Using social robots to help consumers
overcome feelings of embarrassment
The moderating role of embarrassment on the
evaluation of human and robotic service providers
By
2
Using social robots to help consumers
overcome feelings of embarrassment
The moderating role of embarrassment on the evaluation of
human and robotic service providers
Author
Akkelyna de Haan (S2673371)
a.b.a.de.haan@student.rug.nl
Tsjerkepaed 22
9043VM Wier
+31653677939
University of Groningen
MSc Marketing Management
Department of Marketing
Faculty of Economics and Business
First supervisor
Second supervisor
3
Abstract
The increasing integration of personal service robots in service settings changes the interplay
between customers and organizations. It is expected that customer service experiences of the future will be particularly shaped by the extent to which technology engages customers on a social level. Social robots can be a full substitute of human service providers or they can be used besides human service providers. This study investigates the potential value of integrating social robots in
embarrassing service settings. Service providers are the primary source of feelings of embarrassment in service encounters. Consumers feel embarrassed because they are afraid to be evaluated by the service providers. Social robots might eliminate the fear of being evaluated because machines do not judge and might therefore be preferred above human service providers in embarrassing service settings. This study aims to investigate how embarrassment influences the likeability of and perceived safety with a human service provider, a machine-like robotic service provider and a human-like robotic service provider. A survey-based experiment showed that people like the human service provider significantly more than the human-like robotic service provider and machine-like service provider in a non-embarrassing service setting. Furthermore, the study showed that embarrassment significantly influences the perceived safety with the service providers. People felt safer with the human service provider in a non-embarrassing situation but when they felt more embarrassed they felt safer with the machine-like robotic service provider. An extra analysis showed that empathy can explain the influence of embarrassment on the perceived safety with the service provider. In sum, this study shows the potential positive role that social robots can play to help consumers overcome feelings of embarrassment caused by service providers.
4
Table of content
1. Introduction ... 6
2. Theoretical background ... 10
2.1.1. Defining social robots ... 10
2.1.2. Social robots that elicit the uncanny valley effect ... 10
2.1.3. Consumer evaluation of robotic service providers ... 11
2.1.4. Social robots in embarrassing services ... 12
2.2 Conceptual model ... 15
2.3 Hypotheses ... 16
2.3.1. The effect of the service provider type on the consumer evaluation of the service provider ... 16
2.3.2. The influence of embarrassment on the evaluation of a service provider ... 17
3. Research design ... 20 3.1. Research method ... 20 3.2. Data collection ... 20 3.2.1. Sample ... 20 3.2.2. Procedure ... 20 3.3 Measures ... 20
3.3.1. Service provider type ... 20
3.3.2. Level of embarrassment ... 21
3.3.3. Consumer evaluation of the service provider ... 21
3.3.4. Empathy and social judgment ... 22
3.3.5. Attention check ... 22 3.3.6. Control variables ... 22 3.4. Plan of analysis ... 23 4. Results ... 24 4.1 Descriptive statistics ... 24 4.1.1. Demographic variables ... 24 4.1.2. Randomization check ... 24 4.1.3. Descriptive statistics ... 25 4.1.4. Correlations ... 25
4.1.5. Analysis of manipulation checks ... 27
4.2 The influence of the service provider type on the evaluation of the service providers ... 28
5
4.2.2. Do the likeability and perceived safety differ for the service providers? ... 28
4.2.3. Conclusion ... 29
4.3 The influence of embarrassment on the evaluation of the service providers ... 30
4.3.1. Assumption checks ... 30
4.3.2. Manipulation check of embarrassment in the pharmacy setting ... 30
4.3.3. Effect of embarrassment on the likeability and perceived safety ... 31
4.3.4. Conclusion ... 32
4.4 The role of empathy and social judgment ... 32
4.4.1. Mediating role of empathy ... 33
4.4.2. Mediating role of social judgment ... 34
4.4.3. Conclusion ... 35
5. Discussion ... 36
5.1.1. Moderating influence of embarrassment on the perceived safety with the service providers ... 36
5.1.2. Can empathy explain why respondents feel safer with the robotic service providers? ... 36
5.1.3. Can social judgment explain why respondents feel safer with the robotic service providers? ... 37
5.1.4. Higher likeability for the human-like robotic service provider then the machine-like robotic service provider ... 37
5.2. Practical implications ... 38
5.3 Limitations and future research ... 38
6. Conclusion ... 40
References ... 41
Appendix A – Questionnaire ... 46
Appendix B – Randomization check ... 51
Appendix C – Analysis normal distribution hypothesis 1 ... 54
Normality tests ... 54
Appendix D – Normal Q-Q plots hypothesis 1 ... 55
Appendix E – Analysis normal distribution hypothesis 2 ... 56
6
1. Introduction
Imagine that you walk into the reception of the hotel and you are being welcomed by a life-like robot. Next, the robot gives you your room key and explains you where your room and the other hotel facilities are. This might sound like a scene from some kind of science-fiction movie, but the opposite is true. This is a real life situation in a luxury hotel in the city of Hefei in China where the robot works in the lobby of the hotel (Daily mail, 2018). However, this hotel in China is not the only service organisation that uses robots. Robots are used in elderly care (Independent, 2018),
restaurants (FoxNews, 2014), hotels (Daily mail, 2018) and theme parks (PCC Courier, 2018). The integration of robots in service settings is rapidly increasing and is changing the nature of service experiences and service interactions for consumers (Kirby, Forlizzi and Simmons, 2010; Van Doorn, Mende, Noble, Hulland, Ostrom, Grewal and Petersen, 2017).
The robots used in the above called service settings are personal service robots (U.N. and I.F.R.R. 2002). These robots assist or entertain people in domestic settings or in recreational activities (Thrun, 2004). The integration of personal service robots in service settings changes the interplay between customers and organizations. It is expected that customer service experiences of the future will be particularly shaped by the extent to which technology engages customers on a social level. This means that technologies can engage in meaningful social encounters and can develop lasting relationships with humans (Van Doorn et al., 2017). The ability of the new social technologies to engage customers on a social level differentiates them from self-service technologies. Meuter, Ostrom, Roundtree and Bitner (2000) define self-service technologies as ‘’technological interfaces that enable customers to produce a service independent of direct service employee involvement.’’ Hence, self-service technologies create marketspace transactions where there is no interpersonal contact between the service provider and the customer. The new social technologies, like the personal service robots, do facilitate interpersonal contact through the interaction between a customer and a social agent (Van Doorn et al., 2017).
In this way, social technologies will have substantial influences on customers’ experiences. An important element in these future customer service experiences is the human-robot interaction. This is the way humans and robots communicate with each other during the social encounters, either verbally or non-verbally (Murphy, Nomura, Billard and Burk, 2010). It is expected that humans and social robots will interact in a collaborative and socially enriching way. To increase the effectiveness of these human-robot interactions it is valuable to research how the communication between humans and robots needs to be designed (Van Doorn et al., 2017).
