UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Designing social interactions with animated avatars and speech output for Product Recommendation Agents… Qiu, Lingyun 2006

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
831-ubc_2006-199039.pdf [ 14.09MB ]
Metadata
JSON: 831-1.0092949.json
JSON-LD: 831-1.0092949-ld.json
RDF/XML (Pretty): 831-1.0092949-rdf.xml
RDF/JSON: 831-1.0092949-rdf.json
Turtle: 831-1.0092949-turtle.txt
N-Triples: 831-1.0092949-rdf-ntriples.txt
Original Record: 831-1.0092949-source.json
Full Text
831-1.0092949-fulltext.txt
Citation
831-1.0092949.ris

Full Text

DESIGNING SOCIAL INTERACTIONS WITH ANIMATED AVATARS AND SPEECH OUTPUT FOR PRODUCT RECOMMENDATION AGENTS IN ELECTRONIC COMMERCE by Lingyun Qiu Master of Science, University of B r i t i s h Columbia, 2002 Master of Engineering, Shanghai Jiao Tong University, 1998 Bachelor of Engineering, Shanghai Jiao Tong University, 1994 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY i n THE FACULTY OF GRADUATE STUDIES (Business Administration) THE UNIVERSITY OF BRITISH COLUMIBA March 2006 ® Lingyun Qiu, 2 006 A B S T R A C T Product Recomrnendation Agents (PRAs) and other web-based decision aids are deployed extensively by online vendors, to provide virtual advising services to their customers. While the design of PRA functionality has received increasing amount of attention in academic studies, the social aspects of human-PRA interactions are comparatively less studied. This dissertation investigates the potential of enhancing users' social experiences with PRAs by developing and analyzing an anthropomorphic interface, which has humanoid embodiment and voice output. This dissertation first investigates the importance of choosing appropriate demographic embodiments for a humanoid PRA. The two demographic variables that have been assessed are ethnicity and gender. As suggested by similarity-attraction theories and social-identity theories, results of a laboratory experiment have revealed that users apply similar social stereotypes in human-human communications as they apply to evaluate humanoid agents. PRAs that match the ethnicity of users are perceived by the users as more sociable, more competent, and more enjoyable to interact with than PRAs that do not match users' ethnicity; as well, same-gender PRAs are perceived as more competent and more honest than opposite-gender agents. In addition, the "match-up" effects of ethnicity appear to be more significant among female users than among males. Two interface components are also empirically investigated in this dissertation: 1) presence of a humanoid embodiment and 2) output modalities (text, computer-synthesized voice, or human voice). Results from a laboratory experiment demonstrate that humanoid embodiments increase consumers' perception of a PRA's social presence, their beliefs in its competence, and the enjoyment they derive from interaction with the PRA. A human voice also appears to be significantly more effective than on-screen text and computer-synthesized voice in improving the PRA's perceived social presence and enjoyment. Furthermore, the important role of social relationships in influencing user adoption of agents is tested by integrating social presence, trust, and perceived enjoyment with the Technology Acceptance Model (TAM). Social presence appears to be a common ii antecedent of both trust and perceived enjoyment. Trust exerts a direct impact on user intentions to adopt PRAs, as well as an indirect impact via user perceptions of PRAs' usefulness. Perceived enjoyment also influences adoption intentions through perceived usefulness and perceived ease of use. iii T A B L E O F C O N T E N T S ABSTRACT ii TABLE OF CONTENTS iv LIST OF TABLES vii LIST OF FIGURES ix ACKNOWLEDGEMENTS x DEDICATION xi CHAPTER 1: INTRODUCTION 1 1.1 Research Objectives and Questions 3 1.2 Method and Main Findings 6 1.3 Structure of the Dissertation 8 CHAPTER 2: LITERATURE REVIEW 10 2.1 Product Recommendation Agents (PRAs) 10 2.2 Recommendation Agents as Social Actors 15 2.3 Anthropomorphic PRA Interface 18 2.3.1 Humanoid Embodiment (by Animated Human-Like Faces) 25 2.3.2 Voice 27 2.4 Chapter Summary 29 CHAPTER 3: EVALUATING HUMAN-PRA SOCIAL INTERACTIONS: THE DEPENDENT VARIABLES 31 3.1 Social Presence 31 3.2 Trust : 34 3.3 Perceived Enjoyment 40 3.4 Chapter Summary 42 CHAPTER 4: INFLUENCES OF A PRA'S DEMOGRAPHIC EMBODIMENT ON USERS' PERCEPTIONS 44 4.1 Introduction 44 4.2 Hypotheses Development 49 4.3 Research Method 54 4.3.1 Experimental Design 54 4.3.2 The Agent 56 4.3.3 Dependent Variables 61 4.3.4 Control Variables 62 4.3.5 Participants, Incentives, and Experiment Procedure 63 4.4 Data Analysis and Findings 66 4.4.1 Demographic Data 66 iv 4.4.2 Ethnicity 67 4.4.2.1 MANCOVA Results 68 4.4.2.2 ANCOVA Results 68 4.4.3 Gender 73 4.4.3.1 MANCOVA Results 73 4.4.3.2 ANCOVA Results 73 4.4.4 Three-Way Interactions: Ethnicity * Gender 78 4.5 Discussion 82 4.6 Chapter Summary 84 CHAPTER 5: INFLUENCES OF PRA'S HUMANOID EMBODIMENT AND OUTPUT MODALITIES 86 5.1 Introduction 86 5.2 Hypotheses Development 87 5.2.1 Impacts on Social Presence 87 5.2.2 Impacts on Trusting Beliefs 91 5.2.3 Impacts on Perceived Enjoyment 95 5.3 Research Method 99 5.3.1 Experimental Design 99 5.3.2 The Agent 100 5.3.3 Dependent Variables and Control Variables 106 5.3.4 Participants, Incentives, and Experimental Procedure 107 5.4 Data Analysis and Findings 109 5.4.1 Demographic Data 109 5.4.2 MANCOVA Results 110 5.4.3 ANCOVA Results 113 5.4.3.1 Social Presence 113 5.4.3.2 Trust 114 5.4.3.3 Perceived Enjoyment 115 5.5 Discussion 117 5.6 Chapter Summary 119 CHAPTER 6: HUMAN-PRA SOCIAL INTERACTIONS AND USERS' ADOPTION INTENTIONS 120 6.1 Introduction 120 6.2 Hypotheses Development 121 6.2.1 Social Presence and Trust 121 6.2.2 Social Presence and Perceived Enjoyment 123 6.2.3 Trust and TAM 124 6.2.4 Perceived Enjoyment and TAM 125 6.3 Data Analysis and Findings 127 6.3.1 Measurement Model 128 6.3.2 Structural Model 131 6.4 Discussion 133 6.5 Chapter Summary 134 CHAPTER 7: CONCLUSIONS AND FUTURE RESEARCH 135 7.1 Conclusions 135 7.2 Contributions 137 7.3 Limitations 139 7.4 Future Research 141 REFERENCES 144 APPENDICES 161 Appendix A: Measurement Items 161 Appendix B: Interaction Scripts of the Experimental Recommendation Agent 165 L I S T O F T A B L E S Table 2-1: A Summary of Selected Empirical Evaluation Studies on Anthropomorphic Interface Agents 20 Table 3-1: A Summary of Empirical Studies of Trust in Recommendation Agents 38 Table 4-1: Experimental Design of Study One 55 Table 4-2: Experimental Design of Study One (Gender Factor) 55 Table 4-3: Experimental Design of Study One (Ethnicity Factor) 56 Table 4-4: Study One - Demographic Data 66 Table 4-5: Study One - Cronbach's Alpha and Correlations of Constructs 67 Table 4-6: Study One - MANCOVA Results on Ethnicity 68 Table 4-7: Study One - Group Means and Standard Deviations by Ethnicity 69 Table 4-8: Study One - ANCOVA Results by Ethnicity (Social Presence) 69 Table 4-9: Study One - ANCOVA Results by Ethnicity (Competence Beliefs) 69 Table 4-10: Study One - ANCOVA Results by Ethnicity (Benevolence Beliefs) 70 Table 4-11: Study One - ANCOVA Results by Ethnicity (Integrity Beliefs) 70 Table 4-12: Study One - ANCOVA Results by Ethnicity (Perceived Enjoyment) 70 Table 4-13: Study One - MANCOVA Results on Gender 73 Table 4-14: Study One - Group Means and Standard Deviations by Gender 74 Table 4-15: Study One - ANCOVA Results by Gender (Social Presence) 74 Table 4-16: Study One - ANCOVA Results by Gender (Competence Beliefs) 74 Table 4-17: Study One - ANCOVA Results by Gender (Benevolence Beliefs) 75 Table 4-18: Study One - ANCOVA Results by Gender (Integrity Beliefs) 75 Table 4-19: Study One - ANCOVA Results by Gender (Perceived Enjoyment) 76 Table 4-20: Study One - A Summary of Hypothesis Testing Results 78 Table 4-21: Study One - ANCOVA Results by Ethnicity and Gender (Social Presence) 79 Table 4-22: Study One - Ethnicity Interaction by Subject's Gender (Social Presence).. 80 Table 4-23: Study One - ANCOVA Results by Ethnicity and Gender (Perceived Enjoyment) 80 Table 4-24: Study One - Ethnicity Interaction by Subject's Gender (Perceived Enjoyment) 81 Table 4-25: Study One - Ethnicity Interaction by Agent's Gender (Perceived Enjoyment) 81 Table 5-1: Study Two - 3*2 Full Factorial Experiment Design 100 Table 5-2: Study Two - Demographic Data 110 Table 5-3: Study Two - Group Means and Standard Deviations I l l Table 5-4: Study Two - Reliabilities and Bivariate Correlations of Dependent Variables : I l l Table 5-5: Study Two - MANCOVA Results 112 vii Table 5-6: Study Two - ANCOVA Results (Social Presence) 113 Table 5-7: Study Two - Scheffe Test on PRA Modalities (Social Presence) 113 Table 5-8: Study Two - ANCOVA Results (Competence Beliefs) 114 Table 5-9: Study Two - ANCOVA Results (Benevolence Beliefs) 114 Table 5-10: Study Two - ANCOVA Results (Integrity Beliefs) 115 Table 5-11: Study Two - ANCOVA Results (Perceived Enjoyment) 115 Table 5-12: Study Two - Scheffe Test on PRA Modalities (Perceived Enjoyment) 116 Table 5-13: Study Two - A Summary of Hypothesis Testing Results 116 Table 6-1: Construct Attributes 129 Table 6-2: Loading and Cross-Loadings of Measures 130 Table 6-3: PLS Analysis - A Summary of Hypothesis Testing Results 132 L I S T O F F I G U R E S Figure 4-1: Taxonomy of Non-Verbal Cues (De Meuse 1987) 45 Figure 4-2: The Four Animated Characters Used in the Experiment 58 Figure 4-3: Study One - Screenshot of the Agent's Question Page 60 Figure 4-4: Study One - Screenshot of the Agent's Result Page 61 Figure 4-5: Study One - Ethnicity Interactions on Social Presence 71 Figure 4-6: Study One - Ethnicity Interactions on Trust in Competence 71 Figure 4-7: Study One - Ethnicity Interactions on Perceived Enjoyment 72 Figure 4-8: Study One - Gender Interactions on Trust in Competence 76 Figure 4-9: Study One - Gender Interactions on Trust in Integrity 77 Figure 5-1: Study Two - Research Model 87 Figure 5-2: Study Two - Screenshot of the Question Page (Condition 1) 103 Figure 5-3: Study Two - Screenshot of the Result Page (Condition 1) 103 Figure 5-4: Study Two - Screenshot of the Question Page (Conditions 2 and 3) 104 Figure 5-5: Study Two - Screenshot of the Question Page (Condition 4) 105 Figure 5-6: Study Two - Screenshot of the Question Page (Conditions5 and 6) 106 Figure 6-1: The Structural Model 121 Figure 6-2: PLS Structural Model Testing Results 132 ix A C K N O W L E D G E M E N T S I want to thank a lot of people who had offered their invaluable help and support throughout my endeavors to get this dissertation completed. I am deeply grateful to my advisor, Professor Izak Benbasat, for his guidance and continuous support over the years. He is and will continue to be my role model of scholarly thoughtfulness, pedagogical professionalism, and personal kindness. I would also like to thank my committee members, Dr. Jai-Yeol Son and Dr. Darren Dahl, for their great insights and enormous help to my work. I am also grateful to my dissertation examiners - Dr. Carson Woo, Dr. Brian Fisher, and Dr. Hock Chuan Chan, who spent a lot of time reviewing this dissertation and provided constructive comments. I feel lucky to be surrounded by many supportive colleagues and friends - Weiquan Wang, Dongmin Kim, Young-Eun Lee, Zhenhui Jiang, Ron Cenfetelli, Sameh Al-Natour, Palash Bera, Steven Glover, Bo Xiao, and Chee Wee Tan - they are insightful and kind people. I also want to thank the people who worked on the experiments: Victor Wong spent extensive efforts on designing the experimental websites; Natasha Park and Bruce Franklin did a great job in recording the speech manuscripts; Joyce Hou and Nan Wang conducted many of the experiments; lots of other friends (too many to list here) helped me recruit participants. Moreover, many thanks to Steve Doak, who has been particularly helpful in proofreading and editing the manuscript. Most important of all, I am deeply obliged to my parents. Without their persistent love and support, this dissertation could not have been possible. To my parents xi Chapter 1: Introduction C H A P T E R 1 : I N T R O D U C T I O N Electronic commerce allows consumers to search more extensively for their preferred products before purchasing. Nevertheless, as studies in social psychology (e.g., Schwartz et al. 2002) have suggested, consumers may feel dissatisfied as the number of product options available to them increases. To alleviate consumers' cognitive efforts required to gather and to screen vast amounts of product information, online vendors have increasingly adopted various forms of electronic decision aids to help shoppers specify their needs and to identify the products that best fit their requirements (Alba et al. 1997; Maes et al. 1999). These tools have been found helpful in improving both the decision quality and efficiency (Haubl and Trifts 2000). To some extent, the role played by software agents is similar to human salespersons in physical stores who are employed to advise customers what to buy (Komiak et al. 2004/2005). Therefore, software-based product recommendation and decision support services are considered to be critical components of the e-services portfolios that online vendors offer to their consumers (Rust and Kannan 2003), given that they could not provide as interactive and interpersonal product-related pre-purchase services as their brick-and-mortar competitors do. This dissertation focuses on one particular type of online decision aid: content-filtering product recommendation agents (PRAs) (Ansari et al. 2000). More specifically, content-filtering PRAs are conceptualized as a category of online decision support software that "(a) attempts to understand a decision maker's multi-attribute preference with respect to a particular domain or product category based on a learning phase during which the customer reveals her subjective preference information to the agent; and (b) makes recommendations in the form of a sorted list of alternatives to the l Chapter 1: Introduction customer based on its understanding of that individual's preference structure of products 'features and characteristics" (Haubl and Trifts 2000; Haubl and Murray 2003). Currently, this kind of PRA can be found either on dedicated third-party infomediary websites, such as myproductadvisor.com and activedecisions.com, or as a built-in value-added module provided by online retailers like target.com and futureshop.ca. A significant amount of research effort has been dedicated to various facets of PRA design. Most of these works have focused on the design of functional features that can be incorporated into PRA software. For example, some studies have focused on the potential effects of a PRA's underlying algorithm and choice strategies on consumers' decision-making processes (Haubl and Murray 2001; Russo 2002; Ariely et al. 2004; Wang 2005), some have focused on the relationships between users' knowledge about a class of products and the searching strategies employed by PRAs (e.g., Pereira 2000), and others have questioned whether or not the explanatory information provided by a PRA can make an agent more trustworthy and its recommendations more credible (Herlocker et al. 2000; Sproule and Archer 2000; Wang 2005). Diverging from studies that have focused on task-oriented PRA features, this dissertation investigates another important, however under-studied, aspect of PRA designs: the social interactions between a PRA and its users. According to the Merriam-Webster 11th Collegiate Dictionary, the word "social" is defined as: "...2) a: marked by or passed in pleasant companionship with one's friends or associates; b: sociable; c: of, relating to, or designed for sociability; 3) of or relating to human society, the interaction of the individual and the group, or the welfare of human beings as members of society...". All of these definitions imply that social interactions typically take place among human-2 Chapter 1: Introduction beings. Nevertheless, recent studies have proposed that, under certain circumstances, technological artifacts such as software agents are treated by human users as social actors. According to the theory of Computers as Social Actors (CASA), individuals mindlessly attribute human-like characteristics to technology and media representations, and apply social rules and expectations when they interact with technologies (Reeves and Nass 1996; Nass and Moon 2000). Inasmuch as people may behave socially in their interactions with a PRA, it is important for website and agent developers to understand the potential impacts of emerging multimedia technologies that can be used to strengthen or alleviate these responses so that a more satisfying user-PRA interaction experience can be achieved and users' intentions to utilize a PRA can be enhanced. In addition, studies of consumers have established that in brick-and-mortar stores, customers of various demographic backgrounds may prefer to consult with and may exhibit greater trust towards salespersons with similar characteristics, such as age, gender, appearance, etc., suggesting that online consumers' evaluations of humanoid PRA are also likely to be influenced by these individual factors. These are the issues, in a broad sense, that this dissertation addresses. 1.1 Research Objectives and Questions The primary objectives of this research are 1) to study consumers' social and utilitarian evaluations of a multimedia and multimodal PRA in an online shopping environment, 2) to empirically investigate the effectiveness of two categories of web-based multimedia technologies, animated avatars and speech output, on users' acceptance of online recommendation agents, and 3) to examine how users' individual 3 Chapter 1: Introduction characteristics influence their evaluations of an embodied PRA with salient social identity characteristics. In this dissertation, two important agent design issues are examined: 1) whether or not a PRA be designed with particular anthropomorphic features, such as an animated face and speech capabilities; and 2) if an online store decides to employ a humanoid agent, whether or not the agent be customized to match some of the users' demographic characteristics, such as gender and ethnicity. The study of these two issues is motivated and justified by several considerations. First, this study extends previous research on how online stores should use emerging technologies to enrich online consumers' shopping experiences and make them comparable or even superior to what physical stores could offer (Qiu 2002). Shopping activities, on one hand, are analytical tasks during which a shopper tries to identify her needs and makes decisions on purchasing the products that best satisfy her; on the other hand, they are also social experiences which consist of interacting with other people, such as friends, salespeople, and other shoppers. A significant number of the social interactions involved in shopping occur between shoppers and sales associates, when shoppers inquire about product attributes or ask for product recommendations directly. Nevertheless, when consumers shop online, the lack of social interactions and the absence of a "human touch" are significant disadvantages or even inhibitors. Although advanced real-time online service features, such as "Live Help", can, to some extent, emulate consumer-to-serviceperson interactions (Qiu and Benbasat 2006), the obstacles inherent in employing stand-by servicepersons and associated higher labor costs of doing so make it financially infeasible for most online vendors to provide these human-4 Chapter 1: Introduction supported services. A good alternative is to use web-based decision support software; however, research on whether or not PRAs should be designed with capabilities for interacting with shoppers in a socially appealing manner is still sparse. Among the few empirical investigations that have addressed this issue, Cassell and Bickmore (2000) have found that an embodied conversational agent that can simulate interaction rituals among humans, such as greetings, small talk, and conventional leave-taking, along with specific elements of these interactions in speech and in embodied conversational behavior, increase users' trust in agents. Research in social psychology has clearly established that people generally prefer to interact with and are more inclined to trust others with similar social identities as themselves (Clark and Maass 1988; Lee 1993). Furthermore, recent studies (e.g., Nass et al. 2000) have attempted to study whether or not computer users would perceive the arguments provided by an embodied conversational agent of higher quality when an agent's ethnicity matches those of its users' ethnicity. By manipulating the salient characteristics of a humanoid PRA, this dissertation reinvestigates the significance of "similarity-attraction" effects in human-agent interaction. Correspondingly, the main effects of demographic variables, such as users' gender, can be more thoroughly investigated. Realizing the multi-faceted characteristic of users' social experiences with PRAs, this dissertation focuses on three outcome variables: 1) social presence, which is defined as the extent to which a PRA is perceived as social and personal; 2) users' trusting beliefs directed towards agents; and 3) users' perceptions of enjoyment, a construct measuring users' feelings that their interactions with a PRA are playful and pleasant. Besides 5 Chapter 1: Introduction exploring how a PRA's multimedia and anthropomorphic interfaces can shape and affect users' perceptions of these outcome variables, this dissertation attempts to integrate them into the well-established Technology Acceptance Model (TAM) (Davis 1989) to demonstrate how a PRA's interface components can ultimately influence shoppers' intentions to adopt an agent as a decision aid. This research also provides an opportunity to appraise the value of anthropomorphism in human-computer interface design. The question of whether or not the implementation of humanoid interface agents would be beneficial in human-computer interaction has been debated through many studies, but unequivocal results are still extremely rare (see Dehn and van Mulken 2000). Perhaps the appropriate question we should ask is not whether or not such design is universally useful, but rather, "what kind of animated agent used in what kind of domain influences what aspects of the user's attitudes or/and performance" (Rickenberg and Reeves 2000). For online vendors, what matters may be a PRA of what anthropomorphic interface features could be most influential in building an affinity relationship with the agent, boosting consumers' trust in the agent, or convincing customers to adopt the agent as a decision aid for future purchase. 1.2 Method and Main Findings To investigate the above research questions, two laboratory experiments were conducted. Experiment 1 addressed the importance of demographic embodiment of a PRA, namely ethnicity and gender, in influencing users' first impressions of the agent's sociability and trustworthiness. Experiment 2 addressed the distinct and combined effects of anthropomorphic embodiments and output modalities in the implementation of a humanoid PRA. 6 Chapter 1: Introduction The major findings of the two experiments are summarized as follows: • Similar to the effects of social stereotypes in human-human communications, "in-group" favoritism also emerges in human-agent interactions. Ethnicity-matched PRAs are perceived as more sociable, more competent, and more enjoyable to interact with than their mismatched counterparts. People also exhibit stronger trusting beliefs in a PRA of the same gender as themselves, compared with an opposite-gendered PRA. More interestingly, the results, for the first time, illustrate a relationship between the two demographic factors: ethnicity-interactions are more significant among females than males on social presence and perceived enjoyment. • The results confirm the significance of both "persona-effect" and "modality-effect" in the design of anthropomorphic agent interfaces. The use of humanoid avatars augments consumers' perceptions of social presence, their beliefs about the competence of an agent, and their perceived enjoyment. The use of a human voice also appears to be significantly more effective than on-screen text in creating a social and enjoyable interaction experience. • By integrating the constructs of social presence, trust, and perceived enjoyment with the Technology Acceptance Model, the important role of social relations in influencing user adoption of the agents is confirmed. Social presence is found to be a common antecedent of both trust and perceived enjoyment. Trust exerts a direct impact on intentions to adopt recommendation agents, as well as an indirect impact on adoption via users' perceptions of the usefulness of the agents. 7 Chapter 1: Introduction Perceived enjoyment also influences adoption intentions, as mediated by perceived usefulness and perceived ease of use. 1.3 Structure of the Dissertation The remainder of the dissertation is structured as follows. Chapter 2 reviews the existing literature about online recommendation agents, and establishes the importance of social interactions in the design of PRAs. The issue of whether or not employing anthropomorphic interface benefits human-computer interactions (HQ) is also briefly reviewed. Chapter 3 introduces the three key constructs employed in the present study: social presence, trusting beliefs, and perceived enjoyment. The importance of these constructs in evaluating the social aspect of human-PRA interactions as well as the relationship between these three constructs and traditional TAM constructs are discussed. Chapter 4 introduces the first experiment, which investigates the effects of PRAs' social identities on consumers' evaluations of the agents. Two identity factors are investigated: gender and ethnicity. We propose that an identity-matched PRA will be perceived more favorably than a mismatched one. Through a laboratory experiment, four implementations of PRAs with different gender and ethnicity combinations are compared. The hypothesized interaction effects are tested utilizing both multivariate and univariate analyses of covariance. Chapter 5 concentrates on the influences of certain anthropomorphic interface features on users' perceptions of social presence, trusting beliefs, and perceived enjoyment. Through a controlled laboratory experiment, six types of recommendation 8 Chapter 1: Introduction agents with different interface features are compared. In this experiment, the individual and joint influences of humanoid embodiment and output modalities in facilitating human-agent social interactions are investigated. Chapter 6 discusses the interrelationship between social presence, trusting beliefs, and perceived enjoyment, as well as how users' social, relational, and affective perceptions of interactions with a PRA influence their behavioral intentions of adopting the agent. These hypotheses are tested by performing a Partial Least Squares (PLS) analysis with the data collected from the two aforementioned experiments. Finally, Chapter 7 summarizes the findings from both studies, outlines the major contributions offered by this dissertation, and provides suggestions for future research. 9 Chapter 2: Literature Review C H A P T E R 2 : L I T E R A T U R E R E V I E W This chapter discusses the importance of social interactions in human-PRA communications, and the possibility of achieving this goal using emerging multimedia technologies. A definition and taxonomy of PRAs are provided first, followed by a comparison between PRAs and other traditional information sources from which shoppers are likely to seek information before purchasing. After discussing the importance and implications of social interactions in designs for PRAs, prior research on anthropomorphic interfaces utilized in human-computer interaction are briefly reviewed. Finally, two multimedia technologies that have been investigated in the research for this dissertation, namely animated faces and speech output, are introduced. 2.1 Product Recommendation Agents (PRAs) 1 Originally designed to filter information and to automate repetitive tasks (Maes 1994), software agents have evolved into important tools for electronic commerce (Maes 1999). Maes et al. (1999) have classified e-commerce software agents into six major categories by their primary functionalities: • need identification; • product brokering; • merchant brokering; • negotiation; • payment and delivery; and 1 Researchers in the field of Artificial Intelligence (Al) have been working on the development of software programs, such as Chatbots, which interacts socially with people using naturalistic language, for several years. Chatbots are computer programs designed to converse as much as possible like natural persons. However, designs progressing from the first well-known "Eliza", designed by computer scientist Joseph Weizenbaum in 1966, to the most recent "Alice", developed by Dr. Richard Wallace, have not yet been perfected; even though the Natural Language Processing (NPL) capability of these designs have improved over time, modern chatbots are still prone to getting lost in free-content chat, and nowhere near ready for commercial applications. PRAs discussed in this dissertation are not chatbots, but rather decision-support tools. 10 Chapter 2: Literature Review • service and evaluation. In a similar vein, Vaidyannathan and Aggarwal (2002) have examined the roles that shopping agents can play at different stages of the consumer decision-making process: problem recognition, information acquisition, alternative evaluation, and purchasing. Most recommendation systems currently employed by online vendors belong in the category of product brokering, and most agents are used primarily in the stage of alternative evaluation; their key functions involve retrieving and organizing relevant product information, and helping users choose what to purchase. Based on the mechanisms at the heart of their design, product recommendation agents can be categorized into two major types: collaborative fdtering systems and attribute-based content filtering systems (Ansari et al. 2000). Other categorization schemes have been introduced in various studies, most of which involve similar divisions; for example, the conjoint-inferred-preference system and self-explicated-preference system, discussed by Vaidyannathan and Aggarwal (2002), and the profile-based agent and self-expressed preference-based agent categories discussed by Lee et al. (2002). Collaborative filtering systems suggest products of interest to consumers based on the preferences of similar consumers, using a variety of data mining algorithms. Attribute-based systems, on the other hand, usually ask consumers to provide information regarding attributes that are most important to their purchase decisions. The most significant difference between these two types of recommendation systems is whether explicit "conversations" occur between the agent and the shopper. For this reason, attribute-based systems are also referred to as "Query-Based Decision Aids" (Pereira 2001). Chapter 2: Literature Review This dissertation focuses on attribute-based content filtering recommendation systems. An attribute-based system develops a preference structure for each particular user by inquiring about the product attributes which she considers important and the relative weight she would ascribe to each of these attributes. Based on this information, the PRA computes a preference matching score for each of the alternatives in the database. The preference matching score provides an indication of how closely each of the alternative products in the product set matches the preferences expressed by the user. In order to elicit shoppers' preferences and to offer relevant advice in response, attribute-based recommendation agents need to extract sufficient information from potential buyers to ensure that the recommended products fit the needs, uses, and preferences of each consumer. Usually this information extraction process is implemented through a series of question-and-answer interactions. Aggarwal and Vaidyannathan (2003) have proposed two types of preference elicitation mechanisms: conjoint-type full-profile ratings, through which consumers' preferences for various attributes and specifications are inferred from the consumers' ratings of several alternative products, and self-explicated ratings, in which consumers are directly asked about their evaluations of various attributes and levels. Another classification system has been proposed by Russo (2002), who has identified two types of interaction mechanisms as: 1) static surveys, which comprise a fixed set of questions to be answered by a buyer; and 2) dynamic surveys that use Artificial Intelligence (Al) technology to contextualize questions, such that each of the questions that are asked are based on the responses to the previous questions. A few empirical studies have evaluated various methods of preference elicitation. Among them, Komiak (2003) has compared needs-based dialogue, 12 Chapter 2: Literature Review in which agent-users are asked questions about how they intend to use the products, and attribute-based dialogue, in which users are asked to state their preferences for each attribute explicitly. The results reveal that needs-based dialogue increases consumers' perceived personalization of the agents, which in turn increases their cognitive trust in the agent. Olson and Widing (2002) have compared interactive agents (which ask users to identify and weigh their opinions of the relative importance of particular attributes) with passive agents (which apply equal weight to each attribute); passive agents appear to perform as well as or even better than interactive agents in terms of decision quality, consumer approval of the agent, and the time required to make decisions. Treating a recommendation agent as a decision support system (DSS) designed particularly for shoppers in e-commerce domains, another stream of PRA studies has been devoted to the more-traditional topics of an agent's underlying algorithm and its strategy choices (e.g., Grenci and Todd 2002). Among these studies, Pereira (2000) has compared four different search strategies (elimination by aspect, weighted average, profile building, and simple hypertext), revealing that subjects with relatively extensive knowledge of a particular class of products have more positive affective reactions towards agents that use the weighted average method and the elimination by aspect method, while subjects with low product class knowledge prefer the profile building based agents. Furthermore, researchers have also confirmed that explanation facilities can play an important role in users' acceptance of an agent (Wang 2005), echoing the findings of prior DSS research (Gregor and Benbasat 1999). 2 Perceived personalization is a consumers' impression that an agent is operating specifically in their interests, on the basis that the agent has adopted the preferences and desires of the consumer; this type of impression increases consumers' trust insofar as the consumers believe that the agent is pursuing their interests rather than trying to compel them to purchase specific products. 13 Chapter 2: Literature Review In evaluations of the impact of recommendation agents on consumer behavior, empirical research in marketing has demonstrated that the availability of an attribute-based recommendation agent can provide several benefits: 1) an agent can reduce the amount of effort consumers spend in collecting and analyzing pre-purchase information; 2) it can improve consumer perceptions of the accuracy of the decisions they make; and 3) it can influence their purchase decisions (Haubl and Trifts 2000; Pereira 2001). More specifically, Haubl and Murray (2003) have found that the inclusion of an attribute in a recommendation agent renders this attribute more prominent in a consumer's purchase decisions Furthermore, Diehl et al. (2003) have discovered that a smart quality-screening agent can increase consumers' price sensitivity. Other factors that may moderate the impact of recommendation agents on decision quality have also been examined, such as customer product class knowledge (Pereira 2000), product category risk and product complexity variables (Swaminathan 2003), and source websites (Senecal and Nantel 2004). Another important but under-studied area of recommendation agent research is the design of agent interfaces. Although several reports have discussed how an agent's interface might affect users' performance and perceptions (e.g., Carroll and McKendree 1987; Bickmore and Cassell 2001), only a few studies have been dedicated to empirical analysis of the interfaces of recommendation agents in online shopping environments (McBreen and Jack 2001; Keeling et al. 2004). This issue will be revisited in Section 2.3, where more detailed reviews on the anthropomorphic features of interface agents are presented. Chapter 2: Literature Review 2.2 Recommendation Agents as Social Actors Consumers' purchase decisions are usually influenced by information collected (either passively or actively) from various sources. The impacts of information sources on consumer decisions in traditional commerce have been studied extensively in marketing literature. For instance, four types of recommendation sources have been identified by Andreasen (1968): 1) impersonal advocate sources (e.g., commercials or advertisements in mass media); 2) impersonal independent sources (e.g., third-party publications such as Consumer Reports); 3) personal advocate sources (e.g., salespersons); and 4) personal independent sources (e.g., friends). In other words, most consumers are used to seeking help from either friends/family members or salespersons if they want to get personalized and customized advice. However, recommendation systems implemented on e-commerce websites have made it possible that a personalized recommendation can be also provided by a technological artifact (Senecal and Nantel 2004). Now the question is whether consumers would treat recommendation agents more like a person, and consequently evaluate its performance and credibility with the same set of rules based on their past interpersonal experience, or they would simply utilize the agent as a complicated but inanimate calculator. When consumers seek advice from a knowledgeable friend or a family member, established relationships and past experiences may make the consumers believe that the information provider is impartial and competent. When the product recommendation is provided by a salesperson and consumers do not have adequate product expertise to evaluate the credibility of the information provided, their impressions of and their social relationships with the salesperson could exert a significant influence on their evaluations 15 Chapter 2: Literature Review of the recommender, especially during their first encounter when no past experience or third-party testimonies could serve as a reference for their assessments. Salespersons that can effectively employ various social cues along with their product expertise are more likely to create favorable perceptions and build a higher quality relationship. In the dyadic salesperson-customer interactions, non-task linguistic cues can be used to strengthen the social relationship between the two parties. For example, Jacobs et al. (2001) have identified two types of self-disclosure or information exchange in customer-salesperson interactions: exchange-specific self-disclosures, which directly relate to a pending transaction (e.g., disclosing to a camera salesperson concerns about picture resolution or zoom capabilities); and social self-disclosures, which are incidental to the pending transaction (e.g., sharing with a camera salesperson the details of one's recent vacation). It has been demonstrated that social self-disclosure is related positively to the antecedents of commitment for future business. Research has also found that social conversations grant customers a special intimate status (Berry 1995) and open communications foster customer trust (Morgan and Hunt 1994). On the other hand, nonverbal cues can also influence selected dimensions of consumers' initial perceptions of a salesperson during initial and subsequent encounters, whether the presentation is made in person or via a video-taped sales presentation; these cues, defined as "behavior other than words themselves that form a socially shared coding system" (Burgoon 1994, pp. 231), may include eye movements, speech hesitations, gestures, clothing, and posture (Leigh and Summers 2002). Thus far, most PRA research has focused on the product-attribute-related information exchange between agents and their users; while the social interactions have been largely 16 Chapter 2: Literature Review ignored. One possible explanation for this imbalance is that software agents are usually treated as software tools rather than potential social actors, due to their obvious lifeless and mechanical nature. However, recent studies in social psychology have found contrary results. According to Reeves and Nass (1996), due to years of continuous interactions with other people, consumers generally possess mental models that they would apply to computers or other technological artifacts, ascribing social behavior and social attributions to the technology, especially when the technology possesses a set of characteristics normally associated with human behavior. Moreover, people tend to engage in socially desirable behavior when interacting with computers. To investigate how people's social responses to computers equipped with certain "social" characteristics, various empirical studies have examined: 1) the use of language (Clark 1996), such as flattery (Fogg and Nass 1997), apology (Tzeng 2004), politeness (Mayer et al. 2006), self-disclosure (Moon 2000), the position or job title of the source of information (Sundar and Nass 2000), and perceived personality (Nass et al. 1995); 2) the use of nonverbal cues, such as demographic identities (Nass et al. 1997; Nass et al. 2000; Lee 2003); facial expressions (Bartneck and Reichenbach 2005), forms (Lee and Nass 2002); voices (Nass and Lee 2001; Lee and Nass 2003), and postures (Isbister and Nass 2000); and 3) consistency between verbal and non-verbal cues (Isbister and Nass 2000). Results of these empirical studies provide strong support to the "Computers as Social Actors" (CASA) paradigm. Recommendation agents, as suggested by CASA theory, could also be treated socially by online shoppers, which implies that agent developers could create favorable consumer perceptions by designing and manipulating various social cues. For example, in their 17 Chapter 2: Literature Review studies of embodied interface agents, Cassell et al. (1999) and Cassell and Bickmore (2000) have found that non-task linguistic cues, such as small talk, can enhance users' trusting beliefs about an agent's benevolence. In contrast, this dissertation focuses more on the impacts of non-verbal cues, especially those afforded by an anthropomorphic interface, in human-PRA social interactions. 