Video piracy priming and dishonest behaviour

Video piracy priming and dishonest behaviour by

Roelof van Dijk 10452362

A master thesis submitted to the Faculty of Science of the University of Amsterdam in partial fulfilment of the requirements for the degree of

MASTER OF SCIENCE

Economics (Specialization: Behavioral Economics and Game Theory) Examiners:

Prof. Dr. Joep H. Sonnemans October 2015

εὖ γὰρ ἴσθι, ἦ δ᾽ ὅς, ὦ ἄριστε Κρίτων, τὸ μὴ καλῶς λέγειν οὐ μόνον εἰς αὐτὸ τοῦτο πλημμελές, ἀλλὰ καὶ κακόν τι ἐμποιεῖ ταῖς ψυχαῖς.

'For know well', he said, 'o dearest Kriton, that to not speak well is not only sinful by itself, but lets evil intrude into the soul.'

Table of contents

VERKLARING EIGEN WERK ... 4

LIST OF FIGURES ... 4

LIST OF TABLES ... 4

ABSTRACT ... 4

1. INTRODUCTION ... 5

1.1. Internet piracy ... 7

1.2. Prime, but prime with caution ... 10

1.3. Priming in dishonesty research ... 12

1.4. Explaining human dishonesty: models of honesty ... 15

1.5. Research Question and motivation ... 16

2. MATERIALS AND METHODS ...17

2.1. Participants ... 17

2.2. Design of the experiment and relation to literature ... 17

2.3. Analysis ... 21

3. RESULTS ...22

4. DISCUSSION ...28

5. CONCLUSION ...34

REFERENCES ...34

APPENDIX ...42

I Gender ratio differences between the groups ... 42

Verklaring eigen werk

Hierbij verklaar ik, Roelof van Dijk, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.

Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden genoemd.

De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.

List of figures

Figure 1: The random number generator on http://random.org. 20

Figure 1: Cumulative distribution of the reported number (all participants). 23

Figure 2: Cumulative distribution of the reported number of both control groups. 24

Figure 3: Cumulative distribution of the reported numbers of female participants. 25

Figure 4: Cumulative distribution of the reported numbers of male participants. 26

List of tables

Table 1: Recent findings in dishonesty research. 14

Table 2: Coding of piracy frequency answers. 21

Table 3: Participant demographics. 22

Table 4: Means of reported numbers in both treatments. 22

Table 5: Means of reported numbers in both treatments by gender. 23

Table 6: Average alienation scores for the participants in the four different groups. 27

Table 7: Reported number means of female participants by recent piracy activity. 27

Table 8: Reported piracy frequency by group and gender 27

Acknowledgements

I am grateful to Prof. Dr. Joep Sonnemans for his patience and guidance. In addition, I am indebted beyond measure for the unwavering support I received from my partner Liese Van Gompel.

Abstract

Small acts of dishonesty such as petty theft, tax fraud or piracy are committed on a large scale, and collectively cause significant economic damage. Internet piracy is prevalent, especially among young adults. Behavioral research regarding human dishonesty has established that dishonesty can be influenced by priming with subconscious contexts or cues. Recent research has given evidence that priming participants with a software piracy memory increases dishonesty in the form of cheating rates (Chiou, W.-B., Wan, P.-H. & Wan, C.-S. (2012), "A new look at software piracy: Soft lifting primes an inauthentic sense of self, prompting further unethical behavior". International Journal of Human-Computer Studies, Vol. 70 No. 2, pp. 107–115). I extended this research to video piracy, one of the most common forms of Internet piracy, using an online experiment with 573 participants. It was found that priming participants with a recollection of a recent video piracy act might decrease dishonesty in male participants (p < 0.1), but has no effect female participants. This finding contradicts earlier findings, and can neither be explained using perceived authenticity as a mediator nor using the common theoretical frameworks of moral self-regulation or moral disengagement. Furthermore, it was found that considering both genders collectively no effect can be observed. This underlines the importance of considering gender differences in behavioral priming research.

1. Introduction

Honesty is a mysterious phenomenon. According to classical economical theory, a homo economicus would never pass on an opportunity to gain an economical advantage for herself, if potential benefits outweigh any negative consequences. Therefore, from this traditional point of view, it is surprising that people do not always cheat when given the opportunity to cheat without punishment. Yet, while swindlers and fraudsters such as Bernie Madoff are unmasked occasionally, a large majority of the general population does not commit large-scale fraud. One could even state that most humans are honest most of the time. At the same time, numerous real-life examples show that humans are far from morally perfect: the damage caused by small dishonest deeds like tax and insurance fraud, retail theft and employee theft, might be small for individual cases. Summed up, these small, individual transgressions cause significant economic damage and impose a large cost on society. For retail companies, it is estimated that losses of goods before the point of sale, the so-called inventory shrinkage or shrink, is around

1.3% (Checkpoint: Global Retail Theft Barometer 2014), and across all industries it is estimated that companies typically lose around 5% of their total revenue to fraud each year. The latter represents an economic damage of more than $3.5 trillion each year. (Association of Certified Fraud Examiners: Global Fraud Study 2012). The phenomenon of small-scale individual theft carries over to the online domain, and individual acts of the illicit distribution of copyrighted digital material, or Internet piracy, have been blamed to collectively cause significant economic damage.

Attempts to explain dishonest behavior and attempts of moral guidance to reduce it are of all times. The Bible, Torah and Koran all explicitly condemn and forbid lying, and philosophers such as Plato and Kant have written treatises on dishonesty (Kant; Plato). In recent decades, scientific research into cheating, dishonesty and deceptive behavior has intensified. This research is two-pronged: attempts have been made to find theoretic models (e.g. Mazar et al. 2008) that explain dishonest behavior, and experiments have been designed to measure its prevalence and identify circumstantial levers that influence dishonest behavior. Such experiments are based on the concept of subconscious priming, which denotes the manipulation of human decision-making by certain contexts or cues. This research usually has the dual aim of understanding human behavior and decision-making on the one hand and finding effective measures to decrease dishonesty on the other hand.

In the last decade the interest in policy decisions based on research findings has increased tremendously. Examples of this trend are the establishment of the Behavioral Insights Team by the UK cabinet office in 2010 and the Social and Behavioral Sciences Initiative of the US White House Office of Science and Technology Policy in 2014 (both are also commonly referred to as ‘Nudge Unit’, after the popular book by Thaler and Sunstein (Thaler & Sunstein 2008). Behavioral-based public policy is so popular that the 2015 world development report by the World Bank Group was titled “Mind, Society and Behavior” (World Bank Group: World Development Report 2014). US president Barack Obama signed an executive order titled “Using Behavioral Science Insights to Better Serve the American People” in September 2015 (C.F.R. 2015), which explicitly encourages all parts of the US government to identify and implement policies based on behavioral economics research.

