annotated bibliography

lady2005

Privacyconcerns.pdf

Home >Engineering homework help >Telecommunications Engineering homework help >annotated bibliography

Privacy Concerns for Use of Voice Activated Personal Assistant in the Public Space

Aarthi Easwara Moorthy and Kim-Phuong L. Vu Department of Psychology, California State University, Long Beach, Long Beach, California, USA

A review of popular technology adoption models identified sev- eral factors that are likely to influence Voice Activated Personal Assistant (VAPA) use in public spaces. To inform design decisions of how to make the private use of the VAPA in public spaces more acceptable from the users’ point of view, an online sur- vey was conducted to investigate the likelihood of usage of the smartphone VAPA such as Apple’s Siri (compared to the usage of smartphone keyboard) as a function of location (private vs. public) and type of information (private vs. nonprivate). Responses from participants showed that users were more cautious of transmitting private than nonprivate information. This effect of type of infor- mation was amplified in the social context of public locations and when using conspicuous methods of information input such as the VAPA. Participants also preferred using the VAPA in private loca- tions and showed no preference of location for keyboard entries. Correlations between likelihood of usage of VAPA and the social acceptability ratings were positive and predicted similar patterns of smartphone usage.

Voice interface is becoming a standard feature for many mobile computing devices. All major mobile phone platforms have introduced a native Voice Activated Personal Assistant (VAPA) feature in their smartphones: Apple—Siri, Google— Google Now, and Samsung—S Voice. Third-party voice assis- tant smartphone applications such as Vlingo, Maluuba, and Evi are also on the rise. These smartphone voice applications are becoming increasingly capable of allowing users to perform simple tasks such as calling a contact, setting a reminder, and sending a text message through voice input. Voice recognition technology is predicted to serve as the default method to control more complex tasks not only in smartphones but also in auto- mobiles and other home appliances (Knight, 2012). Therefore, it is critical to understand the factors that influence the use of voice-activated applications.

Address correspondence to Kim-Phuong L. Vu, Department of Psychology, California State University Long Beach, 1250 N Bellflower Boulevard, Long Beach, CA 90840, USA. E-mail: kim.vu@ csulb.edu

Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/hihc.

Although models of technology adoption exist (e.g., Davis, Bagozzi, & Warshaw, 1989; Venkatesh, Morris, Davis, & Davis, 2003) and have been applied to many domains including the adoption of mobile devices (e.g., Van Biljon & Kotzé, 2007), research on VAPA user preferences has not extensively explored social factors governing its use given that an intelligent voice- assistant application on smartphones is relatively new. There are also unique concerns with using VAPAs. Notably, many users are concerned with the propriety of public use of the VAPA in front of strangers. For example, interacting with the VAPA by talking and issuing voice commands in public might make users believe that they are being watched or judged for their behavior. This mind-set might cause users discomfort when using the VAPA in public. Moreover, while surrounded by oth- ers in a public space, users might not want to do the following: (a) Draw public attention. Setting reminders to buy grocery items while waiting with other patients to see the doctor might draw stares. (b) Disrupt the environment. Asking the VAPA to check game scores in a quiet classroom will most likely dis- turb the peace in a classroom. (c) Intrude the personal space of others. Trying to ask for directions to the nearest coffee shop in a crowded elevator will illustrate this dilemma. (d) Verbalize information of a private nature. For example, users might not want to disclose their credit card number while dictat- ing an e-mail in a quiet restaurant, home address when looking up directions at a supermarket line, or social security num- ber while sending text messages through voice commands in a crowded bus.

The goal of the present study is to gain a basic under- standing of the usage patterns of VAPA in public spaces. To understand the factors influencing VAPA use in public, popular technology adoption models are described in the sub- sequent section to highlight the role of social influence and other potential determinants and moderating elements on the use of VAPA. Then, factors relating to mobile voice service use in public locations and past studies in this area are reviewed to identify the effects of social settings. Finally, an overview of prior studies on privacy concerns due to physical and remote interactions of users with strangers in the mobile phone domain is reviewed prior to presenting the results of an online survey.

307

mailto:kim.vu@csulb.edu

www.tandfonline.com/hihc

308 A. EASWARA MOORTHY AND K.-P. L. VU

1. TECHNOLOGY ADOPTION AND USAGE MODELS To identify factors contributing to the usage of VAPA, espe-

cially in public settings, it is critical to investigate technology adoption literature that can explain predictors leading to usage of mobile phone applications. Many studies have looked at mobile phone acceptance patterns (W. J. Lee, Kim, & Chung, 2002; Roberts & Pick, 2004). However, a literature review on smartphone technology adoption showed that there is a “lack of in depth analysis of factors affecting users’ accep- tance of technology” and a “need for studying user adoption of smartphone in a holistic approach that reflects users’ perspec- tives on smartphone as whole product/services” (Aldhaban, 2012, p. 2764). Therefore, it is necessary to turn to classical well-tested theories on adoption and usage of information sys- tems such as the technology acceptance model (TAM; Davis et al., 1989) and the unified theory of acceptance and use of technology (UTAUT; Venkatesh et al., 2003), which identify potential constructs that predict usage of traditional desktop- based computer technology.

Theories explaining specific technology trends such mobile phone adoption patterns also help understand factors governing VAPA usage. For example, the widespread rise in mobile phone usage has prompted development of more focused models such as the mobile phone technology adoption model (Van Biljon & Kotzé, 2007) that explain factors contributing to mobile phone use. Although these models established important constructs that can predict VAPA use such as social influence, facilitat- ing conditions, and behavioral intention to use, they are yet to be streamlined to predict factors influencing reception of specialized mobile technologies that employ voice interaction. In the following sections, these models are reviewed to under- stand their contributions and limitations in explaining factors that would likely play a role in the public usage of VAPA.

1.1. Technology Acceptance Model The most widely cited model, based on information systems

theory, the TAM, posits that users’ perception of the usefulness and ease of use of a system play an important role in their actual usage of technical systems (Davis et al., 1989). Specifically, it states that external variables, which include user character- istics such as demographic variables, cognitive variables, and

user training; system characteristics such as system design and behavior; and organizational characteristics such as the internal system implementation process and support for users influence the users’ perceptions of how useful and how easy their inter- action will be with the technology. Perceived usefulness and perceived ease of use directly drive the users’ positive or neg- ative attitudes about using the technology, which then impacts users’ conscious behavior to use the technology and their final actions of technology use or rejection. Perceived ease of use also has the ability to affect perceived usefulness, which in turn can directly impact behavioral intention. This pathway to technology use is illustrated in Figure 1.

Numerous studies have been conducted to validate the effects of these factors on system use. For example, as proposed by the TAM, studies have found perceived ease of use to influence perceived usefulness (Szajna, 1996; S. Taylor & Todd, 1995). Perceived usefulness (Agarwal & Prasad, 1999; Mathieson, 1991) and perceived ease of use (Mathieson, 1991; S. Taylor & Todd, 1995) have an effect on attitude. Perceived usefulness has been found to be a determinant of behavioral intention (Agarwal & Prasad, 1999; Igbaria, Zinatelli, Cragg, & Cavaye, 1997). Attitude also impacts behavioral intention (Davis et al., 1989; Mathieson, 1991), which in turn predicts use (Davis et al., 1989; S. Taylor & Todd, 1995; Venkatesh & Davis, 2000).

The key construct identified by the TAM that is important for this thesis is the behavioral intention to use the technical system. This construct presents a way to predict actual usage by asking potential users of a technical system (such as the VAPA) their likelihood of using it. However, one of the main problems of the TAM is that it does not look at specific external variables such as social factors or type of information to be processed that can influence the critical constructs of perceived usefulness and ease of use. Malhotra and Galletta (1999) suggested that “atti- tude towards adopting a technology is believed to be the result of personal and social influences, and the fact that TAM does not account for social influence is a limitation” (Van Biljon & Kotzé, 2007, p. 115). For the purpose of this study, the construct of social influence is a key factor that might play a role in the public use of the VAPA. Specifically, it is critical to examine the effect of social settings on VAPA usage. Another shortcoming is that the TAM presents no specific construct to incorporate the type of information to be verbally processed in the VAPA.

FIG. 1. Illustration of the technology acceptance model.

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 309

1.2. Unified Theory of Acceptance and Use of Technology Another popular technology adoption model, the UTAUT,

which is based on motivational and behavioral models in psy- chology and technology, furthers the TAM by specifying the external variables of social influence and facilitating conditions as main determinants of system use, in addition to perceived usefulness and perceived ease of use. It also proposes the fol- lowing set of moderators that act on these determinants: gender, age, experience, and voluntariness of use.

According to UTAUT, the determinants of performance expectancy (the perceived usefulness of the system), effort expectancy (the perceived ease of use), and social influence (“the degree to which an individual perceives that important others believe he or she should use the new system”) drive the behavioral intention to use the technology, which directly influences the users’ final use or rejection of the technology (Venkatesh, 2013). The fourth determinant of facilitating con- ditions (“the degree to which an individual believes that an organizational and technical infrastructure exists to support use of the system”) influences only the users’ final use or rejec- tion of the technology (Venkatesh, 2013). The moderators of gender, age, experience, and voluntariness of use (“the extent to which potential adopters perceive the adoption decision to be non-mandatory”) each influence specific determinants (Venkatesh, 2013). Gender influences performance expectancy, effort expectancy, and social influence. Age has an impact on all the determinants. Experience influences effort expectancy, social influence, and facilitating conditions. Finally, voluntari- ness of use affects only social influence. The technology use pathway proposed by the UTAUT model is shown in Figure 2.

Prior studies have shown results consistent with the UTAUT model. They showed that intention to use technology is influ- enced by performance expectancy (Chang, Hwang, Hung, & Li, 2007; Laumer, Eckhardt, & Trunk, 2010), effort expectancy (Gupta, Dasgupta, & Gupta, 2008; Hung, Wang, & Chou, 2007), social influence (Hung et al., 2007; Laumer et al., 2010),

and facilitating conditions (Laumer et al., 2010). Studies have always found that user adoption can be directly driven by per- formance expectancy (Chang et al., 2007; T. Zhou, Lu, & Wang, 2010), effort expectancy (Chang et al., 2007), social influence (T. Zhou et al., 2010), and facilitating conditions (Gupta et al., 2008; T. Zhou et al., 2010). For a complete documentation of findings from all studies that have used the UTAUT, refer to Williams, Rana, Dwivedi, and Lal (2011).

For the purpose of exploring factors influencing public VAPA use, several elements identified by the UTAUT are important. Particularly, the incorporation of social influence highlights the possibility that presence of company could influ- ence public use of VAPA. The unique addition of facilitating conditions indicates that factors such as availability and quality of mobile phone services and devices, respectively, could influ- ence use of VAPA. The construct of behavioral intention to use technical systems (recognized in the TAM) is also retained in the UTAUT.

However, the UTAUT might have limited ability to explain public usage of the VAPA. The UTAUT has been primar- ily applied to study adoption of general stationary computing systems in the domains of communication systems, general- purpose systems, office systems, and specialized business sys- tems (Williams et al., 2011). The infrastructure within which these stationary systems operate is different from the mobile framework in which voice applications such as the smartphone VAPA is used. For example, unlike stationary systems, mobile phones are used under ever-changing physical, social, and tech- nological contexts as proposed by Van Biljon and Kotzé (2007). Therefore, the role of some determinants, such as facilitating conditions, might be more pronounced in the mobile context.

The UTAUT also does not account for the more impactful elements of social influence and type of information transmitted in the mobile context that could affect public usage of VAPA. Information verbalized by mobile phone users is usually more salient to others in their surroundings, compared to actions of

FIG. 2. Sample illustration of the unified theory of acceptance and use of technology. Note. The bars of age, gender, and experience overlap the factors that the variables influence (based on the description provided by Van Biljon & Kotzé, 2007).

310 A. EASWARA MOORTHY AND K.-P. L. VU

traditional stationary computer system users. Thus, the type of information to be transmitted might affect decisions to use or reject this technology. Another implication of this saliency is that perceived attitudes and beliefs of those in their colocated space might have a greater potential to alter mobile phone user preferences compared to stationary computer users. Therefore, the social influence of people in the colocated space might not affect users of traditional stationary computer systems to the degree or in the manner they affect mobile phone users. Thus, there is a need to study factors, especially the roles of social influence and facilitating conditions that can govern adoption of mobile phone services, which might be different from those identified by the TAM and the UTAUT.

1.3. Mobile Phone Technology Acceptance Model The mobile phone technology acceptance model

(MOPTAM) was proposed in order to present a unified theory addressing factors specifically affecting mobile phone use. This model was referred because the main user action with both the VAPA and mobile phone voice calls is verbalizing information. Because of this similarity, the MOPTAM might be able to better predict usage patterns for the VAPA than general technology models such as the TAM and the UTAUT.

Ruuska-Kalliokulju, Schneider-Hufschmidt, Väänänen- Vainio-Mattila, and Von Niman (2001) noted the following reasons that make the mobile context different from the settings in which general, stationary, office-based technology is used. The physical, social, and cultural contexts strongly influence the method of user interaction with the mobile phone interface. Mobile phones can be used anywhere and anytime, unlike stationary computers. The needs of the users change according to the context of mobile phone use, and mobile services and applications are designed to fulfill these needs. Users have to deal with increasingly task-specific devices that communicate with each other.