2.3 Anthropomorphic PRA Interface3 The word "agent" conventionally refers to an individual, for example, a real-estate agent, a librarian, or a salesperson, someone who is empowered to act for or to provide suggestions to the principal. This relationship applies analogously in online shopping environments as well. For example, Alba et al. (1997) have likened a screening agent to a "supersalesperson " with excellent knowledge of both the qualities of the inventory and the tastes of the buyer; Russo (2002) has contrasted a PRA with a "wise uncle". However, whether or not the personification analogy should be extended from a semantic symbol to explicit interface design remains to be investigated. Anthropomorphizing an agent entails adding human-like characteristics, such as facial expressions, body gestures, speech recognition and output, auditory and kinesthetic feedback, models of human discourse and emotions, and social intelligence (i.e., emulations of common social conventions). When anthropomorphic agents are able to interact with users using natural language and to carry on "conversations", they are also known as Embodied Conversational Agents (ECAs). As defined by Cassell et al. (2000), 3 In the literature, two frequently mentioned concepts regarding animated embodiments are anthropomorphic agents and avatars. According to Badler (1997), an anthropomorphic agent is a virtual human figure representation that is created and controlled by computer programs, whereas an avatar is a virtual human controlled by a live participant. In the present research, we use the terms anthropomorphic agent and avatar interchangeably to refer to software-powered product recommendation agents that are presented with humanoid appearance and behavioral characteristics. 18 Chapter 2: Literature Review ECAs are computer-generated characters "that demonstrate many of the same properties as humans in face-to-face conversation, including the ability to produce and respond to verbal and nonverbal communication". Multiple ECA prototypes have been designed for academic research, for example, the real estate agent REA (Cassell and Bickmore 2000; Bickmore and Cassell 2001), Greta (de Rosis et al. 2003; Berry et al. 2005), and various pedagogical agents (Lester et al. 1997; Lester and Stone 1997). There are also a growing number of commercial ECAs, such as those developed by Artificial Life, Oddcast, and Haptek. These systems vary greatly in their linguistic capabilities, input modalities, and task domains, but all share the common feature that they attempt to engage users in natural conversation, with an on-screen face or body. A majority of previous anthropomorphic interface and ECA research has focused on the embodiment design of agents, such as the agent's realism (McBreen and Jack 2001), facial expressions (Koda 1996; Link et al. 2001; Bartneck and Reichenbach 2005; Berry et al. 2005), demographic features (Baylor 2005; Cowell and Stanney 2005), communication modalities (Moreno et al. 2001), or nonverbal behavior (Bickmore and Cassell 2001; Cowell and Stanney 2005). The effects of ECAs have also been evaluated in various application contexts, such as education (Lester et al. 1997; Lester and Stone 1997; Dempsey and Van Eck 2003) and electronic commerce (McBreen and Jack 2001; Keeling et al. 2004). A selected summary of empirical evaluations of anthropomorphic interface is provided in Table 2-1. 19 Chapter 2: Literature Review Table 2-1: A Summary of Selected Empirical Evaluation Studies on Anthropomorphic Interface Agents' Paper Independent Variables and Moderators (Interface Components) Dependent Variables and Results Application Context (Walker et al. 1994) Synthesize Talking face (Face + Text-to-Speech voices) vs. Text Stern face vs. Neutral face • Time: face > text; stern face > neutral face; Mistake: face < text; stern face < neutral face; • Comments: face > text; stern face > neutral face; • Likeability: stern face > neutral Face Interview / Questionnair e (Takeuchi and Naito 1995) Synthesized Face vs. Arrow • Engagingness: face > arrow • Usefulness: face < arrow Poker Game (Sproull et al. 1996) Synthesized Talking Face (Synthesized Face + Text-to-Speech voices) vs. Text • Perceived intelligence: no difference • Social evaluation: talking face < text • Sociability: talking face < text • Arousal: talking face > text Counselor Interview (Koda and Maes 1996) Face (Caricature or Realistic) vs. No Face • Intelligence: no difference • Likeability: face > no face • Engagingness: face > no face • Comfortability: face > no face • Attention attraction: face > no face Poker Game (Lester et al. 1997) Animated pedagogical agents with various level of communicative behaviors • Believability: expressive agent > muted agent • Usefulness: expressive agent > muted agent Tutoring system (van Mulken et al. 1998) No character vs. with character (persona); all with Text-to-Speech voices Moderator: Technical vs. non-technical information • Engagingness: persona > no persona (technical info only) • Usefulness: persona > no persona (technical info only) Tutoring system 4 In this table, only studies that empirically evaluate anthropomorphic interfaces are included. Papers only describing the design of such interface as well as general discussion/review ones are not listed. Chapter 2: Literature Review Table 2-1: A Summary of Selected Empirical Evaluation Studies on Anthropomorphic Interface Agents (Cont.) Paper Interface Components (Independent Variables) Dependent Variables and Results Application Context (Nass et al. 2000) Ethnicity of an interface agent (simulated with real-human video): Caucasian vs. Korean; subjects are all Korean males • Value Congruence: in-group (Korean agent) > out-group (Caucasian agent) • Socially attractiveness: in-group > out-group • Trustworthiness: in-group > out-group • Argument quality: in-group > out-group Collaborative choice-dilemma task (Bickmore and Cassell 2001) Task-oriented dialogue vs. Social dialogue generated by an embodied character with Text-to-Speech voice; Moderator: introvert vs. extravert subjects • Trust: Social dialogue > Task dialogue (for extravert subjects only) • Engagingness: Social dialogue > Task dialogue Information Presentation (Linketal. 2001) Various pitch and speed of Text-to-Speech voice various facial expression of an agent • Feedback rating: no difference on voices; mouth curve up > mouth curve down Learning feedback (McBreen and Jack 2001) Video vs. a 3-D talking head vs. an image with quasidynamic facial expressions vs. a still image vs. a disembodied voice; all with human voice. Moderator: human-like vs. cartoon-like • Overall rating (human-like agents): Video > Voice only > still image > image with facial expressions > 3D talking head • Overall rating (cartoon-like agents): no difference. Online Shopping (Moreno et al. 2001) With animated character vs. without character Text vs. voice • Knowledge transfer: character> no character • Study interests: character > no character; voice >text • Retention: voice > text Tutoring system (Moundridou and Virvou 2002) With embodied (face + Text-to-Speech voice) agent vs. without embodied agent (no face + text) • Enjoyment: with face > without face • Ease of use: with face> without face • Usefulness: with face > without face Tutoring system (Baylor and Ryu 2003) With image vs. without image Static image vs. animated image All with both voice and on-screen text. • Credibility: with image > no image • Person-like: animation > static • Instructor-like: animation > static • Engagingness: animation > static Tutoring system Chapter 2: Literature Review Table 2-1: A Summary of Selected Empirical Evaluation Studies on Anthropomorphic Interface Agents (Cont.) Paper Interface Components (Independent Variables) Dependent Variables and Results Application Context (Nowak and Biocca 2003) No image vs. less-anthropomorphic image vs. highly anthropomorphic image • Telepresence: low-anthropomorphic image > high-anthropomorphic image • Copresence: low-anthropomorphic image > high-anthropomorphic image • Social presence: low-anthropomorphic image > high-anthropomorphic image Virtual environment (Baylor 2005) Gender and ethnicity of the agent; Cartoon-like agent vs. realistic agent; Moderator: agent's role as Expert vs. Motivator • Subjects tend to choose an agent of the same ethnicity when giving a chance to choose. • Usefulness / interesting / satisfaction: male agents > female agents • Intelligence, knowledge, expert-like: male > female (motivator agent) • Competence: male > female (expert agent) • Enjoyment / enthusiasm / motivation: African-American > Caucasian (motivator agent) Tutoring system (Berry et al. 2005) Human vs. animated agent (face + Text-to-Speech Voice) vs. Text-to-Speech Voice-only vs. Text-only • Ease of understanding: text-only > agent, TTS-only, human • Trustworthiness: TTS-only, text-only, agent > human • Helpfulness: human, agent > TTS-only • Likeability: human > TTS-only Information Presentation /Persuasion (Cowell and Stanney 2005) Demographic embodiments of a humanoid agent; Non-verbal interaction styles • Users prefer to interact with characters that match their ethnicity and are young looking (when only the picture of an agent was presented); • Credibility: agent with trustworthy facial non-verbal behavior > agent without non-verbal behavior; no difference on postures / gestures. Collaborative task (photo sorting) Chapter 2: Literature Review In HCI research, the effectiveness of animated interface agents is historically controversial (Dehn and van Mulken 2000). Proponents argue that an animated agent enjoys the benefits of social interfaces, natural and comfortable human-computer interaction, conversational communication, and user-to-agent delegation (Laurel 1990). The advocates also believe that an animated agent renders a computer system more human-like, which can make it more engaging and motivating in knowledge-based learning environments (Elliot et al. 1997; Lester and Stone 1997). However, critics have countered that human-like systems may induce users to form false mental models of the system and to develop erroneous (typically too high) expectations about a system's cognitive and emotional behavior (Norman 1994; Shneiderman and Maes 1997). In addition, animated agents may be distracting and may unnecessarily consume users' cognitive resources, which may hamper the users' performance (Walker et al. 1994). As Dehn and van Mulken (2000) have pointed out, one of the reasons for inconsistent results in previous empirical studies of anthropomorphic agent applications is that these studies have been based on different experimental treatments, control conditions, dependent variables, and experimental settings, which make a meta-comparison across studies rather difficult. As illustrated in Table 2-1, most empirical studies on anthropomorphic interface/agents are conducted in the context of either entertainment or computer-aided tutoring. The effectiveness of such interface in online shopping environment and for the task of product advising has not been empirically validated. Compared with other dedicated systems such as computer games or computer-aided learning software, humanoid interface agents hosted on online shopping websites are limited by various 23 Chapter 2: Literature Review technological and economic constraints. For example, to display a fully-embodied animated character within Internet browsers, users may have to download and install additional software; however, owing to security and privacy concerns, shoppers are usually reluctant to do this when they visit unfamiliar e-commerce websites. On the other hand, the resources it takes an online vendor to design and maintain a highly vivid and expressive anthropomorphic interface could be another concern, especially when no solid evidence is available to show these applications could in fact contribute to customer experiences and worth the extra cost. Therefore, with the implementation feasibility in consideration, we examine in this dissertation a design of humanoid agents that are: 1) based on a widely-adopted Internet multimedia platform so that no additional plug-ins would be required for ordinary online shoppers; and 2) embodied with animated human-like faces but not the full body, which could not only simplify the design but also exclude potential confounds of body look and gestures. Actually, a recent study has shown that the marginal contributions from the gestures/pastures of animated characters on top of those induced by facial expressions are not significant (Cowell and Stanney 2005). What is more, most anthropomorphic agents employed in prior studies are presented with an animated character and voices (either human speech or computer-synthesized speech) simultaneously, which makes it difficult to identify the unique contribution of embodiment effects and modality effects, as well as the possible interactions between these two factors. For online vendors, however, whether to use on-screen text or give the agent a speech interface is a decision that has to be made wisely, as either setting up a text-to-speech engine for computer-synthesized voices or hiring professional voice talent for pre-recorded human speech would incur extra costs. 24 Chapter 2: Literature Review Based on the above discussion, we select the following two interface components in the following empirical studies: embodiment (whether an agent should be represented with an animated human-like face) and modality (whether an agent should use voice to interact with users). 2.3.1 Humanoid Embodiment (by Animated Human-Like Faces) When engaged in unmediated interpersonal communication, people use physical characteristics of their communication partners as the basis for identification and mental model formation (Ichheiser 1970; Argyle 1988). Among various bodily characteristics, faces are probably the most human-specific component. Infants are born with information about the structure of faces; at birth infants exhibit preferences for face-like patterns over others (Bond 1972). It has also been found that a large portion of the brain is allocated to face recognition (Eckman 1973). Faces can also induce appropriate behavior in social situations. For example, particularly attractive faces are found to be effective in improving people's responses to advertisements (Baker and Churchill Jr. 1977); heightened physical attractiveness of faces in turn generally enhances an individual's effectiveness as a social influence agent (Chaiken 1981) . Past research has demonstrated that, when interacting with computers, although to some degree people respond socially to all interfaces (Reeves and Nass 1996) regardless of the use of humanoid embodiments (Klein et al. 2002), users' social responses appear to be stronger with more "human"-looking images (Turkle 1995; Koda 1996; Nass et al. 1998), and people are more likely to interact with human-looking interfaces in a social way - the so-called persona effect (Lester et al. 1997). Chapter 2: Literature Review Human faces have been used in human-computer interaction in a variety of ways. Some research has investigated the effects of static images or pictures. For example, Riegelsberger et al. (2003) have found that photographs of staff members on an e-commerce website increase consumer perceptions of the trustworthiness of poorly performing vendors while decreasing that of vendors with good reputations. Other research has employed taped video-clips or computer-generated facial animation. For example, Walker et al. (1994) have found that when using a "talking face" to administer a questionnaire, users spent more time, made fewer mistakes, and wrote more comments, compared with those who answered questions presented via on-screen text displays. The same pattern appeared when users were presented with a face that was "stern", compared with a "neutral" face. Sproull et al. (1996) have further observed that users' responses to a text-display interface differed from their responses to a talking-face interface, to which they attributed some personality traits that caused the users to become more attentive. Lee and Nass (2002) have found that users attribute greater competence, social attractiveness, and trustworthiness to communication partners represented by anthropomorphic characteristics than those represented by textboxes or "stick figures". One notable limitation in the aforementioned literature is that most animated faces employed have been combined with speech output, which makes it difficult to distinguish the effects of anthropomorphic embodiments and multimodal (visual and audio) presentations. Recent research has discovered that faces and voices are processed holistically by human beings (Massaro 1998; Nass and Gong 2000). The McGurk Effect, which is used to demonstrate that syllables are perceived differently if an individual's lips suggest a different sound, is a notable example of how humans tend to integrate voices 26 Chapter 2: Literature Review and faces when both are present (Massaro 1998). Nass and Gong (1999) have further demonstrated significant inter-relations between faces and voices: synthesized speech with a synthesized face and a recorded speech without a face are evaluated more favorably compared with synthesized speech alone or a synthesized face with a recorded speech. To avoid this potential confound, the present dissertation investigates the distinct effects of an agent's facial animation and those of its speech output, as well as the potential interactions between the two, through a full-factorial experimental design. 2.3.2 Voice 5 Voice carries much richer information than the literal meaning of its denotative content. For example, psychologists have found that people can detect differences between emotions based on the tone of a voice almost as well as from facial expressions (cf. Scherer 2003). People can also perceive social status, personality, and attitudes from others' speech styles (Giles and Powesland 1975), and they can use vocal communications to determine group identity characteristics, such as age, gender, occupation, and geographic origins (Pittam 1994), as well as emotions (Pittam and Scherer 1992; Pittam and Scherer 1993; Scherer 2003) and attitudes (Giles 1970; Ryan and Giles 1982). As claimed by Nass and Scott (2005), "As a result of human evolution, humans are automatic experts at extracting the social aspects of speech". When speech is applied in computer interfaces, it is quite natural that users will behave towards and make 5 A fully duplex speech interface includes both input and output channels. Automatic Speech Recognition (ASR) technologies are widely implemented for input functions, whereas pre-recorded human-speech or computer-synthesized speech is used to generate for output. In the present study, voice interface is based on the output channel only. 27 Chapter 2: Literature Review attributions about the voice systems using the same rules and heuristics they would normally apply to other humans (Reeves and Nass 1996; Nass and Scott 2005). These social responses to speech interface may even be extended from human speech to computer-synthesized speech. Computer-synthesized speech generators, also known as Text-To-Speech (TTS) systems, are computer-based systems designed to read text aloud (Dutoit 1997). The fundamental difference between a TTS system and any other talking machine (e.g. a cassette-player) is the ability of the former to automatically produce new sentences. In other words, a TTS system does not simply concatenate isolated words or parts of sentences, but it also automatically produces speech through a grapheme-to-phoneme transcription of sentences (Dutoit 1997). Compared with pre-recorded human speech, TTS systems enjoy a higher level of flexibility, since the speech can be generated on-the-fly and "produce" almost infinite arrays of content. They are implemented widely in the area of automated voice response services, Interactive Voice Response (IVR) systems, computer games, intelligent agents, and Virtual Reality Environments (VRE). Successful applications of these systems include information access, customer ordering, information for drivers, and interfaces for the disabled. Nevertheless, TTS has some intrinsic deficiencies because it lacks the quality and prosody of natural human speech. A comparative study by Beutnagel et al. (1999) has demonstrated that currently available TTS systems perform pretty well in terms of their intelligibility, although they are still not satisfactory in terms of their naturalness. Computer-generated voices tend to exhibit unnatural pauses, accents, and word emphases, as well as discontinuities between phonemes and syllables (Nass and Lee 28 Chapter 2: Literature Review 2001). Even with the best TTS synthesizers on the market, such as AT&T Natural Voices, after listening to a couple of sentences, it is still quite easy for users to distinguish it from the lowest-fidelity presentation of human speech. Limited research has explored users' responses to such "unnatural" machine voices, and the results of the research are mixed. On the one hand, due to its mechanic nature, synthetic speech is not as socially arousing as human speech. For example, users who listen to synthetic speech interface tend to find significantly less communicability and less comfort with a synthetic voice system compared to users presented with text-based or recorded-speech presentations (Nass et al. 2003). On the other hand, a series of studies by Nass and his colleagues (Lee et al. 2000; Nass and Lee 2000; Nass and Lee 2001; Lee 2003) have revealed that people attribute some social characteristics to computer-generated voices. For example, users regard computer voices as more attractive, credible and informative, and they tend to evaluate products more positively, if the perceived personality of the voice matches their own personalities (Nass and Lee 2001; Lee 2003). It has also been observed that users attempt to perceive social cues embedded in voices even when they knew that the voice is not from a real person (Nass and Lee 2001; Lee 2003; Lee and Nass 2004). 2.4 Chapter Summary This chapter introduces general functionalities of a PRA and discusses the implications of social interaction design of PRA interface. Prior empirical studies on anthropomorphic interface/agent are also reviewed. More specifically, the potential impacts of two anthropomorphic components, namely humanoid embodiment and voice output, are described. In the next chapter, we will elaborate on the key dependent 29 Chapter 2: Literature Review variables, namely social presence, trust, and perceived enjoyment, which are employed to evaluate the social aspects of human-PRA interactions. 30 Chapter 3: The Dependent Variables C H A P T E R 3 : E V A L U A T I N G H U M A N - P R A S O C I A L I N T E R A C T I O N S : T H E D E P E N D E N T V A R I A B L E S The major goal of this study is to investigate whether the anthropomorphic interface of a PRA indeed has an effect on consumers' perceived social interactions with the agent. Recent research on anthropomorphic interface agents has shown that multi-modal information channels of an embodied agent, e.g., the capabilities of exchanging information via voice, gestures, gaze and facial expressions in addition to language, have the potential of providing users with natural means to foster affinity, trust, and pleasure as well as to exchange information with computers (Laurel 1990; Nagao and Takeuchi 1994; Cassell and Bickmore 2000; Cassell et al. 2000). In this research, we posit three constructs that are important for evaluating human-PRA social interactions, particularly in the context of online shopping: 1) social presence; 2) trusting beliefs; and 3) perceived enjoyment. This chapter introduces the three dependent variables and discusses the roles they play in users' perceptions of their social interactions with a PRA. 3.1 Social Presence Through a series of experimental studies, Reeves and Nass (1996) have convincingly demonstrated that people's spontaneous responses to computers follow the principles of human-human social interactions very closely and this social nature of human behaviors toward computers are universal and consistent. However, it is also acknowledged that to elicit such "mindless" social response, individuals must be presented with an entity that has enough cues to lead the person to categorize it as worthy of social responses (Nass and Moon 2000). In other words, computer or software endowed with various interface cues can differ in their capabilities of inducing social responses. 31 Chapter 3: The Dependent Variables In this research, we investigate social presence, a construct that refers to the feeling of "being with another" and is primarily used to measure how users sense the existence of other people in distant locations (Biocca et al. 2003), to describe the "quasi-social" relationships between the agent and its users and to evaluate the users' perceptual differences of the agent's social characteristics which is likely to be influenced by the anthropomorphic interface components. In this study, social presence is defined as the extent to which a PRA is perceived as a social actor, that is, a human advisor or salesperson. The term social presence was first used in studies of interpersonal communications within organizations, to compare the effects of various types of telecommunication technologies for remote business meetings. It was originally developed to measure people's subjective perceptions of other people when the communication is mediated by communication medium. Defined as the subjective evaluation of a communication medium on "the degree of salience of the other person in the interaction and consequent salience of the interpersonal relationships" (Short et al. 1976, pp. 65), it also describes the extent to which a medium is perceived as sociable, warm, sensitive, personal or intimate when it is used to interact with other people (Lombard and Ditton 1997). In traditional social presence theory, social presence is defined as the characteristic that describes the capacity of each medium to give immediate feedback as well as a variety of communication cues. It is conceptualized as a subjective measure of a communication medium, which is exclusively determined by the characteristics of the medium (Short et al. 1976). But recent viewpoints (e.g., Biocca et al. 2003) have tended to conceptualize social presence "as a transient phenomenological state that varies with 32 Chapter 3: The Dependent Variables medium, knowledge of the other, content of the communication, environment, and social context". We concur with Biocca et al. (2003) that users' feelings of social presence are formed through communication processes; therefore, the users are influenced not only by the medium, but also by the communication content and their possible courses of action as well. In other words, social presence is a subjective measure developed over the entire course of the interactions, rather than arising from a static inventory of various features of a communication medium. Earlier studies of social presence have generally investigated technologically mediated representations of other human beings, while recent studies have extended the topic to consider artificial representations of humanoid or animal-like intelligence, such as virtual human agents, computers, and robots (Biocca 1997). Biocca et al. (2003) have proposed that the theory of social presence should be extended beyond the context of intra- or inter-organization communications and to apply it in studying people's various social behaviors in all kinds of mediated environments. Lee and Nass (2003) also projected that social presence can be created through agents or through the machine itself, especially when the way of how people make sense of virtually presented others are undergoing significant changes with more frequent experiences in mediated virtual environments (via agents and avatars) and increasing communications among projected selves. In (Gefen and Straub 2003), it is argued that online shoppers' will treat the e-service provider as a social actor with whom the communication is mediated by the website system. Therefore, social presence could be increased by adding a personal touch, such as a personalized greeting to the user, or through picture and text content that convey a personal presence in the same way as photographs and personal letters can. 33 Chapter 3: The Dependent Variables Kumar and Benbasat (2004) also adopted social presence to assess the relational aspect of the communication between that web site and its customers. Social presence is also particularly relevant in online shopping environment. Gefen et al. (2003) have examined the effects of social presence on consumer trust in e-services, finding that social presence affects consumer trust, and increased trust subsequently has a stronger effect on purchase intentions than TAM beliefs do. In their study of IT-enabled support for personalization systems and virtual communities in online shopping websites, Kumar and Benbasat (2004) have found that social presence and perceived usefulness both significantly affect customer loyalty. 3.2 Trust Trust is very important in the adoption of new technologies such as online shopping and recommendation agents. In their study of research on experienced repeat online shoppers, Gefen et al. (2003) have found that consumer trust is as important to online commerce as perceived usefulness and perceived ease of use. In their studies of online recommendation agents, Wang and Benbasat (2005) have suggested that users' trusting beliefs in an agent exert significant effects on users' adoption intentions as well as user perceptions of the agent's usefulness. However, trust is employed in this research as a core dependent variable not only because of its critical role in influencing users' agent adoption decisions, but more importantly, its close relationship with the perceived social characteristics of a PRA. When interacting with a PRA for shopping advice, consumers delegate some tasks (e.g., find a product that fits their needs and preferences) to the recommendation agents and the agents work on behalf of the consumers. As a common risk in almost all 34 Chapter 3: The Dependent Variables commercial transactions, there are always possibilities of encountering opportunistic behavior by the agent which is difficult to discern. For example, without engaging in further information search activities, users may have a variety of concerns on the services provided by the agent, such as there may be models that fit their requirements better than those included in the list of recommendations, or the agents might be biased toward certain manufacturers. Despite the fact that PRA users can evaluate the agent's recommendations by examining whether or not all of their expressed requirements are met before deciding whether to follow the agent's suggestions or not, the knowledge asymmetry in consumer-PRA interactions makes consumers' trusting beliefs, especially the formation of initial trust, a leap of faith rather than a rational process solely based on cognitive analysis. Trust is an important indicator reflecting the quality of interpersonal social relationship (Blau 1964; Luhmann 1979; Fukuyama 1995). As a social construct that originates from interpersonal relationships (Sztompka 1999), trust is one of the most desired qualities in any close relationship (Rempel et al. 1985), as well as one that significantly influences users' intentions to adopt technological artifacts when they are perceived as social actors (Gefen and Straub 2003; Wang and Benbasat 2005). Due to the extensive research related to trust in a variety of fields, this section only gives a brief review of trust in traditional commerce, computer-mediated communications, and electronic commerce. A more comprehensive review of trust has been presented by (McKnight et al. 2002; Gefen et al. 2003). Trust in online recommendation agents is considered as an extension of interpersonal trust (Wang and Benbasat 2005). In traditional commerce, trust usually refers to the 35 Chapter 3: The Dependent Variables interpersonal trust between consumers and salespeople. In marketing studies, the concept of relationship quality, which to a great extent determines future sales opportunities (Crosby et al. 1990), is composed of two dimensions: (1) trust in a salesperson (Swan and Trawick 1988) and (2) satisfaction with a salesperson (Crosby and Stephens 1987). Swan and Bowers (1999) have further posited that customer trust in a salesperson has two components: affect and cognition. Affect is a feeling of security or insecurity about relying on a salesperson, and cognition is a belief that a salesperson has both the necessary competence and motivation to support user reliance. Komiak and Benbasat (2004) use emotional trust in replacement of affect and expand it to encompass a customer's feelings about relying on an entity, while dividing the cognitive components of trust into (1) cognitive trust in the competence of a salesperson, and (2) cognitive trust in the goodwill of a salesperson. Prior research on online consumer trust has focused on consumers' trust towards either a website or a company (e.g., Hoffman et al. 1999; McKnight et al. 2002; Gefen and Straub 2003). Recent research has circumscribed trust from four major perspectives: 1) viewing trust as a belief or a collection of beliefs that the trustee has attributes that are beneficial to the trustor (Bhattacherjee 2002); 2) viewing trust as emotional feelings of security, comfort, and lack of fear (Komiak and Benbasat 2004); 3) viewing trust as an intention or the willingness of a party to be vulnerable to the actions of another party (Mayer and Davis 1995); and 4) viewing trust as a combination of these elements (McKnight et al. 2002). In the present research, we adopt the belief approach presented by McKnight et al. (2002) and Wang and Benbasat (2005) and define trust in a PRA as user's beliefs in an 36 Chapter 3: The Dependent Variables PRA's competence, benevolence, and integrity. These three trusting beliefs have been well accepted in many recent studies (e.g., McKnight et al. 2002; Wang and Benbasat 2004; Wang and Benbasat 2005). According to McKnight et al. (2002) and Bhattacherjee (2002), competence-beliefs are users' beliefs that a PRA has the expertise and skills to do what users need; benevolence-beliefs are users' beliefs that an agent cares about them and acts in their interests; and integrity-beliefs are an individual's beliefs that the agent is honest and that it will keep its promise. There has been a debate about whether belief constructs such as trust can be applied to a technological artifacts, especially in terms of beliefs about benevolence and integrity (e.g., Chopra and Wallace 2003; Corritore et al. 2003). However, if users' perceptions, thought processes, and emotional responses are similar or identical to those found in human-human interaction when a medium or an actor within a medium is treated as a social entity, it is natural to argue that trusting beliefs are not an exception. Recent studies have indicated that although the formation processes of customer trust/distrust beliefs in "virtual salespersons" (software agents) and human salespersons are slightly different, trust in an embodied conversational agent also contains beliefs in competence, benevolence, and integrity (Cassell and Bickmore 2000), similar to trust in a human salesperson (Komiak et al. 2005). A summary of empirical studies of trust in recommendation agents is listed in Table 3-1. 