A lot of research has been done on human-robot interactions in service settings. Most of this research focuses on the design of the robot (Siegel, Breazael and Norton, 2009; Goetz, Kiesler and Powers, 2003), the behaviour/movements of the robot (Gockley, Forlizzi and Simmons, 2007; Bartneck and Forlizzi 2004) and the personality/character the robot needs to have (Gockley, et al., 2005). One approach that is often used by researches on human-robot interaction is an
anthropomorphic design of the robot (Bartneck and Forlizzi, 2004; Duffy, 2003; Fink, 2012). An anthropomorphic design means that the robot has a humanlike design with ‘human social’
7 more familiar, explainable or predictable. This makes that people feel more empathy for the robots and therefore will like them more and accept them more easily (Fink, J. 2012). However, artificial service agents that too closely resemble a human could be perceived as creepy which gives people a feeling of eeriness. This phenomenon is known as the ‘uncanny valley effect’ (Mori, MacDorman and Kageki, 2012). Trying to increase the degree of human likeness of a robot creates the risk of falling into the uncanny valley. Therefore Mori et al. (2012) recommend designing robots with a nonhuman, machine-like design that evoke a safe level of affinity by humans and that prevent the uncanny valley effect from happening. Thus, the uncanny valley effect might explain why people will prefer
machine-like social service robots above human-like social service robots in future service encounters.
However, little research is done about the uncanny valley effect yet (Gray and Wegner, 2012; Kätsyri, Förger, Mäkäräinen and Takala, 2015). One question that Gray and Wegner (2012) raise for future research is investigating whether humans will always be unnerved by human-like social robots. The occurrence of the uncanny valley effect can be influenced by a number of factors. Rosenthal-von der Pütten, Krämer, Becker-Asano, Ogawa, Nishio and Ishiguro (2013) found that some participants in their study virtually ignored the robot in the restaurant because it was not of high interest for them. The participants stated that they did recognize the robot as a robot but that getting their coffee was a higher priority for them than exploring the robot. Another research found that people’s interest in robots in a train station differed depending on whether they were in a hurry or not (Hayashi, Sakamoto, Kanda, Shiomi, Koizumi, Ishiguro, Ogasawara and Hagita, 2007). These findings suggest that situational factors can play a role in whether humans are interested in robots and possibly whether people feel unnerved by a robot. Situational factors might create a situation where the uncanny valley effect evokes but people are less aware of that because some other issue is of higher interest for them at that moment (Rosenthal-von der Pütten et al., 2013). Consequently, the
unnerving appearance of a human-like robot might become less pronounced for the consumer in a particular situation, thereby preventing that the consumer rejects the human-like robotic service provider.
Being in an embarrassing service setting might be a situation where the unnerving appearance of a human-like robot is of less importance for a consumer. Consumers often feel embarrassed in a service setting with high personal contact (Moore, Moore and Capella, 2005). Personal contact in services can take several forms, from customers interacting with the service provider, to interacting with other customers in the service setting, to just being in a crowded service place. Research has shown that as one experiences a service, interpersonal encounters can add or detract from evaluations of the service and influence patronage decisions (Moore et al., 2005). The presence of another person in a service setting, being a service provider or another customer, can be a sufficient condition eliciting thoughts that one is being evaluated. This can elicit feelings of embarrassment when unwanted events communicate undesired information about oneself to others (Dahl, Manchanda and Argo, 2001).
Embarrassment is a self-conscious emotion in which a person feels awkward or flustered in other people’s company or because of the attention of others, as, for example, when being observed engaging in actions that are subject to mild disapproval from others (VandenBos, 2006).
8 Aydinoglu, 2015). This thesis focusses on category 1 ‘public embarrassment’ of the typology of Krishna et al (2015). They define public embarrassment as ‘embarrassment felt in a public context where a person is both seen and appraised by others’. Using this definition, embarrassment is used as a situational variable.
One mechanism of embarrassment is the social evaluation model which explains that negative evaluations from others trigger embarrassment because individuals are concerned about how others might evaluate them (Krishna, et al., 2015). The reason why people are concerned about how others evaluate them is because we create some ideal social identity for ourselves. As social identity theory explains, an individual strives to achieve a satisfactory concept or image of him: the social identity (Stets and Burke, 2000). Embarrassment is a response to threats upon ones social identity and it creates a concern for how one is being appraised by others (Krishna, et al., 2015). Therefore, at that moment, the threat to someone’s social identity might be of higher interest for a customer than the unnerving appearance of a human-like robotic service provider. Consequently, in an embarrassing service setting, the uncanny valley effect will have less influence on the evaluation of the robot. Some service settings that create feelings of embarrassment are healthcare, financial services and pharmacies (Grace, 2009). Consumers often avoid such services because they feel embarrassed and are afraid how they will be evaluated by others. Consumers do not buy condoms because they find it embarrassing to purchase them (Krishna, et al., 2015). Furthermore, HIV patients do not collect their HIV-medicines out of embarrassment (Bearingpointcarribbean.com, 2015). Also, many people who would benefit from wearing hearing aids do not purchase or use them because they feel
embarrassed (Lacobucci, Calder, Malthouse and Duhachek, 2003). However these consumer behaviours, like not getting the medicines they need, often pose a threat to the consumers’ well-being. But consumer will not change their behaviour as long as these service settings keep eliciting feelings of embarrassment for the consumer. In this way, embarrassing service settings will discourage consumers to increase their own well-being.
Research has shown that the service provider is the primary source of feelings of embarrassment in service encounters (Grace, 2009). The service provider induces the fear of negative evaluations by consumers which poses a threat to the consumers’ social identity. Eliminating the social evaluation element from the service setting might prevent individuals from decreasing their well-being because then the service will not pose a threat to one’s social identity anymore. A possible way to do this is introducing social service robots in embarrassing service settings. A social robot is used because people feel more comfortable when confronted with a more socially communicative version of the robot (Heerink, Kröse, Evers and Wielinga, 2006).
9 robots above human service providers. Therefore, this thesis will focus on the following research questions: ‘To what extent does the consumer preference for a machine-like and a human-like robotic
service provider differ from a human service provider?’ and ‘To what extent does the level of felt embarrassment influences the preference for the different types of service providers?’
This research is theoretically relevant because it can enrich the scientific knowledge about the uncanny valley by investigating whether there are some situations where encountering a possible unnerving robot is of less importance for consumers (Gray and Wegner, 2012; Rosenthal-Von der Pütten et al., 2013). It also creates deeper understanding about how people react to robots in situations where they are embarrassed (Coeckelbergh, 2011). Furthermore it gives some insights in how social judgment influence the evaluation of social robots and if there are situations where consumers do prefer robots above humans (Bartneck, Bleeker, Bun, Fens and Riet, 2010). Finally, it broadens the low number of scientific research about the effect of embarrassment within the service encounter, which contains key factors such as service providers, customers and other customers (Grace, 2009). This research is socially relevant because it can enrich the scientific knowledge about how we can help consumers to overcome feelings of embarrassment (Dahl et al., 2001). This is important because purchase embarrassment can have serious consequences for the consumer well-being (Moore, Dahl, Gorn, and Weinberg, 2006; Bartneck et al., 2004).
