1 Help, my financial adviser is a robot!
The effect of mechanistic dehumanization on saving
Giannina Smit (11635746)
Master Thesis MSc Business Administration Specialization: Digital Business
University of Amsterdam Academic year: 2017-2018 Supervisor: Dr. A.N. Weihrauch 22-06-2018
2 Statement of originality
This document is written by Student Giannina Smit 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 Table of contents Abstract ... 4 Introduction ... 5 2. Literature review ... 10 2.1 Dehumanization ... 10 2.1.1 Mechanistic dehumanization ... 10
2.1.2 An expectation to adopt a mechanistic approach ... 11
2.2 Decision-making ... 13
2.2.1 Affective and analytical decision-making ... 13
2.2.2 Mechanistic dehumanization and analytical thinking ... 15
2.2.3 Thinking style and achievability ... 17
2.3 Consumer financial decision-making ... 18
2.3.1 Emotional biases in financial decision-making ... 19
2.4 Saving ... 21
2.4.1 Rational saving ... 22
2.5 Conceptual model ... 25
3. Data & Method ... 26
3.1 Data collection ... 26
3.2. Stimulus design and pretest ... 26
3.3 Procedure ... 27
4. Results ... 30
4.1 Descriptive and frequencies statistics ... 30
4.2 Reliability analysis for scales ... 32
4.3 Hypotheses testing virtual telepresence ... 35
4.4 Hypotheses testing humanoid ... 38
5. Discussion ... 41
5.1 Overall conclusion ... 41
5.2 Theoretical contribution ... 42
5.2 Limitations and future research ... 44
5.3 Managerial implications ... 46
References ... 48
APPENDIX A ... 63
4 Abstract
Due to rapid technological developments, the boundaries between humans and machines are blurring. This research complements the literature on consumer behavior and mechanistic dehumanization, which makes humans more like machines. The research explores the consequences of exposure to mechanistic dehumanization on financial decision-making. This relationship is especially interesting to study in the financial industry because banks increasingly expose consumers to mechanistic dehumanization stimuli, such as humanoids and teleconferencing financial advisers. In addition, consumers often make irrational financial decisions, which creates interest in understanding whether the use of mechanistic dehumanization can push consumers toward an analytical thinking style. Through exposing consumers to mechanistic dehumanization stimuli (external appearance and physical movement), the effects of these stimuli on saving were documented. In the study, 291 respondents participated via an online between-subjects experiment and were assigned to one of two mechanistic dehumanization conditions (virtual telepresence or humanoid) or the human condition serving as a reference group. The results displayed an unexpected pattern. Consumers perceived no difference between a teleconferencing financial adviser and a humanoid, categorizing both as machines. Furthermore, the results displayed that the exposure to humans-as-machines does not lead to a heightened expectation of a mechanistic approach to saving and does not affect saving. The findings contribute to research on mechanistic dehumanization and consumer behavior, consumer financial decision-making, saving, affective and analytical thinking styles, and the uncanny valley literature. The findings have practical implications for banks and the communication of financial decision-making.
Keywords: (mechanistic) dehumanization, consumer financial decision-making, affective and analytical thinking styles, achievability, saving
5 Introduction
“Humans need to become machine-like in order to survive” (Belk, 2017).
We used to see machines as objects that could help us in our needs (Shibata, 2004). Due to the fast pace of technological developments, however, the boundaries between humans and machines are disappearing (Waytz, Heafner & Epley, 2014). Machines are increasingly looking like humans (e.g., humanoids and androids) (Ferrari, Paladino & Jetten, 2016) and humans are increasingly looking like machines. For instance, paralyzed people are being connected to a mind-controlled robot arm to restore their sense of touch (Devlin, 2016). Human enhancement technologies, such as transcranial direct-current stimulation, improve cognitive performance and make people look and feel like machines (Castelo, Schmitt & Sarvary, 2016). Moreover, virtual telepresence machines are increasingly used in business and social interactions, extending the representation of humans as machines (Stoll et al., 2018; Kristoffersson, Coradeschi & Loutfi, 2013). Gradually, the idea of humans as machines is becoming more normal in consumers’ daily lives (Castelo et al., 2016).
Portraying humans as machines is termed mechanistic dehumanization (Haslam, 2006). Scholars have been examining dehumanization to different domains (e.g., ethnicity and race, modern medicine), but despite the increased prevalence of humans as machines in consumers’ daily life, only one study to date has looked at the effect of mechanistic dehumanization on consumer behavior (Weihrauch & Huang, 2017). This study empirically showed that consumers with high health self-control make healthier food choices upon exposure to mechanistic dehumanization stimuli. Furthermore, it showed that this exposure triggers mechanistic expectations regarding food choices (e.g., rule obeying, unemotional, analytical) (Haslam, 2006), in which consumers with low health self-control made unhealthier food choices (Weihrauch & Huang, 2017). This indicates that exposure to mechanistic dehumanization could have consequences for consumers.
6 Although the effect of mechanistic dehumanization can affect many consumer behavior domains (Castelo et al., 2016), the financial domain is especially interesting to study. First because the financial industry has also been affected by the digital revolution. As mobile banking is lessening the need for physical branches (Berges et al., 2016), banks are experimenting with smart branches. In smart branches, financial advisers are presented through remote or virtual technology (Barbier et al., 2016) and humanoids (human-like machines) serve as the branch’s customer service (Eha, 2017; Dirican, 2015).
Second, people often make irrational financial decisions (Bernartzi & Thaler, 2001; Boshara et al., 2010; McKenzie & Liersch, 2011). For example, consumers expect savings to grow linearly resulting in underestimating the benefits of starting to save early (McKenzie & Liersch, 2011), and people keep investing in projects regardless of returns, just because they already invested so much (Hammond, Keeney & Raiffa, 2006). Such biases happen when consumers approach the decision emotionally (Stango & Zinman, 2009; Hammond et al., 2006). These emotional biases are likely to cause risks (Bernatzi & Thaler, 2007; Kuhnen & Knutson, 2011; Hammond et al., 2006). For instance, when people feel they are knowledgeable about a decision domain, regardless of their objective knowledge, they are willing to take more risk (Hadar, Sood & Fox, 2013). The general belief therefore is that financial decisions benefit from rational decision-making and that emotional decisions can lead to costly errors (Milkman, Chugh, & Bazerman, 2009; Bazerman & Moore, 2008). In addition, consumers also perceive financial decisions as incompatible with feelings (Park & Sela, 2017).
Scholars and government are therefore exploring ways to decrease suboptimal financial decision-making (Fernandes, Lynch & Netemeyer, 2014). In this paper it is proposed that exposure to mechanistic dehumanization stimuli could be such an intervention, because it increases the expectation to adopt a mechanistic approach to decisions (Weihrauch & Huang, 2017), which triggers people to apply a cold analytical thinking style (Park & Sela, 2017; Inbar,
7 Cone & Gilovich, 2010; Novak & Hoffman, 2008). Decision styles can be cued by context (Novak & Hoffman, 2008) and machines are associated with cold analytical features. Most machines are built on algorithms and therefore analytical in their essence (Lee, 2018). Furthermore, although they increasingly look like humans (Ferrari et al., 2016; Hall et al., 2014; Minato et al., 2006; Saygin et al., 2011), they are still perceived as lacking intuition, subjective judgements capabilities (Lee, 2018), and experience (Gray et al., 2007). Moreover, in business and media, machines are portrayed as cold objects. For example, Nestle’s tagline “Feeling like a machine?”, in which people who practice their work unemotionally and automatically are presented as working machines. Similarly, in business, successful sales people who just focus on reaching their targets are portrayed as sales machines (Adamson, Dixon & Toman, 2013).
