individual users, these two terms differ in important respects. Tailoring refers broadly to adapting a single message (by a computational system or by a human) on the basis of knowledge that has been obtained about an individual a priori (see Cappella, 2017). It need not occur in situ and it focuses on adaptation for specific persons (or classes of persons). Contingency, on the other hand, requires multiple messages. It occurs ad hoc and situationally, and reflects adaptation to emerging conversational content. It may or may not focus on a person but references the topic and what is said about the topic. An example of tailoring would be a retailer’s message that incorporates data about an individual’s characteristics or past behavior that may have occurred at any time in history, such as, “Based on the other products you viewed, you might be interested in this lamp.” In contrast, contingency adapts to the contextual content derived from utterances, rather than the person, and does not exploit data outside an immediate interaction sequence, e.g.,    A1: “Here is a lamp you may like.”    B: “It’s too big for my living room.”    A2: “Do you want to see a smaller version?” 2 All discarded items showed strong alpha-if-item-deleted scores, and were negatively-, rather than positively-worded, which appears to have introduced measurement artifact (see Tomás, Oliver, Galiana, Sancho, & Lila, 2013). 3 Perceptions of the organization were also measured with scales assessing top management’s ability (Mayer & Davis, 1999), and willingness to recommend the organization to others (Zeithaml, Berry, & Parasuraman, 1996), with no differences in significance patterns using these alternative measures. 4 There was no theoretical reason for the specification of task before social attraction in this test, and the reverse order provided identical patterns of significance. 5 Other approaches to CMC focus on perceptions of interactivity as the proximal cause of evaluative reactions to differences in interactivity, either exclusively or in conjunction with bona fide interface differences (e.g., Sundar et al., 2016). The present approach, in contrast, focuses on linguistic and temporal causal factors affecting evaluative outcomes, rather than perceptual ones. Nevertheless, for the sake of comparability across studies, replication of the perceptual approach was also undertaken. Participants rated the agent’s synchronicity using scales adapted from Liu’s (2003) perceived synchronicity scale. There was a main effect of the response latency factor on perceived synchronicity, t(129) = 2.45, p = .008, 1-tailed. Mediation analysis found a significant indirect effect of actual response latency on organizational relationship with perceived synchronicity, social attraction, task attraction, and satisfaction as serial mediators, point estimate = .069, SE = .041, 95% CI = [.016, .189]. Participants also completed seven items from Sundar et al.’s (2016) perceived contingency measure regarding the agent’s statements. Participants perceived the stimuli that were designed to reflect more contingency as possessing more contingency, t(129) = 3.86, p < .001. Mediation tests found a significant indirect effect of actual message contingency on organizational relationship with perceived contingency, social attraction, task attraction, and satisfaction as serial mediators, point estimate = .123, SE = .074, 95% CI = [.025, .332]. References Anderson , J. F. , Beard , F. K., & Walther , J. B. ( 2011 ). Turn-taking and the local management of conversation in a two-way computer-mediated communication system. Language@Internet. Retrieved from http://www.languageatinternet.org/articles/2010/2804 Basu , A. ( 2014 , March 13). 5 Zopim shortcuts that will save you time. Retrieved from http://blog.zopim.com/2014/03/13/7-zopim-shortcuts-that-will-save-time/ Bellur , S. , & Sundar , S. S. ( 2017 ). Talking health with a machine: How does message interactivity affect attitudes and cognitions? Human Communication Research, 43, 25 – 53 . doi:10.1111/hcre.12094 Google Scholar Crossref Search ADS WorldCat Burgoon , J. K. , Bonito , J. A., Bengtsson , B., Ramirez , A. , Jr., Dunbar , N., & Miczo , N. ( 2000 ). Testing the interactivity model: Communication processes, partner assessments, and the quality of collaborative work . Journal of Management Information Systems, 16, 35 – 38 . doi:10.1080/07421222.1999.11518255 Google Scholar OpenURL Placeholder Text WorldCat Burgoon , J. K. , Bonito , J. A., Ramirez , A., Dunbar , N. E., Kam , K., & Fischer , J. ( 2002 ). Testing the interactivity principle: Effects of mediation, propinquity, and verbal and nonverbal modalities in interpersonal interaction . Journal of Communication, 52, 657 – 677 . doi:10.1093/joc/52.3.657 Google Scholar Crossref Search ADS WorldCat Cappella , J. N. ( 2017 ). Vectors into the future of mass and interpersonal communication research: Big data, social media, and computational social science . Human Communication Research, 43, 545 – 558 . doi:10.1111/hcre.12114 Google Scholar Crossref Search ADS PubMed WorldCat Cherny , L. ( 1999 ). Conversation and community: Chat in a virtual world . Stanford, CA : CSLI Publications . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Chow , L. Y. ( 2016 , January 26). Chat routing: How your agents manage high chat volume and increase satisfaction. Zopim Blog. Retrieved from http://blog.zopim.com/2016/01/26/chat-routing/ Cole , N. ( 2017 , Apr 25). The power of live chat. Inc. Retrieved from https://www.inc.com/nicolas-cole/the-power-of-live-chat-5-surprising-statistics-that-show-how-consumers-want-thei.html Condon , S. L. , & Cech , C. G. ( 2001 ). Profiling turns in interaction: Discourse structure and function. Proceedings of the 35th Hawaii International Conference on System Sciences (pp. 1–10). Los Alamitos, CA: IEEE Press. Croes , E. A. J. , Antheunis , M. L., Schouten , A. P., & Krahmer , E. J. (2018). Social attraction in video-mediated communication: The role of nonverbal affiliative behavior . Journal of Social and Personal Relationships . Advance online publication. doi:10.1177/0265407518757382 OpenURL Placeholder Text WorldCat De Wulf , K. , Odekerken-Schröder , G., & Iacobucci , D. ( 2001 ). Investments in consumer relationships: A cross-country and cross-industry exploration . Journal of Marketing, 65, 33 – 50 . doi:10.1509/jmkg.65.4.33.18386 Google Scholar Crossref Search ADS WorldCat Dou , X. , & Sundar , S. S. ( 2014 , March). Cultivating relationships through a mobile website: The importance of modality interactivity and message interactivity. Paper presented at the annual meeting of the International Public Relations Research Conference, Miami, FL. Flavián , C. , Guinalíu , M., & Gurrea , R. ( 2006 ). The role played by perceived usability, satisfaction, and consumer trust on website loyalty . Information & Management, 43, 1 – 14 . doi:10.1016/j.im.2005.01.002 Google Scholar Crossref Search ADS WorldCat Hayes , A. F. ( 2013 ). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach . New York , NY: Guilford Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Herring , S. C. ( 1999 ). Interactional coherence in CMC . Journal of Computer-Mediated Communication, 4, 1 – 13 . doi:10.1111/j.1083-6101.1999.tb00106.x Google Scholar OpenURL Placeholder Text WorldCat Kalman , Y. M. , & Rafaeli , S. ( 2011 ). Online pauses and silence: Chronemic expectancy violations in written computer-mediated communication . Communication Research, 38, 54 – 69 . doi:10.1177/0093650210378229 Google Scholar Crossref Search ADS WorldCat Kalman , Y. M. , Scissors , L. E., & Gergle , D. ( 2010 ). Chronemic aspects of chat, and their relationship to trust in a virtual team. Proceedings of the Mediterranean Conference on Information Systems, paper 46. Retrieved from http://aisel.aisnet.org/mcis2010/46/ Kalman , Y. M. , Scissors , L. E., Gill , A. J., & Gergle , D. ( 2013 ). Online chronemics convey social information . Computers in Human Behavior, 29, 1260 – 1269 . doi:10.1016/j.chb.2012.12.036 Google Scholar Crossref Search ADS WorldCat Kellermann , K. , & Sleight , C. ( 1989 ). Coherence: A meaningful adhesive for discourse. In J. A. Anderson (Ed.), Communication yearbook 12 (pp. 95 – 129 ). Newbury Park, CA : Sage . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Klimczak , S. (n.d.). Online customer service best practices. Retrieved from https://www.livechatinc.com/livechat-resources/online-customer-service-best-practices/ Levin , L. ( 2017 ). 5 reasons why live chat is the untapped potential for your business. Kissmetrics Blog. Retrieved from https://blog.kissmetrics.com/live-chat/ LiveChat (n.d.). Chat tools. Retrieved from https://www.livechatinc.com/features/chat-tools/ Liu , Y. ( 2003 ). Developing a scale to measure the interactivity of websites . Journal of Advertising Research, 43, 207 – 216 . doi:10.2501/JAR-43-2-207-216 Google Scholar Crossref Search ADS WorldCat Liu , Y. , & Shrum , L. J. ( 2002 ). What is interactivity and is it always such a good thing? Implications of definition, person, and situation for the influence of interactivity on advertising effectiveness . Journal of Advertising, 31, 53 – 64 . doi:10.1080/00913367.2002.10673685 Google Scholar Crossref Search ADS WorldCat Mayer , R. C. , & Davis , J. H. ( 1999 ). The effect of the performance appraisal system on trust for management: A field quasi-experiment . Journal of Applied Psychology, 84, 123 – 136 . doi:10.1037/0021-9010.84.1.123 Google Scholar Crossref Search ADS WorldCat McCroskey , J. C. , & McCain , T. A. ( 1974 ). The measurement of interpersonal attraction . Speech Monographs, 41, 261 – 266 . doi:10.1080/03637757409375845 Google Scholar Crossref Search ADS WorldCat McGrath , J. E. ( 1990 ). Time matters in groups. In J. Galegher, R. E. Kraut, & C. Egido (Eds.), Intellectual teamwork Social and technical foundations of cooperative work (pp. 23 – 61 ). Hillsdale, NJ : Erlbaum . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC McLaughlin , M. ( 1984 ). Conversation: How talk is organized . Newbury Park, CA : Sage . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC McLaughlin , M. L. , & Cody , M. J. ( 1982 ). Awkward silences: Behavioral antecedents and consequences of the conversational lapse . Human Communication Research, 8, 299 – 316 . doi:10.1111/j.1468-2958.1982.tb00669.x Google Scholar Crossref Search ADS WorldCat McMillan , S. J. , & Hwang , J.-S. ( 2002 ). Measures of perceive interactivity: An exploration of the role of direction of communication, user control, and time in shaping perceptions of interactivity . Journal of Advertising, 31, 29 – 42 . doi:10.1080/00913367.2002.10673674 Google Scholar Crossref Search ADS WorldCat van Noort , G. , Vliegenthart , R., & Kruikemeier , S. ( 2016 ). Return on interactivity? The characteristics and effectiveness of web sites during the 2010 Dutch local elections . Journal of Information Technology & Politics, 13, 352 – 364 . doi:10.1080/19331681.2016.1230921 Google Scholar Crossref Search ADS WorldCat O’Sullivan , P. B. , Hunt , S. K., & Lippert , L. R. ( 2004 ). Mediated immediacy: A language of affiliation in a technological age . Journal of Language and Social Psychology, 23, 464 – 490 . doi:10.1177/0261927x04269588 Google Scholar Crossref Search ADS WorldCat Pang , A. , Shin , W., Lew , Z., & Walther , J. B. ( 2018 ). Building relationships through dialogic communication: organizations, stakeholders, and computer-mediated communication . Journal of Marketing Communications, 24, 68 – 82 . doi:10.1080/13527266.2016.1269019 Google Scholar Crossref Search ADS WorldCat Park , E. K. , & Sundar , S. S. ( 2015 ). Can synchronicity and visual modality enhance social presence in mobile messaging? Computers in Human Behavior, 45, 121 – 128 . doi:10.1016/j.chb.2014.12.001 Google Scholar Crossref Search ADS WorldCat Rafaeli , S. ( 1988 ). Interactivity: From new media to communication. In R. P. Hawkins, J. M. Wiemann, & S. Pingree (Eds.), Advancing communication science: Merging mass and interpersonal processes (pp. 110 – 134 ). Newbury Park, CA : Sage . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Rafaeli , S. , & Sudweeks , F. ( 1998 ). Interactivity on the nets. In F. Sudweeks, M. McLaughlin, & S. Rafaeli (Eds.), Network & netplay: Virtual groups on the Internet (pp. 173 – 189 ). Menlo Park, CA : AAAI Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Ramirez , A. , Jr., Zhang , S., McGrew , C., & Lin , S.-F. ( 2007 ). Relational communication in computer-mediated interaction revisited: A comparison of participant-observer perspectives . Communication Monographs, 74, 492 – 516 . doi:10.1080/03637750701716586 Google Scholar Crossref Search ADS WorldCat Reeves , B. , & Nass , C. I. ( 1996 ). The media equation: How people treat computers, television, and new media like real people and places . New York, NY : Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Reinsch , N. L. , Jr., Turner , J. W., & Tinsley , C. H. ( 2008 ). Multicommunicating: A practice whose time has come? Academy of Management Review, 33, 391 – 403 . doi:10.5465/amr.2008.31193450 Google Scholar Crossref Search ADS WorldCat Ruppel , E. K. , Gross , C., Stoll , A., Peck , B. S., Allen , M., & Kim , S.-Y. ( 2016 ). Reflecting on connecting: Meta-analysis of differences between computer-mediated and face-to-face self-disclosure . Journal of Computer-Mediated Communication, 22, 18 – 34 . doi:10.1111/jcc4.12179 Google Scholar Crossref Search ADS WorldCat Rosenthal , R. , & Rosnow , R. L. ( 1985 ). Contrast analysis: Focused comparisons in the analysis of variance . New York, NY : Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Sheldon , O. J. , Thomas-Hunt , M. C., & Proell , C. A. ( 2006 ). When timeliness matters: The effect of status on reactions to perceived time delay within distributed collaboration . Journal of Applied Psychology, 91, 1385 – 1395 . doi:10.1037/0021-9010.91.6.1385 Google Scholar Crossref Search ADS PubMed WorldCat Sundar , S. S. , Bellur , S., Oh , J., Jia , H., & Kim , H.