A study on how the framing of outsourcing and botsourcing waste separation affects sustainable behaviour

(1)

A study on how the framing of outsourcing and

botsourcing waste separation affects sustainable

behaviour

Julienne Hollanders Student number: 12594393

23 January 2020 Master thesis

MSc Business Administration - Marketing track University of Amsterdam

EBEC number: 20191204111232 Supervisor: Andrea Weihrauch

(2)

Statement of Originality

This document is written by Julienne Hollanders who declares to take full

responsibility for the contents of this document.

I declare that the text and the 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

An increasing number of occupational fields are currently dealing with robotic developments and botsourcing: outsourcing human jobs to robots. The recycling sector, more specifically the waste separation task performed at recycling facilities, is being disrupted by these phenomena as well. Extensive research has identified many factors that influence recycling behaviour and general sustainable behaviour, but no research has yet been conducted on how the perception of human and robotic waste workers affects one’s sustainable behaviour. Since message framing can influence how information is perceived and processed, a ‘helping’ and ‘serving’ frame, indicating different levels of personal responsibility, were included. Hence, this research studied the effect of exposure to waste workers on self-reported intention to perform and subsequent actual performance of sustainable behaviour, which was measured through willingness to participate in a lottery as well as the monetary amount donated to charity. A 2 (human vs. robot) x 2 (help vs. serve) between-subjects online experiment was conducted with 198 respondents. Results show no direct relationship between exposure and intended and performed sustainable behaviour; including the message frame as moderator did not significantly affect these behaviours either. General environmental concern, which was included as a control variable, however, did show to be a significant predictor. Theoretically, this study adds to the existing body of literature on which factors do and do not have predictive power of sustainable behaviour. The insights gained from this study lead to advise for governments, NGOs, institutes or companies in general that seek to increase individual’s performance of sustainable behaviour.

Keywords: Botsourcing, Recycling, Exposure, Message Framing, Sustainable behaviour

(5)

1. Introduction

McKinsey Global Institute (2017) reports expectations that by 2030, up to 800 million global jobs will be replaced by robotic automation, affecting up to one-fifth of the global work force (BBC News, 2017). According to this report, most affected activities are those most susceptible to automation, including physical ones in predictable environments like operating machinery. Research shows that McKinsey’s predictions are in line with public expectations – people across different occupational groups more strongly agree than disagree with the idea that robots will steal people’s jobs in the future, which results in them perceiving robots as a threat to human jobs (Granulo, Fuchs & Puntoni, 2019). People prefer that human workers are employed rather than robots – at least, for jobs not considered dirty, dull or dangerous. This reasoning is related to research on pro-social behaviour, which documents that other individuals’ wellbeing is of importance to humans (Batson & Powell, 2003).

One of the sectors in which the phenomenon of robotic development is rising at a rapid pace, is the recycling sector (Environment Journal, 2018). Recycling is one of the main ways in which people can contribute to the development of a more sustainable world - an important aim in the current era of global concern. A distinction between two ways of recycling can be made: it can either take place at the business and household level, where a number of categories are identified to sort waste into, or at specialised sorting facilities (Czajkowski, Kadziela & Hanley, 2014). In most facilities, human labour workers perform the task of removing items that cannot be recycled from the conveyor by hand (Suez in United Kingdom, n.d.). This human labour in recycling facilities is considered a ‘low-paid, dirty, monotonous and physically demanding job, some which are physically dangerous’ (Gregson, Crang, Botticello, Calestani & Krzywoszynska, 2016, p. 551). The nature of this work leads to an increased number of start-ups developing recycling robots. Recycling facilities can purchase these robots to join human labour workers in the waste separation task, picking up recyclables that humans overlook. When companies can manage to develop robots that outperform human workers – which would be the case when the robotic worker can pick two or three times as many objects – it might even be economically justified to use them as replacement for human workers (Scientific American, 2019).

Extensive research has identified existing factors that predict consumers’ willingness to perform and actual performance of recycling behaviour. Barr, Gilg and Nord (2001)

(6)

place factors into the broader categories of environmental values, situational variables and psychological variables; Hage, Söderholm and Berglund (2009) and Schwartz (1970) elaborate on the role of social and moral norms.

Even though many factors that affect recycling behaviour have been researched extensively, less research has focused on the direct consequences of individuals not performing recycling behaviour. Since the waste separation task performed at recycling facilities is a result of households and businesses not separating waste, it can be seen as a direct consequence of individuals not taking personal responsibility for performance of this task. How an individual’s willingness to perform sustainable behaviour is affected by exposure to performance of the waste separating task at recycling facilities has not been researched yet.

The aim of this master thesis is to identify how exposure to performance of the waste separation task at recycling facilities affects individuals’ intention to perform sustainable behaviour and actual performance of this behaviour. A distinction will be made between human labour workers and robotic workers. Since individuals might perceive human and robotic workers differently, being exposed to either a human labour worker or a robotic worker could elicit different reactions.

To investigate whether this perception could potentially be affected by how information is presented to individuals, the research involves the additional component of message framing. According to Graber (2004), the way in which a message is framed can influence to what extent it is attended to, the knowledge an audience gains from it and how positively or negatively the message is evaluated. Either profiling the facility work as ‘helping’ or ‘serving’, might affect the extent to which people are willing to perform sustainable behaviour.

Framing the recycling facilities’ work as ‘help’ could indicate that separating waste becomes a shared effort between individuals and waste workers. This would indicate that individuals, to a certain extent, are personally responsible for separating waste. Factors like the overall perception of human waste workers (e.g. Douglas, 1966; Tannock, 2015) and robots (Gnambs & Appel, 2019), kinship (Maner et al, 2002) and perceived similarities (Noval, Molinsky & Stahl, 2018) could play a potential role.

Using the ‘serve’ frame might cause that the individual experiences a certain sense of superiority (Hughes, 1962) and social distance (Mutlu, Osman, Forlizzi, Hodgins & Kiesler, 2016) when comparing itself to the human waste workers, which might cause

(7)

individuals to feel more comfortable with outsourcing the task. The main objective of this paper is to answer the following research question:

What is the effect of seeing a robot (vs. a human labour worker) engage in recycling tasks on an individual’s self-reported intention to perform and actual performance of sustainable behaviour and how is this moderated by framing the

recycling activity as service (vs. help) to humans?

