Pandora’s Box?
A Structural Equation Model of Loot Box Monetization and Its Ethical Implications
by Leonard Delank
(Student Number Groningen: S4077237 / Newcastle: 190575865)
Programme: DDM - Advanced International Business Management & Marketing
Module:
EBM719A25: Master’s Thesis International Business and Management NBS8199: Dual Award Dissertation
Supervisors:
Dr. H.U. Hammad Haq (University of Groningen) Prof. Jonathan Sapsed (Newcastle University Business School)
Submitted on 7th December 2020
Abstract
Over the past decades, the psychological and social phenomenon of video games has attracted a global audience. The rapid growth of this industry has facilitated the
exploration of novel business models, which capitalize on additional content sold within games. One of which are loot boxes. These virtual items, which contain randomized in-game content, are mostly bought for real-world money and have become a significant revenue stream for the video games industry. Yet, this type of monetization has drawn criticism from the public, scholars and legislators, due to its mechanistic similarities to traditional gambling activities. Still, games incorporating the mechanic are marketed towards children and remain unregulated in most international markets. To date the limited academic research on the matter has focused on separately linking individual factors like problem-gambling and gaming addiction to loot box purchases. Thereby, the combined effects of them, other relevant constructs like flow, cognitive dissonance, perceived behavioural control and their ethical implications were overlooked. For the purpose of assessing the holistic impact of these constructs on the engagement with loot boxes and their post-purchase manifestations, a structural equation model (SEM) was synthesized. Subsequently, this model was analysed using self-reported, cross-national data from players, which have purchased loot boxes (N = 202). The results of the path analysis showed that the clinical conditions of problem-gambling and gaming addiction were predictive of increased cognitive dissonance. Flow experienced by playing the game, on the other hand, lowered the mental distress inflicted through cognitive dissonance. Problem-gambling was further established to negatively impact perceived behavioural control. No significant effects of flow and gaming addiction on perceived behavioural control were detected. Under Kant’s categorical imperative, these findings suggest that the current implementation of loot boxes is ethically highly questionable as they violate the autonomy and dignity of individuals. Recommendations on how to bring loot boxes in line with ethical business conduct and how to regulate them in the future are made.
Acknowledgements
I would first like to express my gratitude to Dr. Hammad Haq and Prof. Jonathan Sapsed, who have provided me with their guidance and invaluable knowledge throughout the process of writing this thesis. Their feedback and advice helped me further refine my work to significantly improve the end result.
Furthermore, I would like to thank Dr. Raymond Zaal who introduced me to the field of business ethics and expressed interest in the loot box subject matter at an early stage of my studies. I am also highly appreciative of the support from Dr. Paulus Aditjandra who provided me with the research tools necessary and was very supportive of my topic choice. The affirmations of both lead me to further pursue this route.
Table of Contents
1. Introduction 6
2. Definition and Operation of Loot Boxes 11
3. Status Quo of Regulation
3.1 International Legal Debate 14
3.2 Rating Boards and Self-Regulation 18 4. Theoretical Model & Research Hypotheses 20
4.1 Cognitive Dissonance 24
4.2 Perceived Behavioural Control 26
4.3 Problem-Gambling 27
4.4 Gaming Addiction 29
4.5 Flow 31
4.6 The Model 34
5. Methodology 34
5.1 Data Collection & Sample Characteristics 35
5.2 Operational Measures 38
5.2.1 Gaming Addiction Scale 38
5.2.2 Problem-Gambling Severity Index 39 5.2.3 Perceived Behavioural Control 39
5.2.4 Flow 40
5.2.5 Cognitive Dissonance 40
5.3 Data Analysis 40
7. Discussion 48
8. Implications 51
9. Limitations & Future Research 54
10. Conclusion 55
11. References 57
12. Appendix
Appendix 1: List of games 83
Appendix 2: List of participants’ nationalities 84
Appendix 3: Questionnaire in entirety 85
Appendix 4: EFA results for flow and PBC 104
Appendix 5: List of alternative models 106
Tables & Figures
Figure 2.1: Exemplary visual representations of loot boxes 12 Figure 3.2.1: A selection of content descriptors 19
Figure 4.6.1: Theoretical model 34
Table 5.1.1: Composition of sample 37
Table 6.1: Results of initial reliability check 42 Table 6.2: Summary statistics and Shapiro Wilk test results 44
Table 6.3: Measurement model parameters 45
Table 6.4: Correlation matrix and Cronbach’s alpha 46
Table 6.5: Path analysis results 46
1. Introduction
In recent years video games have become increasingly sophisticated and bigger in the scope of their content. This has led them to become a global phenomenon and an entertainment medium that is projected to be enjoyed by 2.6 billion individuals
worldwide in 2020 (Statista, 2017). Due to this major global audience, the corporations developing, and marketing video game titles became part of an influential industry. Within a 60-year span, small passion projects carried out by a singular person without commercial intent (American Physical Society, 2008) have become the products of listed, multinational corporations, which are expected to cumulatively generate $159.3 billion in global revenue (Field Level Media, 2020).
The emergence of such a vast international market has prompted managers to consider novel strategies and business models to capitalize on it (Marchand and Hennig-Thurau, 2013). While traditional business models provided the consumer with an unlimited time of play for a fixed price, alternative ways to derive economic value have been explored by corporations (ibid.). In recent years, the marketing of virtual goods has become a focal point in the value creation of the video games industry. Nowadays, virtual goods are offered via in-game storefronts in addition to the main game. In the advent of these types of monetization, it has come into question whether the ever-growing revenues within the industry are solely driven by an increasing amount of engaged international consumers or if they potentially originate from the business practices having become more exploitative (Lizardi, 2012).
Reddit history1, as Electronic Arts (EA)2 representatives tried to defend the
implementation of the monetization by stating that the mechanic intended to “provide players with a sense of pride and accomplishment for unlocking different heroes” (EACommunityTeam, 2017). The title, aiming to capitalize on the popular Star Wars trademark, thereby, prompted not only outrage within the gaming scene but also raised the attention of legislators and scholars alike (Lewis, 2017).
However, loot boxes are not an entirely new phenomenon and existed prior to the 2017 incident. Nor, was it their last implementation. The first utilization of loot boxes is reported to have taken place in 2004 (Kyu-man, 2016). Yet, the mechanic only enjoyed broad adoption a few years later. A recent study by Zendle, Meyer, Cairns, Waters, and Ballou (2019) found that 36% of the 50 most-played games on Steam, the world’s largest digital distribution platform for PC titles (Edwards, 2013), contained loot boxes. Current research suggests that these mechanics radically increase revenue for video game developers and publishers at the psychological expense of consumers (Drummond and Sauer, 2018). What makes this type of monetization problematic, is its structural similarities to traditional gambling activities. These mechanistic similarities
predominantly lie in the exchange of money for rewards based on chance (ibid.). Yet, to date, no national authority has classified loot boxes as gambling and regulated them as such, with the exception of the Belgian government.