In addition to the aforementioned justifications, several stud- ies on mobile applications have shown that the original rela- tionships of constructs underlined in the TAM and UTAUT may not be applicable to the mobile context. For example, W. J. Lee et al. (2002) found that social influence has a direct effect on the perceived usefulness and perceived ease of use, which in turn drive actual usage. Roberts and Pick (2004) showed that mobile-specific infrastructural factors or facilitating conditions such as technology-related variables (security, reliability, digi- tal standards, and web connectivity) and nontechnology-related factors (customer service) played a role in corporate adoption of mobile phones. Note that the original TAM did not specify the external variable affecting perceived usefulness and perceived ease of use as social influence, and the UTAUT proposes that social influence affects only the behavioral intention to use a system.

Thus, contextual differences in the case of mobile phone usage justified the development of a model specifically

predicting acceptance for mobile phones. Van Biljon and Kotzé (2007) incorporated these findings to develop the MOPTAM. If mobile phone technology is extended to include mobile phone applications such as the VAPA, then, according to the MOPTAM, social influence drives perceived usefulness and perceived ease of use. Perceived usefulness and perceived ease of use impact behavioral intention, which shape the use of a mobile application such as the VAPA. Perceived ease of use also affects perceived usefulness. Facilitating conditions, which refer to mobile infrastructure variables (i.e., the availability of the VAPA for tasks, accuracy and adequacy of the VAPA’s answers, and time taken for VAPA to answer queries), have an effect on perceived usefulness, perceived ease of use, behavioral intention, and use. Personal factors, which “refer to personal preference and user’s beliefs about the benefit of technology” (such as a user’s belief that she might be viewed as a “geek” for using VAPA in public); demographic factors or “variables like age, gender, education and technological advancement” (such as the user’s knowledge that Siri can display movie show times by zip code without having to open a web browser); and socioeconomic factors, which include “variables like job sta- tus, occupation and income,” act as mediators (Van Biljon & Kotzé, 2007, p. 157). These mediating elements of personal factors, demographic factors, and socioeconomic factors influ- ence the determinants of social influence, perceived usefulness, perceived ease of use, facilitating conditions, and behavioral intention to use mobile phone. The MOPTAM pathway is shown in Figure 3.

The MOPTAM is particularly relevant to the present the- sis because it presents a potential framework that can explain predictors contributing to usage of mobile phone applications such as the VAPA. Restructuring the pathway for the mobile context by positioning social influence as the precedent of per- ceived usefulness and ease of use, and amplifying the reach of facilitating conditions to have an effect of perceived usefulness, perceived ease of use, behavioral intention, and actual use are important contributions of the MOPTAM that allow it to poten- tially better predict VAPA usage. In addition, it also retains from the UTAUT, the critical attribute of behavioral intention to use systems.

However, the MOPTAM has not been extensively validated. Although it has been shown to explain mobile phone adoption pattern in demographic groups such as the elderly (Renaud & Van Biljon, 2008), it is yet to be tested in depth for factors gov- erning usage of different types of mobile applications. Like the other general technology adoption models (the TAM and the UTAUT), it also does not specifically take into consideration the type of verbal information to be processed, a crucial aspect unique to the mobile phone context.

1.4. Summary Technology adoption models have identified several impor-

tant constructs that could predict usage patterns of VAPA,

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 311

FIG. 3. Sample illustration of the mobile phone technology acceptance model (based on the description provided by Van Biljon & Kotzé, 2007).

especially in public settings. The TAM establishes the construct of behavioral intention to use technical systems. The UTAUT specifies social influence and facilitating conditions as exter- nal variables that could impact behavioral intention. Finally, the MOPTAM’s unique contribution is to restructure the path- ways to general technology usage from the TAM and UTAUT into a framework for explaining usage of mobile phone services by positioning social influence as the precedent of all factors. Across the models, social influence, facilitating conditions, and behavioral intention to use were the most important elements that can help predict actual usage of the VAPA in public settings. To provide an understanding of the user preferences for public VAPA use, the next section covers research on publicly using mobile voice services, which also require users to verbalize information like with the VAPA.

2. USING MOBILE VOICE SERVICES AND VOICE APPLICATIONS IN PUBLIC SETTINGS

There has been widespread interest in evaluating the effects of publicly talking on a mobile phone on others colocated in the public space. Studies have concluded that cell phone use in public for voice calls is annoying and intrusive for those in the colocated space (Lasén, 2006; Monk, Carroll, Parker, & Blythe, 2004; Wei & Leung, 1999). Previous research has iden- tified many factors such as location, time, user characteristics, environmental, and cultural factors that influence talking on the mobile phone, particularly in social or public settings.

2.1. Location and Time One of the major advantages with mobile phones is that they

present users with the opportunity to use them anywhere and at anytime. However, talking on the phone in certain locations and times might be inappropriate. In a study on cell phone use in public, Wei and Leung (1999) reported that approximately 80% of 834 participants from Hong Kong found cell phone use in

restaurants, libraries, airports, and train stations irritating. In the United States, public signs strongly encourage users to silence or even turn off their mobile phones in theaters, restaurants, hospitals, classrooms, and office spaces. Campbell and Russo (2003) showed that participants in the United States were par- ticularly appalled by the use of mobile phones in classrooms and movie theaters.

2.2. Cultural Factors Although many studies agree on a global trend in the rejec-

tion of mobile phone use in common public spaces, only a few studies have also observed attitudinal differences between cultures on the acceptability of public mobile phone use (Campbell, 2007; Ito & Okabe, 2005). For example, Ito and Okabe (2005) tracked the adoption of the social norm “no voice, e-mail okay” in public transportation in Japan (Campbell, 2007, p. 740). They argued that the widespread use of mobile phones for social connectivity, its rampant use by Japanese teens, and availability of e-mailing through mobile Internet as an alterna- tive for voice calls led the Japanese to adopt a policy in public transportation such as commuter trains, where the use of mobile phones for e-mails is acceptable but voice calls is discouraged; voice calling is discouraged through public announcements and social pressure of others who might nonverbally signal the user to not talk. This social norm is different from that of the United States, which does not place any formal or informal restrictions for taking voice calls in crowded public ground transportation, unless the mobile phone user is explicitly intruding on the per- sonal space of others or disrupting the environment. Campbell (2007) found empirical evidence for these differences between cultures. He showed that Japanese participants “tended to be more tolerant of mobile phone use in a classroom, but less tolerant of use on a sidewalk and, especially, on a bus” than participants from mainland United States, Hawaii, Taiwan, and Sweden; on the other hand, “Taiwanese participants tended to report more tolerance for mobile phone use in a theater,

312 A. EASWARA MOORTHY AND K.-P. L. VU

restaurant, and classroom than did participants from the other cultural groupings” (p. 738).

2.3. Environmental Factors Mobile phone use in public spaces might be regulated by

environmental factors such as the law. For example, “New York City passed a law in 2003, fining people whose mobile phones ring in ‘places of public performance’” (Srivastava, 2005, p. 123). A study of public attitude in the United Kingdom, France, Germany, Italy, and Spain showed that participants from these countries were least likely to leave their mobile phones turned on during events like plays or shows, as they wanted to adhere to rules in those settings (Ling & Haddon, 2003). “Despite the fact that [cell phone] jammers [that restrict usage] are illegal in most countries, more and more countries, such as Japan and France and Mexico are approving their use in public” (Khalil & Connelly, 2005, p. 198). Presence of an audience familiar to a mobile phone user (such as friends) increased user acceptability of performing gesture-based and voice-based interactions in public settings (Williamson, 2012). A familiar context or location (such as a street the user takes everyday) might also increase user acceptability of voice-based interactions in public (Williamson, 2012). Finally, facilitating conditions such as availability and quality of mobile phone services/devices might be determined by environmental vari- ables such as country of user residence (Verkasalo, 2007).

2.4. User Characteristics Demographic and cognitive variables related to users could

also play a role in the user adoption of mobile phones in public. Turner, Love, and Howell (2008) concluded that public mobile phone use annoyed men more than women. Young people use the discreet input method of texting more than older mobile phone users (A. S. Taylor & Harper, 2003). In a study support- ing the role of prior experience length in mobile phone adoption, Palen, Salzman, and Youngs (2001) showed that people became more accepting of its use in social settings as their own usage increased. Love and Perry (2004) showed that the type of affective responses that participants exhibited while overhear- ing mobile conversations could predict their own attitudes and behavioral strategies of taking calls in public.

Most of these studies give insight into the attitudes of people in the colocated space of a mobile phone user taking a voice call in public. However, few studies have probed into the preference and acceptability of public voice service and voice applica- tion usage from the perspective of users by explaining “the reasons why in certain contexts some remain uncomfortable with it” (Turner et al., 2008, p. 203). The following studies on multimodal research might hint at why users avoid smartphone voice applications such as the VAPA in public to complete tasks.

2.5. Preference for Voice Applications in Public Settings Multimodal interfaces are defined as “interfaces that specif-

ically exploit the capabilities and affordances of more than one modality, either used together or separately as part of one interface” (Williamson, 2012, p. 21). Research on multimodal systems has examined preference for use of voice mode in social settings, and these studies have shown that users tend to avoid using voice interaction in public. For example, Reis, de Sá, and Carriço (2008) studied users’ preference of different interaction modalities (voice, gestures, touch screen, keypad) on a PDA device under various contexts (home, park, subway, and driv- ing) and conditions (lighting, noise, position, movement, type of content, number of people surrounding the user, and time constraints). Reis et al. found that usage of voice application in public spaces decreased with increasing number of strangers in the surroundings. “After, when surrounded by more people (as in the subway’s case study), users completely stopped using voice interaction because they were embarrassed of speaking to a device in front of so many people” (p. 68). They also cited privacy concerns for rejecting voice interaction in the park.

In public situations, users are more likely to perform other types of interactions with their mobile devices. Williamson (2012) showed that gesture-based interactions were considered to be more socially acceptable over voice-based interactions when interacting with a mobile device in front of strangers on the sidewalk. However, this study did not ask participants to elaborate on why they avoided voice interaction in public.

Thus, prior studies have shown that users will avoid voice input when surrounded by strangers in public spaces. Although these findings hint that privacy concerns could be one rea- son for avoiding voice interaction in social and public settings, they have not been supported by empirical research. The next section on mobile phone privacy reviews existing literature on this subject.

3. PRIVACY CONCERNS WITH MOBILE PHONE USE Goffman (1959) and Ling (1997) proposed that mobile

phone users are acting on double front stages when talking on a mobile phone around others. They physically interact with the bystanders who surround them and remotely interact with the person on the other side of the conversation. Prior studies have explored privacy related to both of these interactions.

3.1. Privacy Concerns for Physical Interactions There has been much interest in privacy issues arising from

the use of mobile phones in the public and the extension of social norms in public spaces as it relates to interaction with a person engaged in a cell phone conversation (Humphreys, 2005; Ling, 2002; Srivastava, 2005). Past studies found that cell phone users taking a voice call and those surrounding them in the colo- cated public space usually tried to create a “private environment

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 313

within a public one” by adopting various physical behaviors (Srivastava, 2005, p. 123). Murtagh (2002) argued that these behaviors served to mitigate embarrassment associated with the public audibility of private conversations.

For example, Humphreys (2005) found that users talking on the cell phone in public use body language to create a pri- vate space around them by turning their head and upper body away from others, or taking the cell phone elsewhere to talk. In addition, he observed that others around the cell phone user in the colocated space also helped create a private space by physically turning away, looking away, or engaging in other distracting activities. He predicted that the surrounding com- pany could be behaving in a way to show that they were not eavesdropping. Goffman (1963) argued that middle-class Americans use such activities to maintain “civil inattention” to politely distance themselves from socially inappropriate situ- ations in public places (Humphreys, 2005, p. 818). However, these behaviors usually failed to prevent eavesdropping, a com- mon occurrence (Humphreys, 2005; Srivastava, 2005). Thus, these studies confirm that mobile phone users would like to establish a private domain in the public space to keep their voice calls private from others in the colocated space. However, these studies on public mobile phone use did not specifically assess user preferences for privacy, especially as it related to the type of information users transmitted.

3.2. Privacy Concerns for Remote Interactions With the introduction of GPS-enabled smartphones that can

constantly relay user information, the latest research in mobile phone privacy has focused on user preferences for sharing their contextual information, especially location with others in their social circles (i.e., spouse, family, friends, and colleagues) and third parties. Nissenbaum’s (2004) theory of contextual integrity argues that

there are no such thing [sic] as universal privacy norms but that these are distinct to each situation, and assist in maintaining contex- tual integrity, which is “a desirable state that people strive towards by keeping perceived-private information private according to the context.” (Barkhuus, 2012, p. 369)

Individuals change the type and amount of information that is revealed to others depending upon the context and the relation- ship between them. For example, in the United States, one might be ready to discuss salary details with a potential employer but not with neighbors.