In the following chapter, the body of literature underlying the phenomenon that is the focus of this master’s thesis will be discussed extensively. Since the hypotheses of this study are formulated based on combined findings from different authors and research backgrounds, these will be formulated at the end of the literature review, accompanied by a conceptual framework that illustrates the variables’ potential relationships. The chapter covering the literature review is followed up by a data and method section, in which the created stimuli are discussed, as well as what data are used, how they are collected and how they are operationalized and measured. The gathered data will elaborately be analysed and presented in the results section. This thesis will end with a discussion chapter, which includes a discussion of the main results, actionable advice for managers and theoretical implications, a description of the limitations of the current research and suggestions for future research.

(8)

2. Literature review

2.1 Botsourcing

An increasing amount of jobs is being automated and outsourced to robots (Chui, Manyika & Miremadi, 2016). This concept is called botsourcing, which is defined as ‘the use of robots or robotic technology to replace human workers’ (Waytz & Norton, 2014, p. 434). Botsourcing is often confoundedly used with the umbrella term Robotic Process Automation (RPA), which is defined as ‘an emerging form of business process automation technology based on the notion of software robots or artificial intelligence (AI) workers’ (Madakam, Holmukhe & Jaiswal, 2019, p.2). These AI workers, who used to be perceived as a tool, are increasingly being perceived as teammates and autonomous agents (Dignum, 2018), improving productivity, safety and health on a daily basis (Stone et al., 2016).

From service to health to warfare: the concept of botsourcing is rapidly emerging in a wide range of domains (Friedman, 2011; Gates, 2007 in Waytz & Norton, 2014). Burgess (2015) argues that robotic process automation is mainly applied to tasks that are rules-based, repetitive and frequent. Chui et al. (2016) also identified a number of factors on which the potential of an occupation to be automated depends. The most important one is the technical feasibility of automating the activities performed in a specific occupation. Technical feasibility refers to ‘the percentage of time spent on activities that can be automated by adapting currently demonstrated technologies’ (p.3). With 78%, this potential is found to be highest in occupations that belong to the category of predictable physical work, where workers perform pre-defined actions in settings that are well known.

Waytz and Norton (2014) looked at how comfortable people are with botsourcing jobs that are either emotion- or cognition-related. People’s belief that only humans are capable of jobs involving emotion leads to worker’s preference for botsourcing cognition-related tasks rather than emotion-related tasks. When a job typically requires emotion, people express more comfort with botsourcing if a robot appears to convey more emotion. Since the repetitive work performed in a recycling facility is more likely to require basic knowledge rather than emotion, botsourcing would be likely to be generally excepted for this specific domain and skill set.

Granulo et al. (2019) have also investigated human attitudes towards the increasing role of botsourcing. Their findings show an interesting shift in perspective taking: people prefer other humans (vs. robots) to replace human workers when it concerns someone else’s job since they would rather see human workers employed than robots, but this effect is

(9)

reversed when people were asked to consider the prospect of losing their own job. This phenomenon can be explained by the fact that different psychological reactions are triggered; replacement by robots forms a less immediate threat to people’s self-worth. Since participants in this research are not directly threatened with losing their own job, it would be expected that there is a preference for humans performing the waste separation job over robots. As argued by Granulo et al. (2019), following this line of reasoning is consistent with existing research on pro-social behavior and caring about the well being of other individuals. Preferring human workers, however, would only be the case for jobs not considered dirty, dull or dangerous. This would indicate that for recycling tasks specifically, robots would be preferred. The concept of recycling will be elaborated on in the next paragraphs.

2.2 Recycling

Before elaborating on recycling and the sector in general, it is important to make a clear distinction between the concepts of environmentally responsible behaviour (ERB), recycling and separating waste. ERB, which is often referred to as sustainable or pro-environmental behaviour, suggests that individuals want to perform behaviour that enhances environmental sustainability (Fraj & Martinez, 2007). Among other methods like e.g. carpooling, using energy-efficient appliances and driving hybrid vehicles (Hergat Huffman, van der Werff, Henning & Watrous-Rodriquez, 2014), recycling is one of the most frequently measured ERBs since it involves relatively simple, economically feasible behaviour (Iyer & Kashyap, 2007) and has shown to be a global concern since the topic has been examined internationally. On the broader spectrum, recycling can therefore be positioned as one of the activities that contribute to enhancing the broader concept of environmentally responsible behaviour, which will be referred to as sustainable behaviour in the remainder of this study. Some confusion might exist with regard to the exact meaning of the term recycling. Even though this term is used very commonly, it often refers to the activity of sorting/separating waste. The actual meaning of recycling is ‘recovering materials from discarded goods, which are then recycled through further rounds of manufacturing’ (Gregson et al., 2016, p. 544). This process happens at specialized manufacturing companies. Before this process can be started, however, sorting waste into recyclable categories has to happen at either the individual level or at a materials recovery facility (MRF). MRFs receive, separate and prepare unsorted, recyclable materials before shipping them to specialized recycling facilities (The Balance

(10)

Small Business, 2019); these are the facilities that will be referred to in this study. In summary, the separation of waste is an indispensable activity for enabling recycling to happen. Therefore, the concepts of recycling and sorting/separation of waste will be used confoundedly in the remainder of this study.

2.2.1 The recycling sector

Recycling is argued to be one of the main ways to actively tackle global environmental problems (Crociata, Agovino & Sacco, 2015). It both reduces waste and turns it into valuable resources, and since various countries around the world cope with amounts of household waste increasing at a surprising speed (Chen & Tung, 2010), gaining insight into how people can best be motivated to sort waste is of great importance.

Bruvoll, Halvorsen and Nyborg (2002) distinguish between six categories waste can be sorted into: (1) paper and cardboard, (2) glass excluding return deposit, (3) drinking cartons, (4) food waste/compost, (5) metals excluding return deposit, and (6) plastic excluding return deposit. The activity of sorting waste can be organised and handled in two different ways. The first is called dual stream recycling and refers to households, municipalities and businesses taking responsibility for the separation of recyclable materials from non-recyclables. In the second option, single stream recycling, all types of waste are collected without being pre-separated; the task of sorting waste is handled at centralized materials recovery facilities. Both approaches will be explained briefly.

