The Effects of Robot Shape, Size, and Color on Robot Preferences

(1)

The Effects of Robot Shape, Size, and Color

on Robot Preferences

Differences between communal and exchange service settings

(2)

2

MSc Marketing

Management & Intelligence Thesis

The Effects of Robot Shape, Size, and Color on Robot Preferences

Differences between communal and exchange service settings

University of Groningen Faculty of Economics and Business

Department of Marketing PO Box 800, 9700 AV Groningen (NL) January 2018 by Miranda Leijen s2355205 J.C. Kapteynlaan 53a 9714 CN Groningen +31613016422 j.c.h.leijen@student.rug.nl

(3)

3

ABSTRACT

Service robots are becoming increasingly important in the contemporary service environme nt. Drawing on prior research indicating the relevance of robot appearance in influencing user responses, this paper aims to provide meaningful insights into the influence of robot shape, size, and color on robot preference. To this end, a Choice-based Conjoint Analysis is conducted including these central attributes of robot appearance to measure respondents’ preference. Since previous studies present that preferences of service users differ depending on the type of service situation, respondents were randomly divided into two conditions (representing a communal or exchange service setting). According to the responses, a middle-sized, machine- like robot is most preferred, regardless of the service condition. On the other hand, color has no effect on preference. In addition, findings show that distinct feelings of warmth and competence are experienced for the different alternatives of robot shape and size. Thus, while results indicate that robot preference is not depending on the type of service setting, warmth and competence judgments are identified as important drivers of preference.

Keywords: robot appearance, shape, size, color, service setting, communal, exchange, warmth,

(4)

4

PREFACE

(5)

5

1. INTRODUCTION

Nowadays, the development of Artificial Intelligence (AI), several new technologies and devices such as advanced robotics, Intelligent Agents and the Internet of Things (IoT) are drastically changing the interactions between customers and service personnel and thereby affecting the roles of all the involved actors (Larivière et al., 2017). According to the market research report ‘Top Robotics Market’ by Industrial Robotics, the market for industrial robots is estimated to be valued at 79.58 billion USD, with a growth rate of 11.92% between 2016 and 2022. Simultaneously, it is expected that the market for top service robots reaches a value of 20.7 billion USD by 2022, with a growth rate of 14.71% between 2016 and 2022 (Markets and Markets, 2017). Within the field of robotics, a distinction can be made between different types of robots based on the categories defined by the United Nations in their world robotics survey, namely industrial, professional

service and personal service robots (U.N. and I.F.R.R., 2002). These categories are mainly defined by the application domain of the robot. Also, they represent different technologies and refer to different phases in the history of robot development and commercialization (Thrun, 2004; Vaussard et al., 2014). Industrial robots reflect the earliest commercial success in robotics and are the most widely distributed robot type in today’s world of automation. It has three main characteristics: it manipulates its physical environment, it operates in industrial settings, and it is computer-controlled (Thrun, 2004). Typical applications of industrial robots are assembly, packaging, and material handling.

On the other hand, professional service robots represent a much less practiced field, which is growing at a fast pace. Similar to industrial robots, the professional service robot manipulates and navigates its physical environment. However, the application domain of professional service robots is often outside industrial settings, such as hospitals. Professional service robots typically assist people in achieving their professional goals (Thrun, 2004; Bartneck & Forlizzi, 2004). Lastly, personal service robots are expected to have the highest growth rate and usually assist or entertain people in domestic, institutional, or recreational settings (Thrun, 2004). The focus of this paper is the professional service robot. From the human-robot interaction perspective, service robots can be distinguished from industrial robots through higher levels of autonomy and interaction with the user (Murphy et al., 2017).

(7)

7

design of the service interface is of critical importance and raises a number of challenges within the field of service robotics (Bartneck & Forlizzi, 2004; Thrun, 2004). Moreover, Hegel and colleagues (2009) state that robot appearance plays a critical role within the human-robot interaction, in the sense that visual appearance influences expectations of the user about the specific applications and functionalities of the robot. Furthermore, their results show that visual robot appearance strongly correlates with robot acceptance, likeability, and pleasure. Presented findings confirm traditional literature, arguing that appearance is a powerful design component that assists people in forming expectations about others (Nelson & Bowen, 2000). For instance, attractive people are considered more skilled and sociable than unattractive people (Lennon & Miller, 1984; Lapitsky & Smith, 1981). In a similar manner, preferences for robots are often solely based on their appearance (Hegel et al., 2009). According to Hiroi & Ito (2009), the main characteristics determining robot appearance are shape, color, and size.

A robot can take many shapes, ranging from a more machine- like to a more human- like appearance. With regard to the preferred robot shape, careful consideration of the service situatio n and related needs of the user is essential. In general, it is believed that a robot’s degree of human-likeness should match its intended function (Fong et al., 2003). In an ideal situation, the robot reflects an appropriate balance of illusion and functionality. Specifically, this means that the robot should reflect a certain amount of human-likeness to maintain the illusion of being sophisticated in areas where the user will not encounter robot failings. Simultaneously, a certain amount of machine- likeness is necessary to reflect the functionality of the robot, which demonstrates its capabilities to support the human-robot interaction (DiSalvo et al., 2002; Fong et al., 2003). Despite the existing knowledge, there is no clear set of design guidelines for the optimal robot shape, due to difficulties in finding the perfect balance between machine- and human-likeness.

(8)

8

statement about the optimal size from a human-robot communication point of view (Hiroi et al., 2012).

Lastly, concerning the design feature color, current literature mentions the use of differe nt colors in the design of robots (Hiroi & Ito, 2009). Still, no convincing evidence exists for the psychological effects of robot color. Traditional marketing research reveals that colors generate influential associations which explain human physiological responses (Nelson & Bowen, 2000). According to Priluck Grossman & Wisenblit (1999), consumers form color preferences for products based on their associations created through experience. Specifically, this means that consumers prefer a color resulting from either a favorable experience with the particular color or because consumers feel that certain colors are more appropriate for certain product categories. Consequently, color preference depends on both the particular situation and the underlying associations (Priluck Grossman & Wisenblit, 1999). Considering the field of robotics, these insights from marketing literature suggest the relevance of identifying associations service users form about robots. Knowledge about these associations could be used to find appropriate colors in the design process, which in turn could lead to desirable user responses.