Ever since the Star Wars Battlefront 2 incident, the practice has been met with much criticism from public figures in the gaming space. Many have proclaimed that this form of virtual good is unethical, yet only few have substantiated their claims with actual ethical theory. Interestingly enough the bespoke vocal criticism of consumers has seemingly not translated into actual purchasing behaviour. For example, the revenue
1 With about 1.5 billion unique monthly visitors, Reddit is one of the largest social networks worldwide (Statista, 2020). It gives consumers the opportunity to upload content in either visual or text form and to communicate with one another. Via up- and downvotes users can express their (dis-)pleasure of the published content. The comment issued by EA’s official account in an answer to a post saying
“Seriously? I paid 80$ to have Vader locked?”, garnered 667.827 downvotes and earned the title “Most downvoted comment on Reddit” awarded by Guinness World Records (Leskin, 2019).
generated by FIFA’s Ultimate Team mode, which heavily relies upon the purchase of loot boxes, regularly surpasses that of physical copies of the actual FIFA game. Loot box purchases are estimated to have accounted for $30 billion in revenue worldwide (Juniper Research, 2018). This leads to the question of why these attitudes towards these mechanics have not negatively impacted their profitability. Recent research has suggested that it may be due to the mechanic preying on addictive tendencies and, thereby, locking people into its system. Hence, one might wonder if the video games industry has unleashed a digital pandora’s box.
Due to the relative novelty of this digital business model, little is known about the phenomenon, and research on it is considered to be “still in its infancy” (Kristiansen and Severin, 2020; Drummond et al., 2020). Given the rising importance of the sector as a whole, virtual goods, and specifically loot boxes, further investigation concerning the factors of within-game purchase decisions is warranted (Marchand and Hennig-Thurau, 2013). To date, the majority of it has focused on linking the psychological condition of problem-gambling to engagement with loot boxes. Most of the bespoke research has established a clear link between the addiction to traditional gambling activity and the purchase of loot boxes. Studies by Drummond and Sauer (2018), Zendle and Cairns, (2018, 2019) Zendle, Meyer, and Over (2019) have all concluded that loot box purchasing is associated with increased levels of problem-gambling severity. These results were further corroborated by Brady and Prentice (2019), who uncovered that while purchasing a loot box, gamers display the same physiological response as
gamblers do when participating in games of chance. Recently, another disorder has been added to the list of psychological constructs influencing the players’ interaction with loot boxes, namely gaming addiction. This inclination to obsessively play video games has also been found to affect loot box spending (King and Delfabbro, 2018). Up to this point, these scientific insights mostly coexist without being concisely brought into relation with one another. Yet, in light of the lacking understanding concerning loot boxes and the reluctant regulatory response, it seems valuable to synthesize a
worthwhile to investigate whether loot boxes are a psycho-structural issue and if these factors can combinatorially affect a singular person.
To the author’s knowledge, the only two attempts to create such a model were made by Li, Mills, Nower (2019) and Drummond, Sauer, Ferguson, Hall (2020). Their work successfully established relationships between loot box purchases, mental distress, and the symptomology of problem-gambling as well as gaming addiction. Still, it is argued that the structural approach to the matter should further be augmented by the inclusion of psychological constructs of flow and perceived behavioural control. Flow has in the past been shown to be a valuable instrument in the evaluation of players’ experience with a video game (Jin, 2011; Sweetser and Wyeth, 2005). It could either contribute to or counter the effects of the outlined psychological conditions on the mental state of the players. Furthermore, the generalized mental distress measures used in the two
predominant culture (Prinz, 2009). It can, therefore, be argued that another theory is better suited to evaluate the ethics of a digital business model, which is employed in a global market with a set of diverse cultures. For this reason, the theory of Kant’s categorical imperative was used in an effort to contribute to the ethical perspective on this type of video game monetization. Neely (2019) has previously investigated the ethics of microtransactions, loot boxes, and free-to-play games from a Kantian perspective. Still, her argument lacks the insight of actual players, which engage with the virtual good. While her ethical argumentation is not without merit, she ultimately fails to go beyond the technicalities of the mechanics themselves and to take into account the actual experiences of players. Thereby, it is not elaborated on how they actually affect the players’ enjoyment of the product and how their autonomy may be compromised by the implementation of the bespoke systems. Yet, the concept of autonomy is central to Kant’s categorical imperative (see chapter 4). Therefore, this study intended to expand upon her argument by incorporating the experience of paying consumers and the psychological constructs into this research’s design.
To address these outlined shortcomings, the study at hand took a structural equation modelling (SEM) approach to the subject matter of loot boxes, which incorporated the psychological constructs of flow, gaming addiction, problem-gambling, perceived behavioural control, and cognitive dissonance. This way it aimed to further unravel which cognitive structures underly players’ engagement with loot boxes, how they manifest themselves post-purchase, and whether they call the ethics of loot boxes into question. To answer this threefold research question, a review of the literature on loot boxes, the current international legal position as well as the above listed theoretical concepts has been conducted. Based on these insights a theoretical model was
developed. Said model was then tested via structural equation modelling. Using a cross-national, self-selected sample of players, which have purchased loot boxes, a
2. Definition and Operation of Loot Boxes
Loot boxes are a type of in-game purchasing system within video games, which aim to monetize additional content sold as a complementary good to the main game. A
purchase of this additional content is technically not required to play the game. Still, the implementation of such a system is prone to severely impact the game’s design
philosophy, as it tries to incline players to make the purchase. Sometimes loot boxes can be obtained for free by progressing in the game. However, this has to be considered a promotional method, which aims to provide a product sample to incentivize additional purchases. Originating in the mobile games market, the model is now employed in free-to-play games3 as well as full-price titles. The term “loot box” is commonly used for “items in video games that may be bought for real-world money, but which provide players with a randomized reward of uncertain value” (Zendle, Meyer, Cairns, Waters and Ballou, 2020, p. 1768). Visually, they mostly present themselves as treasure chests, crates, or card packs (see Figure 2.1). The contents encapsulated within the boxes can range from purely cosmetic items to customize in-game characters to temporary boosts, which speed up progression or grant other gameplay advantages. The value of these items often correlates with their rarity within the game. Often, this scarcity is colour-coded and further indicated by items being attributed descriptive adjectives such as “common”, “rare”, “epic” or “legendary”. The animation of opening a loot box often includes flashing lights, fireworks, confetti falls, or the sound of coins crashing (Bailey, 2018). These audiovisual stimuli are reminiscent of those used in regular slot machines and seemingly contribute to the induction of the physiological response identified by Brady and Prentice (2019).
Figure 2.1: Exemplary visual representations of loot boxes
The item that players receive is determined by an algorithm, which computes the number X (Drummond and Sauer, 2018). In this instance, X represents the average number of openings required to be rewarded a high-value item (ibid.). This number is often even disclosed to consumers ahead of their purchase. However, due to the chance-based nature of loot boxes, the exact number of openings required to obtain a valuable item will vary (ibid.). This practice has been termed a variable-ratio schedule of reinforcement (Ferster and Skinner, 1957, p. 391) and has been identified to underlie a multitude of traditional gambling activities (Rachlin, 1990). The continued wagers constitute a string for which the single occurrence of a win or loss does not matter (ibid.). Solely, the length of said string of losses, eventually resulting in a win, is relevant (ibid.). People that engage with such a system are, therefore, prone to use an accounting system, which only accumulates losses once a string ends with the
occurrence of a win (Thaler, 1981). Afterwards, the system is reset, leaving the
chance. In a 2018 presentation, a user experience researcher working for Epic Games4 said the following concerning mechanics like loot boxes: “The reason these are famous is they are quite often the ratios used in gambling. The response they tend to get is a very constant and high-level response and that’s because you don’t know how many times you have to respond before you get the reward, so people tend to keep it up.” (Lewis-Evans, 2018)
Furthermore, it is in most cases unclear how many unique items of a certain grade exist. Hence, if a person desires a specific unique item, he/she would have to research the total number of items of that category and then calculate the exact odds of being rewarded the desired item independently. Therefore, the disclosure of chances (number X) is of limited transparency.