Large corporate scandals such as the infamous Enron case are frequently characterized as the cumulative escalation of small dishonest acts increasing in scope (McLean &

Elkind 2013). Along the same vein, recent experimental findings in dishonesty research point towards the potential of minor acts of dishonesty to trigger a slippery slope (Welsh et al. 2015). A related experimental result is that reminding people of their own past dishonest behavior can exacerbate their dishonesty momentarily, for instance by increasing experimental cheating rates (Mazar et al. 2008). In light of behavioral-based policy planning, this raises questions about potentially nefarious second-order effects of widespread illegal behavior such as Internet piracy: while individual acts of illegal downloading are often perceived as petty, the damage caused by them could transcend the direct economic losses if these acts cause large swaths of the population to cheat, on average, more. In comparison with people that do not partake in Internet piracy, those who pirate have been found to demonstrate less ethical concern and to be more likely to engage in other illegal behavior (Robertson et al. 2012). A recent publication finds evidence of a dishonesty-increasing effect of priming participants with a recent act of illegal software use, or soft-lifting (Chiou et al. 2012). However, for large parts of the population, software piracy is much less common than piracy acts concerning consumption of entertainment forms such as movies or TV shows, music or (e-)books. In the following, I will first discuss the prevalence of Internet piracy and attitudes towards it. Subsequently, I will expand on the priming research methodology, its application to dishonesty research in general and Internet piracy in particular. I conclude with my research hypothesis, which was tested with an online experiment.

1.1. Internet piracy

The protection of Intellectual property rights (IPR) can be considered a cornerstone of modern economies - in Europe, 89% of total external trade and 39% of total economic activity was generated by industries that lean heavily on IPR protection (EUROPOL 2015). A particular phenomenon of our times is the rampant breach of IPRs in the on- and offline world. Counterfeit goods are readily available in most countries, and authorities report an exponential increase in the amount of seized counterfeit products. The OECD has estimated that international trade volume of counterfeit goods was $250 billion dollar in 2007, and this number does not take domestic counterfeit production into account (EUROPOL 2015). The rise of the Internet has transformed the distribution of entertainment media, and has created a new type of IPR infringement: Internet piracy. This umbrella term generally refers to the unauthorized distribution, or pirating, of digital products such as movies, music recordings and software. In the last

decennium, increasing connection speeds, improved interfaces, new devices and an increase in digital literacy have promoted the prevalence of digital piracy (Gopal et al. 2004; Bezmen & Depken 2006; Andrés 2006; Peitz & Waelbroeck 2006).

Damage caused by Internet piracy

Estimating the total amount of economic damage caused by digital piracy is far from trivial, since it is in many cases unclear whether the infringing consumer would pay the full price for the pirated content had it not been available for free. Example figures on yearly economic damage caused by piracy amounts to up to $12.5 billion for pirated music (Siwek 2007, US firms only), 20.5 billion for pirated movies (Siwek 2006, US firms only) and $62.7 billion for pirated software (BSA and The Software Alliance 2014, worldwide). However, there are at least four major problems that undermine the credibility of studies that estimate losses caused by digital piracy. Firstly, the numbers are generally published by parties that have a commercial interest in overemphasizing the economic damage, such as the Recording Industry Association of America (RIAA) or the International Federation of the Phonographic Industry (IFPI). Therefore, these figures cannot always be taken at face value. Secondly, the estimation of such numbers relies heavily on the assumption that illicit downloads replace legal products. Thirdly, since the numbers refer to leisure media spending, it is not unreasonable to assume that the funds saved by digital pirates are partly spent on other leisure expenses. Moreover, it has been postulated that piracy represents a sampling function for consumers to make an informed purchasing decision, which does not have to be detrimental to sales (Chellappa & Shivendu 2005). Some researchers indeed conclude that no significant negative effect of piracy on sales can be found (Hammond 2014; Oberholzer‐Gee & Strumpf 2007; Martikainen 2014). Piracy can aid firms to retain a high price by serving customers with a low willingness to pay without compromising the price level needed to sustain development (Cremer & Pestieau 2009) and increased piracy prevention can stifle innovation (Jaisingh 2009; Mason 2008). However, others conclude that piracy does have a negative total impact on media sales based on rigorous analysis using country-level panel data (Zentner 2010). In a review of the relevant literature, Smith and Telang conclude that all of the publications in major peer-reviewed journals unanimously find a significant negative effect on sales (Smith & Telang 2012). For a more skeptical review the reader is referred to Kariithi (Kariithi 2011).

Prevalence of Internet piracy

Internet piracy is a widespread phenomenon: in 2012, the illicit transfer of copyrighted material represented around 24% of the total bandwidth used by all Internet users in North America, Europa and Asia-Pacific. Almost 26% of the total Internet population in these regions sought explicitly for infringing content, and this figure seems is on the rise (Price & Envisional 2011). At the same time, Internet piracy is everything but homogeneous: not only do prevalence rates depend strongly on demographic factors and cultural differences, but also on the legal situation and Internet penetration. Different types of pirated content (e.g. songs, movies and TV show episodes, books, software and academic publications) also imply different consumption patterns ranging from daily use (software), occasional consumption (academic publications and music) to one-time consumption (books, video content). It has been estimated that 43% of all software installed on personal computers is not properly licensed (BSA & The Software Alliance 2014). Movie and music piracy are similarly prevalent, yet reliable numbers are hard to obtain. Results of self-reporting are potentially biased through self-selection of the respondents, with respondents either not answering truthfully or refraining to answer at all. Nevertheless, samples among high school and college students frequently find that more than half of respondents admit to partaking in Internet piracy (Robertson et al. 2012; Morris & Higgins 2009; Gunter et al. 2010; Siegfried 2004; Cox & Collins 2014; Yu 2010). Contrary to criticism of the representativeness of student samples, there are indications that these findings are consistent for both student and non-student samples (Krawczyk et al. 2015). Video piracy is one of the most common form of piracy: in the Netherlands, 25.8% of the population was estimated to have pirated at least one movie in the year 2014, and this figure rises to 56% for the age category of 16-26 years. Motivations for and attitudes towards piracy