Dual stream recycling

This recycling approach describes the process of households and businesses separating fiber items (paper and cardboard) from containers (glass, plastic and can), which allows for a maximum recovery rate. ‘Household recycling has increasingly been seen as a means for reducing the amount of waste for landfill and to increase the reuse of materials in general’ (Bruvoll et al. 2002, p. 337), eventually leading to the achievement of sustainable waste management (Barr et al., 2001). This process leads to lower levels of contamination and is not only considered to be the most cost effective way, but also automatically leads to higher quality and more valuable recovered material (VangelInc., 2019). Hence, a situation in which all households and business would take personal responsibility for performing the waste separation task would be ideal. This, however, is not the case. Different elements that affect the decision to (not) separate waste will be discussed later on.

(11)

Single stream recycling

The second recycling option is called single stream recycling; a recycling method in which plastic, paper, cardboard, glass and metal are placed in the same bin and are not separated before the moment of pickup and transport to facilities. Besides households, many municipalities and businesses also adopt this approach (VangelInc., 2019). Benefits of this collecting method are the fact that it is easier and faster and it cuts costs of collection, making the collection process more convenient. Collecting commingled materials, however, entails the risk of materials becoming unrecyclable due to certain materials (e.g. glass or metals) accidentally damaging others. When choosing this approach, the task of separating waste is performed in facilities. Traditionally, human waste workers remove items that cannot be recycled at the facility, e.g. plastic bags, electronics, toys and hazardous materials, from the conveyor belt by hand.

2.2.2 Robotic developments in recycling

Before elaborating on how newly developed robotic applications execute the task of correctly identifying and separating waste, it is important to touch upon the difference between artificial intelligence and machine learning. ‘AI is considered a broader category of computer systems developed to perform tasks normally requiring human intelligence, including visual perception and follow-up decision-making’ (Recycling Today, 2019). Within the broader category of artificial intelligence technology, machine learning has been defined as ‘an application of AI that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves’ (Expert System, n.d.). In the context of correctly identifying and separating materials in a recycling facility, machine learning specifically plays a leading role.

Since robotic developments have only recently started revolutionizing the recycling sector, a number of companies can be identified as pioneers in this segment. Machinex’ SamurAI, AMP’s cortex robotics system and MIT CSAIL’s Rocycle robot approach the waste separation task in a similar way: their developed division systems are able to learn from experience and can sort, pick and place different materials for precise product recovery (Machinex, 2019), despite being smashed, folded, torn or dirty (AMP Robotics, n.d.). The gripper of AMP’s robot has two fingers that reach down and grasp; there’s a strain sensor on each finger and a pressure sensor at the grasping portion. Based on how

(12)

these sensors are constructed, it can also identify metal objects, since it changes the electric field it is measuring (AMP Robotics, n.d.; CNBC, 2019). Rocycle operates in a similar way: sensors in the soft Teflon ‘fingers’ of Rocycle’s hand enable the robot to determine the nature of an item and sort it correctly through detecting object size and stiffness (Engadget, 2019). RoCycle is not completely accurate yet, which motivates its developers to work on the next step: combining the existing touch system with a camera-based computer vision to optimize the robot’s recycling capability. Machinex also does not claim that their SamurAI and its applications make human involvement completely redundant, mentioning that their product is ‘a perfect solution to reduce the dependence on manual sorting within your recovery facility’ (Machinex, 2019).

2.2.3 Factors influencing recycling intention and behaviour

Barr et al. (2001) have summarized extensive research and identified a number of factors that predict both behavioural intention to recycle and actual environmental behaviour. These factors are placed into three broader categories: environmental values (which is included in this study as general environmental concern, which will be discussed in the method section), situational variables and psychological variables. Variables in the second category are defined by an individual’s personal circumstances at a specific time, and are therefore unpredictable and susceptible for situation-to-situation changes. Examples are contextual and spatial differentiation in recycling services and provision (e.g. Derksen & Gartell, 1993), abstract knowledge (general environmental knowledge) and concrete knowledge (knowing what to recycle and were)(Schahn & Holzer, 1990) and behavioural experience (Daneshvary, Daneshvary & Schwer, 1998). The third category, psychological variables, contains perceptions and personal traits of the individual. Altruistic tendencies (Hopper & Nielsen, 1991), intrinsic motives (De Young, 1986), perception of the waste problem as self-threating (Baldassare & Katz, 1992), environmental citizenship (feeling a strong personal responsibility towards the environment, Selman, 1996) and perceptual factors like time, convenience and storage place (Huhtala, 2010) belong to this category. Barr et al. (2001) identified acceptance of the norm to recycle as additional important predictor of willingness to recycle: ‘people are more willing to recycle if those around them do as well, such as neighbours, friends, and peers’ (p. 2041).

As mentioned, both willingness to recycle and actual behaviour turn out to be norm based. Hage et al. (2009) distinguish between two types of norms. The first category is

(13)

social norms, which are ‘enforced by explicit approval or disapproval from others’ (p. 156), especially by other people or social groups that are considered important to the person (Oom do Valle, Reis, Menezes & Rebelo, 2004). Lang et al. (2014), who applied the cognitive constructs of Ajzen’s (1991) theory of planned behaviour to the recycling sector, used the name subjective norm to explain the exact same concept.

The second type of norm identified by Hage et al. (2009) is moral norms, which imply that individuals sanction themselves. Schwartz (1970) elaborates on moral norms, arguing that ‘the individual must also feel a personal responsibility to recycle; they should not believe that it is (solely) some other actors’ responsibility to solve waste management problems’ (Hage et al., 2009, p. 156-157). Responsibility is defined as ‘the extent to which decision makers feel a sense of ownership of the outcome and so may credit themselves for good and blame themselves for bad outcomes’ (Botti & McGill, 2006, p. 212). Moral norms can lead to internal conflicts in situations of making self-serving decisions. As applied to the recycling context by Mazar, Amir & Ariely (2008), a conflict exists between the desire to feel like a fair and good person that behaves responsibly and does not harm others and the desire to benefit from the rewards of the self-serving act by saving time and convenience.