37 Chapter 3: The Dependent Variables Table 3-1: A Summary of Empirical Studies of Trust in Recommendation Agents Paper Independent Variables Moderator Findings (Herlocker et al. 2000) (presence of) explanation interface NA • The addition of explanation facilities increased the acceptance of an online recommendation agent (Pereira, 2000) Agent's search strategy: elimination by aspects vs. weighted average vs. profile building Users' product knowledge: high vs. low • Users with high product class knowledge had more positive affective reactions (trust and satisfaction) to the content-filtering RAs than the collaborative-filtering ones. The reverse was true for users with low product class knowledge (Bickmore and Cassell 2001) (presence of) social dialogue from an embodied agent Personality traits of agent users: extrovert vs. introvert • Small talk had a significant effect on the trust assessment of extroverts but essentially no effect on how introverts assessed trust; (Sinha and Swearingen 2002) Perceived transparency of the recommender system NA • Users like and feel more confident in recommendations systems that are perceived as transparent. (Swearingen and Sinha 2001) Source of recommendations: friends vs. recommendation agent NA • While users generally trusted recommendations provided by their friends more than those from an RA, they found online RAs capable and useful • Familiar recommendations played an important role in establishing user trust in an RA • User trust in an RA increased when the RA provides detailed product information (Komiak 2003) Internalization: Needs-based agent vs. attribute-based agent Familiarity: One-time use vs. repeated use Customer's product knowledge • For users with low product knowledge, a needs-based recommendation agent resulted in higher trust than an attribute-based agent; • Users' familiarity with the workings of an RA increased users' trust in an RA's benevolence and integrity, but not its competence Chapter 3: The Dependent Variables Table 3-1: A Summary of Empirical Studies of Trust in Recommendation Agents (Cont.) Paper Independent Variables Moderator Findings (Senecal and Nantel 2004) Recommendation source: none, human expert, other consumers, recommendation agent Product type: experience product vs. search product; Website type: seller vs. commercially linked third-party vs. independent third-party • Subjects who consulted product recommendations selected recommended products twice as often as subjects who did not consult recommendations. • Recommendation agent was more influential than "human experts" and "other consumers". • The type of product also had a significant influence on the propensity to follow product recommendations. • The type of a website (i.e., seller, commercially linked third-party, or independent third-party) that provides an RA did not affect the perceived trustworthiness of the RA. (Wang and Benbasat 2004) Type of explanation facilities provided: how, why, guidance NA • The use of how explanations increased consumers' beliefs in an RA's competence and benevolence; • The use of why explanations increased their beliefs in an RA's benevolence; • The use of decisional guidance increased their beliefs in an RA's integrity. (Komiak et al. 2005) Trust towards salespeople vs. trust towards recommendation agents NA • Trust in a recommendation agent contains trust in competence, benevolence, and integrity, similar to trust in a human salesperson. • The formation processes of trust in virtual salespersons, trust in human salespersons, distrust in virtual salespersons, and distrust in human salespersons are different. (Wang 2005) Perceived restrictiveness; Perceived transparency NA • Explanations enhanced perceptions of the transparency of an RA's inner workings, resulting in increased user trust in the RA; • RAs that were perceived by users as more restrictive were considered less trustworthy and useful by them. Chapter 3: The Dependent Variables 3.3 Perceived Enjoyment Perception of enjoyment is another construct employed to evaluate the quality of consumer's social interactions with the agent. Prior research on animated interface agents (e.g., Koda 1996; Lester and Stone 1997) have used constructs such as engagement and entertainment to measure users' affective reactions when they were interacting with an anthropomorphic agent. In the present dissertation, the construct perceived enjoyment is employed to describe the hedonic value of an anthropomorphic PRA interface. Perceived enjoyment in this research refers to the extent to which the activity of interacting with the recommendation agent is perceived to be enjoyable in its own right aside from the utilitarian value of the agent (Davis et al. 1992). It is closely related to the concept of intrinsic motivation, which refers to "the performance of an activity for no apparent reinforcement other than the process of performing the activity per se" (Davis et al. 1992 pp. 1112). The importance of intrinsic motivation as a lever to create favorable user perceptions has been empirically demonstrated by (Venkatesh 1999; Venkatesh 2000; Venkatesh et al. 2002). For example, prior research proposed enjoyment as a determinant of behavioral intention (Davis et al. 1992; Venkatesh et al. 2002) and as a determinant of ease of use (Venkatesh 2000; Venkatesh et al. 2002). In human-computer interaction (HCI) and information systems (IS) literature, perceived enjoyment has been considered as a sub-dimension of various constructs measuring users' intrinsic motivation towards or enjoyable experience from using a technology. For example, in (Moon and Kim 2001), it was presented as a dimension of perceived playfulness, an intrinsic belief or motive formed from an individual's subjective experiences using a technology such as the Internet. Perceived enjoyment is also included 40 Chapter 3: The Dependent Variables as a component of flow, which was initially presented as an optimal and enjoyable experience in which people are completely absorbed in their activities (Csikszentmihalyi 1975). The flow construct was then adapted in the context of information technology and computer-mediated environments as a subjective psychological experience, in which the human-computer experience as characterized as playful and exploratory (Webster et al. 1993). Enjoyment is also recognized as one of five dimensions of cognitive absorption, a similar construct which describes a state of deep involvement with software (Agarwal and Karahanna 2000). In recent years, perceived enjoyment has received increasingly interests in HCI and IS research, as researchers start to acknowledge that how enjoyable an information system is may be as important as how usable and useful it is (Blythe et al. 2004). Fun and enjoyment in the home and leisure context are as important as productivity and efficiency in the work context. The influence of enjoyment perceptions on users' technology adoption intentions have been empirically examined in applications such as general computer usage (Igbaria et al. 1995), Internet usage (Teo et al. 1999; Moon and Kim 2001; Van der Heijden 2003), instant messaging tools (Li et al. 2005), Internet-based learning medium (Yi and Hwang 2003; Lee et al. 2005), and online shopping (Koufaris 2002; Van der Heijden 2004). In marketing literature, the hedonic aspects of shopping are also drawing researchers' interests. Although many motivations exists as shopping goals (Westbrook and Black 1985), most marketing research considers instrumental and hedonic motivations as fundamental to understanding consumer shopping behavior because they maintain a basic underlying presence across consumption phenomena (Babin et al. 1994). Hedonic 41 Chapter 3: The Dependent Variables shopping motives are similar to the task orientation of utilitarian shopping motives, only the task is concerned with hedonic fulfillment, such as experiencing fun, amusement, fantasy, and sensory stimulation (Babin et al. 1994). Past studies have indicated that the hedonic and immersive aspects of an online shopping website are as important predictors of shopper's attitude as the instrumental aspects are (Childers et al. 2001). As a result, shopping enjoyment can be an important determinant of online customer loyalty (Jarvenpaa and Todd 1997; Koufaris 2002). Feeling of enjoyment could be induced by various activities during online shopping. Interacting with a product recommendation agent is expected to contribute to users' perceptions of enjoyment not only through its outcome - personalized suggestions on product choice - but also from the interaction per se. In traditional retailing environments, salespeople are found to influence how customers feel when shopping, through buyer-seller interaction (Lee and Dubinsky 2003). Research has shown that when customers experience positive emotions, they tend to be more satisfied with their buyer-seller relationships (Reynolds and Beatty 1999b), and they are more likely to complete a purchase (Reynolds and Beatty 1999a). When some part of a salesperson's work (i.e., product advising) is performed by PRAs, it is also possible for system designers to use animated interface agents to influence users' affective response by inducing human social interpersonal reactions toward interface agents (Takeuchi and Katagiri 1999; Takeuchi et al. 2000). 3.4 Chapter Summary In this chapter, we introduced the three key dependent variables that are utilized to evaluate various facets of users' social interactions with a PRA. The definitions of the 42 Chapter 3: The Dependent Variables constructs were clarified and their implications in PRA interface design were also discussed. In Chapter 4, we will present the first empirical study of this dissertation, in which the impacts of a humanoid PRA's demographic embodiments on these key dependent variables are investigated. 43 Chapter 4: Influences of a PRA's Demographic Embodiment CHAPTER 4: INFLUENCES OF A PRA'S DEMOGRAPHIC EMBODIMENT ON USERS' PERCEPTIONS 4.1 Introduction Social evaluation of a humanoid agent is highly dependent on the particular implementation used (Dehn and van Mulken 2000). To elaborate on whether or not PRAs embodied with animated avatars and/or speech output capabilities would be perceived more positively by users than those with text-only interfaces, a comparatively favorable implementation needs to be identified first. To compare various implementations, prior research on anthropomorphic interface had focused on factors such as the characters' non-verbal behaviors (Lester and Stone 1997; Cassell et al. 1999; Cowell and Stanney 2005), appearance (McBreen and Jack 2001; Baylor and Ryu 2003), language (Bickmore and Cassell 2001), personality (Isbister and Nass 2000), emotion (Brave et al. 2005). However, one fundamental issue, the agent's demographic embodiments, has received fairly little attention; furthermore, conclusive results based on rigorously designed empirical investigation are extremely scarce in the literature. Compared with an agent's behavioral interface features such as facial expressions and body movements, its demographic characteristics are relatively static and stable, and therefore, technically easier to implement and manipulate. However, the importance of picking the "right" embodiment should not be underestimated, as it affords important non-verbal cues to PRA users. A humanoid PRA can convey considerably more non-verbal cues than a textual agent. A taxonomy schematized by De Meuse (1987) on non-verbal cues employed in interpersonal communications categorized these cues in a two-dimensional matrix (as 44 Chapter 4: Influences of a P R A ' s Demographic Embodiment shown in Figure 4-1). One dimension is individual control, indicating to what extent a non-verbal cue can be controlled by the person. For example, gender and ethnicity are considered as cues that are difficult to control, because they are inherent and relatively enduring in nature and consequently an individual who engages in conversation cannot modify these cues. In contrast, physical appearance variables, such as make-up, hair color, mode of dress, and physical attractiveness are subject to frequent changes, and thus they are considered to be cues that are high in individual control. The other dimension is whether or not the non-verbal cues are behavioral, in other words, whether the cues originate from certain activities. Examples of behavioral non-verbal cues include facial expressions, eye contact, posture, and gestures. Behavioral Non-verbal behavioral variables Origin Non-behavioral 3 Demographic variables Physical appearance variables L o w High Individual Control Figure 4-1: Taxonomy of Non-Verbal Cues (De Meuse 1987) Given that De Meuse's taxonomy is applied in interpersonal interactions, demographic variables are identified as low in individual control. However, P R A designers could easily assign various overt demographic characteristics to the agent 45 Chapter 4: Influences of a PRA's Demographic Embodiment without modifying its underlying algorithm. One objective of this study is to find out whether PRA's demographic variables such as gender may have a direct impact on users' evaluations of the social interactions in the specific task domain of advice-giving in online shopping environment. More importantly, this study aims at examining the interaction effects in regards to demographic variables in dyadic human-PRA interactions. As the same salesperson could be appraised in different ways by individual customers, the influence of the agent's demographic embodiments may not work monotonically; instead, they could operate in conjunction with certain characteristics of PRA users. Nowadays, online vendors could collect intelligence about certain demographic information of their customers through historical transaction records, tracking methods such as "cookies", or users' voluntary disclosure. Finding out whether or not PRAs with particular demographic identities will be perceived more favorably by customers of certain category can help PRA designers employ a more personalized design and improve users' interaction experience. Among the few studies that have attempted to explore this issue, Nass et al. (2000) have investigated the effect of ethnicity as an in-group cue in human-computer interactions. During the experiment, 40 Korean users were asked to work with one of the two embodied agents (one looks like a Korean and the other appears to be a Caucasian) to solve several choice-dilemmas1. The results suggest that the Korean agent was perceived to be more socially attractive and trustworthy than Caucasian agent, and users tended to The task was described as to evaluate the agent's decision and arguments for a hypothetical situation. There are usually two alternative courses of actions for the situation; one usually has the potential for both greater benefit and greater harm. 46 Chapter 4: Influences of a PRA's Demographic Embodiment perceive greater value congruence with Korean agent. In addition, the Korean agent was generally perceived as providing higher quality arguments, which led to greater conformity to the recommendation provided by the agent. Cowell and Stanney (2005), in a study of anthropomorphic computer characters' credibility, have found that a significant majority of participants selected a character matching their ethnicity, while there was no significant difference in their preference for the character's gender. In their study of embodied pedagogical agents, Baylor (2005) has also found that, when given a chance to select an agent with various demographic characteristics (e.g., ethnicity and gender) before actually interacting with the agent, subjects tended to choose an agent of the same ethnicity as themselves. When an agent was randomly assigned to the subjects, male agents were generally preferred by subjects. In addition, there were interaction effects between the agent's gender and the role it plays: female agents were preferred when the agents were performing the role of "motivator" to facilitate the learning process while male agents were rated higher when serving the role of an expert. Nevertheless, a number of validity issues may have significantly reduced the explanatory power of these studies2. In (Nass et al. 2000), the agent users, were interacting with is a "full-motion video of a person" rather than a program-based agent. Subjects were simply told that their interaction partner is a computer agent. The use of human video in replacement of the agent reduces the generalizability of the conclusion as it is technically infeasible to build an agent that looks and acts as a real human. This also weakened the predictive power of "computers as social actors" theories. In addition, as 2 (Baylor 2005) was not included in this section of discussion because no detailed statistics were included in the original paper. 47 Chapter 4: Influences of a PRA's Demographic Embodiment only one group of subjects (Korean participants) were involved in the study, it is difficult to exclude the main effects of the agent's ethnicity (due to other differences on the agent's interface, the Caucasian agent could be perceived worse than the Korean agent by Caucasian participants as well). In (Cowell and Stanney 2005), when asked to indicate their preferred agent, subjects were simply shown the pictures of the agent candidates but did not actually interact with the agent and did not complete a collaborative task. This proxy could undermine the conclusiveness of the results, as people may behave differently towards a static picture than they would after interacting with an agent in a specific task domain to perform a task. Also, as the investigation of demographic embodiments was included merely as a pretest for the main study (on the agent's non-verbal behaviors), only 45 subjects with an unbalanced distribution on gender (20 males and 25 females) and ethnicity (20 Caucasians, 9 Asians, and 16 African Americans) were invited. The lack of statistical power creates another threat to the validity of the findings, especially those insignificant ones. In addition, the dependent variable employed in the demographic embodiment evaluation is the subject's preference for a specific implementation (picture) of the agent, which gave little information on why a user preferred to interact with a certain type of agent. Based on the discussion above, we deem that the research inquiry of how a humanoid PRA's demographic embodiments influence users' evaluations on the human-agent social interactions warrants a more rigorous investigation. Employing similarity-attraction hypothesis and social identity theory, this chapter first provides theoretical justifications for the differences in users' perceptions of social presence, trust, and enjoyment towards a PRA that is embodied with matched or mismatched demographic identities, namely 48 Chapter 4: Influences of a PRA's Demographic Embodiment gender and ethnicity3. A series of hypotheses on the potential "match-up" interaction effects are then proposed, and a controlled laboratory experiment was designed and conducted to empirically test these hypotheses. The results demonstrate that humanoid agents are indeed better perceived by users of similar demographic backgrounds, and these interaction effects are more significant and wide-spread in regards to ethnicity than gender. More interestingly, some unexpected three-way gender-ethnicity interactions reveal that the similarity-attraction effects on ethnicity may differ across genders. 4.2 Hypotheses Development The effects of gender and ethnicity on people's perceptions towards others have been widely studied in social psychology. Two theories have been frequently used to explain these gender or ethnicity interaction effects: similarity-attraction theories and social-identity theories. The similarity-attraction theory predicts that people are attracted to and seek membership in groups having members who are similar to themselves (Byrne 1971). Similarity between individuals includes attributes ranging from personalities to demographic variables, such as gender, race, and age. Such similarities have been found to increase interpersonal attraction and to influence whether one individual likes the other (Byrne et al. 1966). In the retail environment, Crosby et al. (1990) have proposed that similarities between salespersons and customers, as measured by variables such as appearance, lifestyle, and status, are positively related to relationship quality and sales 3 The factor of age was not included in this study because the perception of age could be relatively subjective and it could be easily influenced by other factors such as make-up, hair-style, dressing, etc. Also, the introduction of the third dimension will make a full-factorial design (as introduced in Section 4.3.1) overly complicated. However, we acknowledge that it is another fundamental and important demographic variable that deserves through investigation in future research. 49 Chapter 4: Influences of a PRA's Demographic Embodiment effectiveness. Similarity in gender and age also appears to be positively related to the relationship quality of the buyer-seller dyad (Smith 1998). Moreover, salesperson performance was found to be significantly related to visible similarities (e.g., gender, age, height, and race) (Churchill et al. 1975), and salespeople were found to be more attracted to prospects that are similar to themselves in terms of age and gender (Dwyer et al. 1998). The potential interaction in regards to demographic variables could also be explained by social identity theories (Tajfel 1969; Tajfel 1970; Tajfel 1978; Tajfel 1981). Social identity theories posit that membership in a group creates a psychological state that confers social identity. In effect, the perception of belonging to a group leads a person to identify with that group and with other members of the group. The psychological processes associated with social identity generate distinct group behavior, such as solidarity within one's group and discrimination against people who are not members of the group (Tajfel 1982; Tajfel and Turner 1986). An important and integral aspect of social identity theory involves self-categorization, a process that recognizes and emphasizes similarities between individuals in the same category and differences between individuals in dissimilar categories. Self-categorization is often based on social categories such as gender, race, and age (Messick and Mackie 1989). People tend to use these readily observable physical cues to categorize others as either "in-group" or "out-group" (Biernat and Vescio 1993), and to apply stereotypes to the others. Such stereotyping behavior usually involves in-group favoritism, inter-group discrimination, and affective preference for the in-group over the out-group (Allport 1954; Hogg and Abrams 1988; Brewer 1999). For example, individuals tend to agree with in-group members more than with out-group members (Clark and Maass 1988); furthermore, 50 Chapter 4: Influences of a PRA's Demographic Embodiment in-group members are perceived as more socially attractive (Lee 1993), more trustworthy (Clark and Maass 1988), and more competent (Stephan and Beane 1978). Previous research has also shown that similar sources are considered more credible than dissimilar ones (Simons et al. 1970), and that individuals are more persuaded by the same information from similar rather than from dissimilar others (Hogg et al. 1990). In this study, we posit that the salient ethnic characteristics of a humanoid PRA are likely to be noted by online consumers and used as a non-behavioral cue in their evaluation of the user-agent dyadic relationship. The agent is more likely to be treated as an "in-group" member when its ethnicity matches that of the users. As a result, the stereotypical in-group favoritism will make an "in-group" PRA be perceived as more trustworthy in terms of its competence, benevolence, and integrity. In addition, the attraction induced by the perceived similarity between the user and the PRA will lead to a desire for interaction and increased attention in the human-agent interaction. This focused and selective attention should lead to increased feeling of presence (Fontaine 1992). Also, as similarity may have a more direct effect of enhancing pleasurable and enjoyable interactions (Berscheid and Walster 1978), interacting with a PRA of identical ethnicity will be experienced as more enjoyable than interacting with a PRA of dissimilar ethnicity. Thus, we predict that: HI a: Users perceive stronger social presence from interacting with an "ethnicity-matched" PRA than with an "ethnicity-mismatched" PRA. H l b (c/d): Users have stronger competence (benevolence/integrity) beliefs regarding an "ethnicity-matched" PRA compared to an "ethnicity-mismatched" PRA. 51 Chapter 4: Influences of a PRA's Demographic Embodiment Hie: Users feel more enjoyable when interacting with an "ethnicity-matched" PRA than with an "ethnicity-mismatched" PRA. Beyond the interaction effects on the factor of ethnicity, the interaction between users' gender and the "gender" of a PRA could affect users' evaluation as well. As discussed in Section 4.1, gender is also an important and salient social category. The process of categorizing others in terms of gender is both habitual and automatic (Bern 1993). Therefore, in-group favoritism is expected to arise in same-gender interactions as well. However, the interaction effects of communicators' gender in interpersonal communications are not as clear-cut as those of ethnicity. One possible explanation is that such "in-group" effects are comparatively weak and, therefore, easily subordinated to other factors (Taynor and Deaux 1973). For example, Rudman and Goodwin (2004) argued that even though men, historically and cross-culturally, have been the dominant sex, they lack a mechanism that bolsters automatic own-group preferences, whereas women's automatic in-group bias is remarkably stronger. Men could show a pro-female bias due to maternal bonding, male intimidation, or positive attitudes towards sex. Supporting evidence has been found by Cabellero and Pride (1984), whose studies demonstrated that both male and female consumers tend to prefer highly attractive females, and they exhibited significantly more response to a catalogue offer made by an attractive female. A large portion of previous research on performance appraisals and the influence of ratee gender has indicated that gender-role stereotypes may be another confounding factor (De Meuse 1987). In marketing studies of personal selling, it has been found that 52 Chapter 4: Influences of a PRA's Demographic Embodiment product-related gender-role stereotypes can put salespeople of particular gender in a disadvantageous position when attempting to gain customer acceptance in certain product categories. Gender roles have been defined conceptually as the degree to which men and women are perceived to possess any one of a set of traits (Broverman et al. 1972). A gender-role stereotype is a collection of two subsets of masculine and feminine traits. Masculine traits are those that a high proportion of individuals agree are more descriptive of males than females; the reverse is true for feminine traits (Broverman et al. 1972). As a result, if few females are seen as experts or salespeople for certain product categories in the real world, it is likely that those roles are seen as male occupations. In studies of advertising and celebrity endorsements, the notion of advertising schemas (Lynch and Schuler 1994) has been employed to explain the "match-up" hypothesis, which argues that advertising response is influenced by a perceived match (or similarity) between an endorser's image attributes, brand image attributes, and the function of a product. Such a schema is defined as "a hypothetical cognitive structure that guides perception, thought, and action based on prior knowledge of stimuli gained through experience, media exposure, etc" (McDaniel 1999). The "match-up" effects have even been found in human-computer interactions: "male" computer interfaces elicit greater conformity in regards to masculine topics while "female" computer interfaces are considered more trustworthy regarding feminine topics (Lee 2003). In this study, we expect that before a user interacts with a PRA, a certain schema related to the image of a product expert or advisor may have been formed based on the user's past experience, and gender is an important part of this schema. A PRA would receive a higher rating on its performance when its gender matches the "ideal" gender in 53 Chapter 4: Influences of a PRA's Demographic Embodiment a user's mind. However, when users' stereotypical opinions of a PRA's gender is controlled as a covariate, we still expect the "in-group" effects of the aforementioned dependent variables to arise, coinciding with social identity theory which suggests that demographic similarities will have positive effects on performance by increasing interpersonal attraction and decreasing cognitive biasing (Linville and Jones 1980). Therefore, we propose following exploratory hypotheses: H2a: When their gender-role stereotype is controlled for, users will feel stronger social presence when interacting with a same-sex PRA than with an opposite-sex PRA. H2b (c/d): When their gender-role stereotype is controlled for, users will develop stronger beliefs in the competence (benevolence/integrity) of a same-sex PRA than in an opposite-sex PRA. H2e: When their gender-role stereotype is controlled for, users will perceive interactions with a same-sex PRA as more enjoyable than with an opposite-sex PRA. 4.3 Research Method A laboratory experiment was conducted to test the hypotheses listed in section 4.2. Section 4.3.1 describes the experimental design. The design of the humanoid agent, both its mechanism and interface, is introduced in Section 4.3.2. Dependent variables and control variables are listed in Sections 4.3.3 and 4.3.4. Section 4.3.5 describes the participants, incentives, and experiment procedures. 4.3.1 Experimental Design A 2 * 2 x 2 x 2 factorial experimental design (i.e., agent's gender x agent's ethnicity x user's gender x user's ethnicity) was employed for the study. A simulated product 54 Chapter 4: Influences of a PRA's Demographic Embodiment recommendation agent was designed with four permutations on two demographic variables (Caucasian, Asian; Male, Female); four groups of participants were recruited for the experiment, whose gender and ethnicity were also permuted. By collapsing the results on gender and ethnicity, it could be treated as two 2><2 factorial designs (as shown in Table 4-2 and Table 4-3). Although the focus of this study is to examine the significance of two first-order interactions (gender x gender and ethnicity x ethnicity) as well as gender differences on users' perceptions of social presence, a 2 x 2 x 2 x 2 full factorial design make it possible to explore all interactions between gender and ethnicity4. Table 4-1: Experimental Design of Study One Participant Male Female Agent Asian Caucasian Asian Caucasian Male Asian IA IB 2A 2B Caucasian IC ID 2C 2D Female Asian 3A 3B 4A 4B Caucasian 3C 3D 4C 4D Table 4-2: Experimental Design of Study One (Gender Factor) Male Subject Female Subject "Male" Agent Condition 1 Condition 2 "Female" Agent Condition 3 Condition 4 4 More specifically, what we were trying to identify is whether or not the two hypothesized first-order interactions may be moderated by a third factor; i.e., whether a potential three-way interaction may affect our interpretations on the two-way interactions. Also, we have no intentions to propose and test any four-way interactions as they are not only extremely complicated to interpret but also beyond the relationships we planned to examine. 55 Chapter 4: Influences of a PRA's Demographic Embodiment Table 4-3: Experimental Design of Study One (Ethnicity Factor) Asian Subject Caucasian Subject "Asian" Agent Condition A Condition B "Caucasian" Agent Condition C Condition D According to Cohen (1988), the required sample size for an experiment is decided by four factors: 1) the research design; 2) the critical value for statistical significance (a); 3) the desired level of power (1-P); and 4) the estimated effect size. To the best of our knowledge, no prior studies have empirically studied both gender- and ethnicity-interactions in human-computer interfaces, therefore, a medium effect size was assumed (Cohen 1988). Based on the DF-ES-POWER-ALPHA table (Cohen 1988 pp. 311-314), it was determined that a minimum number of 144 subjects are needed in order to ensure sufficient statistical power (power = 0.8) at the significance level of .05 for both main effects and interactions effects (including two-way, three-way, and four-way interactions). We fulfilled this requirement by recruiting a total number of 160 participants (see Section 4.3.5 for more details on experiment participants), which corresponds to 40 subjects per group for either gender x gender or ethnicity x ethnicity interactions and 20 per group for any three-way interaction analysis. 4.3.2 The Agent The agent designed for the experiment was a recommendation agent for digital cameras. It makes recommendations based on the preferences and needs specified by consumers, applying a content-filtering method. Digital cameras were chosen as the target product because of the complexity of their product attributes, the large number of alternative models available on the market, and the short lifespan of each generation of 56 Chapter 4: Influences of a PRA's Demographic Embodiment products. Without a recommendation source, consumers would have to spend extensive cognitive effort to learn about the product attributes for each camera, to compare the alternatives, and to make a final decision about purchasing a camera. Previous research has shown that the content-filtering based agent technology works best for relatively complex products (Russo 2002); in real-life practices, most content-filtering product recommendation agents (such as myproductadvisor.com) have been designed for technology-oriented products, such as computers, electronics, and cars. The underlying algorithm of the agent was designed to simulate the recommendation agents on ActiveDecisions.com and myproductadvisor.com; the former is currently the leading agent provider and its software is widely used by many websites, such as RadioShack.com and Sony.com; the latter is a popular independent product advising service provider. To elicit users' preferences and needs, an agent-user dialogue was used to simulate dialogues presented in other studies (Russo 2002) and in commercial applications. Users were required to answer all ten product-related questions and to indicate the importance level of each attribute in their decisions before a recommendation list was provided by the agent (the questions are listed in Appendix B). The humanoid characters were designed with Oddcast Sitepal, a Flash-based animation design tool that can integrate both animated characters and speech output with textual information. Sitepal provides a wide-array of animated characters, and each character can be further customized in terms of its physical appearances (e.g., face, eyes, nose, hairstyle, make-up, attire, jewelry and accessories, etc.). As one of the most widely-used plug-ins for Internet browsers, Macromedia Flash is considered an ideal platform for online stores to integrate rich-media elements with HTML-based contents. 57 Chapter 4: Influences of a PRA's Demographic Embodiment In order to select four animated characters exhibiting most salient gender and ethnic characteristics, two rounds of pretests were conducted. A sample of 25 graduate students and faculty members were asked to evaluate pictures of 20 candidate characters on their ethnic and gender characteristics. Based on their feedback, four characters (Asian Male, Asian Female, Caucasian Male, and Caucasian Female) were picked. In the second round of pretests, 70 undergraduate and graduate students were asked to evaluate the ethnicity, gender, physical attractiveness, and professionalism of these four characters. The results showed that all of the participants could correctly identity the character's ethnicity and gender, and there is no significant difference in the agent's physical attractiveness (F (3, 207) = 1.044, p > 0.30)5 and professionalism (F (3, 207) = 1.560, p > 0.20f. These four characters are shown in Figure 4-2. Asian Female Asian Male Caucasian Female Caucasian Male Figure 4-2: The Four Animated Characters Used in the Experiment 5 Group means of these four agents are: Asian Female = 4.49, Asian Male = 4.37, Caucasian Female = 4.63, Caucasian Male = 4.59, with a 7-point Likert scale. 6 Group means of these four agents are: Asian Female = 5.10, Asian Male = 5.13, Caucasian Female = 4.90, Caucasian Male = 5.26, with a 7-point Likert scale. 58 Chapter 4: Influences of a PRA's Demographic Embodiment The screen layout of the agent was designed to simulate commercial websites such as discoveryouride.com. A screen shot of the Question page with the Caucasian Male agent is shown in Figure 4-3. The animated character is placed on the lower-left corner of the screen. The user is asked to select an answer to the question and to indicate the importance level of this attribute in her decision. Most questions are "needs-based" rather than "attribute-based", as they ask users how they will use the digital camera rather than requirements on technical specifications such as the number of megapixels for resolution. If the user needed to know more information for this attribute or feature, she could click the "About This Question" button beside the agent to obtain additional explanatory information. In order to make the human-PRA interaction feel more like a conversation, whenever a user selected an answer, the agent would provide some feedback comments, such as acknowledging user's input or providing additional decisional guidance. After the user completed all ten questions, the agent provided a list of recommendations, as shown in Figure 4-4. The user could return to the question pages at any time to revise her answers, and an updated recommendation list was provided accordingly. 59 Chapter 4: Influences of a PRA's Demographic Embodiment Question 1 How much are you planning to spend on this digital camera? A small amount (under $300) A medium amount (between $250 and $650) A large amount (over $600) No preference Importance of this Feature Figure 4-3: Study One - Screenshot of the Agent's Question Page The agents used pre-recorded human speech as their communication modality. A clip of speech would be played: 1) to greet the user and give a brief self-introduction when the agent was activated; 2) to read aloud the question when each question page was loaded for the first time; 3) to remind users if a certain question was not answered or the importance level was not indicated; 4) to give explanatory information when users clicked the "About this Question" button; and 5) to provide feedback comments when users clicked an answer (the interaction scripts are included in Appendix B). The speech is recited with automatic lip-synchronization, as well as with facial expressions and head movements. Two versions of human speech were recorded (male and female); both narrators are native English speakers. 