Due to the highly private nature of location information and the possibility of constant location tracking with GPS-enabled mobile phones, a majority of the latest mobile phone privacy research has focused on users’ privacy preferences and con- cerns on location tracking and sharing with particular people in their social groups (Barkhuus & Dey, 2003; Tang, Lin, Hong, Siewiorek, & Sadeh, 2010; Tsai et al., 2009). These studies have focused on privacy concerns relating to only one type of contex- tual information—location. Although user privacy preferences

for sharing other types of personal and contextual informa- tion with different social circles have been studied (see, e.g., Brandtzæg, Lüders, & Skjetne, 2010), the topic is yet to be fully examined in the mobile domain.

It is imperative to understand sharing patterns of various types of personal and contextual information in order to under- stand their differential private nature. A study on mobile phone sharing behaviors found that users were more likely to let oth- ers use their phones to make calls and access photos, games, and the web; however, they had been less permissive of lending their phones for using “applications that contained personal informa- tion, such as voicemail, notes, files, email, SMS, and calendars” (Karlson, Brush, & Schechter, 2009, p. 1649). One of the rea- sons for their unwillingness to share was data privacy. They also seemed to restrict access more to strangers and acquain- tances than close friends and family. Khalil and Connelly (2006) conducted an in situ study in which participants were asked whether they would disclose contextual information (about current location, current activity, presence of others in surround- ings, and engagement in a conversation) to a potential caller from their different social groups (significant other, friend, boss, colleague, family, or unknown person) during different times of the day over a period of 10 days. Participants were more willing to share whether they had company or were in a conversation than their location and activity across social groups, suggesting that the former two types of information were considered less private information than the latter two types. Analysis of disclosure rates showed that participants grouped social relations into high-sharing (significant other, friend, and family), medium-sharing (boss and colleague), and low-sharing (unknown callers) groups. This study also found that the percentage of incoming calls that participants con- sidered appropriate for the situation linearly decreased as the number of people surrounding them increased.

In the case of mobile phone texting, Marques, Duarte, and Carriço (2012) found that the theme of text message and the receivers’ relationship status influenced privacy concerns of the sender. This retrospective survey found that participants con- sidered approximately 60% of their recently sent text messages as requiring privacy (i.e., they did not want others in the sur- roundings to understand the content of the message). They strategically ensured their seclusion from others while sending these private messages to prevent them from accessing mes- sages. More messages related to romantic relationships were considered private compared to messages about school, job, places, and other types of information seeking, planning, and chatting. Participants required privacy for 29% of messages sent to a receiver with whom they had a romantic relationship. However, they required privacy or both privacy and secrecy (i.e., they did not want others to detect sending the message) for only 14% of messages sent to family. Finally, participants’ level of privacy concern might be a product of their personality (e.g., some participants were naturally prone to keep their personal information away from others).

314 A. EASWARA MOORTHY AND K.-P. L. VU

These studies in the mobile context support findings from an important privacy study in the general human–computer inter- action domain. Olson, Grudin, and Horvitz (2005) identified clusters of personal information (e.g., demographics, personal contact information, work data, health and preferences, income, financial information) that people were hypothetically willing to share with clusters of social groups (e.g., spouse, family, manager, coworkers, and the public, which included telemar- keters and websites) in interpersonal settings. They found that participants were most unwilling to share private information with the public than all other groups and that sharing levels changed between the various social groups for the different types of information. Within the data shared with the public group, participants seemed to be more comfortable disclos- ing certain types of personal information such as work e-mail and work phone number than other types such as credit card information and past transgressions.

In summary, prior privacy studies exclusively examined pri- vacy concerns due to the sharing of private information with various social groups. From these studies, one can infer that mobile phone users might be more unwilling to disclose per- sonal information (such as personal contact information, past transgressions, work contact information, and credit card infor- mation) and contextual information (such as current location, activity, presence of company, and engagement in a conversa- tion) in front of strangers in public than to closer relations like their significant other, family, friends, and coworkers due to privacy concerns. Within these different types of personal infor- mation, participants would hesitate more about sharing certain private information (such as credit information, past transgres- sions, current location, and current activity) than other types of personal data (such as personal contact information, work contact information, presence of company, and engagement in a conversation). However, privacy concerns due to the verbal transmission of different types of information to social circles are yet to be extensively investigated. The present study exam- ines user privacy concerns when using the VAPA in the public to input private information and provide design recommenda- tions to increase public use of VAPA for disclosing private information.

4. PRESENT STUDY To help inform design decisions to make the private use of

VAPA in public spaces more acceptable by VAPA users, the present study had U.S. smartphone users with prior experience of using a VAPA complete an online survey to capture their rat- ings for likelihood of VAPA use in various locations. It can be inferred from the literature review on technology adoption that publicly using the VAPA will be influenced by social context. Social context might discourage users from interacting with the VAPA in public, because participants might not want to publicly disclose private information audible to others in the colocated space.

Therefore, the online survey (see the appendix) was designed to capture reports of the participants’ likelihood of using their VAPA as a function of location type (public and private loca- tions) and task content type (private and nonprivate informa- tion). As a control input condition, participants were also asked how likely they were to use their smartphone’s keyboard to input the same information in the same locations. Likelihood of VAPA and smartphone keyboard use in a location were rated on a scale ranging from 1 (not likely at all to use) to 7 (extremely likely to use). The survey also required the participants to rate how socially acceptable they thought using the VAPA was in different locations on a scale from 1 (not acceptable at all) to 7 (very acceptable). These ratings were expected to be corre- lated with those of VAPA likelihood of usage and thus serve as a manipulation check for likelihood of usage ratings.

5. HYPOTHESES Based on the literature review, the following seven hypothe-

ses were examined in the present study:

H1: Participants will be more likely use their smartphone key- board than the VAPA to transmit information. Because the acceptability and propriety of use of voice services are dependent on a number of factors such as type of server (e.g., Fischer, Price, & Sears, 2005), location and time (Campbell & Russo, 2003; Wei & Leung, 1999), culture (Campbell, 2007; Ito & Okabe, 2005), and environment (Srivastava, 2005; Williamson, 2012), users would per- ceive it more convenient to use their smartphone keyboard as the default for information transmission regardless of the location.

H2: Participants will be more likely to transmit information in private than in public locations. In public spaces, partici- pants are more likely to be surrounded by strangers com- pared to private locations. Perceived attention of strangers to their smartphone-related activities (Khalil & Connelly, 2006; Marques et al., 2012; Murtagh, 2002; Olson et al., 2005) would prevent users from transmitting information in public locations.

H3: Participants will be more likely to transmit nonprivate information than private information. Previous studies have shown that smartphone users are more protective of their private digital information compared to nonprivate information (Karlson et al., 2009; Khalil & Connelly, 2006; Marques et al., 2012). Therefore, users would be more likely to transmit nonprivate compared to private information.

H4: Participants will be more likely to use their VAPA in pri- vate locations than in public locations. However, they will be equally likely to use the keyboard in private and pub- lic locations. Users’ perceived attention of strangers in the colocated public space to their conspicuous VAPA activi- ties would prevent them from using VAPA more in public

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 315

than in private locations. The discreet nature of keyboard information entry would make the smartphone keyboard ideal to use at any location.

H5: Participants will be more likely to use their VAPA to enter nonprivate information than private information. However, participants will be equally likely to use the keyboard for nonprivate and private information entry. Because users would be more protective of their private information compared to nonprivate information, they would be more unwilling to speak out loud and transmit their private infor- mation using a VAPA. However, the discreet nature of smartphone keyboard would allow users to safely transmit any type of information.

H6: Participants will be more likely to enter nonprivate information than private information in public locations. However, they will be equally likely to enter nonprivate and private information in private locations. The pres- ence of strangers and their perceived attention to users’ smartphone-related activities would prevent users from transmitting private information more, because users would think private information needs to be more guarded from strangers’ scrutiny compared to nonprivate infor- mation. There would no threat of information breach in private locations, where strangers will not be present. Therefore, users would not distinguish between the types of information to be transmitted.

H7: Public use of VAPA will be less likely than private use of VAPA to input all types of information. However, partici- pants will be even less likely to publicly use the VAPA to input private information. In the keyboard condition, par- ticipants will be equally likely to use the keyboard at both locations to input both types of information. The combined effects of the conspicuous nature of information entry using the VAPA and perceived attention from strangers to users’ smartphone activities would prevent users from using the VAPA more in public locations. In these cir- cumstances, the negative impact of information breach is higher when transmitting private information, which is even more guarded compared to nonprivate informa- tion. The inconspicuous nature of information entry using the smartphone keyboard will negate effects of perceived social scrutiny and danger of information breach.

6. METHOD

6.1. Participants Data from 76 participants, who met quality control crite-

ria, for an online survey about their VAPA usage preferences are included in the present study. Participants were recruited from Amazon Mechanical Turk (AMT)—an online crowd- sourcing marketplace through which individuals can sign up to complete human intelligence tasks (HITs) posted by research entities. Only AMT workers who were residents of the United

States were allowed to participate. Residency restrictions were employed to ensure that all participants were exposed to the same availability of services and devices (Verkasalo, 2007) and legal regulations (Khalil & Connelly, 2005; Ling & Haddon, 2003; Srivastava, 2005) regarding mobile phone use in pub- lic spaces, factors that have been shown to affect acceptability of public mobile voice services usage. Participants were also required to be smartphone users with the prior experience of using a smartphone VAPA such as Siri, S Voice, or Google Now to eliminate effects of novelty from their responses.

The participant pool was roughly divided in gender, with males constituting 55% and females constituting 45%. The par- ticipants were relatively young (with 78% younger than 35 and 95% younger than 45 years of age). The racial makeup of the participants was 84% White, 11% Black, and 8% Asian. Note that participants could check more than one category to classify their racial identity; as a result, the cumulative percent- age exceeds 100%. A majority of them (84%) reported having some college education or more. Finally, 70% of all participants had an average household income range between $25,000 and $100,000. For a complete profile of the participants’ demo- graphics, see Table 1.

6.2. Design This study employed a 2 (location: public and private spaces)

× 2 (type of content information: private and nonprivate infor- mation) × 2 (mode of input: VAPA verbal input and smartphone keyboard manual typing) factorial design.

6.3. Materials This survey was listed as a HIT on the online crowdsourc-

ing service AMT. Using screening criteria incorporated into AMT, this HIT was made visible and accessible only to U.S. AMT workers with a task approval rating of 95% or above. The HIT description also included instructions to complete the HIT. First, the workers were instructed to access the survey from the listed web address. In the survey, they were required to answer questions on their smartphone usage habits and demographics and provide ratings for using their smartphone in various con- texts. They were required to pay attention to all questions in the survey, as one of them contained the special code that had to be entered into the verification text box provided in the HIT description. The special code had to be correctly entered for them to receive payment. This verification mechanism was nec- essary to confirm that participants had completed the survey on the external website, in which the survey was listed.

The 28-question online survey (see the appendix) was titled “Smartphone Usage Preferences Survey.” The first page of the survey served as the electronic consent form. It reiterated the survey’s eligibility requirements, purpose, procedures, and pay- ment. Risks, benefits, and confidentiality issues associated with the survey were also mentioned. The main survey consisted entirely of close-ended questions and divided into the following

316 A. EASWARA MOORTHY AND K.-P. L. VU

TABLE 1 Participant Demographic Information

Demographic N %

Gender Male 42 55.3 Female 34 44.7

Age 18–24 29 38.2 25–34 30 39.5 35–44 13 17.1 45–54 2 2.6 55–64 2 2.6

Race White 64 84.2 Black or African American 8 10.5 Asian 6 7.9 American Indian or Alaska Native 2 2.6

Education Less than high school 1 1.3 High school/GED 11 14.5 Some college 25 32.9 2-year college degree (Associates) 9 11.8 4-year college degree (B.A., B.S.) 22 28.9 Master’s degree 5 6.6 Professional degree (M.D., J.D.) 3 3.9

Average household income $0–$24,999 13 17.1 $25,000–$49,999 25 32.9 $50,000–$74,999 13 17.1 $75,000–$99,999 15 19.7 $100,000–$124,999 4 5.3 $125,000–$149,999 1 1.3 $150,000–$174,999 1 1.3 $175,000–$199,999 1 1.3 $200,000 and up 1 1.3 Decline to answer 2 2.6

five sections. (Note that the term “voice assistant” instead of VAPA was used throughout the survey to keep the language simple and understandable to all participants.)

Section 1 probed participants about their smartphone usage information through five questions. Specifically, they were asked to choose the smartphone device type and voice assis- tant application they used from a list of popular smartphone types and VAPAs, select length of smartphone usage, and indi- cate their comfort level with using their smartphone VAPA and keyboard on a scale from 1 (not comfortable at all) to 7 (very comfortable to use). The multiple-choice questions in this section were a modified version of questions from Siftar (2012a) on Siri. The present study asked participants to reflect on their habits and experiences with their current smartphone voice assistant instead of Siri.

In section 2, participants answered 12 questions. They were instructed to imagine themselves at six locations and rate the likelihood of using their voice assistant at each of the loca- tions by saying out loud the given phrases to perform different smartphone tasks through voice mode.