2.2.4 Self-reported and actual recycling behaviour

As explained in the previous section, extensive research has been conducted to identify factors that contribute to both individuals’ intention to perform and actual performance of recycling behaviour. The majority of studies that focused on recycling, measure this behaviour via self-report (e.g. Aguilar-Luzón, García-Martínez, Calvo-Salguero & Salinas, 2012; Andersson & von Borgstede, 2010; Corral-Verdugo, 1997), since they are generally more cost-effective and less time-consuming to gather (Paulhus & Vazire, 2007). In general, attitudes, which form the base for self-report, are predictors of future behaviour (Kraus, 1995; Maio, Olson & Cheung, 2013). Self-report, however, is only partially capable of predicting actual behaviour, since self-reports are often overstatements of observed recycling behaviour (Chung & Leung, 2007; Barker, Fong, Grossman, Quin & Reid, 1994) due to being affected by cultural norms or social expectations (Hergat Huffman et al., 2014). However, research by Warinner, McDougall and Claxton (1984) affirms the notion that a significant relationship exists between self-report and observational data. Other recycling studies also support this by finding a weak, but

(14)

statistically significant positive correlation between self-report and observed behaviour (Corral-Verdugo, 1997; Gamba & Oskamp, 1994).

2.3 Human attitudes towards human waste workers

According to research, labour work in the recycling sector is associated with the four D’s: ‘it is dirty, often demeaning, physically demanding and in some cases, dangerous; added to which it is extremely lowly paid’ (Gregson et al., 2016, p. 543). The concept of ‘dirty work’ has existed for some time, referring to occupations and tasks that are most likely perceived as degrading or disgusting (Hughes, 1951). This type of work is historically associated with and executed by marginalised and foreign workers (Zimring, 2004), since the task of separating waste, also known as resource recovery, is work that locals in the Northern EU member states are often unwilling to do (Tannock, 2015). More specifically, the work is particularly associated with both non-EU nationals and workers from the A8 member states (being the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia, who joined the EU in 2004).

According to Douglas (1966), waste work is closely linked to the image of being contaminating, impure and symbolically damaging. The fact that the work is very physical brings workers and the rarely cleaned discarded goods (including the materials they release) into close proximity, with the potential of negatively affecting the workers’ health and wellbeing. Additionally, due to lingering in receptacles before moving to onward sorting, materials and goods deteriorate in quality and can attract animal life. This also leads to the development of a very recognisable pungent smell (Gregson et al., 2014), which makes the type of work even less attractive.

The unattractiveness of the work itself is transferred to the waste workers performing the job, who are by many associated with physically impure materials. As explained by Hughes (1962, in Ashforth & Kreiner, 1999, p. 413), ‘society delegates dirty work to groups who act as agents on society’s behalf, and that society then stigmatizes these groups, effectively disowning and disavowing the work it has mandated. Group members are seen to personify the dirty work such that they become, literally, ‘dirty workers’.’ Projecting these negatively qualities that are associated with dirt onto waste workers, helps society to continue regarding themselves as clean and, therefore, superior. This leads to waste work becoming a mean to inscribe distinctions between workers from the EU-15 and non-EU countries and the previously explained A8 countries.

(15)

Dehumanization

One concept that could potentially come into play when discussing individual’s perception of waste workers is dehumanization. Kelman (1976) characterizes this concept by denial of two things: a person is denied ‘identity’, which means the person is ‘as an individual, independent and distinguishable from others, capable of making choices’ (p. 301), and ‘community’, which means being ‘part of an interconnected network of individuals who care for each other’ (p. 301). Dehumanization represents people in an objectified fashion and causes that they are treated as a means towards vicious ends (Haslam, 2006). Haslam’s theoretical model of dehumanization distinguishes between lacking human uniqueness and lacking human nature. When lacking uniqueness, people are linked to animals and are perceived to lack refinement, civility, moral sensibility and higher cognition. Rozin, Haidt and McCauley (2000) found that phenomena that remind us of our animal nature fundamentally elicit disgust, an emotion commonly and previously linked to ‘dirty work’. When lacking human nature, people are described as mechanistic and perceived to lack emotionality, warmth, cognitive openness, individual agency and depth. Compared to lacking human uniqueness, lacking human nature is more likely to imply indifference rather than disgust. Both components can be applicable to attitudes towards human waste workers, since the repetitive nature of waste work could be viewed as mechanic, while the associations with waste in general can activate associations with animals due to eliciting feelings of disgust.

Blocking identification with victims by seeing them as sub-human objects rather than actual persons with feelings, hopes and concerns (Bandura, 2002) can selectively disengage moral self-sanctions. Opotow (1990) introduces ‘moral exclusion’, a process in which others are placed ‘outside the boundary in which moral values, rules and considerations of fairness apply’ (p. 1). Linked to the outsourced responsibility of separating waste, this might imply that moral exclusion can be used as a tool that ‘makes up’ for the waste workers suffering as a direct result of not taking personal responsibility for separating waste.

Empathy and kinship

Empathy is defined as ‘an affective response more appropriate to someone else’s situation than to one’s own’ (Hoffman, 2001, p. 4). In addition to shared group identity (Smith, Coats & Walling, 1999), relational closeness (Aron, Aron, Tudor & Nelson, 1991) and perspective taking (Davis, Conklin, Smith & Luce, 1996), kinship is identified as a

(16)

factor that leads to empathic concern for others (Maner et al, 2002). The higher the degree of social kinship, the more concern people show (Davis, 1994). This also means that the more (or less) we can see ourselves in others, the more (or less) relevant the other’s welfare is (Cialdini, Brown, Lewis, Luce & Neuberg, 1997). Empathic concern is a fundamental factor in other-oriented prosocial tendencies (Eberly-Lewis & Coetzee, 2015). The empathic concern we feel for other human beings is related to willingness to help through the empathy-altruism hypothesis (Batson, 1991), which ‘holds that empathic concern for another leads to truly selfless motivation to help that other’ (Maner et al, 2002, p. 1601). The (dis)similarities we see between ourselves and others can also influence how comfortable we feel about performing self-serving behaviour, a matter related to the concept of Motivated Dissimilarity Construal (MDC). This means that when a person anticipates feelings of discomfort (e.g. guilt and/or dissonance) about performing self-serving behaviour, focusing on the dissimilarities between themselves and the ‘victim’ of their behaviour helps them to distance themselves psychologically from the victim and lower these feelings of discomfort (Noval et al, 2018).

