Original Article
Do sincere apologies need to be costly? Test of a costly signaling model of apology

https://doi.org/10.1016/j.evolhumbehav.2008.09.004Get rights and content

Abstract

The present study examined a costly signaling model of human apology. The model assumes that an unintentional transgressor is more motivated to restore the relationship with the victim than an intentional transgressor who depreciates the relationship. The model predicts the existence of a separating equilibrium, in which only sincere apologizers will pay a certain cost to restore the relationship, while dishonest apologizers will not. Accordingly, we hypothesized that the receivers of an apology would be sensitive to the cost involved in the apology. 2 Experiment 1, 3 Experiment 2 were vignette experiments, in which participants imagined that they were victims of an interpersonal transgression and received either a costly or no-cost apology. The costliness of the apology was manipulated by the presence of an apology gift in Experiment 1, and by inconvenience voluntarily experienced by the transgressor to make an apology in Experiment 2. In both experiments, participants found the costly apologizer to be more sincere than the no-cost apologizer. Experiment 3 employed a modified dictator game, in which a fictitious partner behaved in an unfair manner and apologized to the participants. The apology cost was manipulated as a fee for sending the apology message. The results of 2 Experiment 1, 3 Experiment 2 were replicated. In addition, when given a chance to send a complaint message to the unfair person, participants in the costly apology condition abstained from doing so. Implications of the study are discussed in relation to applications of the costly signaling theory to interpersonal behavior.

Introduction

The communicative abilities of animals/humans and their evolutionary origins have engaged the interests of many scholars from divergent perspectives (Hauser, 1997). In the animal signal literature, one of the most important issues is the reliability or honesty of signals (Zahavi & Zahavi, 1997). Honesty of signals deserves both theoretical considerations and empirical investigations, as honest communication systems are vulnerable to deceptive signalers and thus unlikely to exist without some mechanisms to keep them honest. Zahavi's (1975) handicap principle (also known as the costly signaling theory) explains a mechanism whereby honesty of a signaling system becomes evolutionarily stable: high-quality individuals can credibly communicate their quality by voluntarily accepting some handicaps (or cost) that low-quality individuals cannot bear (see also Grafen, 1990).

Costly signaling theory has been successfully applied to some aspects of human behavior, such as altruistic behavior of hunters (e.g., Gurven et al., 2000, Smith & Bliege Bird, 2000, Sosis, 2000), religious behavior (e.g., Irons, 2001, Sosis, 2003, Sosis & Alcorta, 2003) and human courtship behavior (e.g., Griskevicius et al., 2007, Miller, 2000). Although some authors have suggested that costly signaling theory is also applicable to everyday interpersonal communication (Andrews, 2001, Gangestad & Thornhill, 2007), social psychological studies have paid little attention to this theory. Nonetheless, social psychologists have been interested in, and in fact have investigated, deceptive behavior in everyday interpersonal communication (e.g., DePaulo et al., 1996, Ekman, 1985). Having participants keep a diary of their deceptions, for example, DePaulo et al. revealed that people tend not to tell exploitative lies, while they do often tell white lies and prosocial lies (e.g., lies told in order not to embarrass someone).

The DePaulo et al. finding is somewhat puzzling from the perspective of the evolution of communication because human language does not involve any cost that prevents exploitative deceptions — viz., telling a lie is no more costly than telling the truth (Lachmann et al., 2001, Zahavi & Zahavi, 1997). It is known that low-cost signals (or cheap talk) can be honest if signalers and receivers share their interests to a substantial degree or if the signaling game has the coordination game-like incentive structure (Crawford & Sobel, 1982, Farrell & Rabin, 1996, Maynard Smith and Harper, 2003). This explanation might hold in interpersonal communication between relatives or close friends. However, people do not always share their interests. Being deceived and exploited in social exchange is considered to be a serious adaptive problem (Cosmides, 1989). Hence, it is naturally predicted that verbal communication needs to be accompanied by some costly signal when a deceptive incentive is large enough (i.e., when the honesty of low-cost signals is not warranted by shared interests; Zahavi & Zahavi, 1997). As a test case, in the present study, we shall apply the costly signaling theory to human apology.

