Making regulation fit by taking irrationality into account: the case of the whistleblower

Near and Miceli (1985, p. 4) describe whistleblowing as a “disclosure by organization members (former or current) of illegal, immoral or illegitimate practices under the control of their employers, to persons or organizations that may be able to effect action.” This definition is used widely across academic literature. Whistleblowing is characterized as a highly complex decision process (Chiu 2003), including a multitude of situational, individual, as well as institutional factors (Near and Miceli 1996; Mesmer-Magnus and Viswesvaran 2005; Lee and Xiao 2018). Although no formal model has emerged as of today (Lee and Xiao 2018), the general structure of the process can be characterized as shown in Fig. 1. The internal agent observes what might or might not be an actual case of fraud by another person of the organization. The potential whistleblower can now choose to speak up, thus exercising “voice”, remain “silent”, or leave the organization, choosing “exit” (Hirschman 1970). This decision is highly dependent on various factors of individual, situational, and institutional nature. Choosing any other option but “voice” ends the process, and the agent never becomes a whistleblower. Choosing “voice” instead, there are two options for her to choose from in the next stage of the process. She can blow the whistle either internally or come forward to an external channel.¹ The disclosure of information to any third party will evoke one of three reactions by that party: sanction, reward, ignorance, or a combination thereof (Near and Miceli 1985). Depending on the outcome, the whistleblower will either accept or return to the previous stage, repeating the process. For example, in the case of internal whistleblowing: after no action followed the disclosure, the whistleblower might be compelled to take her claims outside of the organization (Callahan and Dworkin 1994; Gray 2004). At the same time, the person accused of the fraudulent act might be compelled to take actions if she becomes aware of the disclosure and the whistleblowers’ identity.

To understand this process and especially the influential factors has been at the core of academic research for decades. Research in the field of social sciences and behavioral economics started with individual and situational factors such as gender, age, occupation, and type of wrongdoing (Miceli and Near 1984; Near and Miceli 1985; Dworkin and Baucus 1998). The last two decades saw a rise in studies concentrating on institutional factors such as culture as defining characteristics. Mixed results suggest that the influence of culture cannot easily be depicted (Sims and Keenan 1999; Keenan 2002, 2007; Park et al. 2008; Cheng, Karim and Lin 2015). Literature outside of the USA on this topic is scarce, which is why scholars argue that research originating from other countries is necessary to advance our understanding of the whistleblowing process (Vinten 2004; Brown et al. 2014; Gao and Brink 2017).

It is usually the USA that is associated with strong whistleblowing laws. The Enron scandal can be seen as a starting point in the development of laws for whistleblower, although the False Claims Act of 1986 already rewarded people who spoke out externally when the government was defrauded by a private entity (Callahan and Dworkin 1994). After the famous Enron case and the subsequent collapse of auditing firm Arthur Andersen, the Sarbanes–Oxley Act 18 U.S.C. § 1514A (SOX) was introduced in 2002 as part of a reformed capital market regulation (Dworkin 2007). Members of the organization in which the alleged fraud such as bank fraud, securities fraud, or any breach of regulation of the Securities and Exchange Commission (SEC) occurred, may report to a designated third party (superior, congress member, SEC). The plaintiff (whistleblower) is entitled to compensation for discharge or discrimination following the disclosure (SOX 18 U.S.C. § 1514A(a)). Scholars such as Dworkin (2007) argue that the protection is inefficient and unlikely to increase reports. Moberly (2007) shows that in the first 3 years after the introduction of SOX only 3.6% of whistleblower won relief through the initial process, concluding that SOX is misapplied to the disadvantage of whistleblower.