Then they had to rate the likelihood of using their smartphone touch screen or physical keyboard for manual typ- ing to perform these same tasks at the same six locations. Ratings were anchored on a scale ranging from 1 (not likely at all to use) to 7 (extremely likely to use). The order of the input conditions presented for likelihood of usage ratings was counterbalanced between participants.

To eliminate the confounding effects of perception of human error or low quality or failure of technology or service, partici- pants were asked to make several assumptions. Before proceed- ing to rate the likelihood of usage of their smartphone voice assistant, participants were instructed to assume that (a) the term “voice assistant” in the survey referred to a smartphone voice assistant application such as Siri, Google Now, or S Voice that enabled the smartphone user to complete tasks through voice mode; (b) their voice assistants could detect users’ input on the first try against background noise; understand their lan- guage and accent; and provide audible, accurate, and adequate feedback/answers; (c) their smartphones had adequate battery power and service reception to run the voice assistant; and (d) they could input voice queries without mistakes and were in a comfortable position to freely operate their smartphones to input voice commands and receive voice feedback from the VAPA. Similarly, before rating keyboard usage likelihood, par- ticipants had to make the following equivalent assumptions about their keyboard: Their keyboard could accurately detect their typing and provide accurate and adequate responses and information, their smartphones had adequate battery power and service reception, and they could input the text query without mistakes and freely operate their smartphone to input manual commands and receive visual feedback.

For each input mode, participants had to rate how likely they were to use it in six locations, specifically three public loca- tions (“at a relatively quiet, but crowded restaurant,” “waiting in a long line at the supermarket,” and “in the lounge at your work”) and three private locations (“alone at home,” “alone in your parked car,” and “at your desk at work”). They were asked to imagine themselves at each of these locations before being shown the rating scales, that is, they were instructed: “Imagine yourself [location]. Rate how likely you are to use your smartphone’s [touchscreen or physical keyboard/voice assistant] to complete the following tasks in this location.” Although the location descriptions did not explicitly state whether the locations were public or private, they indicated the presence or absence of strangers to distinguish public loca- tions (e.g., a relatively quiet but crowded restaurant, waiting in a long line at the supermarket) from private locations (e.g., home alone, alone in a parked car). The same set of three pub- lic locations and three private locations were presented for both

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 317

input conditions. However, order of locations under each input condition was randomized.

At each location, participants rated their preference to use their voice assistant and smartphone keyboard to complete six smartphone tasks (texting, getting directions, calling a con- tact, searching the web, setting up meetings, and setting up reminders). These tasks were derived from common function- alities available to VAPA users and were expected to be familiar to the participants. Multiple tasks were presented to eliminate task-specific effects. The order of tasks was randomized under each location.

For each of the tasks, participants were asked to enter two types of information: private information and nonprivate infor- mation. For each type of information, scenarios were included to illustrate when participants might need to input it. For the task of finding direction for the private information condition, the listed scenario was “Take me home to 1250 Pine Street, Long Beach” and for the nonprivate information condition, the scenario was “Show me directions to the nearest coffee shop”; for the task of searching the web, the private information sce- nario was “Find famous people born on my birthday, July 19, 1973” and for nonprivate information, the scenario was “Search the web for weather tomorrow”; texting task’s private informa- tion scenario was “Text John/Jane ‘My social security number is 652 341 9518’” and nonprivate information scenario was “Text John/Jane ‘How was your day today?’”; calling task’s private information scenario was “Call John/Jane Carpenter” and nonprivate information scenario was “Call John/Jane”; “Set up meeting to change car license plate number 4CTD987 on the 2nd at 9 AM” was the private information scenario and “Set up meeting with John/Jane on the 4th at 3 PM” was the nonprivate information scenario for updating calendar task; “Remind me to add new credit card number 3371 2452 0221 6633 to finance paperwork” was the private information sce- nario and “Remind me to buy cereal on Monday” was the nonprivate information scenario for the setting reminder task. Thus, in section 2, participants provided the likelihood of usage ratings on 144 tasks (12 tasks each for the smartphone keyboard and VAPA conditions, across six different locations).

The private and nonprivate information used in the scenar- ios were derived from data identified as private and nonprivate information in prior privacy studies in the human–computer interaction domain (Khalil & Connelly, 2006; Marques et al., 2012; Olson et al., 2005). They were tested through a pilot survey, in which 16 participants rated the level of privacy of the instructions when spoken out loud in public, on a scale from 1 (not private at all) to 7 (very private). The instruc- tions containing personally identifiable information (PII) were consistently rated between 4 and 7, indicating medium to high privacy, whereas those containing nonprivate information were rated between 1 and 3, which represented the low privacy range. The order of the type of information to be entered for each task was randomized under each location.

Section 3 included four multiple-choice questions on the par- ticipants’ prior experience with their VAPA, such as the tasks

they performed with their voice assistant, reasons for why they used and did not use their voice assistant, and how they felt when using it in front of strangers. These questions were derived from the study on usage of Siri (Siftar, 2012a). They were placed after the likelihood of voice assistant and keyboard usage rating scales in order to eliminate bias from the usage ratings.

Section 4 consisted of a single question, in which a rating matrix was presented to the participants to rate how socially acceptable they thought it would it be to use their voice assistant to complete the same aforementioned tasks in the aforemen- tioned locations. This rating scale also ranged between 1 (not acceptable at all) and 7 (very acceptable to use). This ques- tion was included to assess the correlation between likelihood of usage and perceived social acceptability of use of voice assistants.

The final section, section 5, consisted of six multiple-choice questions on the demographics of the participants: gender, age, race, education, and approximate household income.

6.4. Quality Control Measures As a checking measure to gauge whether participants were

paying attention to all the questions in the survey, quality con- trol questions were placed toward the middle (in section 2) and the end of the survey (in section 5). For example, in section 2, the following quality control question was embedded among the likelihood of usage rating scale questions as shown in the last row of the “Smartphone Voice Assistant Usage” rating scale table under Questions 6 to 11 in the appendix: “For quality control purposes, please select ‘May or may not use’ for this question.” Users had to select 4 (representing “May or may not use”) on the scale from 1 to 7 to indicate that they were being attentive to the survey. The second quality control question in section 5 required participants to select the choice “laptop” from the choices “desktop,” “laptop,” and “mobile phone” and indicated to enter this special code “laptop” to verify their com- pletion of the survey: “For quality control purposes, please select ‘laptop’ as the answer to this question. Enter the special code ‘laptop’ into the verification textbox in the Amazon Turk HIT description to indicate your completion of this HIT.” (See Question 26 in the appendix.) Responses from participants who did not correctly answer the two quality control questions were excluded from data analysis.

6.5. Procedure AMT workers who were U.S. residents with a task approval

rating of 95% or above viewed an HIT on the AMT web- site inviting them to participate in a 30-min online survey for a payment of $0.75. After reading the survey eligibility requirements and instructions, smartphone users among them with previous experience using voice assistants self-selected themselves as survey participants if they wished to complete the HIT.

To access the survey, they clicked on the provided survey hyperlink and navigated to the external website SurveyMonkey,

318 A. EASWARA MOORTHY AND K.-P. L. VU

in which the survey was administered. On the first page of the survey, participants were instructed to pay attention to all the questions, as one of them contained a special code that they had to correctly enter into the HIT verification text box in the AMT website to receive payment. Then they indicated their electronic consent to the terms and conditions of the survey and their vol- untary participation. Next, they completed the aforementioned five sections of the survey and obtained the special code. At the end of the survey, participants were thanked for their partici- pation and again instructed to correctly enter the special code. Finally, they navigated back to the AMT website, entered the special code, and submitted the HIT.

The survey took approximately 30 min to finish, and partic- ipants were paid $0.75 for their time. (This pay is consistent with current rates offered by other researchers for similar tasks with similar durations.) Eligible AMT workers had the option of choosing to participate in this survey or opt out.

7. RESULTS A total of 120 respondents completed the survey. However,

only “quality” responses from 76 participants were included in the data analysis.

7.1. Smartphone Usage The iPhone was the most commonly used smartphone, with

59% of participants indicating that they used it. For the remain- ing participants, 21% reported using a Samsung phone and 20% a non-Samsung Android phone. Participants used the follow- ing VAPAs: Siri (55%), Google Now (37%), and S Voice (7%). A majority of the participants (82%) had been using their smart- phones for more than 1 month but less than 2 years. Touchscreen keyboard (95%) was more common in smartphones than a phys- ical keyboard (1%). Participants also reported feeling more comfortable using their smartphone keyboard (M = 6.5, SD = 0.87) than the VAPA (M = 5.1, SD = 1.61) on a scale from 1 (not comfortable at all) to 7 (very comfortable to use). See Table 2 for a more detailed report of the participants’ smartphone usage information.

7.2. Prior Experience With VAPA The VAPA was most commonly used to perform the follow-

ing functions: make calls (79%), send text messages (68%), find places on a map (74%), get weather information (67%), search the web (70%), set alarm (61%), set reminder (55%). It was considered easier (76%), faster (63%), and more fun (41%) to use for completing the aforementioned tasks. When asked how they felt while using the VAPA in front of unknown people, the most common answer was that they felt uncomfortable (41%). Thirty-two percent felt indifferent, 13% were embarrassed, and only 12% reported feeling comfortable with public use. Privacy concern was the primary reason for not using the VAPA, with

TABLE 2 Smartphone Usage Habits of Participants

Feature N %

Smartphone used iPhone 45 59.2 Samsung phone 16 21.1 Non-Samsung Android phone 15 19.7

Smartphone usage period Less than 1 week 0 0.0 Less than 1 month 2 2.6 More than 1 month but less than 1 year 28 36.8 1–2 years 34 44.7 2–3 years 8 10.5 More than 3 years 4 5.3

VAPA used Siri 41 55.4 S Voice 5 6.8 Google Now 27 36.5 Decline to answer 1 1.4 Other 2

Smartphone keyboard type used Touchscreen 72 94.7 Physical 1 1.3 Both 3 3.9

Note. VAPA = Voice Activated Personal Assistant.

55% of the participants stating it. In addition, being misunder- stood by the application (51%), unsatisfactory answers (50%), and preference for classic methods (24%) were also identified as reasons preventing VAPA use. Table 3 shows a complete record of the participants’ responses to all questions concerning their prior experience using the VAPA.

7.3. VAPA and Keyboard Likelihood of Usage Ratings Participants were asked to rate their likelihood of using either

the VAPA or keyboard to perform certain tasks on a scale of 1 (not likely at all to use) to 7 (extremely likely to use). The mean scores for VAPA and keyboard likelihood of usage rat- ings were calculated across the individual tasks for entering private and nonprivate information in private and public loca- tions. Then they were analyzed using a 2 (location: public and private spaces) × 2 (type of content information: private and nonprivate information) × 2 (input mode: VAPA and keyboard) within-subjects analysis of variance. Table 4 shows a sum- mary of the main results from this analysis of variance. The descriptive statistics of the groups are shown in Table 5.

All main effects were significant. For input mode, the mean likelihood of usage rating of smartphone keyboard was greater (M = 5.0, SD = 0.14) than that of VAPA (M = 3.7, SD = 0.14) to enter information, F(1, 75) = 48.2, MSE = 5.35, p < .001. Thus, Hypothesis 1 was supported: Participants are more

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 319

TABLE 3 Prior Experience of Participants With VAPA

Question N %

Smartphone task performed make calls 60 78.9 send text messages 52 68.4 send e-mails 22 28.9 find places on map 56 73.7 get weather info 51 67.1 get contact info 28 36.8 search the web 53 69.7 play music 31 40.8 set calendar event 24 31.6 set alarm 46 60.5

Smartphone task performed set reminder 42 55.3 take notes 25 32.9 get sports updates 11 14.5 search for local listings (businesses, sites, etc.) 29 38.2 post on social networks 19 25.0 make restaurant reservations 5 6.6 find movies 30 39.5 find friends 13 17.1 get simple facts, conversions, and

calculations 37 48.7

Reason for VAPA use It’s easier 57 76.0 It’s faster 47 62.7 It’s more fun 31 41.3 It’s more challenging 2 2.7 It’s more “Star Trek” 15 20.0 Decline to answer 1 1.3 Other 6 7.89

Feeling when using VAPA in front of strangers Proud 2 2.6 Comfortable 9 11.8 Indifferent 24 31.6 Uncomfortable 31 40.8 Embarrassed 10 13.2

Reasons for not using VAPA Being misunderstood 38 51.4 Unsatisfactory answers 37 50.0 Difficulty of use 10 13.5 Discomfort of use 15 20.3 Language issues 3 4.1 Privacy concerns 41 55.4 Preference for classic methods 18 24.3 I didn’t know it was there 0 0.0 Decline to answer 2 2.7 Other 3 3.95

Note. VAPA = Voice Activated Personal Assistant.

likely to use their smartphone keyboard than VAPA to transmit information.

Participants gave a higher mean likelihood of usage rating for using their smartphone to enter information at private loca- tions (M = 4.8, SD = 0.12) than at public locations (M = 4.0, SD = 0.11), F(1, 75) = 73.8, MSE = 1.55, p < .001. Thus, Hypothesis 2 was supported: Participants are more likely to transmit information in private than in public locations.