Additionally, loot boxes are often employed in combination with a coercive
monetization model. Instead of directly purchasing a virtual good, such as a loot box, with real-world money, an intermediary currency has to be used. In-game currency can come in the form of crystals, gold coins, scrap, or simply points. The two-step process of first purchasing units of said virtual currency and then buying the desired item with it obscures the actual value of the transaction. While the information is technically
available, it is cognitively harder for the consumer to access and comprehend it. Thereby, an information asymmetry to the benefit of game publisher is artificially induced. (Shokrizade, 2013)
It is also to be mentioned that numerous of the games are available to children or even being specifically marketed towards them by:
• the inclusion of characters popular with or likely to appeal to children • cartoon-like graphics
• bright colours
• simplistic gameplay and/or language
• concerning an activity that is likely to appeal to or be popular with children • the game is featured in a children’s section of digital stores
(Office of Fair Trading, 2014, p.20)
3. Status Quo of Regulation
3.1 International Legal DebateIn the aftermath of the Star Wars Battlefront 2 incident, numerous international
governmental bodies have commenced to inquire into the potential issues with modern video game design and their monetization. As a reaction to the mounting public
criticism5, the responsible United Kingdom (UK)’s Gambling Commission responded by examining whether loot boxes constituted gambling under UK law. While expressing its concern for children’s well-being and its awareness of the increasing convergence between gambling and video games, the commission still found that loot boxes did not meet the prevailing legal definition of gambling. This ruling was predominantly due to the notion that most items gained through loot boxes cannot be used or traded outside the game and, therefore, the prizes “won” via a loot box are by legal definition not “money or money’s worth” (House of Lords, 2020, p. 115). As a result, the Gambling Commission could not exert its executive power with regard to this type of virtual good. However, this ruling does not accurately portray reality. While the majority of titles does not allow for trading within the game, there exists a wide array of third-party websites and Facebook trading groups with the sole purpose of giving players the
opportunity to exchange the items gained through loot boxes (Juniper Research, 2018)6. Most of these transfers use the virtual in-game currency, however many also involve real money. These developments have also given rise to middlemen that capitalize on the trades and even advisory e-books, which propose strategies to increase the
profitability of bespoke trades7. Based on this information, one can claim that through loot box mechanics a virtual shadow economy has emerged. As this economy is subject to the same dynamics of supply and demand as the markets for regular goods, these virtual items are seemingly “money’s worth”. (Miller8, 2017)
Yet, in September 2019, the UK’s Department for Digital, Culture, Media and Sport (DCMS) published its report on immersive and addictive technologies (Taylor, 2019). Among other issues relating to digital technology, loot boxes were emphasized to be “one of the most prominent features in the debate about the potential links between game mechanics and gambling” (DCMS, 2019, p.27). In the absence of evidence proving that loot boxes do not cause harm to children, the report recommended that minors should not be permitted to play games, which incorporate virtual goods reliant on an element of chance. In response to the bespoke report, Caroline Dinenage9 and the UK government in power have issued a call for evidence on loot boxes with the aim of further examining the links to gambling-like behaviour and excessive spending (DCMS, 2020). Most recently, the House of Lords has publicised another report demanding immediate action from the government to classify loot boxes as “game of chance” (House of Lords, 2020, p. 115). By this legal definition, the practice could be brought under section 6 (6) of the Gambling Act 2005 and, thereby, be regulated accordingly (ibid.). The House of Lords even went so far as to propose that ministers should make regulations under section 6 (6) without awaiting the government’s wider review of the
6 A multitude of these groups and websites were also encountered while conducting the research at hand. 7 An example for this is this Amazon entry:
https://www.amazon.com/FIFA-21-Market-Guide-
Economists-ebook/dp/B08H134JRC/ref=pd_sbs_351_2/137-4832054- 1869410?_encoding=UTF8&pd_rd_i=B08H134JRC&pd_rd_r=d3fe6806-4365-45e8-a41b-
40257d64221c&pd_rd_w=79Zrq&pd_rd_wg=pG2uj&pf_rd_p=ed1e2146-ecfe-435e-b3b5-d79fa072fd58&pf_rd_r=XD3BW4HJDSD8W7C0QAD2&psc=1&refRID=XD3BW4HJDSD8W7C0QA D2
Gambling Act (ibid.). This proposition clearly echoes the concerns voiced by
researchers as well as the urgency with which further investigative and regulative action should be taken. Despite these initiatives, a review of the Gambling Act 2005 is still pending and consequently, for the time being, loot boxes do not have to obey any kind of gambling-related law within the UK.
However, countries like the Netherlands and Belgium have taken a different regulatory approach. The respective Belgian Gaming Commission analysed, whether loot boxes violated the state’s Gaming Act of May 7, 1999. In its assessment, the commission interpreted the Gaming Act’s definition of gambling more liberally. In simple terms, under Belgian law, gambling is defined as: (1) a game; (2) that involves “a stake of any kind,” (3) results “in a loss or a gain” and (4) involves “at least a minimal degree of chance in the outcome” (Paepe, 2019). Contrary to its UK counterpart, the Belgian commission found that under the above-listed criteria loot boxes constitute a “game of chance”. In this particular ruling, it was irrelevant whether the received item could be converted into money to be considered a win (Gaming Commission, 2018).