The Dodd–Frank Wall Street Reform and Consumer Protection Act 15 U.S.C. §78u-6 (DFA) was passed in 2010. While restricting the capital markets after the financial crisis, Section 922 provides additional regulation in relation to SOX. The DFA overlaps with the SOX in some regards of protection but introduces a reward-based incentive for whistleblower (15 U.S.C. §78u-6(a)). Tips to the SEC concerning violations of securities laws resulting in successful actions against the person or organization charged may be compensated, if the monetary sanctions recovered from the convicted party exceed one million USD. Then the whistleblower is eligible to a reward between 10 and 30% (15 U.S.C. §78u-6(b) also in general Mogielnicki 2011).

Europe, on the other hand, can be separated in two equally large camps. The first one does not have designated whistleblowing laws, while the second does. Figure 2 shows a map where countries within the European Union and adjacent ones of particular interest are classified according to their whistleblowing laws.²

Those countries that do not have designated laws usually have some degree of protection scattered across other laws, most prominently labor law. Internal (formal) channels to go to with information do exist unevenly, and often in the public sector only, probably to combat corruption.

A reason why some countries have whistleblowing laws, while others do not may be found in their attitude toward that behavior. In Germany for example, a word often used in connection with whistleblower is “Denunziant.” It has a strong negative connotation and can most accurately be translated as tattle-tat or squealer (Rauhofer 2007). The perception of whistleblower in Germany is strongly influenced by history, especially Nazi-Germany and later again in the former East, the GDR. Here, spying on neighbors, friends, colleagues, and family was encouraged and even demanded by state authorities such as the Stasi and Gestapo (Gibeaut 2006; Rauhofer 2007). This negative view on whistleblowing is still very much alive, as recent surveys suggest. Bussmann et al. (2013) find that of companies that have a whistleblowing system established, 48% believe that it fosters “squealing.” A similar history shaped the negative view in South Africa. In India, whistleblowers are under threat to be killed (Gibeaut 2006). This negative view might also be a factor in other countries that do not have whistleblowing laws in place. In countries like the USA, on the other hand, that do have whistleblowing laws, they are sometimes referred to as heroes (Grant 2002; Hartman et al. 2009).

Countries that do have designated laws are shown in more detail in Table 1. Most laws are relatively new, passed within the last decade. An exception is the UK. The Public Interest Disclosure Act (PIDA) enacted in 1998 provides employees with protection against retaliation if they make a disclosure in the public interest. However, the law is often characterized as lacking real protection from retaliation, being flawed and inefficient (Bowden 2006; Latimer and Brown 2008). Many countries abandon the “in good faith” provision in whistleblowing laws in favor of a “in public interest” definition. The test whether a whistleblower is protected under the law is therefore less strict: as long as the disclosure is made under reasonable belief of a wrongdoing that is thought to harm the public. The “in good faith” test, on the other hand, often led to courts waiving protection eligibility (Dworkin 2007). Most laws state that internal channels have to be exhausted before making external disclosures. Only a handful of countries allow the public and media as valid external recipients. Ireland is the only country that could—at least indirectly—be considered as having a reward provision. Whistleblower may receive a compensation up to 5 times their annual salary. However, that compensation is only paid in an unfair dismissal case and reduced if the intention was not made in good faith.

In April 2018, the European Commission published a draft directive for the European Parliament on the protection of whistleblower (European Commission 2018). This directive would be in line with international best practice, as it does enable whistleblower to report the misconduct externally even to media and the public. It would be a purely protection-based legislation, which prohibits any form of retaliation if the whistleblower reported the incident to the best of her knowledge and the disclosure is in the public interest.

