1 Introduction

Work environment substantially affects employee satisfaction and behavior. Numerous examples demonstrate that low worker satisfaction decreases productivity. Consider labor disputes: Krueger and Mas (Reference Krueger and Mas2004) showed that the 1994–1996 conflict between a labor union and the management of Bridgestone/Firestone was a major contributing factor to the increase in the number of defective tires produced by the company in that period. Similarly, labor conflicts at the construction equipment manufacturer Caterpillar resulted in a significant decrease in output quality (Mas Reference Mas2008), and compensation disputes in police departments in New Jersey coincided with a decline in the number of arrests by police officers and an increase in crime rates (Mas Reference Mas2006).

Changes in employee behavior as a reaction to the workplace environment might stem from different sources. Workers can alter not only the quality and quantity of effort provision, but also their engagement in unethical behavior. Unethical behavior is widespread in the corporate world and takes different forms, negatively affecting companies’ productivity (see Treviño et al. Reference Treviño, den Nieuwenboer and Kish-Gephart2014, for a review). Reciprocal motives are an important driver of employees’ decision to behave unethically.Footnote ¹ Unethical behavior, or the lack thereof, might serve as a reciprocity device any time a direct reaction (e.g., effort reduction) is not feasible, for example, due to lack of power or fear of retaliation. In many situations, workers might reciprocate negative encounters with deceptive behaviors that hurt the company, such as misreporting the quality of their work or faking sick days. Similarly, after positive experiences within the company, employees might abstain from carrying out unethical acts they would have otherwise engaged in; that is, they might reciprocate with honesty that benefits the company at a cost to themselves. Because deception, by its nature, is difficult to observe, empirical work offers limited guidance in investigating whether and to what extent (a lack of) deception is used as a (reward) punishment mechanism. The vast experimental work in economics on gift exchange has, thus far, mainly explored reciprocity within the domain of effort provision, looking at behaviors that do not involve deception (see e.g., the survey on reciprocal behavior by Fehr and Gächter Reference Fehr and Gächter2000).

In this paper, we use a simple two-player, two-stage game to experimentally study whether deception serves as a reciprocity device. In the first stage, subjects play a variation of the dictator game. In the second stage, they play a deception game (Erat and Gneezy Reference Erat and Gneezy2012), in which the recipient in the dictator game has the opportunity to lie to her counterpart by sending her an untruthful message. Compared to truth-telling, lying reduces the payoff of one’s counterpart if the counterpart follows the message. We use the strategy method (Selten Reference Selten and Sauermann1967) and have the sender choose a message for each possible payoff received in the first stage.

Predictions concerning whether and to what extent deception could be used as a reciprocity device are conflicting. Reciprocity models (e.g., Dufwenberg and Kirchsteiger Reference Dufwenberg and Kirchsteiger2004; Rabin Reference Rabin1993) predict that previous encounters will affect subsequent behavior (see also Charness Reference Charness2004; Charness and Haruvy Reference Charness and Haruvy2002; Charness and Levine Reference Charness and Levine2007; Offerman Reference Offerman2002). However, it is unclear whether previous encounters would affect subsequent behavior when reciprocity occurs via lying. A fast-growing experimental literature has documented that deception entails positive moral costs (Bucciol and Piovesan Reference Bucciol and Piovesan2011; Fischbacher and Föllmi-Heusi Reference Fischbacher and Föllmi-Heusi2013; Gneezy Reference Gneezy2005; Mazar et al. Reference Mazar, Amir and Ariely2008).Footnote ² This literature predicts that, because lying is morally costly, less reciprocal behavior could occur when reciprocating requires individuals to lie than when it does not, regardless of the nature of a previous encounter. To test whether this prediction is true, we compare the results of our baseline game with the results of a game in which individuals can reciprocate without engaging in deception. That is, we modify the second stage of the baseline game such that an identical payoff allocation can be obtained through a direct choice rather than by telling a lie. If there were indeed a moral cost of lying, we would expect lying rates to be lower than direct choice rates.

