1
How can social robots change
the face of consumer banking
Author: Nirvana Naves Student number: 11152362 Supervisor: Dr. A. Weihrauch
A thesis submitted in fulfilment of the requirements for the degree of MSc. in Business Administration – Marketing Track
January 26th, 2018
Keywords:
Consumer banking sector, customer satisfaction, NARS scale, service encounter, service quality, SERVQUAL, social robots, user attitudes
2
Statement of originality
This document is written by student
Nirvana Naves, who declares to take
full responsibility for the contents of
this document. I declare that the text
and work presented in this document is
original and that no sources other than
those mentioned in the text and its
references have been used in creating it.
The Faculty of Economics and Business
is responsible solely for the supervision
of completion of the work, not for the
contents.
3
Abstract
Objective: This paper assesses the impact of
the form of the direct service encounter on perceived service quality under the condition of different levels of job fear by means of an experimental approach (N=216), in which the form of the direct service encounter (service robot vs. human assistant) and job fear (positive vs. negative) were manipulated in an online experiment.
The main objective of this paper is to answer the following research question:
How can banks improve the service encounter by comparing the perceived service quality of social robots and human
employees by consumers and what is the effect of job fear, anxiety and general
attitude towards social robots on the perceived service quality of the two types of
service agents?
Boundaries: The banking industry was
selected because of its history of technology usage in the service encounter and most people use their services on a regular basis. Several existing scales were used and adjusted in order to fit this research, including NARS (to measure the attitude towards social robots), RAS (to measure anxiety towards social robots) and the SERVPERF (to measure perceived service quality for both types of service agents).
Methods: To analyse the research question,
an online experiment was conducted by using a laptop in order to simulate a customer experience with either a service robot or a human assistant. Moreover, participants were randomly assigned to either a negative or a positive article about social robots (no job fear vs. job fear condition).
Results: Attitude towards social robots is positively related to the perceived service quality of social robots. However, attitude towards social robots is not negatively related to the perceived service quality of human employees, indicating that people who have a positive attitude towards social robots are not necessarily dissatisfied with
the perceived service quality of human service employees. Anxiety towards social robots is negatively related to the perceived service quality of social robots. However, anxiety towards social robots is not positively related to the perceived service quality of human employees. This indicated that people who are anxious towards social robots do not feel better about the perceived service quality of human service employees. Users who experience job fear do not have a more negative attitude towards social robots than people who do not. Education is positively related to attitude towards social robots. However, no support was found for a negative relationship between education and anxiety towards social robots.
Only the participants who experienced job fear and watched the robot video, were far more negative about the perceived service quality of social robots. Moreover, only the participants who watched the robot video and experienced job fear, were far more positive about the perceived service quality of human service employees.
Conclusion: Participants still favour human
service employees in terms of perceived service quality. Focusing on the five dimensions of perceived service quality, human service employees score higher on the tangibility, safety and empathy dimensions, while social robots are perceived to be more reliable and more responsive. Banks can improve their service encounter, by implementing social robots on specific tasks which demand reliability and responsiveness, while continue to employ human employees for tasks that demand safety, empathy and tangibility. For the implementation of social robots to be successful this research suggests that the company should invest in educating its consumers to limit anxiety and improve the general attitude towards social robots. Especially, videos are proven to be effective in reducing anxiety towards social robots. Organizations could invest in infomercials or documentaries, where people can become familiar with social robots and be informed about the advantages and their use.
4
Table of contents
Statement of originality ... 2 Abstract ... 3 Table of contents ... 4 Introduction ... 6 Literature review ... 7 1.1 Social robots ... 71.1.1 Social robots definitions and concepts ... 7
1.1.2 User attitudes towards social robots ... 8
1.1.3 Measurement of user attitudes towards social robots ... 8
1.1.4 Anxiety towards social robots ... 9
1.1.5 Job fear ... 9
1.1.6 Education ... 9
1.2 | Service quality in the consumer banking sector ... 9
1.2.1 Importance of service quality ... 9
1.2.2 Job fear and perceived service quality ... 10
1.2.3 Measurement of perceived service quality ... 10
1.2.4 Criticism on the SERVQUAL model ... 11
1.2.5 Comparing SERVQUAL and SERVPERF ... 12
1.3 Conclusion Literature review ... 12
1.4 Conceptual Framework ... 13
Contributions ... 13
Methodology ... 14
2.1 General design ... 14
2.2 Web based experiment ... 14
2.3 Stimuli development ... 15
2.4 Deception technique ... 15
2.5 Research Setting ... 16
2.6 Sampling procedure ... 16
Results and Discussion ... 17
3.1 Participants ... 17
3.2 Instruments ... 17
3.2.1 Dependent variable ... 17
3.2.2 Mediator Variables ... 18
5
3.3 Manipulation check ... 19
3.4 Statistical Analysis of Data and Results ... 20
3.5 Distribution of research variables ... 21
3.6 Hypothesis testing ... 21
3.8 Summary hypothesis testing ... 29
Discussion ... 30
4.1 Limitations and future research ... 31
4.2 Managerial implications ... 31
4.3 Theoretical implications ... 32
4.4 Conclusion ... 33
Bibliography ... 35
Appendix A Negative attitudes towards Robots (NARS) ... 39
Appendix B Anxiety towards Robots (RAS) ... 39
Appendix C Perceived service quality (SERVPERF) ... 39
6
Introduction
Banks are of great importance in the financial and economic development of a country. A competent banking system directly affects the growth of various sectors. It is therefore essential for banks to gain a better understanding of the always changing customer wants and adopt the latest technology in order to cope with these needs (Malhotra and Mukherjee, 2004). Changing customer preferences in combination with new ventures entering the financial sector have resulted in a more dynamic and competitive landscape. Efforts are being made to meet these new challenging needs. To illustrate, new forms of banking have emerged such as internet banking and phone banking (Lau et. al., 2013).
Providing excellent service quality is generally seen as a way to maintain a loyal customer base while searching for ways to acquire new customers (Lau et. al., 2013). In order to keep a competitive advantage, the banking sector must focus on service quality as they offer undifferentiated services in a competitive marketplace (Berry et. al., 1988). Outperformers in terms of service quality can gain a higher revenue, customer loyalty and a higher customer retention rate (Kumar et. al., 2010).
Empirical evidence shows that employee behaviour is one of the most important factors in customers purchase intentions (Koushiki Choudhury, 2013). Therefore, banks are dependent of their employees in order to achieve excellent service quality. Moreover, Buttle et. al. (2002) report significant correlations between service quality and satisfaction. Simultaneously, service employees are becoming obsolete in their traditional positions due to advances in robots, sensor fusion and deep learning algorithms (Larivière et. al., 2017). The benefits of technology driven service encounter are most likely to result in an increase of quality, efficiency, resistance against human-related errors and reduced variability in service performance (Heskett et. al, 2015).