The effects of legislation have been analyzed more prominently after the introduction of the Sarbanes–Oxley Act in 2002 and mostly with a focus on the USA. Dworkin (2007) doubts that SOX would foster whistleblowing due to its ineffective protection. Moberly (2007) shows that only 3.6% of whistleblower won relief after experiencing retaliation through SOX within the first 3 years after its introduction. This is emphasized again by Yeoh (2014) who argues that neither the US (SOX and DFA), nor the UK’s PIDA brings actual whistleblower protection. Schmidt (2005) argues that rewards and protection granted for external whistleblowing may lead to opportunistic behavior that reduces organizational efficiency. He concludes that laws such as the UK’s PIDA or the SOX (and in extension the then not yet introduced DFA) cannot claim legitimacy. Experimentally, Brennan and Kelly (2007) show that protection increased external whistleblowing intention for accounting students in Ireland, while Perreault and Wainberg (2016) find that explicit protection statements in internal channels (hotlines) increase the perceived threat to retaliation and thereby decrease whistleblowing intention.

Mogielnicki (2011) argues that the extension by the DFA would incentivize whistleblower by utilizing rewards for external whistleblowing. This is in line with the experimental findings by Andon et al. (2018) who show with a group of professional auditors that financial rewards lead to higher external (e.g., authorities) whistleblowing intention. However, this effect is mediated by the seriousness of the wrongdoing. That is, rewards are less significant when the seriousness is high enough. However, Stikeleather (2016) only finds a positive influence on internal whistleblowing. Similar results by Brink, Lowe and Victoravich (2017) show that monetary attitude leads to increased whistleblowing intention, although they find no difference between internal and external channels. Rose et al. (2018) only find an influence for very high rewards. A different perspective is given by Cordis and Lambert (2017) who find that whistleblowing laws have a deterrent effect on fraud, which they attribute to the higher perceived detection rate by companies or (potential) wrongdoers.

This analysis shows that the dichotomy of “rewards vs protection” is a recurring theme in legislation and academic research. It is evident that people seem to evaluate whistleblowing laws within these two frames. This is paramount when prospect theory is applied on this issue in the subsequent chapter. In particular, a protective approach aims to minimize or prevent retaliation. The influence of retaliation on whistleblowing has long been studied. Most research finds that whistleblowing intention is reduced by an increase in retaliation likelihood or severity (e.g., Miceli and Near 1984; Mesmer-Magnus and Viswesvaran 2005; Liyanarachchi and Newdick 2009; Mayer et al. 2013, Caillier 2017; Dhamija and Rai 2017). Interestingly, Perreault and Wainberg (2016) find that explicit protection statements in internal channels (hotlines) increase the perceived threat to retaliation and thereby decrease whistleblowing intention.³

Prospect theory was introduced by Kahneman and Tversky in 1979 as a descriptive model of decisions under risk. It gives experimental evidence that people tend to value changes in terms of deviations from a reference point (reference dependence), for example a status quo. From this point, deviations can be classified into gains and losses, where people tend to be risk averse toward gains and risk seeking toward losses, while the latter effect is more pronounced than the former and is called loss aversion (Kahneman and Tversky 1979). Contrary to what economists usually refer to as the “homo oeconomicus”, Kahneman and Tversky find that instead of being purely rational, people can more accurately be described as bounded rational. However, their actions are not completely irrational, that is, they do evaluate options within constant frames and patterns (Kahneman and Tversky 1979, Tversky and Kahneman 1992). While this original version was limited in several regards, a revised version was introduced in 1992, solving issues of stochastic dominance violations, the limitation to two nonzero outcomes that could be examined, as well as extending the theory from decisions under risk to uncertainty. In addition, decision weights can now be assessed separately for gains and losses (Tversky and Kahneman 1992).

The generally known function of expected utility is the sum of probability \(p\) of an outcome multiplied by the outcome value \(x\) for all outcomes, where the added probabilities are 1. In prospect theory this function is altered in several regards. A value \(v\) is assigned to an outcome \(x\): \(V\left( x \right)\), and a decision weight \(w\) is assigned to the probability \(p\) of an outcome: \(W\left( p \right)\). The decision weight may vary for gain and loss prospects, respectively, thus \(w^{ + , - } \left( p \right)\). A typical value function under prospect theory is shown in Fig. 3. The laws introduced in the previous chapter are added to this value function according to their influences: Protection-based legislation is added to the loss (prevention) side and rewards are added to the gain side of the value function.