The apology-making context can be considered one of the situations where the reliability of signals becomes a crucial concern. If a victim unwittingly forgives a transgressor varnishing over his or her exploitative intent with verbal apology (e.g., saying “I am sorry”), he or she may be subject to similar transgressions again in the future. In this section, we shall first develop a formal model of costly apology. In the following sections, we shall briefly review previous studies related to the idea of costly apology and provide an overview of the present study.

As in the standard model of the signaling game, we assume two players: a message sender (S) and a message receiver (R). There is asymmetric information between S and R: S has private information to which R cannot have direct access. In the apology-making context, given some transgression committed by the apologizer, the private information is whether the transgression was committed accidentally. To make the model more concrete, suppose that cooperative S, denoted as SC, has accidentally committed a personal transgression against R, and obtained some benefit, be from it. SC sincerely feels sorry for it and says “I am sorry” to R for her wrongdoing (henceforth, we shall use feminine pronouns for S and masculine pronouns for R). Alternatively, the private information can be defined as whether S sincerely repented her capriciously committed transgression. In either case, it is expected that S's sincerity is correlated with the likelihood of her future cooperation.

Receiving S's apologetic statement, such as “I am sorry,” R needs to be cautious because he expects to receive a similar apology not only from SC but also from an exploitative S, denoted as SE, who does not sincerely feel sorry. SE might also say “I am sorry,” expecting that she will be forgiven and can exploit R again. Although S personally knows which type (SC or SE) she actually is, R does not. In some instances, R may infer from circumstantial evidence that S committed the transgression accidentally (Malle & Knobe, 1997) and assumes that S is SC. In other instances, R may suspect that S did it with exploitative intent and erroneously assume S, who is in fact SC, being SE. In the latter instances, because merely saying “I am sorry” will not work, SC and R need some costly signaling system that prevents SE from producing the deceptive signal.

Assume that both SC and R will gain the benefit of bc from one round of cooperative interaction, and S will gain the benefit of be (>bc) from one round of exploitation regardless of whether it was intentionally committed or not. Their interaction will be repeated with the probability of w if R decides to continue the relationship with his current partner. However, he will terminate the relationship when he thinks that the likelihood of his current partner being SE is too high to justify the continuation of the relationship. Not to lose the potentially beneficial relationship with R, SC has to somehow prove her true identity to R. If she successfully convinces him that she is SC, her net benefit from interactions with R is be+bc×w/(1−w). Here, w/(1−w) is the expected number of future interactions.

Suppose that SC pays the cost of a (≥be) in making her apology. By definition, SE is not willing to pay any cost greater than the benefit from the one-shot exploitation, be. On the other hand, SC has an incentive to pay it if she is better off by paying the cost a to assure the future benefit of bcw/(1−w) than keeping the benefit from one-shot exploitation, i.e., bebea+bcw/(1−w). The model can be summarized in the following inequality:beabcw(1w).When the above inequality holds, SC will make the costly apology while SE will not. Accordingly, R can be assured that anyone paying the apology cost of a is not SE.

The present model assumes that SC's paying a will be offset by the benefit from repeat interactions. A similar idea was proposed by an economist, Nelson (1974), to explain the utility of dissipative advertisements, whose cost, he supposed, is offset by repeat purchases only when the advertiser produces a high-quality good (see also Gintis, 2000, chapter 13). More relevant to the apology-making context, McElreath and Boyd (2007) proposed a similar explanation for why the contrite tit-for-tat (CTFT) strategy works in the noisy repeated prisoner's dilemma game, in which players sometimes defect by mistake. CTFT accepts the partner's defection once when it has accidentally committed a defection in the previous round. McElreath and Boyd argue that a person who accidentally committed a defection can credibly signal his or her good intent by accepting the partner's defection without retaliating.

Regardless of the similarity to the above models, the present model differs from them in its assumption regarding the signaler's type. It is common in the signaling literature that S's type is modeled as her stable trait (e.g., whether one produces a high-quality good is a stable trait of the producer). On the other hand, we do not necessarily assume that S's type is her trait, stable across situations and partners. If it is her trait, R may know a particular S's type not only from her signal but also from other sources, such as her reputation. We maintain that the costly apology is more important when S's type is relation specific. Some S may find the relation with the particular R less valuable (e.g., cooperative interaction with him may bring her a benefit much smaller than bc). Therefore, the sincerity in the present model can be defined as a proximate cue of S's valuation of the current partner, R (see Camerer's, 1988, signaling model of gift-giving for a similar assumption regarding different types of signalers; see also Appendix A for preliminary evidence for this assumption). In summary, the present model assumes that the more valuable S finds a particular relationship, the more likely she is to pay a substantial amount of apology cost. She will be likely to cooperate within the relationship because she highly values it.