In a large survey among inhabitants of Finland, the largest drivers for piracy behavior were financial motivations, access to material before release and the belief that the external effects of piracy are positive (such as direct distribution from artist to customer). The largest inhibitor of piracy behavior was skepticism about the quality of the pirated media, while awareness of legal status and possible consequences was surprisingly not negatively but positively correlated with piracy behavior (Cox & Collins 2014). Tech-savviness, risklessness, and peers participating in piracy were found to be the main motivations correlated with Internet piracy among respondents in the US and UK, while factors such as gender, ethnicity, income and frequency of Internet use were unrelated (Shanahan & Hyman 2010). However, other studies have found that male

participants have a more favorable attitude towards and are more likely to participate in piracy behavior (Coyle et al. 2009; Bhattacharjee et al. 2003; Cox & Collins 2014; Moon et al. 2010; Navarro et al. 2014; Malin & Fowers 2009; Gopal et al. 2004). For factors such as age and income, overall findings are less conclusive, but higher income and higher age have been linked to a lower occurrence of piracy (Cox & Collins 2014). The findings of research into consumer attitude towards media piracy are consistent; a large majority of respondents do not feel guilty about or disapprove of piracy (Stephens et al. 2007; Siegfried 2004; Ingram & Hinduja 2008; Lysonski & Durvasula 2008). Among British teenagers, 70% did not feel guilty about downloading music for free (Human Capital & Marrakesh Records 2009). In a large survey among teenagers and young adults in Belgium, the statement “positive attitude towards torrent downloading” scored 4.98 on a 7-point Likert scale (De Corte & Van Kenhove 2015). The same survey also gives an indication of different attitudes towards piracy and their relative prevalence: the researchers classify their participants into four different groups: anti-pirates (28.5%) that rarely pirate, conflicted pirates (14.7%) that pirate infrequently but still deem piracy ethically unacceptable and feel guilty when pirating, cavalier pirates (27.7%) that have positive views towards piracy and have low levels of guilt even if they recognize piracy as unethical, and die-hard pirates (29.1%) that do not deem piracy unethical, experience the least guilt and pirate the most. In summary, the piracy universe is diverse and opaque, both in prevalence and behavior characteristics (Cox & Collins 2014; De Corte & Van Kenhove 2015).

1.2. Prime, but prime with caution

In the last decade, a host of lab experiments has been designed to identify levers that can subconsciously influence human decision-making. A common technique is using conceptual priming, i.e. the activation of a cognitive representation that influences behavior across contexts (Bargh & Chartrand 2000). This priming can be conscious or non-conscious, with the latter often achieved using disguised experiment designs. Generally speaking, such experiments are structured such that at least one treatment group and one control group are exposed to different conceptual primes. Subsequently, the effect on their decisions is measured. Effects that have been examined using priming techniques are varied and range from the intuitive to the surprising. For instance, reminding people of their goals unsurprisingly results into a more active pursuit of these (Ferguson & Bargh 2004). Priming using symbols, concepts or locations can influence associated behavior, e.g. it has been found that priming with the Apple logo can make people more creative (Fitzsimons et al. 2008), recalling religious concepts can

make people more altruistic (Shariff & Norenzayan 2007), and participants voting at a school are more like to support school funding (Berger et al. 2008). Another common priming technique is to expose participants to certain concepts by asking them to unscramble themed sentences, to read a themed story or to recall a particular situation. Such primes can result in effects like lowered cognitive performance after reading a story about a hooligan (Appel 2011) or a preference for sanitation wipes after recalling an unethical deed (Zhong & Liljenquist 2006).

Nevertheless, priming is much more fickle than a perfunctory overview could make believe. Priming effects can be highly sensitive to variations in subject demographics and methodology, and subtle nuances can elicit vastly different outcomes. A telling example of the intricacies of priming research is the disparity between the effects of priming subjects with a thematical category versus priming with a particular example. While priming participants with the category of superheroes or professors induced higher helpfulness or cognitive performance, participants primed with extreme examples such as Superman and Einstein exhibited lower scores. (Nelson & Norton 2005; Dijksterhuis & van Knippenberg 1998). Strong criticism which has been voiced regarding priming research concerns an alleged lack of reproducibility, which has caused the field at large to become controversial. Specific findings of well-known publications have been questioned (Appel 2011; Banerjee et al. 2012; Aquino et al. 2011) after their results could not be replicated by other researchers using high-powered designs (Shanks et al. 2013; Brandt et al. 2014; Vranka & Houdek 2015). Moreover, a recent large-scale replication study (Open Science Collaboration 2015; Klein et al. 2014) was unable to replicate a significant effect for more than half of the examined 100 well-known studies. Such high-profile nonreplication publications are fanning the recent debate on both the validity of specific priming effects in particular and the validity of the priming methodology in general (Doyen et al. 2012; Pashler et al. 2012; Shanks et al. 2013). Nevertheless, proponents of priming research argue that failures to reproduce findings do not decrease the validity of the initial findings, nor do they discredit the methodology in general. These authors maintain that such reproduction failures should rather serve as signs that there are still unknown modifiers, gaps in the theoretical understanding and possible overestimations of the robustness of priming effects (Dijksterhuis 2014; Cesario 2014; Locke 2015). Therefore, the future of priming research will hinge on theory building (Dijksterhuis 2014; Locke 2015), increased efforts towards self-replication and the moderation of expectations, especially regarding the generalizability of findings. Cesario states that searching for

invariant priming effects is a futile endeavor, since the influence of mediating environmental or contextual factors is inherent to the complexity of the human mind (Cesario 2014). The same author expresses two-fold caution regarding replication practice: Firstly, replication by independent labs can only add value if – and only if – theoretical understanding allows for accurately gauging moderating effects. Secondly, direct replication is to be valued much higher than the far more popular practice of conceptual replication, the credibility of which can be undermined by ‘researcher degrees of freedom: the a posteriori selection of hypotheses, data samples and analysis methods. If many different measures are included and subsequently only significant findings are reported, this freedom can lead to psychological findings that constitute false positives (Simmons et al. 2011). A proposed mitigation strategy would divide research into exploratory and confirmatory experiments, with the latter type performed with methodologies that are publicly announced in advance (Wagenmakers et al. 2012). Along these lines, the present work is to be interpreted as purely exploratory. Another specific point of criticism is targeted at the tendency among researchers to refrain from publishing inconclusive or negative experiment results, which can create a selection bias in the published body of work. This problem is by no means limited to behavioral economics and/or psychological sciences, and exacerbated by the reluctance of prestigious journals to publish such null findings (Simmons et al. 2011).

1.3. Priming in dishonesty research

Despite their recent controversy, priming experiments have been proven to be a popular tool to examine human dishonesty. The core of such experiments often consists of a prime/control stage and an honesty test as dependent measure. More advanced designs use several treatments or include an extra stage that includes a mediating measure, for example perceived authenticity. A powerful way to incentivize dishonest behavior is to allow participants to cheat for a higher experiment payoff, e.g. by misreporting the amount of solved problems (Mazar et al. 2008) or misstating a random event that determines the payoff such as a die roll (Diekmann et al. 2011). Using such experiments, it has been found that primes such as recalling the 10 commandments or being in the presence of other people can decrease cheating rates (Mazar et al. 2008; Kroher & Wolbring 2015), while wearing sunglasses, being mentally exhausted or striking expansive postures can increase them (Gino et al. 2011; Yap et al. 2013; Gino et al. 2010). These and other findings are summarized in Table 1 below. A subset of dishonesty-related priming research is interested in examining the connection between

past dishonest behavior and future dishonest behavior, such as cheating in experiments. However, dishonesty is a dynamic concept, with large variations both between people and within each individual. It is not obvious whether priming dishonest behavior should trigger more or rather less dishonest behavior, in other words, whether humans generally tend towards consistent or towards compensatory behavior. However, this might not be the right question: based on the experiences in priming research on other topics, one must be cautious with such blanket statements. A refined question would ask under what circumstances and which specific primes might bolster or diminish dishonesty behavior.