10
2. Theoretical background
2.1.1. Defining social robots
Introducing the concept of social robots in embarrassing service situations requires a careful
definition of what is meant with a ‘social’ robot. Table 1 presents an overview of key literature. Duffy (2003) defines social robots as ”Physical entities embodied in a complex, dynamic and social
environment sufficiently empowered to behave in a manner conductive to their own goals and those of their community.” This means that a social robot has the capacity to engage in meaningful social interactions and lasting relationships with people (Van Doorn et al., 2017). Social robots can thus act as social agents giving humans the feeling of being with another. They make people feel that they engage with another social entity which are truly social in their appearance and behave in an interactive way (Van Doorn et al., 2017). Social robots can be a full substitute of human service providers or they can be used besides human service providers
To develop a social robot that can function as a fully substitute of a human service provider, the human-robot interaction is an important element in the future customer service experiences.
Therefore researchers try to develop social robots that are able to facilitate collaborative and socially enriching human-robot interactions (Triebel, Arras, Alami, Beyer, Breuers, Chatila, Chetouani,
Cremers, Evers, Fiore, Hung, Islas Ramírez, Joosse, Khambhaita and Kucner, 2016; Duffy, 2003). A key element of a socially acting and interacting robot is the physical appearance. A robot is of little use when it appears unfriendly or threatening, even if the behaviour of the robot fully complies with the socially normative rules. Therefore, a human-like appearance is often chosen for robots that are used in service settings (Triebel et al., 2016). Examples of human-like social robots are Spencer and
Pepper. Spencer is a robot that is used for passenger assistance at airports (Triebel, 2016). Pepper is a humanoid robot with diverse capabilities, including his capability to function as a receptionist (; Perera, Pereira, Connell and Veloso, 2017). But a human-like appearance also has the risk of raising expectations for certain cognitive capabilities which cannot be met by the robot (Triebel et al., 2016). Therefore some researches focus on developing a like social robot. Example of a machine-like social robot is the Human Support Robot of Toyota which has an abstract form that gives the robot the machine-like appearance. It is often used as an assistant in elderly homes (Toyota-global.com, 2018). This robot has a non-human design which makes it less probable that it falls into the uncanny valley.
2.1.2. Social robots that elicit the uncanny valley effect
Mori (1970) describes the uncanny valley effect as “the situation where people experience unsettling feelings for a robot when this robot’s appearance has become too humanlike.” According to Mori et al (2012), it can thus be seen as a certain process where people first experience pleasurable feelings of familiarity when they encounter a robot that possesses some humanlike features. These
11 Mori et al (2012) use the example of a prosthetic arm to illustrate the uncanny valley effect. In the beginning, prosthetic arms differed a lot from real ones. But recently prosthetic arms have improved so much that people cannot distinguish them from real arms at a glance. Prosthetic arms now simulate veins, muscles, finger nails, finger prints, and their colour even resembles human
pigmentation. The arms are so real that if people shake the hand, they are surprised that it is not a real one. This creates a sense of strangeness and no longer a sense of familiarity; it is uncanny. Gray and Wegner (2012) state that the uncanny valley effect of social robots have received relatively little empirical attention so far. The empirical researches that have been done can be divided into two research streams. One stream focuses on finding proof that the uncanny valley effect exists (Lewkowicz and Ghazanfar, 2012; Looser and Wheatly, 2010). Lewkowicz and Ghazanfar (2012) show that the uncanny valley effect starts emerging by infants at 12 months of age, which suggests that perceptual experience with real human faces is crucial for the uncanny valley effect to emerge. The second research stream focuses on finding explanations for why the uncanny valley effect arises (MacDorman, Green, Ho and Koch, 2009; Kätsyri, Förger, Mäkäräinen and Takala, 2015; Gray and Wegner, 2012; MacDorman and Entezari, 2015). Looking at the phenomenon from a simple level, MacDorman et al (2009) demonstrate that the uncanny valley effect describes a relationship
between comfort and the level of human likeness. Furthermore, Gray and Wegner (2012) found that feelings of uncanniness are tied to perceptions of experience, which is seen as fundamental to humans. Also, individual differences can predict someone’s sensitivity to the uncanny valley, like people’s sociocultural constructions and biological adaptations for threat avoidance, such as our fear and disgust system (MacDorman and Entezari, 2015). However, despite the recent findings about the uncanny valley, more research is needed to determine the exact conditions under which the uncanny valley phenomenon arises (Kätsyri et al., 2015). One interesting future research stream is to
investigate if there are situations where the unnerving nature of a human-like social robot becomes of less interest for people resulting in a situation where the human likeness of a robot has less influence on how consumers evaluate the robot (Rosenthal-Von der Pütten et al., 2013; Hayashi et al., 2007).
2.1.3. Consumer evaluation of robotic service providers
Before the influence of robotic service providers on the consumer evaluation can be investigated, it is important to know which dimensions consumers use to evaluate robots. Most researchers look at five key concepts to investigate the consumers’ evaluation of social robots (Li, Rau and Li, 2010; Eyssel, Kuchenbrandt, Hegel and De Ruiter, 2012; Haring, Silvera-Tawil, Takahashi, Watanabe and Velonaki, 2016). These concepts are the anthropomorphism, animacy, likeability, perceived intelligence and perceived safety of the robot (Bartneck, Kulic, Croft and Zoghbi, 2008).
Anthropomorphism refers to the degree that human characteristics or behaviours are attributed to nonhuman things, like robots. Animacy refers to the degree that robots are perceived as lifelike. Likeability refers to the degree that people form positive impressions of the robots. The perceived intelligence refers to the extent that people perceive a robot as intelligent. The perceived safety refers to the user’s perception of the level of danger when interacting with a robot and the level of comfort during the interaction (Bartneck et al., 2008).
12 al., 2010). But researchers not only test how the robot appearance influences the consumer
evaluation of the robot. They also investigate how other design elements of the robots influence the consumer evaluation of the robot. Eyssel et al (2012) found that vocal cues clearly influence
judgments of the robots. Consumers liked a robot with a human voice more than a robot with a computer voice. Besides looking at how robots are evaluated by consumers, research also looks at how robots are perceived in comparison to other technologies. Kidd and Breazael (2004) found that robots are seen as more engaging, more credible and more informative than an animated character. Also social robots are perceived as more enjoyable to interact with (Kidd and Breazael, 2004; Heerink et al., 2006).