It is useful to compare the present model to related models in order to clarify implicit assumptions and boundary conditions of the present model. In fact, some authors argue that no-cost signals work well in certain peacemaking settings. In the present section, we shall explain how the present model differs from the related models.

Silk, Kaldor, and Boyd's (2000) developed a low-cost signaling model of benign disposition and applied it to explain how female monkeys avoid hostile interactions by no-cost signals (grunts and girneys). The Silk et al. model assumes that a signaler is peaceful for p of the time, while it is hostile for (1−p) of the time. Silk et al. proved that if the receiver employs a conditional strategy (i.e., believe the signal until being deceived), the signaler's strategy of honestly signaling its current disposition is evolutionarily stable. Given the receiver's conditional strategy, a single instance of dishonest signal (i.e., disguising a peaceful disposition) will entirely eliminate the signaler's chance of future peaceful interactions with the receiver. Accordingly, insomuch as the expectation of future interactions outweighs the benefit from one-shot exploitation, every female monkey has an incentive to reveal her current disposition honestly, and thus the no-cost signals can be honest. This conclusion partly depends on the Silk et al. assumption that each S's type fluctuates over time even within a specific relationship. On the other hand, the present model assumes that it is constant over time within a specific relationship. Therefore, by definition, SE's expected benefit from future interactions would never exceed the benefit from one-shot exploitation. As such, SE has an incentive for dishonesty, and SC must pay the cost, a, to distinguish herself from SE. Another important difference between the present model and the Silk et al. model is the timing of the signal (i.e., before or after a transgression). The transgression is not yet committed in the Silk et al. model but has been committed in the present model. The timing is important because the pre-transgression signals leave more room for coordination, whereby low-cost signaling systems can be evolutionarily stable.

Recently, an economist, Ho (unpublished data), developed a signaling model of apology whereby he showed that no-cost apology can be honest. There are at least two conceivable reasons why Ho's model and the present model arrived at different conclusions in terms of the effectiveness of no-cost apologies. First, Ho's model deals with R's behavioral reaction toward various apologies, while the present model focuses on R's perception of sincerity. The present model can be considered as a subset of Ho's model (i.e., R's Bayesian inference process upon receiving an apology). According to the Bayesian game framework (Harsanyi, 1967–1968), R is supposed to possess some prior probability of S being SC. Observing S's transgression (i.e., SE-like behavior), R is likely to lower his subjective probability of S being SC. S's apology is thus considered to influence R's subjective probability. According to this framework, whether R forgives S is jointly determined by (1) R's subjective probability of S being SC, (2) his potential benefit from his relationship with S and (3) his potential loss from a further exploitation by S. Our model exclusively focuses on the first factor (i.e., R's subjective probability) referring to it as sincerity perception. On the other hand, if the focus is whether R will forgive S, it is logically predicted that cheap apologies are sufficient when the potential benefit is large and the potential loss is negligible (see Appendix A for preliminary evidence that R's perception of sincerity and willingness to forgive are dissociable under a certain condition).

The second difference between Ho's model and the present model is the assumption regarding the presence of some exogenous cost. Even when R is suspicious about S's type, the no-cost apology equilibrium can be sustained by the presence of exogenous cost, such as punishment against deceptive apologies (B. Ho, unpublished data). We agree with Ho that the presence of exogenous cost plays an important role to sustain honesty in human communication (Lachmann et al., 2001; Y. Ohtsubo, F. Masuda, & E. Watanabe, unpublished data). However, we are dubious about the effectiveness of punishment in the apology-making context. Suppose that a receiver accepts a no-cost apology but threatens the signaler by stating that “I will punish you if you do it again” (i.e., if the apology turns out to be dishonest). The signaler may not take the receiver's threat seriously because the receiver signals his disinclination to punishment by failing to inflict it exactly when the threat is being conveyed (Schelling, 1960).