Consequently, all forms of loot boxes that required a purchase using real-world
currency were outright banned in the country, declaring the practice criminal gambling. Thereby, developers were forced to react and remove the mechanic from their current as well as future releases. Disobeying the strict prohibition could result in “a prison
sentence of up to five years and fines of up to €800.000 for a first violation” (Koen Geens, 2018). In the case of the player, which opens a loot box, being below the age of 18, the sanctions could double. (Cermak, 2019)
market value. However, on the grounds of the Dutch legislation, this is only the case if the acquired items can be traded outside the game. Thereby, solely loot boxes whose content is transferable are banned in the Netherlands. Furthermore, it has recently been revealed that based on this ban, EA has been fined €10 million for not complying with the Betting and Gambling Act by making the necessary, requested adjustments to its games (Sinclair, 2020). (Netherlands Gaming Authority, 2018)
The concerns outlined above are also shared among regulators and the public in the United States of America (US). Yet, the legal environment regarding loot boxes is more complex than in the aforementioned countries. This circumstance occurs due to
gambling not being subject to federal law, but to the legislation of the individual states. Hence, each state has its own statute concerning gambling activity, which leads to a great variety of definitions and increased vagueness. This also raises the question of whether certain practices could constitute gambling in one state while not being considered gambling in another. In addition, the US system is heavily reliant on case law. In comparison to the remaining countries mentioned in the study at hand, the US displays a larger number of cases with regards to loot boxes. However, up until this point in time, courts have failed to definitively rule whether loot boxes are gambling. As with the considerations in other countries, the main obstacle seems to be the
3.2 Rating Boards and Self-Regulation
Contingent on the market(s) it is released in, each video game is subject to the assessment of a rating board that intends to inform consumers which age groups the respective title is appropriate for. The most prominent of these ratings boards include the Pan European Game Information (PEGI) and the Entertainment Software Rating Board (ESRB), which cover the European market and North America, respectively. Furthermore, the International Age Rating Coalition (IARC), a collaboration of the Australian, Brazilian, North American, South Korean, Pan European, and German rating authorities, has been formed in 2013 (IARC, 2020). The partnership streamlines the age classification process by utilizing an algorithm (ibid.). Through answering a questionnaire, which reflects the distinct standards of each participating board, video game releases are automatically assigned the appropriate rating for each territory (ibid.) Thereby, the process becomes significantly more efficient for publishers, which release their games on a global scale. In addition to the age rating, many rating authorities also specify through labels which kind of content is subject to the game. For example, the PEGI’s content descriptors include labels for in-game depictions of violence, sex, drug use, discrimination, among others (PEGI, 2017). Moreover, a “gambling” label exists. However, it is reserved for “simulations of games of chance that are normally carried out in casinos or gambling halls” (ibid.). When this label is applied, it automatically leads to an age classification of 12 years and above (ibid.). Yet, as loot boxes are presented as treasure chests or card packs instead of slot machines and card tables, the classification is not applicable. The ESRB uses descriptive text on its labels to identify similar themes.
Figure 3.2.1: A selection of content descriptors (incl. the new labels concerning in-game purchases)
Due to the negative public response, EA has recently introduced FIFA Playtime to its highly successful soccer title. Playtime is a suite of tools, which gives players the opportunity to limit their weekly time spent in-game, spending on the respective virtual currency (FIFA points), and number of loot box purchases (EA, 2020). Again, the actual monetary value is not displayed and only indirectly controlled. In addition, it has been shown that consumer engagement with these types of consumer protection tools is very low in the context of internet gambling (Procter et al., 2019)
It is of the utmost importance to recognize that these rating boards are not necessarily independent. They rather are self-regulative bodies, predominantly comprised of major industry players (PEGI, 2020; ESA, 2020). The boards’ members include multinational corporations like Sony Interactive Entertainment, Nintendo, Microsoft, Tencent,
The inconsistent international initiatives in combination with the deeply flawed self-regulative measures clearly demonstrate that loot boxes are currently a legal grey area, which demands additional scientific inquiry. Moreover, the discussed highly profitable video game publishers are likely to be apathetic about the prohibition of loot boxes issued by individual countries. This will especially be the case if the respective countries are comparatively small markets like the Netherlands and Belgium. Only when unified action by a large number of nation states is taken, these corporations’ profits are affected to a degree that will cause them to more meaningfully react. While the international legal debate is gaining traction, it should be evident that there exists a legislative vacuum within which loot boxes act and that leaves consumers as well as regulators vulnerable. Some of it is owed to a lack of understanding concerning these mechanics and their influence. Because of these deficiencies, the research at hand intends to deepen this understanding by answering Caroline Dinenage’s call for evidence.
4. Theoretical Model & Research Hypotheses
Video games have been established to inherently be a psychological phenomenon (Wood et al. 2004; King, Delfabbro, and Griffiths, 2010). So much so that its economic success depends on the medium’s degree of psychological stimulation (King, Delfabbro, and Griffiths, 2010). Video game play has even been suggested to positively impact general well-being through its satisfaction of “basic psychological needs for
Informed by additional review of the literature the psychological constructs described in this section were identified. It is proposed that they meaningfully contribute to the experience of players, which purchase loot boxes. Apart from their scientifically proven relevance to the matters of video games, traditional gambling, and loot boxes, the choice of constructs included in the proposed model was further shaped by their applicability to the ethical theory of Kant’s categorical imperative. His considerations have, in the past, been used to evaluate the morality of business activity. Hence, his proposed principles are also employed within the confines of this particular research to address the question of whether loot boxes can be deemed an ethically permissible virtual good. To briefly lay out the groundwork required to later discuss the statistical results in accordance with Kant’s theory the following points about his principle of the categorical imperative should be made.
Kant’s ethical considerations are part of the deontological stream within the discipline. Theories, which belong to the deontological school of thought, are not concerned with the outcome of an action, but with the motive for it. In accordance with these theories, it is believed that one should act out of a moral obligation to do the right thing, no matter what the actual result turns out to be. Kant is the most prominent philosopher within deontology.
(Gibson, 2012, pp. 38). The concept of autonomy is, therefore, fundamental to Kant’s assessment of moral worth.
As outlined in chapter 2, individuals may be deprived of their free choice through the concealment of necessary information concerning the purchase and the stimuli used to prey on addictive tendencies. Due to loot boxes possible infringement of moral
autonomy, a proxy for measuring this elusive construct is employed. Hence, the construct of perceived behavioural control was used as an endogenous11 variable to capture autonomy as described in Kant’s publications, meaning the freedom to choose without alien influence. It is proposed that this potentially impaired freedom is a consequence of the conditions associated with loot box engagement. It is for this very reason that perceived behavioural control is viewed as endogenous in the theoretical model developed throughout the research at hand.
Secondly, Kant proposes that the morality of an action is to be judged by whether the maxim under which it occurs is universally applicable without contradiction. This means, when a person considers taking a specific action, he/she should consider whether it would be acceptable to them, if the maxim, by which they act, would become
universal law and would be enacted upon them. Kant constructs a hypothetical scenario to test whether the universalizability is violated and the action, as a result, immoral. For example, if all businesses broke contracts, no more contracts would be made, and global economic activity be severely impaired. Hence, the maxim is self-defeating, not
consistently universalizable and, therefore, the action of breaking a contract immoral. (Bowie, 1998)
The third and last axiom, stemming from Kant’s second formulation of the categorical imperative, is based on personal autonomy as it demands other individuals to be treated
11 The terminology within SEM differs slightly from that used in more traditional statistics.
as “ends-in-themselves” (Kant, 1785). Everyone has to be respected as a person and cannot be used as a mere instrument in the quest to achieve one’s own goals. He states that “Every human being has a legitimate claim to respect from his fellow human beings and is in turn bound to respect every other.” (Kant, 1785, p. 209). Dignity, thereby, restricts freedom of action again. However, this does not mean that economic activity with the intention of profit generation is inherently unethical (Neely, 2019). The
expectation of adequate compensation for the risk one takes in the creation of a product, is reasonable (ibid.). It solely becomes unethical if the desires and needs of the
consumer are disregarded (ibid.). Within an economic transaction, the involved parties can still treat each other with respect and dignity (Bowie, 1998). Furthermore, such transactions are allegedly voluntary and should, therefore, be viewed as agreements between responsible rational adults (ibid.). Nor is Kant opposed to the idea of private wealth creation. He acknowledges that individuals “whose possessions are sufficient for his [/her] needs” (Kant, 1775) have greater self-respect and are provided with greater independence as well as pleasure (Bowie, 1998). On the grounds of these conceptions, Freeman (2017) advocated for a departure from neoliberalist capitalism driven by shareholder value towards the simultaneous satisfaction of all key stakeholders. Freeman’s approach is renowned as “Kantian capitalism” (Evan and Freeman, 1988). To adhere to this ideal, capitalism must support the autonomy of economic subjects and go beyond the perspective of viewing individuals solely as human capital in the pursuit of profit maximization.
addiction are permanent for the respective individuals12, cognitive dissonance is, for the purpose of this research, regarded as a post-purchase state, which can occur with the action of buying a loot box. Due to this temporal difference, cognitive dissonance is hypothesized to be a reverberation of these conditions, which have been shown to facilitate continuous loot box spending. For this reason, like perceived behavioural control, cognitive dissonance has been used as an endogenous variable within the proposed model. Meanwhile, problem-gambling and gaming addiction were treated as exogenous.