Nowadays, the need of achieving higher service standards has triggered wide research interest in implementing social robots in the service encounter of the banking sector, but
the literature concerning social robots in the banking sector remains limited. Clearly, there is a growing need of research on the impact of social robots on organizational life (Wasen, 2010). This research is aimed at supporting the development of customer friendly service encounters which benefits both customer and bank. This paper will solely focus on socially interactive robots. Socially interactive robots function as partners, peers (Kanda et. al., 2004) or assistants (Wasen, 2010) meaning that they need to be adaptable and flexible in order to interact with humans. However, more knowledge on the performance and benefits of socially interactive robots, especially from a human perspective, are necessary before humans can exploit them (Breazeal, 2003).
The main objective of this paper is to answer the following research question:
How can banks improve the service encounter by comparing the perceived service quality of social robots and human
employees by consumers and what is the effect of job fear, anxiety and general
attitude towards social robots on the perceived service quality of the two types
of service agents?
First, the theoretical background of service robots and customer satisfaction will be examined. Secondly, a research model and hypotheses will illustrate the relationships between the variables. In the next section, the methodology including research design and data collection will be described. In the last part of this paper, the results will be discussed.
7
Literature review
.
1.1 Social robots
1.1
1.1.1 Social robots definitions and concepts
The subfield of artificial intelligence that studies how robots can communicate and assist humans is called social robots. What differentiates social robots from other types of robots is that they can relate to humans in a personal way. A social model is applied by humans in order to understand and communicate with them. Social robots are embodied, meaning that they have a physical body in which they operate (Dautenhahn and Billard, 1999). Social robots possess autonomy (Kiesler and Hinds, 2004). Kiesler and Hinds (2004) found three reasons why autonomous robots differ from any other computer technology. First, people perceive autonomous robots in a different manner than other computer technology. The human mental models of autonomous robots are often more anthropomorphic than of other systems (Friedman, Kahn, & Hagman, 2003). Second, autonomous robots are even more likely to mobile, bringing them closer to other people, roots and objects. Lastly, autonomous make their own decisions (decision autonomy), are capable of learning and emphasizing (behavioural autonomy), and they exert some power over the information they process (informational autonomy). (Kiesler and Hinds, 2004; Schulz-Schaeffer, 2008). Social robots are multifunctional: they can function as assistants (Wasen, 2010) social partners and tutors (Kanda et.al., 2004). For example, in a 18-day field trial two robots successfully communicated in English with first and sixth grade pupils (Kanda et. al., 2004). However, on a critical note, it is crucial
to avoid a technological mediated
heteronomy, meaning robotics being under the rule of another person, organization or government. This can result in a lack of transparency, therefore sufficient control over the robotics system must be guaranteed (Decker, 2012).
Service robots are considered successful when they meet the following three criteria:
first, a successful service robot must be completely autonomous. Second, the robots must be able to interact relevant information. Lastly, it must successfully manage its operations (Behan and O’Keeffe, 2008). Previous work doubted if robots are capable to perform tasks that humans routinely do. Their sensory system was relatively poor in terms of perceptual, cognitive and behavioural abilities compared to humans (Sakagami et. al., 2002).
Due to recent advances in the sensor technology the capabilities of social robots are growing. For example, Scholl (2013) introduced the connection of wireless sensor networks to the robot operating system. Furthermore, new research efforts are focused on the development of programming techniques of robots similar to the way humans are taught (Stoica, 1995). These techniques would narrow the gap between human and robot capabilities. Meuter and Bitner (1998) note two negative effects when limiting interpersonal contact. Reducing personal contact may lead to inability to recover from a service failure or difficulty of developing strong bonds. However, robots can be designed as a social companion. Welge and Hassenzahl (2016) identified six psychological superpowers of robots that are rooted in their thingness rather than
humanness. Robots are void of
competitiveness, have the ability to contain themselves, do not take things personally, have endless patience, can be unconditionally subordinated, and can assume responsibility. These qualities all relate to companionship and play an important role in the service encounter, however are difficult for humans to realize on a daily basis. Moreover, numerous studies suggest that in general humans interact with computers in a similar way as they do with other humans (Kanda et. al., 2004; Stoica, 2001). This theory is also known as the theory of Social Responses to Communication Technologies (Wager, van der Loos and Leifer, 2000). Furthermore, people may find the robots attractive due to its convenience or in order to avoid contact with personnel (Meuter and Bitner, 1998).
8
Research has shown that people tend to have either extremely positive or extremely
negative attitudes towards new
communication technologies. Social robots can be perceived as a new technology, therefore there is a chance that people will have either extremely negative attitudes or emotions. Such negative attitudes and emotions toward social robots can result in an unwillingness to adapt and communication avoidance behaviour toward social robots. It is therefore important to identify to what degree people hold negative attitudes and anxiety towards social robots, and to what extent this influences their perception of service quality provided by a social robot.
1.1.2 User attitudes towards social robots
Attitudes are psychologically defined as a relatively stable and enduring predisposition to be have or react in a certain way toward persons, objects, institutions, or issues. The source of these attitudes is cultural, familial and personal (Chaplin, 1991).
The human robot interaction (HRI) literature found that humans often create expectations regarding the interactions with social robots. Humans’ impressions concerning the robots knowledge, capabilities and autonomy are formed by intrinsic and extrinsic factors (Fong, Nourbakhsh, and Dautenhahn, 2003). Naturally, pre-knowledge and experience will affect the way humans perceive social robots. Expectations also play a significant role in human’s acceptance. The difference between the users’ expectations regarding the function of the robot and the actual function they perceive would significantly influence their behaviour. This difference was named the adaption gap. The results showed that the participants with positive adaptation gap signs had a significantly higher acceptance rate than those with negative ones (Komatsu, Kurosawa, and Yamada, 2012). Moreover, the design of the robot has aninfluence on the user’s attitude. Humans are inclined to sympathize based on resemblance (Stoica, 2001). Therefore, the exterior of the robot must be humanoid, in order to feel natural for people to communicate in a human-like way (Billard and Hayes, 1997). Furthermore,
people’s attitudes are strongly influenced by science fiction (Khan, 1998). In this study, he found that people prefer a round shaped robot with machine-like appearance and a serious personality. Secondly, verbal communication with a human-like voice is more positively rated (Khan, 1998).