Applied to the context of recycling, the waste worker would be the victim of an individuals’ self-serving behaviour, which in this case would be translated to choosing the convenient and time saving option of not sorting waste single-handedly. According to theory, dissimilarities seem to form the basis of the extent to which humans feel comfortable about performing this type of behaviour. Whether humans perceive human waste workers differently than robotic waste workers, could therefore potentially affect how comfortable an individual is with outsourcing the waste separation task.

2.4 Human attitudes towards robots

Anthropomorphism

How humans perceive robots is of great influence of the extent to which they are able to relate to them. One important concept that comes into play when comparing humans to robots is anthropomorphism. This term refers to ‘attributing characteristics that people intuitively perceive to be uniquely human to nonhuman agents or events’ (Waytz, Epley & Cacioppo, 2010, p.58). The most important implication of anthropomorphism is the fact that, as soon as a non-human agent is perceived as human, it is rendered worthy of moral care and consideration (Gray, Gray & Wegner, 2007). Anthropomorphizing robots might be a useful tool to improve human-robotic interaction, given the fact that negative attitudes towards robots are more and more likely due to their increasing role in people’s daily lives

(17)

(Gnambs & Appel, 2019) and the harmony between humans and technological creations appears to still be inherently fragile (Stein, Liebold & Ohler, 2019). This is confirmed by Smith and Anderson (2017), which found that American adults are twice as likely to worry (vs. being enthusiastic) about robotic involvement in the workplace. In the case of recycling, the extent to which humans anthropomorphize robotic waste workers can affect how comfortable they are with botsourcing the waste separation task due to the extent to which they perceive these robots as worthy of moral care and consideration.

Social distance

Social distance is ‘the extent to which people perceive a lack of intimacy with individuals with different characteristics such as ethnicity, race, religion, occupation, and so on’ (Park, 1924 in Kim & Mutlu, 2014, p. 784). It can affect robotic design by manipulating dimensions like task design (cooperation vs. competition), role design (supervising vs. subordinate) and work distance (close proximity vs. distance). Research shows that the cooperating (vs. competing) robot was rated as more sociable and intellectual (Mutlu et al., 2006).

Service vs. industrial robots

Perception of a robot can also be influenced by how they are framed in usage. A distinction can be made between service robots and industrial robots. Jörling, Böhm and Paluch (2019) have looked at perceived outcome responsibility in encounters with service robots. These robots are linked to the previously explained concept of anthropomorphism by being perceived as social agents due to their high level of agency and physical embodiment. Schraft, Hägele and Wegener (2004, in Decker, Fisher & Ott, 2017, p. 348) explain that ‘a service robot is a freely programmable mobile device carrying out services either partially or fully automatically. Services are activities that do not contribute to the direct industrial manufacture of goods, but to the performance of services for humans and institutions’. Service robots, however, differ from industrial robots. According to Sprenger & Mettler (2015, p. 271), ‘the key differentiator between an industrial robot and a service robot is not the robot itself, but rather the context it is operating in.’ Service robots interact with people and serve human needs by performing tasks in the human environment. Which category robots in the recycling sector belong to, has not been clearly determined (yet). However, based on the previous description of a service robot, this is not the category a recycling robot would typically fit into. An important question is how the public perceives

(18)

recycling robots. Since presenting information in a certain frame influences individuals’ responses (Tversky & Kahneman, 1981), framing the robot as providing service (instead of help) can influence human’s perception, leading to complete outsourcing of their moral responsibility of recycling.

2.5 Message framing

It is expected that the way in which an audience responds to a particular message can partly depend on how this message is composed and encoded by the recipient (Pelletier, Lavergne & Sharp, 2008). Hence, message framing is often used to shape construct meaning and perceptions (Cheng, Woon & Lynes, 2011). When a message is framed in a manner that is consistent with the audience’s knowledge, goals and beliefs, it is associated with enhanced attention and learning. Message framing should therefore be capable of influencing the way in which individuals respond to messages that promote environmental behaviour.

This has been confirmed by several researchers who have identified message framing as a tool to make an appeal to people’s feeling of responsibility for eco-friendly behaviour. Bamberg and Möser (2007) researched whether a loss frame or gain frame is most persuasive when comparing the emotions guilt and shame, since ‘both emotions are known to drive consumers to feel morally responsible to behave pro-environmentally’ (in Baek & Yoon, 2017, p. 440). Xu, Arpan & Chen (2015) found that environmentally framed benefits of saving energy led to more positive attitudes than economically framed benefits. More research about the effect of message framing has been conducted in the field of climate change (Scannel & Gifford, 2013; Morton, Rabinovich, Marshall & Bretschneider, 2010) and adoption of innovative, sustainable products (Moon, Bergey, Bove & Robinson, 2016).

Besides the usage of message frames in the context of environment and sustainable behaviour, a small amount of research has focused on its role in attitudes towards technology. Kurila, Lazuras and Ketikidis (2016) have looked at a loss vs. gain frame in the context of message framing’s effect on technology acceptance intentions. Shen (2015) investigated how a regulatory focus messaging (prevention vs. promotion frame) moderates the consumers’ attitude toward using apps for smartphones. Even though it can be concluded that message framing has been deemed relevant in multiple studies regarding environmental behaviour and technology, it has not yet been specifically linked to the recycling sector and the way it affects performance of sustainable behaviour.

(19)

3. Hypothesis development and conceptual model

By combining previously discussed theories and findings, a number of hypotheses can be formulated.

Since separating waste is perceived as a task requiring basic knowledge rather than emotion, which are normally tasks for which botsourcing is generally excepted (Waytz & Norton, 2014), being comfortable with completely outsourcing responsibility would be more likely to occur when being exposed to a robotic worker (vs. a human worker). This might also be explained by the negative attitudes people have towards robots (Gnambs & Appel, 2019; Smith and Anderson, 2017). On the other hand, human workers are often directly associated with the dirty work they perform (Hughes, 1962), which can lead to applying theories of dehumanization (Kelman, 1976; Haslam, 2006), moral exclusion (Opotow, 1990) and psychological distance (Noval et al, 2018). Theories about kinship (Maner et al, 2002), norms (Hage et al., 2009) and desire to perceive oneself as a fair and responsibly behaving person (Mazar et al., 2008), however, lead to the expectation that individuals’ are more affected by exposure to human waste workers (vs. robotic workers) performing the waste separation. This leads to the development of hypothesis 1 and 2:

H1: Exposure to humans performing the waste separation task in recycling facilities leads to a higher self-reported intention to perform sustainable behaviour than exposure to robotic workers performing this task

H2: Exposure to humans performing the waste separation task in recycling facilities leads to higher performance of sustainable behaviour through participation in a lottery than exposure to robotic workers performing this task.