Despite the increasing prevalence of humans as machines in the financial context, the risks of emotional biases in financial decisions and the effect mechanistic dehumanization can have on decisions, no research has yet examined how consumers in a financial context react to humans-as-machines. This paper seeks to answer: “How mechanistic dehumanization influences consumer financial decision-making?” More specifically, because consumer financial decision-making involves many decisions, this paper will focus on saving.
Saving is considered as a typical financial decision (Park & Sela, 2017) and has often been the subject of financial decision-making research (e.g., Ülkümen & Cheema, 2011; Soman & Zhao, 2011; Hershfield et al., 2011). In addition, many Americans fail to save at sufficient rates (Dholakia, 2016), which is alarming because the amount people save for the future influences their well-being (Thaler, 1994). Saving reflect the dilemma of “I should” and “I want” (Hsee et al., 2015). People who emphasize the “should” (i.e., reason-based thinking) show more actual savings (Hsee et al., 2015), whether emotions are shown to cause impulsive purchases (Baumeister, 2002). It is therefore interesting to study whether people who are
8 mechanistically dehumanized show increased saving, because they expect to approach decisions more cold and analytical.
Accordingly, this paper aims to answer three questions: (1) Does representing humans as machines encourage saving? (2) Does the sense of feeling expected to approach decisions mechanistically drive this effect? (3) Is there a difference in how consumers respond to these stimuli? Previous research showed that the effect of mechanistic expectation on food choices depends on consumers’ level of health self-control. Consumers with low health self-control perceived the mechanistic approach as less achievable (Weihrauch & Huang, 2017). Furthermore, in the case of financial decisions consumers seem to avoid such decision when they perceive themselves as affective thinkers, because cold, analytical decisions are not them (Park & Sela, 2017). Taken together, it is argued that the effect of a mechanistic approach toward saving depends on a person’s thinking style (i.e., affective or analytical).
This research has practical importance for managers and public policy makers. First, if people who are exposed to mechanistic dehumanization stimuli do show increased saving, public policy makers could use this information for saving interventions. Second, it is important for managers and the government to understand the possible effects of incorporating virtual and robotic technologies in smart branches. For instance, it is showed that mechanistic exposure can backfire for consumers with low-health self-control, causing them to make unhealthier food choices (Weihrauch & Huang, 2017). If the same effect holds for affective thinkers (i.e., increased suboptimal financial decision-making), this could have consequences for their saving rates and the economical growth and stability (Ülkümen & Cheema, 2011). Finally, if it has no effect, banks can use that information to consider fully automated branches.
The rest of this paper is structured as follows. The next chapter will review the relevant literature on mechanistic dehumanization, (financial) decision-making processes and saving. Moreover, the chapter provides the conceptual framework. Next, Chapter 3 discusses the
9 research design and method, followed by results in Chapter 4. Finally, Chapter 6 presents the conclusions, limitations and managerial and academic implications of the research.
10 2. Literature review
This chapter provides an overview of the relevant literature regarding the concepts of the paper. The first section of the literature review is devoted to dehumanization and its influence on consumer behavior. The subsequent section elaborates on the process of decision-making and discusses affective and analytical thinking. Next, the emotional biases in financial decision-making are analyzed. After that, saving and impulsive purchases are highlighted in the context of analytical and affective thinking. The chapter ends with the conceptual framework.
2.1 Dehumanization
2.1.1 Mechanistic dehumanization
The phenomenon of humans being portrayed as machines increasingly occurs in the daily lives of consumers (Dautenhahn, 2007). Human and machine physical features are merged in the form of cyborgs and humanoids, and mechanistic approaches to the human body are advocated by media (e.g. The incredible human machine by National Geographic) and commercial industry (e.g. Fitbit). Humans being portrayed as machines is called mechanistic
dehumanization (Haslam, 2006; Haslam & Loughnan, 2014). It is part of Haslam’s dual model
of dehumanization, which denies two senses of humanness: human nature (HN) characteristics, which include emotionality, depth and cognitive openness, and uniquely human (UH) characteristics including rationality and civility. When mechanistic dehumanization occurs, human nature (HN) characteristics, which separate humans from machines, are denied and people are perceived as inert, cold, rigid, passive, and superficial and behavior is explained in causal terms (i.e., robot-like). In contrast, when animalistic dehumanization occurs, uniquely human (UH) characteristics that separates humans from animals are denied and people are perceived as course, uncultured, childish, irrational, and lacking self-control and behavior is
11 explained by desires and wants (i.e., animal-like) (Haslam, 2006). Dehumanization can be classified in different degrees. It can occur in blatant and severe forms, which has been related to the context of violence (Kteily et al., 2015). It can also take subtle and mild forms, such as infrahumanization, which occurs when individuals perceive the out-group as less human (Leyens et al., 2000). Haslam’s dual model broadens dehumanization’s scope by classifying different behaviors that make humans look like objects such as robots and machines (Haslam, 2006; Haslam and Loughnan, 2014).
The dual model of dehumanization is pertinent to study now that almost all industries are digitizing. Especially the financial sector is fast changing the way it operates and interacts with consumers (PWC, 2014). Banks, for example, are increasingly creating smart branches, where consumers interact with remote or virtual financial advisers on video and interactive displays (Barbier et al., 2016), and in some branches consumers interact with real humanoids (Eha, 2017; Dirican, 2015). However, while the financial industry is also (increasingly) portraying humans as machines, little is known regarding the effect of mechanistic dehumanization on consumers. In this paper, it is argued that exposing consumer to mechanistic dehumanization stimuli has an impact on their saving behavior, because it changes their beliefs on the appropriate way to behave in a specific situation (Haslam, 2006).
2.1.2 An expectation to adopt a mechanistic approach
In the anthropomorphism literature, where machines are being portrayed as human, it is proven that changing physical features can elicit certain models and schemas that stimulate congruent behavior and beliefs (Aggarwal & McGill 2007, 2012; Kim & McGill 2011). For example, when a robot vacuum cleaner (e.g., a Roomba) moves at a human speed, it is perceived as more sensitive and thus more human (Sugano, Morita & Tomiyama, 2012). People have more trust in anthropomorphized cars because their competence is perceived as human-like
12 (Waytz et al., 2014), and thoughts written by machines are perceived as human when they are expressed with a human voice, but not when they are only visually presented (Schroeder & Epley, 2016).
Similarly, this process also applies in the mechanistic dehumanization literature. When humans are portrayed as machines, they are not only perceived but also expected to behave like machines. In medicine, for example, dehumanizing patients allows medical professionals to dissociate from them by expecting them to feel less pain (Haque & Waytz, 2012). Similar, consumers who dehumanize customer service employees treat them more harshly, expecting them to feel less emotions (Henkel et al., 2018). Furthermore, individuals who are exposed to the similarities between the human body and machines feel they are expected to adopt a mechanistic approach to food (Weihrauch & Huang, 2017). Therefore, exposing consumers to mechanistic dehumanization stimuli elicits mechanistic schemas and changes their expectations of how they should behave in a specific context (Weihrauch & Huang, 2017; Haslam 2006). Accordingly, the first hypothesis of this paper is:
H1: Exposure to humans-as-machines stimuli leads to an expectation that one should adopt a mechanistic approach to saving.