Table 1: Recent findings in dishonesty research, 1_{same amount as human} 2_{temporary effect.}

Publication Lever/Prime Effect

(Mazar et al. 2008) Reminder of (non-existent) honor code Decreased cheating

idem Recall: Ten Commandments

(Gillath et al. 2010) Recall: friendship

(Jordan et al. 2011) Recall: personal immoral act

(Aveyard 2014) Recall: Athan (Islamic call to prayer)

(Hoffman et al. 2015) Social control: human in the room

(Bering et al. 2005) Social control: story of ghost in room

(Hoffman et al. 2015) Social control: robot in the room1

(Kroher & Wolbring 2015) Social control: sharing a die2

(Gino & Mogilner 2013) Self-awareness: mirror

(Vincent et al. 2013) Self-awareness: mirror

(Vohs & Schooler 2008) Encouraging belief in free will

(Gino et al. 2009) Awareness: cheating allowed

(Caruso & Gino 2011) Closing eyes

(Shu et al. 2012) Signing form at top instead of bottom

(Randolph-Seng & Nielsen 2007) Sentence unscrambling: religions words

(Jordan et al. 2011) Recall: personal moral act Increased cheating

(Cohn et al. 2013) Recall: criminal identity

(Chiou et al. 2012) Recall: software piracy act

(Vincent et al. 2013) Recall: positive experience

(Cohn et al. 2014) Recall: professional identity as a banker

(Gino & Pierce 2009) Money prime: seeing large pile of cash

(Kouchaki et al. 2013) Money prime: descrambling sentences

(Gino & Mogilner 2013) Money prime: song lyrics

idem Money prime: counting money

(Mead et al. 2009) Willpower: exhaustive task

(Gino et al. 2011) Willpower: exhaustive task

(Gino et al. 2009) Social control: destroying form

(Zhong et al. 2010) Social control/anonymity: Darkness

idem Social control/anonymity: Wearing sunglasses

(Gino et al. 2013) Social control: choice to avoid control

(Kroher & Wolbring 2015) Social control: Increasing anonymity

(Gino et al. 2009) Learning: directly observing cheating

(Diekmann et al. 2011) Learning: being told that others have cheated

(Fischbacher & Föllmi-Heusi 2013) Learning: repeating experiment

(Kroher & Wolbring 2015) Learning: seeing others have cheated

(Mazar & Zhong 2010) Buying green products

(Gino et al. 2010) Authenticity: wearing fake sunglasses

(Mazar et al. 2008) Distance: tokens as intermediary reward

(Schweitzer et al. 2004) Unmet goals

(Vohs & Schooler 2008) Encouraging belief in determinism

(Houser et al. 2012) Feeling treated unfairly

(Yap et al. 2013) Expansive postures

(Welsh et al. 2015) Slowly increase payoff

1.4. Explaining human dishonesty: models of honesty

Dishonest behavior can be explained using classical economic theory: according to expected utility theory, a perfectly rational human will weigh the benefit gained from acting dishonestly against possible repercussions (Von Neumann & Morgenstern 1944). Therefore, the choice involves a trade-off between economic gain and potential negative consequences, social or otherwise (Hechter 1990; Lewicki 1983; Becker 1968). In this trade-off, potential punishment plays an important role, in line with deterrence theory (Ehrlich 1973). A second established theoretical model is Ajzen’s theory of planned behavior (Ajzen 1991), which is applied to dishonest behavior in (Beck & Ajzen 1991). According to this model, any action is based on three distinct sources: behavioral intentions based on social pressure, norms and values (Campbell 1964); personal attitudes towards the action; and expected self-efficacy in performing the action. A third theoretic model is based on a self-concept (Aronson 1969) or moral identity (Blasi 1980; Shao et al. 2008; Hardy & Carlo 2005) that individuals strive to maintain through their actions to avoid cognitive dissonance (Festinger et al. 1978; Mazar et al. 2008). In an attempt to explain why people do cheat, but not to the maximal extent, one central outcome of this model is that people cheat enough to enjoy an increased payoff, but not so much that it would threaten the perception of their own honesty. That last part is essential: it has been known for decades that not only do preferences influence behavior, but also the opposite: that the self-perception is influenced by past behavior (Bem 1972). This fact is emphasized again in the theory of self-concept maintenance by (Mazar et al. 2008), which shares elements with interpreting actions as a form of self-signaling (Bodner & Prelec 2003; Mijović-Prelec & Prelec 2010; Chiou et al. 2012). How do people react when confronted with their own dishonest behavior? Well-know findings are in line with the theory of self-concept maintenance: increased cheating rates were observed after reminding participants of their criminal identity (Cohn et al. 2013), professional identity as a banker (Cohn et al. 2014) or a recent software piracy act (Chiou et al. 2012). What is more, recalling dishonest behavior can not only lead to increased cheating, but also to increased memory lapses concerning moral rules (Shu et al. 2011). However, there also is substantial experimental evidence for a dishonesty-boosting effect of priming with positive actions or memories. For instance, participants buying green products were more likely to cheat or steal (Mazar & Zhong 2010), participants that first helped a foreign student donated less money (Khan & Dhar 2006) and participants recalling moral behavior cheated more, while participants recalling immoral behavior cheated less (Jordan et al. 2011). The contradictory findings that

negative primes can both increase and decrease dishonest behavior can be explained using the theoretical model of moral self-regulation, which consists of both moral highlighting (consistency) and moral licensing (compensatory) effects. (Dhar & Simonson 1999; Fishbach et al. 2009; Miller & Effron 2010; Susewind & Hoelzl 2014; Vohs & Baumeister 2005; Zhong et al. 2009). Central to this model are the findings that consistent behavior is observed for participants primed with abstract concepts or temporally distant behavior, while compensatory behavior or licensing is observed after exposure to specific, personal or temporally close primes, such as recent immoral behavior (Jordan et al. 2011). A central tenet of this theory therefore is that the nature of the effect (highlighting vs. licensing) depends on the distance between the participant and the prime.

1.5. Research Question and motivation

Immoral acts of low perceived significance can pave the way for further dishonest behavior, and priming participants with a personal immoral deed has been shown to decrease their subsequent honesty. Video piracy is a common act of copyright infringement, especially among young adults. Dishonesty priming has been applied in a piracy context before, yet only limited to software piracy (Chiou et al. 2012). In their experiment the researchers primed participants that had already admitted of participating in software piracy by asking them to recall a specific, recent piracy deed. Thusly primed participants exhibited high cheating rates: they inflated their performance on a performance-paid test by 43%, which was significantly more than the rate with which the participants in the control group inflated their performance (11%). Furthermore, the researchers found that this increased dishonesty was mediated by a less authentic sense of self. To the best of my knowledge, the connection between video piracy and dishonesty has not been examined to date. This forms the basis of the following research question.