Based on existing research about empathic concern and willingness to help humans (Batson, 1991) and positive evaluations of cooperative robots (Mutlu et al., 2006), the following hypotheses with regard to the influence of message framing are developed:

H1b: Framing the waste separation task as ‘helping’ will lead to higher intentions for performing sustainable behaviour than ‘serving’.

(20)

H2b: Framing the waste separation task as ‘helping’ will lead to higher willingness to perform sustainable behaviour through participation in a lottery than ‘serving’.

As argued previously, self-report is only partially capable of predicting actual behaviour due to cultural norms and social expectations (Chung & Leung, 2007; Barker et al., 1994; Hergat Huffman et al, 2014). A number of studies, however, found a significant relationship between self-report and observational data (Warinner et al, 1984; Corral-Verdugo, 1997; Gamba & Oskamp; 1994). Even though the relationship is expected to be weak due to an existing gap between intention and behaviour, a statistically significant correlation between self-reported intention and actual sustainable behaviour, which in this study is measured as participating in a lottery, is expected. Based on this knowledge, a separate hypothesis is developed:

H3: There is a positive, but weak relationship between self-reported intention to perform sustainable behaviour and actual sustainable behaviour

To find out whether an individual’s intention is in line with their actions once they agree to participate in the lottery, meaning that people who report high intentions of performing sustainable behaviour are willing to donate a high amount of the money, an additional hypothesis is developed:

H4: Among lottery participants, there is a positive, but weak relationship between self-reported intention to perform sustainable behaviour and the amount of money donated to charity

The conceptual framework that visualizes the potential relationships between variables is attached in appendix 1A.

(21)

4. Data and method

In the following section, the methodological considerations of this study will be discussed. An operationalization of the main variables is given, as well as an explanation of the procedure followed in the main study.

4.1 General design

This study uses an online experiment with a 2 (exposure: human waste worker vs. robotic worker) x 2 (message frame: help vs. serve) between-subject design to test the hypotheses.

The independent variable was exposure to this task; the moderating variable was message framing; the dependent variables were self-reported intention to perform sustainable behaviour and actual sustainable behaviour. These variables will be operationalized in section 4.3 Operationalization of variables. Online survey tool Qualtrics was used for the development of this survey. Respondents were collected through usage of online participant recruitment tool Prolific as well as personal efforts to distribute the survey. A control variable was created to control for potential differences among participants who were recruited in different ways; this is as well as other demographical information will be reported in section 5.1 Descriptives and frequencies.

4.2 Procedure and materials

This section describes the online experiment in detail. In the debriefing respondents were asked to answer a number of questions about a newspaper article and an accompanying video related to the opening of a new recycling facility. The article was presented as online news article on website DutchNews. Even though DutchNews is an actual website reporting Dutch news to an English audience, both the newspaper article and the video were solely created for the purpose of this study.

Each participant (N = 198) was randomly assigned to one of the four conditions, being exposure to human waste workers helping, human waste workers serving, robotic waste workers helping and robotic waste workers serving in performing the task of separating waste in a recycling facility. The video in the robotic condition existed of combined footage of two videos uploaded to YouTube by WCCO – CBS Minnesota (2017) and CNBC (2019). The fragments were cut, combined and accommodated with a basic sound in online platform Kapwing; so was the video in the human condition for which elements

(22)

from a YouTube-video by Cristian Lazar (2016) were used. Both the newspaper and video were accommodated with a title in which either help or serve (depending on the condition) was stated in bold. Stressing this frame more clearly was a way of making participants notice this detail specifically among the other information stated in the title. Except for the title and explaining different processes taking place in the facility in the content of the article, all other information included was equal across the different versions.

The article participants in the human-help condition were redirected to, mentioned the opening of a new recycling facility centre in which human labour workers helped Dutch households with separation waste (see appendix 1B). The video illustrated this process by displaying human workers picking waste by hand from a large conveyer belt. Participants in the human-serve condition were redirected to the same video as the human-help condition, but the article differed. Even though the content of the news article also mentioned the opening of the new recycling facility centre, it stressed the fact that human labour workers served Dutch households in this task (see appendix 1C). Participants in the robotic-help condition were redirected to an article that covered the same content as the articles in the human conditions, even though this article stressed that this recycling facility approaches operations differently by exclusively using robotic workers to ‘help’ perform the task of separating recyclable waste (see appendix 1D). The video displayed robotic workers performing the waste separation task. The robotic-serve condition was designed with the same approach: all other content equal to the robotic-help condition, this article stressed the robot’s task execution as ‘serving’ Dutch households (see appendix 1E). The video used for this category was the exact same as the other robotic video.

After being exposed to the manipulation, the survey continued with three attention check questions to examine whether the participant had paid sufficient attention and read the article carefully. Respondents were asked to re-state the title of the article and answer two additional questions related to the content of the article. Literally re-stating the article’s title would indicate that the respondent had at least been exposed to the priming of using either help or serve.

The survey continued with nine 5 Likert scale items, measuring the respondents’ self-reported planned sustainable behaviour for the next month. To check for the gap between planned sustainable behaviour and actual sustainable behaviour, respondents were then presented with the option to voluntarily participate in a lottery in which 50 euros could be won. In case the respondent wished to participate, he or she had to move a slider to indicate what amount of the money would be donated to organisation PlasticOceans, and

(23)

what amount would be kept for personal purposes. Lastly, it was stated that the researchers were ‘interested in some additional information about the participant and some demographic characteristics’. To serve as a control variable, the survey ended with three more 5 Likert scale items on self-reporting general environmental concern. Age and gender were also included in the survey to serve as control variables.

To ensure participants were unaware of the study’s true purpose of examining recycling behaviour, which would entail the risk of them providing socially desirable answers, it was decided to utilize a deception strategy by creating new stimuli containing fake information, solely created to serve the purpose of this study. After participation, respondents were clearly informed and apologized to that they had been deceived by being exposed to both a news paper article and a video that were solely created for the purpose of this study.