The costly signaling model of apology predicts that people in the victim's perspective are sensitive to the presence of the apology cost in determining the apologizer's sincerity. Accordingly, it predicts that people will be more likely to forgive a costly apologizer than a no-cost apologizer, although we have noted that whether one perceives the apologizer as sincere cannot be equated with whether one forgives the apologizer. Supportive evidence has been accumulated by social psychological research on apology: if a transgressor voluntarily offers some compensation or gift, he or she is more likely to be forgiven (Darby & Schlenker, 1982, Eaton et al., 2006, Gauché & Mullet, 2005). A similar observation was made by a primatologist, de Waal (1989), who noted that gift-giving is a uniquely human way of peacemaking. Notice that an apologizer inevitably pays some cost in making compensation or giving a gift to a victim. We first tested the effect of the apology gift on the sincerity perception in a vignette experiment (Experiment 1).

The effect of the apology gift (or compensation) is, however, subject to an alternative explanation. Because the cost paid in the form of a gift or compensation is transferred to the victim, the materialistic value of a costly apology is greater for the victim than that of a no-cost apology. Accordingly, those who received a costly apology might perceive the apology more positively (i.e., more sincere) because they are happier than those who received a no-cost apology. This problem pertains not only to Experiment 1 but also to the previous research revealing positive effects of apology gifts and compensation. Our model, nonetheless, predicts that the cost per se is crucial for a victim to evaluate the apologizer's sincerity. Another vignette experiment, Experiment 2, was thus conducted to test the prediction that apology cost that is not transferred to a victim is sufficient for him or her to perceive the sincerity in the apology. For this purpose, we devised scenarios describing the cost such that it would reduce the transgressor's payoff but not increase the victim's payoff (e.g., apologizer's canceling a part-time job to make an apology as soon as possible).

Experiment 3 tested the same prediction with a behavioral experiment. In Experiment 3, participants played a modified version of the dictator game with a fictitious partner (see Camerer & Fehr, 2004, for an explanation of the standard dictator game). Participants learned that the partner allocated them only 200 of 1000 Japanese yen (JPY) and then apologized for the unfair allocation. Costliness of the apology was experimentally manipulated as a fee to send an apology message. The cost paid by the dictator was not transferred to participants. In addition to the perceived sincerity, Experiment 3 assessed participants' behavioral reaction to the costly/no-cost apologizer by asking them to indicate their willingness to send a complaint message to the partner. Recall that the present model does assume that the perceived sincerity influences R's behavioral reaction, while it does not assume that the perceived sincerity is the sole determinant of the behavioral reaction.

One might criticize Experiment 3, noticing that it violated an assumption of the model — viz. participants did not engage in the repeated interactions. We consider that the exclusion of the repeat interactions was a rather desirable feature to test the present, evolutionary hypothesis. There are three recent studies that compared the effect of costly/no-cost apology in the repeated economic game settings (Bottom, Gibson, Daniels, & Murnighan, 2002; B. Ho, unpublished data [available at: http://ssrn.com/abstract=944161]; Schweitzer, Hershey, & Bradlow, 2006). Interestingly, they differ in their conclusions regarding the effect of no-cost apology. Investigating the effect of different types of apology on trust restoration, Schweitzer et al. found little facilitative effect of the no-cost apology. Bottom et al. found a small but positive effect of no-cost apology. More importantly, however, the Bottom et al. study revealed that costly apology was more effective than no-cost apology. Ho also found a positive effect of no-cost apology and did not find a significant effect of the level of apology cost. The contradictory results might be partly attributable to the methodological features of the previous studies. All three studies measured the apology receivers' behavioral reaction in some repeated game setting. The repeated game allowed the participants to behave according to their prospects of success in the game, and thus whether no-cost apologies were effective might have depended on the specific game structure that those researchers employed in their experiments. This argument suggests that forgiving S who made a no-cost apology might have been a product of rational calculation in these studies. Exclusion of the repeat interactions from Experiment 3 allowed us to eliminate variants of rational choice explanations for the observed results.

Moreover, we argue that the present model in fact predicts the effect of apology cost on the perceived sincerity in the absence of the repeat interactions. The present model's specific prediction is that people have some innate sensitivity to the cost involved in apologetic signals. Such a cost-sensitive psychology is supposed to have evolved in the environment of evolutionary adaptedness (EEA), in which the repeat interactions along with the other game structures existed. Once evolved in the EEA, our innate psychological mechanisms are triggered by proximate cues that were correlated with the hypothesized game structures in the EEA (Hagen & Hammerstein, 2006, Haley & Fessler, 2005). We consider that plausible proximate cues for the sincerity assessment are the contextual frame (i.e., apology-making context) and cost involved in apologetic signals. We do not consider the repeat interactions as a plausible proximate cue for the sincerity assessment. In the EEA, most social interactions probably occurred among the same community members; thus most interactions were repeated (see Maynard Smith & Harper, 2003, p. 126, for a similar argument). Something that is always present cannot serve as a good criterion on which one bases his or her decision. Accordingly, we considered that Experiment 3 was rather a suitable setting to test the evolutionary hypothesis derived from the costly signaling model.