Does priming with a recent memory of video piracy elicit increased dishonesty?

Moreover, I hypothesize that if an effect similar to the findings of Chiou et al. can be found, this effect will again be mediated by a reduced feeling of authenticity.

2. Materials and methods

An experiment was designed in order to test the priming effect of recalling video piracy act on participants’ honesty. Participants were asked to complete a short survey, which included a short priming task based on recalling an act of video piracy, an honesty measurement, an authenticity measurement, a piracy frequency test and demographic questions.

2.1. Participants

The experiment was conducted using an online survey (Appendix II) designed in the Qualtrics Research Suite (http://www.qualtrics.com). Respondents were recruited online using a special survey link, which was distributed personally via e-mails, on the Facebook wall of the author, in various Facebook groups (http://www.facebook.com) and via posters on the campus of Utrecht University. The target demographic was specified as adults (15-40) in Western European countries. The survey was conducted between the 1st _{of September and the 3}rd _{of October 2015.}

2.2. Design of the experiment and relation to literature

The design of the experiment resembled the design used by (Chiou et al. 2012). These authors firstly use a disguised survey to identify participants familiar with software piracy, which were then divided into three groups primed. These groups were then using a recall task on legal software use, illegal software use and shopping experience. Subsequently, the authors measured feelings of authenticity using standardized questions. The next part consisted of a dishonesty test using the coded matrix-task test designed by (Mazar et al. 2008). My design used comparable elements, but exhibits six important distinctions: the experiment medium, the participant selection, the cohorts, the choice of control question topic, the nature of the dishonesty test and the order of the experiment elements:

1. This experiment was conducted using an online survey and not in a lab. This implied changing certain procedures such as the dishonesty test.

2. No initial selection of participants according to familiarity with video piracy was made. The familiarity with video piracy of the participants was tested in the last part of the survey.

3. I limited the design to two cohorts instead of three: a priming treatment consisting of a video piracy recall task and a control group. The main reason for this was to increase the power of the result for a limited number of participants. 4. A different control question was chosen: Chiou et al. used a control group that

recalled recent shopping experience. This was deemed problematic, since priming with monetary cues been shown to decrease pro-social behavior (Review: Vohs 2015) in terms of generosity (Chatterjee et al. 2013; Gasiorowska et al. 2012; Gąsiorowska & Hełka 2012; Park & Vohs 2013; Vohs et al. 2006), willingness to volunteer (Banerjee et al. 2012; Pfeffer & DeVoe 2009; Vohs et al. 2006), helpfulness (Gasiorowska et al. 2012; Gąsiorowska & Hełka 2012; Guéguen & Jacob 2013; Vohs et al. 2006) and empathy (Molinsky et al. 2012). More specifically, priming participants with money has been experimentally shown to explicitly increase cheating (Gino & Mogilner 2013) and lying (Kouchaki et al. 2013). The original control topic could therefore potentially increase dishonesty in the control group, reducing the difference between both groups. In this experiment participants in the control group were therefore were asked to recall a moment when they read a paper book, newspaper or magazine. This choice has the advantage of keeping the treatment and the control question context similar, since both concern leisure media consumption. Furthermore, in contrast to recalling a shopping experience the act of recalling reading a paper medium is less directly connected to the expense of money. Lastly, the question explicitly excluded e-books, which could have potentially be pirated.

5. A different dishonesty test was chosen. In an online environment, it is challenging to alleviate participants’ concerns about control of their possible dishonest behavior, since it is not unjustified to assume that every interaction with a survey is tracked. This would threaten the experiment results, since an increased feeling of social control has been shown to strongly decrease dishonesty in both lab and field experiments (Kroher & Wolbring 2015; Gino et al. 2013; Haley & Fessler 2005; Mazar et al. 2008; Zhong et al. 2010; Olken 2007). Therefore I did not replicate the matrix task as dishonesty test, but chose a test based on reporting a random event.

6. The order of the elements differed from the structure chose by Chiou et al. (Chiou et al. 2012). In their experiment the authors placed the authenticity questions after the priming task and before the dishonesty test. Furthermore, question about the frequency of piracy behavior were part of a preliminary survey. Due to concerns of additional, undesirable priming through these personal questions

concerning piracy behavior and perceived authenticity were both positioned after the dishonesty task.

The survey therefore consisted of the following parts: A. Subject priming / control

B. Honesty test

C. Authenticity measurement D. Familiarity with video piracy E. Demographic questions F. Payout scheme

Subject priming

All visitors to the website containing the survey were randomly assigned to the treatment or the control group. In the video piracy treatment group (henceforth: piracy group), participants were asked to recall a specific moment in which they had watched a movie or TV show episode that was not purchased through legal means. Afterwards, they were asked to give details such as the name, the duration and five keywords associated with the movie. The control groups answered the same questions about an experience reading a paper book, magazine or newspaper. The recall technique is a common priming technique in behavioral research (Vohs et al. 2006; Zhong & Liljenquist 2006; Williams & Bargh 2008; Chao et al. 2011; Chiou et al. 2012).

Honesty test

The dishonesty task in this experiment was to randomly generate the possible personal lottery prize between 1 and 100 Euros. At this step in the survey, participants were asked to go to a third party website (http://www.random.org/) to generate a number between 1 and 100, which they were asked to enter into the survey (Figure 1). This design made it easy for participants to deceive by either fabricating a high number or repeatedly generate a number until a sufficiently high number occurs.

Figure 1: The random number generator on http://random.org.

The fact that this number generator was located on a different, independent website outside of the survey strengthens the perceived lack of control on dishonest behavior. Differences in honesty between the two groups were expected to emerge between the distributions of the reported numbers, which can be compared to the real, uniform distribution of the random generator. This design resembles a laboratory experiment design which is based on reporting the outcome of a die roll (Fischbacher & Föllmi-Heusi 2013; Shalvi et al. 2012; Shalvi et al. 2011). Previous experiments have repeatedly found that participants cheated on average, but not the maximum extent. Therefore, the larger range of answers in this design (1-100) offered participants the ability to cheat on a more nuanced scale than the range offered by a six-sided die. Secondly, this design allows the possible payoff to directly correspond to the reported number, therefore potentially increasing saliency.

Authenticity measurement

The authenticity measurement was performed using four established statements for measuring self-alienation from (Wood et al. 2008), which the participant had to grade on a 7-point Likert-scale (1: Not at all to 7: Very much): ”Right now, I don’t know how I really feel inside”, ”Right now, I feel as if I don’t know myself very well”, ”Right now, I feel out of touch with the ‘real me’”, and ”Right now, I feel alienated from myself.” Two bogus questions were added to measure attentiveness.