4.3 Operationalization of variables

Exposure to waste separation tasks in recycling facilities

The independent variable used in this study is exposure to waste separation tasks in recycling facilities. Exposure is defined as ‘the fact of experiencing something or being affected by it because of being in a particular situation or place’ (Cambridge Dictionary, 2020). This definition incorporates an important criterion for the way the concept is operationalized in this thesis. Participants are exposed to a newspaper article and an accompanying video of either humans or robots performing the waste separation task, which makes them experience what the work performed in such a facility is like. Exposure was manipulated in the research to study the effects on the dependent variables measuring sustainable behaviour.

Message framing

The moderating variable, message framing, is operationalized by framing the title of this article and video as either helping or serving households in performing this task of separating waste. As mentioned before, the way in which a message is framed can influence to what extent it is attended to, the knowledge an audience gains from it and how positively or negatively the message is evaluated (Graber, 2004). To ensure the participant was exposed to either of two message frames, an attention check question was included in which participants were asked to directly re-type the title of the article. To prevent participants from realizing they were asked this question for specific attention check

(24)

measures, the question was formulated as ‘to check your attention and ensure this survey is answered by a person and not a bot, please re-type the title of the article here’.

Self-reported intention to perform sustainable behaviour

Self-reported intention to perform sustainable behaviour, the first dependent variable of this study, was measured by nine items in which participants were asked to indicate the behaviour they are planning to perform in the next month, e.g. ‘I will recycle glass’ and ‘I will buy products with less packaging’. These items are part of a pro-environmental behaviour scale originally created by Preisendörfer (1988), but translated by and more extensively discussed by Bamberg and Rees (2015). These items were measured on a 5-point Likert scale (1 = totally disagree, 5 = totally agree).

Actual sustainable behaviour

Actual sustainable behaviour, the second dependent variable of this study, was measured through participant’s willingness to participate in a lottery. Leaving an e-mail address was required for participation, since that was the only way to reach out to the winner. In case of participation, respondents were asked to indicate to what extent the money would be kept for personal purposes or donated to non-profit organisation PlasticOceans, which raises awareness about plastic pollution to inspire behavioural change.

General environmental concern

General environmental concern was measured with three selected items from Taufique, Siwar, Talib and Chamhuri (2014), being ‘I consider the potential environmental impact of my actions when making many of my decisions’, ‘I would describe myself as environmentally responsible’ and ‘I worry about environmental problems’. Participants answered these items on a 5 point Likert scale (1 = totally disagree, 5 = totally disagree).

(25)

5. Results

5.1 Descriptives and frequencies

The file imported from Qualtrics existed of 212 people in total, of which 14 people were deleted due to missing data. Qualtrics indicated that these incomplete cases were finished for 58% and are therefore qualified as accepted, even though in practice this means that participants decided to leave the survey straight after agreeing to participate, so right before they were presented with the article they were asked to read. Deleting these cases resulted in a total of 198 respondents. These respondents were recruited through usage of online recruitment platform Prolific and through personal distribution of the survey. The binary variable recruitment was created to control for potential differences between these groups. Respondents were randomly assigned to either the human (N = 102) or the robotic condition (N = 96). Within the human conditions, 53 respondents were assigned to the condition with the help message frame; the remaining 43 respondents were conditioned with the serve message frame. Within the robotic condition, this distribution was 49 respondents for the help, and 47 for the serve message frame.

Three manipulation check questions were included per condition to check for paid attention during exposure to the news article and the video. After controlling for whether the title of the article was re-typed correctly, 43 respondents remained in the ‘human help’ condition, 37 in the ‘human serve’ condition, 30 in the ‘robotic help’ condition and 34 in the ‘robotic serve’ condition. A dummy variable was created to separate the participants who correctly identified the title from those who did not (HelpServe_CorrectTitle). A new variable was computed for the second attention check question across all four conditions (AttentionCheck1); this was also done for the third attention check question (AttentionCheck2).

The dependent variable self-reported intention to perform sustainable behaviour is a scale variable existing of nine items (Q155_1 to Q155_9) that were measured on a 5 point Likert scale (1 = totally disagree; 5 = totally agree). The mean for ‘In the next month, I plan to recycle paper’ (Q155_1) was highest (M = 4.31, SD = .96); the mean for ‘In the next month, I plan to use public transport as an alternative to using a car’ (Q155_9) was lowest (M = 3.03, SD = 1.45). As concluded from the negatively skewed histogram and Q-Q plot and the significant Kolmogorov-Smirnov and Shapiro-Wilk tests, self-reported

(26)

skewness = -.945). This indicates a relatively high intention to perform sustainable behaviour.

The dependent variable actual sustainable behaviour is a binary variable since participants could either answer ‘yes’ (N = 154) or ‘no’ (N = 44) to whether they wished to participate in the lottery. Since the participants answering ‘yes’ were asked to decide which amount of 50 euros they wished to donate, amount is continuous and treated as an additional dependent variable in an analysis ran with only lottery participants (M = 16.85,

SD = 16.17, range: 0–50). Data were not normally distributed based on previously

mentioned tests. A positively skewed distribution suggests a tendency towards not donating or only donating a limited amount of the 50 euros (skewness = .819).

Age (Q167) of the participants ranged between 18 and 76 (M = 33.25, SD = 12.94),

although more than half of the participants were between 20 and 30 years old. In terms of

gender (Q165), 104 were female, 89 were male and 5 identified as others. Since the

majority of the respondents were female, it was decided to create a dummy variable and use the female category as baseline group; males and others were treated as non-female (Female).

As a third control variable general environmental concern was included, a scale variable existing of three 5 point Likert scale items (Q163_1 to Q163_3). The mean for ‘In general, I worry about environmental problems’ (Q163_3) was highest (M = 4.17, SD = 1.03); the mean for ‘In general, I describe myself as environmentally responsible’ (Q163_2) was lowest (M = 3.64, SD = 1.01). Data were, again, not normally distributed. A negatively skewed distribution suggests high self-reported general environmental concern (M = .3.91, skewness = -1.23).

Since people who report to be high in general environmental concern might be less open to manipulation in the conditions due to their strong predisposition, a median split was performed to create a variable including only those participants who reported to be low in general environmental concern. GenConLOW was created to reproduce analyses in which general environmental concern could play an important role.