The origin can be interpreted as the status quo or the neutral reference point. The function for positive (negative) outcomes is concave (convex), and the function for losses is steeper than for gains, depicting loss aversion. The value function in parametric form can be written aswhere \(\lambda\) is the parameter signifying loss aversion, if \(\lambda > 1\). This “new version” (1992) has been shown to give better results in regard to explaining human behavior not only compared to the original version (Fennema and Wakker 1997), but also in comparison with other versions of utility descriptions (Kothiyal et al. 2014).

Prospect theory’s introduction has sparked numerous academic researches that try to apply it to economic problems. Most notably in the area of management research, especially strategic management, organizational behavior, and human resource management, targeting executive compensation, negotiations, relations between organizational risks and return and firm risk-taking behavior. However, Holmes et al. (2011) find in their review of these studies that most either misconstrue prospect theory or only apply aspects of it on the issue. Outside of management, an obvious application is the behavior of people in finance, in relation to portfolio and investment decisions and insurance contexts, especially insurance schemes and preferences (for a review see Barberis 2013). More recent studies try to extend these applications to other areas. Yang et al. (2017) use prospect theory to explain traveler’s choices under travel time variability. Weingarten et al. (2018) use it to show that a positive deviation in one area (examinations) does not compensate for a failure in another area. However, research mostly uses prospect theory to describe observed behavior, instead of using it in a prescriptive way (Barberis 2013). Its application lack outside of pure economic contexts is staggering. Thus, the actual impact of prospect theory is still out for debate, even 40 years after its initial introduction.

One reason for lack of applicability may be due to the vague definition of the reference point (Köszegi and Rabin 2007; Barberis 2013). Literature refers to it mainly as the current wealth or status quo (Kahneman and Tversky 1979; Samuelson and Zeckhauser 1988; Tversky and Kahneman 1992), as expectations (Cherry et al. 2003; Köszegi and Rabin 2007) or goals (Weingarten et al. 2018). For whistleblowing in particular, it may be interpreted as expectations about what would happen under “normal” circumstances, a continuation of the current state of nature, that is expectations about the future: The person goes to work and does her job. To discover a fraud case is most likely unexpected or unanticipated. The subsequent consequences of going public with that information are a deviation from the expectation about the future. It involves a deviation from the expected or current wealth, e.g., her salary and employment position.

A second reason for the lack of applicability might be the common understanding of prospect theory as decision under risk—and not under uncertainty, which captures a wider variety of actual problems (Kothiyal et al. 2014). Originally, prospect theory was devised as a theory to describe decisions under risk. However, due to Tversky and Kahneman’s (1992) extension of prospect theory, it is able to capture situations under uncertainty, also called ambiguity in the Knightian sense (Knight 1921). Ellsberg (1961, p. 657) famously showed that people prefer risk over ambiguity, which he defines as “a quality depending on the amount, type, and ‘unanimity’ of information, and giving rise to one’s degree of ‘confidence’ in an estimate of relative likelihoods.” However, as Kothiyal et al. (2014, p. 2) lament, “there is still a widespread misunderstanding that prospect theory could only be applied to risk. The popularity of prospect theory for risk may have contributed to this misunderstanding,” which may be one of the reasons why there are so few experiments of prospect theory under ambiguity. Notable exceptions are Kothiyal et al. (2014), Abdellaoui et al. (2016), and Baltussen et al. (2016), which I will refer to in the experimental section.