Familiarity with video piracy

To assess the frequency with which the participants consumed pirated movies or TV show episodes, participants were asked to recall the last three movies or TV show episodes and indicate via what medium they had consumed it. To elicit honest

responses, neutrally formulated answer choices were given for legal ways (i.e. cinema, regular TV programming, a website with a paid account, an official TV-network website or a bought DVD) and illicit ways (a file downloaded through the BitTorrent protocol, a streaming website without a paid account, a file downloaded via an unpaid direct download). For all online media example websites were given. Afterwards, the answers were coded for legality (Table 2).

Table 2: Coding of piracy frequency answers.

Legal High chance of piracy Unknown

Cinema Streaming with paid account YouTube.com/Vimeo.com

Regular TV Downloaded via unpaid _{direct download} Other

Streaming website with

paid account Downloaded through the BitTorrent protocol

Official TV-network website

Bought DVD

Demographic questions

The following demographic information was requested: age, highest degree, student status, country of residence, country of origin. Additionally, participants could indicate whether they had answered the survey attentively and completely

Personal ID and payout

Each participant generated a personal ID consisting of eight digits. One participant number was chosen as winner and the first five digits of the winning ID were published on a designated website. It was the participants’ responsibility visit this website after a designated date to check whether they had been chosen as the winner, and confirm their identity by reproducing the other three digits. This design was necessary due to the anonymous nature of the survey.

2.3. Analysis

The results of the survey were analysed in Microsoft Excel 2011 (Microsoft 2010) and R version 3.2.2 (R Development Core Team 2012). The distribution of reported numbers between different groups were compared to each other and to the known uniform distribution generated by the random number generator. Significance of differences between the distributions was determined using Mann-Whitney U tests.

3. Results

3.1. Participant demographics

In total, 653 respondents completed the survey. Of these, 80 participants were excluded due to entering an invalid negative number (1), age (10), reporting never having committed video piracy acts while being in the piracy group (25), self-reported lack of conscientiousness (4) or due to indicating that they did not want to be paid (40). This yielded 573 remaining participants (Table 3).

Table 3: Participant demographics.

Control group Piracy group All

Participants 304 269 573

Age 23.8±3.9 23.5±3.9 23.7±3.9

Female 64% 53% 59%

Students 76% 75% 76%

From the Netherlands 44% 47% 45%

3.2. Dishonesty results

In the experiment, dishonest behavior in the form of cheating was universal: participants in both the piracy and control group on average over-reported the randomly generated number corresponding to their (Table 4), and both number distributions differed significantly from the real uniform distribution generated by the random number generator (One-sided Mann-Whitney U test against uniform distribution, control group: p = 0.00008, piracy group: p = 0.0007). The average reported number in the control group was higher than the average reported number in the piracy group.

Table 4: Means of reported numbers in both treatments.

Measures Control group Piracy group Uniform distribution

Participants 304 269

Mean reported number 63.65 (28.56) 61.97 (28.96) 50.5 (28.58)

Proportion 91-100 20% 20.6% 10%

The large range of the random number generator allows for a more specific analysis of the dishonest behavior. A cumulative distribution of the reported numbers makes the extent of dishonest behavior visible – in this representation, the distance to the line representing the cumulative distribution of the true uniform distribution represents the magnitude of the underreported lower numbers, and the slope of corresponds to the magnitude of under- or over-reporting of numbers in a certain range. Therefore, if

cheating occurs, the reported cumulative distributions of the reported numbers are expected lie below the uniform distributions for lower numbers and to catch up for higher numbers. The cumulative distributions (Figure 2) indicate cheating in both groups, but both are close across the whole range.

Figure 2: Cumulative distribution of the reported number (all participants).

A two-sided Mann-Whitney U test between the number distributions reveals that there is no significant difference between the piracy group (median = 68) and the control group (median = 71), Z = 0.71, p = 0.475, r = 0.03.

Gender differences

Although participants were assigned randomly to both groups, the gender ratios between the two groups differ (piracy group: 53% female, control group: 64% female). This discrepancy originated in a lower survey completion rate among male participants in the control group and a slightly higher number of necessary exclusions among male participants in the control group (See Appendix I).

Based on this discrepancy, both genders were analysed separately. Splitting up the participant pool into male (n=227) and female (n=336) participants reveals a marked difference in treatment effect (Table 5).

Table 5: Means of reported numbers in both treatments by gender.

Gender Measures Control group Piracy group Uniform distribution

Female Participants Mean reported number 61.74* (28.47) 64.39* (27.14) 192 144 _{50.5 (SD = 28.58)} 10 %

Male Participants Mean reported number 66.29* (28.69) 59.80* (30.48) 105 122

Proportion 91-100 25% 24%

The reported numbers of all four groups differs significantly from the uniform distribution, therefore, dishonesty is likely to be present in all groups. However, the dishonesty levels between the two control groups seem to differ. The cumulative distribution graph for both control groups (Figure 3) confirm that cheating is more prevalent among male participants.

Figure 3: Cumulative distribution of the reported number of both control groups.

A one-sided Mann-Whitney U test (Z = 1.46, p = 0.072, r = 0.085) between the female control group (median = 68) and the male control group (median = 75), indicates that this effect is only significant at the p = 0.1 level.

A second observation is that the treatment seems to have an opposite effect on participants of different genders. For female participants, the priming effect of the piracy group seems to increase dishonesty, marked by the higher average reported number for the piracy group. However, for male participants the effect is the opposite, as indicated by a lower average reported number (Table 5). To display the actual distribution of reported numbers, again cumulative distributions are used. For female participants, the piracy recall treatment seems to slightly increase the dishonesty of the

0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 Cumulative proportion Reported number Uniform distribution Control-Female (n=192) Control-Male (n=105)

participants, as illustrated by the downwards shift of the cumulative distribution function (Figure 4).

Figure 4: Cumulative distribution of the reported numbers of female participants.

However, a one-sided Mann-Whitney U test (Z = 0.68, p = 0.249, r = 0.037) between the female piracy (median = 70) and the female control groups (median = 68), indicates that this effect is not significant. It therefore has to be concluded that there is no significant observable effect of the treatment on female participants. For male participants, the piracy recall treatment seems to elicit a more honest response in male participants, as illustrated by the upwards shift of the cumulative distribution (Figure 5). 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 Cumulative proportion Reported number Uniform distribution Control group (n=192) Piracy group (n=144)

Figure 5: Cumulative distribution of the reported numbers of male participants.

A one-sided Mann-Whitney U test (Z = 1.46, p = 0.073, r = 0.097) between the male piracy group (median = 62.5) and the male control group (median = 75), indicates that this effect is only significant at the p = 0.1 level.

In summary, three effects can be observed:

• there is dishonesty in all four groups: the numbers reported by all four groups are significantly higher (p < 0.01) than those generated by the random number generator.

• the baseline dishonesty level is higher for male participants than for female participants: the numbers reported by the male control group are significantly higher (however only at p < 0.1) than the numbers reported the female control group.

• the treatment reduces dishonesty in male participants but not in female participants: the numbers reported by the male piracy group are significantly lower (however only at p < 0.1) than the numbers reported the male control group, whereas for female participants they are not.