5.2 Reliability analysis for scales

A reliability analysis was performed for both scales included: self-reported intention to

perform sustainable behaviour and general environmental concern. Cronbach’s Alpha (α)

should be greater than .70 for acceptable internal consistency of the items (Tavakol & Dennick, 2011). Self-reported intention to perform sustainable behaviour consisted of 9

(27)

items showing high reliability (α = .806). The corrected item-total correlations indicate that all items had a good correlation with the total score of the scale (α > .30), except for ‘I want to use public transport as an alternative to using a car’ (Q155_9, α = .246). Since this indicates a very low correlation with the total score of the scale and deleting this item would improve Cronbach’s Alpha to .828, it was decided to exclude this item from the scale and the analysis. A new item, PlannedTOT8, was created to continue the analyses with. With α = .773, the scale for general environmental concern showed high reliability as well (GeneralConcernTOT).

For both scales, recoding of counter-indicative items did not apply to the items included.

5.3 Differences across conditions

To test any significant differences in the relationships between the four conditions and the binary variables (Female, Recruitment, HelpServe_CorrectTitle, AttentionCheck1 and

AttentionCheck2), Chi-Square tests were performed. No significant differences were found

for distribution across the four conditions in terms of gender (X2

= 1.53, p = .68), recruitment approach (X2

= 3.48, p = .32), correct title (X2

= 5.35, p = .15), correct answer to attention question 2 (X2

= 6.14, p = .11) and attention question 3 (X2

= 6.23, p = .10). One-way ANOVA tests were performed to show whether general environmental concern (GeneralConcernTOT) and age (Q167) differed significantly across the four conditions, which turned out to not be the case for both variables (respectively F(3, 194) = 1.24, p = .30 and F(3, 194) = .92, p = .43). It can therefore be concluded that these six variables are not responsible for any between-group differences in self-reported intention

to perform sustainable behaviour.

5.4 Correlations

A correlation test was performed including all main variables: self-reported intention to

perform sustainable behaviour, actual sustainable behaviour (lottery participation), amount donated, exposure, message frame, message frame corrected for title, gender, general environmental concern, age, attention check 1, 2 and recruitment. Table 1 shows

(28)

Variables M SD 1 2 3 4 5 6 7 8 9 10 11 12 1. Self-reported intention 3.84 .76 (.83) 2. Actual sustainabl e behaviour (lottery) .80 . 42 .17* - 3. Amount donated 16.85 16.17 .11 - - 4. Exposure .52 .50 -.02 -.01 .01 - 5. Age 33.25 12.94 .07 -.13 .08 -.10 - 6. Gender .53 .50 .17* .05 .05 .01 .09 - 7. General Environm ental Concern 3.91 .83 .69** .16* .20* -.06 .07 .16* (.77) 8. Message frame .52 .50 -.04 -.06 -.04 .01 .05 -.07 -.10 - 9. Message frame corrected for title .73 .45 .05 .14 -.08 .13 -.34* * .03 .01 -.03 - 10. Attention Check1 .99 .10 -.09 -.05 -.05 .10 .02 .01 -.11 -.10 .17 * -

(29)

11. Attention Check2 .98 .14 .01 .01 -.00 .08 .01 .08 .04 -.07 .23 ** .70 ** - 12. Recruit- ment .49 .50 -.02 .70* * -.18 -.06 -.32* * .09 .12 -.18 .04 - .1 2 -

Table 1: Descriptive statistics and correlations between key variables

* Correlation is significant at the 0.05 level tailed) /** Correlation is significant at the 0.01 level (2-tailed)

Two separate dependent variables were included in the research to measure momentous performance of sustainable behaviour after exposure to the stimuli. The first variable,

self-reported intention, indicates participants’ intended sustainable behaviour; the second

variable, actual sustainable behaviour, measures real-life behaviour and is measured through willingness to participate in a lottery. As expected, these two variables are positively correlated (r = .17, p = .02) since they reflect the same dependent variable: sustainable behaviour. According to standards set by Cohen (1988), the correlation is considered small since a small to medium effect size is r = .10 to .36. This phenomenon can potentially be explained by the existing gap between intention and actual performed behaviour, which will be elaborated on in the discussion section.

The main model also includes a control variable, general environmental concern, which measures participant’s general predisposition about sustainability. This control variable showed to have significant expected correlations with both dependent variables

self-reported intention (r = .69, p < .01) and actual sustainable behaviour through lottery

participation (r = .16, p = .03). Among those participants willing to participate in the lottery, general environmental concern also showed to have an expected correlation with the amount donated (r = .20, p = .02).

Since gender (r = .17, p = .02) turns out to have an expected correlation with

self-reported intention, this control variable will also be included in further analysis regarding

this dependent variable. Recruitment will be treated in the same way since it is expected to have a correlation with actual sustainable behaviour (r = .70, p < .01), meaning this variable will be included in further analysis regarding lottery participation.

(30)

5.5 Hypothesis testing

Hypothesis 1

‘Exposure to humans performing the waste separation task in recycling facilities leads to a higher self-reported intention to perform sustainable behaviour than

exposure to robotic workers performing this task’

A one-way ANOVA test was performed to show whether the mean of self-reported

intention to perform sustainable behaviour significantly differed for people exposed to the

human (M = 3.83) vs. the robotic (M = 3.86) condition. The results show that this is not the case, since exposure equates to F(1, 196) = .08, p = .78.

Since the one-way ANOVA solely tested whether there is a significant relationship between the independent variable exposure and the dependent variable self-reported

intention, a hierarchical linear regression was performed to test the entire model including

other variables that could have a relationship with self-reported intention to perform

sustainable behaviour. This regression analysis included the independent variable,

moderator and relevant covariates (as concluded from the correlation matrix) to predict this dependent variable. The regression results are reported in table 2.

R R2 R2 change B SE β t Step 1 .04 .00 -.01 Exposure -.03 .11 -.02 -.28 Message framing -.06 .11 -.04 -.52 Step 2 .05 .00 -.01 Exposure -.06 .16 -.04 -.35 Message framing -.08 .16 -.05 -.51 Inter_HumanHelp .05 .22 .03 .22 Step 3 .69 .48* .47* Exposure .08 .11 .06 .73 Message framing .11 .11 .07 .95 Inter_HumanHelp -.11 .16 -.07 -.71 General environmental concern .63 .05 .69* 12.84

A study on how the framing of outsourcing and botsourcing waste separation affects sustainable behaviour