On the other hand, there is also evidence that consumers see the employment of social robots in the banking sector as a threat. Some consumers are uncertain about the resolution of technology issues (Meuter and Bithner, 1998). Other prefer to communicate with people as they consider the service encounter as a social matter (Zeithaml and Gilly, 1987). For some customers the costs of learning how to deal with robots will be too high (Gatignon and Robertson, 1991).
H1a: Attitude towards social robots is positively related to the perceived service quality of social robots.
H1b: Attitude towards social robots is negatively related to the perceived service quality of human employees.
1.1.3 Measurement of user attitudes towards social robots
User attitudes can be measured using a psychological scale developed by Nomura et. al. (2014), named Negative Attitudes toward Robots Scale (NARS). It shows the mental states reflecting the opinions that humans have about robots. The original Japanese scale contains 14 items which are classified into three subscales: negative attitude toward interaction with the robot, negative attitude toward social influence of the robot and
negative attitude toward emotional
interaction with robots (Nomura et. al., 2014). Numerous studies used the NARS scale, for example, Wang (2010) found that Chinese subjects were more negative towards robots and had less trust in robots as compared to US subjects. This highlights the importance of a culturally sensitive design.
9
1.1.4 Anxiety towards social robots
Nomura et. Al defines robot anxiety as fear preventing individuals from interaction with robots having function of communication in daily life. The RAS scale was developed to measure anxiety evoked by robots. The RAS scale consists of three subscales: anxiety toward communication capability of robots, anxiety toward behavioural characteristics of robots, and lastly the anxiety toward discourse with robots. Each questionnaire item was assigned to a six-point sale score, ranging from 1: I do not feel anxiety at all to 6: I feel anxiety very strongly. In their research, Nomura et. Al hypothesized that robot anxiety was caused by factors: anxiety
toward technological products and
communication apprehension. However, both psychological factors were not proven to be related to robot anxiety. In this research, it is argued that people that experience high anxiety towards social robots
H2a: Anxiety towards social robots is negatively related to the perceived service quality of social robots.
H2b: Anxiety towards social robots is positively related to the perceived service quality of human employees.
1.1.5 Job fear
There is a large debate about the introduction of advanced technologies in the workplace such as machine learning, artificial intelligence etc. Such technologies are often argued to be detrimental for future
employment opportunities (Fred and
Osborne, 2013; Ford, 2015). The current public debate is largely centred on robotics. The media outlets have constantly covered the impact of robots on jobs, posting headlines such as ‘Coming to an office near you’ (The Economist, 2014) and ‘Technology Will Replace Many Doctors, Lawyers, and Other Professionals’ (Harvard Business Review, 2016), where robots are described as outperformers and replacers which are more efficient and effective than human beings (Floridi, 2017). A question that immediately
arises is, to what extent job fear and a negative attitude towards social robots is related and how.
Hence, the following hypothesis is formulated:
H3: Users who experience job fear have a more negative attitude towards social robots than people who do not.
1.1.6 Education
Dekker et. Al (2017) found that a fear of robots at work can partly be understood along the lines of self-interest. Managers, professionals and the higher educated, who often hold positions in the labour market that are unlikely to be affected negatively by the introduction of robotics, fear robots at work less than manual and white-collar workers and the less well educated. In terms of the cost benefit theory, higher educated participants tend to feel less threatened by the introduction of robots in the workplace and so feel less anxiety towards them. Moreover, they will focus more on the perceived benefits that social robots can bring to the workplace, therefore higher educated people will have a more positive attitude towards social robots.
H4: Education is positively related to attitude towards social robots and negatively related to anxiety towards social robots.
1.2 | Service quality in the consumer banking sector
1.2.1 Importance of service quality
There is a lot of debate on how to conceptualise the term service quality. Academicians have reached consensus that a complete definition is difficult to formulate. They all agree on the fact that service quality is a dynamic and multidimensional concept, using past and present service experiences (Parasuraman et al., 1985; Carman, 1990; Bolton and Drew, 1991). In general,
10
numerous studies acknowledge the following definition: service quality is an attitude of overall judgement about service superiority. However, the boundaries of this definition are still debatable.
Previous research shows that service quality has a positive effect on satisfaction. This relationship is especially of importance in high involvement industries such as the banking industry (Cronin and Taylor, 1992). Moreover, Brady et. al. (2005) indicate that service quality directly affects the satisfaction in five countries, including the Netherlands. Service quality has also been proven to be related to costs (Crosby, 1979), profitability (Buzzell and Gale, 1987; Rust and Zahorik, 1993), customer satisfaction (Bolton and Drew, 1991) and customer retention (Reichheld and Sasser, 1990). Therefore, many banks are interested on how to achieve excellent service quality. Social robots can overcome two important difficulties concerning the interpersonal service encounters: First, the personality and mood of both the employee and the customer can have a negative effect of the service encounter (Schneider and Bowen, 1985). By introducing technology in the service encounter, the bank can guarantee a more constant level of service. Consequently, the customer will have clear expectations from the encounter and have similar experiences. Second, many banks deal with perishability of shifts in demand for their services. Allocating staff to fluctuations in demand can be a time
consuming and costly process. The
introduction of social robots will enable banks to handle varying levels of demand without continually adjusting staffing levels (Curran, Meuter, and Surprenant 2003).
1.2.2 Job fear and perceived service quality
While many studies have covered the impact of technology on employment, less is known
about how consumers experience
technological change in the workplace. The social exchange theory originating from the social psychology field, provides a framework which outlines the interpersonal processes underlying the formation of relationships. This framework assumes that
human relationships are similar to the market. Humans tend to continue relationships if the perceived benefits outweigh the perceived risks and alternatives in comparison to the current relationship are perceived less positive. The perceived benefit theory has also been positively linked with the adoption of and attitudes towards new technologies (Lee, 2009; Wang et. Al. 2008), while perceived risk has negative effects on public acceptance of and trust towards technology. Given that interacting with a robot in a service delivery context involves both the formation of a relationship and the adoption of a new technology, this study argues that users’ perceptions of a robot can be predicted based on the basic premise of social exchange theory, such that users are more likely to form positive perceptions of the robot when they perceive their relationship with the robot to be beneficial rather than risky and costly. Moreover, Dekker et. Al (2017) suggest that habitants from countries where the economic conditions indicate better labour prospects (abundant job opportunities and GDP growth rate) and insulating institutional conditions (high trade union density) are less afraid for the implementation of robots in the work environment.
H5a: Users who experience job fear will be more negative about the perceived service quality of social robots
H5b: while people who do not experience job fear will be more positive about the perceived service quality of social robots. H6a: Users who experience job fear will be more positive about the perceived service quality of the human employees
H6b: while users who did not experience job fear will be more negative about the perceived service quality of human employees.