60 Chapter 4: Influences of a PRA's Demographic Embodiment MyDCAdvisor 1 havt 10 r«comm»n<Ution* tor you R«comm«ndation No. 1 (Fit Scoro 79%) Cute Exilim EX-P800 CM.lt log* ftfMitl » « » . » Cltefcloe*ld«U»a Brand Rocomm«n<iation No 2 (Fit Seoro 79%) Casio Exilim EX-PJ-OO l*itcB Zoom ^UH^^L Rrcomme Rocommondation No. 3 (Fit Scoro 74%) Fuji FlnoPlx E5S0 ( M ii> o« *n«» Irani Mini Zoom Rocommondation No. 4 (Fit Scor* 74%) Sam«ung Digimax WOO CicKtogotdoUio tWaml 2o«m irnaMcK Rtcommondation No,« (Fit Scor* 72%) Sony Cyo.r-.hot DSC-P150 AbOUt My tooMOOM. (rand Recommendations t"tisum\ sow Pine Zoom ::: &ae« to »w ^ u«sKm::: Koptid fmooapowt 4xopiit*s rmogootiol dmoptkol (rramapoMil Figure 4 - 4 : Study One - Screenshot of the Agent's Result Page 4.3.3 Dependent Variables This study used validated scales for all dependent variables, with minor changes on wording (all measurement items are listed in Appendix A). The measures for social presence were adapted from (Gefen and Straub 2003; Kumar and Benbasat 2004). The original scales were designed to measure how users perceive an e-commerce website as a social actor. In the present study, the website has been replaced by the agent. The three trusting beliefs in recommendation agents have been adapted from (Wang and Benbasat 2004; Wang 2005), which in turn relied on web trust measures developed by (McKnight et al. 2002). The scales in (Wang and Benbasat 2004; Wang 2005) had already been 61 Chapter 4: Influences of a PRA's Demographic Embodiment adapted for product recommendation agents, therefore, we kept them intact. The measures for perceived enjoyment were adapted from (Koufaris 2002) and (Van der Heijden 2004), both of which have been used to measure the extent to which users feel a sense of fun from interacting with an online shopping website. In the present study, the shopping website was replaced with the recommendation agent. In the questionnaire, we also included the measures of Perceived Usefulness, Perceived Ease of Use, and Agent Adoption Intention measures for the PLS analysis discussed in Chapter 6. These items were adapted from (Davis 1989), which has been adopted and validated in numerous TAM studies. Since the constructs were measured by multiple items, summated scales based on the average scores of the multi-items were used in the analysis. Responses were recorded on a seven-point Likert scale (1 = "strongly disagree"; 7 = "strongly agree"). 4.3.4 Control Variables Prior research on trust has suggested that individuals' propensity to trust others should be a controlled factor, due to its potential influence on trust. As a personality trait, trust propensity can affect the likelihood that an individual will trust an other (Mayer et al. 1995; Lee and Turban 2001). As consumers develop trust, they look for cues and information about the trustees. The trust propensity of individual consumers magnifies, or reduces, the effectiveness of the cues and information provided by trustees (Lee and Turban 2001). McKnight et al. (2002) have further confirmed that consumers' disposition to trust likely influences their trusting beliefs. Therefore, trust propensity is included in this study as a control variable in analyzing the impact of explanations on trusting beliefs. The scales for this propensity have been adapted from (McKnight et al. 2002), who measured individuals' dispositions to trust. As these scales were designed for 62 Chapter 4: Influences of a PRA's Demographic Embodiment interpersonal trust, we employed another trust propensity measure - users' trust propensity in computers - which was adapted from (Lee and Turban 2001). The other control variable, which is particularly important when examining gender interaction effects, is users' product-related gender-role stereotypes regarding a digital camera expert/salesperson. As discussed in section 4.2, social psychology research has found that both genders are more easily persuaded when message content is relevant to the opposite gender's stereotypical social roles than to their own gender's role (Sistrunk and McDavid 1971). Furthermore, conformity to established majority opinions is greatest when the gender of the majority matches the gender-orientation of the issue in question (Feldman-Summers et al. 1980); for example, female speakers are perceived to be more trustworthy than males regarding messages about child care, whereas the opposite is true for the gun control messages (Falbo et al. 1982). It is very likely that these stereotypes exist and influence how customers interact with salespeople (Fischer et al. 1998). As digital cameras, the product selected for this experiment, are usually related to male buyers (Dholakia and Chiang 2002), consumers are more likely to link the role of digital camera experts or salespeople with male roles. As a result, this stereotype is likely to affect consumer evaluations of the capability of an agent with salient gender characteristics. A scale of four items was specifically designed for this study. All measurement items for control variables are listed in Appendix A. 4.3.5 Participants, Incentives, and Experiment Procedure As discussed in Section 4.3.1, a total of 160 students or staff members were recruited campus-wide from the University of British Columbia through online and offline advertisement. Forty participants were included in each of the four demographic groups: 63 Chapter 4: Influences of a PRA's Demographic Embodiment Asian Female, Asian Male, Caucasian Male, and Caucasian Female. Ten participants from each demographic group were randomly assigned to each of the four agent interfaces. As there are a variety of ethnic groups in Asia, only participants originating from East Asia, i.e., Mainland China, Hong Kong, Taiwan, Japan, and Korea, were invited to participate. As the agent interface required users to listen to the agent for explanations and comments, to avoid any potentially confounding effects from language, for East Asian participants, we only included volunteers of East Asian ethnicity who were either born in North America (Canada or the United States) or who had immigrated to North America in their early childhood and had completed elementary school and high school here7. The Caucasian participants were born in either Canada or the U.S., or at least completed their elementary school and high school in North America. To avoid potential biases in their evaluations and to make interactions with the agent more necessary to complete the task, only individuals who did not own or had not purchased a digital camera before, as well as those who had owned or purchased a digital camera but with low expertise were invited (Wang and Benbasat 2005). Subjects were asked to report their expertise level on digital cameras when signing up for the study. A 7 Our pretest demonstrated that international Asian students who had stayed in their mother country before they moved to Canada for their university education exhibited mixed ethnic identifications towards an agent who looks like an Asian but speaks fluent North-American accented English. Therefore, we decided to recruit only those Asians who have spent an extended period of time in North America. Although it was found that ethnic identity was weaker among those who arrived at a younger age and had lived longer in the new country (Rogler et al. 1980), a recent study (Rosenthal and Feldman 1992) has found that among adolescent Chinese immigrants, there was no change in the importance or positive valuation of ethnicity, despite the consideration that ethnic knowledge and behavior decreases between first and second generations. 64 Chapter 4: Influences of a PRA's Demographic Embodiment seven-point scale was used, ranging from "Not at all" (1) to "Expert" (7). Only those with a score of 4 or below were invited. The experiment was administered during a single one hour session for each participant. Each participant was welcomed and given an informed consent form to review. Each participant was then asked to fill out a short online questionnaire which included demographic questions and instruments to measure all control variables. A research assistant then trained participants on how to use the agent by playing a video clip (with narration) that explained the interface and all functionalities of the experimental PRA. Afterwards, the participant was asked to complete a simulated shopping task. To reduce the potential confound that the subject may have purchased a digital camera for a similar situation before, two shopping tasks were prepared and the participant was randomly assigned to one of them. One task asked the subject to choose a digital camera for a close friend (a photography enthusiast) as a wedding gift; the other Q asked the subject to select a model for a close family member (an amateur user) . Participants were informed that there was no time limit for the task, but most participants spent 20 to 30 minutes to complete the task. After the task, participants were asked to complete a post-task questionnaire which included measures of all dependent variables. Each participant was guaranteed a monetary compensation for his or her participation ($10). In order to motivate participants to view the experiment as a serious online shopping session and to increase their involvement, they were told that their choices would be evaluated and ranked by a digital camera expert on its appropriateness for the Post-hoc tests indicated no significant differences between tasks on all dependent variables. 65 Chapter 4: Influences of a PRA's Demographic Embodiment requirements described in the task scenarios. They were told that the top 25% best-performers would receive an extra bonus ($50). 4.4 Data Analysis and Findings This section first reports the demographic data of the participants in the experiment. The main and interactions effects of gender and ethnicity are examined separately in Section 4.4.2 and Section 4.4.3. In each section, a multiple analysis of covariance (MANCOVA) is firstly conducted to test whether or not the experimental treatments on gender or ethnicity exhibited an overall direct impact on the dependent variables (Hair et al. 1998). Based on the MANCOVA results, a follow-up analysis of covariance (ANCOVA) is performed on individual variables. Section 4.4.4 reported MANCOVA and ANCOVA results of potential interactions between gender and ethnicity based on a 2 x 2 x 2 x 2 factorial design. 4.4.1 Demographic Data Table 4-4 outlines the characteristics of the participants who volunteered in the experiment. The majority of participants were undergraduate students, and their ages were generally around 20. All participants had at least four years of Internet experience, and most had used the Internet for more than seven years. Table 4-4: Study One - Demographic Data # of Participants Percentage Age 18-20 86 51.25% 21-25 62 38.75% 26-30 10 6.25% 31-35 1 0.625% >35 1 0.625% Year in University 66 Chapter 4: Influences of a PRA's Demographic Embodiment Sophomore 58 36.25% Junior 43 26.875% Senior 42 26.25% Graduate 12 7.5% Staff 5 3.125% Years of Internet Experience 4-6 34 21.25% 7-10 116 72.5% >10 10 6.25% In the background questionnaire, participants were asked about their product expertise, attitudes towards computers, and perceived risk of online shopping. Their comfort levels with the Internet and online shopping were also assessed. No significant differences were found between subjects randomly assigned to the four experimental groups for each of these variables. Also, no significant differences were found between groups with respect to participants' age and years of Internet experience. 4.4.2 Ethnicity Potential main and interactions effects on gender and ethnicity were examined separately. The descriptive data and bivariate correlations of the dependent variables are shown in Table 4-5. The measures for all dependent variables have good reliabilities (Cronbach's alphas > .70). Table 4-5: Study One - Cronbach's Alpha and Correlations of Constructs Mean S.D. Cronbach's Alpha (a) 1 2 3 4 5 1. Social Presence 4.40 1.13 0.88 1.00 2. Competence Belief 5.63 0.80 0.79 0.35" 1.00 3. Benevolence Belief 5.47 0.96 0.83 0.57" 0.54" 1.00 4. Integrity Belief 5.64 0.76 0.73 0.40" 0.51" 0.77" 1.00 5. Perceived Enjoyment 4.62 1.09 0.89 0.68" 0.45" 0.47" 0.30" 1.00 ** Correlation is significant at the 0.01 level (2-tailed). 67 Chapter 4: Influences of a PRA's Demographic Embodiment 4.4.2.1 MANCOVA Results Since we predicted ethnicity-interaction on multiple dependent variables (i.e., social presence, trusting beliefs, and perceived enjoyment), MANCOVA was applied first to test whether the experimental treatments have an overall impact on the dependent variables (Hair et al. 1998). This study also tested several control variables, as discussed in Section 4.3.3. The four dispositional trust beliefs (competence, benevolence, integrity, and trusting stance), and user's trust propensity towards computer are included. The MANCOVA results on ethnicity are presented in Table 4-6. No main effects were significant, but the overall interaction effects were significant (Wilks' Lambda = 0.909, F (5, 147) = 2.946,/? < 0.02), which allows us to examine the potential ethnicity-interactions on individual dependent variables via ANCOVA. Table 4-6: Study One - MANCOVA Results on Ethnicity Wilks' Lambda D F F-Value P-Value Agent's Ethnicity 0.973 5, 147 0.821 0.537 Subject's Ethnicity 0.973 5, 147 0.810 0.544 Agent's Ethnicity * Subject's Ethnicity 0.909 5, 147 2.946 0.015 Dispositional Trust - Competence 0.954 5, 147 1.426' , 0:218 Dispositional Trust -Benevolence 0.951 : 5, 147 1.511 0.1 «MI Dispositional Trust - Integrity 0.987 5, 147 0.379 0.863 Dispositional Trust - Trusting Stance 0.996 5, 147 0.131 0.985 Trust Propensity - Computer 0.956 5, 147 1.363 0.242 4.4.2.2 ANCOVA Results Group means and standard deviations of dependent variables are reported in Table 4-7. ANCOVA results are shown in Tables 4-8 to 4-12. 68 Chapter 4: Influences of a PRA's Demographic Embodiment Table 4-7: Study One - Group Means and Standard Deviations by Ethnicity Agent User (N) Social Competence Benevolence Integrity Perceived Perceived Adoption Presence Beliefs Beliefs Beliefs Enjoyment Usefulness Intentions Asian Asian 4.68 5.69 5.53 5.71 4.91 5.61 5.23 (40) (1.17) (0.80) (0.85) (0.69) (1.05) (0.88) (1.37) Caucasian 4.11 5.50 5.58 5.74 4.32 5.41 4.67 (40) (1.20) (0.85) (0.86) (0.78) (1.07) (0.92) (1.58) Total 4.40 5.59 5.55 5.73 4.61 5.51 4.95 (80) (1.21) (0.83) (0.85) (0.73) (1.09) (0.90) (1.50) Caucasian Asian 4.08 5.48 5.13 5.43 4.37 5.05 4.59 (40) (1.16) (0.86) (1.12) (0.81) (1.22) (1.13) (1.61) Caucasian 4.73 5.83 5.67 5.68 4.89 5.64 5.22 (40) (0.83) (0.62) (0.93) (0.75) (0.90) (0.85) (1.47) Total 4.40 5.66 5.39 5.55 4.63 5.35 4.90 (80) (1.05) (0.77) (1.06) (0.78) (1.10) (1.03) (1.57) Total Asian 4.38 5.58 5.33 5.57 4.64 5.33 4.90 (80) (1.20) (0.83) (1.01) (0.76) (1.16) (1.04) (1.52) Caucasian 4.42 5.67 5.62 5.71 4.60 5.53 4.94 (80) (1.07) (0.76) (0.89) (0.76) (1.02) (0.89) (1.54) Total 4.40 5.63 5.47 5.64 4.62 5.43 4.93 (160) (1.13) (0.80) (0.96) (0.76) (1.09) (0.97) (1.53) Table 4-8: Study One - ANCOVA Results by Ethnicity (Social Presence) Source DF Sum of Squares Mean Square F P-Value Agent's Ethnicity 1 0.001 0.001 0.001 0.977 Subject's Ethnicity 1 0.064 0.064 0.053 0.848 Agent's Ethnicity * 1 14.884 14.884 12.329 0.001 Subject's Ethnicity Error 156 188.330 1.207 Table 4-9: Study One - ANCOVA Results by Ethnicity (Competence Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Ethnicity 1 0.021 0.021 0.035 0.853 Subject's Ethnicity 1 0.176 0.176 0.294 0.588 Agent's Ethnicity * Subject's Ethnicity 1 2.348 2.348 3.933 0.049 Dispositional Trust -Competence (Covariate) 1 1.924 1.924 3.223 0.075 Dispositional Trust-Trusting Stance (Covariate) 1 <0.01 < 0.(»l 0.000 0.998 Trust Propensity -Computer (Covariate) 1 775 1.775 2'.973 * 0!087„ Error 153 91.351 0.597 69 Chapter 4: Influences of a PRA's Demographic Embodiment Table 4-10: Study One - ANCOVA Results by Ethnicity (Benevolence Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Ethnicity 1 0.822 0.822 1.001 0.319 Subject's Ethnicity 1 1.879 1.879 2.288 0.132 Agent's Ethnicity * Subject's Ethnicity 1 2.003 2.003 2.439 0.120 Dispositional Trust -Benevolence (Covariate) 1 4.890 , 4.8'M) 5.955 0.016 Dispositional Trust -Trusting Stance (Covariate) 1 0.135 0.135 0.164 0.686 Trust F'ropcnsity -Computer (Covariate) 1 7.935 7.935 9.663 0.002 Error 153 0.821 Table 4-11: Study One - ANCOVA Results by Ethnicity (Integrity Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Ethnicity 1 1.228 1.228 2.246 0.136 Subject's Ethnicity 1 0.188 0.188 0.345 0.558 Agent's Ethnicity * Subject's Ethnicity 1 0.306 0.306 0.559 0.456 Dispositional Trust -Integrity (Covariate) 1 0.154 0.154 0.282 0.596 Dispositional Trust -TrustingiStance (Covariate) 1 0.155 0.155 0.284 0.595 Trust Propensity -ComputerXCovariate) 1 3.871 3.871 7.083 0.009 Error 153 0.546 Table 4-12: Study One - ANCOVA Results by Ethnicity (Perceived Enjoyment) Source DF Sum of Squares Mean Square F P-Value Agent's Ethnicity 1 0.016 0.016 0.014 0.906 Subject's Ethnicity 1 0.064 0.064 0.056 0.813 Agent's Ethnicity * Subject's Ethnicity 1 12.100 12.100 10.643 0.001 Error 156 1.137 70 Chapter 4: Influences of a PRA's Demographic Embodiment 4.80 H 4.70 H 4.60 H c j? re 4.40-s •o & 4.30-re E W 4.20-Ul 4.10-4.00H Asian Subject Ethnicity Asian Caucasian 1 Caucasian Agent Ethnicity Figure 4-5: Study One - Ethnicity Interactions on Social Presence 5.80 H c re c o> k_ n E •a S 5.60-n E *3 in m 5.50 H ;-:: 01 / Subject Ethnicity / Asian f Caucasian l i i i i i i i i i i i i i i i i i i i i i j j i i i i i i i i i i i i i ^ I Asian i Caucasian Agent Ethnicity Figure 4-6: Study One - Ethnicity Interactions on Trust in Competence Chapter 4: Influences of a PRA's Demographic Embodiment 5.00-4.90- I l l l l l f f l i l l i (0 c g 4.80-s "5 .E 4.70-E 1 (0 S ^ 4.60-£ !!!»! 111! Ill IIIIIll 1 fe! !« i i l i p i i I l l illlillll 1 llillll / Subject Ethnicity \ / Asian X Caucasian «iiiH'iiiiiiii:«;»Ni«iiirsiiiii«iiii«:iii:!B!i«a!«!7i:,HHi!iiBi«n!.««;iH«riiiiiiiiB«r«i r / i i s i :;;v;;;;;;;i;: n ipsg i s ! n • J 4 .50-(0 UJ ^mviMmiim i x/ •••• 4.40- mmmmm mmmmmmm. 4.30-i i Asian Caucasian A g e n t Ethnicity Figure 4-7: Study One - Ethnicity Interactions on Perceived Enjoyment As shown in Figure 4-5 to 4-7, the ANCOVA results reveal significant ethnicity-interaction effects, at the.05 level, for social presence (F (1,156) = 12.329, p < 0.01), competence beliefs (F (1,153) = 3.933,/? < 0.05), and perceived enjoyment (F (1,156) = 10.643,/? < 0.01). No significant interaction effects were found regarding benevolence beliefs (F (1,153) = 2.439, p > 0.10) or integrity beliefs (F (1,153) = 0.559, p > 0.10). Therefore, hypotheses HI a, Hlb, and Hie are supported while Hlc and Hid are not. When interacting with a same-ethnicity PRA, users not only perceived the agent as more sociable and the experience more enjoyable, but they also exhibited stronger beliefs in the agent's competence. No main effect of the PRA's ethnicity or user's ethnicity was found to be significant. Chapter 4: Influences of a PRA's Demographic Embodiment 4.4.3 Gender 4.4.3.1 MANCOVA Results The MANCOVA results on gender are presented in Table 4-13. Users' gender stereotypes regarding digital camera experts and salespeople are included as an additional covariate. No main effects were found to be significant, but the overall interactions effects were significant (Wines' Lambda = 0.916, F (5, 146) = 2.687, p < 0.03), which allows us to examine the potential gender-interactions on individual dependent variables via ANCOVA. Table 4-13: Study One - MANCOVA Results on Gender Wilks' DF F-Value P-Lambda Value Agent's Gender 0.950 5, 146 1.538 0.181 Subject's Gender 0.974 5, 146 0.773 0.571 Agent's Gender * Subject's Gender 0.916 5, 146 2.687 0.024 Gender Stereotype 0.979 5, 146 0.638 0.671 Dispositional Trust - Competence 0.949 5, 146 1.566 0.173 Dispositional Trust - Benevolence 0.947 5, 146 1.647 0.151 Dispositional Trust - Integrity 0.985 5, 146 0.442 0.8 IX Dispositional Trust - Trusting Stance 0.987 5, 146 0.370 ; 0.868 Trust Propensity-Computer 0.950 5. 146 1.553 0.177 4.4.3.2 ANCOVA Results Group means and standard deviations of dependent variables are reported in Table 4-14. ANCOVA results are shown in Tables 4-15 to 4-19. 73 Chapter 4: Influences of a PRA's Demographic Embodiment Table 4-14: Study One - Group Means and Standard Deviations by Gender Agent User(N) Social Competence Benevolence Integrity Perceived Perceived Adoption Presence Beliefs Beliefs Beliefs Enjoyment Usefulness Intentions Female Female 4.33 5.83 5.43 5.71 4.56 5.53 5.32 (40) (1.10) (0.78) (1.11) (0.83) (1.24) (1.11) (1.50) Male 4.09 5.43 5.26 5.50 4.40 5.27 4.46 (40) (1.17) (0.78) (1.03) (0.71) (0.96) (0.87) (1.75) Total 4.21 5.63 5.34 5.60 4.48 5.40 4.89 (80) (1.14) (0.80) (1.07) (0.77) (1.11) (1.00) (1.68) Male Female 4.58 5.46 5.47 5.54 4.91 5.33 4.79 (40) (1.05) (0.80) (0.94) (0.85) (1.15) (0.96) (1.44) Male 4.59 5.78 5.74 5.83 4.62 5.59 5.13 (40) (1.17) (0.77) (0.67) (0.61) (0.97) (0.92) (1.29) Total 4.59 5.62 5.60 5.68 4.77 5.46 4.96 (80) (1.10) (0.80) (0.83) (0.75) (1.06) (0.94) (1.37) Total Female 4.45 5.64 5.45 5.62 4.73 5.43 5.05 (80) (1.07) (0.80) (1.02) (0.84) (1.20) (1.04) (1.49) Male 4.34 5.61 5.50 5.66 4.51 5.43 4.80 (80) (1.19) (0.79) (0.90) (0:68) (0.96) (0.90) (1.56) Total 4.40 5.63 5.47 5.64 4.62 5.43 4.93 (160) (1.13) (0.80) (0.96) (0.76) (1.09) (0.97) (1.53) Table 4-15: Study One - A N C O V A Results by Gender (Social Presence) Source DF Sum of Squares Mean Square F P-Value Agent's Gender 1 5.709 5.709 4.534 0.035 Subject's Gender 1 0.443 0.443 0.352 0.554 Agent's Gender * 1 0.576 0.576 0.457 0.500 Subject's Gender Gender-Role Stereoty pe 1.402 1.402 1.113 0.293 (Covariate) , . , . Error 155 1.259 Table 4-16: Study One - A N C O V A Results by Gender (Competence Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Gender 1 <0.01 <0.01 0.000 0.996 Subject's Gender 1 0.100 0.100 0.171 0.679 Agent's Gender * Subject's Gender 1 4.634 4.634 7.954 0.005 Gender-Role Stereotype (Covariate) 1 0.593 0.503 1.018 0.315 Dispositional Trust -Competence (Covariate) 1 1.984 1.984 3.405 " 0.067 74 Chapter 4: Influences of a PRA's Demographic Embodiment Dispositional Trust -Trusting Stance (Covariate) 1 0.291 0.291 0.499 0.481 Trust Propensity -Computer (Covariate) 1 1.483 1.483 2.545 0.113 Error 152 0.583 Table 4-17: Study One - ANCOVA Results by Gender (Benevolence Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Gender 1 3.022 3.022 3.693 0.057 Subject's Gender 1 0.337 0.337 0.412 0.522 Agent's Gender * Subject's Gender 1 2.602 2.602 3.180 0.077 Gender-Role Stereotype (Covariate) 1 0.01-1 0.044 0.054 0.817 Dispositional Trust -Benevolence (Covariate) 1 4.643 4.643 5.675 0.018 Dispositional Trust -Trusting Stance (Covariate) 1 0.165 0.165 0.202 0.654 Trust Propensity -Computer (Covariate) 1 7.637 7.637 9.334 0.003 Error 152 0.818 Table 4-18: Study One - ANCOVA Results by Gender (Integrity Beliefs) Source DF Sum of Squares Mean Square F P-Value Agent's Gender 1 0.389 0.389 0.716 0.399 Subject's Gender 1 0.050 0.050 0.091 0.763 Agent's Gender * 1 2.374 2.374 4.372 0.038 Subject's Gender Gender-Role Stereotype 1 0.058 0.058 0.107 0.743 (Covariate) Dispositional Trust -Integrity (Covariate) 1 0.022 0.022 0.041 0.840 Dispositional Trust -Trusting Stance (Covariate) 1 0.726 0.726 1.338 0.249 Trust Propensity -Computer (Covariate) 1 3.618 3.618 6.664 0.011 Error 152 0.543 75 Chapter 4: Influences of a PRA's Demographic Embodiment Table 4-19: Study One - ANCOVA Results by Gender (Perceived Enjoyment) Source DF Sum of Squares Mean Square F P-Value Agent's Gender 1 3.416 3.416 2.892 0.091 Subject's Gender 1 1.956 1.956 1.656 0.200 Agent's Gender * 1 0.169 0.169 0.143 0.706 Subject's Gender Gender-Role Stereotype 1 u.898 0.8'>X 0.760 0.385 (Covariate) Error 155 1.181 Figure 4-8: Study One - Gender Interactions on Trust in Competence 76 Chapter 4: Influences of a PRA's Demographic Embodiment 5.90H (0 c 5.80-m a> 2 ft) c "5> <5 5.70-S T3 Q) 4-* re E w 5.60-LU 5.50H X Subject Gender Female Male / Female Male Agent Gender Figure 4-9: Study One - Gender Interactions on Trust in Integrity As shown in Figure 4-8 and 4-9, the ANCOVA results reveal significant gender-interaction effects on competence beliefs (F (1,152) = 7.954, p < 0.01) and integrity beliefs (F (1,152) = 4.372,/? < 0.05). However, gender-interaction is not significant on benevolence beliefs (F (1,152) = 3.180, p > 0.05). Also, no significant interaction was found regarding social presence (F (1,155) = 0.457,/? > 0.10) and perceived enjoyment (F (1,155) = 0.143, p > 0.10). Hypotheses H2b and H2d are thus supported, while H2a, H2c, and H2e are not supported. In other words, users generally exhibited stronger trusting beliefs in the competence and integrity of a same-gender agent compared to an opposite-sex one; however, gender match or mismatch had no effects on users' perceptions of social presence, benevolence beliefs, and enjoyment. Unexpectedly, there were significant differences in measurements of social presence according to the agent's 77 Chapter 4: Influences of a PRA's Demographic Embodiment gender (F (1,155) = 4.543,p < 0.05). Users perceived male agents as more sociable than female agents, even after their gender-role stereotype was controlled as a covariate. A summary of the hypotheses testing results is listed in Table 4-20. Table 4-20: Study One - A Summary of Hypothesis Testing Results Hypothesis Supported? H i a: Users perceive stronger social presence from interacting with an "ethnicity-matched" PRA than with an "ethnicity-mismatched" PRA. Yes Hlb: Users have stronger competence beliefs regarding an "ethnicity-matched" PRA compared to an "ethnicity-mismatched" PRA. Yes Hlc: Users have stronger benevolence beliefs regarding an "ethnicity-matched" PRA compared to an "ethnicity-mismatched" PRA. No H i d : Users have stronger integrity beliefs regarding an "ethnicity-matched" PRA compared to an "ethnicity-mismatched" PRA. No Hie: Users feel more enjoyable when interacting with an "ethnicity-matched" PRA than with an "ethnicity-mismatched" PRA. Yes H2a: Users will feel stronger social presence when interacting with a same-sex PRA than with an opposite-sex PRA. No H2b: Users will develop stronger beliefs in the competence of a same-sex PRA than in an opposite-sex PRA. Yes H2c: Users will develop stronger beliefs in the benevolence of a same-sex PRA than in an opposite-sex PRA. No H2d: Users will develop stronger beliefs in the integrity of a same-sex PRA than in an opposite-sex PRA. Yes H2e: Users will perceive interactions with a same-sex PRA as more enjoyable than with an opposite-sex PRA. No 4.4.4 Three-Way Interactions: Ethnicity x Gender Although the focus of this study involved separately examining the potential similarity-attraction effects in human-PRA interactions on dimensions of ethnicity and gender, the full factorial subject assignment scheme (equal number of males and females 78 Chapter 4: Influences of a PRA's Demographic Embodiment and Asians and Caucasians assigned to each of the four animated characters) makes it possible for us to further explore the potential high-level interactions, that is, the interactions between gender and ethnicity factors, especially those high-level interactions that may affect our interpretations on the results of gender- or ethnicity- interactions. ANCOVA tests were conducted with both ethnicity and gender factors on all dependent variables. No four-way interaction was found. Three-way interactions were only found in relation to social presence and perceived enjoyment. The ANCOVA results on social presence are shown in Table 4-21. Table 4-21: Study One - ANCOVA Results by Ethnicity and Gender (Social Presence) Source DF Sum of Squares Mean Square F P-Value AG1 1 5.675 5.675 5.023 0.027 SG 1 0.459 0.459 0.406 0.525 AE 1 0.001 0.001 0.000 0.983 SE 1 0.007 0.007 0.006 0.938 AG * SG 1 0.576 0.576 0.510 0.476 AG* AE 1 0.487 0.487 0.431 0.513 SG* AE 1 0.593 0.593 0.525 0.470 AG * SG * AE 1 8.039 8.039 7.114 0.009 AG * SE 1 0.722 0.722 0.639 0.425 SG* SE 1 0.834 0.834 0.738 0.392 AG * SG * SE 1 0.098 0.098 0.087 0.769 AE* SE 1 15.175 15.175 13.429 < 0.001 AG * AE * SE 1 3.028 3.028 2.680 0.104 SG * AE * SE 1 4.534 4.534 4.013 0.047 AG * SG * AE * SE 1 0.035 0.035 0.031 0.860 Gender-Role Stereotype T 0.445 0.445 0.394 0.531 (Covariate) Error 143 1.130 1: A G = Agent's Gender; SG = Subject's Gender; AE = Agent's Ethnicity; SE = Subject's Ethnicity For social presence, significant three-way interactions were found on both Agent's Gender x Subject's Gender * Agent's Ethnicity (F (1,143) = 7.114, p < 0.01) and Agent's Ethnicity x Subject's Ethnicity x Subject's Gender (F (1,143) = 4.013,/? < 0.05). There is a three-way interaction whenever a two-way interaction differs depending on the 79 Chapter 4: Influences of a PRA's Demographic Embodiment level of a third variable. For social presence, the ethnicity interaction is significant while the gender interaction is not. Therefore, we examined the potential interactions between (Agent's Ethnicity x Subject's Ethnicity) and Subject's Gender to see whether the hypothesized ethnicity interaction may have been moderated by subject's gender. The group means of social presence were decomposed accordingly as shown in Table 4-22. Table 4-22: Study One - Ethnicity Interaction by Subject's Gender (Social Presence) Male Participant Female Participant Participant Participant Asian Caucasian Asian Caucasian Agent Asian 4.32 4.25 Agent Asian 5.04 3.97 Caucasian 4.18 4.62 Caucasian 3.97 4.83 Results in Table 4-23 suggested that similarity-attraction on ethnicity is significant among female subjects (F (1, 76) = 19.606, p < 0.001) but not significant among males (F (1, 76) = 0.904,/? > 0.10). The ANCOVA results on perceived enjoyment are shown in Table 4-23. Table 4-23: Study One - ANCOVA Results by Ethnicity and Gender (Perceived Enjoyment) Source DF Sum of Squares Mean Square F P-Value AG1 1 3.383 3.383 3.314 0.071 SG 1 1.997 1.997 1.957 0.164 AE 1 0.010 0.010 0.010 0.922 SE 1 0.107 0.107 0.105 0.746 AG* SG 1 0.169 0.169 0.166 0.685 AG* AE 1 0.109 0.109 0.107 0.744 SG* AE 1 0.005 0.005 0.005 0.944 AG * SG * AE 1 8.763 8.763 8.584 0.004 AG* SE 1 0.208 0.208 0.203 0.653 SG*SE 1 0.048 0.048 0.047 0.828 AG * SG * SE 1 0.972 0.972 0.952 0.331 AE* SE 1 12.206 12.206 11.956 0.001 AG* AE* SE 1 4.287 4.287 4.199 0.042 SG * AE * SE 1 9.703 9.703 9.505 0.002 AG * SG * AE * SE 1 0.650 0.650 0.637 0.426 Gender-Role Stereotype 1 0.114 0.114 0.111 0.739 80 Chapter 4: Influences of a PRA's Demographic Embodiment (Covariate) Error 143 1.021 1: A G = Agent's Gender; SG = Subject's Gender; AE = Agent's Ethnicity; SE = Subject's Ethnicity Significant three-way interactions are found on Agent's Gender x Subject's Gender x Agent's Ethnicity (F (1, 143) = 8.584,/? < 0.01), Agent's Gender x Agent's Ethnicity x Subject's Ethnicity (F (1, 143) = 9.505, p < 0.01), and Subject's Gender x Agent's Ethnicity x Subject's Ethnicity (F (1, 143) = 4.199, p < 0.05). Similarly, further analyses were only conducted for the potential interaction between (Agent's Ethnicity x Subject's Ethnicity) and Agent's Gender, as well as that between (Agent's Ethnicity x Subject's Ethnicity) and Subject's Gender. The group means of perceived enjoyment were decomposed accordingly as shown in Table 4-24 and Table 4-25. Table 4-24: Study One - Ethnicity Interaction by Subject's Gender (Perceived Enjoyment) Male Participant Femai 'e Participant Participant Participant Agent Asian Caucasian Agent Asian Caucasian Asian 4.52 4.47 "Asian\ 5.29 4.16 Caucasian 4.50 4.54 Caucasian^ 4.25 5.23 Results in Table 4-24 suggested that similarity-attraction on ethnicity is significant among female subjects (F (1, 76) = 18.418,/? < 0.001) but not significant among males (F (1, 76) = 0.042,/? > 0.10). Table 4-25: Study One - Ethnicity Interaction by Agent's Gender (Perceived Enjoyment) Male Agent Female Agent Participant I I'ar ticipant Agent Asian Caucasian Agent Asian Caucasian Asian 4.90 4.56 Asian 4.91 4.07 Caucasian 4.75 4.85 Caucasian 4.00 4.92 81 Chapter 4: Influences of a PRA's Demographic Embodiment Results in Table 4-25 suggested that similarity-attraction on ethnicity is significant when subjects were interacting with a female agent (F (1, 76) = 14.523,/? < 0.001) but not significant when interacting with a male agent (F (1, 76) = 0.834,/? > 0.10). 4.5 Discussion The purpose of this study was to examine the potential influences of different demographic embodiments of an anthropomorphic PRA on users' perceptions. The outcomes of this study are relevant for anthropomorphic recommendation agent design in a multicultural context, given the vast demographic diversity of Internet shoppers and increasing number of online vendors that have multinational presence and serve customers of a great variety of ethnic and cultural backgrounds. Although recent studies (e.g., Nass et al. 2000; Baylor 2005; Cowell and Stanney 2005) have already acknowledged the importance of demographic factors in the design of embodied conversational agents, to the best of our knowledge, this study is the first one that has used a controlled laboratory experiment to rigorously investigate the impacts of both gender and ethnicity in human-agent interface design. In addition, the results of this study were used to justify the agent selection in Study Two (discussed in Chapter 5) in which the embodiment and modality effects of a PRA's multimedia interface are examined separately. As proposed by the Computers as Social Actors (CASA) paradigm, when online consumers treat recommendation agents as social actors, they will interact with and evaluate the agents in a similar way as buyers treat salespeople in conventional retail environments. Similar-attraction theories and social identity theories were employed to predict the potential ethnicity- and gender-interactions. Overall, the empirical results 82 Chapter 4: Influences of a PRA's Demographic Embodiment suggest that similarity-attraction and "in-group" favoritism indeed influence users' evaluations of an embodied PRA. Such similarity-attraction effects are particularly significant regarding the selection of an agent's ethnic identity. Our results revealed that when interacting with an agent of the same ethnicity, users not only perceive interactions with the agent to be more sociable and enjoyable, they also tend to give more positive evaluations on the agent's competence. It is noteworthy that the results on users' benevolence beliefs and integrity beliefs are not significant, which is probably a consequence of the agency relationship in this specific context of online shopping. A PRA of same-ethnicity may not convey strong enough cues to effectively alleviate users' concerns on the possible opportunist activities taken by the agent. In contrast, attractions due to gender similarity are not as salient as those for ethnicity. As discussed in Section 4.2.2, such attractions could be easily subordinated to other factors such as users' gender-role stereotypes and their attitudes towards sex. This probably explains why we did not find a significant interaction effect on more relationship and affect oriented outcomes such as social presence and perceived enjoyment. Nevertheless, for those more cognitive oriented outcomes, such as trusting beliefs in the agent's competence and integrity, gender is still a significant "in-group" cue. With a large sample size and completely-balanced factorial design, we could, for the first time, explore the potential gender-ethnicity interactions by analyzing some significant three-way interactions9. Our results suggest that social and affective responses arising from ethnicity similarities are in fact more significant in females than in males. In 9 We acknowledge that it is risky to interpret a three-way interaction arbitrarily without a theoretically-sound hypothesis proposed beforehand; therefore, the results were analyzed and presented in an exploratory manner. 83 Chapter 4: Influences of a PRA's Demographic Embodiment other words, female users tend to perceive a same-ethnicity agent to be more sociable and their interactions more enjoyable than a different-ethnicity agent, but this pattern is not as evident among male users. Although the interpretations of three-way interactions are usually challenging, previous research in cross-cultural studies may lend some support to this finding. The small amount of research that has addressed gender and ethnicity interactions suggests a greater involvement in ethnic traditions by women than by men. For example, Ting-toomey (1981) found that women are more oriented to their ancestral culture than men are in a study of Chinese-American college students. Ullah (1985) found that, among Irish adolescents in England, girls are significantly more likely than boys to adopt an Irish identity. Masuda et al. (1973) also showed that Japanese girls and women tend to score higher than boys and men on Japanese ethnic identity. Our results suggest that this pattern also holds true in human-agent interactions and it is likely that female users, being more aware of their own ethnic identity as well as that of the agent, are more prone to use ethnicity as cues of "in-group" connections. However, it remains an open question why such gender difference is only significant in affective outcomes such as social presence and perceived enjoyment, while it is insignificant in relation to cognitive-oriented variables such as trusting beliefs. Further investigations should be conducted before a solid conclusion can be drawn. 4.6 Chapter Summary This chapter presents the first empirical study of this dissertation. Based on similarity-attraction hypothesis and social identities theory, we proposed and empirically tested the role a humanoid PRA's demographic embodiments can play in shaping users' perceptions towards the agent. In the next chapter, we will investigate the separate and 84 Chapter 4: Influences of a PRA's Demographic Embodiment joint impacts of two PRA interface components, namely humanoid embodiments and output modalities, on users' perceptions of social presence, trusting beliefs, and perceived enjoyment. 85 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities CHAPTER 5: INFLUENCES OF PRA'S HUMANOID EMBODIMENT AND OUTPUT MODALITIES 5.1 Introduction The results of the previous study have supported most of the hypotheses concerning how people respond socially towards a humanoid PRA, and that certain social-interaction stereotypes are clearly relevant to human-PRA interactions, such as similarity attractions and in-group favoritism. However, it remains unclear whether or not anthropomorphic interfaces are worth the efforts of online vendors to establish and maintain them, as well as whether the animated avatar or the speech output of the PRA interface is relatively more important and should receive higher priority in implementation. In the previous study, all PRAs are equipped with both animated human-like face and human speech. Is it possible that a disembodied PRA may be perceived the same as or even better than the one with a "matched" face? Or the social interactions will be just as effective when human voices are replaced by on-screen text or computer-synthesized voices? To answer these questions, we conduct another study to investigate the distinct impacts of two interface components: humanoid embodiments and output modalities. As summarized in Section 2.3, most prior studies on anthropomorphic interface/agents were conducted in the context of either entertainment or computer-aided tutoring. Research on their impacts in the environment of online shopping is still quite sparse. This chapter presents the second part of this dissertation, which examines the unique and conjoint influences of both humanoid embodiments and output modalities of a PRA on users' social-psychological perceptions, measured by social presence, trusting beliefs, and perceived enjoyment. 86 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities The remainder of this chapter is organized as follows. Section 5.2 develops hypotheses for the proposed model. The research method and results of hypothesis testing are reported in section 5.3, and the chapter concludes with a discussion of the results. PRA Interface Embodiment (Animated Face) Output Modalities (Voice) Figure 5-1: Study Two - Research Model 5.2 Hypotheses Development The research model for Study Two is presented in Figure 5-1. 