Section snippets

Participants and design

Participants were 60 undergraduates (23 males and 37 females) at a small private university in Japan. They volunteered to participate in a set of experiments for a monetary reward (1000 JPY). The costliness of apology (costly apology vs. no-cost apology) was a between-participants factor: 30 participants were assigned to the costly apology condition, and the other 30 to the no-cost apology condition. This study was followed by an unrelated experiment. Each experimental session involved a

Participants and design

Participants were 77 undergraduates (46 males and 31 females) at a small private university in Japan. The study was conducted in a large classroom as part of the course requirements. The costliness condition was a between-participants factor. To test the generalizability of the results, three new vignettes were written and included as repeated measures. All three scenarios presented to each participant were from the same condition, either the no-cost apology or costly apology condition. The

Participants

Participants were 42 (19 males and 23 females) undergraduates who were enrolled in an evolutionary psychology class at a small private university in Japan. They participated in the study for partial credit in the course and a small monetary reward (i.e., 200 JPY), whose exact amount had not been specified in advance.

Procedure

Forty-two participants were assembled in a large room. Each participant received an ID card on which his or her ID number was printed either in red or in blue ink. Participants were

General discussion

The present study demonstrated that a costly apology effectively communicates the apologizer's sincerity to the victim. In Experiment 1, we showed that the apology with a gift was more effective in communicating the apologizer's sincerity than the apology without it. In Experiment 2, we showed that an apology cost that was experienced as inconvenience merely to the apologizer was also effective in communicating sincerity. Experiment 3 tested the effect of a costly apology in an experimental

Acknowledgments

We are grateful to Keiji Iwasaki, Kimiko Shiode and Shin'itsu Tomohiro for their help in data collection, and to Martin Daly, Daniel Fessler, Tatsuya Kameda, Masanori Takezawa and anonymous reviewers for their helpful comments on earlier drafts. We would like to extend our thanks to our research assistants, Mina Adachi, Fumiko Masuda, Kazumi Nakae and Akihiro Shuda.

References (47)

  • SosisR.

    Costly signaling and torch fishing on Ifaluk atoll

    Evolution and Human Behavior

    (2000)
  • ZahaviA.

    Mate selection: A selection for a handicap

    Journal of Theoretical Biology

    (1975)
  • AmesD.R.

    Inside the mind reader's tool kit: Projection and stereotyping in mental state inference

    Journal of Personality and Social Psychology

    (2004)
  • BaumeisterR.F. et al.

    Guilt: An interpersonal approach

    Psychological Bulletin

    (1994)
  • BottomW.P. et al.

    When talk is not cheap: Substantive penance and expressions of intent in rebuilding cooperation

    Organization Science

    (2002)
  • CamererC.

    Gifts as economic signals and social symbols

    American Journal of Sociology

    (1988)
  • CamererC.F. et al.

    Measuring social norms and preferences using experimental games: A guide for social scientists

  • CrawfordV.P. et al.

    Strategic information transmission

    Econometrica

    (1982)
  • CrosbyF.

    A model of egoistical relative deprivation

    Psychological Review

    (1976)
  • DarbyB.W. et al.

    Children's reactions to apologies

    Journal of Personality and Social Psychology

    (1982)
  • de WaalF.B.W.

    Peacemaking among primates

    (1989)
  • DePauloB.M. et al.

    Lying in everyday life

    Journal of Personality and Social Psychology

    (1996)
  • EatonJ. et al.

    The mediating role of perceptual validation in the repentance–forgiveness process

    Personality and Social Psychology Bulletin

    (2006)
  • Cited by (110)

    • A pull versus push framework for reputation

      2023, Trends in Cognitive Sciences
    View all citing articles on Scopus

    This research was supported by the Japan Society for the Promotion of Science (grant no. 18730398).

    View full text