There is a growing knowledge about how social robots are evaluated by consumers. However, little research is done about how social robots are evaluated when encountered in different situations or service settings. Broadbent et al (2009) showed that healthcare robots are evaluated more positive and are accepted more when the robots’ role, appearance and behaviour match the human user’s needs. These needs differ for each service setting. So more research can be done to investigate how consumers react to social robots in different situations (Broadbent et al., 2009). Embarrassing services are one situation type where investigating the influence of social robots’ evaluation has a high relevance. There is a high relevance because consumers often avoid embarrassing services which can have detrimental consequences for the consumer well-being (Lacobucci et al.,2003). 2.1.4. Social robots in embarrassing services
Embarrassment is a common emotion in service encounters and service providers can play an important role in creating a feeling of embarrassment (Grace, 2009; Wu and Mattila, 2013; Lau-Gesk and Drolet, 2008). Looking at consumer embarrassment, a distinction between public and private embarrassment can be made (Krishna et al., 2015). Public embarrassment is defined as felt embarrassment in a public context where the consumer is both seen and appraised by others. Private embarrassment is defined by Krishna et al (2015) as “felt embarrassment when others are not present or when the audience present does not observe the embarrassing transgression but you imagine being evaluated.” Public embarrassment is the dominant view of embarrassment and is therefore most used in research about consumer embarrassment.
Little research about consumers and embarrassment is done so far (Krishna et al., 2015). Researchers who have investigated consumer embarrassment found that the elicitation of consumer
embarrassment is socially shaped by service providers’ actions (Wu and Mattila, 2013). Furthermore, customers are even more likely to boycott the service when the embarrassment experienced by them is triggered by the service provider rather than being due to a consumers’ own action (Grace, 2009). There are also individual differences to the sensitivity of feeling embarrassed in a certain situation. Consumers with a high public self-consciousness feel more embarrassed than consumers with a low public self-consciousness (Lau-Gesk and Drolet, 2008).
13 people often feel embarrassed when they need to undress in front of a doctor because they are afraid for what the doctor might think of them. It is expected that the robotic doctors might eliminate this social evaluation element, thereby preventing that patients feel embarrassed during the examination. The results of the study show that the respondents were less embarrassed when interacting with a technical box than with a highly human-like robot. However Bartneck et al (2010) only compared the feelings of embarrassment for a machine-like and a human-like robot and did not include a human service provider in the experiment. In sum, it is not known if there are situations where the level of felt embarrassment differs between a human service provider, a machine-like robotic service provider and a robotic service provider. Further research can investigate this to see if robots can really help consumers to overcome feelings of embarrassment (Bartneck et al., 2010).
Table 1 Literature review
Authors Key concept(s) Summary of findings
2.1.1. Defining social robots
Duffy (2003) Social robots Incorporating human features in social robots can facilitate the development of an enriching social interaction between humans and social robots
Triebel et al. (2016) Social robots A human-like appearance of social robots needs to be balanced in such a way that social robots can meet the expectations that people expect based on their appearance.
Van Doorn, Mende, Noble, Hulland, Ostrom, Grewal and Petersen (2017)
Social robots Social robots can act as social agents giving humans the feeling of being with another. They make people feel that they engage with another social entity which are truly social in their appearance and behave in an interactive way.
2.1.2. Social robots and uncanny valley effect
Mori (1970) Uncanny valley A humanlike appearance of a robot leads to pleasurable feelings of familiarity among people until the robot becomes too much a resemblance of a human, resulting in feelings of eeriness.
Hayashi et al (2007) Social robots and uncanny valley
The uncanny valley effect only happens when an individual shows interest in the robot and is not distracted by some other goal at that moment.
MacDorman, Green, Ho and Koch (2009)
Uncanny valley and human-likeness of the robot
The uncanny valley effect arises because people do not feel comfortable anymore with the level of human-likeness of the robot.
Looser and Wheatly (2010)
Social robots and uncanny valley
People are very sensitive to human forms and especially the design and human-likeness of the robot’s face has a big influence on the perceived animacy of a robot.
Lewkowicz and Ghazanfar (2012)
Uncanny valley The uncanny valley effect can already emerge at infants of 12 months of age, which shows that people need perceptual experience with real human faces before the uncanny valley can arise.
Mori, MacDorman and Kageki (2012)
Uncanny valley and human-likeness of the robot
14 Gray and Wegner
(2012)
Social robots and uncanny valley
Perceptions of experience can explain why the uncanny valley effect arises. It also shows that people have a general aversion against social technologies, like social robots.
Rosenthal-Von der Pütten et al (2013)
Social Robots and uncanny valley
People are only receptive to the unnerving
appearance of human-like robots when people take the time to look at the robot.
Kätsyri, Förger, Mäkäräinen and Takala (2015)
Social Robots and uncanny valley
Using a human-like design does not automatically lead to the uncanny valley effect. The uncanny valley effect arises under very specific conditions but more research is needed to identify these conditions.
MacDorman and Entezari (2015)
Social robots and uncanny valley
Individual differences, like culture and religion, influence how sensitive people are for the uncanny valley effect.
2.1.3. Evaluation of robotic service providers
Kidd and Breazael (2004)
Evaluation of social technologies
They physical embodiment of a social robot positively influences the consumer’s perceived credibility and likeability of the robot. This also explains why social robots are liked more than animated characters.
Heerink, Kröse, Evers and Wielinga (2006)
Evaluation of social robot People feel more comfortable with social communicative robots and also expressed themselves more while communicating with a social robot.
Broadbent, Stafford and MacDonald (2009)
Evaluation of social robot Ways to increase the acceptance of healthcare robots are to match the needs of a human user with the role, appearance and behaviour of the robot.
Li, Rau and Li (2010) Evaluation of social robot The appearance of a robot influences the judgments of the robot while the task of a robot influences people’s behaviour towards the robot.
Eyssel, Kuchenbrandt, Hegel and De Ruiter (2012)
Likeability of social robot The type of voice of a social robot influences the consumers’ likeability of the robot.
Haring, Silvera-Tawil, Takahashi, Watanabe and Velonaki (2016)
Evaluation of social robot The appearance of a robot influences the extent to which consumers like the robot, perceive it as intelligence and feel safe with the robot.
2.1.4. Embarrassment in services and for robots
Lau-Gesk and Drolet (2008)
Embarrassment in services and social judgment
The extent to which people feel afraid for negative social judgment differs between people with a high public self-consciousness and people with a low public self-consciousness.
Grace (2009) Embarrassment in services
The service provider is an important source of embarrassment and influences consumers’ repurchase intentions for a service setting.
Bartneck et al. (2010) Anthropomorphism, embarrassment in robot interactions
Using a technical box as doctor decreases the level of embarrassment felt during the medical
15 Wu and Mattila
(2013)
Embarrassment in services
Consumer embarrassment arises through the actions of service providers and the presence of other customers in the service setting.
Krishna et al (2015) Embarrassment in services
Consumer embarrassment can be divided into public and private embarrassment.