1.2.3 Measurement of perceived service quality
One of the most used measurement tools in the service quality literature is SERVQUAL (Parasuraman et al., 1988; Ladhari, 2009) . The assessment is conceptualized as a gap between the customer expectations of the
11
service quality from a class of service providers (in the context of this research the bank providers) and their perceptions of the performance of a particular service
provider (f.e. ING bank). This instrument is a multidimensional concept. The SERVQUAL model consists of 22 items which are compiled in the following five dimensions:
- Tangibles (appearance of equipment, physical facilities and personnel) - Reliability (the quality of being
trustworthy or of performing consistently well)
- Responsiveness (the promptness of the service, the willingness to help) - Safety (knowledgeable employees
who inspire trust and security) - Empathy (the ability to understand
and offer personalized care)
It is important to consider each dimension by its own weight. For example, Zhou (2004) found evidence that the reliability and safety dimensions have a more significant relationship to satisfaction. Tangibles, empathy and responsiveness were found to have no substantial effect on satisfaction. The SERVQUAL instrument has been used by numerous researchers who replicated or adjusted the model (Ladhari et. al., 2011; Kumar et. al, 2010; Petridou et. al., 2007). Therefore, criticism must be seen within this broader context of strong endorsement (Buttle, 1996). The following five dimensions have been adjusted to the context of this research:
- Tangibles (appearance of the robot / service employees)
- Reliability (performance and
trustworthiness of the robot / service employees)
- Responsiveness (the promptness and willingness to help of the robot / service employees)
- Safety (knowledge and feelings of security and trust of the robot / service employees)
- Empathy (feelings of understanding and offering personalized help of the robot / service employees)
1.2.4 Criticism on the SERVQUAL model
The SERVQUAL instrument has been subject to theoretical and operational criticism. On a theoretical level, critics questioned its conceptual suitability claiming that the five dimensions are not applicable to every service industry. For example, Lam (2002) used the SERVQUAL model for the banking sector in Macau. His results identified six rather than five dimensions, in which empathy was divided into two different dimensions. Tacit understanding of needs was labelled empathy 1 and convenient operating hours was labelled as empathy 2. Seth et. al. (2005), recognized that the measurement of the tool is dependent on the type of service provided, time, competition, number of encounters and needs.
Critics question the way service quality is
calculated (the difference between
expectations and perceptions). Peter et. al. (1993) report that difference scores do not have sufficient reliability and Babakus and Boller (1992) report that the perceptions scores are the dominant contributor to the perception-minus expectation scores, because of a generalized response tendency to rate expectations high. Moreover, SERVQUAL does not incorporate previous social science research, such as economics, statistics and psychology (Andersson, 1992). Parasuman et. al.’s work is highly inductive in that it moves from historically situated observation to common theory (Buttle, 1996). Moreover, Parasuman’s management theory does not consider the expenses of increasing service quality. The marginal revenue of the improved service quality does not always exceed the marginal costs of improvement (Aubrey and Zimbler, 1983). Lastly, the SERVQUAL model has no response to changing expectations. Customers learn from experiences and so naturally expectations rise over time. Nevertheless, expectations can also fall over time (Buttle, 1996). Currently, the Dutch population are facing doubts about the quality of healthcare due to rising expenditures as a consequence of an aging population and government cuts (Pollitt et. al., 2007)
12
On an operational level, critics questioned its item composition. In both the 1988 as the 1991 version, each factors is composed out of four or five items. This is often insufficient to truly measure the variance within each dimension. Moreover, the perceived service encounters are moments of truth (Buttle, 1996). Carman (1990) found evidence that customers evaluate service quality by reference to these multiple encounters. However, Parasuraman et al. (1988) stated that the service quality is not directly connected to particular incidents. The operationalization of the term expectations has also raised questions. Parasuraman et al (1988) defined expectations as “desires or wants of consumers, i.e. what they feel a service provider should offer rather than would offer”. Since the expectations component lacked discriminant validity, Parasuraman et. al. redefined expectations as “the service customer would expect from excellent service organizations, rather than normative expectations of service providers”. Iacobucci (1994) wants to exclude the term expectations and include the term standard in the service quality literature. According to Iacobucci (1994), this term is generic and multiple standards can work simultaneously. SERVQUAL does not incorporate customer perceptions based on absolute standards of service quality. The tool uses the perceptions of service providers within a specific class in which measures are relative rather than absolute. This way, customers will perceive a service more positively as their expectations are met or exceeded. This calculation does not take into account whether the pre-experience expectations were high or low and regardless whether the absolute level of the service was negative or positive (Iacobucci et al., 1994). To illustrate, a consumer may have low expectations based on a previous experience, if those expectations are met, there is no gap and the service quality is considered excellent.
1.2.5 Comparing SERVQUAL and SERVPERF
Cronin and Taylor (1992) showed that their
measure of service performance
(SERVPERF) functions better which resulted in more reliable estimations, greater discriminant validity, greater explained variance, and consequently less bias than the SERVQUAL. In SERVPERF perception is the only contributor to service quality rather than perception – expectation in the SERVQUAL. Both SERVQUAL and SERVPERF are the most commonly used scales of service quality measurement (Gilmore and McMullan, 2009), however SERVQUAL is still the most common used. (Duff and Hair, 2008; Ladhari, 2009). SERVQUAL is superior in its diagnostic power and is therefore the preferred choice of measurement tool to identify service quality shortfalls, however, another problem considering using the SERVQUAL is the lengthy data collection (Jain and Gupta, 2004). For this research, the SERFPERF is the preferred scale for this research due to its
psychometric soundness and greater
instrument parsimoniousness. The
SERVPERF scale is superior in measuring overall perceived service quality and is the best choice when undertaking service quality comparisons between different types of service agents.
1.3 Conclusion Literature review
There is still a wide gap in the literature concerning the relationship between job fear and perceived service quality of the both types of service agents. Moreover, it is still unknown how anxiety and attitude towards social robots affect these relationships. Therefore, the following conceptual framework was built in order to gain a wider understanding concerning the research constructs. It includes an overview of the research hypotheses.