-S. ( 2016 ). Theoretical importance of contingency in human–computer interaction: Effects of message interactivity on user engagement . Communication Research, 43, 595 – 625 . doi:10.1177/0093650214534962 Google Scholar Crossref Search ADS WorldCat Tidwell , L. C. , & Walther , J. B. ( 2002 ). Computer‐mediated communication effects on disclosure, impressions, and interpersonal evaluations: Getting to know one another a bit at a time . Human Communication Research, 28, 317 – 348 . doi:10.1111/j.1468-2958.2002.tb00811.x Google Scholar Crossref Search ADS WorldCat Todman , J. , Rankin , D., & File , P. ( 1999 ). The use of stored text in computer-aided conversation: A single-case experiment . Journal of Language and Social Psychology, 18, 287 – 309 . doi:10.1177/0261927x99018003004 Google Scholar Crossref Search ADS WorldCat Tomás , J. M. , Oliver , A., Galiana , L., Sancho , P., & Lila , M. ( 2013 ). Explaining method effects associated with negatively worded items in trait and state global and domain-specific self-esteem scales . Structural Equation Modeling: A Multidisciplinary Journal, 20, 299 – 313 . doi:10.1080/10705511.2013.769394 Google Scholar Crossref Search ADS WorldCat Turner , J. W. , & Reinsch , N. L. , Jr. ( 2010 ). Successful and unsuccessful multicommunication episodes: Engaging in dialogue or juggling messages? Information Systems Frontiers, 12, 277 – 285 . doi:10.1007/s10796-009-9175-y Google Scholar Crossref Search ADS WorldCat Walther , J. B. ( 1992 ). Interpersonal effects in computer-mediated interaction: A relational perspective . Communication Research, 19, 52 – 90 . doi:10.1177/009365092019001003 Google Scholar Crossref Search ADS WorldCat Walther , J. B. ( 1995 ). Relational aspects of computer-mediated communication: Experimental observations over time . Organization Science, 6, 186 – 203 . doi:10.1287/orsc.6.2.186 Google Scholar Crossref Search ADS WorldCat Walther , J. B. ( 2017 ). The merger of mass and interpersonal communication via new media: Integrating meta-constructs . Human Communication Research, 43, 559 – 572 . doi:10.1111/hcre.12122 Google Scholar Crossref Search ADS WorldCat Walther , J. B. , Loh , T., & Granka , L. ( 2005 ). Let me count the ways: The interchange of verbal and nonverbal cues in computer-mediated and face-to-face affinity . Journal of Language and Social Psychology, 24, 36 – 65 . doi:10.1177/0261927x04273036 Google Scholar Crossref Search ADS WorldCat Walther , J. B. , & Tidwell , L. C. ( 1995 ). Nonverbal cues in computer-mediated communication, and the effect of chronemics on relational communication . Journal of Organizational Computing, 5, 355 – 378 . doi:10.1080/10919399509540258 Google Scholar Crossref Search ADS WorldCat Walther , J. B. , Van Der Heide , B., Ramirez , A. Jr., Burgoon , J. K., & Peña , J. ( 2015 ). Interpersonal and hyperpersonal aspects of computer-mediated communication. In S. S. Sundar (Ed.), The handbook of psychology and communication technology (pp. 3 – 22 ). West Sussex, England : Wiley-Blackwell . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Weisband , S. , & Atwater , L. ( 1999 ). Evaluating self and others in electronic and face-to-face groups . Journal of Applied Psychology, 84, 632 – 639 . doi:10.1037/0021-9010.84.4.632 Google Scholar Crossref Search ADS WorldCat Zhao , X. , Lampe , C., & Ellison , N. B. ( 2016 ). The social media ecology: User perceptions, strategies and challenges. Proceedings of the 2016 CHI conference on human factors in computing systems, pp. 89 – 100 . doi:10.1145/2858036.2858333 Zeithaml , V. A. , Berry , L. L., & Parasuraman , A. ( 1996 ). The behavioral consequences of service quality . The Journal of Marketing , 60 ( 2 ), 31 – 46 . doi:10.2307/1251929 Google Scholar Crossref Search ADS WorldCat Author notes Accepted by Nicole Krämer © The Author(s) 2018. Published by Oxford University Press on behalf of International Communication Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
TI - Interactivity in Online Chat: Conversational Contingency and Response Latency in Computer-mediated Communication
JF - Journal of Computer-Mediated Communication
DO - 10.1093/jcmc/zmy009
DA - 2018-07-01
UR - https://www.deepdyve.com/lp/oxford-university-press/interactivity-in-online-chat-conversational-contingency-and-response-qGLVooQAcR
SP - 201
EP - 221
VL - 23
IS - 4
DP - DeepDyve
ER -