The remaining exogenous factor, which likely impacts, perceived behavioural control and cognitive dissonance, is that of flow. Flow has in the past been shown to underly a multitude of recreational activities, including video games (Sweetser and Wyeth, 2005). Due to its perceived pleasurable experience, it could indirectly counter the effects, increasing cognitive dissonance. Yet, other scholars have suggested that flow could lead to a loss of control and, thereby, also take on a negative role (Huang, Hsieh, & Wu, 2014). Because of these ambiguous characteristics, flow demands further exploration in the loot box context. Hence, it was included as the third exogenous variable.
In the following, each utilized variable and the hypothesized relationships of it are explained in further detail to ultimately formulate the theoretical model tested via structural equation modelling.
4.1 Cognitive Dissonance
According to the theory, an individual holds a number of cognitive elements that constitute knowledges about itself, its environment, its attitude, its opinion, and past behaviour (Oshikawa, 1969). The term cognitive dissonance refers to a “psychologically uncomfortable state” (Sweeney, Hausknecht and Soutar, 2000, p.369), which occurs when two of the aforementioned cognitive elements are in conflict with one another (Carpenter, 2019). This negative arousal can be evoked “ (1) after making an important
and difficult decision, (2) after being coerced to say or do something which is contrary to private attitudes, opinions, or beliefs, and (3) after being exposed to discrepant information” (Oshikawa, 1969, p.44). To rectify this unpleasant emotion, subjects may engage in dissonance reducing behaviours. These can include attitude change,
avoidance of dissonant information, perceptual distortion, and behavioural change (Brehm and Cohen, 1962). The originator of the theory, Festinger (1957), hypothesized that decision-making almost always induces cognitive dissonance as it necessitates the rejection of the remaining alternative(s). The magnitude of the experienced dissonance is contingent on the importance of the decision and the relative attractiveness of the alternatives that have been rejected in the process (Oshikawa, 1969). The theory has been applied to consumer behaviour by examining the dissonance experienced post-purchase. In this context cognitive dissonance is often referred to as “buyer’s remorse” (Rosenzweig and Gilovich, 2012). A multitude of circumstances in which dissonance may occur has been identified. The experience of dissonance may be induced when the consumer is exposed to novel, discrepant information that may ascribe negative
attributes to the chosen alternative or positive attributes to the rejected alternative (Straits, 1964). Typically, subjects positively reevaluate their purchase decision, when novel information is confirmatory (Cohen and Goldberg, 1970). Meanwhile,
information, which is inconsistent with the purchase decision, leads to negative reevaluation (ibid.). The occurrence of the phenomenon is, furthermore, contingent on how public the commitment to a purchasing decision is (Cohen, Brehm and Latane, 1959). As a consumer publicly claims his choice to have been a good one, his ego-involvement concerning the purchase becomes stronger (Oshikawa, 1969). Scholars Knox and Inkster (1968) have previously applied the theory in the context of a traditional gambling activity, namely horse betting. Their post-decision observations revealed another possible dissonance-reducing behvaiour. After making the financial commitment, bettors reappraised their choice and grew more confident in it without yet knowing the outcome (ibid.).
recognize who has obtained certain prestigious items via loot boxes (Evers, van de Ven and Weeda, 2015). As many in-game items are cosmetic in nature, players that engage with them are likely to be motivated by a desire to express themselves (Cleghorn and Griffiths, 2015). Commitment is further increased when the purchase is irreversible, as this circumstance heightens psychological importance of the decision (ibid.).
Irreversibility of the purchase can be assumed for loot boxes, as refunds for them are hard to obtain. The moment in which the purchased loot box is opened, and its contents are revealed is critical to the concept of cognitive dissonance. In this instant, it is determined for the consumer whether the information is supportive or discouraging to their purchase decision. They are either awarded an item that is of value to them or something they deem useless. Based on the knowledge that the negative strings in variable-ratio schedules of reinforcement are longer (see chapter 2), it is expected that for the vast majority of players, the information is mostly disconfirmatory.
4.2 Perceived Behavioural Control
a desired behaviour, but also in desisting from socially undesirable or potentially unhealthy behaviours. Given these research results, it is evident that perceived behavioural control plays a major role in the engagement of activities, which have addictive potential.
4.3 Problem Gambling
Gambling can be defined as risking something of value in the hopes of gaining something that offers higher value. The likes of horse racing, lotteries, and chance-based card games constitute traditional gambling activities. While most people come into contact with a gambling activity at one point in their life and many engage in gambling on a frequent basis, only a few satisfy the criteria of a gambling disorder. Within the diagnosis of a gambling disorder, psychologists often differentiate between problem gambling and the more severe form pathological gambling (King and
Delfabbro, 2013). Both of these conditions are commonly identified by the negative consequences the excessive gambling behaviour creates for the individual itself or those in its social network (Ferris and Wynne, 2001). In accordance with the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) these morbidities are classified as impulse control disorders. Among other characteristics, the illness can manifest itself as:
• a need to gamble with increasing amounts to achieve the same rush • irritability when it is attempted to gamble less frequently
• lies, which try to conceal the severity of the habit
• chasing losses, meaning the inclination to return on another day to recoup previous losses (American Psychiatric Association, 2018)
the strong correlation between adolescence and problem-gambling in the traditional form of the activity. As of now, the direction of causality has not been sufficiently established. It is unclear whether purchasing loot boxes acts as a gateway to gambling, or whether increased spending on loot boxes has greater appeal to problem-gamblers (Zendle and Carins, 2018).
Based on these scientific insights, problem gambling was included in the proposed theoretical model to test if the above results would be echoed. Due to it being an impulse control disorder, the path analysis tested whether the degree of problem
gambling severity affects perceived behavioural control. Previous research by St. Pierre et al. (2015) found that a lack of perceived behavioural control influences intention to gamble in adolescents. It has further been established that a poor sense of behavioural control over gambling refusal increases vulnerability to excessive gambling
involvement (Wu et al., 2013). The negative relationship between problem-gambling and perceived behavioural was further empirically supported by Martin et al. (2010), who concluded that a lowered sense of control was predictive of higher frequency of gambling. Consistent with these observations, Wu and Tang (2012) provided further evidence by identifying a positive effect of low control over gambling refusal resulted in increased intention to gamble among Chinese college students. The only converse results reported, were those of Martin et al. (2011). In this study, differences between problem and non-problem student gamblers were found. While perceived behavioural control positively impacted gambling frequency for the non-problematic group, it had no effect for problem-gamblers. Due to the majority of the previous scientific insights supporting the notion of problem-gambling significantly affecting perceived
behavioural control, the following hypothesis was proposed:
H1: Problem-gambling severity has a direct and negative effect on perceived behavioural control.