3.3. Other results

In the latter part of the survey, perceived authenticity was measured and certain demographic information was requested.

Authenticity results 0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97 10 0 Cumulative proportion Reported number Uniform distribution Control group (n=105) Piracy group (n=122)

In general, the participants’ answers to the authenticity questions were highly consistent (Cronbach’s α = 0.81). Therefore, the answers to the four questions were averaged to obtain an indication of perceived authenticity (with lower levels corresponding to higher perceived authenticity). However, T-tests showed that there was no significant difference (p > 0.1) between the reported alienation scores between any of the four groups (Table 6).

Table 6: Average alienation scores for the participants in the four different groups.

Gender Control group Piracy group

Female 2.68 (1.15) 2.69 (1.23)

Male 2.61 (1.15) 2.65 (1.05)

Piracy frequency results

On average, male participants reported a higher piracy rate than female participants (Table 8), and this difference was highly significant (two-sided T-test, t = 3.06, df = 561, p = 0.002). However, no significant effect of individual piracy frequency on the treatment effect could be determined for either gender. Nevertheless, the dishonesty levels of female participants that did pirate at least one of the three most recent consumed pieces of video content was significantly (at p < 0.1) lower than the dishonesty of female participants that did report not having pirated any of the last three consumed pieces of video content (Table 6), in both the control group (Two-sided Mann-Whitney U-Test, Z = 1.43, p = 0.077, r = 0.103) and the piracy group (Two-sided Mann-Whitney U-Test, Z = 1.84, p = 0.033, r = 0.153). For the male participants, the differences were not significant at the p < 0.1 level.

Table 7: Reported number means of female participants by recent piracy activity.

Gender # pirated out of last 3 Control group Piracy group

Female _1-3 0 65.44 (29.61) 73.37 (21.10) _{59.38 (27.72) 61.17 (28.52)}

Curiously, participants in the piracy groups of both genders reported significantly (p < 0.001) higher piracy rates than participants in the control groups (Table 8), which hints at an effect of the treatment on honesty levels in disclosing piracy activity.

Table 8: Reported piracy frequency by group and gender

Gender Control group Piracy group All

Male 1.4 ± 1.2 1.9 ± 1.1 1.6 ± 1.2 Other demographic factors

Neither age (17-27 vs. 27-37), nor student status (student vs non-student), nor country of origin (NL vs. non-NL) were found to have any significant effect on the dishonesty average (p > 0.12 for all gender-separated, two-sided Mann-Whitney U tests) with the exception of age for the female piracy group (two-sided Mann-Whitney U test, p = 0.0926).

Prize

The winning participant ID was published according to the specified protocol. However, as of late October 2015 the prize had not been claimed yet.

4. Discussion

The aim of this study was to extend the previous finding that participants primed with a software piracy memory exhibit higher cheating rates than a control group (Chiou et al. 2012). Due to the high prevalence of video piracy in the Netherlands (Leenheer & Poort 2014), this experiment was conducted with the memory of a recent video piracy act as priming treatment. Based on their refusal to participate in the lottery, 40 participants were excluded from the sample, since their salience for dishonesty differs strongly form the rest of the sample.

Initial analysis revealed that cheating was present in the control group. The average reported number in the control group was 26% higher than the average of the real, uniform distribution. This compares to previous findings of average inflation of die roll results in laboratory experiments of +40.8% (Fischbacher & Föllmi-Heusi 2013), +30.2% (Shalvi et al. 2012, Experiment 1) and +27.1% (Shalvi et al. 2011, Experiment 1). When analyzing participants of both genders separately, a more nuanced picture emerges: male participants in the control group (+31.3%) cheated significantly (at p < 0.1) more than female participants in the control group (+22.3%).

Piracy treatment effect

Regarding the piracy treatment effect, no statistically significant difference in cheating between the control group (+26%) and the piracy group (+22.7%) could be found. However, analysis of only the male participants revealed a statistically significant (at p < 0.1) dishonesty-reducing effect: participants in the piracy group (+18.4%) on average

reported a number that was much lower than participants in the control group (+31.3%). In the initial analysis of participants of both genders, this effect was masked by an opposite trend displayed by the more numerous female participants in the piracy (+27.5%) and control (+22.3%) groups. This latter difference was not, however, significant at p < 0.1.

The finding that male participants cheat to a higher degree than female participants is a marked difference to Chiou et al. (Chiou et al. 2012), who did not find a significant difference between the two genders. In their control condition, the authors find that male participants cheat more at a p = 0.1 level of significance, yet deem this finding not significant. The level of significance could be related to their total participant pool numbering only 105 participants across three different conditions (software piracy, legal software, control group). Nevertheless, the results of the present research analyzing five times as many participants (573) failed to achieve the commonly accepted significance level of p = 0.05.

The finding of the present experiment – that male participants’ dishonesty levels are significantly decreased by priming with a video piracy memory – directly contradicts the findings of Chiou et al., who observe a strong increase in cheating rates in their piracy treatment group (+43%) compared to their control group (+12%). This difference could be attributable to a number of differences between the experiments.

Firstly, the participants in the original experiment originate from Taiwan, whereas the majority of the participants in the present research originate from Western Europe. Not only are piracy rates higher in many Asian countries than in Western Europe (Price 2013), but Asians students have also been shown to be significantly more likely to justify digital piracy (Yu 2013) than American students. The latter share more cultural commonalities to Western European students participating in this research. It is therefore possible that Western European students perceive piracy as more immoral than Taiwanese students. With lower feelings of guilt, Taiwanese students might lean towards consistency (moral highlighting) whereas Western European students towards a compensatory (moral licensing) reaction to being primed with a recent piracy memory.

Secondly, the nature of the dishonesty measure differs between Chiou et al. and the present research: Chiou et al. measure dishonesty of participants overstating their performance in a matrix test, while the present research asks participants to report a randomly generated number corresponding to their potential reward for completing a

survey. The participants’ likelihood of cheating might be influenced by a different sense of entitlement towards a compensation and by the fact that the random number generator used in the present research can generate low numbers, which might be perceived as unfairly low.

Thirdly, Chiou et al. prime their participants with a software piracy memory, while the present research is focused on video piracy. Potential differences in both prevalence among and attitude towards software versus video piracy in the target demographic could lead to different reactions. However, available figures for piracy prevalence in the Netherlands according to content type are limited to movie, music and games (Schermer & Wubben 2011; Huygen et al. 2012).

Ultimately, the observed opposite nature of the treatment effect also undermines an explanation following the argumentation of Chiou et al. based on higher levels of alienation or lower levels of perceived authenticity, respectively (Chiou et al. 2012). Moreover, in this experiment no significant differences between perceived authenticity between the control and piracy groups could be observed for either gender.