5.2.1 Impacts on Social Presence Social presence has been identified by Biocca (1997) as a user's access to any form of intelligence and sensory experience of another intelligent being. Prior empirical research has suggested that the concept of social presence is as applicable to people's interactions with any medium, machine, or software as in their interpersonal relations (Nass and Gong 2000; Nass and Moon 2000; Biocca et al. 2003; Lee and Nass 2003; Nowak and Biocca Social Presence Trusting Beliefs Perceived Enjoyment 87 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities 2003) . Social presence is about the extent to which a PRA can convey to its user a sense of human warmth and sociability. Media richness theories have proposed that there is consistency in ratings of diverse media in terms of their capabilities to convey social presence (Rice 1992; Walther 1992). As the bandwidth of media narrows, the communication is generally perceived as being less friendly, emotional, personal, and more serious, official, impersonal, and task-oriented, while media high in bandwidth represent the opposite directions. When an agent is embodied with an animated character or can "speak" with human voice, the perceived communication bandwidth, even though it develops one-way from the agent to the user, is significantly broader, which may contribute to users' perception of social presence. Reciprocity has been posited as another important source of social presence (Lee 2004) . When interacting with human-like virtual social actors, people reciprocate first before thinking about the ontological nature of the virtual actors. "Social agency theory" proposed by Mayer et al. (2003) has argued that social cues in a multimedia message can prime the social conversation schema in receivers. Once the social conversation schema is activated, learners are more likely to act as if they are in a conversation with another person. When a PRA is embodied with an animated face or speech capability, these obvious human characteristics convey apparent cues of interpersonal communication, which is commonly experienced in one's social life and implies that the user should apply social norms reciprocally. In the meantime, those anthropomorphic features impede users from carefully elaborating the ontological nature of the virtual agent; instead, an almost-automatic response is elicited to apply various social rules acquired from everyday understandings of complicated social life and human behavior to judge, evaluate, and 88 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities respond to the personality cues presented by the audio and the animated character. Therefore, we expect that when users are interacting with an embodied PRA, they will treat the interaction more as a conversation rather than just a case of information delivery (Mayer 2001). More specifically, a PRA with humanoid embodiment can engage in interactions that are more similar to face-to-face conversations, as users are provided with more visual signals from the agent. In particular, the animated face may exhibit a number of nonverbal cues, such as eye contact, gazing, and expressive facial expressions like lip movements and smiles, all of which contribute to feelings of social interactions. Therefore, we propose that: H3a: Users will perceive greater social presence interacting with a PRA with a humanoid embodiment than a PRA without a humanoid embodiment. The use of voice is another potent social cue and has been shown to elicit perceptions that a single computer can support multiple distinct entities (Nass and Steuer 1993), and to evoke gender stereotypes (Nass et al. 1997). Speech suggests social presence due to human evolutionary psychological traits: before the invention of telephone, humans use speech only when they interact with human who is physically close (Clark 1996; Nass and Gong 2000). According to communications research, voice is powerful indicator of "presence of another human being", and its incorporation in an interaction might promote richer processing by the incorporation of the additional attitudes and beliefs that are attached to the agent. Other than voice from human beings, computer-synthesized (text-to-speech) voice is being increasingly used in the interface of various technological artifacts, such as 89 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities computers and electronic devices, especially when there is a need for a quick and easy transformation of time-sensitive textual information (e.g., news, traffic condition, weather) into audible information. If a text-to-speech (TTS) voice sounds exactly like a human voice, listeners would have no way to make a distinction, and therefore, they will respond precisely as if it was produced by a human individual. However, because complete syntactic and semantic understanding is extremely difficult, TTS systems inevitably speak in a manner that is "nonhuman" unless the text is marked-up for all important characteristics by an actual person (Nass and Scott 2005). Although TTS voice is still far from satisfactory in naturalness compared with a real-human voice, which restricts the social cues that TTS can convey (Mayer et al. 2003), the limited prosodic features, such as intonation and phrasing, may be more discernable when units of speech are selected from a large speech inventory that best matches the prosody predicted by the TTS system (Jilka et al. 2003). People are primed to distinctive social cues, such as gender, personality, and emotional status, embedded in human speech to the extent that they unconsciously attribute some of these cues to computer-generated speech. For example, empirical evidence indicates that people exhibit similarity-attraction and consistency-attraction toward computer-generated speech (Nass and Lee 2001). In this present study, we expect that PRAs with speech output will elicit stronger social responses than PRAs with text-only interfaces, but the effects of human speech would be stronger than computer-generated speech. H3b: Users will perceive greater social presence when interacting with a PRA with human speech output compared to the same PRA with textual output. H3c: Users will perceive greater social presence when interacting with a PRA with human speech output compared to the same PRA with TTS output. 90 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities H3d: Users will perceive greater social presence when interacting with a PRA with TTS output compared to the same PRA with textual output. 5.2.2 Impacts on Trusting Beliefs As mentioned in Chapter 3, trusting beliefs investigated in this research refer to the initial trust in the PRA. Online shoppers will have their initial trust formed after their first experience with a PRA. Initial trust is different from "experienced trust" (Tseng and Fogg 1999) in that it is built before an online shopper engage in trust-related behaviors, such as following the agent's advice and purchasing the recommended product, and have the opportunity to assess the trustworthiness of the agent by observing the consequences of those behaviors (McKnight 2002). When consumers form their initial trusting beliefs in a PRA, the perceived quality of the information provided by the agent is an important factor. Users make inferences on the agent's competence and intentions by reflecting on issues like whether or not the questions raised by the agent are relevant; whether or not the agent provides adequate information to help users identify their preferences; or how well the recommended products conform to the preference structure they have specified. For example, Wang and Benbasat (2004) have found that users' trusting beliefs in an agent's competence, benevolence, and integrity could be effectively enhanced when the agent provides additional information to explain their underlying reasoning process and cognitively justify their recommendations. In addition to the product-related information provided by the agent, other sources are also used by consumers to form their initial trust, especially by consumers who have limited product knowledge and therefore cannot accurately appraise the completeness and 91 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities integrity of the information provided by the agent. As explained in Section 2.2, if the agent is treated as a social actor, consumers will predictably employ certain means that are applied in interpersonal communications to help them make judgments. When people interact with others for the first time, the initial phase of trust building is based on 'whatever information is available' (Meyerson et al. 1996 pp. 170). The information available often comes in the form of small signals or cues, such as the physical appearance (Riker 1971) or trustee's voice (Baldwin 1992). For example, individuals wearing glasses are rated as more intelligent and industrious than those not wearing glasses (Thorton 1944); physically attractive person is more often placed into the "good" person category than the one who was not attractive (also known as the 'what is beautiful is good' stereotype) (Dion et al. 1972). A similar pattern has been found in human-agent interactions as well. For example, when an embodied agent exhibits social and interpersonal cues that are widely used in building up interpersonal trust, it was perceived as more benevolent and credible (Cassell and Bickmore 2000). When the PRA is embodied with anthropomorphic interface, the three following factors could influence users' competence belief in the agents. First, it could facilitate users to attribute some human intelligence to the agent. As demonstrated by King and Ohya (1996) and Koda and Maes (1996), humanoid visualization of an interface is generally perceived as having higher intelligence. In the shopping context, as suggested in (Price and Feick 1984), people are more willing to get shopping advice from other human beings, be them friends or salespeople, as they expect that a personal source could understand their needs better and give more suitable suggestions than an impersonal source, such as Consumer Reports. Likewise, when an agent is perceived more like a 92 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities human advisor rather than a piece of preprogrammed software, PRA users will have higher levels of trusting beliefs in the agent's competence. Second, when the agent looks like a human being, people will apply various stereotypes of physical appearance to evaluate the competence of the agent. For example, an agent with professional attire could convey additional signals of competence, as several studies have indicated that professional dress tends to create heightened perceptions of speaker competence, credibility, and professionalism (Bassett 1979; O'Neal and Lapitsky 1991). Third, the humanoid interface could also be interpreted as a signal that the agent is designed with more advanced technology (compared with a text-only interface) and therefore, more technically competent. Such inferences among online consumers have been found in prior studies. For instance, in a study of consumers' initial trust in a web vendor, (McKnight 2002) indicate that perceived website quality is an antecedent of trusting beliefs because a well-designed website could imply that the company is solid and well-managed. It has also been noted that a good interface design could enhance the formation of consumer trust in online banking (Kim and Moon 1998). On the other hand, very limited research had been conducted comparing the perceived competence of output modalities. One exception is (Gong et al. 2001), which demonstrates that a voice system with consistent synthesized speech is superior to the one that mixes synthesized speech with human speech with respect to liking, perceived competence, and credibility. Thus, no hypothesis on output modalities is proposed. In summary, we hypothesize that: 93 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities H4: Users will have stronger trusting beliefs in the agent's competence when interacting with a PRA with a humanoid embodiment than a similar PRA without a humanoid embodiment. Cassell et al. (1999) and Cassell and Bickmore (2000) have argued that when people interact with an embodied interface agent, the use of social cues such as small talk can enhance the users' trusting beliefs regarding an agent's benevolence and credibility. The underlying explanation is that people's trust in the agent can be established using the same myriad social cues people use in face-to-face conversation, such as looking people in the eye, shaking their hands, and watching them make presentations. Studies on interpersonal trust have also found that users' trusting beliefs toward each other can be influenced by the communication medium they use. Face-to-face meetings promote greater trust and potential cooperative behavior more effectively than telephone conversations, which in turn are more effective than email and fax correspondences (Jensen et al. 2000). When the medium moves from text to audio to audio/visual modalities, enhanced perceived interactivity elicits corresponding increases in trust and credibility (Burgoon et al. 2002; Burgoon et al. 2003). In their study of pedagogical agents, Baylor and Ryu (2003) have found that the presence of an agent image led to the agent being perceived as significantly more credible than when it was absent. The authors proposed that, when the information is provided by an embodied agent, subjects might infer that the agent had more personal ownership of the information and therefore more accountable for the credibility of the information. We expect that personifying agents can make human-computer interaction more similar to human-human interaction, in which the trusting beliefs regarding an agent can 94 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities also be strengthened as the communication media is getting richer. As discussed in the previous sections, agent with animated face could carry more non-verbal cues than one without; agent with human speech output could convey more social cues than TTS voice, which in turn would be more capable than textual output. Therefore, we propose: H5a: Users will have stronger trusting beliefs in the agent's benevolence when interacting with a PRA with a humanoid embodiment than a similar PRA without a humanoid embodiment. H5b: Users will have stronger trusting beliefs in the agent's benevolence when interacting with a PRA with human speech output than a similar PRA with textual output. H5c: Users will have stronger trusting beliefs in the agent's benevolence when interacting with a PRA with human speech output than a similar PRA with TTS output. H5d: Users will have stronger trusting beliefs in the agent's benevolence when interacting with a PRA with TTS output than a similar PRA with textual output. H6a: Users will have stronger trusting beliefs in the agent's integrity when interacting with a PRA with a humanoid embodiment than a similar PRA without a humanoid embodiment. H6b: Users will have stronger trusting beliefs in the agent's integrity when interacting with a PRA with human speech output than a similar PRA with textual output. H6c: Users will have stronger trusting beliefs in the agent's integrity when interacting with a PRA with human speech output than a similar PRA with TTS output. H6d: Users will have stronger trusting beliefs in the agent's integrity when interacting with a PRA with TTS output than a similar PRA with textual output. 5.2.3 Impacts on Perceived Enjoyment Creating fun and enjoyment within a software application has become a desirable objective in software engineering not only for computer games but for business applications as well (Carroll 2004). However, how software designers can generate the 95 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities feeling of enjoyment is a question that technology alone cannot answer. As mentioned by Carroll (2004, p. 39), "Things are fun when they attract, capture, and hold our attention by provoking new or unusual perceptions, arousing emotions in contexts that typically arouse none, or arousing emotions not typically aroused in a given context". However, "...a design does not evoke fun merely because it incorporates color and animation, sound and music, or graphical fantasy content". In other words, whether or not an interface can create an enjoyable experience for its users is influenced by both the technology-based factors as well as its appropriateness for a specific context. Researchers have discovered that the presence of an anthropomorphic agent can significantly increase the entertainment value of an interface (Dehn and van Mulken 2000), but such effects seem to be domain-specific. One such application is in the entertainment domain. For example, in a study conducted by Koda and Maes (1996), a computer opponent in a poker game visualized by a face was rated as more entertaining than a non-anthropomorphic counterpart. Such effect could be largely explained by the "fantasy" factor proposed by Malone (1982) in his summary of the design heuristics for enjoyable computer game interfaces. A system with fantasy elements can evoke mental images of physical objects or social situations that are not actually present. For example, computer poker opponent with a simulated face is more likely to be imagined by game players as a human opponent. By using metaphors of physical objects or situations with which the users are already familiar, designers could create fantasies to satisfy users' emotional needs and make the experience more enjoyable. Another application domain that has seen the entertainment value of an anthropomorphic agent is multimedia tutoring system. Quite a few empirical findings suggest that the presence of humanoid agent could 96 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities effectively enhance students' motivation and interests in the learning process and make the learning process perceived as more enjoyable and less effortful. For example, the study by Lester and Stone (1997) with different versions of a software tutor revealed that the presence of a lifelike character in a learning environment can have strong positive effect on student's perception of the agent's utilitarian and entertainment value. Van Mulken et al. (1998) performed an empirical study to examine the "persona effects" both on subjective and objective measures. The results of their study indicate that the presence of the agent has neither a positive nor a negative effect on comprehension and recall performance. However, even the mere presence of the anthropomorphic agent causes presentations and tests to be experienced by users as more entertaining and less difficult. More interestingly, Moundridou and Virvou (2002) have noted that the entertainment value is more significant when the agent serves to explain technical information than non-technical information, which implies that the benefits of entertainment is more likely due to the additional cognitive effort that is motivated by the interface. In addition, researchers have obtained consistent results regarding the effects of attention attraction. For example, Takeuchi and Naito (1995) have observed that a facial display evokes more eye contact than an arrow, which implies that the face attracts more attention. Van Mulken et al. (1998) have also reported a significantly higher rating of perceived concentration in anthropomorphic agent condition than in no agent condition. In information search contexts, heightened engagement or involvement may lead to a stronger perception of play, a sensation related to enjoyment and escape (Mathwick and Rigdon 2004). Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Based on the above discussions, we deem that interacting with an agent to complete a shopping task is a cognitive activity as it requires shoppers to invest significant amount of mental effort in order to process the information provided by the agent. The presence of an animated character could make the interaction intrinsically interesting, as well as enhance users' motivation and involvement. However, these cognitive efforts are spent in the context of entertainment due to the hedonic nature of shopping activities (as discussed in Section 3.4). An animated character could provide shoppers with an additional feeling of fun. Therefore, we hypothesize that: H7a: Users will perceive greater enjoyment when interacting with a PRA with a humanoid embodiment than a similar PRA without a humanoid embodiment. Voice carries a great deal of information beyond the nominal messages in the words of a narration. Voice has long been found to be able to generate physiological and affective arousal (Scherer 1986; Scherer 2003). The use of human speech can significantly strengthen the "interpersonal conversation" nature of an interaction and enhance users' perceptions of human-to-human conversation. Research has shown that learners reported higher levels of motivation and interest when interacting with an animated agent with voice output than an agent with textual output (Moreno et al. 2000). Nonetheless, it should be noted that the valence of perceived voice quality also exerts significant influence on users' affective responses (Gobi and Chasaide 2003). As suggested by McBreen and Jack (2001), for human speech output to be used in the agent interface, the voice should sound "friendly" and conversational with intonation. A professional voice talent should be employed to instill these cues into his or her speech. Compared with human speech, computer-generated voices tend to have unnatural pauses, accents, and word emphases, as well as discontinuities between phonemes and 98 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities syllables (Nass and Lee 2001). It has been demonstrated that users exhibit significantly less self-disclosure and less comfort with a system that uses synthetic speech than one that uses text-based or recorded-speech (Nass et al. 2003). Such results could be explained by the difficulties involved in projecting social cues unique to real-human speech onto any existing TTS platforms, given the monotonous pitch and intonation of computer-generated voices. Also, listeners of TTS may have to spend extra effort in order to understand the computer-generated voice, which could further restrain the feelings of enjoyment. Therefore, we expect that, even though PRAs with TTS voice might provoke higher levels of arousal than on-screen text, they will induce significantly weaker perceptions of enjoyment than human voices. Thus, we posit: H7b: Users will perceive greater enjoyment when interacting with a PRA with human speech output than a similar PRA with textual output. H7c: Users will perceive greater enjoyment when interacting with a PRA with human speech output than a similar PRA with TTS output. 5.3 Research Method A laboratory experiment was conducted to test the hypotheses listed in Section 5.2. Section 5.3.1 describes the experimental design. The design of agent interface in each experiment group is introduced in Section 5.3.2. Dependent variables and control variables are listed in section 5.3.3. Section 5.3.4 describes the participants, incentives, and experiment procedures. 5.3.1 Experimental Design The two independent variables are 1) the presence or absence of a humanoid embodiment (in the form of an animated character) and 2) the output modality of the 99 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities agent. A three (output modality: text, TTS voice, and human voice) by two (humanoid embodiment: with and without) full factorial design was employed (as shown in Table 5-1)-Table 5-1: Study Two - 3*2 Full Factorial Experiment Design Output Modality Text Only TTS Voice Human Speech Embodiment Without Avatar Group 1 Group 2 Group 3 With Avatar Group 4 Group 5 Group 6 5.3.2 The Agent The simulated recommendation agent used in this study is designed to provide purchasing advice on digital cameras and it shares the same algorithm design as the agent used in Study One. To elicit users' preferences and needs, an agent-user dialogue was used to simulate dialogues presented in other studies (Russo 2002) and in commercial applications. Users were required to answer all ten product-related questions and to indicate the importance level of each attribute in their decisions (the interaction scripts are included in Appendix B). Based on users' individual preference, a "fit score" of each digital camera in the database was then calculated and only those with a fit score above 70% were included in the recommendation list. The agent interface was designed with Oddcast Sitepal, a Flash-based animation design tool that integrates both animated characters and speech output with textual information. Four animated characters (Asian Female, Asian Male, Caucasian Female, and Caucasian Male), the same ones designed for Study One, were used in each of the three with-avatar conditions. Under the three without-avatar conditions, the agents were 100 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities also designed with apparent gender cues: users were able to discern gender of the agent voice in TTS voice condition and human voice condition, and by the agent's name (Max for the male agent and Maria for the female agent) in the text-only condition. A manipulation check after the experiment revealed that all of the participants had correctly identified the agent's gender (and ethnicity in the with-avatar conditions). Given that the purpose of this study is to examine the effects of the PRA's humanoid embodiment and output modality assuming users will interact with the best possible implementation of a PRA, experiment results from the first study were used to decide which agent (in regard to its "gender" and "ethnicity") should be assigned to individual participants. As explained in Section 4.4.3, results of Study One suggested that "ethnicity-match" PRAs were generally better perceived than mismatched ones. Users interacting with an ethnicity-matched PRA reported significantly higher social presence, competence belief, and perceived enjoyment than those with a mismatched PRA. Users of ethnicity-matched PRAs also scored higher on benevolence and integrity beliefs even though the differences are not statistically significant. In other words, an ethnicity-matched PRA would generally benefit rather than impair the perceived quality of human-PRA social interactions. As for the PRA's gender, Study One results revealed that users have stronger trusting beliefs in the agent's competence and integrity when they interacted with a PRA of same gender. There is no significant main or interaction effect on perceived enjoyment. A preliminary analysis on users agent adoption intentions (data collected for the PLS 101 Chapter 5: Influences o f P R A ' s Humanoid Embodiment and Output Modalities analysis described in Chapter 6) also indicated significant same-gender favoritism (F (1,155) = 6.321,/? < 0.02). Based on the above analysis, we decided to assign a P R A with matched-gender (and with matched-ethnicity in with-avatar conditions) to each participant. • Condition 1: N o Avatar + Text-Only The screen layouts of the Question page and result page of Condition 1 are presented in Figure 5-2 and Figure 5-3. Each user was asked to select an answer for each of the ten questions and to indicate the importance level o f this attribute in her decision. I f the user needed additional information for this attribute or feature, she could click the "About This Question" button beside the agent to obtain additional explanatory information on why this question is important for decision and how each answer w i l l influence the agent's recommendation. This information was displayed in the text-box on the upper-left corner o f the screen. In order to make the human-PRA interaction feel more like a conversation, whenever a user selected an answer, the agent would provide some feedback comments, such as acknowledging user's input or providing additional decisional guidance. These comments were also presented through the textbox. After the user completed all ten questions, the agent provided a list o f recommendations, as shown in Figure 5-3. The user could return to the question pages at any time to revise her answers; an updated recommendation list would be provided accordingly. 102 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities MyDCAdvisor X a m asluifcj tbis qucst iou to mi.lei Maud y o m price i iiniit Pl lvcs of most C J t U K l A . t i l l between $200 a i d $<>00. L o w e r cost earner*; um m.iljy do no I have m a n y advouced ifeanires; mi the ether h a n d , for o c p i u i c i s « i occasional user*, a simple poml-ani l -shoot m o d e l could do a p u t t y good j o b on most of tasions R c i u c m b e i . I m a y r e c o m m e n d tamei.i< tbat are d o s e to y o u r About This Question Question 1 How much are you planning to spend on this digital camera? A small amount (under $300) A medium amount (between $250 and $650) A large amount (over $600) No preference Importance of this Feature I 2 3 1 S 6 7 8 9 not important very important Next Question p p Question: 1 2 3 4 s 6 ? 8 9 10 Figure 5-2: Study Two - Screenshot of the Question Page (Condition 1) MyDCAdvisor m Ttx». W&ini Cftjmwa N o w . I K I e t mties the del l? I iCvOltttlKIid t i l * tuUv>VVHii! niotkls'. w l m l i are m u s i c o i i m t e n t w i th y o u r i*ichitact* Y o u t o u M d t e v k detaUeif spcc&cat iotts b y <. I K Luis tlie UM^t oft c:unei:i or Uic m o d e l uau ic P lease note, i f tlk« fit s « « * is b e l o w 100. it tmia l iy I I W M W thai there cou ld be some d i i i e i c i u e * between y o u r i cfpai eiucut^ stud the models in our inventory. A f i the m o d e l s I I have 20 recommendat ion* tor you Recommendat ion N o . 1 (Fit Score 79%) Fuji FtowPlx F « 1 0 Bt MHl ftto* Zoom N0 HtfcM eiopttui Recommendat ion H o . 2 (FM Seore T*%) tasiot-xi lrniEX-PftOO • tMMl f tk* Zuotn Recommendat ion No. 3 (Fit S t o r e 71%) C a « i o E * W m EX-P700 Prtee ( ( H M zeow « t « p U t 4 A b o u t M y Recornrrwxiaoont j Recommendat ion No, 4 (FK S c o r e 7*%) C anon PowerShot S70 0>M»e Pi t t* Zeem frKeM a * Recommendat ion No. S (Fit S c o r e 79%) Sony Cyber- thot D S C - V 3 2eeen •tte* AKt(etn«Quieeen( '3 Figure 5-3: Study Two - Screenshot of the Result Page (Condition l ) Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities • Conditions 2 and 3: No avatar + TTS Voice / Human Voice The screen layout of the Question page of Conditions 2 and 3 is illustrated in Figure 5-4. The only difference between the interfaces of Conditions 2 and 3 and Condition 1 is that users heard explanations and comments through computer speakers. The speech was either generated by a TTS synthesizer or pre-recorded by human narrators. The TTS engine was AT & T Natural Voices, and the two voices that had been chosen for corresponding male and female agents are voice "Mike" and "Crystal"1. Two versions of human speech were recorded by two native English speakers. MyOCAdvisor tfr» OtyHai Camera E»fNH*t About This Question Question 1 How much are you planning to spend on this digital camera? , A small amount {under $300) A medium amount (between $250 and $650) A large amount (over $600) No preference importance of this Feature 1 2 3 4 5 6 7 8 9 not important very important Next Question j Question: 1 2 3 4 5 6 7 8 9 10 Figure 5-4: Study Two - Screenshot of the Question Page (Conditions 2 and 3) 1 These are the names of AT&T Natural Voices fonts, which are used to differentiate TTS of various characteristics (for example, language and gender). 104 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities • Condition 4: With Avatar + Text-Only The screen layout of the Question page of Condition 4 (with an Asian Female agent) is shown in Figure 5-5. The animated character was placed on the lower-left corner of the screen. Users could read explanatory information and feedback comments from the agent through the textbox on the upper-left corner. M y D C A d v i s o r I a m asking this ques t ion to understand y o u r pr ice range . Pr ices o f m o s t c a m e r a * l i d between $200 a n d $900 . L o w e r cos t c a m e r a s n o r m a l l y do not h a v e m a n y a d v a n c e d features; o n the o ther h a n d , f o r befrjnners o r o c c a s i o n a l users , a s imple potnt -and-shoot m o d e l could d o a pret ty g o o d j o b o n m o s t occas ions . R e m e m b e r , I m a y r e c o m m e n d cameras that arc c lose to y o u r Question 1 How much are you planning to spend on this digital camera? A small amount (under $300) A medium amount (between $250 and $650) A large amount (over $800) No preference Importance ot [His Feature 5 rot important very important About This Question Next Question | Question: 1 2 3 4 5 6 7 8 9 10 Figure 5-5: Study Two - Screenshot of the Question Page (Condition 4) • Conditions 5 and 6: With Avatar + TTS Voice / Human Voice The screen layout of the Question page of Conditions 5 and 6 (with an Asian Male agent) is shown in Figure 5-6. The animated character was placed on the lower-left corner of the screen. The agent used either TTS voice (in condition 5) or pre-recorded human speech (in condition 6) for explanatory information and feedback comments. The 105 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities speech was played with automatic lip-synchronization, as well as with facial expressions and head movements. MyDCAdvisor ^ i Question 1 How much are you planning to spend on this digital camera? , A small amount (under $300)  medium amount (b tween $250 and $650) A large amount (over $600) fSjB/tk No preference ,;C— / Importance of thi! \ . 1 2 3 4 5 ^^^^ te^^^ j^^^ not important • j K n Question Feature 6 7 8 9 very important Next Question 4 5 6 7 8 9 10 Figure 5-6: Study Two - Screenshot of the Question Page (Conditions5 and 6) 5.3.3 Dependent Variables and Control Variables All dependent variables used in Study One were employed again in Study Two. All of them were measured on validated scales with minor adaptation on their wording. The measures for social presence were adapted from (Gefen and Straub 2003; Kumar and Benbasat 2004). The three trusting beliefs in recommendation agents were adapted from (Wang and Benbasat 2004; Wang 2005), which in turn relied on web trust measures developed by (McKnight et al. 2002). The measures for perceived enjoyment were adapted from (Koufaris 2002) and (Van der Heijden 2004). The questionnaire also 106 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities included measurement items of Perceived Usefulness, Perceived Ease of Use, and Agent Adoption Intentions, which will be used for PLS analysis explicated in Chapter 6. Those measures were adapted from (Davis 1989), which has been adopted and validated in numerous TAM studies. All measurement items are listed in the Appendix A. Since the constructs were measured by multiple items, summated scales based on the average scores of the multi-items were used in the analysis. Responses were recorded on a seven-point Likert scale (1 = "strongly disagree" to 7 = "strongly agree"). All control variables had been used in Study One, including: 1) users' dispositional trust on competence, benevolence, and integrity, all adapted from (McKnight et al. 2002); 2) users' trust propensity in regards to computers, which was adapted from (Lee and Turban 2001); and 3) product expertise, which was adopted from (Flynn and Goldsmith 1999). 5.3.4 Participants, Incentives, and Experimental Procedure A total of 168 students and staff members were recruited campus-wide from the University of British Columbia through online and offline advertisement. Based on (Cohen 1988), this sample size (28 subjects per group) could ensure sufficient statistical power (about 80%) at the significance level of .05 when medium effect size (f = 0.25) was assumed for both the main effects and the two-way interaction effects. To control potential influences of users' demographic backgrounds, the subjects' gender and ethnicity were completely balanced in each treatment group. Only subjects from the following four demographic groups were recruited: Asian Female, Asian Male, Caucasian Male, and Caucasian Female. Seven participants from each of these four demographic group were randomly assigned to one of the six treatment groups. 107 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities To avoid potential biases in their evaluations and to ensure that their interactions with the agent were necessary for the task, only individuals who did not own or had not purchased a digital camera before, as well as those who had owned or purchased a digital camera but with low expertise were invited (Wang and Benbasat 2005). Subjects were asked to report their expertise level on digital cameras when signing up for the study. A seven-point scale was used, ranging from "Not at all" (1) to "Expert" (7). Only those with a score of 4 or below were invited. Similar procedures as used in Study One were used in Study Two. The experiment process was administered during a single one hour session. Each participant was welcomed and given an informed consent form to review. Each participant was then asked to fill out a short online questionnaire which included a demographic questions and instruments related to all control variables. A research assistant then trained participants how to use the agent by playing a pre-recorded video clip (with narration) that explained the interface and all functionalities of the experimental PRA. Afterwards, the participant was asked to complete a simulated shopping task. To reduce the potential confound that the subject may have purchased a digital camera for a similar situation, two shopping tasks were prepared and the participant was randomly assigned to one of them. One task asked the subject to choose a digital camera for a close friend (a photography enthusiast) as a wedding gift; the other asked the subject to select a model for a close family member (an amateur user)2. Participants were informed that there is no time limit for the task, but most participants spent 20-30 minutes to complete the task. After the task, participants 2 Post-hoc tests indicated no significant differences between tasks on all dependent variables. 108 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities were asked to complete a post-task questionnaire which includes the measures of all dependent variables. Participants were randomly assigned to one of the treatment groups. They were guaranteed monetary compensation for their participation ($10 each). In order to motivate participants to view the experiment as a serious online shopping session and to enhance their involvement, they were told that the camera model they picked would be evaluated and ranked by a digital camera expert on its appropriateness for the requirements described in the task scenarios. They were told the top 25% best-performers would receive an extra bonus ($50). 5.4 Data Analysis and Findings This section first reports the demographic data of the participants in the experiment. A multiple analysis of covariance (MANCOVA) was conducted to test whether or not PRA's anthropomorphic embodiment and output modality exert an overall direct impact on the dependent variables (Hair et al. 1998). Based on the MANCOVA results (presented in section 5.4.2), follow-up analysis of covariance (ANCOVA) was performed on individual dependent variables, as discussed in Section 5.4.3. 5.4.1 Demographic Data Table 5-2 outlines the characteristics of the participants who participated in the experiment. Similar to the participants in Study One, the majority of participants were undergraduate students around 20 years old. All participants had at least four years of Internet experience, and most had used the Internet for more than seven years. Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Table 5-2: Study Two - Demographic Data # of Participants Percentage Gender / Ethnicity Asian Female 28 25.0% Asian Male 28 25.0% Caucasian Female 28 25.0% Caucasian Male 28 25.0% Age ' 18-20 89 52.98% 21-25 69 41.07% 26-30 4 2.38% 31-35 4 2.38% >35 2 1.69% Year in School ' " Sophomore 57 33.93% Junior 51 30.36% Senior 47 27.98% Graduate 8 4.76% Staff 5 2.97% ...Years of Internet Experience <4 2 1.19% 4-6 31 18.45% 7-10 119 70.83% >10 16 9.53% In the background questionnaire, participants were asked about their product expertise, attitudes towards computers, and perceived risk of online shopping. Their comfort levels with the Internet and online shopping were also assessed. No significant differences were found between subjects randomly assigned to one of the experimental groups for each of these variables. Also, no significant differences were found between groups with respect to participants' age and years of Internet experience. 5.4.2 M A N C O V A Results The descriptive data of the dependent variables are shown in Table 5-3. The measures for all dependent variables have good reliabilities (Cronbach's alphas > .70 as indicated in table 5-4). n o Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Table 5-3: Study Two - Group Means and Standard Deviations Avatar Voice Social Competence Benevolence Integrity Perceived Presence Beliefs Beliefs Beliefs Enjoyment Without Text 3.82 5.46 5.24 5.29 4.09 (84) (28) (1.11) (0.75) (0-71) (0.69) (0.71) TTS 3.48 5.57 5.37 5.51 4.17 (28) (1.14) (0.65) (0.90) (0.75) (0.95) Human 4.64 5.61 5.61 5.74 4.86 (28) (1.27) (0.94) (1.10) (0.78) (1.29) Total 3.98 5.55 5.40 5.51 4.37 (84) (1.26) (0.79) (0.92) (0.75) (1.06) With Text 4.25 5.69 5.44 5.71 4.66 (84) (28) (0.76) (0.61) (0.82) (0.67) (0.61) TTS 4.11 5.66 5.45 5.57 4.68 (28) (0.83) (0.75) (0.89) (0.76) (0.88) Human 4.73 6.02 5.77 5.75 4.91 (28) (0.91) (0.56) (0.69) (0.72) (0.86) Total 4.36 5.79 5.56 5.68 4.75 (84) (0.87) (0.66) (0.81) (0.71) (0.79) Total Text 4.04 5.57 5.34 5.50 4.38 (168) (56) (0.97) (0.69) (0.76) (0.71) (0.71) TTS 3.79 5.61 5.41 5.54 4.43 (56) (1.04) (0.69) (0.89) (0.75) (0.95) Human 4.69 5.82 5.69 5.75 4.88 (56) (1.09) (0.79) (0.91) (0.74) (1.08) Total 4.17 5.67 5.48 5.60 4.56 (168) (1.10) (0.73) (0.86) (0.74) (0.95) Table 5-4: Study Two - Reliabilities and Bivariate Correlations of Dependent Variables Cronbach's Alpha 1 2 3 4 5 1. Social Presence 0.887 1.00 2. Competence Beliefs 0.755 0.27 1.00 3. Benevolence Beliefs 0.794 0.42 0.48" 1.00 4. Integrity Beliefs 0.726 0.30 0.53" 0.70 1.00 5. Perceived Enjoyment 0.872 0.63 0.27" 0.32 0.34 1.00 Correlation is significant at the 0.01 level (2-tailed). i l l Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Since we predicted the effects of humanoid embodiment and output modality on multiple dependent variables (i.e., social presence, trusting beliefs, and perceived enjoyment), MANCOVA was applied first to test whether the experimental treatments have an overall impact on these dependent variables (Hair et al. 1998). This study also tested several control variables as described in Section 5.3.4. The four dispositional trust beliefs (competence, benevolence, integrity, and trusting stance) and users' trust propensity towards computers are included in MANCOVA test. The MANCOVA results are presented in Table 5-5. Significant main effects were found on both humanoid embodiments (Wilks' Lambda = 0.928, F (5, 153) = 2.387, p < 0.05) and output modality (Wilks' Lambda = 0.847, F (10, 306) = 2.646, p < 0.01). No interaction effect was found to be significant. The MANCOVA results allow us to examine potential main effects on individual dependent variables via ANCOVA. Table 5-5: Study Two - MANCOVA Results Wilks' Lambda D F F-Value P-Value Embodiment 0.928 5, 153 2.387 0.041 Modality 0.847 10,306 2.646 0.004 Embodiment * Modality 0.937 10,306 1.005 0.439 Dispositional Trust -Competence (Covariate) 0.951 5, 153 1.561 0.174 Dispositional Trust -Benevolence (Covariate) 0.986 5, 153 0.447 0.815 Dispositional Trust -Integrity (Covariate) 0.982 5, 153 0.558 0.732 Dispositional Trust -Trusting. Stance (Covariate) 0.971 5, 153 0.916 0.472' Trust Propensityltoward Computersi(Gdvariate) 0.925 5,153: ! 2.488 0.034 112 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities 5.4.3 A N C O V A Results 5.4.3.1 Social Presence The ANCOVA results for the impact on social presence are listed in Table 5-6. Users reported a stronger sense of social presence when the agent was embodied with an animated character (F (1, 162) = 5.871,/? < 0.05). Output modality also appeared to exert a significant influence on social presence (F (2, 162) = 11.495,/? < 0.01). To compare the three output modalities, a Scheffe test for multiple comparisons was conducted (Table 5-7), the results of which indicate that the difference between human voice and text, as well as the difference between human voice and TTS voice, is significant, while no significant difference was found between text and TTS voice. Therefore, Hypotheses H3a, H3b, and H3c are supported, while H3d is not. Table 5-6: Study Two - ANCOVA Results (Social Presence) Source DF Sum of Squares Mean Square F P-Value Embodiment 1 6.095 6.095 5.871 0.016 Modality 2 23.869 11.934 11.495 < 0.001 Embodiment * Modality 2 2.110 1.055 1.016 0.364 Error 162 188.330 1.207 Table 5-7: Study Two - Scheffe Test on PRA Modalities (Social Presence) Group A Group B Mean Difference (A-B) Significance Text Only TTS Voice 0.243 0.453 Human Voice -0.650 0.004 TTS Voice Text Only -0.243 0.453 Human Voice -0.893 < 0.001 113 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities 5.4.3.2 Trust The ANCOVA results of three trusting beliefs are shown in Tables 5-8 to 5-10. The only significant difference arose in regards to competence beliefs: users had stronger beliefs in the agent's competence when it was embodied with an animated character (F (1, 159) = 4.870, p < 0.05). Humanoid embodiments had no significant effects on benevolence beliefs (F (1, 159) = 2.136, p > 0.10) and integrity beliefs (F (1, 159) = 2.825, p > 0.09). Output modality shows no significant difference on competence (F (2, 159) = 1.792,/? > 0.17), benevolence (F (2, 159) = 2.358,/? > 0.09), and integrity (F (2, 159) = 1.469, p > 0.20). Therefore, among the hypotheses on trusting beliefs, only H4 is supported. Table 5-8: Study Two - ANCOVA Results (Competence Beliefs) Source DF Sum of Squares Mean Square F P-Value Embodiment 1 2.556 2.556 4.870 0.029 Modality 2 1.882 0.941 1.792 0.170 Embodiment * Modality 2 0.686 0.343 0.653 0.522 Dispositional Trust -Competence (Covariate) 1 0.231 0.231 0.440 . 0.508 Dispositional Trust -Trusting Stance (Covariate) 1 0.149 0.149 0.283 0.596 Trust Propensifyitoward Computers (Covariate) 1 0.192 0.192 0.366 0.546 Error 159 83.467 0.525 Table 5-9: Study Two - ANCOVA Results (Benevolence Beliefs) Source DF Sum of Squares Mean Square F P-Value Embodiment 1 1.480 1.480 2.136 0.146 Modality 2 3.268 1.634 2.358 0.098 Embodiment * Modality 2 0.070 0.035 0.050 0.951 Dispositional Tru-.t -Benevolence (Covariate) 1- 0.092 0.092 0.1.V, 0.716 Dispositional Trust -Trusting Stance (Covariate) V 1 " 0.988 0.988 1 126 0.234 114 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Trust Propensity toward Computers, (Covariate) J | | 1 P 5.818 50818 8.3% (1.004 Error 159 110.164 0.693 Table 5-10: Study Two-^ v N C O V A Results (Integrity Beliefs) Source DF Sum of Squares Mean Square F P-Value Embodiment 1 1.341 1.341 2.825 0.095 Modality 2 1.394 0.697 1.469 0.233 Embodiment * Modality 2 0.845 0.422 0.890 0.413 Dispositional Trust -Integrity (Covariate) 1 0.238 0.238 0.501 0.480 ^Dispositional Trust -Trusting Stance (Covariate) 1 ' ?. 1.279 1.279 2.695 0:103 Trust Propensity toward Computers (Covariate) 1 5.739 5.739 12.095 0.001 Error 159 0.475 5.4.3.3 Perceived Enjoyment The ANCOVA results for the impact on perceived enjoyment are shown in Table 5-11. Users felt that their interactions were more enjoyable when the agent was embodied with an animated character (F (1, 162) = 7.105,/) < 0.01). Output modality also exerted a significant influence on social presence (F (2, 162) = 5.295,/? < 0.01). To compare the three output modalities, a Scheffe test for multiple comparisons was conducted (Table 5-12) and the results indicate that the difference between human voice and text and the difference between human voice and TTS voice, are significant, while no significant difference was found between text and TTS voice. Therefore, Hypotheses H7a, H7b, and H7c are supported. Table 5-11: Study Two - ANCOVA Results (Perceived Enjoyment) Source DF Sum of Squares Mean Square F P-Value Embodiment 1 5.869 5.869 7.105 0.008 Modality 2 8.749 4.374 5.295 0.006 Embodiment * Modality 2 2.225 1.112 1.347 0.263 Error 162 133.819 0.826 115 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Table 5-12: Study Two - Scheffe Test on PRA Modalities (Perceived Enjoyment) Group A Group B Mean Difference (A-B) Significance Text Only TTS Voice -0.050 0.959 Human Voice -0.507 0.014 TTS Voice Text Only 0.050 0.959 Human Voice -0.457 0.031 A summary of hypotheses testing results are provided in Table 5-13. Table 5-13: Study Two - A Summary of Hypothesis Testing Results Hypothesis Supported? H3a: Social Presence (with embodiment) > Social Presence (without embodiment) Yes H3b: Social Presence (human voice) > Social Presence (text only) Yes H3c: Social Presence (human voice) > Social Presence (TTS) Yes H3d: Social Presence (TTS) > Social Presence (text only) No H4: Competence Belief (with embodiment) > Competence Belief (without embodiment) Yes H5a: Benevolence Belief (with embodiment) > Benevolence Belief (without embodiment) No H5b: Benevolence Belief (human voice) > Benevolence Belief (text only) No H5c: Benevolence Belief (human voice) > Benevolence Belief (TTS) No H5d: Benevolence Belief (TTS) > Benevolence Belief (text only) No H6a: Integrity Belief (with embodiment) > Integrity Belief (without embodiment) No H6b: Integrity Belief (human voice) > Integrity Belief (text only) No H6c: Integrity Belief (human voice) > Integrity Belief (TTS) No H6d: Integrity Belief (TTS) > Integrity Belief (text only) No H7a: Perceived Enjoyment (with embodiment) > Perceived Enjoyment (without embodiment) Yes H7b: Perceived Enjoyment (human voice) > Perceived Enjoyment (text only) Yes H7c: Perceived Enjoyment (human voice) > Perceived Enjoyment (TTS) Yes 116 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities 5.5 Discussion The second study provides strong evidence proving that humanoid embodiments and output modalities can be employed to enhance social interactions between a PRA and its users. As expected, an embodied PRA can provoke significantly stronger perceptions of social presence than a disembodied one. This finding demonstrates that humanoid social interface is helpful in the context of online shopping and recommendation agent design. Moreover, our results suggest that an animated face can make an agent more fun to use, which is an important design factor for non-workplace information systems such as recommendation agents. PRA's voice output also plays an important role in inducing stronger social and affective responses from users. As we have expected, the rich social cues embedded in human voices greatly enhance users' social responses and enjoyment. In contrast, computer-generated voices, which could only carry limited social cues, were perceived no different from text. It is noteworthy that users in TTS conditions reported even weaker (although not statistically significant) social presence than those from the text-only conditions (MeanxTS= 3.79, MeanText = 4.04, 7-point scale). This suggested that the mere addition of voice channel in the agent interface could not automatically benefit the social interactions. What actually matters are the characteristics of the voice, such as pitch, syllable speed, and interphrase pauses (Chattopadhyay et al. 2003). Even though people may be able to attribute certain social characteristics to computer-generated voices, as demonstrated in prior studies (Lee et al. 2000; Nass and Lee 2000; Nass and Lee 2001; 117 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities Lee 2003), it is still premature to apply TTS voice in recommendation agent interface, where people expect a much more "human" tone from a voice interface. Mixed results have emerged regarding the impacts of PRA interfaces on users' trusting beliefs. Firstly, the presence of an animated character could enhance users' trusting beliefs in an agent's competence. This supported our hypothesis that online shoppers still perceive that a PRA that looks more like a human expert could understand their needs better and give them higher quality recommendations. In contrast, for both benevolence and integrity beliefs, although group means of with-avatar conditions and with-voice conditions are higher than those of no-avatar conditions and text-only conditions respectively, the differences were not statistically significant. One explanation is that for PRA users to identify the benevolence and integrity from the agent, more subtle social cues are needed to reinforce users' beliefs that the agent had not put the online store's interests first by giving inappropriate advice to obtain a higher profit margin. As most customers may have similar concerns when interacting with a salesperson in a physical store, they are more sensitive to the "hidden" information embedded social interactions that are commonly used by people to manifest benevolence and credibility and to convey goodwill and amity, such as small talk (Cassell and Bickmore 2000), facial expressions, eye contact, and paralanguage (Cowell and Stanney 2005). Ceiling effects could be another possible explanation, as the results indicate that users in the no-avatar/text-only condition reported relatively high scores on benevolence and integrity (Benevolenceno avatar/text = 5.24, Integrity,,,, avatar/text = 5.29, 7-point scale), which implies that during the experiment, a lot of concerns a shopper may have in regarding to the agent's motives and intentions were already addressed by the verbal 118 Chapter 5: Influences of PRA's Humanoid Embodiment and Output Modalities information (questions, answers, explanations, or comments) provided; as a result, limited room was left for significant improvements from the interface components. 5.6 Chapter Summary This chapter discusses the distinct impacts of two PRA interface components, namely humanoid embodiments and output modalities, on users' perceptions of social presence, trusting beliefs, and perceived enjoyment. In the next chapter, we will investigate the interrelationship between these dependent variables, as well as how users' perceptions of social interactions with the PRA will ultimately affect their intentions to adopt the agent. A Partial Least Square (PLS) analysis was conducted to empirically examine these relationships by using the data collected from both experiments. 119 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions C H A P T E R 6 : H U M A N - P R A S O C I A L I N T E R A C T I O N S A N D U S E R S ' A D O P T I O N I N T E N T I O N S The results from both Study One and Study Two have revealed that social presence, trusting beliefs, and perceived enjoyment are highly correlated. What are the potential inter-relationships between these constructs? More importantly, how do perceptions of social experience ultimately influence users' intentions to adopt the agent? In this chapter, we employ a structural equation modeling method to investigate these relationships based on the data collected from both studies. 6.1 Introduction Users' intentions to accept a new technology have been traditionally explained either by attitude towards that behavior as in the theory of planned behavior (Ajzen 1991) or by a set of beliefs as in the technology acceptance model (TAM) (Davis 1989). As one of the well-established adoption theories in IS research, TAM suggests that users' decisions to accept a new information technology are based on two rational assessments of its expected outcomes: 1) Perceived Usefulness (PU), which directly affects the decision to use the new technology; and 2) Perceived Ease of Use (PEOU), which affects intended use primarily through PU (Davis 1989). Although the applicability of TAM had been established by numerous studies (see (Venkatesh et al. 2003) for a comprehensive review), its cognitive-based orientation have inevitably excluded other important components of attitudes. In recent years, the original TAM model has been adapted to accommodate additional constructs in a wide range of contexts, notably Internet use and online shopping (Venkatesh 2000; Gefen et al. 2003; Venkatesh et al. 2003). To explain how 120 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions people's intentions of adopting a PRA are influenced by an anthropomorphic interface, this chapter proposes an extended model which integrates social presence, trusting beliefs, and perceived enjoyment with the TAM constructs (the model is as shown in Figure 6-1). The remainder of this chapter is organized as follows. Section 6.2 develops hypotheses for the proposed model. Results of the measurement model and the structural model are reported in section 6.3, and the chapter concludes with a discussion of the results. Trusting Hll Perceived Beliefs IJ Usefulness H13 \ / H 1 0 \ . Intentions / \H12 H15 to adopt a PRA H16 / Perceived Perceived H14 Enjoyment H17 Ease of Use Figure 6-1: The Structural Model 6.2 Hypotheses Development 6.2.1 Social Presence and Trust In exploration of the effects of social presence on consumer trust and the relative importance of trust in TAM beliefs, Gefen and Straub (2003) have argued that, in B2C e-services, social presence can be an enabler of trust-building cues, inasmuch as 121 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions untrustworthy behavior is more easily hidden in lean social presence situations; furthermore, social presence can build trust in its own right. This relation arises because the feeling of interacting with a human sales representative suggests that the vendor behind the website is not hiding information and is investing in the potential relationship with its customers. Lee and Nass (2004) have identified a significant mediating effect of social presence on website content credibility in their study of the effects of multiple synthetic voices in e-commerce context. Another study by Hertzum et al. (2002) has also revealed that the trusting beliefs of users of e-commerce websites are affected significantly by the visual appearance of virtual agents, but the valence of the responses in the study was not uniform: agents with a visual appearance that is perceived as cold or out of place relative to the provided service can become a threat to trustworthiness, whereas agents with more realistic personifications are perceived more engaging, lively, and convincing than virtual agents represented by a text field only. In the present research, we agree with Gefen and Straub (2003) that the social presence a PRA affords can positively contribute to consumer trust. In face-to-face human interactions, trustworthiness is usually manifested by a plethora of trivial but important social cues. As inexperienced users interact with an agent for the first time, they have very few cues to judge an agent's trustworthiness. As Fogg and Tseng (1999) have pointed out, the impressions formed through simple inspection of "surface attributes" can directly influence users' attitudes and behavior. When the interaction with an agent is perceived as socially warm, the general social convention that "a nice person is usually trustworthy" is applied automatically. In addition, when consumers attempt to solicit product recommendations from a PRA, it is very likely that they will contrast their 122 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions experiences to the process of obtaining advice from friends, salespeople, or human experts. A socially-rich experience can therefore permit a recommendation agent to function as an information source that not only gives personalized advice but also exhibits a human touch, which is considered more credible than other impersonalized sources (Price and Feick 1984). In summary, we propose that: H8: Users 'perceptions of social presence will positively affect their trusting beliefs. 6.2.2 Social Presence and Perceived Enjoyment Researchers studying physical presence have long found that technologies that afford a strong sense of presence can provide enjoyment and delight to their users (Fleeter 1995; Lombard and Ditton 1997). However, little research has explored the impacts of social presence on enjoyment. Some theories and findings in social psychology can probably lend some support to address this gap. A major portion of people's lives concerns dealing with other people, and the importance of personal relationships in the real world is well accepted. In general, people like to learn and conduct business with other people present. This is not only convenient and often necessary; it is also intrinsically desired (Baumeister and Leary 1995). Culturally, we are taught from youth how to be social, and we are told that a sociable person is generally well-liked. Psychologically, humans have evolved to facilitate social relationships (Lee 2004). Therefore, we hypothesize that in human-PRA interactions, people will enjoy interacting with a PRA that carries more social cues and social intelligence. H9: Users' perceptions of social presence will positively affect their perceived enjoyment. Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions 6.2.3 Trust and TAM The connections between trust and TAM have been widely discussed in previous studies (Gefen et al. 2003; Pavlou 2003; Wang and Benbasat 2005; Wu and Chen 2005). Gefen et al. (2003) have integrated consumer trust into the traditional TAM model in the context of online shopping, arguing that trust, conceptualized as a set of trusting beliefs towards an online vendor, would directly affect intentions to use a B2C website, along with perceived usefulness (PU) and perceived ease of use (PEOU). Trust in a merchant can also affect PU in both the short term and the long term. In addition, PEOU has been hypothesized to exert a positive influence on trust, as PEOU can help promote consumers' favorable impressions of e-vendors in the initial adoption, and it can enhance consumers' willingness to make an investment and a commitment in buyer-seller relationships. Wang and Benbasat (2005) have extended this integrated Trust-TAM model to the context of online recommendation agents. Similarly, they have found that initial trust in online recommendation agents can positively affect intentions to adopt agents as well as the consumers' perceived usefulness of the agent. These recent findings, as well as the relationship represented in the original TAM model (Davis 1989), are investigated in the present study: H10: Users' trusting beliefs in a PRA will positively affect their intentions to adopt the agent. Hll: Users' trusting beliefs in a PRA will positively affect perceived usefulness of the agent. H12: Perceived ease of use in a PRA will positively affect trust in the agent. H13: Perceived usefulness of a PRA will positively affect consumers' intentions to adopt the agent. 124 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions H14: Perceived ease of use of a PRA will positively affect consumers' intentions to adopt the agent. HIS: Perceived ease of use of a PRA will positively affect perceived usefulness of the agent. 6.2.4 Perceived Enjoyment and T A M Perceived enjoyment is considered to be an important addendum to the original TAM (Davis et al. 1992). In IS research, some studies have demonstrated that enjoyment influences usefulness through the mediation of ease of use (Venkatesh 2000); others have found that, while intrinsic motivation has no direct influence on intention to use technology, it can serve as an important predictor for both perceived usefulness and perceived ease of use (Venkatesh et al. 2002; Yi and Hwang 2003). Recent research on non-workplace information systems, such as the Internet, online shopping, and video-games, has suggested a more important role for perceived enjoyment. It is not only an immediate antecedent to perceived usefulness, but it is also more important to adoption of information systems (Teo et al. 1999; Venkatesh 1999; Moon and Kim 2001; Van der Heijden 2004; Li et al. 2005; Shang et al. 2005). Van der Heijden (2004) have argued that the aforementioned systems share common characteristics with hedonic systems, which aim to provide self-fulfilling value to users, in contrast to utilitarian systems that aim to provide instrumental value to users. The encouragement of "prolonged use", rather than "productive use", should be the dominant design objective of hedonic systems. Although in (Van der Heijden 2004), information systems are dichotomized into either utilitarian or hedonic category, we contend that many information systems possess 125 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions the characteristics of both aspects at the same time, especially for systems developed for home use, such as computer-aided tutoring systems and online shopping websites. Marketing research has suggested that individuals' shopping behavior is driven by both instrumental and hedonic motivations (Babin et al. 1994). Consumers' shopping values are subjective, characterized by their interactions with an environment and indicated by both the event's usefulness and an appreciation of its activities. In online shopping studies, the effects of shopping enjoyment have been identified as important determinants of online customer loyalty (Jarvenpaa and Todd 1997; Koufaris 2002); furthermore, the immersive, hedonic aspects of online shopping environments have been found to play at least an equal role as the instrumental aspects of the environments (Childers et al. 2001). The current study proposes that for users of a recommendation agent, which is a decision support tool used in shopping context, perceived enjoyment will exert an indirect influence on their adoption intentions. As argued in (Van der Heijden 2004), the unmediated impacts of perceived enjoyment on adoption intentions are most manifest only among hedonic-oriented systems, in which the instrumental values of the system are less important than the pleasurable experience a user could acquire from the usage. In contrast, the primary role of a PRA is to help shoppers complete a cognitive task of identifying a specific product among hundreds of alternatives, the extrinsic motivation -find the right product - is still the primary incentive for a user to interact with an agent. When the agent is fun to use, users will perceive the task as being less effortful; in addition, users are likely to be more engaged in the task and therefore more likely to get Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions higher quality output. Therefore, the impacts of perceived enjoyment on behavioral intentions will be fully mediated by perceived usefulness and perceived ease of use. HI6: Users' perceptions of enjoyment will positively affect perceived usefulness of the agent. HI 7: Users' perceptions of enjoyment will positively affect perceived ease of use of the agent. 6.3 Data Analysis and Findings In addition to the measures of social presence, trust, and enjoyment, subjects in both Study One and Study Two were asked to complete a questionnaire with measurement items of PU, PEOU, and intentions to adopt. These measures are adapted from Davis's scale (Davis 1989) (all measurement items are listed in Appendix A). The results collected from both studies are pooled together (with a total sample size of 300) and Partial Least Squares (PLS) analysis was used to assess the measurement properties of the dependent variables and the structural model. PLS is a technique that uses a combination of principal components analysis, path analysis, and regression to simultaneously evaluate theory and data (Pedhazur 1982). The impact of all the exogenous variables is determined together on the endogenous variable. The path coefficients in a PLS structural model are standardized regression coefficients, while the loadings can be interpreted as factor loadings. The measurement properties of the dependent variables are presented in section 6.3.1 and the results of the structural model are reported in section 6.3.2. Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions 6.3.1 Measurement Model All dependent variables were modeled as reflective constructs. As part of the PLS analysis, we used PLS-Graph 3.0 to examine: 1) individual item reliability; 2) internal consistency; and 3) discriminant validity (Barclay et al. 1995). Internal consistency was examined by the composite reliability developed by Fornell and Larcker (1981), as a measure of reliability similar to Cronbach's alpha. Both composite reliability and Cronbach's alpha are reported in Table 6-1. The benchmark for acceptable reliability is 0.7 and all constructs met this criterion, indicating that the measures are internally consistent. In PLS, discriminant validity can be examined by two criteria. One criterion is that a construct should share more variance with its own measures than it shares with other constructs in a model. The measure of Average Variance Extracted (AVE), suggested by Fornell and Larcker (1981), was used, in which the square root of AVE of a construct should be greater than the correlations between the construct with other constructs (Barclay et al. 1995). Table 6-1 shows the square roots of AVEs and correlations between constructs. The results met this criterion. The second criterion for discriminant validity is that no item should load more highly with a construct other than the construct it intends to measure. The loadings and cross-loadings of measures are shown in Table 6-2. An examination of the matrix reveals that all items satisfied this criterion. 128 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions Table 6-1: Construct Attributes Variable Mean S.D. Cronbach Alpha Composite Reliability 1 2 3 4 5 6 7 8 1. Social Presence 4.24 1.13 0.887 0.917 0.83a 2. Competence Beliefs 5.61 0.77 0.769 0.858 0.29** 0.78a 3. Benevolence Beliefs 5.45 0.93 0.816 0.893 0.49" 0.49" 0.86a 4. Integrity Beliefs 5.61 0.75 0.732 0.834 0.35" 0.50" 0.73" 0.76a 5. Perceived Enjoyment 4.56 1.03 0.882 0.914 0.65 0.36" 0.40" 0.31" 0.83a 6. Perceived Usefulness 5.39 0.95 0.864 0.908 0.33 0.54" 0.47** 0.45" 0.49** 0.84a 7. Perceived Ease of Use 5.91 0.60 0.755 0.837 0.19 0.32 0.29" 0.31" 0.30" 0.33 0.75a 8. Agent Adoption Intentions 4.81 1.44 0.961 0.975 0.35 0.50 0.40" 0.39" 0.42** 0.73 0.27 0.96a a The scores in the diagonal of the matrix are square roots of AVEs, while the lower triangle represents the correlations between constructs. Correlation is significant at the 0.01 level (2-tailed). Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions Table 6-2: Loading and Cross-Loadings of Measures SP TCA TBA TIA PE PU PEOU AAI SP1 .861 .248 .428 .308 .528 .301 .193 .316 SP2 .810 .248 .371 .270 .520 .271 .176 .303 SP3 .872 .249 .419 .291 .602 .315 .251 .301 SP4 .790 .266 .379 .279 .522 .269 .123 .271 SP5 .817 .213 .424 .301 .517 .255 .163 .246 TCA1 .189 .812 .297 .301 .275 .433 .275 .368 TCA2 .281 .857 .394 .395 .345 .459 .342 .431 TCA3 .270 .849 .490 .482 .314 .467 .415 .472 TCA4 .156 .565 .360 .391 .178 .343 .182 .282 TBA1 .377 .510 .842 .690 .318 .452 .270 .422 TBA2 .422 .326 .865 .640 .345 .386 .217 .271 TBA3 .455 .429 .865 .607 .367 .384 .323 .342 TIA1 .267 .426 .655 .831 .237 .454 .302 .332 TIA2 .285 .401 .558 .751 .304 .363 .223 .306 TIA3 .293 .376 .613 .806 .249 .346 .319 .320 TIA4 .191 .293 .395 .581 .149 .210 .200 .211 PE1 .564 .336 .360 .299 .866 .451 .313 .380 PE2 .528 .254 .299 .165 .770 .418 .169 .361 PE3 .574 .335 .365 .294 .812 .381 .363 .313 PE4 .432 .286 .308 .275 .779 .414 .272 .322 PE5 .566 .287 .312 .274 .893 .411 .258 .336 PU1 .333 .495 .394 .416 .505 .856 .317 .665 PU2 .191 .406 .388 .379 .324 .799 .297 .540 PU3 .269 .430 .392 .373 .391 .882 .276 .654 PU4 .332 .505 .430 .416 .451 .835 .426 .602 PEOU1 .214 .345 .274 .284 .322 .362 .808 .296 PEOU2 .048 .023 .085 .094 .104 .075 .531 .043 PEOU3 .196 .348 .264 .312 .272 .326 .854 .237 PEOU4 .140 .335 .257 .289 .243 .312 .783 .268 AAI1 .333 .488 .378 .362 .396 .698 .294 .967 AAI2 .330 .513 .424 .416 .402 .718 .326 .954 AAI3 .338 .462 .371 .364 .401 .696 .293 .970 130 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions 6.3.2 Structural Model PLS was also used to analyze the structural model. The statistical significance of the path coefficients was determined with a bootstrapping technique using 100 samples (Chin 1998). The objective of a PLS analysis is to explain variance in endogenous constructs, rather than replicating the observed covariance matrix as is the case with covariance structure techniques (such as LISREL). Trust can be modeled as a reflective second order factor, because the three trusting beliefs are closely correlated to each other (McKnight et al. 2002). This second order construct of trust in online recommendation agents is composed of three sub-constructs (i.e., competence, benevolence, and integrity), which are also measured as reflective. According to previous studies that have investigated trust in recommendation agents (Wang and Benbasat 2005), we have also used factor scores of each first order trusting belief as indicators for the second-order constructs of trust in agents in PLS. Figure 6-2 illustrates the results to the structural model. Social presence significantly influences both trust and perceived enjoyment. Trust significantly correlates with perceived usefulness, perceived ease of use, and agent adoption intentions; it also partially mediates the influence of PEOU on adoption intentions. Perceived enjoyment correlates with both perceived usefulness and perceived ease of use. As part of the classical TAM, PU positively influences adoption intentions, and it mediates the influence of PEOU on adoption intentions. In summary, Hypotheses H8, H9, H10, H l l , H12, H13, H15, H16, and H17 are supported, while H14 is not supported. Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions Figure 6-2: PLS Structural Model Testing Results A summary of hypotheses testing from PLS analyses are provided in Table 6-3. Table 6-3: PLS Analysis - A Summary of Hypothesis Testing Results Hypothesis Supported? H8: Social Presence -> Perceived Usefulness H9: Social Presence -> Perceived Enjoyment Yes Yes H10: Trusting Beliefs Agent Adoption Intentions HI 1: Trusting Beliefs -> Perceived Usefulness HI 2: Perceived Ease of Use Trusting Beliefs Yes Yes Yes HI3: Perceived Usefulness Agent Adoption Intentions HI4: Perceived Ease of Use -> Agent Adoption Intentions Yes No 132 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions HI 5: Perceived Ease of Use -> Perceived Usefulness Yes HI6: Perceived Enjoyment -> Perceived Usefulness H17: Perceived Enjoyment Perceived Ease of Use Yes Yes 6.4 Discussion PLS results of the structural model supported our hypotheses that social presence is positively associated with both trust and perceived enjoyment. It corroborated the argument that, in the context of online shopping, social presence can directly contribute to trust (Gefen and Straub 2003). On the other hand, this is one of the first studies that demonstrated a positive association between social presence and enjoyment. The proposed relationship between trust and TAM constructs (Gefen et al. 2003; Wang and Benbasat 2005) once again received empirical support from our data, showing the robustness of the Trust-TAM model in the context of online shopping and recommendation agents. Trust in an online recommendation agent positively affects perceived usefulness of the agent and, at the same time, exert a direct influence on agent adoption intentions. In addition, an easy-to-use agent could lead to higher levels of initial trust. PLS results also support the hypotheses that perceived enjoyment exerts positive effects on both perceived usefulness and perceived ease of use. As expected, effects of perceived enjoyment on adoption intentions are mostly mediated by perceived usefulness and perceived ease of use. The impact of perceived ease of use (PEOU) on intentions to adopt agents is fully mediated by perceived usefulness (PU) and trust, which corroborates with many other TAM studies (e.g., Davis 1989) have found that PEOU is mediated by PU and Trust-133 Chapter 6: Human-PRA Social Interactions and Users' Adoption Intentions TAM studies (Gefen et al. 2003; Wang and Benbasat 2005) that PEOU is mediated by both PU and trust. 6.5 Chapter Summary In this chapter, the inter-relationships between the three core constructs of human-PRA social interactions were discussed and empirically examined with PLS analysis. Social presence has been found a common antecedent of both trust and perceived enjoyment. In addition, the influences of trust and perceived enjoyment on users' adoption intentions were demonstrated by integrating these constructs with TAM. In so doing, not only have we found additional empirical support for the Trust-TAM model proposed by (Wang and Benbasat 2005) for the context of online recommendation agents, but also illustrated the importance of perceived enjoyment in PRA design. In addition to the theoretical framework proposed in Chapter 3, the PLS results empirically justified our emphasis on the social interface design for online recommendation agents. In the next and final chapter, we will summarize all findings of this dissertation, discuss possible limitations, and propose future research directions. 134 Chapter 7: Conclusions and Future Research CHAPTER 7: CONCLUSIONS AND FUTURE RESEARCH 7.1 Conclusions Online recommendation agents are designed to help shoppers analyze vast amount of product information and to provide shoppers relatively small number of recommendations that fit their needs best. While most prior research on recommendation agents has focused on the technical features of PRAs, such as the underlying computing models and algorithms, the social aspects of user-PRA interactions have been understudied. Just as a salesperson is sometimes preferred by shoppers not only because of her professional expertise, but also the way she interacts with her customers, a PRA with certain social intelligence could make it more trustworthy and users' shopping experiences more enjoyable. The overall objective of this dissertation is to understand the importance of social perceptions in user acceptance of online recommendation agents, as well as what design factors could be employed to build and enhance favorable social experiences. Two empirical studies were conducted to examine: 1) the demographic-fit effects between a humanoid PRA and its users; and 2) the distinct and conjoint effects of anthropomorphic embodiment and output modalities in the implementation of humanoid PRAs. Based on similarity-attraction theories and social-identity theories, this dissertation has investigated whether a PRA will be better perceived regarding its sociability, trustworthiness, and enjoyableness when its demographic characteristics are designed to match those of its users'. The two demographic variables examined are ethnicity and gender, traits that are most likely to be used by people as social identity cues when forming social relationship with others. As we expected, people appear to apply similar 135 Chapter 7: Conclusions and Future Research social stereotypes in human-human communications as they apply to evaluate a humanoid agent: the "in-group" favoritism held true in human-agent interactions. Ethnicity-matched PRAs were perceived as more sociable, more competent, and more enjoyable to interact with than their mismatched counterparts. Also, users indicated higher level of trusting beliefs in the agent's competence and integrity of a same-sex PRA than an opposite-sex PRA. More interestingly, our results have illustrated for the first time the interactions between the two demographic factors: ethnicity-interactions are more significant among females than males. This finding corroborates observations from cross-cultural studies that have found that women exhibit stronger adherence to ethnic identities than men do. Also, the experimental results confirm that the use of virtual embodiments increases consumers' perceptions of social presence, their beliefs about agent competence, and their perceived enjoyment. Human voice was found to be significantly more effective than on-screen text and computer-synthesized voice in creating a social and enjoyable interaction experience. Therefore, both "persona-effect" and "modality-effect" are shown to be effective in anthropomorphic PRA design. Furthermore, by integrating TAM with the constructs of trust and perceived enjoyment in recommendation agents, the important role of social relations in influencing user adoption of the agents is confirmed. Social presence is a common antecedent of both trust and perceived enjoyment. Trust exerts a direct impact on intentions to adopt recommendation agents, as well as an indirect impact via perceived usefulness of the agents. Perceived enjoyment also influences adoption intentions through perceived usefulness. 