Whistleblowing seems to be better described by a situation under ambiguity than under risk. Risk proposes that the probabilities within a prospect are known. However, this is highly unlikely for a whistleblowing scenario—or any real-life problem. For example, the US SOX on paper gives a 100% protection from retaliation if certain conditions are met. Moberly (2007) shows that only 3.6% of whistleblower won relief. The US SEC awarded rewards to 46 individuals between 2012 and 2017 only, even though they received over 4400 tips in 2017 alone (SEC 2018). Many of these whistleblowers probably genuinely believed they would be eligible for a reward. This contributes to the interpretation that probabilities are unknown or hard to estimate. Every fraud and whistleblowing case is unique, and comparing one’s own case to that of another whistleblower is problematic. In this paper, the situation is therefore interpreted as a decision under ambiguity. In addition, this helps to advance our understanding of the validity of prospect theory for ambiguous contexts and adds to the scarce experimental literature in this area.

A third issue is whether prospect theory is a laboratory phenomenon or an actual behavioral trait (Tversky and Kahneman 1992; Barberis 2013). Laboratory experiments often use low stakes and small amounts of money, while they are also hypothetical decision problems and mostly restricted to monetary outcomes. However, it has repeatedly been shown that this way of decision making stays constant across different problems, such as decisions involving monetary outcomes (Hershey and Schoemaker 1980) as well as human lives (Tversky and Kahneman 1981). Kachelmeier and Shehata (1992) show that there are no significant differences between hypothetical and real payoffs, or large and small amounts of money. Studies found no significant differences between a flat compensation of the participants in the experiments and a payoff-based pay (Camerer 1989, Kachelmeier and Shehata 1992). However, literature on large amounts of money remains scarce, a point where this study aims to provide further insights, using high (hypothetical) payoffs that seem more realistic in the whistleblowing context (rewards: SEC 2018; losses: GBES 2016).

Prospect theory seems very much suited to evaluate whistleblowing decision making, considering the framing of legislation in terms of losses (protection) and gains (reward). Each whistleblowing case is highly individual, and therefore, outcomes are ambiguous. The reference point as the expectation of the future or status quo and the detection of fraud as an unexpected deviation seem reasonable.

Given the implications of prospect theory and the laws framed as protection- and reward-based incentive structures, I assume that people exhibit—in general—in a whistleblowing scenario a similarly s-shaped value function, that is convex (concave) for losses (gains). They evaluate outcomes in terms of gains and losses, with losses weighing stronger than gains of the same magnitude (loss aversion).

This general behavior should be irrelevant of the person’s intention to whistleblow, since loss aversion is assumed to be a “general” behavioral trait in humans (Tversky and Kahneman 1992).

However, the application of prospect theory also implies that if losses weigh heavier than gains, then a protection from a loss, e.g., retaliation or sanction, should influence a whistleblower stronger than a reward of the same monetary value.

Not only loss aversion in prospect theory suggests that losses should have a stronger impact on people than gains, but also results of whistleblowing research discussed earlier. In particular, the many studies that suggest retaliation to have a strong influence on the whistleblowing intention and the findings by Andon et al. (2018) and Rose et al. (2018) in relation to the rather mixed influence of rewards substantiate the idea that losses weigh heavier than gains. Thus, a loss is a stronger dis-motivation to whistleblow, than a gain increases intention to whistleblow.

The value function in prospect theory is s-shaped: For positive (negative) outcomes, it is concave (convex). To measure the curvature of the function, \(v\left( x \right)\) can be normalized to 1 by scaling \(V\left( {x_{6}^{ + } } \right) = 1\) (\(V\left( {x_{6}^{ - } } \right) = - 1\)) and for any other value V \(\left( {x_{j}^{ + } } \right) = j/6\) (V \(\left( {x_{j}^{ - } } \right) = - j/6\)) for all \(j \in \left\{ {1, \ldots 5} \right\}\) (Abdellaoui et al. 2016).⁶ As an example, the mean function over all subjects normalized is shown in Fig. 4 and without normalization in Fig. 5. When Fig. 5 is compared to Fig. 3, one can already see that the shape is very similar to that of the typical value function proposed in prospect theory.