Furthermore, the likelihood of usage to enter nonprivate information (M = 4.9, SD = 0.12) was higher than for private information (M = 3.9, SD = 0.11), F(1, 75) = 72.5, MSE = 2.14, p < .001. Thus, Hypothesis 3 was supported: Participants will be more likely to transmit nonprivate information than private information.

All two-way interactions of the variables were also signif- icant, but the three-way interaction of all variables was not, as summarized in Table 4. There was a significant interac- tion effect between input mode and location, F(1, 75) = 32.3, MSE = 1.67, p < .001 (see Figure 4). Mean likelihood of usage rating for VAPA was greater for entering information in private locations (M = 4.5, SD = 0.15) than in public loca- tions (M = 3.0, SD = 0.15). However, this difference was less pronounced in the keyboard condition, in which the rating dif- ference between using the smartphone in private locations (M = 5.2, SD = 0.16) than public locations (M = 4.9, SD = 0.16) was lower. A simple effects analysis was conducted to determine if these differences in ratings were significant. It showed that the difference in these ratings was significant for the VAPA, F(1, 75) = 116.9, MSE = 0.70, p < .001, whereas it was not signif- icant for the keyboard, F(1, 75) = 3.1, MSE = 0.91, p = .084. Thus, Hypothesis 4 was supported: Participants will be more likely to use their VAPA in private locations than in public loca- tions. However, they will be equally likely to use the keyboard in private and public locations.

A similar pattern was found in the interaction between input mode and information type, such that the difference between the mean likelihood of usage ratings for entering nonprivate infor- mation (M = 4.4, SD = 0.17) and private information (M = 3.1, SD = 0.12) was greater with the VAPA compared to the difference between nonprivate information (M = 5.4, SD = 0.14) and private information (M = 4.6, SD = 0.17) while using the keyboard, F(1, 75) = 17.6, MSE = 0.49, p < .001. An analysis of the simple effects revealed that these differences between location ratings were significant for the VAPA, F(1, 75) = 94.4, MSE = 0.63, p < .001, as well as the keyboard, F(1, 75) = 33.0, MSE = 0.69, p < .001, conditions (see Figure 5). Thus, Hypothesis 5 was partially supported: Participants will be more likely to use their VAPA to enter nonprivate information than private information. Although participants were also more likely to use the keyboard for nonprivate over private informa- tion, the difference in usage was less than with the VAPA.

320 A. EASWARA MOORTHY AND K.-P. L. VU

TABLE 4 Analysis of Variance Table for Voice Activated Personal Assistant and Keyboard Likelihood of Usage Ratings

df MSE F Value p Value Partial η2

Input mode 1,75 5.35 48.17 <.001 0.391 Location 1,75 1.55 73.80 <.001 0.496 Type of information 1,75 2.14 72.49 <.001 0.491 Input Mode × Location 1,75 1.67 32.30 <.001 0.301 Input Mode × Type of Information 1,75 0.49 17.57 <.001 0.190 Location × Type of Information 1,75 0.43 6.69 .012 0.082 Input Mode × Location × Type of Information 1,75 0.22 1.56 .216 0.020

TABLE 5 Descriptive Statistics of Experimental Conditions

Input Mode Location Information type Usage M Usage SD Accept M Accept SD

VAPA Private Private 3.946 1.415 4.904 2.415 Nonprivate 5.007 1.475 5.217 2.262

Public Private 2.298 1.125 2.637 1.957 Nonprivate 3.729 1.776 3.607 2.203

Keyboard Private Private 4.841 1.547 Nonprivate 5.523 1.402

Public Private 4.480 1.743 Nonprivate 5.342 1.351

Note. Usage ratings represent Likelihood of Usage ratings. Accept ratings represent Acceptability ratings. VAPA = Voice Activated Personal Assistant.

FIG. 4. Mean likelihood of usage rating as a function of input mode and location. Note. VAPA = Voice Activated Personal Assistant.

Finally, the interaction between location and information type was significant, F(1, 75) = 6.7, MSE = 0.43, p = .012. Participants gave a higher rating for entering nonprivate infor- mation (M = 5.3, SD = 0.13) than private information (M =

FIG. 5. Mean likelihood of usage rating as a function of input mode and information type. Note. VAPA = Voice Activated Personal Assistant.

4.4, SD = 0.13) in private locations. However, this differ- ence between nonprivate information rating (M = 4.5, SD = 0.14) and private information rating (M = 3.4, SD = 0.12) was higher in public locations (see Figure 6). A simple effects analysis showed that the difference in ratings between private

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 321

FIG. 6. Mean likelihood of usage rating as a function of location and information type.

information and nonprivate information in both private loca- tions, F(1, 75) = 59.8, MSE = 0.48, p < .001, and public locations, F(1, 75) = 62.5, MSE = 0.80, p < .001, was signif- icant. Thus, Hypothesis 6 was partially supported: Participants will be more likely to enter nonprivate information than private information in public locations. Although this pattern was evi- dent for the private locations, the difference in usage was not as strong.

As noted earlier, the three-way interaction between input mode, location, and information type was found to be not signif- icant, F(1, 75) = 1.6, MSE = 0.22, p = .216. Thus, Hypothesis 7 was not supported (see Table 6 for summary of hypotheses).

7.4. Correlation Between VAPA Usage Likelihood and Social Acceptability

Ratings on the social acceptability of using a VAPA to complete the six smartphone tasks were correlated with the cor- responding VAPA likelihood of usage ratings for each location and each information type (for a total of 72 correlated data pairs; see Table 5 for descriptive statistics and Table 7 for the compre- hensive set of correlation data). The correlation analysis found that social acceptability and likelihood of usage of VAPA were significantly positively correlated in 44 of 72, or 61.1%, of the pairs.

Tasks requiring participants to enter nonprivate information showed more significant correlations (26 of 32 tasks, or 81.3%) than tasks involving participant input of private information (17 of 32 tasks, or 53.1%). In the nonprivate information con- dition, the number of significant correlated ratings was higher in public locations such as the supermarket (six of six pairs, or 100%), work lounge (six of six pairs, or 100%), restaurant (five of six pairs, or 83.3%), and the work desk (five of six pairs, or 83.3%) compared to private locations such as home (two of

six tasks, or 33.3%) and car (one of six tasks, or 16.7%). This pattern was not found within private information tasks.

However, it was found that public locations had fewer signif- icant correlations in the private information condition compared to the nonprivate information condition. The following compar- isons illustrate this trend within public locations: supermarket (16.7% vs. 100%), work lounge (66.7% vs. 100%), restaurant (50% vs. 100%), and work desk (50% vs. 83.3%), which could be a public space. This trend is reversed for private locations, with the private information condition receiving higher corre- lations compared to the nonprivate information condition as listed: home (66.6% vs. 33.3%) and the car (33.3% vs. 16.7%).

Finally, among the private information tasks, texting, finding directions to a location, and calling contacts received more sig- nificant correlated ratings (66.6% each) compared to searching the web (two of six, or 33.3%), updating calendar (one of six, or 16.7%), and setting reminder (one of six, or 16.7%). In the nonprivate information condition, all the tasks had at least 50% significant correlations.

8. DISCUSSION Previous research investigating privacy concerns in the

mobile domain identified social context as a factor affecting use of mobile voice calls (Humphreys, 2005; Murtagh, 2002; Srivastava, 2005) and mobile voice input applications (Reis et al., 2008; Williamson, 2012). In addition, prior studies have shown that the type of information that is being transmitted has an influence on the data-sharing patterns of smartphone users (Khalil & Connelly, 2006; Marques et al., 2012; Olson et al., 2005). To examine whether these factors also affected usage patterns of smartphone input modes, an online survey was con- ducted to gather usage ratings for VAPA and keyboard input to transmit private versus nonprivate information as a function of the location (private vs. public space). Table 6 provides a quick summary of hypotheses that were supported and those that were not supported. Results from the survey were consistent with findings from previous research indicating that social con- text and type of information influences the likelihood of using VAPA technology, as is explained next. These findings could be applied to inform smartphone design decisions to increase usage of the VAPA, especially for disclosing private information in public locations.

8.1. Type of Information Transmitted As predicted, participants preferred to enter nonprivate infor-

mation compared to private information. This finding is con- sistent with previous research, which showed that smartphone users were more unwilling to disclose digital private informa- tion (Karlson et al., 2009; Khalil & Connelly, 2006; Marques et al., 2012). It also reiterates the fact that smartphone users distinguish between the different types of information they transmit and that they are more protective of their private infor- mation. However, the degree of willingness to transmit private

322 A. EASWARA MOORTHY AND K.-P. L. VU

TABLE 6 Summary of Hypotheses and Results

Hypothesis Result

H1: Participants will be more likely use their smartphone keyboard to transmit information.

Supported

H2: Participants will be more likely to transmit information in private than in public locations.

Supported

H3: Participants will be more likely to transmit nonprivate information than private information.

Supported

H4: Participants will be more likely to use their VAPA in private locations than in public locations. However, they will be equally likely to use the keyboard in private and public locations.

Supported

Partially supported. Participants were more likely to use the VAPA and the keyboard for nonprivate compared to private information entry; however, this difference was lower in the keyboard condition.

Partially supported. Participants were more likely to enter nonprivate information than private information in both public and private locations; however, this difference was lower in the private locations condition.

H7: Public use of VAPA will be less likely than private use of VAPA to input all types of information. However, participants will be even less likely to publicly use the VAPA to input private information. In the keyboard condition, participants will be equally likely to use the keyboard at both locations to input both types of information.

Not supported.

Note. VAPA = Voice Activated Personal Assistant.

information is dependent on the input mode used to enter infor- mation and where the user is located, as is discussed in the next sections.

8.2. Input Mode for Information Transmission Participants preferred using the smartphone keyboard more

than the VAPA to enter information into their smartphone, suggesting that manual typing of information is favored over verbalizing it. This result is contradictory to the finding that users would be more likely to use personal information manage- ment agents supporting natural language through voice mode than those that did not (L. Zhou, Mohammed, & Zhang, 2012). Apart from the assistive system design differences between the studies that could have contributed to the contradictory results, there are several other reasons for the participants’ higher usage ratings for the keyboard over the VAPA in the present study.

As reported in their smartphone usage habits, participants were more comfortable using the keyboard (with a comfort level rating of 6.5 on a scale of 1 to 7, with 7 standing for very

comfortable) than the VAPA (with a comfort level rating of 5.1). Lower comfort ratings might be evidence to the novelty of the VAPA application due to its recent inclusion as a stan- dard feature in smartphones. Participants might not be used to interacting with their smartphones using voice. However, use of VAPA was expected to be more prevalent, given that it is a more efficient means to complete tasks than use of the key- board. This is especially the case given that the participants were instructed to assume that technological and human errors associated with current VAPAs were not an issue here. Thus, participants might not be discounting technical issues encoun- tered when using the VAPA (which was given as a reason for avoiding the use of VAPA by at least 51% of the participants), reporting a relatively moderated optimism about the effective- ness voice technology. Previous studies have shown that tasks could be completed faster with the VAPA than with the key- board on intelligent assistive systems (Cox, Cairns, Walton, & Lee, 2008; L. Zhou et al., 2012). The participants in this sur- vey also reported noticing this benefit of VAPA. That is, in the survey, when probed about their previous experience using the

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 323

TABLE 7 Correlation Between Voice Activated Personal Assistant Likelihood of Usage and Social Acceptability

Texting

Finding Directions to a

Location Calling a Contact

Searching the Web

Updating Calendar

Setting Reminder

Private information Restaurant 0.345∗∗ 0.261∗ 0.474∗∗ 0.208 0.192 0.176 Home 0.269∗ 0.272∗ 0.342∗∗ 0.329∗∗ 0.098 0.072 Work desk 0.315∗∗ 0.231∗ 0.481∗∗ 0.156 0.176 0.143 Supermarket 0.151 0.215 0.385∗∗ 0.192 0.116 0.159 Car 0.263∗ 0.100 0.170 0.265∗∗ −0.040 0.098 Work lounge 0.214 0.449∗∗ 0.374∗∗ 0.173 0.278∗ 0.420∗∗

Nonprivate Information Restaurant 0.267∗ 0.488∗∗ 0.374∗∗ 0.290∗∗ 0.250∗ 0.349∗∗ Home 0.237∗ 0.329∗∗ 0.220 0.100 0.078 0.187 Work desk 0.299∗∗ 0.573∗∗ 0.418 0.376∗∗ 0.411∗∗ 0.283∗ Supermarket 0.371∗∗ 0.621∗∗ 0.359∗∗ 0.320∗∗ 0.306∗∗ 0.400∗∗ Car 0.111 0.302∗∗ 0.078 0.181 0.048 0.100 Work lounge 0.375∗∗ 0.545∗∗ 0.540∗∗ 0.391∗∗ 0.381∗∗ 0.362∗∗

Note. VAPA = Voice Activated Personal Assistant. ∗p < .05. ∗∗p < .001.