This design allows us to test Hurkens and Kartik’s (Reference Hurkens and Kartik2009) conjecture that individuals have either a zero or infinite moral cost of lying—the former types always lie if they prefer the outcome from lying and the latter types never lie. In such a case, the conditional probability of lying (the ratio of lying to direct choice) would be constant regardless of the type of previous encounter.Footnote ³ On the other hand, moral costs might be malleable. The morality literature has suggested that moral costs decrease when one can blame the other party (Bandura et al. Reference Bandura, Barbaranelli, Caprara and Pastorelli1996; Bandura Reference Bandura1999, Reference Bandura2002, Reference Bandura, Moghaddam and Marsella2004; Cubitt et al. Reference Cubitt, Drouvelis, Gächter and Kabalin2011; Ditto et al. Reference Ditto, Pizarro and Tannenbaum2009). In that case, we would expect that the more unfair the previous interaction, the lower the moral cost of lying. In other words, deceiving someone who was previously unkind would be easier than deceiving someone who was previously kind. Additionally, kind encounters could be rewarded with honesty from people who would otherwise lie. Thus, the conditional probability of lying would decrease with increasing kindness.

Further, moral costs of lying might be sensitive to whether individuals have monetary incentives to lie. There are conflicting predictions. On the one hand, the moral cost of lying may increase with one’s payoff from lying. When reciprocating unkind behavior requires individuals to lie and lying leads to a financial benefit, people might worry about the signal they convey to themselves or to others about their own morality, as it can be unclear whether individuals lie to gain a personal material benefit or to punish preceding unkind behavior. This concern may be present when lying is beneficial, but it cannot be present when lying is costly in terms of payoffs. Indeed, work on perceptions of generosity in psychology has found that people morally discredit prosocial actions if those actions result in monetary benefits for the self but not if they result in a monetary loss or no benefit for the self (e.g., Carlson and Zaki Reference Carlson and Zaki2018; Lin-Healy and Small Reference Lin-Healy and Small2012; Newman and Cain Reference Newman and Cain2014). If the moral cost of lying increases with one’s own payoff from lying—when lying is used as a tool to reciprocate unkind behavior—the conditional probability of lying would be smaller if lying were beneficial to the self than if it were costly. On the other hand, work on excuses to behave selfishly (e.g., Babcock et al. Reference Babcock, Loewenstein, Issacharoff and Camerer1995; Dana et al. Reference Dana, Weber and Kuang2007; Exley Reference Exley2015; Gneezy et al. Reference Gneezy, Saccardo, Serra-Garcia and van Veldhuizen2017; Konow Reference Konow2000) suggests that when lying is beneficial, self-serving biases might decrease one’s moral cost of lying. When lying is costly for the self, however, a self-serving bias cannot affect the size of moral costs. Hence, this mechanism would predict that the conditional probability of lying is higher when lying is beneficial to the self than when it is costly.

To investigate whether lying costs are malleable in the presence of reciprocal motives as well as different incentives for lying, we study two domains of lies: In the Selfish domain, lying increases one’s own payoff at a cost to the receiver, and in the Spiteful domain, lying is costly for the self but imposes a larger cost on the receiver (see the taxonomy of lies by Erat and Gneezy Reference Erat and Gneezy2012). In organizations, selfish lies would be akin to over-reporting the quality of one’s work or taking credit for someone else’s work. An illustration of spiteful lies is misrepresenting one’s private information in order to sabotage a project that has a small value to oneself and a high value to one’s counterpart.