In addition, it was hypothesized that problem-gambling severity would increase
to showcase greater frustration with their purchase decision, as they fail to break the cycle and experience more psychological distress as a result. Taormina and Chong’s 2015 study suggested that even though they do gamble, gamblers hold a multitude of negative attitudes towards the chance-based activity they engage in. The scholars found that increased cognitive dissonance was strongly associated with higher levels of gambling. They even went so far as to advocate for the usage of cognitive dissonance for therapeutic purposes, meaning that the subject’s negative beliefs should be
channeled to deter it from engaging in excessive gambling. Hence, it was tested whether this relationship would prevail among those who purchase loot boxes. Therefore, the second hypothesis read:
H2: Problem-gambling severity has a direct and positive effect on cognitive dissonance.
4.4 Gaming Addiction
Gaming addiction, also referred to as internet gaming disorder, has in recent years been highly debated among clinician scientists. To date, the disorder has not been included as a formal disorder in the DSM-5. Yet, the American Psychiatric Association (APA) is still considering its future inclusion and recognizes that the condition warrants further clinical research (2013). Meanwhile, the World Health Organization (WHO) has recently included gaming disorder as a psychopathological condition in its 11th revision of the International Classification of Disease (ICD-11). This publication is universally used by practitioners in diagnosis and informs the planning of public health strategies (WHO, 2018).
evokes pleasant emotions (ibid.). This reward circuitry plays a major part in the
conditioning of maladaptive behaviour and can ultimately turn initial affinity for a game into obsession. Abstinence from play can in severe cases lead to the occurrence of withdrawal symptoms (APA, 2013). Previous research conducted by King and
Delfabbro (2018) has hypothesized that the presence of a gaming addiction may result in the individual dedicating larger financial resources to the game. They view an investigation of the relationship between gaming addiction symptoms and the different types of monetization as the next logical step in research (King and Delfabbro, 2018). The proposed theoretical model incorporated their recommendation in regard to loot boxes. Given that gaming addiction is considered to be compulsive and linked to higher impulsivity (Kim et al., 2017), it is inferred that perceived behavioural control is
diminished as the severity of the gaming addiction increases. A similar relationship was previously established by Haagsma et al. (2013), whose results indicated that perceived behavioural control was the most important factor in the prediction of video gaming addiction. Yet, within the research at hand the reverse was tested, as it was viewed as more reasonable to assume that the underlying condition of gaming addiction causes an increasing sense of losing control than vice versa. These considerations led to the formulation of the following hypothesis:
H3: Gaming addiction has a direct and negative effect on perceived behavioural control.
Similar to the proposed relationship between problem-gambling and cognitive dissonance, it was expected for gaming addiction to also increase post-purchase
problematic behaviour. In the presence of sufficient external justification13, individuals likely experience little to no cognitive dissonance (Elliot & Devine, 1994). It was, conversely, expected that the absence of the bespoke justification would lead to
heightened states of cognitive dissonance. The study at hand, therefore, followed up on Chin-Sheng and Chiou’s (2007) recommendation and, therefore, incorporated their rationale concerning gaming addiction and cognitive dissonance as the following:
H4: Gaming addiction has a direct positive effect on cognitive dissonance. 4.5 Flow
The construct of flow was pioneered by psychologist Csíkszentmihályi in 1990. The term refers to a state of consciousness, in which individuals are completely absorbed in an activity and concentrated on performing the behaviours associated with that activity (Csíkszentmihályi, 1990). Subjects predominantly perceive the experience as
pleasurable, as the phenomenon manifests itself as an exhilarating feeling of
transcendence, which blocks out emotional problems and prompts temporal distortion (ibid.). Characteristics of flow often recognized by individuals include a sense of strength, alertness, effortless control, unselfconsciousness, and a feeling of being at the peak of one’s abilities (ibid.). Csíkszentmihályi (1990), therefore, deems flow the optimal experience, which brings about deep enjoyment that may even be described as happiness.
on a day-to-day basis (Wanner et al., 2006). This dimension of balance between challenge and skill marks the most important precondition among the dimensions identified14 (ibid.) Previous research has concluded that flow positively influences the willingness to use hedonic information technologies (Kwak, Choi, and Lee, 2014). In the context of video games, it has been shown that flow affects the players’ perceived enjoyment and can impact the adoption of online video games (Weibel, Wissmath, Habegger, Steiner and Groner, 2008). Similar results were reported by Hsu and Lu (2004), who found evidence of flow to be a main contributing factor in the intention of users to play an online game. Therefore, the entertainment medium offers a prototypical context for flow (Sherry, 2004). The concept was also applied to media usage,
suggesting that flow leads to repeated and potentially excessive media consumption (Lee and LaRose, 2007).
However, other scholars have suggested that not all outcomes are necessarily beneficial to the individual experiencing flow (Partington, Partington, and Oliver, 2009). It has been established that for a minority of individuals the emotions experienced during recreational activities can become addictive (Jacobs, 1989). The positive experience of flow has been argued to be a possible precursor to compulsion (Faber, O’Guinn, & Krych, 1987; Huang, Hsieh, & Wu, 2014). When applied to gambling, Wanner et al. (2006) found that pathological gamblers experienced lower mean levels of flow than those who engage in the activity for recreational purposes.
As it has been established, that loot boxes are psychologically akin to gambling, it is of interest to test whether this relationship would prevail in the context of video games and loot boxes. With regards to video games specifically, it has been suggested that the experience of a flow state may prompt the self-regulatory system of an individual to malfunction as the functions of judgmental process and self-reflective influence fail to effectively control gaming behaviour. Paradoxically, in this instance, flow leads to a
lack of self-control, despite it typically being described as involving a sense of control. Csíkszentmihályi (1990) himself recognized this inconsistency and termed it the “paradox of control”. Further empirical evidence for the existence of this paradox within a video game context was provided by Khang, Kim, and Kim (2013). They reported that among the variables tested flow had the strongest negative correlation with self-control. Due to the theoretical similarity of self-control and perceived behavioural control (Wolfe and Higgins, 2008), it was expected that the same would hold true for the effect of flow on perceived behavioural control. The fifth hypothesis included in the theoretical model was, therefore:
H5: Flow has a negative direct effect on perceived behavioural control.
It was further proposed that the construct would lower cognitive dissonance, due to the positive emotions associated with flow. Due to flow frequently being regarded as a desired attribute of a video game and it even being used as a measurement to assess a game’s quality (Cowley et al., 2008; Sweetser et al., 2012; Sweetser and Wyeth, 2005), it was expected to lessen the post-purchase dissatisfaction measured through cognitive dissonance. Furthermore, flow is associated with, increased life satisfaction, higher self-esteem as well as a strengthened sense of social and place identity (Mao et al., 2020). Cognitive dissonance, on the other hand, has been shown to pose a threat to an individual’s self-esteem (Steele, 1988). These previous findings imply the following relationship:
4.6 The Model
Based on the above considerations the following theoretical model has been formulated:
Figure 4.6.1: Theoretical model
5. Methodology
For this particular research, an overall quantitative approach was taken. The assumption underlying this mode of inquiry is that phenomena have an objective ontological
mechanic that may negatively affect a larger portion of players. It is argued that, due to the above-outlined characteristics, quantitative research is a better methodological fit for this type of research question.