Although most research has focused on the negative effects of mechanistic dehumanization (Christoff, 2014; Rudman and Mescher, 2012; Bastian & Haslam, 2011; Haslam & Loughnan 2014), dehumanization has also shown to have positive effects. In the medical field, for example, health workers who dehumanized their patients suffered from fewer burnout symptoms (Vaes & Muratore, 2013) and consumers with high health self-control made healthier food choices upon exposure to mechanistic stimuli (Weihrauch & Huang, 2017). The latter shows that mechanistic dehumanization can be advantageous for decision-making. The
13 next sections elaborate on the types of decision-making (analytical and affective) and will further explain how they are linked to mechanistic dehumanization.
2.2 Decision-making
Before examining the relation between dehumanization and financial decision-making, the process of decision-making itself must be defined. Humans make many decisions in their lives, but the decision-making process is not consistent or simplistic. For instance, others can perceive the decisions by intelligent people as inaccurate, or people stick to their feeling-based choice, even though they know it is not optimal (Phillips et al., 2016). Consequently, the decision-making literature has expanded its paradigm of a rational human decision maker to an explanation where decision-making is also subject to bias and heuristics. For example, humans are overconfident in the reliability of their judgements (Lichtenstein, Fischhoff & Phillips, 1982), the likelihood estimations of events are biased by their representativeness (Kahneman & Tversky, 1984), and losses weigh heavier than gains (Tversky & Kahneman, 1992).
2.2.1 Affective and analytical decision-making
Consumer decision-making has been defined in dual process theories that describe behavior as a reflection of two distinct forms of information processing. Although dual process theories have been proposed in several ways, most authors agree on a difference between an unconscious, rapid, automatic process and a conscious, slow and deliberative process (Evans, 2008). Some examples of different names attributed to these thinking styles are intuitive and analytic (Hammond, 2000), experiential and rational (Epstein, 1994) or system 1 and system 2 processing (Kahneman & Frederick 2002; Stanovich 1999). The last is the most used and neutral term founded in the literature (Evans, 2008). Like animalistic and mechanistic dehumanization, the different information processing styles have also been identified with unique features. Processes grouped as system 1 are characterized as unconscious, rapid and
14 automatic in nature as well as with associative and lead by gut feelings and intuitions. Conscious, slow and deliberative system 2 processing is additionally characterized as analytic, controlled, effortful, and rule based (Evans, 2008).
Due to the many comparable variants of processing theories, one should be selected that is most suitable to the decision type and domain. In this paper, in which the effect of mechanistic dehumanization on saving is examined, the financial domain is most relevant. Archetypal financial decisions such as saving, investing and debt management are associated with a cold analytical mode of thinking (Park & Sela, 2017). In addition, more than other specialized, complex and important domains, financial decision-making has been related to an analytical thinking style that is incompatible with emotions and feelings (Park & Sela, 2017). For example, financial and medical decision-making are both perceived as equally important and consequential and therefore associated with systematic processing (system 2). However, when comparing the decision-making process in the financial and medical domains, most affective thinkers prefer a generic decision task over a financial decision task. For analytical thinkers this was not the case. In contrast, people in a similar medical versus generic decision task show no difference in their task preference, regardless of their perceived thinking style (Park & Sela, 2017). This indicates that financial decision-making goes beyond deliberate (system 2) and heuristic (system 1) processing and can be better explained in terms of affective and analytical thinking (Park & Sela, 2017).
There are several other reasons why affective and analytical processing can be distinguished from system 1 and 2 processing. First, affect can indeed serve as a heuristic cue, for example when a person prefers a food product that is promoted with affective words such as “new” “natural” and “improved” (Slovic et al., 2007). However, making decisions based on feelings does not implicitly indicate that affective thinking is heuristic with less cognitive capacity or response time. Negative affect could involve substantial deliberation (Schwarz,
15 1990). Moreover, affective decisions could take as long as analytical ones (Lee et al. 2015) and when people experience uncertainty individuals increasingly rely on affect and less on heuristic processing (Faraji-Rad & Pham, 2016). Therefore, affective processing is not the same as system 1 processing and should be considered in terms of thought content instead of effort or speed (Park & Sela, 2017).
Returning to financial decision-making, the research from Park and Sela (2017), which links financial decision-making to a cold, analytical mode of thinking, agrees with the general belief that emotions do not foster positive outcomes for financial decisions. For example, media gurus advocate keeping emotions separated from financial matters (Barton, 2015; Palmer, 2014; Triffin, 2014). Bankers and other financial workers are portrayed in popular culture as “cold fish” who lack morals and empathy (Admati 2016; Lewis, 2010; Luyendijk, 2015). Such framing may promote the incompatibility of affective thinking and positive financial decisions (Park & Sela, 2017). Building upon academic research and a western social cultural view, in this paper financial decision-making is explained in terms of affective and analytical thinking, in which emotions are the main input for affective thinking and cognitive reasoning and logic are the main input for analytical thinking (Epstein et al., 1996).
2.2.2 Mechanistic dehumanization and analytical thinking
The question remains how mechanistic dehumanization relates to affective and
analytical thinking. In this paper it is argued that the dual process model of dehumanization is compatible with affective and analytical processing. Exposure to animalistic dehumanization stimuli, in which behavior represents desires and wants, could trigger affective thinking, whereas exposure to mechanistic dehumanization stimuli, associated with cold, analytical behavior, could trigger analytical thinking.
16 As Aristotle argued, humans are rational animals (Joachim & Rees, 1952). Humans and animals both experience affective states, which are based in the mammalian part of the brain (Panksepp, 2005). The difference lies in the ability of humans to better regulate their affective states (Panksepp, 2011). Because humans perceive themselves as more sophisticated than animals (LaLand & Brown, 2002), behavior that is dominated by desires and wants (affective states) are associated with animals. For instance, animal metaphors are used when someone disobeys the social rules (e.g., stupid cow) (Van Oudenhoven et al., 2008) or shows extreme emotions (e.g., lovebirds). In the media, education and religion, animals are linked to affective behavior. For example, Magnum’s tagline “release the beast”, in which consumers are encouraged to be more like intuitive animals and indulge with a Magnum ice-cream. Moreover, in Dutch education, children have to read the story van den vos Reynaerde, in which animals are used to represent non-rational behaviour of humans. Lastly, in religion it was the snake who seduces Eve to surrender to her desires. All these examples show that animals are associated with affective behavior.
In contrast, machines are associated with cold, analytical behavior. First, most machines are analytical in their essence (Lee, 2018). Machines function on algorithms, which is a computational formula that makes decisions automatically based on statistical models or rules. Although machines can look or behave like humans (Minato et al., 2006; Saygin et al., 2011; Hall et al., 2014), people perceive them as lacking intuition, subjective judgements capabilities (Lee, 2018) and experience (Gray, Gray & Wegner, 2007). This view is also found in business and media. Successful sales people have been compared to machines because they ignored their judgements and focused on reaching targets (Adamson et al., 2013). Nestle’s tagline “Feeling like a machine?” portrays people who are doing their jobs unemotionally and automatically, like machines. These examples show how machines are associated with cold, analytical entities.