We also measure the extent to which lying rates change in response to increasingly kinder previous interactions, and whether the effect is driven by intentions rather than initial payoff allocations. The issue is quite elusive, because even in settings where reciprocity doesn’t involve deception, evidence is mixed. Some studies (Blount Reference Blount1995; Falk et al. Reference Falk, Fehr and Fischbacher2008) point to intentions, while others (Bolton et al. Reference Bolton, Brandts and Ockenfels1998; Charness Reference Charness2004) highlight distributional concerns as the main driver of reciprocity. In our baseline game, a response to both intentions and initial payoff allocations—including inequity aversion and an income effect—would predict lying and direct-choice rates of the same option to decrease with increasing payoffs from the first stage. To identify whether intentions drive the reciprocal response, we compare our baseline treatment with a treatment in which the experimenter rather than one’s counterpart implements the first-stage outcome.

We find that individuals use lies in order to reciprocate (un)kind behavior in both the Selfish and Spiteful domains: lying rates are highest after an unkind encounter and decrease with increasing kindness of the encounter. For selfish lies, the proportion of individuals who lie is always lower than the proportion of individuals who directly choose the selfish payoff allocation, and the conditional probability of lying decreases as a function of the counterpart’s kindness. This finding is in line with the predictions of a positive moral cost of lying that increases with increasing kindness, refuting the hypothesis of Hurkens and Kartik (Reference Hurkens and Kartik2009). Surprisingly, for spiteful lies, lying rates are similar to the percentage of direct choices. This result provides evidence against a moral cost of lying when lying is costly for the self and is used as a punishment device, demonstrating the malleability of the moral cost of lying.

For selfish lies, both intentions and initial payoff allocations affect lying rates, but the latter effect is larger. We find evidence for positive reciprocity, whereas the evidence for negative reciprocity is only directional and not statistically significant. For spiteful lies, only intentions matter. When the experimenter implements the first-stage payoffs, lying rates are close to zero in all cases; however, individuals punish unkind intentions by lying. As expected, lying rates are close to zero after kind encounters.

Our paper adds to the growing literature on deception by investigating the drivers of deception within an interacting pair after an initial encounter.Footnote ⁴ Previous work by Ellingsen et al. (Reference Ellingsen, Johannesson, Lilja and Zetterqvist2009) showed that honest revelation of private information in bilateral bargaining is affected by whether individuals previously cooperated or defected in a prisoner’s dilemma game. Several features distinguish our experiment from theirs. First, we systematically vary the levels of kindness of the initial encounter to show whether the moral cost of lying affects retaliation via deception. We provide the first evidence of reciprocal deception that can be attributed to intentions. More importantly, we compare whether reciprocating through lying is conceptually different from reciprocity that doesn’t involve deception, and whether the presence (absence) of financial costs from deception affects the retaliation behavior of the deceiver. Finally, we contribute to the recent stream of papers that investigate the structure of lying costs (e.g., Abeler et al. forthcoming; Dufwenberg and Dufwenberg Reference Dufwenberg and Dufwenberg2018; Gneezy et al. Reference Gneezy, Kajackaite and Sobel2018; Khalmetski and Sliwka Reference Khalmetski and Sliwka2017). Our results show that there is an interaction between reciprocity concerns, lying costs, and incentives from lying, and highlight the need to combine insights from both the reciprocity and the lying literature in order to fully understand the drivers of deception.

2 Experimental design and procedures

2.1 Experimental design

In the experiment, two subjects (Player A and Player B) are matched for two consecutive stages. This information is common knowledge. In the first stage, Player A (dictator) is given an endowment of 10 tokens and has to decide how much to send to Player B (receiver). Player B does not make any decision in the first stage. Player A can choose between sending 0, 2, 4, 6, 8, or 10 tokens to Player B. Player A keeps the amount not sent to Player B. The amount sent is tripled. We chose to triple the amount sent because doing so makes being kind to Player B easier for Player A, compared to a conventional dictator game. In this setup, we consider an interaction as being increasingly kind the higher the amounts sent in the first stage. Therefore, sending nothing is clearly unkind, and the kindness of the interaction increases with the amount sent.