5.1 Data Collection & Sample Characteristics
To empirically test the proposed theoretical model, a questionnaire was administered online. The survey was designed and distributed using the Qualtrics software. The utilization of the online medium for primary data collection offers a number of advantages over its offline counterpart. First of all, most gamers have access to the internet and are proficient in using it. Thereby, access to a potentially global pool of respondents was gained (Buchanan, 2000). These respondents, additionally, did not have to be in the same location as the researcher (Whitty, 2004; Wood et al., 2004). The pool of potential participants was further enlarged by the chance that “socially
unskilled” individuals, who would not have considered participating in an offline interview, may partake in an online environment (Wood et al., 2004; Wood and Griffiths, 2007). Lastly, the recruitment process is simplified, as the project can be advertised on a multitude of websites (Wysocki, 1998). This is used to the advantage of this particular study, as it was advertised across a vast variety of Facebook groups. The targeted groups were dedicated to games, which incorporate loot boxes15. Given this nonrandom technique, the method should be classified as purposive sampling (Etikan, Musa, and Alkassim, 2016). A deliberate choice was made to recruit members of these gaming groups, as they would possess greater ability to provide meaningful information by virtue of their loot box-related knowledge and experience (Bernard, 2017). After an initially disappointing response to the posted survey16, an incentive in form of a prize draw17 was introduced to increase the response rate. The inclusion of the prize draw could affect the results of the survey, yet the majority of research examining the impact of incentives on response quality has found no such effects (Singer and Ye, 2013).
15 A list of the games featured in this particular sample can be found in the Appendix (1).
16 After one week only 15 responses had been recorded, despite the survey being posted in a large number of Facebook groups.
This measure proved effective as a sufficient18, self-selected sample from 43 different nationalities was obtained. The heterogenous sample, thereby, mirrored the diverse international gaming community. The majority of responses originated from the United Kingdom (n=36; 17,82%), Germany (n=29; 14,36%) and the United States of America (n=23; 11,39%)19. To combat ethical concerns concerning data protection, the responses were recorded anonymously. A total of 274 responses was registered.
A subsequent data screening was conducted to eliminate responses, which would not meaningfully contribute to the analysis. In the first step, 60 respondents, which have either chosen the option “I have never spent directly or indirectly spent any real money on a loot box.” or “Mobile Phone” as their gaming system of choice, were excluded. This is due to the reason that the behaviour and experience of players who have spent money on virtual goods vastly differ from those, which did not dedicate financial resources to them (Paavilainen et al., 2013). Moreover, as established in the formulation of the theoretical model (see chapter 4), for buyer’s remorse to occur a commitment in form of an actual purchase has to be made (Rosenzweig and Gilovich, 2012). As per definition, engagement with loot boxes could only possibly constitute gambling if the subject puts something at risk. The decision to exclude players on mobile platforms has been made, as they are still considered a market with a different player audience that is separate from consoles and PC. While a mobile game may in essence offer the same gameplay as its console and PC counterparts, the technical performance can
significantly digress due to the hardware differences. This circumstance could,
therefore, affect the results. In the second stage, unengaged responses were eliminated. These were identified by examining the responses to the inverse questions, which were specifically included in the survey to serve this purpose. Moreover, using descriptive statistics, the standard deviation of the responses was computed. Responses, which showed a standard deviation below a value of 0.5, became subject to further investigation as it could originate from participants consistently choosing the same answer for all items just to quickly get through the survey (Pamu, 2017). It was decided
on a case-to-case basis whether the respective answers seemed unengaged. As a result of these two procedures, 12 further responses were disregarded. A test for outliers concerning the sole continuous variable “Age” was conducted. Yet, no outliers were detected. In addition, the data was screened to detect missing values. Incomplete responses with more than 10% of values missing, would be rejected. However, all answers that were retained up to this point were complete. Consequentially, none were excluded in this step. As a result, 202 observations remained within the sample after the completion of this process. The resulting sample was composed as follows:
Individual-level variables N (= 202) Percent (rounded) Gender
Male 194 96.04
Female 8 3.96
Age in Years (Mean=28.3)
≤ 19 26 12.87
20-30 118 58.42
31-40 33 16.34
41-50 13 6.44
50 ≥ 12 5.94
Weekly Hours of Play
Problem-Gambling Severity
Non-problem Gambler 85 42.08
Low-risk Gambler 50 24.75
Moderate-risk Gambler 43 21.29
Problem-Gambler 24 11.88
Table 5.1.1: Composition of sample 5.2 Operational Measures20
All scales used in the survey were adapted from previous studies. Their wording was slightly modified to match the video game and loot box context. A few of them were converted from a 5-point to a 7-point Likert scale. This was done in an effort to increase consistency, thereby, making the questionnaire more comprehensive and ultimately improving reliability.
5.2.1 Game Addiction Scale
least “Sometimes” (4)21. As a result, a binary variable was created. It made the distinction between those displaying addictive behaviour and those that did not. No alterations to the wording of the items were made, as they already matched the video game context. (Lemmens, Valkenburg, and Peter, 2009)
5.2.2 Problem Gambling Severity Index (PGSI)
Developed by Ferris and Wynne, the PGSI was first introduced in 2001 to meaningfully measure problem-gambling. The 9 items included in the index are assigned a score to, subsequently, produce a prevalence rate for problem gambling (ibid.). Depending on their answers, participants are allocated to 1 of 5 groups (“non-gambling, non-problem gambling”, “low-risk gambling”, “moderate risk gambling”, “problem gambling”). Each of the 4 possible answers is assigned a score22. It is the sole instrument in the analysis
whose range has not been modified to match the 7-point scale. The rationale for this decision was that this conversion would have required the inclusion of a neutral point (a 4 in a 7-point scale). To stay consistent with the remaining scales this middle point would have had to be “Sometimes”. However, in accordance with its original design, “sometimes” would already have to contribute one point to the overall score. Hence, the conversion would have constituted a severe interference with the original design, which could potentially have led to the loss of the instrument’s desirable properties. Prior research has established that the PGSI is preferable over the widely used DSM instrument. The PGSI was shown to have greater internal reliability, uni-dimensionality, as well as better item-response characteristics (Orford et al., 2010).
5.2.3 Perceived Behavioural Control
A 4-item semantic differential scale was adapted from Conner and MacMillan’s (1999) publication to capture perceived behavioural control. Respondents had to indicate how much in control they feel, as well as how easy and likely it is for them to purchase additional loot boxes in the next 2-week period. They were tasked with giving a score
21 A stricter monothetic mode also exists, which demands for all 7-items to be endorsed to grant a positive diagnosis.
between 1 and 7. Higher scores signaled greater perceived behavaioural control. In their study, Conner and MacMillan (1999) reported a Cronbach’s alpha of 0.9, indicating high internal consistency.
5.2.4 Flow
The Likert scale utilized in the measurement of flow captured 3 aspects of the flow experience and is grounded in Csikszentmihalyi’s (1975) theory. These first-order constructs included enjoyment, merge of action and awareness, and concentration in playing video games (Lee and LaRose, 2007). Furthermore, the original wording of the 9 items has been left intact. Respondents were solely asked to answer the questions with regards to the specific game they have purchased loot boxes for.