2.2 Conceptual model
The literature review in the previous paragraphs shows that there is little research done about how consumers’ evaluate social service robots in different situations. However, using the existing research about the uncanny valley as a guideline, it is expected that, in general, consumers will evaluate machine-like robots more positive than human-like robots. MacDorman et al (2009) demonstrate that the uncanny valley effect describes a relationship between comfort and the level of human likeness. So the higher the human likeness of a social robot is, the lower the felt comfort towards the robot will be. Furthermore people are still a bit unnerved by the whole idea of integrating robots in services (Gray and Wegner, 2012). Therefore, it is expected that consumers prefer the human service providers above both robotic service providers. Additionally, the uncomforting human-likeness of a human-like social robot determines a more negative consumer evaluation of a human-like robotic service provider than a machine-like robotic service provider.
When looking at embarrassment in services, research has shown that embarrassment is a common emotion in services and that service providers can play an important role in creating feelings of embarrassment. Wu and Mattila (2013) showed that the actions of a service provider can elicit feelings of embarrassment among consumers. Consequently service providers are a source of embarrassment than can cause irrational behaviour of consumers. It is therefore highly relevant to investigate the effect of a robotic service provider on the felt embarrassment in a service situation (Grace, 2009). In order to investigate if the level of felt embarrassment in a situation has an influence on the consumer evaluation of a service provider, embarrassment is added to the conceptual model as a moderator. Main interest is to test whether and how the evaluation of the different service providers varies between a low and high embarrassment situation. The conceptual model is illustrated below (figure 1).
16
2.3 Hypotheses
2.3.1. The effect of the service provider type on the consumer evaluation of the service provider
As the conceptual model in figure 1 illustrates, it is expected that different types of service providers affect the consumer evaluation of the service providers. This effect is derived from existing literature about anthropomorphism and the uncanny valley effect. The concept of the uncanny valley effect has been introduced by Masahiro Mori in 1970. He noticed that, as robots appear more humanlike, the level of felt familiarity towards this robot increases until we come to a valley (Mori et al., 2012). Consequently, artificial entities that too closely resemble a human can be perceived as creepy and thus create a feeling of eeriness among people (Mori et al., 2012). MacDorman et al (2009) therefore describe the uncanny valley as a simple relation between the comfort level and the human likeness. People feel uncomfortable when robots become too human like.
One concept that is related to the uncanny valley effect is anthropomorphism. Anthropomorphism is the tendency to attribute human characteristics to inanimate objects or animals and helps us
rationalise the actions of these objects or animals (Duffy, 2003). Anthropomorphism is a strategic action of humans used to interpret the behaviour of a (strange) entity (Duffy, 2003). Present social robots are very human-like in appearance but their movements and behaviour fall short in simulating the movements and behaviour of humans. This creates a situation where humans perceive the robots as humans through anthropomorphizing their appearance but their behaviour identifies them as robots (Walters, Syrdal, Dautenhahn, Te Boekhorst and Koay, 2007). Using this rationale, human-like robots create strange, confused and eerie feelings by humans leading the uncanny valley effect to arise.
However, there are also researchers that investigate whether there are situations where higher human likenesses of social robots lead to a more positive evaluation of the robot (Fink, 2012; Niculescu, Van Dijk, Nijholt, Li and See, 2013) Fink (2012) states that people feel more empathy for a human-like robot and therefore like a human-like robot more and accept them more easily. Brave, Nass and Hutchinson (2005) found evidence for this statement by showing that modelling
empathetic emotions in an agent increased the positive ratings concerning its likeability and trustworthiness. Modelling empathetic emotions in an agent means designing the robot to show empathetic behaviour towards the customer (Niculescu et al. 2013). Other studies have found that empathy in agents also increases comfort (Bickmore and Schulman, 2007) and user’s satisfaction (Prendinger, Mori, Iszhisuka, 2005). This research is related to the question of Gray and Wegner (2012) if people will always be unnerved by human-like social robots. The example of empathy shows that anthropomorphism provides powerful physical and social features that might make the
unnerving appearance of a human-like robot less important (Duffy, 2003)
Despite the fact that physical and social features of human-like robots can be designed in such a way that the likeability for human-like robots increases, their likeability is still much lower than the likeability of machine-like robots (Bartneck, Kanda, Ishiguro and Hagita, 2007). This is because people form certain expectations regarding the capabilities and behaviour of a human-like robot (Zlotowski, Proudfoot, Yogeeswaran and Bartneck, 2015; Epley, Waytz and Cacioppo, 2007). Consumers’
17 meet these expectations. Machine-like robots lead to lower expectations which can be fulfilled, leading to a higher consumer satisfaction and likeability of the robot.
Although machine-like robots lead to a higher likeability than human-like robots, consumers still have some general aversion against robots. People are still adapting to the idea of integrating robots in their social lives (De Graaf and Allouch, 2013). As Gray and Wegner (2012) describe it: “people may refer to their car as upset or their spouse as robotic, but when a car really is upset or a spouse really is a robot, people find it unnerving.” Therefore it is expected that consumers will prefer a human service provider above any robotic service provider, both a machine-like and human-like robotic service provider.
Summarizing, due to the influence of the uncanny valley, anthropomorphism, the role of expectations and the adoption of robots, it is expected that consumers prefer human service providers above machine-like robotic service providers, leading the human-like robotic service providers as being the least preferred service provider. Thus, the following hypotheses can be formulated:
H1a: Consumers will evaluate a human service provider more positive than a machine-like robotic service provider.
H1b: Consumers will evaluate a machine-like robotic service provider more positive than a human-like robotic service provider.
2.3.2. The influence of embarrassment on the evaluation of a service provider
Following the theory of the uncanny valley and relating this to the concepts of anthropomorphism, the role of expectations and the adoption of robots, it is expected that consumers will prefer the human service provider above both robotic service providers in a non-embarrassing situation. However, as the conceptual model in figure 1 illustrates, main interest is to test whether and how the evaluation of the different service providers vary between a low and high embarrassment situation. The level of embarrassment might be a situational variable that makes the unnerving nature of a robot less important because the threat to the consumer’s social identity in the embarrassing situation is of higher relevance for the consumer at that moment.
This is expected because an individual strives to achieve a satisfactory concept or image of the self. Social identity theory explains that a social identity is the knowledge that you belong to a social category or group (Stets and Burke, 2000). The social identity is someone’s social or public image (Krishna et al., 2015). We use a social comparison process to categorize persons who are similar to the self, called the in-group, and to categorize persons who differ from the self, called the out-group (Stets and Burke, 2000).
18 One response people have when their social identity becomes threatened due to a negative
evaluation from others, is feeling embarrassed (Krishna et al., 2015). Particular service settings can trigger people to experience social identity threats because the received service can communicate undesired information about themselves (Dahl et al., 2001; Grace, 2009). Consumers are afraid of how the service provider might evaluate them. This leads to public embarrassment because they feel that their social identities become threatened. In this way, the elicitation of consumer
embarrassment is socially shaped by service providers (Wu and Mattila, 2013). However, using social robots as service providers might reduce consumer’s feelings of
embarrassment caused by service providers. Feelings of embarrassment can be reduced because social robots can eliminate the social judgment part (Bartneck et al., 2010). Consumers are not afraid that they are evaluated by the robot so no threat to their social identity is posed. This is because social evaluation is a human characteristic which people do not expect by a robot. Protecting their social identity is the highest priority of consumers in an embarrassing service situation (Krishna et al., 2015). Therefore, machine-like robotic service providers will be preferred by consumers above human service providers in an embarrassing situation. This is because in this situation, avoiding negative evaluations of the human service provider and consequently threats to the social identity has a higher relevance than the consumer’s willingness to adapt to the unnerving robot.