Although there is sufficient support for the relevance of robot appearance within the service context (e.g., Thrun, 2004; Hegel et al. 2009; Hwang et al., 2013; Hiroi et al., 2012), general conclusions about the optimal design of the robot in terms of shape, size, and color are lacking. However, as these particular attributes are demonstrated to be the critical features of robot appearance (Hiroi & Ito, 2009), it seems worthwhile to investigate preferences for these attributes. Robot designers could take obtained insights into account during the design process, which will eventually improve the quality of the human-robot interaction.

(9)

9

is no obligation of returning a comparable benefit for the receiver (Clark & Mills, 1994). Thus, it is evident that social aspects are more important in communal rather than exchange service settings. With regard to robot interactions, Hinds and colleagues (2004) show that human-like robots are treated more politely and more socially interactive than machine-human-like robots. Consequently, it is assumed that human-like robots are preferred to machine- like robots in communal service situations as human-like robots are perceived as more sociable. Drawing on these inferences, investigating the role of service settings seems promising, as it could provide robot designers the opportunity to match design features to the preferences observed within a particular service setting.

This paper attempts to analyze the influence of service robot appearance on robot preference within different types of service conditions. It addresses the following research questions:

 How does appearance of the service robot influence robot preference?

 Specifically, what is the preferred robot design in terms of shape, size, and color?  Lastly, how do communal and exchange service settings influence user preference for

the different robot design features?

This study provides the following main contributions to literature: first, this paper explains how service robot appearance shapes robot preference. Second, insights are provided about the preferred design of robots in terms of shape, size, and color from the service user perspective. Third, this study is the first to investigate the differences in robot preference under two varying service conditions: communal and exchange service settings.

(10)

10

2. THEORETICAL BACKGROUND

2.1 Definitions of the Robot over the Years

In existing literature, various definitions of robots can be found, changing over the years as new types and applications of robots emerge. The paper of Bartneck and Forlizzi (2004) provides an overview of the different definitions, after which a new definition is proposed. According to their framework, the robot made its first appearance in the early 1920s, leading to a negative connotation. Specifically, robots were described as ‘evil machines that would subsume mankind ’. Subsequently, the Robot Institute of America described a robot as ‘a reprogrammab le, multifunctional manipulator designed to move material, parts, tools, or specialized devices through various programmed motions for the performance of a variety of tasks’. In addition, the International Standard Organization defines a robot as follows, ‘an automatically controlled, reprogrammable, multipurpose manipulator programmable in three or more axes, which may be either fixed in place or mobile for use in industrial automation applications’. Clearly, these definitions apply to industrial robots. However, the human-robot interaction (critical for service robots) is not an integral part of these definitions. Consequently, the International Federation of Robotics introduced a preliminary definition of the service robot, namely ‘a robot which operates semi or fully autonomously to perform services useful to the well-being of humans and equipment, excluding manufacturing operations.’ Still, the human-robot interaction is not made fully explic it. This was first done by Engelhardt & Edwards (1992), defining service robots as ‘systems that function as smart, programmable tools, that can sense, think and act to benefit or enable humans or extend/enhance human productivity’. Bartneck and Forlizzi (2004) state that productivity is not always the ultimate goal, for instance in case of entertainment. Therefore, a new definition was proposed for the service robot: ‘An autonomous or semi-autonomous robot that interacts and communicates with humans by following the behavioral norms expected by the people with whom the robot is intended to interact’. According to this definition, a service robot has a physical embodiment, and communication and interaction with humans play a critical role (Bartneck & Forlizzi, 2004).

2.2 Robot Appearance

(11)

11

and the user. According to Bergmann et al. (2012), users’ first impressions of the robot are essential in accomplishing these goals. Along the same line, Powers & Kiesler (2006) propose that it is important to create an unambiguous first impression of the robot’s expertise and personality, especially for robots operating in public settings and interacting with strangers. Bergmann and colleagues (2012) argue that visual appearance is one of the major cues in forming first impressions. In addition, Hegel and colleagues (2009) suggest that people have preferences for robots due to their visual appearance. A proposed explanation is provided by the attractive ness bias (Dion et al., 1972): whereas unattractive robots arouse negative emotions, attractive robots are perceived as more likable, enjoyable, and usable. Hegel and colleagues (2009) were not the first to investigate the effect of appearance. In fact, a growing body of literature provides evidence for the influence of appearance on how people form perceptions, expectations, and preferences for robots (Bergmann et al., 2012).

Extending on these insights, Goetz et al. (2003) argue that the appearance of the robot provides cues and influences people’s perceptions about the capabilities of the robot. This effect is explained by means of two hypotheses. First, the positivity hypothesis proposes that the higher the perceived attractiveness of the robot, the more willing people are to accept and comply with the robot. Second, the matching hypothesis, suggests that the appearance of the robot should match the seriousness of the situation and task. Thus, in addition to confirming the attractiveness bias, Goetz and colleagues (2003) provide evidence for the task-dependent relationship between robot appearance and user acceptance and cooperation.

2.3 Robot Design Features

While the crucial role of robot appearance is widely discussed, much less has been reported on how to design a robot to ensure desirable user responses. According to DiSalvo et al. (2002), most work on the design of robots focuses on improving the current state of technology. However, since service robots are expected to grow in the foreseeable future, it is of even greater importance to investigate the social implications of robot appearance.

(12)

12

the (un)familiarity with the robot’s morphology has profound effects on its accessibilit y, desirability, and expressiveness. Other design considerations are artificial emotions, speech and facial expressions of the robot. The primary goal of using artificial emotions is to facilitate a believable human-robot interaction. Moreover, robot voice can be used to support emotiona l communication, whereas facial expressions communicate information and make the behavior of the robot more predictable and understandable (Fong et al., 2003).

Since it is argued that first impressions are crucial in forming positive user’s expectations, perceptions, and preferences, the structure and form of a robot are considered to be the most important design elements forming robot appearance. That is, first impressions are created within a few seconds based on visual appearance (Bergmann et al., 2012), and unlike most of the other design elements, a robot’s structure and form stand out immediately.

Within the field of robotics, the form or shape of the robot is often associated with the degree of human- likeness. According to Goetz and colleagues (2003), the human-likeness of a robot contains cues that elicit automatic perceptions of lifelikeness in a robot. Subsequently, people will make attributions of the ability and personality of the robot based on these perceptions. In turn, these attributions will elicit social responses and expectations. Among researchers, finding the appropriate robot shape remains a controversial issue. Some argue that a human- like interface is preferred, since rules of human-interaction can be easily transferred, which makes the conversation more intuitive. However, opponents argue that human-like shapes can result in unrealistic high expectations or even cause fear (Bergmann et al., 2012). Hence, further research is essential to resolve the controversy surrounding robot shape.