In the second stage of the experiment, participants play the deception game (Gneezy Reference Gneezy2005; Erat and Gneezy Reference Erat and Gneezy2012). The experimenter rolls a six-sided die in front of each Player B. Afterwards, Player B is asked to send one of six possible messages to Player A. The six messages are “The outcome of the roll of die was x,” where x can be any number from the set {1, 2, 3, 4, 5, 6}. After receiving Player B’s message, Player A declares what she believes to be the outcome of the die roll. Player A’s choice and the actual die outcome determine the payoffs for both players. There are two payment options. Option 1 yields equal payments to both participants and is the same across domains, whereas Option 2 varies between the Selfish and Spiteful domains as described below. Both players are informed that if Player A’s choice coincides with the real outcome, Option 1 is implemented. Otherwise, Option 2 is implemented. Player A is not informed of the possible payoffs associated with the two options; she learns her own earnings from the task only at the end of the experiment. Player B knows the payoffs associated with Options 1 and 2 for both players and also knows that Player A does not and will not know the payoffs, aside from her own final earnings.

If Player A chooses the number that is the actual outcome of the die roll, Option 1 is implemented and both players earn 10 tokens. Otherwise, Option 2 is implemented. We vary the payoff from Option 2 for Player B while keeping it constant for Player A as follows. In the Selfish domain, deception is beneficial for Player B; Player B earns 12 tokens and Player A earns 4. In the Spiteful domain, deception is costly for Player B; she earns 9 tokens and Player A earns 4.Footnote ⁵ Thus, in both cases, Option 2 is worse for Player A. Table 1 summarizes the payoffs in the two domains.

Table 1

Payoffs in the Selfish and Spiteful domains

	Selfish	Spiteful
Option 1	10, 10	10, 10
Option 2	4, 12	4, 9

The richness of the message space in the deception game eliminates strategic truth-telling (Sutter Reference Sutter2009), because with such a rich space it is unlikely that participants will choose to tell the truth in order to obtain the unequitable payoff (see also the discussion in Erat and Gneezy Reference Erat and Gneezy2012). Further, we assume Player B generally expects her recommendations to be followed. This assumption rests on empirical evidence showing that telling the truth is unlikely to result in Option 2, even if senders expect receivers not to follow the message (van de Ven and Villeval Reference Van de Ven and Villeval2015). As we will show in detail in Sect. 4.3, in our setup there is also a very small probability that telling the truth implements a lie. Therefore, we assume Player B would send an untruthful message in order to either increase her own payoff (Selfish domain) or to decrease Player A’s payoff (both Selfish and Spiteful domains). In this paper, we classify all messages that contain a number different from the real outcome of the die roll as a lie.

Using a 2 × 3 design, we conducted six treatments, which are presented in Table 2. In all of our treatments, we used the strategy method (Selten Reference Selten and Sauermann1967) for the second-stage decisions, and therefore Player B had to choose her action conditional on all possible first-stage outcomes. In a separate pilot study, we conducted the Selfish-Intentions treatment both using a between-subjects design and the strategy method. The study showed similar results, suggesting that in our setting using the strategy method leads to the same qualitative results obtained by the standard direct-response method. A detailed discussion of these results can be found in “Appendix C” in ESM.

Table 2

Overview of the treatments in the Selfish and Spiteful domains

		Intentions	Direct Choice	No-Intentions
Selfish domain (Option 2: 4, 12)	Stage 1	Dictator game Player A chooses how much to give to Player B	Dictator game Player A chooses how much to give to Player B	Experimenter The experimenter informs Player B of her and Player A’s initial allocation
	Stage 2	Deception game Player B is the sender	Dictator game Player B is the dictator	Deception game Player B is the sender
Spiteful domain (Option 2: 4, 9)	Stage 1	Dictator game Player A chooses how much to give to Player B	Dictator game Player A chooses how much to give to Player B	Experimenter The experimenter informs Player B of her and Player A’s initial allocation
	Stage 2	Deception game Player B is the sender	Dictator game Player B is the dictator	Deception game Player B is the sender