VAPA, only 24% of the participants reported a preference for classic keyboard methods of input over VAPA; 76% of the par- ticipants had reported that VAPA was easier to use, and 62.7% indicated that it was faster to use (also shown by L. Zhou et al., 2012). Yet participants leaned toward the keyboard for enter- ing information. This pattern suggests that perceived ease of use, a classic construct advocated by the TAM, UTAUT, and the MOPTAM (see, e.g., Venkatesh, 2013), traditionally applied to standard desktop applications and simple mobile phone ser- vices, might not play as much of a significant role in the usage of smartphone voice applications such as the VAPA.

It is unclear what aspect of the VAPA (e.g., verbalization, voice command phrases failing to match users’ habits, inter- action with the robotic assistant, or conspicuousness of the verbal input) prevents users from using it. The act of having to speak out the voice commands might pose problems for users. The voice command phrases presented to them in the study might significantly differ from the language they would use to interact with the VAPA. The differences could cause them to avoid using these phrases and therefore decrease the chances of VAPA usage. Having to interact with a robotic personal assistant might evoke privacy concerns, inducing performance changes as a function of the user knowing that they are being observed, as in the Hawthorne effect (Sullivan, 2009, p. 232). For example, Calo (2009) made the following assessment in his discussion on user attitude toward interacting with anthropo- morphic technology: “One of the well-documented effects on users of interfaces and devices that emulate people is the sensa- tion of being observed and evaluated. Their presence can alter

[users’] attitude, behavior, and physiological state” (p. 809). Users might also want to use discreet methods of information entry such as the keyboard more than conspicuous ones such as the VAPA.

Any of the aforementioned reasons could justify the find- ing that participants were more likely to use the VAPA to enter nonprivate information than private information. However, it was surprising to find this effect of information type repeated in the keyboard condition, although the effect was not as strong. For private information, participants were only somewhat likely to use their smartphones to transmit information, and this did not depend on the input modes. Although the keyboard seems more suited for transmitting this type of information (there is little risk of other people overhearing the information as it is being inputted), the fact that users are reluctant to use the key- board indicates that PII is a type of information that people do not want transmitted over mobile devices in general. In addi- tion, correlation analyses between mean likelihood of usage and social acceptability of using the VAPA indicate that one reason why private information is not being transmitted is that users find it to be socially unacceptable. Specifically, it was found that tasks involving participant input of nonprivate information showed more significant positive correlations (i.e., high usage with high social acceptability) compared to tasks involving transmission of private information.

Participants’ sensitivity to the type of information, especially in the VAPA context, shows that the privacy construct might play a crucial role in the usage of smartphone voice applica- tions. Privacy is reflected not only in the type of information

324 A. EASWARA MOORTHY AND K.-P. L. VU

being transmitted but also in the mode of information input and the location where the users are transmitting the informa- tion. Thus, a technology adoption model for smartphones might have to account for information type and input modes to assess how they will be used, along with the location of the user, as explained in the next section.

8.3. Location for Information Transmission Participants were more likely to transmit information in pri-

vate locations than in public locations. This finding is probably due to the possibility that participants are more likely to be surrounded by strangers in the colocated public space than in private locations. Because users had been asked to assume that the VAPA would detect their voice with the first issuance of the voice command, the confounding effect of background noise was eliminated. From this point of view, this result is consis- tent with findings from prior studies that found that mobile phone users were more guarded of their spoken conversations (Murtagh, 2002) and digital information (Khalil & Connelly, 2006; Marques et al., 2012; Olson et al., 2005) in the presence of strangers. Therefore, presence of strangers might prevent users from using their smartphones to perform tasks.

The technology adoption models UTAUT and MOPTAM define the construct of social influence to mean “the degree to which an individual perceives that important others believe he or she should use the new system” (Venkatesh, 2013). However, the present findings indicate that they might have to incorporate the presence of strangers as part of social influence, as it was found that being in a public space influences a user’s intention to use or reject mobile voice applications such as the VAPA. Another interpretation is to consider strangers in social settings as “important others” who can drive behavioral intention to use mobile phone voice applications.

As predicted, location also had an effect on the input mode chosen to complete tasks. The hypothesis that participants would be more likely to use the VAPA in private than public locations, but no effect of location when using the keyboard was supported. This finding suggests that presence of people in the colocated space and the resulting users’ perception of attention to their conspicuous actions deter users from engag- ing the VAPA feature of their smartphone. At both public and private locations, participants were equally likely to use their keyboard to type—a prevalent common action that would go unnoticed. However, they were more likely to use their VAPA in private than public locations, because they would not want to conspicuously verbalize information, especially private infor- mation. This argument can be supported by the fact that 41% of the participants reported feeling uncomfortable and 13% felt embarrassed using the VAPA in front of unknown peo- ple, with 55% of the participants reporting privacy concerns as the primary reason for avoiding VAPA use. These findings on user attitude toward public use of VAPA are consistent with those reported by Siftar (2012b). Thus, we can infer that using

the VAPA is perceived to draw attention from others in the colocated space. Therefore, it can be expected that potential attention to information might prevent users more from disclos- ing private information than nonprivate information in public locations. It follows that appropriate measures need to be taken to protect sharing of private information, especially with the VAPA and in the public context as described in smartphone design implications.

However, it was surprising to see that participants were also more unwilling to disclose certain private information, even in private locations. This finding indicates that participants are generally averse to sharing private information, in this case PII. Again, this finding implies that the tasks that are to be supported by the VAPA should take into account the type of information that is to be transmitted.

Correlation analyses also followed this pattern indicating that entering nonprivate information with VAPA was acceptable in public locations; however, low significant correlation ratings for nonprivate information in private locations might be due to the insignificance of social acceptability, as there is no pres- ence of social company to judge smartphone usage behavior. In these analyses, the work lounge and work desk locations fol- lowed the correlation patterns for public locations, indicating that even when one is alone, the work environment is considered a public location. Finally, all the tasks received a higher number of significant positive correlations for entering nonprivate infor- mation compared to private information, again indicating that users found it more socially acceptable to transmit nonprivate than private information through their mobile phones.

In case of entering private information, there was a lower number of significant correlations in public locations, as partici- pants were unlikely to use the VAPA to enter private information in these places. The restricted range in VAPA usage scores might have contributed to the nonsignificant findings. However, there was a higher number of significant, positive correlations in private locations, which might indicate that participants find it acceptable to enter private information in private locations. Finally, the tasks of texting, finding directions to a location, and calling contacts received a high number of significant, positive correlation scores for the transmission of private information, indicating that these are the common smartphone tasks where users find it socially acceptable to transmit private information.

8.4. Limitations of the Study The present study used an online survey to gather user

reported usage and preference data. The survey method is based on the self-report of participant behavior and may not reflect data patterns gathered by more realistic field studies. In addi- tion, users may not be attending to all aspects of the survey when filling it out. We found that a large proportion (36.6%) of data collected from AMT did not meet the quality require- ments of the survey. Therefore, we limited this possibility in the present study and recommended that researchers embed

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 325

quality control questions in their online surveys to ensure that the participants are paying attention to the survey questions.

The present survey also asked users to make various assump- tions that eliminated technological and human errors related to the smartphone tasks when making their assessments about their likelihood of VAPA and keyboard usage. However, voice recog- nition errors can be common, and the phrase used to input the command may influence whether it would be interpreted cor- rectly by the technology. Moreover, other factors can influence users’ preference for voice input (e.g., K. Lee & Lai, 2005). Given that voice recognition is a relatively recent introduction, substantial improvements in natural language processing and associated technology should dramatically reduce these errors in the future. In the present study, we had participants assume that their commands would be understood by the VAPA. As a result, the participants’ usage reports may reflect usage under the best case scenario rather than actual usage.

8.5. Summary and Conclusion This survey examined whether the social context and type of

information transmitted in a task influenced the usage patterns of the smartphone VAPA. Participants reported the likelihood of their using the VAPA and the smartphone keyboard to enter private versus nonprivate information as a function of the loca- tion (private vs. public space). Results from this survey showed that participants were more cautious of disclosing private infor- mation than nonprivate information. This effect was amplified in the social context of public locations and when using con- spicuous methods of information input such as the VAPA. Correlations between likelihood of usage of VAPA and the social acceptability of using it were positive and followed simi- lar patterns of smartphone usage. Thus, social context and type of information were found to have an influence also on the ver- bal transmission of information through the VAPA. Findings from this study will contribute to smartphone design decisions to optimize usage of the VAPA, especially to transmit private information in the social context.

8.6. Smartphone Design Implications It should be noted that users are sensitive to the nature of both

verbal and digital information they transmit through their smart- phones. Increasing user trust and comfort for sharing private information through classic input methods such as the keyboard and new innovations such as the VAPA should be considered when designing smartphones. Users tend to restrict relaying their private information in general, and even more so in the presence of strangers. Thus, the tasks supported by VAPA must take these factors into account. For example, all tasks—texting, getting directions, calling a contact, searching the web, set- ting up meetings, and setting up reminders—highly supported

the information entry of nonprivate information. On the other hand, only texting, finding directions to a location, and call- ing contacts were considered acceptable for transmitting private information.

Social context also played a role in the type of input mode used at different locations. Because the VAPA tends to be used in private settings, users might benefit more from functional- ities and design for use in the private domain. For example, results from this study are consistent with research on voice- activated smartphone applications to control automobiles and home devices as described by Knight (2012).

To optimize VAPA and smartphone keyboard usage for dis- closing private information in public locations, the following design guidelines should be followed:

1. Users should be able to input PII such as their credit card number and home address (especially using the keyboard) in their smartphone Personal Information Management system on a one-time basis for later retrieval.

2. PII should be securely stored at the network level in order to prevent collocated strangers from hearing or seeing the information while user is interacting with the smartphone.

3. Prestored PII should be available for retrieval by smartphone programs and applications. However, they should not be able to track or record this information during transmission.

4. PII should be retrieved through generic voice commands (e.g., the voice command “Take me home” should prompt the smartphone GPS to guide the user to the prestored home address) or keyboard shortcuts (e.g., @home for home address) that do not require users to verbally repeat or type the details.

5. Information output through the VAPA should ensure that PII details are not verbalized (e.g., VAPA feedback for GPS route to the home address could be “Taking you home” instead of “Taking you to 1250 Pine Street, Long Beach”). Feedback visualized on the screen when using the VAPA and keyboard should also only show such generic details.

6. Various privacy layers should be implemented to protect users’ private information from the general public. For example, the smartphone’s GPS could be used to track the location of the users, evaluate their social context, and auto- matically switch to a corresponding public mode (in which entry and display of PII would be controlled) or a pri- vate mode (in which entry and display of PII would not be restricted).

ACKNOWLEDGEMENTS A subset of the data from this project was presented at

the Human–Computer Interaction International Convention in 2014 and appear as part of its proceedings (Easwara Moorthy & Vu, 2014).

326 A. EASWARA MOORTHY AND K.-P. L. VU

REFERENCES Agarwal, R., & Prasad, J. (1999). Are individual differences germane to

the acceptance of new information technologies? Decision Sciences, 30, 361–391.

Aldhaban, F. (2012). Exploring the adoption of smartphone technology: Literature review. Proceedings of the PICMET 2012 Conference on Technology Management for Emerging Technologies, 2758–2770.

Barkhuus, L. (2012). The mismeasurement of privacy: Using contextual integrity to reconsider privacy in HCI. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 367–376.

Barkhuus, L., & Dey, A. K. (2003). Location-based services for mobile telephony: A study of users’ privacy concerns. INTERACT , 3, 702–712.

Brandtzæg, P. B., Lüders, M., & Skjetne, J. H. (2010). Too many Facebook “friends”? Content sharing and sociability versus the need for pri- vacy in social network sites. International Journal of Human–Computer Interaction, 26, 1006–1030.

Calo, M. R. (2009). People can be so fake: A new dimension to privacy and technology scholarship. Penn State Law Review, 114, 809–855.

Campbell, S. W. (2007). Perceptions of mobile phone use in public settings: A cross-cultural comparison. International Journal of Communication, 1, 738–757.

Campbell, S. W., & Russo, T. C. (2003). The cocial construction of mobile telephony: An application of the social influence model to perceptions and uses of mobile phones within personal communication networks. Communication Monographs, 70, 317–334.

Chang, I., Hwang, H. G., Hung, W. F., & Li, Y. C. (2007). Physicians’ accep- tance of pharmacokinetics-based clinical decision support systems. Expert Systems with Applications, 33, 296–303.

Cox, A. L., Cairns, P. A., Walton, A., & Lee, S. (2008). Tlk or txt? Using voice input for SMS composition. Personal and Ubiquitous Computing, 12, 567–588.

Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989). User acceptance of com- puter technology: A comparison of two theoretical models. Management Science, 35, 982–1003.

Easwara Moorthy, A., & Vu, K.-P. L. (2014). Voice activated personal assis- tant: Acceptability of use in the public space. In S. Yamamoto (Ed.), Human Interface and the Management of Information 2014, Part II, Lecture Notes in Computer Science, 8522, 324–334. Cham, Switzerland: Springer International.

Fischer, A., Price, K., & Sears, A. (2005). Speech-based text entry for mobile handheld devices: An analysis of efficacy and error correction techniques for server-based solutions. International Journal of Human–Computer Interaction, 19, 279–304.