(13)

13

According to a large body of psychological and marketing literature, the appearance of people, products and, service settings can influence consumer feelings and behaviors, either consciously or subconsciously through colors. Colors are found to carry meaning and attract customers (e.g., Elliot & Maier, 2014; Labrecque & Milne, 2012; Labrecque et al., 2013). In addition, careful consideration of the use of colors can lead to beneficial outcomes, such as positive associations (Priluck Grossman & Wisenblit, 1999). Moreover, knowledge about the physiologica l effects of colors is used nowadays by institutions to relax people without the need to insert medication. A study of Costigan (1984) shows that dentists occasionally paint their walls blue to relieve their patients from fears, as blue is considered a calming color. Despite the potential and influential role of robot color, research on the effects of color within the field of robotics is lacking. Nevertheless, Hiroi and Ito (2009) confirm that shape, size, and color are the major features forming robot appearance.

2.4 Conceptual Model

According to the presented literature, the appearance of a robot influences attitudes and preferences towards the robot, most likely through the psychological effects of its shape, size, and color (as is also suggested by the work of Hiroi & Ito (2009)).

It seems worthwhile to investigate the effects of robot shape, as a better understanding of the appropriate degree of human-likeness is expected to benefit the human-robot interaction. Since literature suggests that the appropriate degree depends on the attributed attractiveness and assigned tasks of the robot (Goetz et al., 2003), insights into user- and context-specific factors (influenc ing preference) are of particular relevance. As a consequence, the shape of the robot can be aligned with the wishes of the target user within a specific situation.

(14)

14

Lastly, color literature reveals that colors can elicit calming and fear-releasing effects. Since robots are expected to cause fear and anxiety under certain circumstances (e.g., Bergmann et al., 2012; Hiroi & Ito, 2009), it seems promising to investigate the effects of robot colors on user perceptions, as the use of calming colors might mitigate the threatening effect of a robot.

In sum, shape, size, and color are identified as the central elements of robot appearance and the relevance of investigating their main effects is demonstrated. Although literature suggests that there is a task-dependent relationship between appearance and preference (Goetz et al., 2003), knowledge is lacking about possible changing effects on preference due to different tasks of the robot. In general, the assigned tasks differ depending on the type of service situation. Existing literature reveals that consumers’ preferences within their relationship with service employees vary between communal and exchange service situations (Mende & Bolton, 2011). Since robots are likely to fulfill the role of service assistants in the future, consumers are also expected to have different preferences for service robots, depending on the type of service situation. Hence, investigating the role of service settings within the field of robotics is expected to provide valuable insights into preferences. To visualize the expected relationship between the described concepts , a conceptual framework is derived in Figure 1.

Figure 1: Conceptual Framework

(15)

15

2.5 Robot Shape

2.5.1 The Influence of Robot Shape on Preference

Regarding existing work on the appearance of service robots, most research has been devoted to the design feature shape. Current literature uses different classifications of robot shape. Fong et al. (2003) classify robots into four categories: anthropomorphic, zoomorphic, caricatured and

functional. Anthropomorphism refers to the tendency to attribute human features to objects to

rationalize actions (Duffy, 2003). A zoomorphic robot refers to animal morphology and actual designs are frequently inspired by household animals. Next, caricatured designs commonly use techniques from traditional animation and cartooning, where remarkable features are often exaggerated to create comic effects. Lastly, functional robots are designed based on operational objectives, and therefore reflect the assigned task (Fong et al. 2003). Another popular categorization is based on the work of Gong & Nass (2007) and MacDorman & Ishiguro (2006). Here, the distinction is made between mechanoid, humanoid, and android robots (Walters, 2008). A mechanoid robot has a machine- like shape, with no apparent human-like features. Next, the humanoid robot does not have a realistic human- like appearance. However, it typically possesses some human- like attributes which are often stylised, cartoon-like, or simplified versions of the human equivalent. Lastly, the appearance of the android robot is as close to a real human as technically possible. The ultimate purpose is to be perceived as a human by real human beings. Hitherto, this goal is only achieved for a few seconds. As stated earlier, creating unambiguous first impressions is important, especially when robots are working in public settings and communicate with strangers (Bergmann et al., 2012). The experimental results of Lohse and colleagues (2007) reveal that participants generally imagine zoomorphic robots as being used as a toy or pet for entertainment purposes. Since this paper focuses on professional service robots, which typically operate in public settings (such as banks and hospitals), the use of zoomorphic robots seems inappropriate and not in line with the intended service application. Hence, the categorization of

mechanoid, humanoid, and android robots appears to be more suitable for professional service

robots and is therefore used for further purposes.

(16)

16

robot. On the other hand, if peer interactions play a critical role, a certain amount of ‘human-likeness’ is necessary to make the user feel comfortable during the social engagement with the robot (Fong et al., 2003). However, as people attribute more human qualities to robots when they look more human- like (Hwang et al., 2013), a certain amount of ‘robot-likeness’ is desirable sometimes to avoid false expectations of the robot’s capabilities (Fong et al., 2003).

In practice, finding the perfect degree of human-likeness is challenging. An explanatio n for this difficulty is provided by the Uncanny Valley theory of Mori. According to Mori et al. (2012), the more human- like robots are in motion and appearance, the more positive the users’ emotional reactions towards the robots become, such that familiarity and social acceptance increase. However, the positive relationship between the degree of human-likeness and emotiona l responses was found to be non-linear. The trend suddenly declines at the point where the robot almost resembles a human, as the subtle imperfections of the robot are perceived as disturbing and repulsive (Bartneck et al., 2009). On the other hand, Mori argues that if the robot’s human-like ness can be further increased to almost perfectly match the appearance of a human, familiarity will rise again (Beer et al., 2011). According to another line of research, human- like robots are expected to be the preferred alternative in social settings. This is mainly due to the presence of human- like attributes, which provide advantages when close interactions with humans are necessary. Another reason is that people assume robots to be better at performing human- like skills, such as verbal communication and intelligence, when they look more like real human-beings (Hwang et al., 2013). However, contradicting results are presented by Woods and colleagues (2004), since the children in their study assess human- like robots as aggressive and machine- like robots as friendly.