13
1.4 Conceptual Framework
Contributions
From an academic perspective, this research paper aims to address the research gap within the topic of social robots in the banking sector. Within the writers’ knowledge, no research has related the concept of job fear, to perceived service quality of both human service employees and social robots in the consumer banking sector. Moreover, no study has included anxiety towards social robots
and general attitude towards social robots as constructs in their research design. From a managerial perspective, this research aims to examine whether social robots can replace
human service employees without
compromising the perceived service quality of its customers. Moreover, this research aims to give interesting insights in how anxiety and attitude towards social robots interact with the perceived service quality of the two types of service agents. Naturally, creating a digitalized service encounter can help to be time and cost efficient and less variable in service performance (Heskett et. al, 2015). If the results are positive, the employment of
Legend Direction of relation Indirect relation Hypothesis 𝑥 Mediator
▲ Figure 1 | Conceptual framework Service agent (human
employees, social robots)
H1a: Attitude towards social robots is positively related to the perceived service quality of social robots. H1b: Attitude towards social robots is negatively related to the perceived service quality of human employees. H2a: Anxiety towards social robots is negatively related to the perceived service quality of social robots. H2b: Anxiety towards social robots is positively related to the perceived service quality of human employees. H3: Users who experience job fear have a more negative attitude towards social robots than people who do not.
H4: Education is positively related to attitude towards social robots and negatively related to anxiety towards social robots. H5a: Users who experience job fear will be more negative about the perceived service quality of social robots
H5b: while people who do not experience job fear will be more positive about the perceived service quality of social robots. H6a: Users who experience job fear will be more positive about the perceived service quality of the human employees
H6b: while users who did not experience job fear will be more negative about the perceived service quality of human employees. Degree of Job fear
Perceived Service Quality Control variables (gender, age, educational backgrounds) Attitude towards social robots Anxiety towards social robots
14
social robots will increase strengthen customer’s value which in turn will enhance a stronger brand. The results of this study have large generalizability, since the results are applicable in many different service industries. Research concerning social robots in areas such as health care (Wasen, 2010), education (Behan and O’Keeffe, 2008; Kanda et. al., 2004) and retail (Shi et. al., 2016) noted positive results when experimenting with social robots. Especially in a competitive environment, in which service providers offer undifferentiated services, this can be a competitive advantage.
Methodology
.2.1 General design
We used a 2 (service robot vs. human assistant) x 2 (job fear vs. no job fear) between-subjects experimental design to assess the hypotheses. Each participant (N=216) was assigned randomly to one of the four conditions.
2.2 Web based experiment
One of the largest advantages of web experiments is the researcher is provided with an easy access to a much wider and
geographically diverse participant
population. In web experiments, people from the general population are now accessible to the researcher.
The advantage of eased access is twofold, as web experiments allow people to experience psychological research who would never had the chance to do so due to geographical, cultural and social barriers. Scientific psychology becomes more accessible. Experiments often include a rational of the research at hand and might give an indication of what the main purpose of the experiment is. Since this research used the deception technique, this effect is less applicable to this research. Another advantage of utilizing a web-based experiment is a higher validity. Findings of a laboratory experiments are far more difficult to generalize. The participants are far from their familiar surroundings in an
artificial place they can relate to. Naturally, this can lead to very unnatural responses. In online experiments, participants fill out the questionnaires in their natural surroundings such as work, university or home. Another advantage of web employed experiments is that the added comfort for the respondents. Bringing the experiment to the participants saves time and (travel) costs. A disadvantage linked to web-based experiments is the dependency of a technical interface (computer, smartphone, tablet etc.) to fill out the survey. However, many locations provide access to these interfaces (universities, libraries etc.) and so provide opportunities to participate. Moreover, participants have the freedom to choose whenever to wish to participate, increasing the external validity of web-based experiments as opposed to laboratory experiments.
In order to control for possible cheating, some precautionary measures were taken in order to ensure a minimal control of the researcher. For example, participants could fill out the survey once from their technical interface. Moreover, the fixed response setting was used in order to ensure that the participants filled out every question. Nevertheless, it can not be ensured that the independence of observations is guaranteed. For example, people could log in from multiple technical interfaces to fill out the questionnaire, but this extremely unlikely.
As mentioned before, the distribution of participants were randomized to the experimental conditions. Randomization is highly generalizable as the effect will be the same, dependent on consequent use and sample size. Comparatively high dropout rates is a disadvantage linked to web employed experiments due to its voluntary nature. Drop outs can influence the results of the experiment by dropping out of different conditions for different reasons. However, for the randomization of the videos and articles, an evenly present function was used, making sure that as elements are randomized, they are randomized evenly. This setting keeps careful watch on the completed counts to ensure every element is presented an equal number of times. Two other empirically verified factors that influence dropout are monetary
15
rewards and placement of assessment of personal information. Neither, were used as the scope of this research is analytical and the researcher preferred to create anonymous space, where answers to risky questions are less restrained.
2.3 Stimuli development
An web experiment was conducted using Qualtrics, an online survey tool. The survey was available for completion to English speaking respondents. All respondents were anonymous and could only take part in the survey once. A video of a humanoid robot was chosen, since particularly humanoid robots elicits human like interactions (Kanda et. Al, 2003). The banking industry was selected because of its history of technology usage in the service encounter, for example the use of self-service technologies. The banking industry is also appealing because bank services are widely used and most people use their services on a regular basis.
Participants in the robots-job fear condition were redirected to a video of a service robot helping a customer and an article by The Guardian which highlighted the short span in which robots will replace humans in their work activities.
Participants in the robots-no job fear were redirected to the same video only the article differed from the first condition. This article highlighted the long time it takes to successfully implement service robots in the workplace. Participants in the human assistant-no job fear condition were redirected to a video of a human assistant helping a banking consumer and to the same article mentioned in the previous article. Finally, participants in the human assistants-job fear condition were shown the same video in the previous condition and the article highlighted the long-time span for robots replacing humans in the workplace. After the
respondents were exposed to the
manipulation, they had to agree to the
conditions of the experiment and confirm that met the criteria of selection to participate. Next the research questions were presented, followed up by the manipulation check. Lastly, the demographic questions were presented. It was important to identify a relevant set of questions to measure respective variables and apply statistical tests to measure the impact of each variable on another. After extensive literature review, it was identified that gender, age, education, anxiety and general attitude towards social robots can influence the perceived service quality of the two service agents.
2.4 Deception technique
To conduct this study, a deception technique was chosen in order to answer the research questions. As mentioned in the previous section, the participants were deceived by either reading an extremely positive article (or extremely negative article) about robots in which a long-time span (a short time span) would be needed in order for robots to overtake the work activities of humans. Deception is a widely used technique in the social sciences. Deception is often justified with two arguments. First, deception is a way to create situations of interest which would not easily arise naturally. The second argument for deception is that certain relevant aspects of behaviour can only be studied if the participants are caught off guard (Weiss, 2001; Hertwig & Andreas Ortmann, 2008)Besides, the participants are debriefed after the experiment is over, thereby removing most of the effects of deception. Kelman (1967) has suggested that not only should the subject not leave the laboratory with greater anxiety or lower self-esteem but that he should in some positive way be enriched by the experience, that is, he should come away from it with the feeling that he has learned something, understood something, or grown in some way; (p. 8). Another important element of the debriefing procedure is to develop an understanding of the nature of the experiment and the role which the participant plays in the process. This helps to create a willingness among participants to cooperate in not discussing
16
the experiment with other persons. If the subjects do not refrain from discussing the experiment, it is likely that knowledge about the true purpose will spread to future
subjects, contaminating their results and preventing the data from the experiment from being useful or meaningful. In the debriefing, the basic principles underlying the experiment are explained so that the participants have an honest view of the research.