17 Furthermore, people can be cued by the decision context, to which thinking style they apply (i.e., affective or analytical) (Park & Sela, 2017; Inbar et al, 2010; Novak & Hoffman, 2009). When animalistic dehumanization occurs which prompt schemas of irrational and lack of self-control behavior (Haslam, 2006; Weihrauch & Huang, 2017), people expect to behave accordingly and apply an affective thinking style. Likewise when mechanistic dehumanization occurs, which is associated with unemotional, cold and analytical behavior (Haslam, 2006), people expect to behave accordingly and apply an analytical thinking style.
2.2.3 Thinking style and achievability
The previous sections, argued that people feel expected to adopt a mechanistic to approach to decisions upon exposure to humans-as-machines, which makes them apply an analytical thinking style to saving. Adopting a mechanistic approach to saving, however, can be perceived as difficult. Many consumers struggle to control their emotions in financial decisions resulting in impulsive buying (Baumeister, 2002; Vohs & Faber, 2007; Achtziger et al., 2015). Consumers who perceive themselves as affective thinkers especially struggle with approaching saving analytical, because they already feel challenged controlling their emotions in such financial decisions. For example, these consumers emphasize the hedonic value of products (fun, pleasure) over the utilitarian (useful, practical), are more likely to give charitable donations (Hsee et al., 2015), and are more likely to have debts because they are easier involved in impulsive purchases (Achtziger et al., 2015).
Furthermore, affective thinkers perceive themselves as less capable of making proper financial decisions because they experience self-concept incongruity with the associated analytical and cold processing mode (Park & Sela, 2017; Novak & Hofmann, 2009; Schwarz & Bless, 1991). In other words, the more people perceive themselves as affective decision makers, the more they feel that financial decisions are not them. For this reason, these
18 consumers even start avoiding financial decisions (Park & Sela, 2017). Similar, in this paper it is expected that consumers who emphasize their feelings (i.e., affective thinkers) also find it difficult to approach their saving cold and analytical. This is because the mechanistic approach (e.g., cold, analytical) represents the behaviors to which affective thinkers cannot identify (Park & Sela, 2017). Similarly, consumers with low health self-control found adopting a mechanistic approach to food less achievable, because it represented the behaviors these consumers experienced difficulties with (Weihrauch & Huang, 2017). It even resulted in a backfire effect, causing these consumers to make even unhealthier food choices. Therefore, in this paper it is argued that although all consumers might expect to adopt a mechanistic approach to saving, those who are affective thinkers might consider this as impossible, because the mechanistic approach represents behaviors that do not fit them.
H2: Affective thinkers (vs. analytical) find adopting a mechanistic approach to saving less achievable, upon exposure to humans-as-machines.
2.3 Consumer financial decision-making
A lot of areas of decision-making could be interesting to look at when thinking of mechanistic dehumanization and decision-making style. As explained in the introduction, consumer financial decision-making was selected for two reasons: First, (visual) mechanistic dehumanization is becoming more and more prevalent (e.g., consumers having appointments with financial advisers through teleconferencing (Barbier et al., 2016; Eha, 2017; Dirican, 2015). Second, consumers often make emotional financial decisions, which can cause risks (Bernatzi & Thaler, 2007; Kuhnen & Knutson, 2011). Especially the latter makes it
interesting to understand whether the use of mechanistic dehumanization can push consumers toward an analytical thinking style.
19 2.3.1 Emotional biases in financial decision-making
Although people self-report that they can make proper financial decisions (Lin et al., 2016), research has shown that most consumers do the opposite. Consumers are likely to choose suboptimal contracts and fail to switch, even when mistakes are costly (Agarwal et al., 2015). Consumers do not take advantages of 401(k) matching contribution plans, which have clear financial gains (Choi, Laibson & Madrian, 2011) and they prefer not to be actively be involved in investing despite the chance of lower financial returns (van Rooij, Lusardi & Alessie, 2011). Scholars have been investgating the antecedents of this suboptimal financial behavior and have acknowledged that consumer financial decision-making (CFDM) can be influenced by many different concepts (Lynch, 2011). Cognitive capability (Agarwal & Mazumder, 2013), financial literacy (van Rooij, Lusardi & Alessie, 2011) and framing (Amar et al., 2011) are just a few aspects influencing the process.
Emotions also play a significant role in suboptimal financial decision-making (Lee & Andrade, 2011). Financial decisions often involve emotional aspects of someone’s life, such as saving for child’s education or achieving financial goals for a desired lifestyle (Jackson, Saffell, Fitzpatrick, 2018). Moreover, as many consumers lack financial knowledge (Lusardi 2008; Lusardi & Mitchell 2017), they are more likely to rely on their feelings (Clore, Schwarz & Conway, 1994). Emotions are thus present in financial decisions and come with the risk that consumers tend to pay less attention to relevant logical information (Gilovich, Griffin & Kahneman, 2002). In turn, this biased perspective is likely to lead people to suboptimal choices. For instance, people are known for making choices based on the justification of past invalid choices (i.e., sunk-cost trap). Investing money in a project that does not lead to financial returns, only because of the invested resources, is a non-optimal choice leading to a lot of risk (Hammond et al., 2006). Such emotional biases are not only common among consumers. Even financial professionals suffer from affective biases in financial
decision-20 making and the ones who can regulate their emotional levels perform better than the ones who cannot (Fenton-O’Creevy et al., 2010). Furthermore, emotions not only leads to costly errors that cause risk, but positive emotions also prompts people to take financial risks (Kuhnen & Knutson, 2011). For instance, people are more willing to take risky investments when they feel they are knowledgeable about the subject, regardless of their objective knowledge (Hadar et al., 2013).
Although emotions can clearly influence decisions, especially the financial ones (Jercic et al., 2012), affective processing does not always have to be negative when it comes to decision-making. Scholars also have shown that it can be beneficial. A problem with analytical thinking is that the working memory has its limits and too much complex information could lead to “analysis-paralysis” (Gigerenzer, 2008; Huang, 2017), hindering decision-making (Dane, Rockmann & Pratt, 2012). Some scholars even proved that better complex decisions were made in the absence of analytical thinking (Dijksterhuis et al., 2006). Nevertheless, in the case of CFDM, it is generally believed that an analytical approach is most beneficial, because financial decisions are prone to emotional biases (Milkman et al., 2009; Bazerman & Moore, 2008). Moreover, research that mentions the benefits of emotions in financial decisions often goes with the prerequisites that people can regulate their emotions (Fenton-O’Creevy et al., 2010) or have high expertise in the field (Dane et al., 2012).