2.5.2 The Influence of Robot Shape within Different Service Settings

In communal and exchange service situations, the norms associated with the particular service orientation are used as standards to make social perceptions, evaluate others, and decide on what is appropriate in certain situations (Clark & Mills, 1993). As social perceptions differ across contexts, Scott and colleagues (2013) propose that people tend to put more emphasis on warmth than competence in case of communal relationships, whereas more emphasis is placed on

competence than warmth in case of exchange relationships. According to a number of other

(17)

17

whereas competence cues indicate whether people are skillful and efficacious (Fiske et al., 2007; van Doorn et al., 2017). Moreover, Bergmann and colleagues (2012) suggest that warmth and competence judgments are made during the first moments of contact. In turn, these first impressions are critical in forming preferences and creating a basis for rich relationships.

Since presented literature reveals that the type of service setting influences feelings of warmth and competence and associated first impressions, it is expected that preferences for robot shape depend on the particular service situation. As it is argued that warmth cues are more important in communal service settings, the social aspects of the conversation are assumed to be of substantial influence in forming preferences. In addition, it is known that human- like robots are perceived as more friendly, trustworthy, and sociable (e.g., Li et al., 2010). Along the same line of reasoning, Van Doorn et al. (2017) introduce the concept of automated social presence (ASP), which is defined as the ‘degree to which technology makes customers feel the presence of another social entity’. According to van Doorn and colleagues (2017), high ASP increases perceptions of sociability, which enhances feelings of warmth, especially when customers adopt a communa l service orientation. High ASP refers to robots which are associated with a high degree of human-likeness. In line with the proposition of van Doorn et al. (2017), the social appearance of a humanoid robot is assumed to serve as a warmth cue. Since users in communal service settings tend to focus on warmth cues, it is expected that the humanoid robot activates feelings of warmth, whereby positive traits are attributed to the robot. In turn, this is expected to positively influe nce preference.

(18)

18

Regarding the android alternative, Bartneck et al. (2009) suggest that robots that are almost similar to humans are perceived as creepy and disturbing, based on the Uncanny Valley theory of Mori. However, robot familiarity and social acceptance can rise again if the appearance of the robot almost perfectly resembles a real human (Mori et al., 2012). Within the current state of technology, it is not possible to design a perfect human-looking robot. Therefore, the android robot is assumed to evoke negative emotions, such as anxiety and fear instead of social acceptance and familiarity. Since fear is a powerful emotion, capable of influencing attitudes and decision-mak ing processes (e.g., Coget et al., 2011; Nabi, 2002), it is expected that feelings of fear aroused by the android alternative are more influential on robot preference than warmth and competence judgments. Thus, the android robot is expected to be the least preferred alternative, regardless of the service situation. In sum:

H1: An android robot is least preferred compared to a mechanoid and humanoid robot in both a

communal and exchange service setting.

H2.a: A humanoid robot is most preferred compared to a mechanoid and android robot in a

communal service setting.

(19)

19

mechanoid robot is assumed to serve as a competence cue in exchange service settings, which makes users believe that the robot possesses the right skillset to perform the assigned tasks. As a result, it is expected that user preference for the humanoid robot in communal settings shifts towards the mechanoid robot in exchange service settings, as competence cues are now more important. To conclude:

H2.b: A mechanoid robot is most preferred compared to a humanoid and android robot in an

exchange service setting.

2.6 Robot Size

2.6.1 The Influence of Robot Size on Preference

(20)

20

Within the field of robotics, many actual service robots measure around 1.2 m. Robot designers often base their choice on technical reasons, as the size of the robot should be suitable for performing its tasks. Other examples of robots are known as well, covering sizes both smaller and larger than 1.2 m. Designers always take into account the functional purpose of the robot when making decisions on size. For instance, human-sized robots working in human environments have to be smaller than 2.0 m to be able to pass through the doorway. Next to that, the body-to-limb proportion of robots cannot differ greatly from humans, as the robot should be able to keep balance while walking. On the other hand, small robots in care support with a size lower than 0.6 m are likely to experience difficulties in delivering food or documents (Hiroi & Ito, 2009). In conclusio n, both psychological effects and technical reasons are essential in the design process of the robot. However, the current focus is mainly on the technical aspects of size.

2.6.2 The Influence of Robot Size within Different Service Settings

Preference for robot size is again likely to depend on the type of service setting. In communa l service situations, the social aspects of the robot are assumed to be more important in influenc ing preference, as people are more alert to warmth cues indicating personality traits such as being caring, helpful, and friendly (Scott et al., 2013). According to Hiroi & Ito (2012), people feel most comfortable during the human-robot interaction, when the size of the robot is lower than the user’s eye-height. On the other hand, when the robot is a lot smaller or higher than the user’s eye-height, feelings of discomfort are experienced. Since Williams & Bargh (2008) suggest that feelings of comfort and trust are associated with warmth, the size of the robot is expected to serve as a warmth cue when the robot is (slightly) lower than the user’s eye-height. Therefore, the middle-sized robot (which is typically lower than the user’s eye-height) is expected to be the preferred alternative in the communal setting, as it makes users feel comfortable, whereby feelings of warmth are activated. Consequently, users are more likely to perceive the robot as friendly and caring, which positively influences preference.

(21)

21

2009), a similar negative effect on warmth judgments might be expected for the small robot alternative. On the other hand, Bergmann and colleagues (2012) suggest that a child-like look is typically associated with many positive attributes. In the same vein, Zebrowitz & Motepare (2008) argue that a child-like appearance creates impressions of people’s traits. In particular, adults with baby-faces were found to be perceived as warmer. Since small robots are often as tall as a baby, this might activate similar feelings of warmth, as people in communal settings tend to focus on warmth cues. Therefore, it is expected that the large robot is least preferred compared to the small robot. In sum:

H3.a: A middle-sized robot is most preferred compared to a small-sized and large-sized robot in

a communal service setting.

H3.b: A large-sized robot is least preferred compared to a small-sized and middle-sized robot in

a communal service setting.