There are two practical concerns regarding the use of deception: first, deception
increases the suspicion of future participants and second, deception reduces the trust in research. However, research has shown that the effects of participant suspiciousness are neglible. Moreover, the deceived participants did not become resentful after the
experiment ended. It did not negatively affect their attitudes towards social science and the participants did not have serious objections to use of deception in research (Kimmel, 1996).
Another technique that is widely used to recreate situations of interest in the social sciences is role-playing. In role-playing, the experimenters ask participants to imagine that they are in a particular situation and to respond as if that situation would be a real-life scenario. In terms of effectiveness, many researchers note different results from deception and roleplaying (Martin; 2007, Orne 1970). One of the biggest concerns related to roleplaying is that roleplaying reflects the demand characteristics of the experiment, rather than predicting behaviour in a real- world scenario (Martin; 2007), therefore the deception technique was chosen.
2.5 Research Setting
After extensive literature review, the questionnaire based on previous research was devised to collect data based on convenience and efficiency of the sampling method. In total, there were two independent variables namely Job fear and Type of Service agent, two proposed mediators
including anxiety towards social robots and general attitude towards social robots, and finally the independent variable perceived service quality was measured for both types of service agents. The questionnaire
measured the agreement continuum of the respondents on a multi item instrument using the Likert scale. Likert scales in combination with fixed choice response were designed to measure attitudes and opinions (Bowling, 1997). The questionnaire ended with several demographic questions including age, gender and education.
2.6 Sampling procedure
Convenience sampling refers to
non-probability sample in which participants are chosen based on their accessibility and availability to the research (Bryman, 2012, p.201). Participants were recruited in the University of Amsterdam and on various Facebook platforms. Invitations to
participate was sent out to the student union and various student organizations within the university. Also, university students were approached at the university’s common areas and were asked if they agreed to participate. After they agreed, a link was send which redirected them to the questionnaire. Since the aim of this research is to see how bank customers react to various types of service agents and how this influences their perception of service quality of these two types of agents, the sample was chosen from participants with a bank account. Moreover, only participants who had a job were selected. The rationale behind this decision was that people with a job would be more afraid to lose a job as apposed to people without. Moreover, we chose to exclude people aged below 18, since they might be too young to be financially independent and so not able to experience actual job fear.
17
Results and Discussion
3.1 ParticipantsIn total 223 people participated in this study, of which 7 people were deleted from this study due to missing data or preferred their data removed from this research after the informed consent was shown resulting in a total of 216 responses. Ages of the
participants ranged between 18 and 65 (M =
25.85, S = 8.57). Participants had various education backgrounds as shown in the pie chart below:
With regard to the manipulation, 106 participants read the negative people, while 110 people read the positive article about social robots. Moreover, 105 people saw the human employee video and 111 people the social robot video. The unequal number in display is due to the removal of cases with missing data and/or disagreement to participate after the informed consent was shown. Statistical tests were carried out using SPSS software to analyse the results of the online experiment. SPPS is considered to be the most widely used software for
analysing quantitative data (Bryman, 2012, p.354). All hypotheses are presented separately and either accepted or rejected based on the results of statistical tests.
3.2 Instruments
3.2.1 Dependent variable
The dependent variable examined in this study is the perceived service quality. This variable had two levels since two types of service agents were investigated. The variable was operationalized by the
SERVPERF. The SERVPERF has originally 19 items which are compiled in five
dimensions: tangibles, reliability,
responsiveness, safety and empathy. The items were measured on a seven-point Likert scale from 1 (strongly disagree) to 7
(strongly agree). An example of an item includes ‘A social robot informs the clients exactly when the services are going to be executed’.
Education
Did not complete High School Lower General Secondary Education (MAVO) Higher General Secondary Education (HAVO) Pre-University Education (VWO)
Bachelor Degree Master Degree
18 Perceived service quality of human
employees
3 items were separately removed in order to improve the Cronbach Alpha: A human service employee has favourable service hours for its clients (0.042), Human service employees are always available to help clients (0.014) and Human service
employees can always quickly respond to the requests of the clients (0.095).
Perceived service quality of social robots
Similarly, 3 items were removed, A social robot has favourable service hours for its clients (0.071), A social robot is always
available to help clients (0.213), You receive a fast service from a social robot (0.279). When calculating the Cronbach’s alpha with the 16 items of the applied instrument, a value of 0.849 for the PSQ of human employees and a value 0.871 for the PSQ of social robots were obtained. This means that the instrument is reliable, as the items collect
information consistently on the perceived service quality of social robots. Cronbach alpha was also calculated for each of the five dimensions: reliability, responsiveness, safety, empathy and tangible elements which can be found in the table 1:
Dimensions of PSQ Number of items PSQ human employees PSQ social robots
Reliability 5 0.840 0.881
Responsiveness 2 0.419 0.430
Safety 4 0.802 0.727
Empathy 3 0.750 0.721
Tangible elements 2 0.701 0.684
▲ Table 1: Reliability analysis
According to Hernandez et. Al. (2012, p.302) alpha values higher than 0.5 relate to an instrument with a medium reliability and values higher than 0.75 are pertinent to an acceptable reliability. The decrease of the Cronbach alpha is due to each dimension having different numbers of items. Cortina (1993) indicated that the alpha coefficient depends on the number of items: the more items in the instrument, the greater its reliability. Of the in total three deleted items, two were categorized under responsiveness and one under empathy.
3.2.2 Mediator Variables
In total, two mediator variables were analysed to see their effect on the perceived service quality on different types of service agents, in this study human employees and
social robots. The variable general attitude towards social robots was the first mediator in the questionnaire. To measure this variable the NARS scale was used. Two items were deleted:
- The word robot means nothing to me - I feel that in the future robots will play a small role in society.
Both items had a corrected item-total correlation far below 0.3.