5.2.5 Cognitive Dissonance
To measure the elusive construct of cognitive dissonance, Sweeney, Hausknecht, and Soutar (2000) proposed a 22-item instrument. It, thereby, constituted the most extensive scale used in the questionnaire. The scale measures cognitive dissonance along the 3 dimensions of emotional response, wisdom of purchase, and concern over deal. To accommodate for the relatively small sample size (N = 202) the 22-item underwent an item-parceling procedure (Matsunaga, 2008). Research has suggested that parceled data reduces random error (Little et al., 2002), provides estimates of greater stability, and showed a higher fit with the data than its item-based counterparts (Bandalos, 2002; Holbert and Stephenson, 2002). The mean for each of the 3 dimensions was, therefore, calculated and in the analysis.
5.3 Data Analysis
(Byrne, 2010). The similarity of the two patterns is assessed via the degree of model fit. It can, therefore, be thought of as a mostly “confirmatory approach” (Byrne, 2010). The approach allows researchers to simultaneously study the correlations between several complex constructs and present them in a more comprehensive pictorial form (ibid.; Andreassen, Lorentzen and Olsson, 2006). The advantages of SEM over traditional multivariate techniques lie in the following:
• Its explicit assessment of measurement error.
• The estimation of latent variables via observed variables.
• Model testing, which involves imposing and assessing a theoretical structure regarding its fit to the data. (Novikova et al., 2013)
Cheng (2001) has further suggested that SEM is statistically more efficient and yields greater explanatory power in the testing of parsimonious models and, thereby, excels beyond multivariate linear regression. Additionally, the modification indices within SEM allow for the identification of novel relationships within the model (ibid.). Multivariate linear regression, on the other hand, overlooks such interaction effects among the posited variables and is, hence, not as capable to analyse complex human and behavioural issues as the one at hand (ibid.). As the psychological constructs employed in the research at hand are rather abstract, not directly measurable, and expected to interact with one another, SEM is regarded as the preferable method to capture the essence of the players’ engagement with loot boxes, because of the above-listed characteristics. The software utilized in the process was STATA 16.
6. Results
(EFA)23. The EFA of flow revealed that item 5 loaded heavily (0.88) on a secondary factor. It was presumed that the reverse logic of item 5 (“I am not totally
concentrated.”)24 confused respondents and, thereby, caused inconsistent answers. Given the observed loading and the outlined judgement the decision was made to remove the above item from the measurement model for the purpose of scale purification (Wieland et al., 2017). A second EFA of the remaining items lead to satisfactory results, as all items loaded on the correct singular factor (flow).
Scale Number of
Questionnaire Items Cronbach’s Alpha
Flow 7 0.69
Gaming Addiction 7 0.77
Problem-Gambling 9 0.9
Cognitive Dissonance 22 0.95
Perceived Behavioural Control 4 0.42
Table 6.1: Results of initial reliability check
The severe underperformance of the perceived behavioural control measure was more alerting. Two different factors were identified via the EFA, yet none would display satisfactory internal consistency. Due to the identified unreliability, the decision was made to and assess the degree of perceived behavioural control via a single item, in accordance with Bergkvist and Rossiter (2007). Based on face validity considerations, item 2 (“How much do you feel that whether you buy additional loot boxes is beyond your control?”) was chosen from the set of indicators and employed in the analysis, as it most closely embodied perceived behavioural control (ibid; Dimantopoulos et al., 2012). While some researchers have criticised the use of single-item measures due to varying predictive validity (Dimantopoulos et al., 2012), others have vocally advocated
for it (Petrescu, 2013; Kwon and Trail, 2005) and came to the conclusion that their predictive validity is equal to that of multiple-item measures (Bergkvist and Rossiter, 2007). One advantage of their utilization lies in the avoidance of common method bias, among others (ibid.).
From this point onwards, the modified versions of the variables were used, meaning that flow and perceived behavioural control included the outlined refinements, while the final scores for the gaming addiction scale and the PGSI were computed. Based on this score the subjects were classified, leading to two additional single-item measures. In SEM single indicator constructs are typically treated as observed variables, as was done in this instance (Muthen, 2003; IBM, 2018). Cognitive dissonance was from here on utilized in its parceled form, consisting of 3 items (see chapter 5.2.5). All variables underwent a Barlett test of sphericity and a Kaiser-Meyer-Olkin (KMO) test. The results of both confirmed the suitability of the variables for factor analysis25 (Dziuban and Shirkey, 1974; IBM, 2020).
Subsequently, descriptive statistics for these variables were generated to establish the distribution of the data. In addition, a Shapiro Wilk test of normality was employed. The values for skewness and kurtosis as well as the significance of the Shapiro Wilk test for the majority of the variables indicated non-normal distribution (Table 6.2). Oftentimes, researchers model ordinal data and ignore the non-normal distribution that is, in many cases, associated with this type of data (Finney and DiStefano, 2006, p.276; Muthen and Kaplan, 1985). Consequentially, the default option of normal theory maximum
likelihood (ML) estimation is used to analyse CFA models (Curran, West, and Finch, 1996). However, ML assumes the measured variables to display multivariate normal distribution in the population (ibid.). Due to the violation of this assumption, the estimation of non-normally distributed data using ML can produce severe distortions and ultimately misleading results (ibid.; Yuan and Bentler, 2007). In line with the
recommendations of Bandalos (2014), a robust maximum likelihood estimator (MLR)26 was used instead to accommodate for the non-normal distribution and the inclusion of binary and ordinal variables in the model.
n Mean SD Skewness Kurtosis Shapiro Wilk
Test (p-value) Flow Flow_1 202 5.46 1.33 -1.1 1.2 0.00000 Flow_2 202 5.27 1.32 -0.91 0.91 0.00000 Flow_3 202 4.72 1.69 -0.61 -0.5 0.00052 Flow_4 202 4.44 1.52 -0.42 -0.54 0.02327 Flow_6 202 4.76 1.67 -0.59 -0.61 0.00017 Flow_7 202 5.1 1.27 -0.7 0.47 0.00007 GamAdd 202 0.32 0.47 0.79 -1.38 0.16502 ProbGamb 202 1.03 1.06 0.56 1 0.02446 PBC 202 5.15 2.11 -0.69 -1.01 0.00002 CogDis CD_1 202 2.87 1.3 0.57 -0.54 0.00000 CD_2 202 4.69 1.53 -0.49 -0.4 0.00405 CD_3 202 4.2 1.86 -0.29 -1 0.00087
Table 6.2: Summary statistics and Shapiro Wilk test results
The a priori defined measurement model fit the data well in the joint confirmatory factor analysis using MLR. The standardized root mean square residual (SRMR; = 0.071), as well as the coefficient of determination ( = 0.96), indicated good model fit (Hu and Bentler, 1999)27. To further establish convergent validity the variance extracted and composite reliability for the multi-item constructs flow and cognitive dissonance were investigated (Table 6.3). Fornell and Larcker (1981) recommend the variance extracted (AVE) to be above 0.5 and composite reliability (CR) above 0.7. With 0.36 the