Regarding the exchange service setting, competence cues are expected to cause a change in size preference. Research in communication reveals that people who are taller relative to the perceiver are associated with various positive characteristics (Rae et al., 2013). Specifically, taller people are often perceived as more competent, authoritative, or dominant (Judge & Cable, 2004; Stulp et al., 2012). Since similar psychological effects of height on people’s perceptions are demonstrated to extend into robot-mediated communications (Rae et al., 2013), it is expected that large-sized robots are also perceived as more competent. Thus, the size of a large robot is assumed to serve as a competence cue in exchange service settings. Consequently, a large robot is more likely to be perceived as skillful and efficacious, since competence cues activate competence perceptions. Therefore, the large robot is expected to be the most preferred alternative in exchange service settings, as people tend to focus on competence cues.

(22)

22

service settings, a small robot is expected to be perceived as least competent and thereby least preferred. In summary:

H4.a: A large-sized robot is most preferred compared to a small-sized and middle-sized robot in

an exchange service setting.

H4.b: A small-sized robot is least preferred compared to a middle-sized and large-sized robot in

an exchange service setting.

2.7 Robot Color

2.7.1 The Influence of Robot Color on Preference

(23)

23

Shimp (1991) states that is easier to develop associations for stimuli that are unfamiliar to people than for stimuli that are familiar. Regarding service robots, it seems unlikely that service users have already formed color preferences, as robots are relatively new in the service environme nt. Instead, it is assumed that color preference depends on more general color associations.

Considering existing color associations, traditional color literature reveals that in general cool colors (such as blue) are considered calming, whereas warm colors (such as red) are considered arousing. Furthermore, the color red is associated with feelings of excitement and stimulation, again implying high arousal and pleasure (Priluck Grossman & Wisenblit, 1999). On the other hand, the color blue is associated with secureness, comfortableness, and tenderness, referring to pleasure and low arousal. In addition, Labrecque & Milne (2012) suggest that blue colors are linked to competence and associated with intelligence, communication, trust, efficie nc y, duty, and logic. This link between the cool color blue and perceptions of competence can be explained by the theory of embodied cognition. Embodied cognition proposes that sensations that share similar affect states are typically stored together in memory (Connell, 2007). Drawing on this theory, Mehta and colleagues (2011) reveal that warm colors make people feel warmer, which activates warmth perceptions, whereas cool colors make people feel cooler, which activates competence perceptions.

2.7.2 The Influence of Robot Color within Different Service Settings

Extending on these insights, cool and warm color associations can be related to communal and exchange service settings. Since people tend to focus more on warmth (than competence) cues in communal settings (Scott et al., 2013), it is expected that the color red activates warmth perceptions, as red is a warm color and thus a warmth cue (Priluck Grossman & Wisenblit, 1999; Mehta et al., 2011). However, in exchange service settings, a red color is less likely to activate warmth perceptions, as people tend to focus more on competence (rather than warmth) cues (Scott et al., 2013). In this situation, a blue (cool) color is expected to activate competence perceptions, as the color blue serves as a cue for competence (Labrecque & Milne, 2012; Mehta et al., 2012).

(24)

24

since a red color serves as a warmth cue, the presence of a red robot in a communal service setting activates warmth perceptions, presumably making users perceive the robot as friendly and helpful. On the other hand, being skillful and efficacious is assumed to be more important for service users in exchange service settings (e.g., van Doorn et al., 2017). Therefore, it is expected that a blue robot is preferred to a red robot in these settings, since blue serves as a competence cue, activating competence perceptions of the robot. Summarizing the above:

(25)

25

3. METHODOLOGY

First, this section discusses the applied research method. Subsequently, the data collectio n procedure, the sample characteristics, and the scales used in the survey are described. Lastly, the formula for measuring preferences is presented, followed by a description of the measureme nts applied to assess the performance of the estimated models.

3.1 Research Methods

The applied research method in this study is Conjoint Analysis. A Conjoint Analysis is an experimental method for studying consumer preferences for products and services. There are different ways in which study participants can evaluate the examined products. For this experiment, a Choice-based Conjoint (CBC) is conducted. In a Choice-based Conjoint, respondents repeatedly choose their preferred product among different alternatives within a choice set. Since choices are an integral part of people’s daily life, this approach is very effective. Subsequently, preference measurement examines how respondents evaluate each attribute of the product and produces a quantifiable result. Ultimately, this valuation process results in a utilit y function, that translates the specific features of a product or service into perceived preferences of the respondent (Eggers & Sattler, 2011). Within this study experiment, respondents were asked to indicate their preferred choice among three robot alternatives per choice set. Each robot represents a combination of robot shape, size, and color. According to Eggers & Eggers (2011) consume rs are expected to choose the alternative that maximizes their total utility.

3.2 Data Collection

For measuring robot preferences, an online survey was created using the software Preference Lab. In order to collect the data, the survey link was spread via email, social media, WhatsApp, and personal contact. The first section of the survey consists of questions gathering information about the control variables age, gender, and technology readiness. The next section is devoted to the type of service setting.

(26)

26

manipulated by means of a brief description of a service situation. According to Scott and colleagues (2013), communal relationship norms are typically associated with health-care services, whereas exchange relationships are often associated with financial services, such as banks. Accordingly, a description of a doctor’s visit was used for the communal condition, while a description of a bank visit was used for the exchange condition.

Specifically, the following description was used for the communal service condition: ‘Image you are visiting a doctor for a regular health check. After entering the doctor’s office, you notice the presence of a robot assistant. The health assistant robot is able to automatically record your health data (such as heartbeat, blood pressure, blood glucose, and body temperature) and to perform comprehensive analysis and generate reports. If you were in this situation, which of the following robots would you prefer to perform the health check? Note: the bookcase (0.5 m) at the left of the picture indicates the relative size of the robot.’

For the exchange service condition, the description states:

‘Imagine you are entering the business lobby of a bank. While approaching the reception desk, you notice the presence of a robot to welcome and assist you. The bank service robot is able to answer your questions about banking issues, to guide you to designated service areas of the bank, to fulfill various personal banking service requests, and to promote financial products. If you were in this situation, which of the following service robots would you prefer to assist you? Note: the bookcase (0.5 m) at the left of the picture indicates the relative size of the robot.’

Within the next section of the survey (the conjoint section), participants were asked to indicate their preferred choice among the different robot alternatives. After each choice set, a multi- item measurement scale for warmth and competence was included to test whether respondents actually experienced feelings of warmth and competence. Subsequently, respondents were asked to complete the scale for two of the three indicated robot alternatives (per choice set).