In the Intentions treatments, Player B is asked (in Stage 2) to decide which message to send to Player A, for each feasible action of Player A in Stage 1. The Intentions treatments allow us to detect whether a relationship exists between the amount sent by Player A in Stage 1 and Player B’s lying rates in Stage 2. In the Direct-Choice treatments, Player B’s second-stage decision is a dictator decision, and therefore Player B directly decides whether Option 1 or Option 2 will be implemented without having to lie. The comparison between the lying rates in the Intentions treatments and the direct implementation rates in the Direct-Choice treatments informs us on whether reciprocity via deception is observed less often than reciprocity without deception. In the No-Intentions treatments, the experimenters determine the amount sent in the first stage; this information is common knowledge. By comparing the lying rates in the Intentions treatments with the lying rates in the No-Intentions treatments, we can isolate the effect of intentions from distributional concerns.

In the No-Intentions treatments, we build on previous literature (e.g., Blount Reference Blount1995; Bolton et al. Reference Bolton, Brandts and Ockenfels1998; Charness Reference Charness2004; Falk et al. Reference Falk, Fehr and Fischbacher2008) and determine the payoff outcome based on the distribution of the first-stage outcomes of previous sessions of the other treatments.Footnote ⁶ In this way, we ensure Player B has the same beliefs about Player A’s choice in both the No-Intentions and Intentions treatments. This distribution is made known to Player B without giving her information on how it is derived. Note that the first-stage instructions are the same in the Intentions and Direct-Choice treatments. Therefore, we pooled Stage 1 choices from these two treatments to determine the distribution of Stage 1 payoffs in the No-Intentions treatments. Player B is made aware of this distribution of the first-stage outcomes before she has to make a decision. That is, after Player B has decided on her strategy, the experimenter asks her to draw a chip out of a bag containing 100 numbered chips. The selected chip determines the starting earnings that will be implemented. “Appendix B” in ESM shows the distribution of the possible earnings. This way, it is completely transparent to each Player B that the initial payoff allocation between her and Player A is not determined by Player’s A intentions. After Player B draws the first-stage payoffs, Player A is informed about her own and her matched partner’s payoffs. Player A then decides whether to follow Player B’s message. This procedure ensures the information each player has in each decision-making stage in the No-Intentions treatments is comparable to the information received in the Intentions treatments.

3 Theoretical predictions

In this section, we discuss the theoretical predictions across treatments for the behavior of Player B when taking into account purely selfish preferences, outcome-based fairness models, intention-based fairness models, and lying costs. The predictions are summarized in Table 3 for both the Selfish and Spiteful domains. In the table, b denotes the first-stage payoff of Player B, and Option2 (O2 hereafter) denotes the percentage of Option 2 choices by Player B in the second stage, regardless of whether the choice is implemented by direct choice or lying. Further, for ease of exposition, Table 3 only depicts predictions within each domain.