Goffman, E. (1959). The presentation of self in everyday life. New York, NY: Doubleday Anchor.

Goffman, E. (1963). Behavior in public places: Notes on the social organization of gatherings. New York, NY: Free Press.

Gupta, B., Dasgupta, S., & Gupta, A. (2008). Adoption of ICT in a government organization in a developing country: An empirical study. The Journal of Strategic Information Systems, 17, 140–154.

Humphreys, L. (2005). Cellphones in public: Social interactions in a wireless era. New Media & Society, 7, 810–833.

Hung, Y. H., Wang, Y. S., & Chou, S. C. T. (2007). User acceptance of e-government services. Proceedings of Pacific Asia Conference on Information Systems (Paper 97). Auckland, New Zealand. Retrieved from http://aisel.aisnet.org/pacis2007/97

Igbaria, M., Zinatelli, N., Cragg, P., & Cavaye, A. L. (1997). Personal com- puting acceptance factors in small firms: A structural equation model. MIS Quarterly, 21, 279–305.

Ito, M., & Okabe, D. (2005). Technosocial situations: Emergent structuring of mobile email use. In M. Ito, D. Okabe, & M. Matsuda (Eds.), Personal, portable, pedestrian: Mobile phones in Japanese life (pp. 257–273). Cambridge, MA: MIT Press.

Karlson, A. K., Brush, A. J., & Schechter, S. (2009, April). Can I borrow your phone?: Understanding concerns when sharing mobile phones. Proceedings of the CHI 2009 Conference on Human Factors in Computer Systems, 1647–1650.

Khalil, A., & Connelly, K. (2005). Context-aware configuration: A study on improving cell phone awareness. In A. Dey, B. Kokinov, D. Leake, & R. Turner (Eds.), Modeling and using context (pp. 197–209). Berlin, Germany: Springer.

Khalil, A., & Connelly, K. (2006). Context-aware telephony: Privacy pref- erences and sharing patterns. Proceedings of the 2006 20th anniversary conference on Computer Supported Cooperative Work, 469–478.

Knight, W. (2012, May 29). Where speech recognition is going | MIT Technology Review. Retrieved from http://www.technologyreview.com/ news/427793/where-speech-recognition-is-going/

Lasén, A. (2006). How to be in two places at the same time? Mobile phone uses in public places. In J. R. Höflich & M. Hartmann (Eds.), Mobile commu- nication in everyday life: Ethnographic views, observations and reflections (pp. 227–252). Berlin, Germany: Frank & Timme.

Laumer, S., Eckhardt, A., & Trunk, N. (2010). Do as your parents say?— Analyzing IT adoption influencing factors for full and under age applicants. Information Systems Frontiers, 12, 169–183.

Lee, K., & Lai, J. (2005). Speech versus touch: A comparative study of the use of speech and DTMF keypad for navigation. International Journal of Human–Computer Interaction, 19, 343–360.

Lee, W. J., Kim, T. U., & Chung, J. (2002, July). User acceptance of the mobile Internet. Paper presented at the meeting of M-Business 2002, Athens, Greece.

Ling, R. (1997). One can talk about common manners!: The use of mobile tele- phones in inappropriate situations. In L. Haddon (Ed.), Themes in mobile telephony: Final Report of the COST 248 Home and Work group (pp. 73–96). Stockholm, Sweden: Telia.

Ling, R. (2002, July). The social juxtaposition of mobile telephone conversa- tions and public spaces. Paper presented at the Conference on the Social Consequence of Mobile Telephones, Chunchon, Korea.

Ling, R., & Haddon, L. (2003). Mobile telephony, mobility, and the coordina- tion of everyday life. In J. E. Katz (Ed.), Machines that become us: The social context of personal communication technology (pp. 245–265). New Brunswick, NJ: Transaction.

Love, S., & Perry, M. (2004). Dealing with mobile conversations in public places: some implications for the design of socially intrusive technologies. Proceedings of the CHI 2004 Extended Abstracts on Human Factors in Computer Systems, 1195–1198.

Malhotra, Y., & Galletta, D. F. (1999). Extending the technology acceptance model to account for social influence: theoretical bases and empirical val- idation. Proceedings of the 32nd Annual Hawaii International Conference on System Sciences, 6–19.

Marques, D., Duarte, L., & Carriço, L. (2012). Privacy and secrecy in ubiquitous text messaging. Proceedings of the 14th international confer- ence on Human–Computer Interaction With Mobile Devices and Services Companion, 95–100.

Mathieson, K. (1991). Predicting user intentions: comparing the technology acceptance model with the theory of planned behavior. Information Systems Research, 2, 173–191.

Monk, A., Carroll, J., Parker, S., & Blythe, M. (2004). Why are mobile phones annoying? Behaviour & Information Technology, 23, 33–41.

Murtagh, G. M. (2002). Seeing the “rules”: Preliminary observations of action, interaction and mobile phone use. In B. Brown, N. Green, & R. Harper (Eds.), Wireless world: Social and international aspects of the mobile age (pp. 81–91). London, UK: Springer.

Nissenbaum, H. (2004). Privacy as contextual integrity. Washington Law Review, 79, 101–158.

Olson, J. S., Grudin, J., & Horvitz, E. (2005). A study of preferences for sharing and privacy. Proceedings of the CHI 2005 Extended Abstracts on Human Factors in Computer Systems, 1985–1988.

Palen, L., Salzman, M., & Youngs, E. (2001). Discovery and integration of mobile communications in everyday life. Personal and Ubiquitous Computing, 5, 109–122.

Reis, T., de Sá, M., & Carriço, L. (2008). Multimodal interaction: Real con- text studies on mobile digital artefacts. In A. Pirhonen, & S. Brewster (Eds.), Haptic and audio interaction design (pp. 60–69). Berlin, Germany: Springer.

http://aisel.aisnet.org/pacis2007/97

http://www.technologyreview.com/news/427793/where-speech-recognition-is-going/

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 327

Renaud, K., & Van Biljon, J. (2008). Predicting technology acceptance and adoption by the elderly: a qualitative study. Proceedings of the 2008 annual research conference of the South African Institute of Computer Scientists and Information Technologists on IT research in developing countries: riding the wave of technology, 210–219.

Roberts, K. G., & Pick, J. B. (2004). Technology factors in corporate adop- tion of mobile cell phones: A case study analysis. Proceedings of the IEEE 37th Annual Hawaii International Conference on System Sciences, 90287–90296.

Ruuska-Kalliokulju, S., Schneider-Hufschmidt, M., Väänänen-Vainio-Mattila, K., & Von Niman, B. (2001). Shaping the future of mobile devices. Proceedings of Mobile HCI 2001: Third International Workshop on Human Computer Interaction with Mobile Devices, 1740–1829.

Siftar, J. (2012a). The Adoption and Appropriation of Siri (Unpublished master’s dissertation). University College London, University of London, London, UK.

Siftar, J. (2012b). Siri loves you, but do you love Siri? [PowerPoint slides]. Retrieved from http://www.slideshare.net/JiriSiftar/siri-loves-you-but-do- you-love-siri

Srivastava, L. (2005). Mobile phones and the evolution of social behaviour. Behaviour & Information Technology, 24, 111–129.

Sullivan, L. (Ed.). (2009). The SAGE glossary of the social and behavioral sciences. Thousand Oaks, CA: Sage.

Szajna, B. (1996). Empirical evaluation of the revised technology acceptance model. Management Science, 42, 85–92.

Tang, K. P., Lin, J., Hong, J. I., Siewiorek, D. P., & Sadeh, N. (2010). Rethinking location sharing: exploring the implications of social-driven vs. purpose-driven location sharing. Proceedings of the 12th ACM International Conference on Ubiquitous Computing, 85–94.

Taylor, A. S., & Harper, R. (2003). The gift of the gab?: A design oriented soci- ology of young people’s use of mobiles. Computer Supported Cooperative Work, 12, 267–296.

Taylor, S., & Todd, P. A. (1995). Understanding information technology usage: A test of competing models. Information Systems Research, 6, 144–176.

Tsai, J. Y., Kelley, P., Drielsma, P., Cranor, L. F., Hong, J., & Sadeh, N. (2009). Who’s viewed you?: The impact of feedback in a mobile location-sharing application. Proceedings of the CHI 2009 Conference on Human Factors in Computer Systems, 2003–2012.

Turner, M., Love, S., & Howell, M. (2008). Understanding emotions experi- enced when using a mobile phone in public: The social usability of mobile (cellular) telephones. Telematics and Informatics, 25, 201–215.

Van Biljon, J., & Kotzé, P. (2007). Modelling the factors that influence mobile phone adoption. Proceedings of the 2007 annual research conference of the South African institute of computer scientists and information technologists on IT research in developing countries, 152–161.

Venkatesh, V. (2013, June 7). Technology acceptance. Retrieved from http:// www.vvenkatesh.com/it/organizations/theoretical_models.asp

Venkatesh, V., & Davis, F. D. (2000). A theoretical extension of the technology acceptance model: four longitudinal field studies. Management Science, 46, 186–204.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User accep- tance of information technology: Toward a unified view. MIS Quarterly, 27, 425–478.

Verkasalo, H. (2007). A cross-country comparison of mobile service and handset usage (Licentiate’s thesis). Helsinki University of Technology, Networking Laboratory, Helsinki, Finland.

Wei, R., & Leung, L. (1999). Blurring public and private behaviors in pub- lic space: policy challenges in the use and improper use of the cell phone. Telematics and Informatics, 16, 11–26.

Williams, M. D., Rana, N. P., Dwivedi, Y. K., & Lal, B. (2011, June). Is UTAUT really used or just cited for the sake of it? A systematic review of citations of UTAUT’s originating article. Presented at the European Conference on Information Systems, Helsinki, Finland.

Williamson, J. R. (2012). User experience, performance, and social acceptabil- ity: usable multimodal mobile interaction (Unpublished doctoral disserta- tion). University of Glasgow, Glasgow, Scotland.

Zhou, L., Mohammed, A. S., & Zhang, D. (2012). Mobile personal information management agent: Supporting natural language interface and application integration. Information Processing & Management, 48, 23–31.

Zhou, T., Lu, Y., & Wang, B. (2010). Integrating TTF and UTAUT to explain mobile banking user adoption. Computers in Human Behavior, 26, 760–767.

ABOUT THE AUTHORS Aarthi Easwara Moorthy has a master’s degree in Human Factors Psychology from California State University, Long Beach. She currently enjoys working as a User Experience Researcher at UserTesting, conducting remote unmoderated and moderated UX studies to evaluate websites and mobile applications across a variety of industries every day.

Kim-Phuong L. Vu is Professor of Psychology at California State University Long Beach. She obtained her Ph.D. in Psychological Sciences from Purdue University. She has a broad research program relating to Human Factors, human– computer interaction, and human-systems integration.

http://www.slideshare.net/JiriSiftar/siri-loves-you-but-do-you-love-siri

http://www.vvenkatesh.com/it/organizations/theoretical_models.asp

328 A. EASWARA MOORTHY AND K.-P. L. VU

APPENDIX Survey Questionnaire

Smartphone Usage Preferences Survey Electronic Consent

The purpose of this survey is to understand smartphone users’ preferences when interacting with their smartphone to enter informa- tion at various locations. This is a noncommercial research survey conducted by a graduate student in the Psychology Department of California State University, Long Beach as part of a Master’s thesis. You are invited to take this survey because you identified yourself as a smartphone user living in the U.S. with prior experience using a smartphone voice assistant (such as Siri, S Voice, Google Now, etc.). In this survey, you will be asked to provide your demographic information, smartphone usage information, and preferences for using your smartphone in various contexts. Your responses will be kept anonymous. This survey should take less than 30 minutes to finish and you will be paid $0.75 for completing the survey, provided that you correctly enter a special code into the verification textbox in the Amazon Turk HIT description for this survey. Please pay attention to all the questions in the survey, since one of the questions will contain this special code. Your participation is voluntary. You may choose to withdraw your participation at anytime. There are no foreseeable risks or direct benefits to you from your participation. Knowledge gained by researcher might be used to inform design decisions of smartphone applications.

Smartphone Usage Habits

We are not judging you, or your knowledge and abilities. Therefore, please answer the questions truthfully and accurately.

==QUESTION 1. Choose the smartphone you use. (If you use more than one smartphone, pick the most frequently used one.) 1. iPhone (example: 4S or 5) 2. Samsung phone (example: Galaxy S III, Galaxy S III Mini, Galaxy S4, Galaxy S II Plus, Galaxy Note II, Galaxy Stellar, or

Galaxy Grand) 3. Non-Samsung Android phone (example: Nexus 4, HTC One, LG Optimus, etc.) 4. Decline to answer 5. Other (please specify) ____________________

==QUESTION 2. How long have you been using this smartphone? 1. Less than 1 week 2. Less than 1 month 3. More than 1 month but less than 1 year 4. 1–2 years 5. 2–3 years 6. More than 3 years 7. Decline to answer 8. Other (please specify) ____________________

==QUESTION 3. Choose the voice assistant you have been using on your smartphone. 1. Siri 2. S Voice 3. Google Now 4. Decline to answer 5. Other (please specify) ____________________

==QUESTION 4. Does your smartphone have a touch screen or physical keyboard? 1. Touch screen keyboard 2. Physical keyboard 3. Both 4. None 5. Decline to answer

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 329

==QUESTION 5. Rate how comfortable you are with using the voice assistant and keyboard in your smartphone:

1 - Not comfortable at all to use 2 3

4 – Somewhat comfortable 5 6

7 – Very comfortable

Decline to answer

Voice assistant Keyboard

Smartphone Voice Assistant Usage For each question in this page, imagine yourself at the location mentioned in the question and rate how likely you are to use your smartphone’s VOICE ASSISTANT to complete the listed tasks in that location. You will be SAYING OUT LOUD THE FOLLOWING PHRASES TO YOUR VOICE ASSISTANT in order to complete tasks.