Summarizing, due to the influence of social evaluation and the fear for a threatened social identity, it is expected that consumers feel less socially judged by a machine-like robotic service provider then by a human service provider in an embarrassing situation. Therefore, consumers will prefer the machine-like robotic service provider above the human service provider.
The influence of embarrassment on the evaluation of a human-like robotic service provider is more uncertain. Using the knowledge about anthropomorphism, it is possible that consumers perceive the human-like robot as another person and thus describe certain human attributes to the robot, like social judgment. Social judgment means judging people based on some visible cues, like clothing, skin colour or received services (Miller, 1995). Human-like robots might increase the consumer’s
apprehension of unwanted judgments from the service provider by anthropomorphizing this attribute to the robot which causes embarrassment. In this way, a human-like robot might increase the level of social judgment felt by the consumer (Bartneck et al., 2010). Using this perspective, it is possible that consumers fear a negative evaluation of the human-like service robot in a high embarrassment situation, leading to a less positive consumer evaluation (Miller, 1995).
Consequently, this will lead to a higher preference for a machine-like service robot than for a human-like service robot in an embarrassing situation.
19 people are not embarrassed, they have no need to receive empathy from the robot. This will only decrease the likeability and trustworthiness of the robot (Cramer et al., 2010). Therefore, the overall evaluation of the human-like service robot and the appreciation of the empathy felt depend on the extent to which consumers feel embarrassed. As explained above, avoiding negative evaluations of the service provider has the highest priority for consumers in embarrassing situations. Getting help from an empathic service provider might decrease the apprehension of a negative evaluation which makes the unnerving appearance of the human-like service robot less important for the consumer. This will lead to a more positive evaluation of the human-like robot when there is a higher level of felt embarrassment. It is expected that if consumers perceive the human-like robot as empathetic they will prefer a human-like robotic service provider above a machine-like service provider in an embarrassing situation.
Summarizing, anthropomorphism can lead to different effects on the evaluation of human-like robotic service providers. It is interesting to see how the level of embarrassment influences how consumers anthropomorphize the human-like service robot and perhaps can make the unnerving appearance of the human-like service provider less relevant. Till now only one research has investigated the influence of a human-like compared to machine-like social robot on the consumer embarrassment. Bartneck et al (2010) found that respondents were less embarrassed when
interacting with a technical box than with a more human-like social robot. Based on these results and the social identity theory, it is expected that consumers will prefer machine-like robotic service providers above the human-like robotic service providers and human service provider in an
embarrassing situation. Furthermore, it is expected that consumers prefer human-like robotic service providers above human service providers because it is expected that humans will pose a greater threat to consumers’ social identity than human-like robotic service providers while social evaluation is more related to humans than social robots (Miller, 1995). This leads to the following hypotheses:
H2a: In the situation of high embarrassment, consumers will evaluate a machine-like robotic service provider more positive than a human-like robotic service provider.
20
3. Research design
3.1. Research method
A quantitative research is performed in the form of an experimental research design to test the hypotheses. Different scenarios are developed where the service provider type and the level of embarrassment are manipulated to test their influence on the consumer evaluation of the service providers. To test this, a 3 (human service provider vs machine-like robotic service provider vs human-like robotic service provider) x 2 (high embarrassment situation vs low embarrassment situation) between subjects factorial design is used. This leads to six different versions of the scenario-based questionnaire. Respondents are randomly assigned to the conditions. The questionnaire will be made with Qualtrics and will be distributed online through prolific. Respondents get a small monetary compensation for participating in the survey.
3.2. Data collection
3.2.1. Sample
As mentioned in the previous paragraph, the questionnaire is distributed online through prolific. Prolific is a crowdsourcing platform that helps researchers with recruiting research participants for data collection. The majority of the sample will consist out of respondents from the United Kingdom. In order to be able to get some meaningful results, the researcher strives to get a minimum of 20-30 respondents per experimental condition. The exact sample size and the demographic characteristics of the respondents is analysed in chapter 4.
3.2.2. Procedure
The questionnaire starts with some information about the voluntary participation of the study and the guarantee that all responses will be processed anonymously. The respondents are asked to give consent that they want to participate. Then the respondents receive some information about the scenario that they will get to see. Respondents are asked to imagine that they walk into their local pharmacy for some antibiotics for an ear infection (low embarrassment situation) or for some antibiotics for a sexually transmitted disease (high embarrassment situation). Next, the respondents are asked how much they feel embarrassed for the mentioned medicine. The questionnaire
continues by showing the respondents picture of the pharmacy with a service provider (human vs machine-like robot vs human-like robot) present. Subsequently, the respondents are asked to evaluate the service provider and the questions for the manipulation check are presented. After this, the respondents are asked to rate the level of felt embarrassment for the entire situation. Finally, the attention check and the control variables are presented. The questionnaire ends with thanking the respondent for completing the survey. The complete questionnaire is presented in appendix A.
3.3 Measures
The theoretical concepts introduced in the previous chapter are operationalised into questions that test the influence of the service provider type and the level of embarrassment on the consumer evaluation of the service providers. Some control questions, an attention check and a manipulation check are also added to the questionnaire. All items are presented in appendix A.
3.3.1. Service provider type
21 shown in figure 2. A manipulation check is used to test if the respondents indeed perceive the
robotic service providers as machine-like or human-like and the human service provider as human. To test this, the five-item scale for anthropomorphism is used (Bartneck et al., 2008).The
respondents are asked to evaluate the service provider on five dimensions: ‘fake vs natural’, ‘machinelike vs humanlike’, ‘unconscious vs conscious’, ‘artificial vs lifelike’ and ‘moving rigidly vs moving elegantly’. A seven-point semantic differential scale is used to test the attitudes towards the service providers using the items presented above as the bipolar points (Bartneck et al., 2008).
Figure 2 Service providers: human service provider vs machine-like robotic service provider vs human-like robotic service provider
3.3.2. Level of embarrassment
The level of felt embarrassment is measured for two situations. First of all, the respondents are asked to what extent they associate the emotion of embarrassment with the product. More specifically, respondents are asked to what extent they feel embarrassed for picking up the mentioned medicine (antibiotic for ear infection vs STD). Next, the respondents are asked to what extent they feel embarrassed in the pharmacy situation where they encounter a human, machine-like robotic or human-like robotic service provider that gives them the medicine. The three-item scale of Dahl et al (2001) is used to test the level of embarrassment felt by the respondents both times. Respondents are asked to what extent they agree with the statements that the presented situation made them feel ‘embarrassed’, ‘awkward’ and ‘uncomfortable’. A seven-point Likert answer scale is used, ranging from 1 = ‘strongly disagree’ to 7 = ’strongly agree’.