Table 3

Predictions for the rate of choosing Option 2 or lying in the second stage

		Intentions	No-Intentions	Direct-Choice
Models with social preferences
Purely selfish	Selfish	O2 = 1, ∀b	Same as in Intentions	Same as in Intentions
	Spiteful	O2 = 0, ∀b	Same as in Intentions	Same as in Intentions
Outcome-based fairness	Selfish	O2 decreases with b	Same as in Intentions	Same as in Intentions
	Spiteful	O2 decreases with b	Same as in Intentions	Same as in Intentions
Income effect	Selfish	O2 decreases with b	Same as in Intentions	Same as in Intentions
	Spiteful	O2 = 0, ∀b	Same as in Intentions	Same as in Intentions
Intention-based fairness	Selfish	O2 decreases with b	O2 constant, ∀b O2 lower than in Intentions for low b, and higher for high b	Same as in Intentions
	Spiteful	O2 decreases with b, O2 = 0 for high b	O2 = 0, ∀b	Same as in Intentions
Models with moral cost of lying
Fixed moral cost of lying and Hurkens and Kartik (Reference Hurkens and Kartik2009)	Selfish	O2 constant, ∀b	Same as in Intentions	O2 higher than in Intentions, ∀b $\frac{\text{O}{2}^{\text{Intentions}}}{\text{O}{2}^{\text{Direct-Choice}}}$ constant, ∀b
	Spiteful	O2 = 0, ∀b	Same as in Intentions	Same as in Intentions
Moral costs decrease with unfairness of the other	Selfish	O2 decreases with b	O2 constant, ∀b O2 lower than in Intentions for low b, and higher for high b	O2 higher than in Intentions, ∀b $\frac{\text{O}{2}^{\text{Intentions}}}{\text{O}{2}^{\text{Direct-Choice}}}$ decreases with b
	Spiteful	O2 decreases with b, O2 = 0 for high b	O2 = 0, ∀b	O2 = 0 for high b O2 higher than in Intentions and $\frac{\text{O}{2}^{\text{Intentions}}}{\text{O}{2}^{\text{Direct-Choice}}}$ decreases with b for all other b

Standard models with selfish payoff-maximizing agents predict Player B always chooses the option with the higher payoff for herself, that is, Option 2 in the Selfish domain and Option 1 in the Spiteful one. This choice does not depend on whether a particular option is implemented through a lie or through direct choice.

Purely outcome-based theories of social preferences (e.g., Bolton and Ockenfels Reference Bolton and Ockenfels2000; Fehr and Schmidt Reference Fehr and Schmidt1999) argue that people derive disutility from inequity in monetary payoffs. Such models would predict the percentage of Players B who choose the equitable option, namely Option 1, to be an increasing function of their first-stage payoff, because the higher Player B’s first-stage payoff, the lower Player A’s payoff. Player B’s first-stage payoff is higher than Player A’s if Player A sends four tokens or more, which would result in Player B choosing the equitable option. Note that Player B dislikes advantageous inequality, i.e., earning more than the other. Given Player B’s first-stage payoffs ranging from 0, 6,… 30, the switching point from Option 2 to Option 1 depends on her disutility parameters from inequity. In general, the lower the disutility from being ahead or from having a higher than equitable share, the higher the switching point.Footnote ⁷ Because outcome-based social preference models do not take intentions or moral costs into account, they predict no treatment differences within a given domain. The result of the comparison of the outcome-based social preferences regarding the Selfish and Spiteful domains depends on the distribution of estimated disutility parameters.Footnote ⁸

An income effect predicts a treatment pattern similar to the purely outcome-based social preferences in the Selfish domain: If Player B is more likely to choose the generous option (i.e., the equitable option) when she has more income, then in all the treatments in the Selfish domain, the percentage of Players B who choose the equitable option, namely Option 1, would be an increasing function of their first-stage payoff. An income effect predicts no treatment differences across the three treatments. In the Spiteful domain, an income effect predicts that subjects always choose the equitable option.

Purely intention-based models of reciprocity (e.g., Dufwenberg and Kirchsteiger Reference Dufwenberg and Kirchsteiger2004; Rabin Reference Rabin1993) posit that people respond to intentions by rewarding kind actions and punishing unkind ones. Thus, similar to the outcome-based social preference models, intention-based models would predict that the higher the first-stage payoff, the higher the likelihood that Player B chooses Option 1. However, the rationale is different: The higher first-stage payoff means Player A is kinder, and therefore Player B would reciprocate with a kinder choice. In the Selfish domain, the percentage of liars in the No-Intentions treatment would be the same for all first-stage payoffs. If the previous encounter is unkind, more lying would occur in the Intentions than in the No-Intentions treatment, and vice versa for kind encounters. Finally, no difference would exist between the Direct-Choice and Intentions treatments. In the Spiteful No-Intentions treatment, Player B should always choose according to her self-interest, namely Option 2. If the previous encounter is unkind, punishment would take place in both the Intentions and Direct-Choice treatments and at the same rate. Note that a model that is a mixture of both outcome- and intention-based preferences, such as the one proposed by Falk and Fischbacher (Reference Falk and Fischbacher2006), only changes the predictions for the No-Intentions treatments: The prediction would be to observe a decrease in the percentage of Option 2 choices with increasing first-stage payoffs.