In general, financial decisions are affectively biased, but in the financial industry this can be even more the case because the relation between the financial adviser and customer can also affect the decisions. Due to growing consumer wealth and choice in financial products, financial decisions have become more complex (White, 2005). Most consumers who own mutual funds therefore turn to a financial adviser for help (Foerster et al., 2017). The relationship the adviser and customer have can be more important for the financial decisions than risk or return characteristics (Monti et al., 2014). The quality of the
21 relationship is built on trust (Inderst & Ottaviani, 2012), but trust can be established by simple cues such as a smile and type of clothes (Monti et al., 2014). If there is a lack of trust,
consumers often search for a second opinion. However, when trust is present this behavior decreases, resulting in consumers choosing (expensive) recommended options that might not be the most effective ones (Schwartz, Luce & Ariely, 2011). Taking this together, many scholars suggest pushing people toward analytical thinking, to lessen the chance of costly errors (Milkman, Chugh, & Bazerman, 2009; Bazerman & Moore, 2008).
From the perspective that emotions bias perceptions and cause risks, it is interesting to examine whether exposure to mechanistic dehumanization stimuli can push people toward an analytical thinking style, reducing errors from being made (Milkman, Chugh, & Bazerman, 2009; Bazerman & Moore, 2008). The next section discusses saving and impulsive purchases as consumers’ financial decisions and explains how this can be regarded as analytical and affective behavior.
2.4 Saving
To examine the effect of mechanistic dehumanization on consumer financial decision-making, this paper looks at saving behavior. Saving is a typical financial decision (Park & Sela, 2017) and has often been the subject of financial decision-making research (e.g., Ülkümen & Cheema, 2011; Soman & Zhao, 2011; Hershfield et al., 2011; Haws et al., 2012). Furthermore, saving is considered as rational behavior (Hall, 1978). Scholars perceive the decision to save as the dilemma between “I should” and “I want”. People who save, choose the “I should” option and emphasize their analytical thinking style (Hsee et al., 2015). Therefore it is interesting to examine whether people who are mechanistically dehumanized show increased saving because they are triggered to apply an analytical thinking style. Understanding this effect would not only help consumers ensure a financially sound future, but would also be of interest to banks
22 and government, to support saving rates and economic growth and stability (Ülkümen & Cheema, 2011).
2.4.1 Rational saving
For almost the last 20 years Americans saved on average less than 5% of their yearly income (Federal Reserve Bank of St. Louis, 2018) and reported having difficulties saving money on a regular basis (Dholakia et al., 2016). This is alarming, because the amount people save for the future influences their well-being (Thaler, 1994). Although most people know that they should save to build personal wealth and ensure a prosperous retirement, they are still temped to spend money (Fudenberg & Levine, 2006). The question then arises why consumers fail to save at sufficient rates even if they are aware of its importance and benefits.
The most popular view on saving is as goal-directed behavior. Goals play an important role in spending and saving decisions (Thaler, 1988; Soman & Cheema, 2004). To reach a financial state, such as buying a house or paying for child’s education, people formulate saving goals (Soman & Zhao, 2011). It helps make the trade-off between immediate benefits or delayed gratification (Frederick, 2002; Laibson, 1997). When people fail to reach their saving goals, it is often due to a lack of rational thinking. For example, people expect that saving is an “all-or-nothing” situation (Dholakia et al., 2016), do not have enough self-control (Baumeister, 2002) and are very optimistic about their ability to save more money in the future, which is often not true (Tam & Dholakia, 2011). Moreover, saving can benefit from mechanisms that help reaching optimal saving decisions (Bernartzi & Thaler, 2004; Botti & Iyengar, 2006). For instance, interventions such as feeling powerful (Garbinsky et al., 2014) and picturing one’s older self (Hershfield et al., 2011) or future relevant outcomes (Haws et al., 2012; Nenkov et al., 2008) all increase saving. These studies show that a psychological shift in people’s mindset can lead to increased saving.
23 Another factor that influences saving is a person’s thinking style. As mentioned, saving decisions are a dilemma of what people want to do and what they should do (Hsee et al., 2015), or the trade-off between delayed gratification and immediate benefits (Frederick Frederick, Loewenstein & O'donoghue, 2002; Laibson 1997). “Want” refers to affective states whereas “should” refers to analytical reasons. People who perceive themselves as analytical thinkers emphasize the “should” over the “want” (Bazerman, Tenbrusel & Wade-Benzoni, 1998). This means that they are less likely to base their financial decisions on their feelings and as a result, save more of their income for retirement (Hsee et al., 2015).
The opposite of rational saving is impulsive purchasing. It is regarded as a form of impulsive behavior, which is a consequence of an unregulated, unplanned and spontaneous need. In the case of an impulsive purchase, it includes getting a sudden need to buy something without having the intention to, followed by acting on it and not considering its consequences for one’s goals, such as saving (Baumeister, 2002). Impulsive purchases are accompanied by feelings of pleasure and excitement (Rook, 1987) and the surrendering to it, has been considered as a lack of rationalization (Baumeister, 2002). Failing to resist impulsive purchases occurs mostly due to a lack of self-control. In turn, a breakdown in self-control is often caused by emotional distress. This can be explained by the general need to feel good. If people do not feel good, their priorities shift toward restoring the good feeling. As a consequence, other long-term goals, such as saving money, lose priority (Baumeister, 2002).
Overall, emphasizing analytical thinking leads to saving (Hsee et al., 2015), whereas emphasizing affective thinking leads to biases (Dholakia et al., 2016; Tam & Dholakia, 2011) and increased tendency to deplete self-control (i.e., impulsive purchases) (Baumeister, 2002). As proposed, people who are exposed to humans-as-machines are expected to adopt a mechanistic approach to saving because they are pushed toward an analytical thinking style. People who apply an analytical thinking style save more (Hsee et al., 2015). To measure the
24 effect of a mechanistic expectation on consumer financial decision-making, saving therefore serves as a good measure. This leads to the third hypothesis:
H3: An expectation to adopt a mechanistic approach to saving increases saving, upon exposure to humans-as-machines.
As explained in Section 2.2.3, however, thinking style is likely to negatively influence the relation between mechanistic expectation and saving, because affective thinkers consider the mechanistic approach as less achievable (Weihrauch & Huang, 2017). Following this logic, in this paper it is proposed that when people are exposed to mechanistic dehumanization stimuli, they feel to expect to adopt a mechanistic approach to saving (i.e., unemotional, cold, analytical), which feels not achievable and as a consequence show decreased saving. This leads to the final and most crucial hypothesis:
H4: Consumers with an affective thinking style show decreased saving, upon exposure to humans-as-machines.
25 2.5 Conceptual model
Based on the hypotheses in the previous sections, the following conceptual model is developed.
Figure 1. Conceptual model of the paper.
Mechanistic Dehumanization Stimuli Expectation to Adopt a Mechanistic Approach to Saving Saving Thinking Style H1 H3 H4 Achievability to Adopt a Mechanistic Approach to Saving (H2)
26 3. Data & Method
This chapter provides the empirical part of the study. The study tested the four hypotheses on mechanistic stimuli, mechanistic expectation, thinking style and saving. First, the measures and manipulations of the variables are explained. Second, the characteristics of the operationalization will be discussed.
3.1 Data collection
A total of 2911 US resident adults (Mage = 35, 38.5% female) were recruited from the Amazon MTurk platform and offered $0.30 compensation to fill in the survey. Scholars have shown that MTurk is more representative of the general population than traditional samples (Buhrmester, Kwang & Gosling, 2011) and reliable for consumer research (Goodman, Cryder & Cheema, 2013). An online study was chosen because it is a cost- and time-efficient way to collect data, in which participants have the flexibility to complete the survey when it suits them. Furthermore, online surveys can be distributed anonymous and the total time of each respondent can be measured (Evans & Mathur, 2005).