The area under the normalized curve for gains (losses) can more accurately be measured and interpreted for each subject individually as \(\mathop \smallint \nolimits_{0}^{1} f\left( {x^{ + } } \right){\text{d}}x \left( { - \mathop \smallint \nolimits_{ - 1}^{0} f\left( {x^{ - } } \right){\text{d}}x} \right)\), where an area < .5 indicates a convex (concave), = .5 linear (linear), and > .5 concave (convex) shape. The results are shown in Table 3. Almost half of the subjects (46.2%) can be classified according to prospect theory’s s-shaped value function, while only 15.4% acted according to the traditional idea of constant risk aversion. Almost 70% changed their risk preferences at the reference point, further validating prospect theory’s idea of reference dependence. Results are similar to previous studies (Tversky and Kahneman 1992; Abdellaoui et al. 2016).

This can be seen as a confirmation of hypothesis 1.1 that participants would exhibit behavioral traits consistent with prospect theory in terms of the shape of the value function: It is s-shaped for a majority of subjects (marked in bold) and almost half of the students show concave (convex) value functions in the area of gains (losses).

Several ways to measure loss aversion have emerged over the years. The original method proposed by Kahneman and Tversky (1979) (K/T) assumes loss aversion when \(V\left( x \right) < - V\left( { - x} \right)\) for all \(x > 0\). Since \(V\left( {x_{j}^{ + } } \right) = - V\left( { - x_{j}^{ - } } \right)\), \(x_{j}^{ + } > - x_{j}^{ - }\) implies that \(V\left( {x_{j}^{ + } } \right) < - V\left( { - x_{j}^{ + } } \right)\) holds. This can be measured using a Wilcoxon test for all \(x_{j}\) (Table 4).

The difference is significant for all \(x_{j}^{ + } > \left| {x_{j}^{ - } } \right|\), and the effect is very pronounced (\(r = \left| {Z/\sqrt n } \right|\) > .5 for all Z, is considered a strong effect, while > .3 is considered a medium effect and > .1 is a small effect according to Cohen 1988). Loss aversion can be assumed to be present based on the K/T measure (Kahneman and Tversky 1979).

Another measurement was introduced by Tversky and Kahneman (1992) (T/K) by calculating the loss aversion parameter \(\lambda\):

Given that \(\alpha\) and \(\beta\) are found to be equal across many studies (see Bromiley 2010 for a review), then positive outcomes (ordinate) divided over the absolute values of the negative outcomes (abscissa) for the respective \(x_{j}\) make \(\lambda\) observable with \(x_{j}^{ + } = \lambda ( - x_{j}^{ - } )\). This also holds because \(V\left( x \right)\) is a ratio scale with known values (Abdellaoui et al. 2016). A function can be elicited by linear interpolation with the origin as reference and starting point of the function, shown in Fig. 6, where \(\lambda = 2.5336\). This is in line with prior studies who found \(\lambda\) to be between 2.3 (Abdellaoui et al. 2016) and 2.83 (Baltussen et al. 2016) for ambiguity. For risk, Tversky and Kahneman (1992) elicited a value of 2.25. Other studies found values for risk of 2.47 (Abdellaoui et al. 2011), 2.21 (Abdellaoui et al. 2016), and 1.86 (Baltussen et al. 2016). This deviation between the values for ambiguity and risk may be attributed to the effect found by Ellsberg (1961) that people in general exhibit higher aversion toward uncertainty over probability. Thus, pronounced loss aversion can be assumed to be present based on the T/K measure.