3.3.3. Consumer evaluation of the service provider
22 the consumer evaluation of the service provider, this study focuses on measuring the likeability and perceived safety. If consumers feel that they are socially judged by a service provider, they fear that their social identity is threatened which leads to feelings of embarrassment (Krishna et al., 2015). Consumers also feel less embarrassed and more comfortable when they interact with a service provider that they like (Grace, 2009). Therefore, likeability and perceived safety are included in the questionnaire.
The likeability is measured using five items: ‘dislike vs like’, ‘unfriendly vs friendly’, ‘unkind vs kind’, ‘unpleasant vs pleasant’ and ‘awful vs nice’. The perceived safety is measured using three items: ‘anxious vs relaxed’, ‘agitated vs calm’ and ‘quiescent vs surprised’. A seven-point semantic
differential scale is used headed with the instruction ‘please rate how the service provider made you feel’.
3.3.4. Empathy and social judgment
The process variables of empathy and social judgment are added to the questionnaire to investigate if these variables can explain the received results for the consumer evaluation of the service
providers when respondents prefer the human-like robotic service providers above machine-like robotic service providers. The seven-item scale of Kim et al (2004) is used to test the level of empathy felt. The respondents are asked to what extent they agree with the statements that the service provider is caring, respects their feelings and cares about the well-being of the consumer. All items are presented in appendix A. A seven-item likert answer scale is used ranging from 1 = ‘strongly disagree’ to 7 = ‘strongly agree’.
The eight-item fear of negative evaluation scale of Leary (1983) is used to measure the social judgment. This scale originally consists out of twelve items but research has shown that the four reverse-worded items have a weak relationship with theoretically related measures and therefore should be excluded from the questionnaire (Carleton, Donald, McCreary, Norton and Asmundson, 2006; Rodebaugh, Woods, Thissen, Heimberg, Chambless and Rapee, 2004). To make the questions suitable for this study, the questions are somewhat rephrased by adding the service provider to the questions. The respondents are asked to what extent they agree with the statements. A seven-item Likert answer scale is used ranging from 1 = ‘strongly disagree’ to 7 = ‘strongly agree’. The items are presented in the appendix.
3.3.5. Attention check
An attention check is used to test if the respondents have the right medicine in mind when answering the questions about embarrassment. The medicine that people have in mind influences their feelings and consequently how they answer the questions. Furthermore the attention check is also used to test if people read the instructions. The respondents are asked which medicine they were told to gather in the pharmacy. They were presented with three options: antibiotic for an ear infection, sexually transmitted disease and inflammation of bladder.
3.3.6. Control variables
23 eerie and unnatural. A seven-item Likert answer scale is used ranging from 1 = ‘strongly disagree’ to 7 = ‘strongly agree’.
3.4. Plan of analysis
24
4. Results
4.1 Descriptive statistics
4.1.1. Demographic variables
A total of 362 respondents filled in the survey. There was one respondent who did start the survey but all questions were unanswered. This respondent is removed from the dataset. Furthermore, 4 respondents did not want to participate in the survey and 14 respondents did not pass the attention check. These respondents are also removed from the dataset. There were four respondents that have not noted their age. However they did answer all the other questions. Therefore these
respondents were not removed from the dataset. This leads to a final dataset of 343 responses. It is a diverse sample where respondents come from different continents, like Europe, Oceania and
America. The majority of the respondents currently live in the United Kingdom (34.7%). Other countries that are well represented are the United States of America (16.6%) and Portugal (9.3%). The other demographic characteristics of the sample are shown in table 2 below.
Table 2 Demographic characteristics of the sample
Gender Age Education level Student
Male 195 56.9% Mean 28.94 years Primary school 2% Yes 38.2% Female 144 42% SD 9.64 years Lower General
Secondary Education
1.5% No 61.8%
Minimum 16 years Higher General Secondary Education
21%
Maximum 67 years Pre-University Education 16% Intermediate Vocational Education 2.6% University of Applied Science 8.5% University 48.4% 4.1.2. Randomization check
25 provider’. This is the only condition where there are more women than men in the group. However, the age, education level and student status does not differ from the other conditions and is also equal to the full sample. One possible explanation of the difference in gender is that there are 19 respondents removed from the dataset that might have disturbed the division of men and women in this group. However, it is a small difference and therefore it poses no threat to the random
assignment of the conditions. But it is interesting to see if this group will score differently in the upcoming analyses than the other groups
Table 3 The number of respondents per experimental condition.
Experimental condition Number of respondents Percentage
Low embarrassment and human service provider
57 16.6%
Low embarrassment and machine-like robotic
service provider 59 17.2%
Low embarrassment and human-like robotic
service provider 53 15.5%
High embarrassment and human service provider
58 16.9%
High embarrassment and machine-like robotic
service provider 56 16.3%
High embarrassment and human-like robotic
service provider 60 17.5%
4.1.3. Descriptive statistics
The mean, standard deviation and Cronbach’s alpha of all survey items are shown in table 4. All theoretical concepts were measured using multiple items. These have been computed to one variable. All items were kept when computing the average variable, except for one item in the scale for perceived safety. Perceived safety was measured with three items through asking the
respondents whether they felt (1) anxious vs relaxed, (2) agitated vs calm and (3) quiescent vs surprised with the service provider. When all three items are put together in one factor, they reach an insufficient Cronbach’s Alpha of α =.662. The reliability analysis shows that the factor reaches a sufficient level of Cronbach’s Alpha of α = .910 when the question whether they felt quiescent vs surprised (item 3) with the service provider was removed from the scale. Additionally, an explorative factor analysis has been performed to check if it is indeed justifiable to eliminate the item from the factor. The explorative factor analysis shows that item 3 has a very low loading of .120 compared to high loadings of .885 for item 1 and .893 for item 2. This shows that item 3, whether consumers feel quiescent vs surprised with the service provider, does not load sufficiently in the factor. Item 3 is therefore removed from the scale for perceived safety.
4.1.4. Correlations
26 when respondents are older they also feel more embarrassed in the whole situation where the service provider is present. However, an interesting finding is that they do not feel more embarrassed for the medicine that they need to gather at the pharmacy when they are older. A possible explanation is that there might be another source for the felt embarrassment than the medicine they need to gather when people get older.