3.2. Stimulus design and pretest
Mechanistic dehumanization was manipulated through altering a body’s external appearance/shape and physical movement (Weihrauch & Huang, 2017; Aggarwal & McGill, 2011; Graham & Poulin-Dubois, 1999; Morewedge, Preston & Wegner, 2007). Specifically, a virtual telepresence (VT) machine and a humanoid were used to design the stimuli. As mentioned in the literature review, the first is gaining popularity in consumers’ daily lives and business interactions and the second is getting more frequently used in smart branches. In the
1 There was asked for 300 participants, but Qualtrics documented 291 completed surveys when the data was downloaded.
27 first mechanistic dehumanization condition, the financial adviser’s appearance was illustrated as a robotic skeleton and a human face, just as seen in virtual telepresence machines. In the second mechanistic dehumanization condition, the financial adviser’s appearance was illustrated as a humanoid. In the human condition, the financial adviser’s appearance was illustrated in its regular human form (see Appendix A). To incorporate the dimension of physical movement, participants were then showed a video clip of the financial adviser walking (corresponding to their condition). In the human condition, the movement was fluent/smooth. In the dehumanized conditions, the movement was mechanistic/static (Tremoulet & Feldman, 2000).
For the pretest, participants (N = 38, Mage = 24, 84.6% female) were recruited through email, Whatsapp and face to face. Participants viewed one of the stimuli and rated three 7-point Likert scales adopted from the anthropomorphism literature (Aggarwal & McGill, 2007; Romero & Craig, 2017) to measure the perceived level of mechanistic dehumanization: “ The financial adviser… 1 = looks like a machine, 7 = looks like a human; 1 = does not look alive
at all, 7 = looks very alive; 1 = contains mainly machine-like features, 7 = contains mainly human-like features” (Cronbach’s alpha = .88). The pretest was successful. The VT (M =
2.08, SD = .95) and humanoid (M = 1.67, SD = .65) stimuli were perceived as more mechanistically dehumanized than the human stimulus (M = 5.51, SD = 1.34; F(2,36) = 30.62, p = .000, η2 = .741). The post-hoc test (Tukey HSD) revealed that the VT and humanoid condition were equally perceived as mechanistically dehumanized (p = .549).
3.3 Procedure
The study used a 3 stimuli (2 x mechanistic dehumanization vs. human) x mechanistic approach (measured as a continuous variable) between subject design. In the main study, participants were told to imagine that they received a mail from their bank that they would
28 switch financial adviser. Attached they could find (visual) details about their new financial adviser. Next, they were randomly assigned to one of three conditions (VT, humanoid or human adviser) and saw the related picture of their adviser for 15 seconds and watched the 30-second clip of the agent walking.
Participants then went through a filler survey, in which they evaluated different visuals. The filler task was included to minimize the possibility of a demand effect. Afterward, mechanistic expectation and saving were measured in random order. Participants reported their perceived expectation of adopting a mechanistic approach to saving at this moment on six 7-point Likert scales (1= strongly disagree, 7 = strongly agree : “I feel that I am expected to make my saving choices… unemotional, analytical, cold, rigid (not changing decision rules for a specific context), passive (follow pre-set rules instead of my current intent), and superficial (not doing deep analysis of the specific decision context before making choice). These six dimensions were adopted from the definition of mechanistic dehumanization (Haslam, 2006). In addition, participants reported their perceived achievability of adopting a mechanistic approach to saving at this moment on six 7-point Likert scales (“I feel that it is achievable for me to make my saving choices… unemotional, analytical, cold, rigid, passive, and superficial.” 1 = strongly disagree, 7 = strongly agree. These questions were adopted from Weihrauch and Huang (2017).
Three constructs were used to measure the dependent variable saving. First, all participants had to fill in a money allocation task, which was adopted from Hershfield et al. (2011). They were told that they unexpectedly received $1000,- and were asked to allocate the money among four options: “Use it to buy something nice for someone special”, “Invest in a retirement fund”, “Plan a fun and extravagant occasion” and “Put it into a checking account”. The sum of each option had to be 1000. The retirement and checking account options were used as measures for saving behavior, because they were both perceived as longer in time before one
29 can spend the allocated money than the two other impulsive options (Rudzinska-Wojciechowska, 2017) and they opt for later rewards (Hershfield et al. 2011). The second construct measured saving through a save and spend task, which was created for the purpose of this study. Participants were asked to choose between three scenarios in the following sequential order: “A city trip worth $150” or “Save $192 to your savings account”; “A restaurant coupon worth $70” or “Save $90 to your savings account”; “A movie coupon worth $20” or “Save $25 to your savings account”. The last construct measured saving behavior through a self-reported saving intention scale adjusted from Azjen (2011). Participants reported their saving intentions at this moment on three 7-point Likert scales in the following sequential order (“At this moment, I feel that I should save more.” 1 = strongly disagree, 7 = strongly
agree, “I am intending to save more. “ 1 = very unlikely, 7 = very likely. I am confident that I
can save more money. 1 = strongly disagree, 7 = strongly agree.)
Furthermore, participants filled in their perceived thinking style, which was adopted from the lay rationalism scale (“When making decisions, I focus on objective facts rather than subjective feelings.” 1 = strongly disagree, 7 = strongly agree) (Hsee et al., 2015). The lay rationalism scale measures people’s self-beliefs about making decisions based on
rational/analytical thinking or emotions and feelings. Before exiting the study, they entered control variables such as age, gender and monthly income after paying fixed costs, because these are common predictors of saving (Fernandes et al., 2014; Dholakia et al., 2016). Finally, participants reported any suspicion or question they might have about the survey.
30 4. Results
This chapter discusses the results of the analyses of the data in SPSS. First, a detailed examination of the descriptive and frequency statistics is provided. Second, the reliability of the measures are analyzed. Finally, the hypotheses are tested in two parts. The first part tests the hypotheses for the VT (vs. human) stimulus. The second part tests the hypotheses for the humanoid (vs. human) stimulus.
4.1 Descriptive and frequencies statistics
291 participants participated in the experiment, whereas 13 participants were excluded from further analyses due to outliers in the variables age, income, response time (SD > 3) or answer patterns (leverage, mahalanobis distance), which could bias regression. One participant was excluded due selecting the “other” gender option, making it not possible to equally compare it to the other genders. 277 participants were assigned to one of three conditions to manipulate mechanistic dehumanization: Human (N = 94, reference group), Virtual Telepresence (N = 89, mechanistic dehumanization stimulus) and Humanoid (N = 94, mechanistic dehumanization stimulus). Accordingly, there was no missing data and none of the participants revealed concerns or suspicion about the study.
The pretest showed no difference between the two mechanistic dehumanization stimuli (VT & humanoid). Therefore, it was decided to perform the statistical analyses with the VT condition and execute a second set of analyses with the humanoid condition to generalize the findings. Subsequent two dummy variables were created for the VT condition (0 = Human, 1 =
VT) and the Humanoid condition (0 = Human, 1 = Humanoid) with the Human condition
serving as a reference group in both conditions.