A third method proposed by Köbberling and Wakker (2005) (K/W) measures the kink at the reference point, thus the left derivative over the right derivative at the reference point as \(V_{ \uparrow }^{'} \left( 0 \right)/V_{ \downarrow }^{'} \left( 0 \right)\). Abdellaoui et al. (2016) simplify this assumption by arguing that the fist gain and loss payoffs elicited may be used. Then \(\lambda = V_{ \uparrow }^{'} \left( 0 \right)/V_{ \downarrow }^{'} \left( 0 \right) = V\left( {x_{1}^{ - } } \right)/x_{1}^{ - } /V\left( {x_{1}^{ + } } \right)/x_{1}^{ + }\) and since \(V\left( {x_{1}^{ - } } \right) = - V\left( {x_{1}^{ + } } \right)\), one can simply calculate \(x_{1}^{ + } / - x_{1}^{ - }\). Doing this individually for all 39 subjects, the median (mean) value is 4.7619 (9.5858). Using the median (mean) values for \(x_{1}\) over all subjects for gains and losses, respectively, gives a loss aversion of 7.1429 (3.6527). These values are arguably higher than the one produced by the T/K method. A simple explanation is that the K/W method only takes into account the initial elicited values of the function in either direction, whereas the T/K method measures the entire function. The latter is thus less influenced by outliers and seems more robust. Regardless of method used, participants in general exhibit strong loss aversion. A loss weighs more than twice as heavy as a gain of the same absolute monetary value. This is a confirmation of the hypothesis 1.2 that people exhibit strong loss aversion. Hypotheses 1.1 and 1.2 show that the reference point is important for people to value payoffs in terms of gains and losses in a law context. The results suggest that this holds for high payoffs and ambiguity.

A subsample of the first group (n = 24) was invited to participate in a second part of the experiment. Here, the intention to whistleblow was measured, using five questions (\(int_{j}\)) that asked for the intention to actually come forward with the information in different ways. This is a normal procedure often used in whistleblowing research (Park et al. 2008; Park and Blenkinsopp 2009; Brown, Hays and Stuebs 2016). A 7-point Likert scale (\(1, \ldots ,7\)) was used, where 1 equals “I do not agree at all” and 7 “I fully agree.” The translated questions and their descriptive statistics are given in Table 5.

The questions showed very high internal consistency (Cronbach’s alpha = .911), indicating that the questions actually measured the same underlying aspect. The same case as in the first part of the experiment was given. Each participant was subsequently confronted with her alternatives from the first test, and her (individually elicited) payoffs. For each alternative the participant had to score on a 7-point Likert scale her intention to actually whistleblow, given each prospect. As an example, recall part 1.1 of Table 2. Two prospects were given: \(x_{0}\) and \(G;L_{1}\). The elicited value for \(L_{1}\) from the first experiment was inserted and the participant had to state her intention to whistleblow given the prospect \(x_{0}\) and her intention for the situation where either \(G\) or \(L_{1}\) realizes.

The correlation for the overall intention to whistleblow with loss aversion was identified, using the individual loss aversion parameters W/K and T/K elicited in the first experiment for each subject and their overall whistleblowing intention (\({\text{int}}_{\text{sum}}\)) measured as the sum of each participant’s \({\text{int}}_{j}\) answer. As shown in Table 6 there is no significant correlation between the intention and loss aversion for either measures (\(p > .05\)), indicating that these are separate concepts and not influenced by each other.

This is in line with hypothesis 2 that assumed whistleblowing intention and behavior according to prospect theory, namely loss aversion, to be separate concepts. However, intention should nonetheless change according to the “laws” of prospect theory. That is, a loss should decrease intention to whistleblow stronger than a gain would increase it, as proposed by hypotheses H 3.1 and 3.2.