Table 4 Mean, standard deviation and Cronbach’s alpha of survey items
Variabe Variable items Mean SD α
1. Age - 28.94 9.64 -
2. Education level - 6* 6* -
3. Embarrassment for medicine
Gathering this medicine in the pharmacy makes me feel: - Embarrassed - Awkward - Uncomfortable 3.64 2.08 0.963 4. Anthropomorphism of service provider
Please rate your impression of the service provider on these scales:
- Fake vs natural
- Machine like vs human like - Unconscious vs conscious - Artificial vs lifelike
- Moving rigidly vs moving elegantly
2.80 1.54 0.900
5. Likeability of service provider
Please rate your impressions of the service provider on these scales:
- Dislike vs like - Unfriendly vs friendly - Unkind vs kind - Unpleasant vs pleasant - Awful vs nice 4.36 1.35 0.911
6. Perceived safety with service provider
Please rate how the service provider made you feel:
- Anxious vs relaxed - Agitated vs calm
4.27 1.60 0.910
7. Empathy - The service provider responds to me mechanically - I believe this service provider will try to keep me from worrying
- I believe this service provider will respect my feelings
- I believe this service provider will show interest in me
- I believe this service provider will care about my psychological well-being
- I believe this service provider will show great concern for my well-being
- I believe this service provider cares about me
3.75 1.14 0.796
8. Embarrassment for whole situation
The situation presented made me feel: - Embarrassed
- Awkward - Uncomfortable
3.42 1.70 0.916
9. Social judgment - I worry about what the service provider will think of me even when I know it doesn’t make any difference
- I am frequently afraid of the service provider noticing my shortcomings
27
- I am afraid that the service provider will not approve of me
- I am afraid that the service provider will find fault with me
- When I am talking to the service provider, I worry about what the service provider may be thinking about me
- I am usually worried about what kind of impression I make by the service provider - Sometimes I think that I am too concerned with what the service provider thinks of me
- I often worry that I will say or do the wrong things in front of the service provider
* Education level is a categorical measure which means that the table shows the median and the range for this variable. The median is 6, representing the University of Applied science and the range is 6.
Table 5 Correlations between survey items
Variable 1 2 3 4 5 6 7 8 1. Age 2. Education level .128* 3. Embarrassment for medicine .002 .110* 4. Anthropomorphism of service provider .158** -.001 .021 5. Likeability of service provider -.013 -.005 .091 .573**
6. Perceived safety with service provider .057 -.036 .063 .421** .646** 7. Empathy .127* -.037 .024 .567** .543** .433** 8. Embarrassment for whole situation .108* .114* .414** -.043 -.266** -.397** -.079 9. Social judgment -.082 .046 .461** .174** .139** -.017 .226* * .389** N=343, *P < 0.05, **P < 0.01
4.1.5. Analysis of manipulation checks
4.1.5.1 Are the service providers perceived differently?
Before the influence of the service providers on the evaluation of the service providers can be tested, it is first needed to test if the respondents indeed perceived the service providers different. To test if the respondents indeed perceive the service providers differently, a One-way ANOVA has been performed with the ‘service provider type’ as the factor and the ‘anthropomorphism’ as dependent list. The One-way ANOVA test was significant, F(2,340) = 101.14, p < .001. This means that the type of service provider that respondents see has an influence on how the respondents evaluate the service provider on the anthropomorphism scale. The human service provider scored highest on human-likeness (M = 4.12, SD = 1.519) where the machine-like robotic service provider scored lower on human-likeness (M = 2.14, SD = 0.971) just like the human-like robotic service provider (M = 2.13, SD = 1.107). Planned contrasts showed that the anthropomorphism score for the human service
28 <.001. Furthermore, it showed that the anthropomorphism score for the human service provider differs significantly from the human-like robotic service provider, t(208.55) = -11.35, p = <.001. But, planned contrasts showed that the anthropomorphism score for the machine-like robotic service provider does not differ significantly from the human-like robotic service provider, t(221.15) = -.110, p = .913. This shows that, although it was intended to make the human-like robot more human than the machine-like robot, the human-like robotic service provider was not perceived more human than the machine-like robotic service provider. Nevertheless, the results show that the manipulation was still successful. Respondents perceive the human service provider as human and the robotic service providers as machines.
4.1.5.2 Are the respondents embarrassed?
To analyse if the felt embarrassment differs for the antibiotic for ear infection compared to antibiotic for STD, a One-way ANOVA is performed with ‘medicine type’ as the factor and the ‘felt
embarrassment’ as the dependent list. The One-way ANOVA test was significant, F(1,341) = 313.76, p<.001. This means that the mean for the antibiotic for ear infection (M = 2.18, SD = 1.50) differs significantly from the mean for antibiotic for a STD (M = 5.06, SD = 1.50). Therefore it can be concluded that the manipulation was successful because the respondents indeed feel more embarrassed for the STD than for the ear infection.
4.2 The influence of the service provider type on the evaluation of the service
providers
4.2.1. Assumption checks
The preference for a service provider type was tested by measuring how the respondents evaluate the service providers on two constructs: likeability and perceived safety. Therefore two ANOVA tests are performed to see how the different service providers are evaluated. But, before these tests can be done, it is first needed to perform the assumption checks. First of all it is assumed that the observations are independent from each other. A between-subjects factorial design is used to randomly assign the respondents to the different experimental groups and the respondents are evenly distributed between the groups. Second, A Levene’s test was executed to test if there is equality of variances in the sample. This test was successful for likeability (p = .78) and perceived safety (p = .37). Third, a normality test was executed for both variables. This is done to see if the likeability and perceived safety were normally distributed for the three types of service providers. Both variables were normally distributed. The full analysis is shown in appendix C.
4.2.2. Do the likeability and perceived safety differ for the service providers?
29 significantly lower than the likeability for the human-like robotic service provider (M = 4.22, SD = 1.33), t(340) = 2.43, p = .016. In sum, the respondents like the human service provider more than both robotic service providers. From the two robotic service providers, the respondents like the human-like robotic service provider more than the machine-like robotic service provider.
A second One-way ANOVA has been performed to test the influence of the service provider types on the perceived safety with the service provider. The ‘service provider type’ was the factor and the ‘perceived safety’ was added to the dependent list. The One-way ANOVA test for service provider type on perceived safety was not significant, F(2,340) = 2.45, p = .087. This means that the service provider type does not influence the perceived safety with the service provider. Planned contrast showed that the perceived safety with the human service provider (M = 4.53, SD = 1.63) does not differ significantly from the perceived safety with the machine-like robotic service provider (M = 4.18, SD = 1.60), t(340) = -1.67, p = .095. The perceived safety with the human service provider (M = 4.53, SD = 1.63) is significantly higher than the perceived safety with the human-like robotic service provider (M = 4.09, SD = 1.55), t(340) = -2.09, p = .037. Finally, the perceived safety with the machine-like robotic service provider (M = 4.18, SD = 1.60) does not differ significantly from the perceived safety with the human-like robotic service provider (M = 4.09, SD = 1.55), t(340) = -.43, p = .671. In sum, the respondents feel safer with the human service provider than with both robotic service providers. From the two robotic service providers, the respondents feel safer with the machine-like robotic service provider than with the human-like robotic service provider. The scores are also graphically illustrated in figure 2.
Figure 2 Evaluation of the service providers
4.2.3. Conclusion