The second mediator that was researched was anxiety towards social robots. Anxiety was measured with the following three dimensions: behavioural
characteristics, communication capabilities and discourse. The Alpha Cronbach was 0.571 for the 11 items which indicating a moderately reliability. Specifically, the items concerning the communication capabilities
19
had a corrected item-total correlation far below 0.3. In order to improve the internal consistency of Anxiety towards social robots, it was chosen to remove the dimension communication capabilities. These included:
- Whether the robot talks about relevant things in the middle of the conversation
- Whether the robot might be flexible in following the direction of the conversation
- Whether the robot might understand difficult topics
After running the reliability analysis
excluding these items, Cronbach Alpha went from 0.571 to 0.837, indicating strong reliability.
Appendix A and B contain the items measured for the mediator variables.
Number of items Cronbach Alpha
PSQ for social robots 16 0.871
PSQ for human employees 16 0.849
Negative Attitude towards social robots
12 0.834
Anxiety towards social robots 8 0.837 ▲ Table 2: Cronbach Alpha
3.2.3 Control Variables
Research is mixed on which factors play a significant role on how robots are perceived and responded to by humans. Some
researchers suggest that individual
differenced in terms of personality or gender can play a role. For example, Nomura et. Al found some gender difference on
correlations between the RAS (anxiety towards social robots), STAI (general anxiety) and PRCA–24 (personal report of communication apprehension). This implies the possibility of gender difference on mental relations between personal traits, evoked emotions, and behaviours toward robots in human-robot communication. However, others argue that pre-existing attitudes towards social robots are difficult to only link to demographics or previous experience with technology. This research is especially interested in the traditional demographic variables such as age, gender and education. These variables were measured in order to control for their
potential influence on the dependent variable.
3.3 Manipulation check
Two questions were included in the survey in order to check if the articles had some effect on the dependent variable:
1. Do you expect to lose your job due to robotization?
2. How likely do you think it is (in general) that human jobs will be taken over by robots in the near future?
An independent sample t-test was run both questions, to see if the answers differed between the two grosups (positive/negative article). For the first question, there was no significant difference between the two groups (0.782). However, for the second question, a notable difference was found (0.027) between the groups. This question shows that the manipulation with the article has worked and that the manipulation was successful. A possible explanation for the
20
different outcomes between the two
questions could be that the second question is focusing on a general trend rather than estimating a personal situation.
3.4 Statistical Analysis of Data and Results This section provides the details of the analysis of the data collected through the online questionnaire. The design of the survey questionnaire was described in the literature review and research methodology. First of all, the data was checked if there were any errors in the data. No errors were found. The missing data was solved by only including and analysing the cases that had no missing data in any variable. Five cases were excluded as a result of the revelation of the use of the deception technique. The positive and negative articles about social robots were recoded in 0 (positive article) and 1 (negative article). Similar coding was applied for the videos, in which 0 meant Human Employee and 1 Social Robot. The recoding of counter indicative items applied to items of perceived service quality of humans and social robots and attitude towards robots. Additionally, in order to enable use of gender as a Control variable, it
was needed to adapt the scale from 1 (Male), 2 (Female) and 3 (I don’t identify myself with a certain gender) to 0 (Other) and 1 (Female).
Reliability enables to examine the consistency of measurements. Reliability checks have been run for Perceived service quality, General attitude towards social robots and anxiety towards social robots as described in the research methodology section.
In general, participants tend to prefer human employees over social robots as can be concluded from the histogram below
21
3.5 Distribution of research variables
All research variables were approximately normally distributed. The distributions are illustrated in the plots below:
3.6 Hypothesis testing
The main purpose of this research was to examine the interaction effect of the type of service agent and the degree of job fear on the relationship of perceived service quality, and the way this relationship is mediated by the general attitude towards social robots and anxiety towards social robots. Moreover, control variables were measured and
analysed to see if they had any effect on the research variables. In total six hypotheses were tested. At the end of this section, an overview will be provided with the outcomes of the hypotheses. New variables as a
function of existing variables were created for hypothesis testing. The mean was calculated for all items that was used to describe a variable. Means and standard deviations of all variables are exhibited.
22 * Correlation is significant at the 0.05 level (two-tailed)
** Correlation is significant at the 0.01 level (two-tailed)
In the first correlation matrix, five significant relationships were found. The first positive linear relationship was between the
perceived service quality of social robots and the general attitude towards social robots. Second, there is a significant negative relationship between perceived service quality of social robots and anxiety towards social robots. Third, there is a positive significant between education and general attitude towards social robots. Fourth, there is a negative relationship between video and anxiety towards social robots. Lastly, there is a negative relationship between age and female. This can be explained due to the sample population, in which the female participants are younger than the male participants.
Now, the findings will be compared to the initial hypotheses. A positive correlation was found between attitude towards social robots and the perceived service quality of social robots, therefore supporting H1a. However, no negative relationship was found between the attitude towards social robots and the perceived service quality of human service employees, so H1b is not supported. A negative relationship was found between anxiety towards social robots and the perceived service quality of social robots, supporting H2a. Again, no support was found for a positive relationship between anxiety towards social robots and the
perceived service quality of human employees, thereby rejecting H2b. Moreover, job fear (the article) is not correlated to the attitude towards social robots, rejecting H3. Nevertheless, education is positively correlated to the general attitude towards social robots, partly supporting H4. There was however no correlation observed between education and anxiety towards social robots. No relationship was found between job fear (the article) and the perceived service quality of both types of service agents, so no support was found for H5 and H6 initially.
Mean, Standard Deviations and correlations Mean PSQ for both service agents
Variables M SD 1 2 3 4 5 6 7 8 9 1. PSQ Social Robots 4,3724 0,779 - (,871)
2. PSQ Human Employees 4,6406 0,639 0,059 - (,849) 3. General attitude towards
social robots
2,8827 0,627 0,469** 0,018 - (,834) 4. Anxiety towards social robots 3,3889 0,87
-0,421** -0,081 -0,587 - (,837) 5. Article 0,4907 0,501 -0,031 0,056 -0,028 0,031 - 6. Video 0,5139 0,501 -0,037 -0,031 0,064 -0,143* -0,064 - 7. Female 0,6991 0,459 -0,074 0,085 -0,119 0,038 -0,083 -0,113 - 8. Age 25,86 8,589 0,037 -0,083 0,092 -0,079 -0,049 -0,049 -0,186** - 9. Education 4,72 1,052 0,049 -0,026 0,169* -0,104 -0,027 -0,027 -0,032 0,041 -
23
Instead of looking at overall perceived service quality, two more correlation analyses were run in order to see if there are any notable differences between the
dimensions and their effect on the mediator and/or control variables. Below, two correlation matrixes are shown, one for the perceived service quality for human
employees and one for the perceived service quality for social robots.