Finally, intention-based models of reciprocity would predict that, given an unkind first-stage choice of Player A, the percentage of Players B who choose the inequitable option is smaller in the Spiteful domain than in the Selfish domain because punishing an unkind offer in the Spiteful domain is costlier. Given a kind first-stage choice of Player A, we would expect some players to opt for the selfish option in the Selfish domain but no one to choose the inequitable option in the Spiteful domain. Therefore, given the same first-stage interaction, the percentage of players who choose the inequitable option in the Intentions and Direct-Choice treatments would be higher in the Selfish domain than in the Spiteful one.

With a fixed moral cost of lying, Player B’s decision to lie depends on the lie’s benefits and costs. That is, people are selfish but also incur a cost from lying, and therefore the rate of lying does not depend on the first-stage outcome. In the Selfish domain, the percentage of lies in the Intentions treatment would be lower than the percentage of direct choices in the Direct-Choice treatment. Because intentions do not matter, lying rates would be the same in the Intentions and No-Intentions treatments. In the Spiteful domain, no lying would occur.

Three additional mechanisms also assume a moral cost of lying. First, Hurkens and Kartik (Reference Hurkens and Kartik2009) conjectured that individuals have either a zero or infinite moral cost of lying. Absent lying costs, people always lie if they prefer the outcome from lying to the outcome from not lying. People with an infinite cost of lying never lie. Then the conditional probability of lying (i.e., the ratio of lying to direct choice) would be constant regardless of the type of previous encounter, because a constant proportion of people who directly choose an outcome have an infinite moral cost of lying. This argument assumes no relationship between preferences toward fairness and one’s moral cost type. A second mechanism is based on the premise that moral costs are malleable and decrease when one can blame the other party (Bandura et al. Reference Bandura, Barbaranelli, Caprara and Pastorelli1996; Bandura Reference Bandura1999, Reference Bandura2002, Reference Bandura, Moghaddam and Marsella2004; Ditto et al. Reference Ditto, Pizarro and Tannenbaum2009). Consequently, the unkinder the previous interaction, the lower the moral cost of lying. Thus, the conditional probability of lying would decrease with increasing kindness. Whether subjects would also lie in the Spiteful domain depends on whether they are willing to forgo payoffs to punish unkind behaviors, and if so, the prediction on the conditional probability of lying would be the same as in the Selfish domain. Finally, the moral cost of lying may also depend on whether lying results in a benefit or a loss for the deceiver. Two competing hypotheses arise from prior literature. If the moral cost is higher when the lie is beneficial due to being perceived as selfish (e.g., Carlson and Zaki Reference Carlson and Zaki2018; Lin-Healy and Small Reference Lin-Healy and Small2012; Newman and Cain Reference Newman and Cain2014), the conditional probability of lying would be smaller in the Selfish domain than in the Spiteful domain. Instead, if the moral cost is smaller when the lie is beneficial due to a self-serving bias (e.g., Babcock et al. Reference Babcock, Loewenstein, Issacharoff and Camerer1995; Dana et al. Reference Dana, Weber and Kuang2007; Exley Reference Exley2015; Gneezy et al. Reference Gneezy, Saccardo, Serra-Garcia and van Veldhuizen2017; Konow Reference Konow2000), the conditional probability of lying would be larger in the Selfish domain than in the Spiteful domain.