Analyzing the money allocation task, the option “Invest in a retirement fund” was taken to measure actual savings in US$ (M = 334.72, SD = 263.90, range: 0-1000). Beforehand, it was decided to use both “invest in a retirement fund” and “put money into a checking account”
31 as items to measure saving. However, analyzing the results it was displayed that the two items had a strong negative correlation (r = -.548, p = .000), indicating that they do not measure the same. Hence, retirement was chosen because it was perceived as longer in time before one can spend the allocated money than checking account (Rudzinska-Wojciechowska, 2017), thus opting more for later rewards. Furthermore, in the original experiment the retirement option was used to measure saving (Hershfield et al., 2011). Next, dummy variables were created for the second DV measure “scenarios” (0 = spend, 1 = save) and the mean of the three scenarios was calculated (M = .81, SD = .26) to treat it as a continuous measure. However, while “citytrip” was small positive correlated with “restaurant” (r = .261, p = .000) and “movie” (r = .168, p = .005), there was no correlation between “movie” and “restaurant” (r = .084, p = .161), indicating that there is no relationship between those variables. Furthermore, virtual inspection of a scatterplot, normal probability plot and a plot of standardized residuals vs. standardized predicted values showed no linearity, normal distribution, nor homoscedasticity. From this, it was concluded that the items were not related nor suited for a multiple regression and only “citytrip” was used to measure saving, making it a dichotomous dependent variable (M = .81,
SD = .39). Citytrip was chosen because it was the first in order, not being affected by nonrandom
measurement error (Lavrakas, 2008) and it was the only item correlated with both other items. Finally, the mean was calculated for the third measure, “saving intention” scale (M = 5.77, SD = .98) and for the DV of H2, “mechanistic achievability” (M = 4.39, SD = 1.06).
The mediator “expected mechanistic approach to saving” (M = 4.34, SD = 1.02) had a normal distribution. The moderator “thinking style” (M = 5.46, SD = 1.05) displayed a negatively skewed distribution of -.76 (SE = 1.45), which is still considered as an acceptable value (George & Mallery, 2010). However, it indicates that participants were mostly analytical thinkers. Furthermore, the mean age of participants was 35 (SD = 10.93), in which 62.5% of
32 the participants were male and had a mean income after paying fixed costs of $842.64 (SD = 1019.78). The standard deviation displays are large variety in income.
4.2 Reliability analysis for scales
A reliability check for the “expectation to adopt a mechanistic approach to saving” scale was executed to measure internal consistency, resulting in α = .706 for 6 items. The Cronbach’s Alpha could slightly increase to α = .726 if “I feel that I am expected to make my saving choices
analytical” was deleted. However, this is a crucial part of the research. It was decided to keep
all the six items because it was a validated scale (Haslam, 2006), the difference was not larger than .10 and a Cronbach’s alpha above 7 is sufficient (Tavakol & Dennick, 2011).
Next, a reliability check for the “achievability to adopt a mechanistic approach to saving” scale was executed to measure internal consistency, resulting in α = .733 for 6 items. The Cronbach’s Alpha could slightly increase to α = .746 if “I feel that I am expected to make my saving choices analytical” was deleted. However, this was perceived as not necessary as explained above. Lastly, the reliability of the saving intention scale was measured, resulting in a α = .741 for 3 items. Again the scale could make an increase to α = .790 when “I am confident that I can save more money. However as aforementioned, there was no statistical need to delete this item.
Prior to testing the hypotheses, a correlation analysis was conducted to check whether variables were correlated (see Table 1). Thinking style was both small positive correlated with mechanistic expectation (r = .192, p = .001) and achievability (r = .213, p = .000). Thinking style was furthermore small positive correlated with retirement savings (r = .122, p = .042) and saving intention (r = .205, p = .001). Mechanistic expectation and achievability were strong positive correlated (r = .769, p = .000), indicating that if one expects to adopt a mechanistic approach to saving, one often finds such an approach also achievable. Mechanistic expectation
33 was small negative correlated with citytrip (r = -.142, p = .018), indicating that people who are high on mechanistic expectation can also show less saving. Lastly, mechanistic expectation was small positive correlated with saving intention (r = .139, p = .020). Age was small negative correlated to mechanistic expectation (r = -.144, p = .017) and achievability (r = -.151, p = .000), indicating that older people more often show a lower mechanistic expectation or achievability than younger people. Age was small positive correlated to citytrip (r = .143, p = .017). Finally, gender was small negative correlated with thinking style (r = -.191, p = .001), indicating that females are more often affective thinkers. Although gender (r = .136, p = .024) and income (r = .180, p = .003) were only small positive correlated with retirement savings and age only with small positive with citytrip (r = .143, p = .017), they were still used as control variables because previous research showed that they are important variables to consider (Fernandes et al., 2014; Dholakia et al., 2016).
34
Table 1
Means, Standard Deviations, Correlations
Variables M SD 1 2 3 4 5 6 7 8 9 10 11 1. Gender .62 .49 - 2. Age 34.68 10.93 .05 - 3. Income 842.64 1019.64 -.10 .06 - 4. Virtual Telepresence - - -.08 -.09 .03 - 5. Humanoid - - -.02 .09 .05 - - 6. Mechanistic expectation 4.34 1.02 .00 -.14* -.00 .09 .09 (.71) 7. Mechanistic achievability 4.39 1.06 -.02 -.15* .02 .10 .09 .77** (.73) 8. Thinking style 5.46 1.05 -.20** -.04 .01 .01 .07 .19** .21** - 9. Retirement savings 334.72 263.72 -.14* .05 .18** -.04 -.01 -.06 -.10 .12* - 10. Citytrip .81 .39 .00 .14* .02 -.02 .04 -.14* -.12 .10 0.11 - 11. Saving intention 5.77 .98 -.09 -.01 .02 -.02 -.02 .14* .16** .21** 0.08 .13* (.79)
**. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
35 4.3 Hypotheses testing virtual telepresence
For the dependent variables money allocation and saving intention the assumptions were met for a linear regression. The assumptions of a logistic regression for the dependent variable citytrip were also met.
Next, a regression analysis was conducted to measure participants’ retirement savings using stimulus as a predictor and gender, age and income as control variables. A non significant regression was found, F(4, 178) = 1.44, p = .223 with R² = .03, indicating that the model was not able to predict saving better than the mean of retirement savings. In addition, exposure to VT did not make a difference on retirement savings, β = -.07, t = -1.14, p = .256. A second regression analysis was conducted to measure participants’ saving intention, using stimulus as predictor and gender, age, and income as control variables. There was again no significant regression found, F(4, 178) = .217, p = .929 with R² = .005. Exposure to VT makes no difference on saving intention, β = -.02, t = -.236, p = .814. Lastly, a logistic regression was performed to measure the likelihood that participants would save instead of going on a citytrip, using stimulus as a predictor and gender, age and income as control variables. The full model containing all predictors was non significant, χ2 (4, N = 183) = 8. 33, p = .080, indicating that the model was
not able to distinguish between respondents who saved and who did not. The model explained 7% (Nagelkerke R²) of the variance in saving. The Wald criterion showed that VT was not a significant predictor, β = .11, χ2 (4, N = 183) = .08, p = .772. Concluding, there is no direct
effect of exposure to humans-as-machines on saving.