3.3 Sample Characteristics

(27)

27

(Condition 2). Within Condition 1, 72.55% of the respondents are female compared to 58.23% in Condition 2. Furthermore, the average age of respondents in Condition 1 is 27.35 years, whereas respondents in Condition 2 are typically younger, with an average age of 24.39.

3.4 Study Design and Conjoint Attributes

According to Eggers & Sattler (2011), there are several design efficiency criteria that the study design should meet to ensure research validity. First, the choice design should be balanced, which means that all attribute levels need to be displayed an equal number of times. Next, the design needs to be orthogonal; each combination of attribute levels should appear an equal number of times to prevent attribute correlations. Moreover, there should be minimal overlap between the choice sets, meaning that the alternatives within a choice set are maximally different from one another. Lastly, utilities ought to be balanced, which means that alternatives within a choice set have to be equally attractive to the respondent to avoid dominating choice sets or alternatives. For this study, a full factorial design was used, which means that all the possible attribute level combinations are displayed. As such, a full factorial design is always efficient (Eggers & Sattler, 2011). Since Preference Lab automatically controls for these criteria, a balanced and orthogonal design was created for this study with minimal overlap between the choice sets and alternatives.

For the conjoint part of the survey, the attributes of the conceptual model need to be decomposed into different levels. According to Eggers & Sattler (2011), levels should be realistic and reasonable in number. Typically, a number between two and five is considered acceptable. Table 1 provides an overview of the attributes and levels included in this study.

ATTRIBUTE LEVELS

Shape Mechanoid, humanoid, android

Size Small (0.6 m), middle (1.2 m), large (1.8 m)

Color Red, blue

Table 1: Conjoint Attributes and Levels

(28)

28

3.2.3 Warmth and Competence

3.4 Utility Function

3.5 Model Fit

3.6 Moderating Effect

Figure 2: Survey Example Choice Set

3.5 Measures and Descriptives

For measuring technology readiness of the respondents, the Technology Readiness Index (TRI) by Parasuraman and Colby (2001) was used in the survey. In addition, warmth and competence were measured using a multi- item scale developed by Fiske and colleagues (2002). Appendix A provides an overview of the two scales. Warmth and competence and technology readiness were both measured on a 5-point scale.

3.5.1 Technology Readiness

According to Fiske et al. (2002), four underlying dimensions can be identified for the Technology Readiness Index: innovativeness, discomfort, optimism, and insecurity. In order to test whether the dimensions also exist for the current dataset, a factor analysis was conducted. First, the KMO

Measure of Sampling Adequacy and the Barlett’s Test of Sphericity were performed to check

whether it is appropriate to combine the individual items into factors. As Table 2 shows that both the KMO test (> 0.5) and the Barlett’s test (Sig. = 0.00) indicate that a factor analysis is suitable, three factors were created (after inspection of the scree plot and rotated component matrix).

KMO Measure of Sampling Adequacy 0.79 Barlett’s Test of Sphericity

Chi-square df

(29)

29

Sig. 0.00 Table 2: KMO and Barlett’s Test for the TRI Scale

Table 3 shows the three factors and the corresponding combination of items. Subsequently, a reliability analysis was conducted to test the reliability of each of the factors. As can be seen in Table 3, the Cronbach’s alpha is > 0.6 for each of the factors, which indicates strong interna l consistency. Thus, it is appropriate to continue with the factors instead of the original items.

FACTOR ITEMS CRONBACH’S ALPHA

1: Innovativeness 1, 7 0.65

2: Optimism 3, 9 0.67

3: Insecurity & discomfort 2, 4, 6, 8, 10 0.63 (0.66 if item 6 deleted)

Table 3: Factors TRI Scale

As can be inferred from Table 3, the factors contain the same items as the dimensions identified by Parasuraman and Colby (2001). Note: the original dimensions insecurity and discomfort are taken together in Factor 3. Since the items measuring innovativeness are most relevant for this research, the decision was made to only include the first factor as a covariate for further analysis. With regard to the respondents’ level of innovativeness, a positive value was measured for 50.98% of the sample, whereas a negative value was measured for 49.02% of the respondents. In conclusion, approximately half of the respondents are considered to have a relatively high level of innovativeness, whereas the other half has a relatively low level of innovativeness. Additiona l ly, the average innovativeness of respondents assigned to Condition 1 is -0.04, whereas the average level of innovativeness in Condition 2 is 0.04. Therefore, respondents in Condition 1 are generally more innovative than respondents in Condition 2.

3.5.2 Warmth and Competence

(30)

30

Cronbach’s alpha > 0.6 for each of the factors. Therefore, the factors warmth (including the items: tolerant, warm, good-natured, and sincere) and competence (including the items: competent, confident, independent, competitive, and intelligent) were used for further analysis instead of the original items.

3.6 Utility Function

According to Eggers & Sattler (2011), the assumption can be made that products and services are combinations of attributes and that consumers attach part-worth utilities to each of these attributes. Subsequently, the attribute valuation process of respondents produces a utility function which translates the products attributes into perceived preferences. Accordingly, the following utilit y function can be formulated:

𝑈

𝑗

= ∑ 𝛽

𝑘

𝐾

𝑘=1

𝑥

𝑘𝑗

Herein, U is the systematic utility component representing the total utility among all respondents for the preferred robot alternative j, resulting from the sum of part-worth utilities β of attribute k (=1, 2…K), where

ⅹ

indicates the specific attribute level.

3.7 Model Performance

(31)

31

4. RESULTS

In this section, the statistical outcomes of the study are presented. First, the overall model performance is assessed. Second, the results of the Choice-based Conjoint are presented, includ ing the utility estimates per attribute level. Next, the formulated hypotheses are tested and the relative attribute importance is demonstrated. Lastly, warmth and competence judgments are analyzed, followed by a discussion of different segmentation strategies.

4.1 Model Results

In the following section, the estimation outcomes of the Choice-based Conjoint Analysis are presented and discussed. For testing whether attribute preferences depend on the type of service setting, a model was estimated with two classes. The first class includes all the respondents that were assigned to the communal service setting (Condition 1), whereas the second class contains the respondents within the exchange service setting (Condition 2). For estimation of the full model, the attributes shape, size, and color are included, together with the covariates gender, age and innovativeness.