Gender differences

The finding that male participants in the control group cheat to a higher degree than female participants in the control condition is in accordance with findings in literature that preferences and deceptive behavior differ between the sexes (Fosgaard et al. 2013; Erat & Gneezy 2012; Houser et al. 2012; Croson & Gneezy 2009). It has been found that men are more aware of the potential to cheat (Fosgaard et al. 2013). Furthermore, men are more willing to cheat: male bus passengers were found to be more likely to not have a valid ticket in a field experiment (Bucciol et al. 2013), male restaurant customers were less likely to return excessive change in restaurant visits (Azar et al. 2013), men were more likely to tell a selfish black lie (Dreber & Johannesson 2008; Erat & Gneezy 2012) and, of particular relevance to this study, male participants were found to cheat more in reporting the outcome of a private coin flip (Houser et al. 2012).

The observed dishonesty-decreasing effect for male participant is generally in line with the postulation that priming with temporally and socially close dishonest actions, such as the implemented specific, personal example of piracy, can trigger a compensatory effect (Jordan et al. 2011). However, it is surprising that this effect seems to be stronger for male participants than for female participants. This could have two reasons:

differences in attitude towards piracy between genders, and different sensitivity towards the priming technique itself.

A straightforward, yet fallacious interpretation of this finding in light of moral licensing theory would be that male participants feel more guilt about their recalled piracy behavior than female participants, and are therefore subsequently more honest in reporting their number than similarly primed female participants. However, this directly contradicts the common finding that men are more likely to pirate than women and, more importantly, feel less guilty about it (Mandel & Süssmuth 2010; Chaudhry et al. 2011). In their survey among Belgian young adults, De Corte and Van Kenhove find that male respondents were much more likely to be ‘die-hard pirates’ condoning piracy behavior than ‘anti-pirate’, and the opposite was true for female respondents (De Corte & Van Kenhove 2015). Furthermore, in a recent survey in the Netherlands 35.2% of all male respondents admitted to pirating video content in the last year, which compares to 22.3% for female respondents (Leenheer & Poort 2014). In accordance with these figures, male participants of the present experiment reported a higher piracy rate than female participants. In light of these findings, the premise about the participants’ feelings towards piracy might be flawed. The assumption of this research was that video piracy is perceived as immoral and elicits negative feelings such as guilt. However, it might be that a significant proportion of male participants have no negative attitude towards video piracy at all, but see a recent video piracy experience in a positive light. Potential reasons for this might be benefits of video piracy such as monetary savings, access to material that would otherwise not be available (such as brand-new movies or overseas TV shows) or ease-of-use, but could also be related to the enjoyment of the content itself. Furthermore, the high prevalence of piracy among the peer group of young adult men might cause it to be perceived as the norm, and therefore pirating might not be perceived as aberrant behavior. In such social contexts, having seen a recent new movie or TV-show episode might be perceived as an achievement and a social mark of distinction. Nevertheless, even in this case the positive affect caused by the treatment would be expected to increase dishonesty via moral licensing (Jordan et al. 2011) or moral disengagement (Vincent et al. 2013). Therefore, differences in attitude towards piracy cannot sufficiently explain the findings of this experiment. Gender differences in sensitivity towards dishonesty primes have been demonstrated by Fosgaard et al. in whose experiment men initially are more aware of the potential to cheat. Increasing awareness that cheating is possible increases cheating rates for female

but not for male participants (Fosgaard et al. 2013). In the present experiment, there might be a similar difference in sensitivity towards the prime based on yet unknown differences between the genders.

Limitations of this research

There are several limitations to this research, both in design and in participant selection. First and foremost, great care was taken to address potential shortcoming in the original design of Chiou et al. (Chiou et al. 2012), such as the nature of the control group prime and the order of the questions and. The control prime choice of reading a physical book improves on the money-related control prime used in the original paper. Nevertheless, this prime might cause uncontrolled for and unwelcome priming effects. Since reading a book is connected to learning, knowledge and potentially science, this prime might still cause an honesty-increasing effect such as demonstrated in research priming participants with science-related primes (Ma-Kellams & Blascovich 2013). A further concern is the potential insignificance of piracy behavior as a negative prime – if the primed behavior is pervasive and socially accepted it might not be perceived as dishonest or immoral.

The piracy frequency measurement was located at the end of the present survey, in the demographic information section. This choice was made to prevent possible additional, uncontrolled priming of the participants for the honesty task. Chiou et al. take a different approach – their participant pool is selected from the respondents of an prior survey on piracy habits (Chiou et al. 2012). Their approach yields more reliably individual piracy data at the cost of potential additional priming of the participants before their honesty task. In the present research, an analysis of the piracy frequency results reveals that there is a highly significant effect of the treatment on the reported piracy frequencies (p < 0.001). Since all participants were distributed randomly across the conditions, this implies that participants primed with a piracy memory are more honest in reporting their piracy behavior. Therefore, the validity of the converse analysis – the influence of piracy frequency on dishonesty – is severely threatened.

The participant pool is inhomogeneous in various ways. This survey was partly distributed via Facebook groups, there is a high probability that a significant proportion of participants might know the author personally. These participants might much be less inclined to cheat than strangers, which could disturb the results in two ways. The overall decrease of dishonesty could be responsible for the low significance of the

results. Moreover, since the gender distribution of these biased participants was not controlled for, gender differences might be based on this bias. For instance, if a high number of female acquaintances of the author participated, yet only a low number of male acquaintances, the average dishonesty level of female participants is expected to be lower than that of male participants. Both problems could be addressed by conducting a similar experiment with a larger amount of participants that do not know the researcher(s) personally.

This experiment was conducted online, and the participant group was inhomogeneous regarding age, student status and country of origin. However, these three factors did not change the distributions of the reported numbers significantly (p > 0.5). This was a disguised study, however, there was no way do exclude participants that suspected the intention behind the experiment. Adding an option for the participants to express their suspicions or express familiarity with behavioral research. However, fact that the dishonesty test occurred towards the beginning of the survey aided the disguise, and none of 30 participants in contact with the author after the survey had any suspicion about the objective of this research.

Concerns can be voiced about the salience of the reward: since only one winner was chosen out of a large pool of participants, the individual chances of winning were low. An extremely high reward (expected value of 50 Euro for a survey of 6 minutes) was chosen in order to mitigate this low expected value, and participants did not know the number of other participants. The reward mechanism devised improve anonymity increased the effort for the participants to claim the reward sharply. Therefore, concerns about the integrity of this experiment were voiced. The drawback of this design is that the winner loses her anonymity by claiming the reward – this could reduce the willingness to claim the reward.

Further research

The findings of this experiment highlight the importance of gender in dishonesty research. That the effect found among male participants was masked by the opposite (non-significant) effect among female participants raises questions about the well-known experiments by Jordan et al. and Vincent et al. (Jordan et al. 2011; Vincent et al. 2013) that do not examine or account for gender effects. Further research could expand the covered range of piracy topics and include music and e-book piracy. If for a given population differences in prevalence of and attitude towards different kinds of piracy (such as software, video music and e-book piracy) are known, these differences could