In order to analyze the hypothesized mediation effect (H1 & H3), the PROCESS Procedure for SPSS, written by Andrew F. Hayes, was executed. A bootstrap sample equal to 1000 was used in combination with a 95% confidence level for confidence intervals. The PROCESS analysis was executed on the three dependent variables, controlling for age, gender and income. A significant regression was found, F(4, 178) = 2.91, p = .023 with R² = .06. Results did not reveal the first hypothesized effect. Participants exposed to mechanistic
36 dehumanization stimulus did not feel a greater sense of being expected to adopt a mechanistic approach to saving than participants exposed to a human, b = .14, t(178) = .96, p = .339. The control variable age did make a difference on participants mechanistic expectation, b = -.02,
t(178) = -2.98, p = .003. The older participants were, the less they felt expected to adopt a
mechanistic approach to saving. The hypothesis that exposure to humans-as-machines stimuli leads to an expectation that one should adopt a mechanistic approach to saving (H1) cannot be supported.
Next, there was a non significant regression found for mechanistic expectation on retirement savings, F(5, 177) = 1.14, p = .339 with R² = .03. There is insufficient evidence that the model explains the variation in retirement savings better than its mean. Participants with higher expectations to adopt a mechanistic approach to saving did not show increased retirement savings, b = -.64, t(177) = -.03, p = .973. Subsequent, a non significant regression analysis was found for mechanistic expectation on saving intention, F(5, 177) = .68, p = .642 with R² = .02. Again, there is insufficient evidence that the model explains the variation in saving intention better than its mean. Participants with higher expectations to adopt a mechanistic approach to saving did not show a increased saving intention, b = .12, t(177) = 1.58, p = .115. Lastly, there was a significant logistic regression found for mechanistic expectation on citytrip, χ2 (5, N = 183) = 11.34, p = .045, indicating that the model was able to
distinguish between participants who saved and who did not. The model explained 9% (Nagelkerke R²) of the variance in saving. Mechanistic expectation was not a significant predictor, β = -.34, χ2 (5, N = 183) = 2.89, p = .089. Again, age made a significant contribution
to the prediction (p = .029). The odds ratio indicated that when age is raised by one unit (one year) participants were 1.04 times more likely to save instead of going on a citytrip. Concluding having a heightened sense of adopting a mechanistic approach to saving does not increase saving. Hypothesis 3 cannot be supported.
37 Subsequent it was tested if affective thinkers (vs. analytical) find adopting a mechanistic approach to saving less achievable, upon exposure to humans-as-machines (H2). A regression analysis on the perceived achievability of adopting a mechanistic approach to saving was conducted, using stimulus, thinking style (mean-centered) and their interaction as predictors, with gender, age and income as control variables. A significant regression was found, F(6, 176) = 2.88, p = .010 with R² = .09. The main hypothesized main effect of thinking style on achievability was not found. Affective thinkers did not perceive the mechanistic approach as less achievable than analytical thinkers, β = .10, t = .92, p = .359. The stimulus also did not affect participants perceived achievability, β = .09, t = 1.19, p = .237, nor was the interaction between stimulus and thinking style significant, β = .09, t = .74, p = .440. Age was a significant predictor of achievability. The older participants were the less achievable they perceived a mechanistic approach to saving, β = -.21, t = -2.85, p = .005. Concluding, exposure to humans-as-machines does not influence the perceived mechanistic achievability, regardless of thinking style. Hypothesis 2 cannot be supported. Further exploratory analysis did show a significant regression (F(4, 272) = 5.01, p = .001 with R² = .07) on the direct effect between thinking style and achievability controlling for age, gender and income, β = .22, t = 3.64, p = .000. Meaning that affective thinkers perceived mechanistic traits (e.g., unemotional, cold, analytical) less achievable. The implications of this will be discussed in Chapter 5.
In order to analyze the full conceptual model (H4), a bias corrected moderated mediation PROCESS Procedure for SPSS, written by Andrew F. Hayes, was executed (model 14). A bootstrap sample equal to 1000 was used in combination with a 95% confidence level for confidence intervals. The PROCESS analyses was executed on the three dependent varaibles, controlling for age, gender and income. A significant regression was found, F(7, 175) = 2.10,
p = .046 with R² = .08 for the DV retirement savings. Results did not reveal the hypothesized
38 thinking style, b = 9.16, t(175) = .50, p = .619. In addition, there was a conditional effect of thinking style on retirement savings, b = 51.45, t(175) = 2.95, p = .004. Analytical thinkers saved more for their retirement, when mechanistic expectation was 0. The control variable income was a significant predictor of retirement savings. The higher the income the more participants saved for their retirement, b = .04, t(175) = 2.01, p = .046. However, the effect of income and thinking style on retirement savings are low, due the small regression coefficient of income and the overall variance explained by the model. Due to lacking significant interaction, there was no need to continue the analysis, because no moderation took place.
Next, a non significant regression was found for the DV saving intention, F(7, 175) = .92, p = .500 with R² = .04. The results showed no significant interaction between mechanistic expectation and thinking style, b = -.05, t(175) = -.72, p = .473. There was no further need to continue the analysis. Finally, there was a non significant logistic regression found for mechanistic expectation on citytrip, χ2(7, N = 183) = 13.03, p = .071, indicating that the model
was not able to distinguish between participants who saved and who did not. The model explained 11% (Nagelkerke R²) of the variance in saving and there was no significant interaction , β = .04, χ2 (7, N = 183) = .04, p = .958. The hypothesis that consumers with an
affective thinking style (vs. affective) show decreased saving, upon exposure to mechanistic dehumanization stimuli, cannot be supported.
4.4 Hypotheses testing humanoid
After testing the hypotheses for the VT stimulus, the same hypotheses were tested for the humanoid stimulus for generalizability. The humanoid analyses showed almost no notable different results than the VT. However, although it did not pertain to the main focus of the research, it should be noted that for the moderated mediation a significant regression was found (F(7,180) = 1.60, p = .006), in which thinking style predicted saving intention (b = .26, t(180)
39 = 3.88, p = .000), when mechanistic expectation was 0. The model explained 10% (Nagelkerke
R²) of the variance in saving. Moreover, conditional effects were found for the moderated
mediation on citytrip, χ2(7, N = 188) = 16.76, p = .019. The model explained 14% (Nagelkerke R²) of the variance in saving. A conditional effect was displayed between mechanistic
expectation and citytrip when thinking style was 0 (χ2(7, N = 188) = 6.26, p = .012). The Exp
(B) indicated that when mechanistic expectation is raised by one unit, participants were .65 times more likely to save the money than spend it. In addition, a conditional effect was also displayed between thinking style and citytrip when mechanistic expectation was 0 (χ2(7, N =
188) = 4.75, p = .002). The Exp (B) indicated that when thinking style is raised by one unit and mechanistic expectation is constant, participants were 1.39 times more likely to save the money than spend it. Although mechanistic expectation affected saving on citytrip, it is hard to draw conclusions from it because it is only displayed in one non-validated item. Finally, with respect to control variables, gender predicted mechanistic achievability, β = -.15, t = -2.03, p = .044. Indicating that females perceive a mechanistic approach less achievable. For the ease of reporting, the remaining results can be found in Appendix B (Table 2-8). The full conceptual model can be found in Figure 2, 3 and 4.