Next, the change of intention to whistleblow is measured when moving along the value function. Since the value difference (recall that \(V\left( x \right)\) is a ratio scale and that the intention to whistleblow was elicited for each prospect) between subsequent outcomes is equal, it is possible to compare the intention at given points along the value function. This also holds for changes since equal changes in magnitude along the value function can be measured. For reasons of simplicity the value for the observed loss (gain) is set to − 1 (1). The value at the reference point is 0 by definition (Tversky and Kahneman 1992). Thus, a change from the reference point in either direction is in magnitude equal to 1 as well. Intention at the reference point is given as \({\text{int}}_{\text{ref}}\), for the sure loss (gain) as \({\text{int}}_{\text{neg}}\) (\({\text{int}}_{\text{pos}}\)), and for the uncertain loss (gain) as \({\text{int}}_{{{\text{neg}}\_{\text{unc}}}}\) (\({\text{int}}_{{{\text{pos}}\_{\text{unc}}}}\)). The intention change was calculated when moving from the reference point (prospect: sure \(x_{0}\); part 1.1) to a sure loss (prospect: sure loss \(x_{1}^{ - }\); part 1.2) as \({\text{int}}_{{{\text{neg}}\_{\text{change}}}}\), as well as from the reference point to a sure gain (prospect: sure \(x_{1}^{ + }\); part 1.3) as \({\text{int}}_{{{\text{pos}}\_{\text{change}}}}\). Similarly for a change from the reference point to an uncertain loss (gain), prospect: 1.3 (prospect: 1.2), as \({\text{int}}_{{{\text{neg}}\_{\text{unc}}\_{\text{change}}}}\) (\({\text{int}}_{{{\text{pos}}\_{\text{unc}}\_{\text{change}}}}\)). These positions and changes in intention to whistleblow were then evaluated using a Wilcoxon test. Table 7 shows the comparisons of positions in intention.

Intention to whistleblow changes significantly with a change of the outcome from the reference point toward a sure loss (gain). However, the effect is stronger for a change toward a loss (\(r = .81\)) than a gain (\(r = .52\)). This effect realizes although the monetary value for a gain is already on average more than twice as high as the absolute value of the loss. The change toward an uncertain outcome from the reference point is only significant for losses, not for gains. The change from a certain to an uncertain outcome (recall that the value of the outcome remains at 1) is only significant for losses, not for gains. This can be interpreted as that an uncertain loss increases whistleblowing intention compared to a certain loss, whereas whistleblowing intention is not significantly increased by an uncertain gain toward a certain one, keeping the value of the outcome constant.

Table 8 shows the comparison of changes in intention. There are significant differences between an equal (dis)value change toward the domain of gains and losses from the reference point. An equal value change from the reference point to a loss is felt significantly stronger than the one from the reference point to a gain (\({\text{int}}_{{{\text{neg}}\_{\text{change}}}}\) vs \({\text{int}}_{{{\text{pos}}\_{\text{change}}}}\)). The effect is strong (\(r = .51\)) and realizes although the absolute magnitude of the change is 1, and the value function already controls for the person’s loss aversion. The impact on intention of a change from the reference point toward a sure loss is found to be significantly stronger than toward an uncertain loss, but this effect is not significant for gains. Instead of direct intention changes, the percentage change was calculated as a control. For example, the percentage change from 5 to 7 is less than one from 2 to 4, where the value doubles. The untabulated analysis gives the same results. The asymptotic significance for the difference between an uncertain and certain gain are even less pronounced (\(p = .115\); \(Z = - 1.577\) based on positive ranks).

Hypothesis 3.1 stated that a loss would decrease whistleblowing intention, while a gain would increase whistleblowing intention. This can partially be shown with the analysis above. Deviations toward losses (gains) decrease (increase) whistleblowing intention, but there is no difference in the intention between a sure gain and an uncertain gain, whereas there is a significant difference for the loss side.

Hypothesis 3.2 stated that a loss has a stronger impact on intention than a gain of the same magnitude. Intention to whistleblow is decreased stronger by a loss than it is increased by a gain. Figure 7 illustrates this again graphically for the mean intention positions, where the intentions for the (uncertain) loss and gain are depicted. It can be seen that the impact on the loss side for a decrease in intention is much stronger than the increase in intention for a gain. This is despite the fact that for the same absolute value of the outcome (\(v\left( x \right) = 1\)), the monetary gain on average is already more than twice as high as the absolute loss.