While answering these questions, assume the following: • The term voice assistant refers to your smartphone voice assistant application such as Siri, Google Now, or S Voice that

enables use to complete tasks through voice mode

• Your smartphone voice assistant can:

◦ Detect your voice on first try against background noise ◦ Understand your language/accent ◦ Provide audible responses ◦ Provide accurate and adequate responses and information

• Your smartphone has:

◦ Sufficient battery power to run the smartphone voice assistant ◦ Full service reception

• You:

◦ Can freely operate your smartphone to input voice commands and receive voice feedback. ◦ Input the voice query without mistakes

NOTE: The following questions were asked individually along with the scale in the actual survey. To save space in this publication, the stem questions will be presented below and the scale will be provided afterwards.

==QUESTIONS 6–11:

Imagine yourself AT A RELATIVELY QUIET, BUT CROWDED RESTAURANT. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location. Imagine yourself ALONE AT HOME. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location. Imagine yourself AT YOUR DESK AT WORK. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location. Imagine yourself WAITING IN A LONG LINE AT THE SUPERMARKET. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location. Imagine yourself ALONE IN YOUR PARKED CAR. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location. Imagine yourself IN THE LOUNGE AT YOUR WORK. Rate how likely you are to use your smartphone’s voice assistant to complete the following tasks in this location.

330 A. EASWARA MOORTHY AND K.-P. L. VU

Smartphone Voice Assistant Usage

1 2 3 4 5 6 7 Not likely at

all to use May or may

not use Extremely

likely to use Decline to

answer

Texting private information (example: “Text John/Jane ‘My social security number is 652 341 9518’”)

Texting nonprivate/other information (example: “Text John/Jane

‘How was your day today?’”) Providing private information while finding directions to

a location (example: “Take me home to 1250 Pine Street, Long Beach”)

Providing nonprivate/other information while finding directions to a location (example: “Show me directions to the nearest coffee shop”)

Providing private information while calling a contact (example: “Call John/Jane Carpenter”)

Providing nonprivate/other information while calling a contact (example: “Call John/Jane”)

Searching the web for private information (example: “Find famous people born on my birthday, July 19, 1973”)

Searching the web for nonprivate/other information (example: “Search the web for weather tomorrow.”)

Updating calendar with private information (example: “Set up meeting to change car license plate number 4CTD987 on the 2nd at 9 AM”).

Updating calendar with nonprivate/others information (example: “Set up meeting with John/Jane on the 4th at 3 PM”).

Setting a reminder for private information (example: “Remind me to add new credit card number, 3371 2452 0221 6633, to finance paperwork”)

Setting a reminder for nonprivate/other information (example: “Remind me to buy cereal on Monday”)

For quality control purposes, please select “May or may not use” for this question.

Smartphone Keyboard Usage For each question in the next page, imagine yourself at the location mentioned in the question and rate how likely you are to use your smartphone’s TOUCH SCREEN OR PHYSICAL KEYBOARD to complete the listed tasks in that location. You will be TYPING THE FOLLOWING PHRASES ON THE KEYBOARD in order to complete tasks.

While answering these questions, assume the following:

• Your smartphone touch screen/physical keyboard can:

◦ Accurately detect your typing

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 331

◦ Provide accurate and adequate responses and information • Your smartphone has:

◦ Sufficient battery power to allow typing to complete the tasks ◦ Full service reception

• You:

◦ Can freely operate your smartphone to input manual commands and receive visual feedback ◦ Input the text query without mistakes

Smartphone Keyboard Usage

NOTE: The following questions were asked individually along with the scale in the actual survey. To save space in this publication, only the stem questions will be presented below. The scale was the same as that used for the voice assistant usage questions presented earlier.

==QUESTIONS 12–17: Imagine yourself AT A RELATIVELY QUIET, BUT CROWDED RESTAURANT. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location. Imagine yourself ALONE AT HOME. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location. Imagine yourself AT YOUR DESK AT WORK. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location. Imagine yourself WAITING IN A LONG LINE AT THE SUPERMARKET. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location. Imagine yourself ALONE IN YOUR PARKED CAR. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location. Imagine yourself IN THE LOUNGE AT YOUR WORK. Rate how likely you are to use your smartphone’s touch screen or physical keyboard to complete the following tasks in this location.

For questions on this page, reflect on your previous experiences with your smartphone voice assistant.

==QUESTION 18. What tasks do you perform with your smartphone voice assistant? Check all that applies. � make calls � send text messages � send e-mails � find places on map � get weather info � get contact info � search the web � play music � set calendar event � set alarm � set reminder � take notes � get sports updates � search for local listings (businesses, sites, etc.) � post on social networks � make restaurant reservations � find movies � find friends � get stock info � get simple facts, conversions, and calculations � decline to answer � other (please specify) ____________________

332 A. EASWARA MOORTHY AND K.-P. L. VU

==QUESTION 19. Why do you use your voice assistant for certain tasks? � It’s easier � It’s faster � It’s more fun � It’s more challenging � It’s more “Star Trek” � Decline to answer � Other (please specify) ____________________

==QUESTION 20. How do you feel when using voice assistant in front of unknown people? 1. Proud 2. Comfortable 3. Indifferent 4. Uncomfortable 5. Embarrassed 6. Decline to answer 7. Other (please specify) ____________________

==QUESTION 21. What are your reasons for not using voice assistant? � Being misunderstood � Unsatisfactory answers � Difficulty of use � Discomfort of use � Language issues � Privacy concerns � Preference for classic methods � I didn’t know it was there � Decline to answer � Other (please specify) ____________________

Social Acceptability of Voice Assistants

==QUESTION 22. Imagine yourself in each of locations and rate how socially acceptable you think it would be to use your smartphone’s VOICE ASSISTANT to complete following tasks in each of the following locations on a scale from 1 to 7, with 1 being not acceptable at all to 7 being very acceptable.

At a relatively quiet, but crowded

restaurant When alone

at home At your desk

at work

When waiting in a long line at

the supermarket

Alone in your parked car

In the lounge at your work

Texting private information (example: “Text John/Jane ‘My social security number is 652 341 9518’”)

1 – 7 or Decline to answer

Texting nonprivate/other information (example: “Text John/Jane ‘How was your day today?’”)

1 – 7 or Decline to answer

(Continued)

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 333

(Continued).

At a relatively quiet, but crowded

restaurant When alone

at home At your desk

at work

When waiting in a long line at

the supermarket

Alone in your parked car

In the lounge at your work

Providing private information while finding directions to a location (example: “Take me home to 1250 Pine Street, Long Beach”)

1 – 7 or Decline to answer

Providing nonprivate/other information while finding directions to a location (example: “Show me directions to the nearest coffee shop”)

1 – 7 or Decline to answer

Providing private information while calling a contact (example: “Call John/Jane Carpenter”)

1 – 7 or Decline to answer

Providing nonprivate/other information while calling a contact (example: “Call John/Jane”)

1 – 7 or Decline to answer

Searching the web for private information (example: “Find famous people born on my birthday, July 19, 1973”)

1 – 7 or Decline to answer

Searching the web for nonprivate/other information (example: “Search the web for weather tomorrow.”)

1 – 7 or Decline to answer

Updating calendar with private information (example: “Set up meeting to change car license plate number 4CTD987 on the 2nd at 9 AM”).

1 – 7 or Decline to answer

Updating calendar with nonprivate/others information (example: “Set up meeting with John/Jane on the 4th at 3 PM”).

1 – 7 or Decline to answer

Setting reminder for private information (example: “Remind me to add new credit card number, 3371 2452 0221 6633, to finance paperwork”)

1 – 7 or Decline to answer

Setting reminder for nonprivate/other information (example: “Remind me to buy cereal on Monday”)

1 – 7 or Decline to answer

334 A. EASWARA MOORTHY AND K.-P. L. VU

Demographic Questions

==QUESTION 23. What is your gender? • Male • Female • Decline to answer

==QUESTION 24. What is your age? • 18 to 24 • 25 to 34 • 35 to 44 • 45 to 54 • 55 to 64 • 65 to 74 • 75 or older • Decline to answer • Other (please specify) __________________________

==QUESTION 25. What is your race? Please choose one or more. � White � Black or African American � Asian � Native Hawaiian or other Pacific Islander � American Indian or Alaska Native � Decline to answer � Other (please specify) __________________________

==QUESTION 26. For quality control purposes, please select “laptop” as the answer to this question. Enter the special code “laptop” into the verification textbox in the Amazon Turk HIT description to indicate your completion of this HIT.

• Desktop • Laptop • Mobile phone

==QUESTION 27. What is the highest level of education you have completed? • Less than high school • High school/GED • Some college • 2-year college degree (Associates) • 4-year college degree (B.A., B.S.) • Master’s degree • Doctoral degree • Professional degree (M.D., J.D.) • Decline to answer • Other (please specify) __________________________

==QUESTION 28. What is your approximate average household income? • $0 – $24,999 • $25,000 – $49,999 • $50,000 – $74,999 • $75,000 – $99,999

PRIVACY CONCERNS FOR VOICE ACTIVATED PERSONAL ASSISTANT 335

• $100,000 – $124,999 • $125,000 – $149,999 • $150,000 – $174,999 • $175,000 – $199,999 • $200,000 and up • Decline to answer • Other (please specify) __________________________

End of Survey Thank you for your participation! Correctly enter the special code into the verification text box to receive your payment for this task.

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

ABSTRACT
1. TECHNOLOGY ADOPTION AND USAGE MODELS
TECHNOLOGY ADOPTION AND USAGE MODELS[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0002

1.1. Technology Acceptance Model
Technology Acceptance Model[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2001
1.2. Unified Theory of Acceptance and Use of Technology
Unified Theory of Acceptance and Use of Technology[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2002
1.3. Mobile Phone Technology Acceptance Model
Mobile Phone Technology Acceptance Model[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2003
1.4. Summary
Summary[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2004

2. USING MOBILE VOICE SERVICES AND VOICE APPLICATIONS IN PUBLIC SETTINGS
USING MOBILE VOICE SERVICES AND VOICE APPLICATIONS IN PUBLIC SETTINGS[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0003

2.1. Location and Time
Location and Time[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2005
2.2. Cultural Factors
Cultural Factors[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2006
2.3. Environmental Factors
Environmental Factors[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2007
2.4. User Characteristics
User Characteristics[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2008
2.5. Preference for Voice Applications in Public Settings
Preference for Voice Applications in Public Settings[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2009

3. PRIVACY CONCERNS WITH MOBILE PHONE USE
PRIVACY CONCERNS WITH MOBILE PHONE USE[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0004

3.1. Privacy Concerns for Physical Interactions
Privacy Concerns for Physical Interactions[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2010
3.2. Privacy Concerns for Remote Interactions
Privacy Concerns for Remote Interactions[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2011

4. PRESENT STUDY
PRESENT STUDY[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0005
5. HYPOTHESES
HYPOTHESES[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0006
6. METHOD
METHOD[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0007

6.1. Participants
Participants[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2012
6.2. Design
Design[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2013
6.3. Materials
Materials[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2014
6.4. Quality Control Measures
Quality Control Measures[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2015
6.5. Procedure
Procedure[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2016

7. RESULTS
RESULTS[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0008

7.1. Smartphone Usage
Smartphone Usage[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2017
7.2. Prior Experience With VAPA
Prior Experience With VAPA[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2018
7.3. VAPA and Keyboard Likelihood of Usage Ratings
VAPA and Keyboard Likelihood of Usage Ratings[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2019
7.4. Correlation Between VAPA Usage Likelihood and Social Acceptability
Correlation Between VAPA Usage Likelihood and Social Acceptability[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2020

8. DISCUSSION
DISCUSSION[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0009

8.1. Type of Information Transmitted
Type of Information Transmitted[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2021
8.2. Input Mode for Information Transmission
Input Mode for Information Transmission[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2022
8.3. Location for Information Transmission
Location for Information Transmission[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2023
8.4. Limitations of the Study
Limitations of the Study[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2024
8.5. Summary and Conclusion
Summary and Conclusion[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2025
8.6. Smartphone Design Implications
Smartphone Design Implications[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S2026

ACKNOWLEDGEMENTS
REFERENCES
ABOUT THE AUTHORS
APPENDIX
APPENDIX[]pdfmark=/DEST,linktype=anchor,View=/XYZ H.V,DestAnchor=S0010