136 Chapter 7: Conclusions and Future Research 7.2 Contributions This dissertation makes several theoretical and practical contributions. From a theoretical perspective, there are three major contributions. First, this research provides additional empirical support for the applicability of the "computers as social actor" (CASA) theories in the context of online recommendation agents. Embodied recommendation agents, despite their technological nature, are treated by users as social actors whose "appeared" demographic characteristics, such as gender and ethnicity, carry certain social cues that could significantly influence users' perceptions. Theories from social psychology predicting that in-group members are treated more favorably than out-group members received empirical support in human-agent interactions. This helps us better understand the variety of sources that could stimulate people's social responses to technological artifacts or communication media. Second, this work has examined specific interface features that influence the perception of social presence, trusting beliefs, and perceived enjoyment. The positive effects of humanoid embodiments and speech output in the context of online shopping and product advising have been verified empirically. This dissertation also provides additional evidence attesting the value of anthropomorphic interfaces in human-computer interactions. Third, this research tested the integration of social presence, trust and perceived enjoyment along with TAM, which helps to explain the importance of social interactions in users' adoption of online recommendation agents. The impacts of social presence have been explored in prior studies of online digital products (Gefen and Straub 2003) and email (Straub 1994; Gefen and Straub 1997); this dissertation extends social presence 137 Chapter 7: Conclusions and Future Research research to online recommendation agents. Findings from this study suggest that social presence a common antecedent to both trust and perceived enjoyment. Furthermore, the influence of trust and perceived enjoyment on users' behavioral intentions has been reexamined in the context of online shopping where interacting with recommendation agents could afford both utilitarian and hedonic value to users. This dissertation also makes significant contributions to practice. The results have several implications for online vendors who strive to design PRAs that could provide online shoppers with more positive virtual advising experiences. First, online vendors and PRA designers should pay more attention to the social cues embedded in every facet of the interaction, and try their best to create favorable social perceptions, to build stronger trusting beliefs about the agent, and to make the agent more fun to use. Animated characters and/or speech interface could be employed to reach this goal, but only with a thoughtful implementation that is based on broad knowledge of prospective users. Humanoid interface could make users feel more natural and behave more socially to the agent, which gives designers more opportunities of "borrowing" social norms from interpersonal communications to create a virtual salesperson that is competent on both professional expertise and social interactions. Second, this research calls for more attention to demographic factors in the design of embodied conversational agents. Compared with other more dynamic and subtle non-verbal behaviors, such as facial expressions or body gestures, the demographic features of a humanoid agent are static and thus relatively easy and economical to manipulate, but that does not mean developers could choose them at will. Rather than creating a character based on gut feelings or personal preferences, designers should make the agent appear 138 Chapter 7: Conclusions and Future Research like an "in-group" member in users' eyes. In this study, gender and ethnicity have been demonstrated as important cues for group identification. When available, customers' demographic variables should be used in deciding which version of an agent to present. Alternatively, the website could provide multiple PRAs with varying demographic embodiment and allow customers to pick their preferred embodiment for the advisor. Also, our results have revealed that effective social interactions can be implemented at relative low-cost1. For online vendors, a 2D Flash-based animated face is probably good enough to embody the PRA. On the other hand, human speech is still more effective than either on-screen text or TTS voice technology; however, its inflexibility (as all audio contents need to be pre-recorded) and much higher accompanied cost2 make it not feasible with small or medium online retailers whose PRAs have been designed with large amount of interactive contents. In contrast, although TTS voice was relatively cheaper , it was perceived to be no better than text. Therefore, before a more naturally-sound TTS engine is ready for the mainstream market, on-screen text is probably a safer bet for online vendors if content flexibility is important or pre-recorded human voices are too costly to produce. 7.3 Limitations There are a number of limitations to this study that should be noted. First, construct measures were collected at a single point in time and via one method, which may have 1 Adding a Flash-based virtual character to a website could be as low as US$10.00 per month by using third-party application service providers such as Oddcast Inc. More charges will be applied when audio capabilities are enabled. 2 Hiring a professional narrator for human-speech recording could cost at least US$50 per minute of speech (price quoted by Oddcast Inc.), and it usually take 2-3 business days if not done in-house. 3 For example, US$100 extra per year for TTS functionality and $4 for 1,000 audio streams (price quoted by Oddcast Inc.) 139 Chapter 7: Conclusions and Future Research induced the common method bias (Straub et al. 1995). To address this potential threat, Harmon's one-factor test was conducted to the data collected from both experiments (Podsakoff and Organ 1986). An exploratory factor analysis was conducted on all dependent variables; no single factor was observed and no single factor accounted for a majority of the covariance in the variables. This suggests that common method bias did not posit a threat for either study. On the other hand, other research methods, such as verbal and behavioral protocol analysis, could be employed in future studies to provide a •J richer understanding and to overcome potential bias from common method variance. Second, our studies have examined users' first impressions of various implementations of embodied PRAs. The perceptions of social presence, trust, and enjoyment, as well as their effects on attitude and behavioral intentions may vary as an online consumer becomes more familiar with a PRA. First time versus repeat customers may exhibit different preferences for the design elements that can induce social and affective responses towards a PRA. Therefore, future research should employ a longitudinal design to determine whether the finding still holds when novelty wears off. The controlled laboratory experiments used in both studies, with their advantages on ensuring internal validity, may affect the generalizability of the results. Given the artificial settings of the fictitious shopping tasks and the simulated PRA, one might expect that the decision-making processes employed by the subjects during the experiment may differ if they were actually making a real purchase decision. Our incentive mechanism (particularly the possibility of a $50 bonus based on the appropriateness of the product that was selected) helped increase the task relevance as well as the task involvement. However, it is possible that subjects may have been overly 140 Chapter 7: Conclusions and Future Research involved compared with the way they might behave in a real-world scenario. Lastly, since the participants were mostly university students (whose demographic data generally match with those of average Internet users (Johnson 2005)), the results from our studies may be somewhat different from results obtained using people from other demographic groups. Also, only one type of product was used. Further research with different participant samples and different types of products is warranted. 7.4 Future Research Besides the topics mentioned in Sections 6.2 and 6.3, there are other future research areas which are also important and promising. First, more research is needed to examine the impacts of other non-verbal cues that might be implemented in an embodied PRA. As proposed in De Meuses's model of non-verbal behavior (De Meuse 1987), demographic variables and physical appearance variables of a PRA, including the agent's age, facial and bodily attractiveness, clothing and accessories, could all influence the perceived sociability, credibility, and likeability of an agent. Results from this research stream could provide practitioners more specific design guidelines for embodied agents used in online shopping environment. It is critical that PRA evaluation studies should always keep users' demographic and cultural characteristics in mind. As our studies suggested, there is no "one design fits all" agent. Various social stereotypes should be well thought-out in PRA design, so users can be presented with a social-identity-fit agent. Furthermore, studies manipulating the realism of an embodied agent, as well as the possible use of human/cartoon celebrities in the real world, will provide more insight on how people reacts to virtual characters in online shopping environments. 141 Chapter 7: Conclusions and Future Research Second, it is worthwhile to investigate how the body gestures of a PRA could influence users' evaluations of the social interactions. Although Cowell and Stanney (2005) have found that perceived trust in an embodied agent did not significantly increase when behaviors from bodily regions (posture and gesture) were added to facial expressions, a full-body agent with appropriate body movements might still exert an impact on social presence and perceived enjoyment. Also, even though the characters used in our experiments were quite realistic, they were still based on cartoonish characters. It will be interesting to assess how people respond to photorealistic characters, and whether the effects of "uncanny valley"4 may enhance or inhibit human-PRA social interactions. Third, as users' interactions with PRAs involve behavioral and psychological activities involving cognitive, social, and affective aspects, further research should be conducted to study how these activities may interact with each other. For example, the "dual process" theories of persuasion, in particular, the Heuristic Systematic Model (Chaiken 1984; Chaiken, Liberman et al. 1989) and the Elaboration Likelihood Model (Petty and Cacioppo 1986), both hold that persuasion may take one of two distinct paths. The first path is taken when a receiver carefully evaluates the most important factors in the persuasive situation, so called "systematic processing" in HSM and "central route" in ELM. The other path, also known as "heuristic processing" in HSM and the "peripheral route" in ELM, requires less effort or attention from the receiver who instead focuses 4 The concept of "Uncanny Valley" comes from the Japanese roboticist Masahiro Mori, who argued that as a robot is made more humanlike in its appearance and motion, the emotional response from a human being to the robot will become increasingly positive and empathic, until a point is reached at which the response suddenly becomes strongly repulsive; as the appearance and motion are made to be indistinguishable to that of human being, then the emotional response becomes positive once more and approaches human-human empathy levels. 142 Chapter 7: Conclusions and Future Research upon variables that are not directly central to the attitude object. It would be interesting to investigate whether or not shoppers rely more on the systematic processing of the product-related contents provided by a PRA than non-verbal cues, such as its humanoid interface, to evaluate the agent when they are highly involved in a shopping task. Fourth, research is also needed to examine input modalities on the user side. For example, the effects of the evolution from simple mouse-clicking regarding predefined questions and answers to the use of natural language (either by a keyboard or through speech recognition software), and the consequences for perceptions of social presence has not been empirically investigated. HCI research in this field could provide a more holistic view of the modality-effects in human-PRA interactions. In conclusion, social interactions with online recommendation agents have been recognized as an important issue in online shopping environments. When an agent responds to users' social and affective needs and affords a compelling socially-rich and enjoyable interaction experience, on top of and aligned with its underlying functionalities of product searching and comparisons, it will more likely be accepted and employed by online shoppers. An interdisciplinary perspective that integrates insights from a variety of academic areas such as information systems, social psychology, communication, and computer science will not only facilitate researchers extend their understandings on how people interact with and respond to emerging technologies in everyday life, but also benefit online retailers by creating a more satisfactory experience for their customers. 143 References REFERENCES Agarwal, R. and Karahanna, E. (2000). "Time flies when you're having fun: Cognitive absorption and beliefs about information technology usage," MIS Quarterly 24(4), 665-694. Aggarwal, P. and Vaidyannathan, R. (2003). "Eliciting online customers' preferences: Conjoint vs self-explicated attribute-level measurements," Journal of Marketing Management 19(1/2), 157-177. Ajzen, I. (1991). "The theory of planned behavior," Organizational Behavior and Human Decision Processes 50(2), 179-211. Alba, J., Lynch, J., Weitz, B., Janiszewski, C , Lutz, R., Sawyer, A. and Wood, S. (1997). "Interactive home shopping: Consumer, retailer, and manufacturer incentives to participate in electronic marketplaces," Journal of Marketing 61(July), 38-53. Allport, G. W. (1954). The nature of prejudice. Addison-Wesley, Cambridge, MA, USA. Andreasen, A. R. (1968). Attitudes and customer behavior: A decision model. Perspectives in consumer behavior. H. H. Kassarjian and T. S. Robertson, Eds. Scott, Foresman and Company, Glenview, IL, 498-510. Ansari, A., Essegaier, S. and Kohli, R. (2000). "Internet recommendation systems," Journal of Marketing Research 37(August), 363-375. Argyle, M. (1988). Bodily communication (2nd ed.). International Universities Press, New York, NY, USA. Ariely, D., Lynch Jr., J. G. and Aparicio IV, M. (2004). Which intelligent agents are smarter? An analysis of relative performance of collaborative and individual based recommendation agents. Working paper, Massachusetts Institute of Technology, Cambridge, MA, USA. Babin, B. J., Darden, W. R. and Griffen, M. (1994). "Work and/or fun: Measuring hedonic and utilitarian shopping value," Journal of Consumer Research 20(4), 644-656. Badler, N. I. (1997). Real-time virtual humans. In Proceedings of the IEEE - The Fifth Pacific Conference on Computer Graphics and Applications, 4-13. Baker, M. J. and Churchill Jr., G. A. (1977). "The impact of physically attractive models on advertising evaluations," Journal of Marketing Research 14(4), 538-555. Baldwin, M. W. (1992). "Relational schemas and the processing of social information," Psychology Bulletin 112(3), 461-484. Barclay, D., Thompson, R. and Higgins, C. (1995). "The Partial Least Squares (PLS) approach to causal modeling: Personal computer adoption and use an illustration," Technology Studies 2(2), 285-309. Bartneck, C. and Reichenbach, J. (2005). "Subtle emotional expressions of synthetic characters," International Journal of Human-Computer Studies 62(2), 179-192. Bassett, R. E. (1979). "Effects of source attire on judgments of credibility," Central States Speech Journal 30, 282-285. 144 References Baumeister, R. F. and Leary, M. R. (1995). "The need to belong: Desire for interpersonal attachments as a fundamental human motivation," Psychology Bulletin 117(3), 497-529. Baylor, A. L. (2005). The impact of pedagogical agent image on affective outcomes. In Proceedings of the International Conference on Intelligent User Interface, Workshop on Affective Interactions: Computers in the Affective Loop, San Diego, CA. Baylor, A. L. and Ryu, J. (2003). "The effects of image and animation in enhancing pedagogical agent persona," Journal of Educational Computing Research 28(4), 373-394. Bern, S. L. (1993). The lenses of gender: Transforming the debate on sexual inequality. Yale University Press, New Haven, Connecticut, USA. Berry, D. C , Butler, L. T. and de Rosis, F. (2005). "Evaluating a realistic agent in an advice-giving task," International Journal of Human-Computer Studies 63(3), 304-327. Berry, L. L. (1995). "Relationship marketing of services: Growing interest, emerging perspectives," Journal of the Academy of Marketing Science 23(Fall), 236-245. Berscheid, E. and Walster, E. (1978). Interpersonal attraction. Addison-Wesley, Reading, MA, USA. Beutnagel, M., Conkie, A., Schroeter, J., Stylianou, Y. and Syrdal, A. (1999). The AT&T next-gen TTS system. In Proceedings of the Joint Meeting of ASA, EAA, and DAGA, Berlin, Germany. Bhattacherjee, A. (2002). "Individual trust in online firms: Scale development and initial test," Journal of Management Information Systems 19(1), 211-241. Bickmore, T. and Cassell, J. (2001). Relational agents: A model and implementation of building user trust. In Proceedings of the SIGCHT 01, Seattle, WA, USA, 396-403. Biernat, M. and Vescio, T. K. (1993). "Categorization and stereotyping: Effects of group context on memory and social judgment," Journal of Experimental Social Psychology 29(2), 166-202. Biocca, F. (1997). "The cyborg's dilemma: Progressive embodiment in virtual environment," Journal of Computer Mediated Communication 3(2). Biocca, F., Harms, C. and Burgoon, J. K. (2003). "Towards a more robust theory and measure of social presence: Review and suggested criteria," Presence: Teleoperators and Virtual Environments 12(5), 456-480. Blau, P. (1964). Exchange and power in social life. Wiley, New York, New York, USA. Blythe, M., Hassenzahl, M. and Wright, P. (2004). "More funology," Interactions 11(5), 37. Bond, E. K. (1972). "Perceptions of form by the human infant," Psychology Bulletin 11, 225-245. Brave, S., Nass, C. and Hutchinson, K. (2005). "Computers that care: Investigating the effects of orientation of emotion exhibited by an embodied computer agent," International Journal of Human-Computer Studies 62(2), 161-178. 145 References Brewer, M. B. (1999). "The psychology of prejudice: Ingroup love or outgroup hate?," Journal of Social Issues 55(3), 429-444. Broverman, E. K., Vogel, S. R., Broverman, D. M., Clarkson, F. E. and Rosenkrantz, P. S. (1972). "Sex-role stereotypes: A current appraisal," Journal of Social Issues 28(2), 59-78. Burgoon, J. K. (1994). Nonverbal signals. Handbook of interpersonal communication. M. L. Knapp and G. R. Miller, Eds. Sage Publications, Thousand Oaks, CA, 229-285. Burgoon, J. K., Bonito, J. A., Artemio Ramirez, J., Dunbar, N. E., Kam, K. and Fischer, J. (2002). "Testing the interactivity principle: Effects of mediation, propinquity, and verbal and nonverbal modalities in interpersonal interaction," Journal of Communication 52(September), 657-677. Burgoon, J. K., Stoner, G. M., Bonito, J. A. and Dunbar, N. E. (2003). Trust and deception in mediated communication. In Proceedings of the 36th Hawaii International Conference on System Sciences (HICSS'03), Hawaii, US. Byrne, D. (1971). The attraction paradigm. Academic Press, New York, NY, USA. Byrne, D., Clore, G. and Worchel, P. (1966). "The effect of economic similarity-dissimilarity as determinants of attractions," Journal of Personality and Social Psychology 4(2), 220-224. Cabellero, M. J. and Pride, W. M. (1984). "Selected effects of salesperson sex and attractiveness in direct mail advertisements," Journal of Marketing 48(1), 94-100. Carroll, J. M. (2004). "Beyond fun," Interactions 11(5), 38-40. Carroll, J. M. and McKendree, J. (1987). "Interface design issues for advice-giving expert systems," Communications of the ACM 30(1), 14-32. Cassell, J. and Bickmore, T. (2000). "External manifestations of trustworthiness in the interface," Communications of the ACM 43(12), 50-56. Cassell, J., Bickmore, T., Billinghurst, M., Campbell, L., Chang, K., Vilhjalmsson, H. and Yan, H. (1999). Embodiment in conversational interfaces. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems (CHI'99), Pittsburgh, Pennsylvania, USA, 520-527. Cassell, J., Sullivan, J., Prevost, S. and Churchill, E. (2000). Embodied conversational agents. MIT Press, Cambridge, Massachusetts, USA. Chaiken, S. (1981). Physical appearance and social influence. Physical appearance, stigma, and social behavior: The Ontario symposium volume 3. C. P. Herman, M. P. Zanna and E. T. Higgins, Eds. Lawrence Erlbaum Associates, Hillsdale, New Jersey, USA, 143-177. Chaiken, S. (1984). The heuristic model of persuasion. Social Influence: The Ontario Symposium, Volume 5. M. P. Zanna, J. Olson and C. P. Herman, Eds. Lawrence Erlbaum Associates, Hillsdale, New Jersey, USA, 3-39. Chaiken, S., A. Liberman, et al. (1989). Heuristic and systematic information processing within and beyond the persuasion context. Unintended Thought. J. S. Uleman and J. A. Bargh, Eds. Guilford, New York, NY, USA, 212-252. 146 References Chattopadhyay, A., Dahl, D. W., Ritchie, R. J. B. and Shahin, K. N. (2003). "Hearing voices: The impact of announcer speech characteristics on consumer response to broadcast advertising," Journal of Consumer Psychology 13(3), 198-204. Childers, T. L., Carr, C. L., Peck, J. and Carson, S. (2001). "Hedonic and utilitarian motivations for online retail shopping behavior," Journal of Retailing 77(4), 511-535. Chin, W. (1998). The partial least squares approach to structural equation modeling. Modern methods for business research. G. A. Marcoulides, Ed. Lawrence Erlbaum Associates, Hillsdale, NJ, USA, 295-336. Chopra, K. and Wallace, W. A. (2003). Trust in electronic environment. In Proceedings of the 36th Hawaii International Conference on System Sciences, Big Island, Hawaii. Churchill, G., Collins, R. H. and William, A. S. (1975). "Should retail salespersons be similar to their customers?," Journal of Retailing 51(3), 29-42. Clark, H. H. (1996). Using language. Cambridge University Press, New York, NY, USA. Clark, R. D. and Maass, A. (1988). "The role of social categorization and perceived source credibility in minority influence," European Journal of Social Psychology 18(5), 381-394. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates, Hillsdale, NJ. Corritore, C. L., Kracher, B. and Wiedenback, S. (2003). "On-line trust: Concepts, evolving themes, a model," International Journal of Human-Computer Studies 58(6), 737-758. Cowell, A., J. and Stanney, K. M. (2005). "Manipulation of non-verbal interaction style and demographic embodiment to increase anthropomorphic computer character credibility," International Journal of Human-Computer Studies 62(2), 281-306. Crosby, L. A., Evans, K. and Cowles, D. (1990). "Relationship quality in services selling: An interpersonal influence perspective," Journal of Marketing(Ju\y), 68-81. Crosby, L. A. and Stephens, S. (1987). "Effects of relationship marketing on satisfaction, retention, and prices in the life insurance industry," Journal of Marketing Research 24(4), 404-411. Csikszentmihalyi, M. (1975). Beyond boredom and anxiety. Jossey-Bass, San Francisco, CA, USA. Davis, F. D. (1989). "Perceived usefulness, perceived ease of use and user acceptance of information technology," MIS Quarterly 13(3), 319-339. Davis, F. D., Bagozzi, R. P. and Warshaw, P. R. (1992). "Extrinsic and intrinsic motivation to use computers in the workplace," Journal of Applied Social Psychology 22(14), 1111-1132. De Meuse, K. P. (1987). "A review of the effects of nonverbal cues on the performance appraisal process," Journal of Occupational Psychology 60(3), 207-226. 147 References de Rosis, F., Pelachaud, C , Poggi, I., Carofiglio, V. and De Carolis, B. (2003). "From greta's mind to her face: Modeling the dynamics of affective states in a conversational embodied agent," International Journal of Human-Computer Studies 59(1-2), 81-118. Dehn, D. M. and van Mulken, S. (2000). "The impact of animated interface agents: A review of empirical research," International Journal of Human-Computer Studies 52, 1-22. Dempsey, J. V. and Van Eck, R. (2003). "Modality and placement of a pedagogical adviser in individual interactive learning," British Journal of Educational Technology 34(5), 1-16. Dholakia, R. R. and Chiang, K.-P. (2002). "Shoppers in cyberspace: Are they from Venus or Mars and does it matter?," Journal of Consumer Psychology 13(1&2), 171-176. Diehl, K., Kornish, L. J. and Lynch Jr., J. G. (2003). "Smart agents: When lower search costs for quality information increase price sensitivity," Journal of Consumer Research 30(June), 56-71. Dion, K., Berscheid, E. and Walster, E. (1972). "What is beautiful is good," Journal of Personality and Social Psychology 24(12), 285-290. Dutoit, T. (1997). An introduction to text-to-speech synthesis (text, speech and language technology, v. 3). Kluwer Academic Publishers, Dordrecht. Dwyer, S., Richard, O. and Shepherd, C. D. (1998). "An exploratory study of gender and age matching in the salesperson-prospective customer dyad: Testing similarity-performance predictions," Journal of Personal Selling and Sales Management 18(4), 55-69. Eckman, P. e. (1973). Darwin and facial expression: A century of research in review. Academic Press, New York, NY, USA. Elliot, C , Rickel, J. and Lester, J. C. (1997). Integrating affective computing into animated tutoring agents. In Proceedings of the IJCAI-97 Workshop on Animated Interface Agents: Making them Intelligent, Nagoya, Japan, 113-121. Falbo, T., Hazen, M. D. and Linimon, D. (1982). "The costs of selecting power bases or message associated with the opposite sex," Sex Roles 8, 147-157. Feldman-Summers, S., Montano, D. E., Kasprzyk, D. and Wagner, B. (1980). "Influence attempts when competing views are gender-related: Sex as credibility," Psychology of Women Quarterly 5, 311-320. Fischer, E., Gainer, B. and Bristor, J. (1998). Exploring gendered servicescape. Services capes: The concept of place in contemporary markets. J. Sherry, Ed. NTC Business Books, Lincolnwood, IL. Flynn, L. R. and Goldsmith, R. E. (1999). "A short reliable measure of subjective knowledge," Journal of Business Research 46(1), 57-66. Fogg, B. J. and Nass, C. (1997). "Silicon sycophants: Effects of computers that flatter," International Journal of Human-Computer Studies 46(5), 551-561. 148 References Fogg, B. J. and Tseng, H. (1999). The elements of computer credibility. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems (CHF99), Pittsburgh, Pennsylvania, USA, 80-87. Fontaine, G. (1992). "The experience of a sense of presence in intercultural and international encounters," Presence: Teleoperators and Virtual Environments 1(4), 482-490. Fomell, C. and Larcker, D. (1981). "Evaluating structural equation models with unobservable variables and measurement error," Journal of Marketing Research 18(1), 39-50. Fukuyama, F. (1995). Trust: The social virtues & the creation of prosperity. The Free Press, New York, New York, USA. Gefen, D., Karahanna, E. and Straub, D. W. (2003). "Trust and TAM in online shopping: An integrated model," MIS Quarterly 27(1), 51-90. Gefen, D. and Straub, D. (1997). "Gender differences in perception and adoption of e-mail: An extension to the technology acceptance model," MIS Quarterly 21(4), 389-400. Gefen, D. and Straub, D. (2003). "Managing user trust in B2C e-services," e-Service Journal 2(2), 7-24. Giles, H. (1970). "Evaluative reactions to accents," Educational Review 22, 211-227. Giles, H. and Powesland, P. F. (1975). Speech style and social evaluation. Academic Press, New York, NY, USA. Gobi, C. and Chasaide, A. N. (2003). "The role of voice quality in communicating emotion, mood and attitude," Speech Communication 40(1-2), 189-212. Gong, L., Nass, C , Simard, C. and Takhteyev, Y. (2001). When non-human is better than semi-human: Consistency in speech interfaces. Usability evaluation and interface design: Cognitive engineering, intelligent agents, and virtual reality. M. J. Smith, G. Salvendy, D. Harris and R. Koubek, Eds. Lawrence Erlbaum Associates, Mahwah, New Jersey, USA, 1558-1562. Gregor, S. and Benbasat, I. (1999). "Explanations from intelligent systems: Theoretical foundations and implications for practice," MIS Quarterly 23(4), 497-530. Grenci, R. T. and Todd, P. A. (2002). "Solutions-driven marketing," Communications of the ACM45(3), 65-71. Hair, J. F., Anderson, R. E., Tatham, R. L. and Black, W. C. (1998). Multivariate data analysis. Prentice-Hall, Englewood Cliffs, New Jersey, USA. Haubl, G. and Murray, K. B. (2001). Recommending or persuading? The impact of a shopping agent's algorithm on user behavior. In Proceedings of the EC'01, Tampa, Florida, USA, 163-170. Haubl, G. and Murray, K. B. (2003). "Preference construction and persistence in digital marketplaces: The role of electronic recommendation agents," Journal of Consumer Psychology 13(1&2), 75-91. 149 References Haubl, G. and Trifts, V. (2000). "Consumer decision making in online shopping environments: The effects of interactive decision aids," Marketing Science 19(1), 4-21. Heeter, C. (1995). Communication research on consumer VR. Communication in the age of virtual reality. F. Biocca and M. R. Levy, Eds. Lawrence Erlbaum Associates, Inc., Hillsdale, New Jersey, 191-218. Herlocker, J. L., Konstan, J. A. and Riedl, J. (2000). Explaining collaborative filtering recommendations. In Proceedings of the 2000 ACM conference on Computer supported cooperative work, Philadelphia, Pennsylvania, United States, 241-250. Hertzum, M., Andersen, H. H. K., Andersen, V. and Hansen, C. B. (2002). "Trust in information sources: Seeking information from people, documents, and virtual agents," Interacting with Computers 14(5), 575-599. Hoffman, D. L., Novak, T. P. and Peralta, M. (1999). "Building consumer trust online," Communications of the ACM42(4), 80-85. Hogg, M. and Abrams, D. (1988). Social identification. Routledge, London. Hogg, M. A., Turner, J. C. and Davidson, B. (1990). "Polarized norms and social frames of reference: A test of the self-categorization theory of group polarization," Basic and Applied Social Psychology 11, 77-100. Ichheiser, G. (1970). Appearance and realities: Misunderstanding in human relations. Jossey-Bass, San Francisco, CA, USA. Igbaria, M., Iivari, J. and Maragahh, H. (1995). "Why do individuals use computer technology? A Finnish case study," Information & Management 29(5), 227-238. Isbister, K. and Nass, C. (2000). "Consistency of personality in interactive characters: Verbal cues, non-verbal cues, and user characteristics," International Journal of Human-Computer Studies 53, 251-267. Jacobs, R. S., Hyman, M. R. and McQuitty, S. (2001). "Exchange-specific self-disclosure, social self-disclosure, and personal selling," Journal of Marketing Theory and Practice 9(1), 48-62. Jarvenpaa, S. L. and Todd, P. A. (1997). "Consumer reactions to electronic shopping on the world wide web," International Journal of Electronic Commerce 1(2), 59-88. Jarvenpaa, S. L., Tractinsky, N. and Vitale, M. (2000). "Consumer trust in an internet store," Information Technology and Management 1(1-2), 45-71. Jensen, C , Farnham, S., Drucker, S. and Kollock, P. (2000). The effect of communication modality on cooperation in online environments. In Proceedings of the SIGCHI conference on Human factors in computing systems (CHT00), The Hague, The Netherlands, 470-477. Jilka, M., Syrdal, A. K., Conkie, A. D. and Kapilow, D. A. (2003). Effects on TTS quality of methods of realizing natural prosodic variation. In Proceedings of the 15th International Congress of Phonetic Science (ICPhS 2003), Barcelona, Spain. Johnson, C. A. (2005). The North American consumer: Online retail update, Forrester Research. 150 References Keeling, K., McGoldrick, P., Beatty, S. and Macaulay, L. (2004). Face value? Customer views of appropriate formats for embodied conversational agents (ECAs) in online retailing. In Proceedings of the 37th Hawaii International Conference on System Sciences, Big Island, Hawaii, USA, 178-187. Kim, J. and Moon, J. Y. (1998). "Designing towards emotional usability in customer interfaces - trustworthiness of cyber-banking system interfaces," Interacting with Computers 10(1), 1-29. King, W. J. and Ohya, J. (1996). The representation of agents: Anthropomorphism, agency, and intelligence. In Proceedings of the CHT96, 289-290. Klein, J., Moon, Y. and Picard, R. W. (2002). "This computer responds to user frustration: Theory, design, and results," Interacting with Computers 14(2), 119-140. Koda, T. (1996). Agents with faces: A study on the effects of personification of software agents. Unpublished Master's Thesis, Massachusetts Institute of Technology, Boston, MA, USA. Koda, T. and Maes, P. (1996). Agents with faces: The effects of personification. In Proceedings of the 5th IEEE International Workshop on robot and Human Communication (RO-MAN'96), 189-194. Komiak, S. X. (2003). The impact of internalization and familiarity on trust and adoption of recommendation agents. Unpublished Doctoral Dissertation, The University of British Columbia, Vancouver, BC, Canada. Komiak, S. X. and Benbasat, I. (2004). "Understanding customer trust in agent-mediated electronic commerce, web-mediated electronic commerce, and traditional commerce," Information Technology and Management 5(1&2), 181-207. Komiak, S. X., Wang, W. and Benbasat, I. (2004/2005). "Trust building in virtual salespersons versus in human salespersons: Similarities and differences," e-Service Journal 3(3), 49-63. Komiak, S. X., Wang, W. and Benbasat, I. (2005). Comparing customer trust in virtual salespersons with customer trust in human salesperson. In Proceedings of the 38th Hawaii International Conference on System Sciences, Big Island, Hawaii. Koufaris, M. (2002). "Applying the technology acceptance model and flow theory to online consumer behavior," Information Systems Research 13(2), 205-233. Kumar, N. and Benbasat, I. (2004). The impact of personalization and virtual communities on customers' evaluations of a website. Unpublished Doctoral Dissertation, The University of British Columbia, Vancouver, BC, Canada. Laurel, B. (1990). Interface agents: Metaphors with character. The art of human-computer interface design. B. Laurel, Ed. Addison-Wesley, Reading, MA, USA. Lee, E.-J. (2003). "Effects of "gender" of the computer on informational social influence: The moderating role of task type," International Journal of Human-Computer Studies 58(4), 347-362. 151 References Lee, E.-J. and Nass, C. (2002). "Experimental tests of normative group influence and representation effects in computer-mediated communication: When interacting via computers differs from interacting with computers," Human Communication Research 28(3), 349-381. Lee, E.-J., Nass, C. and Brave, S. (2000). Can computer-generated speech have gender? An experimental test of gender stereotype. In Proceedings of the ACM SIGCHI conference on Human factors in computing systems (CHI'00), The Hague, The Netherlands, 289-290. Lee, K. and Nass, C. (2003). Designing social presence of social actors in human computer interaction. In Proceedings of the ACM SIGCHI conference on Human factors in computing systems (CHI'03), Ft. Lauderdale, Florida, USA, 289-296. Lee, K. M. (2004). "Why presence occurs: Evolutionary psychology, media equation, and presence," Presence: Teleoperators and Virtual Environments 13(4), 494-505. Lee, K. M. and Nass, C. (2004). "The multiple source effect and synthesized speech: Doubly-disembodied language as a conceptual framework," Human Communication Research 30(2), 182-207. Lee, M. K. O., Cheung, C. M. K. and Chen, Z. (2005). "Acceptance of internet-based learning medium: The role of extrinsic and intrinsic motivation," Information & Management 42(8), 1095-1104. Lee, M. K. O. and Turban, E. (2001). "A trust model for consumer internet shopping," International Journal of Electronic Commerce 6(1), 75-91. Lee, S. and Dubinsky, A. J. (2003). "Influence of salesperson characteristics and customer emotion on retail dyadic relationships," International Review of Retail, Distribution & Consumer Research 13(1), 21-36. Lee, W.-P., Liu, C.-H. and Lu, C.-C. (2002). "Intelligent agent-based systems for personalized recommendations in internet commerce," Expert Systems with Applications 22(4), 275-284. Lee, Y. T. (1993). "In-group preference and homogeneity among African American and Chinese American students," Journal of Social Psychology 133(2), 225-235. Leigh, T. W. and Summers, J. O. (2002). "An initial evaluation of industrial buyers' impressions of salespersons' nonverbal cues," Journal of Personal Selling and Sales Management 22(1), 41-53. Lester, J. C, Converse, S. A., Kahler, S. E., Barlow, S. T., Stone, B. A. and Bhogal, R. S. (1997). The persona effect: Affective impact of animated pedagogical agents. In Proceedings of the SIGCHI conference on Human factors in computing systems, Atlanta, Georgia, United States, 359 - 366. Lester, J. C. and Stone, B. A. (1997). Increasing believability in animated pedagogical agents. In Proceedings of the 1st International Conference on Autonomous Agents, Marina Del Rey, CA, USA, 16-21. 