4.2 Model Comparison

First, the full model was compared to a model without any variables (the null model). In order to test whether the inclusion of variables significantly improves the quality of the model, a loglikelihood-ratio (LL) test was conducted. According to the Chi-square statistic, the full model is significantly better than the null model (X2_{= 0.00). In addition, the full model produces a higher} Hit rate (48.86%) compared to the null model (33.66%), indicating better model performance.

4.3 CBC Results and Utility Levels

In order to test whether the two conditions significantly differ, the utility estimates of the two classes are compared. Table 4 provides an overview of the utility estimates, the Wald-values, and

p-values. According to the formal test of equality, given by the Wald(=) statistic, shape (p = 0.53),

(32)

32

color (p = 0.46) does not have a significant effect on preference. With regard to the covariates, age (p = 0.13) and innovativeness (p = 0.54) are not significant, whereas gender (p = 0.09) has a significant effect at a 10% significance level.

ATTRIBUTES Utilities Class 1

Utilities Class 2

Wald p-value Wald(=) p-value

Shape 56.18 0.00*** 1.28 0.53 android -0.40 -0.26 humanoid -0.01 -0.14 mechanoid 0.42 0.40 Size 56.75 0.00*** 3.64 0.16 large -0.58 -0.36 middle 0.36 0.35 small 0.22 0.01 Color 1.55 0.46 1.35 0.25 blue -0.03 0.07 red 0.03 -0.07 COVARIATES Age 0.02 -0.02 2.33 0.13 Gender 2.93 0.09* female 0.19 -0.19 male -0.19 0.19 Innovativeness 0.07 -0.07 0.38 0.54

Table 4: Utility Estimates *** p < 0.01

** p < 0.05 * p < 0.10 4.3.1 Hypothesis Testing

(33)

33

Communal Condition

First, the utility estimates are investigated for the communal condition (class 1) to test hypotheses 2.a, 3.a, 3.b, and 5.a (see Table 5). The most preferred robot shape is the mechanoid robot, thus H2.a cannot be confirmed. In addition, the most preferred robot size is middle-sized, thereby H3.a is supported. On the other hand, the least preferred size alternative is the large-sized robot, which also confirms H3.b. Since the effect of robot color is not significant, H5.a is not supported.

Exchange Condition

In order to test hypotheses 2.b, 4.a, 4.b, and 5.b (see Table 5) the estimation outcomes of the exchange condition (class 2) are analyzed. Since Table 4 reveals that preferences do not differ between the two classes, the mechanoid robot shape and the middle-sized robot are also the most preferred alternatives within the exchange condition. Consequently, H2.a is supported, whereas H4.a cannot be confirmed. In addition, the large-sized robot is again the least preferred alternative, whereby H4.b is not supported. Finally, H5.b cannot be confirmed, since color does not have a significant effect on preference.

HYPOTHESIS

RESULT

H1: an android robot is least preferred compared to a mechanoid and

humanoid robot in both a communal and exchange service setting.

Supported

H2.a: a humanoid robot is most preferred compared to a mechanoid

and android robot in a communal service setting.

Not supported

H2.b: a mechanoid robot is most preferred compared to a humano id

and android robot in an exchange service setting.

Supported

H3.a: a middle-sized robot is most preferred compared to a

small-sized and large-small-sized robot in a communal service setting.

Supported

H3.b: a large-sized robot is least preferred compared to a small-s ized

and middle-sized robot in a communal service setting.

Supported

H4.a: a large-sized robot is most preferred compared to a small-s ized

and middle-sized robot in an exchange service setting.

Not supported

H4.b: a smallsized robot is least preferred compared to a middle

-sized and large--sized robot in an exchange service setting.

Not supported

H5.a: a red robot is more preferred than a blue robot in a communa l

service setting.

Not supported

H5.b: a blue robot is more preferred than a red robot in an exchange

service setting.

(34)

34

4.3.2 Attribute Importance

Figure 3 shows the relative attribute importance for both the communal and exchange condition. It becomes evident that respondents in both conditions consider size to be the most important attribute, closely followed by shape. On the other hand, color is by far perceived as the least important attribute compared to shape and size.

Figure 3: Relative Attribute Importance

4.4 Warmth and Competence Measurements

In order to test whether the effect of the robot appearance on preference can be explained by judgments of warmth and competence, the average warmth and competence measurements per attribute level are analyzed by means of a One-way ANOVA. According to the test results, shape and size have a significant effect on warmth, whereas the effect of color is not significant (shape:

F(2, 1221) = 39.07, p = 0.00; size: F(2, 1221) = 18.32; color: F(2,1221) = 0.38, p = 0.54). Likewise,

shape and size have a significant effect on competence judgments, whereas color does not significantly influence feelings of competence (shape: F(2, 1221) = 23.05, p = 0.00; size:

F(2,1221) = 18.46, p = 0.00; color: F(2,1221) = 0.01, p = 0.92).

Figure 4 provides an overview of the average warmth and competence measurements per attribute level. In line with the expectations, competence measurements are highest for the mechanoid robot (M = 0.21, SD = 1.03) compared to the other shape alternatives, whereas warmth measurements are highest for the humanoid robot (M = 0.30, SD = 1.01). For the android alternative, both warmth (M = 0.02, SD = 0.90) and competence (M = 0.03, SD = 0.96) measurements are relatively low. In contrasts to the expectation that feelings of warmth are most pronounced for the middle-sized alternative, warmth measurements are highest for the small robot

45% 52% 4% 44% 47% 9% 0% 10% 20% 30% 40% 50% 60%

Shape Size Color

(35)

35

(M = 0.17, SD = 0.99). Moreover, for a middle-sized robot both warmth (M = 0.06, SD = 0.97) and competence (M = 0.11, SD = 0.97) are positive, with competence measurements being slight ly higher. In addition, competence measurements are highest for the large robot (M = 0.14, SD = 0.98) compared to the other size alternatives, whereas warmth measurements are the lowest (M = -0.24, SD = 0.98).

Figure 4: Average Warmth and Competence Measurements

Subsequently, a logistic regression was performed to analyze the effect of warmth and competence on preference choice. Results reveal that both warmth (p = 0.00) and competence (p = 0.00) have a positive significant effect on choice at a 1% significance level. Next, the marginal effects were calculated. According to the estimation outcomes, the probability of selecting a robot increases with 0.13 for the average respondent, if warmth increases by 1, whereas this probability increases with 0.11 if competence goes up by 1. Thus, warmth judgments are slightly more important than competence judgments in selecting a robot. In conclusion, both shape and size significa nt ly influence warmth and competence judgments, which in turn have a significant positive effect on preference.