Means, Standard Deviations and Correlations for the five dimensions of PSQ for human employees
* Correlation is significant at the 0.05 level (two-tailed) ** Correlation is significant at the 0.01 level (two-tailed)
In the second correlation matrix, it can be observed that all five dimensions are correlated to each other except for responsiveness and empathy. This can be due to the low Cronbach Alpha of
responsiveness. Two new relationships were found. First, a significant positive
relationship between human tangibles and anxiety towards social robots was found. Second, there is a negative significant relationship between anxiety towards social robots and attitude towards social robots.
Variables M SD 1 2 3 4 5 6 7 8 9 10 11 12 1. PSQtanghuman 4,924 1,015 - (,701) 2. PSQreliabhuman 4,450 0,911 0,299** - (,840) 3. PSQsafetyhuman 4,899 0,884 0,209** 0,538** - (,802) 4. PSQempathhuman 4,767 0,974 0,154* 0,160* 0,314** - (,750) 5. PSQresponshuman 4,127 0,954 0,177** 0,578** 0,348** 0,047 - (,419) 6. General attitude
towards social robots
2,883 0,627 -0,003 -0,01 -0,008 -0,032 -0,007 - (,834) 7. Anxiety towards social robots 3,389 0,87 0,134* 0,051 0,08 0,062 -0,076 -0,587** - (,837) 8. Article 0,4907 0,501 -0,074 0,036 -0,112 -0,118 0,083 -0,028 0,031 - 9. Video 0,5139 0,501 0,06 0,004 0,056 0,055 -0,096 0,064 -0,143* -0,064 - 10. Female 0,699 0,459 -0,085 -0,043 -0,091 -0,03 -0,051 -0,119 0,038 -0,083 -0,113 - 11. Age 25,86 8,589 0,059 0,044 0,107 -0,009 0,095 0,092 -0,079 -0,049 -0,049 -0,186** - 12. Education 4,72 1,052 0,073 0,089 -0,006 -0,102 0,017 0,169* -0,104 -0,027 -0,121 -0,032 0,041 -
24 Means, Standard Deviations and Correlations for the five dimensions of PSQ for social robots
* Correlation is significant at the 0.05 level (two-tailed) ** Correlation is significant at the 0.01 level (two-tailed)
For the perceived service quality of robots, almost all five dimensions were correlated, except for tangibles and empathy. This difference from the second correlation matrix for perceived service quality of human employees where responsiveness and empathy did not correlate. Attitude towards social robots and attitude towards social robots are both correlated to all five dimensions of perceived service quality of social robots. No new relationships were found. Education seems to be the only control variable that is related to the other research variables.
Focusing on the five dimensions of perceived service quality, human service employees score higher on the tangibility, safety and empathy dimensions, while social robots are perceived to be more reliable and more responsive. The safety dimension was in particularly centred around trust and a sense of safety in the transactions with the service agent, while responsiveness focused on providing services in a timely fashion and
keeping promises. The appendix C includes a table with the items per dimension. The explanations concerning the preference of service agent were manually coded. The majority of participants expressed their preference for human service employees which is consistent with the finding that the perceived service quality of human service employees is higher than for social robots. Most participants that favoured human service employees named empathy, social interaction, adaptability (to complex situations) and familiarity as their main reasons. An example includes:
“I feel that a Human Service Employee could easily and better understand a person's emotions such as frustration in the way he/she speaks or acts and adapt the kind
of responses to deal with the problem which in my opinion would be complicated for a
social robot to get.”
Variables M SD 1 2 3 4 5 6 7 8 9 10 11 12 1. PSQtangrobot 4,155 1,116 - (,684) 2. PSQreliabrobot 4,757 1,094 0,402 ** - (,881) 3. PSQsafetyrobot 4,360 0,954 0,321 ** 0,646 ** - (,727) 4. PSQempathrobot 3,617 1,146 0,096 0,193 ** 0,302 ** - (,721) 5. PSQresponsrobot 4,787 0,989 0,218 ** 0,728 ** 0,505 ** 0,205 ** - (,430) 6. General attitude
towards social robots
2,883 0,627 -0,273 ** -0,351 ** -0,483 ** -0,294 ** -0,203 ** - (,834) 7. Anxiety towards social robots 3,389 0,87 0,250 ** 0,369 ** 0,399 ** 0,164 * 0,267 ** -0,587 ** - (,837) 8. Article 0,491 0,501 0,037 -0,017 0,057 0,087 -0,066 -0,028 0,031 - 9. Video 0,514 0,501 -0.044 0,027 0,017 0,091 0,017 0,064 -0,143 * -0,064 - 10. Female 0,699 0,46 0,094 0,01 0,102 0,067 0,019 -0,119 0,038 -0,083 -0,113 - 11. Age 25,86 8,589 -0,22 -0,008 -0,069 0,005 -0,059 0,092 -0,079 -0,049 -0,049 -0,186 ** - 12. Education 4,72 1,052 0,01 -0,065 -0,035 -0,035 -0,009 0,169 * -0,104 -0,027 -0,121 -0,032 0,041 -
25
Participants that preferred social robots named the efficiency gains, reliability and increased capabilities as the primary reasons for favouring robots over human employees. To illustrate, one of the participants
answered:
“Social robots would probably make less mistakes than humans because they don’t
have the feeling of time pressure.” However, a few participants noted that their preference for their service agent is
dependent on the activity that needs to be
performed or were not sure about their preference.
Next, a 2 x 2 x 2 Mixed ANOVA was run, with agents as within levels and the two articles and two videos as the four
conditions. The dependent variables are all approx. normally distributed, the
independent variables are categorical, and the dependent variables are measured using an interval. Below, the Mauchly's Test of Sphericitywill be shown, to demonstrate that the final criteria is also met.
Mauchly's Test of Sphericitya
Measure: PSQ
Within Subjects Effect Mauchly's W
Approx.
Chi-Square df Sig.
Epsilonb
Greenhouse-Geisser Huynh-Feldt Lower-bound
Agents 1.000 .000 0 . 1.000 1.000 1.000
Tests the null hypothesis that the error covariance matrix of the orthonormalized transformed dependent variables is proportional to an identity matrix.
a. Design: Intercept + Article + Video + Article * Video Within Subjects Design: Agents
b. May be used to adjust the degrees of freedom for the averaged tests of significance. Corrected tests are displayed in the Tests of Within-Subjects Effects table.
There is no p-value for the test of sphericity, because there are only 2 levels of repeated measures. As such, there is only one set of difference scores and nothing to compare those difference scores against to indicate a violation of sphericity. Therefore, the assumption of Mauchly's sphericity will be met under this situation.