152 References Li, D., Chau, P. Y. K. and Lou, H. (2005). "Understanding individual adoption of instant messaging: An empirical investigation," Journal of the Association for Information Systems 6(4), 102-129. Link, K. E., Kreuz, R. J., Graesser, A. C. and Group, T. T. R. (2001). "Factors that influence the perception of feedback delivered by a pedagogical agent," International Journal of Speech Technology 4(2), 145-153. Linville, P. and Jones, E. (1980). "Polarized appraisals of outgroup members," Journal of Personality and Social Psychology 38(5), 689-703. Lombard, M. and Ditton, T. (1997). "At the heart of it all: The concept of presence," Journal of Computer Mediated Communication 3(2). Luhmann, N. (1979). Trust and power. John Wiley and Sons, London. Lynch, J. and Schuler, D. (1994). "The matchup effect of spokesperson and product congruency: A schema theory interpretation," Psychology & Marketing 11, 417-445. Maes, P. (1994). "Agents that reduce work and information overload," Communications of the ACM37(7), 31-40. Maes, P. (1999). "Smart commerce: The future of intelligent agents in cyberspace," Journal of Interactive Marketing 13(3), 66-76. Maes, P., Guttman, R. H. and Moukas, A. G. (1999). "Agents that buy and sell," Communications of the ACM 42(3), 81-91. Malone, T. W. (1982). Heuristics for designing enjoyable user interface: Lessons from computer games. In Proceedings of the 1982 conference on Human factors in Computing Systems, Washington D.C., 63-68. Massaro, D. W. (1998). Perceiving talking faces. The MIT Press, Cambridge, Massachusetts, USA. Masuda, M., Hasegawa, R. and Matsumoto, G. (1973). "The ethnic identity questionnaire: A comparison of three Japanese age groups in Tachikawa, Japan, Honolulu, and Seattle," Journal of Cross-Cultural Psychology 4(2), 229-244. Mathwick, C. and Rigdon, E. (2004). "Play, flow, and the online search experience," Journal of Consumer Research 31(2), 324-332. Mayer, R. C. and Davis, J. H. (1995). "An integrative model of organizational trust," Academy of Management Review 20(3), 709-734. Mayer, R. C , Davis, J. H. and Schoorman, F. D. (1995). "An integrative model of organizational trust," Academy of Management Review 20(3), 709-734. Mayer, R. E. (2001). Multimedia learning. Cambridge University Press, New York. Mayer, R. E., Johnson, W. L., Shaw, E. and Sandhu, S. (2006). "Constructing computer-based tutors that are socially sensitive: Politeness in educational software," International Journal of Human-Computer Studies 64(1), 36-42. Mayer, R. E., Sobko, K. and Mautone, P. D. (2003). "Social cues in multimedia learning: Role of speaker's voice," Journal of Educational Psychology 95(2), 419-425. 153 References McBreen, H. M. and Jack, M. A. (2001). "Evaluating humanoid synthetic agents in e-retail applications," IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans 31(5), 394-405. McDaniel, S. R. (1999). "An investigation of match-up effects in sport sponsorship advertising: The implications of consumer advertising schemas," Psychology & Marketing 16(2), 163-184. McKnight, D. H. (2002). "The impact of initial consumer trust on intentions to transact with a web site: A trust building model," Journal of Strategic Information Systems 11(3-4), 297-323. McKnight, D. F£., Choudhury, V. and Kacmar, C. (2002). "Developing and validating trust measures for e-commerce: An integrative typology," Information Systems Research 13(3), 334-359. Messick, D. M. and Mackie, D. M. (1989). "Intergroup relations," Annual Review of Psychology 40, 45-81. Meyerson, D., Weick, K. E. and Kramer, R. M. (1996). Swift trust and temporary groups. Trust in organizations: Frontiers of theory and research. R. M. Kramer and T. R. Tyler, Eds. Sage, Thousand Oaks, California, USA, 261-287. Moon, J.-W. and Kim, Y.-G. (2001). "Extending the TAM for a World-Wide-Web context," Information & Management 38(4), 217-230. Moon, Y. (2000). "Intimate exchanges: Using computers to elicit self-disclosure from consumer," Journal of Consumer Research 26(4), 323-339. Moreno, R., Mayer, R. E. and Lester, J. C. (2000). Life-like pedagogical agents in constructivist multimedia environments: Cognitive consequences of their interaction. In Proceedings of the ED-MEDIA 2000, Montreal, Canada. Moreno, R., Mayer, R. E., Spires, H. A. and Lester, J. C. (2001). "The case for social agency in computer-based teaching: Do students learn more deeply when they interact with animated pedagogical agents?," Cognition and Instruction 19(2), 177-213. Morgan, R. M. and Hunt, S. D. (1994). "The commitment-trust theory of relationship marketing," Journal of Marketing 58(July), 20-38. Moundridou, M. and Virvou, M. (2002). "Evaluating the persona effect of an interface agent in a tutoring system," Journal of Computer Assisted Learning 18(3), 253-261. Nagao, K. and Takeuchi, A. (1994). Social interaction: Multimodal conversation with social agents. In Proceedings of the Twelfth National Conference on Artificial Intelligence (AAAI'94), Seattle, Washington, USA, 22-28. Nass, C. and Gong, L. (1999). Maximized modality or constrained consistency? In Proceedings oftheAVSP'99, Santa Cruz, CA, USA, 1-5. Nass, C. and Gong, L. (2000). "Speech interfaces from an evolutionary perspective," Communications of the ACM 43(9), 36-43. 154 References Nass, C , Isbister, K. and Lee, E.-J. (2000). Truth is beauty: Researching embodied conversational agents. Embodied conversational agents. J. Cassell, J. Sullivan, S. Prevost and E. Churchill, Eds. MIT Press, 374-402. Nass, C , Kim, E.-Y. and Lee, E.-j. (1998). When my face is the interface: An experimental comparison of interacting with one's own face or someone else's face. In Proceedings of the CHI'98, Los Angeles, CA, USA, 148-154. Nass, C. and Lee, K. (2000). Does computer-generated speech manifest personality? An experimental test of similarity attraction. In Proceedings of the SIGCHI conference on Human factors in computing systems (CHI'00), The Hague, The Netherlands, 329-336. Nass, C. and Lee, K. (2001). "Does computer-synthesized speech manifest personality? Experimental tests of recognition, similarity-attraction, and consistency-attraction," Journal of Experimental Psychology: Applied 7(3), 171-181. Nass, C. and Moon, Y. (2000). "Machines and mindlessness: Social responses to computers," Journal of Social Issues 56(1), 81-103. Nass, C, Moon, Y., Fogg, B. J., Reeves, B. and Dryer, D. C. (1995). "Can computer personalities be human personalities," International Journal of Human-Computer Studies 43(2), 223-239. Nass, C , Moon, Y. and Green, N. (1997). "Are computers gender-neutral? Gender stereotypic responses to computers," Journal of Applied Social Psychology 27(10), 864-876. Nass, C , Robles, E., Heenan, C , Bienstock, H. and Treinen, M. (2003). "Speech-based disclosure systems: Effects of modality, gender of prompt, and gender of user," International Journal of Speech Technology 6(2), 113-121. Nass, C. and Scott, B. (2005). Wired for speech: How voice activates and advances the human-computer relationship. MIT Press, Cambridge, MA, USA. Nass, C. and Steuer, J. (1993). "Voices, boxes, and sources of messages: Computers and social actors," Human Communication Research 19(4), 504-527. Norman, D. A. (1994). "How might people interact with agents?," Communications of the ACM 31(1), 68-71. Nowak, K. L. and Biocca, F. (2003). "The effect of the agency and anthropomorphism on user's sense of telepresence, copresence, and social presence in virtual environments," Presence: Teleoperators and Virtual Environments 12(5), 481-494. Olson, E. L. and Widing II, R. E. (2002). "Are interactive decision aids better than passive decision aids? A comparison with implications for information providers on the internet," Journal of Interactive Marketing 16(2), 22-33. O'Neal, G. S. and Lapitsky, M. (1991). "Effects of clothing as nonverbal communication on credibility of the message source," Clothing and Textiles Research Journal 9(1), 28-34. 155 References Pavlou, P. A. (2003). "Consumer acceptance of electronic commerce: Integrating trust and risk with the technology acceptance model," International Journal of Electronic Commerce 7(3), 101-134. Pedhazur, E. J. (1982). Multiple regression in behavioral research. Holt, Rinehart and Winston, New York, NY, USA. Pereira, R. E. (2000). "Optimizing human-computer interaction for the electronic commerce environment," Journal of Electronic Commerce Research 1(1), 23-44. Pereira, R. E. (2001). "Influence of query-based decision aids on consumer decision making in electronic commerce," Information Resource Management Journal 14(1), 31-48. Petty, R. E. and J. T. Cacioppo (1986). Communication and persuasion: Central and peripheral routes to attitude change. Springer-Verlag, New York, NY, USA. Pittam, J. (1994). Voice in social interaction: An interdisciplinary approach. Sage Publications, Thousand Oaks, CA, USA. Pittam, J. and Scherer, K. R. (1992). The encoding of affect: A review and directions for further research. In Proceedings of the Fourth Australian International Conference on Speech Science and Technology, Canberra, Australia, 744-749. Pittam, J. and Scherer, K. R. (1993). Vocal expression and communication of emotion. The handbook of emotion. M. Lewis and J. Haviland, Eds. Guilford, New York, NY, USA, 185-197. Podsakoff, P. M. and Organ, D. W. (1986). "Self-report in organizational research: Problems and prospect," Journal of Management 12(4), 531-544. Price, L. L. and Feick, L. F. (1984). The role of interpersonal sources in external search: An informational perspective. Advances in consumer research. T. Kinnear, Ed. Association for Consumer Research, Provo, UT, 250-255. Qiu, L. (2002). How to provide "live help": The effects of text-to-speech voice and 3d avatar on perception of presence in electronic shopping. Unpublished Master's Thesis, The University of British Columbia, Vancouver, BC, Canada. Qiu, L. and Benbasat, I. (2006). "An investigation into the effects of text-to-speech voice and 3d avatars on the perception of presence and flow of live help in electronic commerce," ACM Transactions on Computer-Human Interactions (Forthcoming). Reeves, B. and Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Cambridge University Press, New York, New York, USA. Rempel, J. K., Holmes, J. G. and Zanna, M. P. (1985). "Trust in close relationship," Journal of Personality and Social Psychology 49(1), 95-112. Reynolds, K. E. and Beatty, S. E. (1999a). "Customer benefits and company consequences of customer-salesperson relationships in retailing," Journal of Retailing 75(1), 11-32. 156 References Reynolds, K. E. and Beatty, S. E. (1999b). "A relationship customer typology," Journal of Retailing 75(4), 509-523. Rice, R. E. (1992). "Task analyzability, use of new medium and effectiveness: A multi-site exploration of media richness," Organization Science 3(4), 475-500. Rickenberg, R. and Reeves, B. (2000). The effects of animated characters on anxiety, task performance, and evaluations of user interfaces. In Proceedings of the CHI '2000, The Hague, Amsterdam, 49-56. Riegelsberger, J., Sasse, M. A. and McCarthy, J. D. (2003). Shiny happy people building trust? Photos on e-commerce websites and consumer trust. In Proceedings of the CHI 2003, Ft. Lauderdale, Florida, USA, 121-128. Riker, W. H. (1971). The nature of trust. Perspectives on social power. J. T. Tedeschi, Ed. Aldine, Chicago, IL, USA, 63-81. Rogler, L. H., Cooney, R. S. and Ortiz, V. (1980). "Intergenerational change in ethnic identity in the Puerto Rican family," International Migration Review 14(2), 193-214. Rosenthal, D. and Feldman, S. (1992). "The nature and stability of ethnic identity in Chinese youth: Effects of length of residence in two cultural contexts," Journal of Cross-Cultural Psychology 23(2), 214-227. Rudman, L. A. and Goodwin, S. A. (2004). "Gender differences in automatic in-group bias: Why do women like women more than men like men?," Journal of Personality and Social Psychology 87(4), 494-509. Russo, J. E. (2002). Aiding purchase decisions on the internet. In Proceedings of the Winter 2002 SSGRR (Scuola Superiore Giuseppe Reiss Romoli) International Conference on Advances in Infrastructure for Electronic Business, Education, Science, and Medicine on the Internet, L'Aquila, Italy. Rust, R. T. and Kannan, P. K. (2003). "E-service: A new paradigm for business in the electronic environment," Communications of the ACM46(6), 36-42. Ryan, E. B. and Giles, H. (1982). Attitude towards language variation: Social and applied context. Edward Arnold, London, Britain. Scherer, K. R. (1986). "Vocal affect expression: A review and a model for future research," Psychology Bulletin 99(2), 143-165. Scherer, K. R. (2003). "Vocal communication of emotion: A review of research paradigms," Speech Communication 40(1-2), 227-256. Schwartz, B., Ward, A., Monterosso, J., Lyubomirsky, S., White, K. and Lehman, D. R. (2002). "Maximizing versus satisficing: Happiness is a matter of choice," Journal of Personality and Social Psychology 83(5), 1178-1197. Senecal, S. and Nantel, J. (2004). "The influence of online product recommendations on consumers' online choices," Journal of Retailing 80(2), 159-169. Shang, R.-A., Chen, Y.-C. and Shen, L. (2005). "Extrinsic versus intrinsic motivations for consumers to shop on-line," Information & Management 42(3), 401-413. 157 References Shneiderman, B. and Maes, P. (1997). "Direct manipulation vs. Interface agents," Interactions 4(6), 42-61. Short, J., Williams, E. and Christie,, B. (1976). The social psychology of telecommunication. John Wiley & Sons, London. Simons, H. W., Berkowitz, N. N. and Moyer, R. J. (1970). "Similarity, credibility, and attitude change," Psychology Bulletin 73,1-16. Sinha, R. and Swearingen, K. (2002). The role of transparency in recommender systems. In Proceedings of the ACM Conference on Human Factors in Computing Systems (CHI'02), Minneapolis, Minnesota, USA, 830-831. Sistrunk, F. and McDavid, J. (1971). "Sex variable in conforming behavior," Journal of Personality and Social Psychology 17(1), 200-207. Smith, J. B. (1998). "Buyer-seller relationships: Similarity, relationship management, and quality," Psychology & Marketing 15(1), 3-21. Sproule, S. and Archer, N. (2000). "Knowledgeable agents for search and choice support in e-commerce: A decision support systems approach," Journal of Electronic Commerce Research 1(4), 152-165. Sproull, L., Subramani, M., Kiesler, S., Walker, J. H. and Waters, K. (1996). "When the interface is a face," Human-Computer Interaction 11(1), 97-124. Stephan, W. and Beane, W. E. (1978). "The effects of belief similarity and ethnicity on liking and attributions for performance," Interamerican Journal of Psychology 12(2), 153-159. Straub, D. W. (1994). "The effect of culture on it diffusion: E-mail and fax in Japan and the U.S.," Information Systems Research 5(1), 23-47. Straub, D. W., Limayem, M. and Karahanna-Evaristo, E. (1995). "Measuring system usage: Implications for is theory and testing," Management Science 41(8), 1328-1342. Sundar, S. S. and Nass, C. (2000). "Source orientation in human-computer interaction programmer, networker, or independent social actor?," Communication Research 27(6), 683-703. Swaminathan, V. (2003). "The impact of recommendation agents on consumer evaluation and choice: The moderating role of category risk, product complexity, and consumer knowledge," Journal of Consumer Psychology 13(1&2), 93-101. Swan, J. E. and Bowers, M. R. (1999). "Customer trust in the salesperson: An integrative review and meta-analysis of the empirical literature," Journal of Business Research 44(2), 93-107. Swan, J. E. and Trawick, F. I. (1988). "Measuring dimensions of purchaser trust of industrial salespeople," Journal of Personal Selling and Sales Management 8(1), 1-9. Swearingen, K. and Sinha, R. (2001). Beyond algorithms: An HCI perspective on recommender systems. In Proceedings of the ACM SIGIR Workshop on Recommender Systems. 158 References Sztompka, P. (1999). Trust: A sociological theory. Cambridge University Press, Cambridge, UK. Tajfel, H. (1969). "Cognitive aspects of prejudice," Journal of Social Issues 25(1), 79-97. Tajfel, H. (1970). "Experiments in intergroup discrimination," Scientific American 223, 96-102. Tajfel, H. (1978). Differentiation between social groups: Studies in the social psychology of intergroup relations. Academic Press, London. Tajfel, H. (1981). Human groups and social categories. Cambridge University Press, New York, NY. Tajfel, H. (1982). Social identity and intergroup relations. Cambridge University Press, New York, NY, USA. Tajfel, H. and Turner, J. C. (1986). The social identity theory of intergroup behavior. Psychology of intergroup relations. S. Worchel and W. G. Austin, Eds. Nelson-Hall, Chicago, IL, USA, 7-24. Takeuchi, A. and Naito, T. (1995). Situated facial displays: Towards social interaction. In Proceedings of the CHI 95, New York, NY, USA, 450-455. Takeuchi, Y. and Katagiri, Y. (1999). Social character design for animated agents. In Proceedings of the RO-MAN99, 53-58. Takeuchi, Y., Takahashi, T. and Katagiri, Y. (2000). Life-like agent design based on social interaction. In Proceedings of the 6th Pacific Rim International Conference on Artificial Intelligence, Melbourne, Australia. Taynor, J. and Deaux, K. (1973). "When women are more deserving than men: Equity, attribution and perceived sex difference," Journal of Personality and Social Psychology 28, 360-367. Teo, T. S. H., Lim, V. K. G. and Lai, R. Y. C. (1999). "Intrinsic and extrinsic motivation in internet usage," Omega: International Journal of Management Science 27(1), 127-145. Thorton, G. R. (1944). "The effect of wearing glasses on judgments of personality traits of people seen briefly," Journal of Applied Psychology 28(June), 203-207. Ting-toomey, S. (1981). "Ethnic identity and close friendship in Chinese-American college students," International Journal of Intercultural Relations 5(4), 384-406. Tseng, S. and Fogg, B. J. (1999). "Credibility and computing technology," Communications of the ACM42(5), 39-44. Turkle, S. (1995). Life on the screen: Identity in the age of Internet. Simon & Schuster, New York, NY, USA. Tzeng, J.-Y. (2004). "Toward a more civilized design: Studying the effects of computers that apologize," International Journal of Human-Computer Studies 61(3), 319-345. Ullah, P. (1985). "Second generation Irish youth: Identity and ethnicity," New Community 12(2), 310-320. 159 References Vaidyannathan, R. and Aggarwal, P. (2002). "The impact of shopping agents on small business e-commerce strategy," Journal of Small Business Strategy 13(Spring), 63-79. Van der Heijden, H. (2003). "Factors influencing the usage of websites: The case of a generic portal in the Netherlands," Information & Management 40(6), 541-549. Van der Heijden, H. (2004). "User acceptance of hedonic information systems," MIS Quarterly 28(4), 695-704. van Mulken, S., Andre, E. and Muller, J. (1998). The persona effect: How substantial is it? In Proceedings of the People and Computers XIII: HCI'98, 53-66. Venkatesh, V. (1999). "Creation of favorable user perceptions: Exploring the role of intrinsic motivation," MIS Quarterly 23(2), 239-260. Venkatesh, V. (2000). "Determinants of perceived ease of use: Integrating control, intrinsic motivation, and emotion into the technology acceptance model," Information Systems Research 11(4), 342-365. Venkatesh, V., Morris, M. G., Davis, G. B. and Davis, F. D. (2003). "User acceptance of information technology: Toward a unified view," MIS Quarterly 27(3), 425-478. Venkatesh, V., Speier, C. and Morris, M. G. (2002). "User acceptance enablers in individual decision making about technology: Toward an integrated model," Decision Sciences 33(2), 297-316. Walker, J. H., Sproull, L. and Subramani, R. (1994). Using a human face in an interface. In Proceedings of the CHI'94, Boston, Massachusetts, USA, 85-91. Walther, J. B. (1992). "Interpersonal effects in computer-mediated interaction," Communication Research 19(1), 52-90. Wang, W. (2005). Design of trustworthy online recommendation agents: Explanation facilities and decision strategy support. Unpublished Doctoral Dissertation, The University of British Columbia, Vancouver, BC, Canada. Wang, W. and Benbasat, I. (2004). Impact of explanations on trust in online recommendation agents. Working Paper, The University of British Columbia, Vancouver, BC, Canada. Wang, W. and Benbasat, I. (2005). "Trust in and adoption of online recommendation agents," Journal of the Association for Information Systems 6(3), 72-101. Webster, J., Trevino, L. K. and Ryan, L. (1993). "The dimensionality and correlates of flow in human-computer interactions," Computers in Human Behavior 9(4), 411-426. Westbrook, R. A. and Black, W. C. (1985). "A motivation-based shopper typology," Journal of Retailing 61(1), 78-103. Wu, I.-L. and Chen, J.-L. (2005). "An extension of trust and TAM model with TPB in the initial adoption of on-line tax: An empirical study," International Journal of Human-Computer Studies 62(6), 784-808. Yi , M. Y. and Hwang, Y. (2003). "Predicting the use of web-based information systems: Self-efficacy, enjoyment, learning goal orientation, and the technology acceptance model," International Journal of Human-Computer Studies 59(4), 431-449. 160 Appendices A P P E N D I C E S Appendix A: Measurement Items Variable Source Item(s) Dependent Variables Social Presence (SP) (Gefen and Straub 2003) SP1:1 felt a sense of human contact in the agent. SP2:1 felt a sense of personalness in the agent. SP3:1 felt a sense of human warmth in the agent. SP4:1 felt a sense of sociability in the agent. SP5:1 felt a sense of human sensitivity in the agent. Trusting Beliefs - Competence (TCA) (Wang and Benbasat 2005) TCA1: The agent was competent in recommending digital cameras. TCA2: The agent performed its role of recommending digital cameras very effectively. TCA3: Overall, the agent was capable of providing suitable digital camera recommendations. TCA4: In general, the agent was verv knowledgeable about digital cameras. Trusting Beliefs - Benevolence (TBA) (Wang and Benbasat 2005) TBA1:1 believe that the agent's dealings with me were in my best interest. TBA2: The agent's dealings with me felt like that it would do its best to help me. TBA3: The agent's dealings with me felt like that it was interested in my well-being, not someone else's. Trusting Beliefs - Integrity (TIA) (Wang and Benbasat 2005) TIA1:1 believe the agent's recommendations to me were truthful. TIA2:1 would characterize the agent's dealings with me as honest. TIA3: The agent appeared to be unbiased. TIA4: The agent dealt with me in an obiective fashion. Perceived Enjoyment (PE) (Koufaris 2002) PE1: Interacting with the recommendation agent is enjoyable. PE2: Interacting with the recommendation agent is exciting. PE3: Interacting with the recommendation agent 161 Appendices Variable Source Item(s) is pleasant. PE4: Interacting with the recommendation agent is interesting. PE5: Interacting with the recommendation agent is fun. Perceived Usefulness (PU) (Davis 1989) PUA1: Using the recommendation agent can improve my shopping performance. PUA2: Using the recommendation agent can increase my shopping productivity. PUA3: Using the recommendation agent can increase my shopping effectiveness. PUA4:1 found using the recommendation agent useful. Perceived Ease of Use (PEOU) (Davis 1989) PEOU1: M Y interaction with the agent is clear and understandable. PEOU2: Interacting with the agent does not require a lot of my mental effort. PEOU3:1 find the agent easv to use. PEOU4:1 find it easv to get the agent to do what I want it to do. Agent Adoption Intentions (AAI) (Davis 1989) AAI1: If I have access to the svstem. I intend to use the agent for my next digital camera purchase. AAI2: If I have access to the svstem, I predict I would use the agent for my next digital camera purchase. AAI3: If I have access to the svstem. I plan to use the agent for my next digital camera purchase. Control Variables Disposition to Trust-Competence (DTC) (McKnight et al. 2002) DTC1:1 believe that most professional people do a very good job at their work. DTC2: Most professionals are very knowledgeable in their chosen field. DTC3: A large maioritv of professional people are competent in their area of expertise. Disposition to Trust-Benevolence (McKnight et al. 2002) DTB1: In general, people really do care about the well-being of others. DTB2: The typical person is sincerely concerned Appendices Variable Source Item(s) (DTB) about the problems of others. DTB3: Most of the time, people care enough to try to be helpful, rather than just looking out for themselves. Disposition to Trust - Integrity (DTI) (McKnight et al. 2002) DTI1: In general, most folks keep their promises. DTI2:1 think people generally try to back up their words with their actions. DTB: Most people are honest in their dealings with others. Disposition to Trust - Trusting Stance (DTS) (McKnight et al. 2002) DTS 1:1 usually trust people until thev give me a reason not to trust them. DTS2:1 generally give people the benefit of the doubt when I first meet them. DTS3: Mv typical approach is to trust new acquaintances until they prove I should not trust them. Trust Propensity toward Computer (TPC) (Lee and Turban 2001) TPC1: It is easv for me to trust computers. TPC2: Mv tendency to trust computers is high. TPC3:1 tend to trust computers, even if I have little knowledge of them. TPC4: Trusting computers is not difficult. Product Knowledge / Expertise -Subjective (PKS) (Flynn and Goldsmith 1999) PKS1:1 know pretty much about digital cameras. PKS2:1 do not feel very knowledgeable about digital cameras. (R) PKS3: Among mv circle of friends, I'm one of the "experts" on digital cameras. PKS4: Compared to most other people. I know less about digital cameras. (R) PKS5: When it comes to digital cameras. I really don't know a lot. (R) Web-Shopping Risk Attitudes (WRK) (Jarvenpaa et al. 2000) WRK1: In general, I feel safe completing commercial transactions over the Internet. (R) WRK2: Generally speaking, there is too much uncertainty associated with shopping on the Internet. WRK3: Generally speaking, buying on the Internet is more risky compared with other ways of shopping. 163 Appendices Variable Source Item(s) Attitudes towards computers (ATC) (Jarvenpaa et al. 2000) ATC1: Computers make work more interesting. ATC2:1 enjov interacting with computers. ATC3: Working with computers is fun. Comfort with the Internet (CFI) (Komiak2003) CFI1:1 feel comfortable with the Internet. Comfort with Online Shopping (CFO) (Komiak 2003) CFOl: I feel comfortable with online shopping. Product-related Gender Stereotype (GS) Developed for this study GS1: Most technology product experts are males. GS2: For technology products such as digital cameras, the salespersons in a physical store (e.g., Future Shop or Staples) are more likely to be males rather than females. GS3: Usually I will consult my female friends for purchasing advice on technology products such as digital cameras. (R) GS4: My male friends are likely to know more about technology products such as digital cameras than my female friends do. 164 Appendices Appendix B: Interaction Scripts of the Experimental Recommendation Agent 1. W e l c o m e Page : (Welcome) • H i there, thanks for stopping by. M y name is M a x (Maria) and I am your digital camera advisor. I hope that my expertise w i l l help you find the best deal! Before showing you any camera models, I am going to ask you about ten questions in order to better understand your preferences. Based on your individual preferences, I w i l l inspect about 200 camera models and tell you which one I think would best fit your requirements. It's that easy! Ok, i f you are ready for the first question, please click the "begin" button now. 2. F low Control : (Flow Answer) • Please tell me your answer to this question before moving to the next one. (Flow CheckBox) • Please indicate your answer by clicking one or more o f the checkboxes. (Flow Importance) • Please indicate the importance level of this feature in your decision. 3. Sea rch in Progress (after clicking the "submit answer" button): (Search) • Thanks a lot for answering these questions so that I can identify the cameras that best meet your preferences. Just relax for a moment while I work on my recommendations. Thank you for your patience. 4. Recommendat ion Results Page : (ResultYes) • Now, here comes the deal! I recommend the following models, which are most consistent with your preferences. Y o u could check detailed specifications by clicking the image o f a camera or the model name. Please note, i f the fit score is below 100, it usually means that there could be some differences between your requirements and the models in our inventory. A l l the models I have recommended here have at least a fit score of 70. 165 Appendices By the way, you can always change some of your answers or the importance level of some attributes by clicking the "Back to the Questions" button. (ResultNo) • I am sorry; I cannot give you any recommendations because there is no camera model that satisfies most of your preferences. Please go back and revise some of your answers or the importance level of some attributes. Simply click the "Back to the Questions" button. 5. Task completed (after clicking the "add to shopping cart" button) (Complete) • Thank you for shopping with me today! I hope you are now ready to shoot, organize, and share your photos with the rest of the world! See you next time. 166 Appendices Question 1: My first question for you is: how much are you planning to spend on this digital camera? (Explanations) • I am asking this question to understand your price range. Prices of most cameras fall between $200 and $900. Lower cost cameras normally do not have many advanced features; on the other hand, for beginners or occasional users, a simple point-and-shoot model could do a pretty good job on most occasions. Remember, I may recommend cameras to you that are close to your price range if those cameras are a great fit to other preferences that you have specified. Answers: 1. A small amount (under $300) (Response) • Okay, so you are looking for a good deal! Let's see what cameras we can find for you that are priced under $300. 2. A medium amount (between $250 and $650) (Response) • Thank you! There are a variety of cameras with a price range of between $250 and $650. 3. A large amount (over $600) (Response) • Well, it seems that you are looking for a high-end model. I can help you find it. Cameras in this range will cost $600 and higher, however they will have the most versatility and produce pictures of highest quality. 4. No preference (Response) • I see that you have no price preference. Why don't you tell me about other attributes that you are considering in your purchase? 167 Appendices Question 2: My second question is related to brands. Which one or more of the following brands do you prefer? (Explanations) • If you already have some brand names in mind, you could help me do a better job by telling me of your favorites or those you have particular interests in. Please check all brands that you think you may consider. When making recommendations, I will favor these brands that you have specified. Please note that it is possible that some suitable models may be ranked lower if they were not your favorite brands. Answers: 1. I prefer one or more following brand names: Canon; Casio; Contax; Epson; Fujifilm; Hewlett Packard; Kodak; Kyocera; Leica; Konica Minolta; Nikon; Olympus; Panasonic; Pentax; Samsung; Sony (Response) • Excellent! Check those brands that you would consider and I will keep your preference in my mind while I search. 2. No preference. (Response) • Okay. Let's move on to the next question. 168 Appendices Question 3: Next, how far away will most of your subjects be from the camera? (Explanations) • I am asking this question to determine how much optical zoom you will need for your camera. Optical zoom refers to a camera's ability to "zoom in" or magnify the object being photographed. Optical zoom is measured by the maximum number of times that an object can be magnified, such as "2X" or "3X". Please keep in mind that cameras big optical zoom (usually 5X or above) will be significantly more expensive. Also, your choices may be more limited, since only around 30% of cameras have an optical zoom over 3x. Better not to over-estimate your needs. Answers: 1. Immediate vicinity (small room or within 8 ft. away from you) (Response) • Good. Those cameras with below 3X optical zoom will suit you just right. 2. A moderate distance or less (big room or within 15 ft. away from you) (Response) • No problem. I will pay more attention to those models with 3X to 5X optical zoom. 3. Far away (backyard or farther than 15 ft. away from you) (Response) • Right on! I will look for cameras with a zoom capability of 5X or greater! 4. No preference (Response) • Okay, I will use other criteria to determine which cameras are best for you. 169 Appendices Question 4: How will you view or print the pictures you have taken? (Explanations) • Your answer will determine the appropriate resolution level of your camera. Resolution refers to the sharpness of a picture. It is expressed in "megapixels" and one megapixel is one million light-sensing dots. Higher resolution cameras allow larger "photo quality" prints. Remember, more isn't always better when it comes to pixels. Many people end up buying way more than they need. If your main task will be sharing pictures on the web, or making 4-inch by 6-inch prints, a three or four megapixel camera is probably good enough. Answers: 1. Most likely I will view them on computers or print them in sizes of 4" x 6" or 5" x 7" (Response) • Okay, I will put higher priority on cameras with a resolution of up to 4 megapixels. 2. Probably I will print some of them in sizes of at least 5" x 7" or even 8" x 10" (Response) • Great. I will pay more attention to cameras with 5 or 6 megapixels resolution. 3. / may print some of them in large sizes (at least 8" x 10") (Response) • Cool! I will find models with 6 megapixels or above. 4. No preference (Response) • All right, let's move on to other criteria. 170 Appendices Question 5: What size of a digital camera do you like? (Explanations) • Digital cameras range in weight from 100 grams for an ultra-compact to more than 600 grams for a professional Single Lens Reflex style model. The smallest cameras can be stuffed into a tiniest pocket, with the largest ones requiring a camera bag to carry them around. In general, more weight means more features. The smallest cameras usually have a 2X or 3X lens and very limited manual controls. In contrast, the larger models that resemble the traditional Single-Lens-Reflex style film camera could offer full manual controls, support for add-on flashes and sometimes add-on lenses. Answers: 1. Small enough to fit into a shirt pocket (Response) • That's neat. I will focus more on the lightweight models. Most of them would be lighter than 130 grams. Be careful not to lose it! 2. Fit well into a waist pack or purse (Response) • Now that is a convenient size. I will find compact size models that weigh somewhere between 130 and 454 grams. 3. Fit well into luggage or a backpack (Response) • Great! I am sure you will look professional with these Single-Lens-Reflex size models. 4. No preference (Response) • Okay, let's consider other factors. 171 Appendices Question 6: What size of LCD screen on your digital camera do you like? (Explanations) • An LCD screen is very useful if you want to frame your shot without using the optical viewfinder or to review pictures that have been taken. The size of LCD screen is measured in inches diagonally across the screen. Cameras with your desired LCD screen size will be given higher priority in my recommendations. Please note, cameras with very large LCD screen, for example, 2 inches or above, are relatively more expensive. Answers: 1. Small-size L CD screen (Response) • Thank you! I will favor models that have a 1.5 inch LCD. 2. Medium-size LCD screen (Response) • All right, I will look into models whose screen sizes are up to 2 inches. 3. Large-size LCD screen (Response) • No problem. I see that you like to see the big picture! 4. No preference (Response) • Okay, let's move on to another feature. 172 Appendices Question 7: What type of camera flash will you use? (Explanations) • You will definitely need a flash if you want to take pictures in low light conditions. Most cameras nowadays come with at least a simple flash with red-eye reduction features. Some advanced models could be connected to a more powerful external flash through flash synchronization and hot shoe features. Remember, the support of external flash is considered an advanced feature for digital cameras; therefore, only relatively more expensive cameras have this feature. Answers: 1. /just want a basic flash with red-eye removal feature (Response) • I see. On most occasions, a simple flash will do the job. 2. / use external flash for different lighting conditions (Response) • No problem. I will look for cameras with external flash support. 3. No preference (Response) • All right, let's move on to the next criterion. 173 Appendices Question 8: What kind of batteries do you prefer to use for your camera? (Explanations) • Digital cameras use two types of batteries: standard AA or proprietary Lithium-Ion. Lithium-Ion batteries are almost always rechargeable, but for cameras using standard AA batteries quite often it's up to you to buy the rechargeable ones. Please note, proprietary Lithium-Ion batteries often last longer than their AA counterparts and take less time to recharge. However, at roughly $8 for a set of four, standard AA batteries are less expensive than proprietary Lithium-Ion, which often costs $50 or more. Answers: 1. /prefer standard AA batteries (Response) • Please make sure to get Nickel-Metal Hydride (Ni-MH) rechargeable batteries because they could offer decent performance with affordable cost. 2. I prefer proprietary lithium-ion batteries (Response) • I see. You will get what you pay for, Lithium-Ion batteries are generally more powerful and allow more photos per charge. 3. No preference (Response) • Okay, let's move on to the next question. 174 Appendices Question 9: How many manual features would you like to have in your digital camera? (Explanations) • While most point-and-shoot cameras could set the aperture and shutter speed automatically and do a good job most of the time, there are occasions you need manual controls that help you override the default settings and take a well-exposed picture. More advanced models may offer control over shutter speed, aperture, white balance, color, and focus. Answers: 1. / prefer simple point-and-shoot cameras (mostly automatic for ease of use) (Response) • Sure! I will look for a straight forward design that's easy to use. 2. I prefer cameras with some manual features (Response) • You bet. It's always amusing to have more features to alter the effects. I will try to get you models with up to 4 manual features. 3. I prefer cameras with lots of manual features (Response) • Looking for all the bells and whistles, aren't you? Models with 5 manual features or more will have higher priority in my recommendations. 4. No preference (Response) • No problem. Please click "Next Question" to continue. 175 Appendices Question 10: And now it's time for my final question: which additional features would you like to have in your digital camera? (Explanations) • Nowadays digital cameras can do more than taking pictures. For example, a camera with movie feature is an easy way to take short video clips without investing in a camcorder. Some cameras record sound along with video segments while others don't. In addition, cameras with PictBridge support let you output pictures from your digital camera to your printer directly, without using a PC or image-editing software. Also, some cameras support conversion lenses. You can get wide-angle, telephoto, or macro lenses that improve the capabilities of your camera. Answers: 1. I want a camera with following additional features: Movie clip; Sound recording; PictBridge; Conversion lenses (Response) • Thank you. I will keep these additional features in my mind. Please click the "submit answer" button to get my recommendations. 2. No preference. (Response) • That's fine. You will get my recommendations now by clicking the "submit answer" button. 176 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0092949/manifest

Comment

Related Items