4.5 Segmentation

Since dividing the sample according to the assigned service condition was not significant, other segmentation methods were applied based on the control variables gender and innovativeness. As

-0,30 -0,25 -0,20 -0,15 -0,10 -0,05 0,00 0,05 0,10 0,15 0,20 0,25 0,30

(36)

36

most of the respondents (75%) are between 17 and 25 years, the variable age was intentiona l ly excluded as a segmentation strategy.

Gender

Gender was selected as a priori segmentation strategy to investigate preference differe nces between males and females. First, this new model (model 2) was compared to the null model and the condition-based model (model 1). Appendix C provides an overview of the performance measures. The AIC(LL) of model 2 is 1345.48 and the BIC(LL) is 1379.60, which are lower than the information criteria of model 1. Furthermore, the Hit rate of model 2 is 50.49%, which is higher than the Hit rate of model 1 (48.86%). In conclusion, a segmentation based on gender results in better model performance than a segmentation based on service condition.

The CBC results (Appendix D) reveal that both shape (p = 0.00) and size (p = 0.00) have a significant influence on preference. However, only size preferences significantly differ between males and females (p = 0.00) at a 1% significance level. With regard to the covariates age and innovativeness, only the latter is found to be significant (p = 0.02) at a 5% significance level. For the female segment (class 1), the preferred size is the middle-sized robot, whereas the least preferred size is the large-sized robot. Moreover, size is the most important attribute (63.82%), followed by shape (34.77%) and color (1.41%). The average innovativeness is -0.18, which indicates that in general female respondents are not very innovative. In addition, the male segment (class 2) also prefers a middle-sized robot. However, in contrast to females, the small robot is least preferred. Furthermore, shape is considered the most important attribute (54.18%), followed by size (38.90%) and color (6.92%). The average innovativeness of males is 0.33, which indicate that males are in general more innovative than females.

Innovativeness

(37)

37

However, these performance measures are slightly worse than those of model 2. To conclude, a segmentation strategy based on gender results in the best performing model.

(38)

38

5. DISCUSSION

In this final chapter, the research findings are discussed. First, the theoretical implications are presented for each attribute, followed by the managerial implications. Lastly, the limitations of this research and future research directions are presented.

5.1 Theoretical Implications 5.1.1 Shape

In line with previous work, current research findings reveal that shape is an important driver of robot preference. Its influence on preference is highly significant and respondents consider it to be one of the most important attributes when forming preferences (only size is slightly more important). According to existing research, the degree of human- likeness of the robot is an essential part of the design process and has to match with the intended tasks of the robot (e.g., Goetz et al., 2003). However, findings demonstrate that a mechanoid robot shape is most preferred, regardless of the service situation. Since the described tasks of the robot differ between the two service conditions, the task-dependent relationship between robot appearance and preference is not supported with regard to shape.

(39)

39

5.1.2 Size

Study results confirm existing work demonstrating the psychological effects of size (e.g., Hiroi & Ito, 2009; Hiroi & Ito, 2012). According to the current findings, size has a highly significa nt influence on preference and is perceived as the most important feature in forming preferences in both service conditions. Specifically, the middle-sized robot was expected to be the most preferred alternative within the communal condition, since it makes users feel comfortable during the interaction and is most likely to activate feelings of warmth. On the other hand, the large-sized robot was expected to be preferred for the exchange condition, as being tall is more likely to be associated with competence. In contrast to the latter expectation, the middle-sized robot (1.2 m) is the preferred alternative for both conditions. Thereby, current findings confirm the work of Hiroi & Ito (2009, 2012) stating that users prefer a robot that is lower than the user’s eye-height. Furthermore, results seem to confirm the assumption that large robots are more likely to evoke feelings of anxiety (in a communal condition), since the large robot is the least preferred alternative. However, as there are no preference differences between the conditions, this assumption appears to be true for the exchange condition as well.

In addition, current findings confirm the expectation that warmth judgments are more pronounced for middle-sized robots, whereas large robots are perceived as more competent. Consequently, it seems that the psychological effect of height during a human conversation (whereby taller people are perceived as more competent) extend to human-robot interactio ns. Furthermore, findings reveal that a small robot is considered as both the warmest and least competent alternative. Thereby, the assumption that the attributed child-like abilities of a small robot enhance feelings of warmth and reduce feelings of competence appears to be true. Lastly, since warmth and competence judgments are both positive for the middle-sized alternative, the middle-sized robot seems to represent the perfect balance between warmth and competence, which possibly explains why respondents prefer this alternative.

5.1.3 Color

(40)

40

On the contrary, results reveal that neither a red nor a blue color significantly influe nces respondents’ preferences for the different robot alternatives. A possible explanation for lack of color preferences could be that service users experience difficulties in building complex color associations for robots due to its novelty within the service environment (Priluck Gross & Wisenblit, 1999). In addition, research on color psychology argues that a red color serves as a warmth cue, whereas a blue color serves as a competence cue (Mehta et al., 2011). However, since color does not have a significant effect on warmth and competence judgments, this finding is not supported by the current research.

In conclusion, results reveal that robot appearance significantly influences robot preference by means of its shape and size, which is in line with previous research findings. Furthermore, this research brings new insights into the optimal robot design, since respondents prefer a middle-sized, mechanoid robot, whereas they are indifferent about the color of the robot. According to the results, preferences do not differ between communal and exchange service settings. Nevertheless, this study is one of the first to demonstrate the effects of warmth and competence judgments on robot preference caused by the central attributes of robot appearance. Specifically, findings indicate that the different alternatives of robot shape and size can serve as warmth and competence cues, which influence perceptions of the robot and ultimately preference.

5.2 Managerial Insights

The Effects of Robot Shape, Size, and Color on Robot Preferences

The Effects of Robot Shape, Size, and Color

on Robot Preferences

MSc Marketing

Management & Intelligence Thesis

The Effects of Robot Shape, Size, and Color on Robot Preferences

ABSTRACT

PREFACE

TABLE OF CONTENTS

1. INTRODUCTION

2. THEORETICAL BACKGROUND

3. METHODOLOGY

𝑈

= ∑ 𝛽

𝑥

ⅹ

4. RESULTS

HYPOTHESIS

RESULT

5. DISCUSSION