«Building and Rebuilding Trust with Promises and Apologies Eric Schniter and Roman Sheremeta and Daniel Sznycer Online at ...»
While the sanctioning of false signals and our tendency to err to caution may reduce the frequency of false signals in a population, those individuals who expect to escape sanctions may be more motivated to use signals deceptively. In economies where opportunity costs of forgone trust-based exchange are larger, receivers tend to tolerate greater proportions of false signals to honest signals. Specifically, the logic of error management theory (for a review see Haselton & Nettle 2006) predicts that despite the existence of false signaling and the costs of receiving false signals, signals will tend to be received when opportunity costs associated with not receiving true signals of trustworthiness (from forgone advantageous exchange) are greater than costs associated with receiving false signals of trustworthiness (i.e., when the consequent exchange produces a loss). This economically justified tolerance of a rate of false signaling also predicts that individuals will exploit opportunities to use deception. This study explores the use of cheap signals (e.g., promises of reciprocation, personalized messages, and apologies) that do not directly affect payoffs of the game, or require monetary costs for production, yet are common features of trust-based interactions. Personalized communication may improve cooperation (Orbell et al. 1988; Bohnet & Frey 1999; Ridings et al. 2002; Zheng et al. 2002; Buchan et al.
2006) by facilitating coordination, decreasing social distance, raising solidarity, and providing the cues of familiarity that are normally associated with trustworthy relationships. Non-binding promises have also been shown to increase cooperation (Rubin & Brown 1975; Kerr & Kaufman-Gilliland 1994; Elingsen & Johannesson 2004; Charness & Dufwenberg 2006). In relationships where trust has been damaged, apologies and explanations have been shown to elicit forgiveness (Ohbuchi et al. 1989; Tavuchis 1991; Lewicki & Bunker 1996; Benoit & Drew 1997; Girard & Mullet 1997; McCullough et al. 1997, 1998; Girard et al. 2002; Witvliet et al.
2002) and promote future trust (De Cremer et al. 2010). These strategies are based on signals that are cheap to produce, raising the questions of how people use signals in these contexts; when the signals achieve their intended effects; and who benefits from their use.
In sum, while cheap signals are helpful for building new trust and rebuilding damaged trust to achieve efficient outcomes, they can be used deceptively and may be distrusted, making their reliability tenuous. Therefore, we expect that in our experiment trustees whose actions have already produced reliable cues establishing their trustworthy reputations (by keeping promises and not succumbing to more profitable opportunism) will be less incentivized (than previously distrusted trustees, or trustees whose reputations indicate untrustworthiness) to spend time and effort constructing messages to persuade investors to trust them, when those messages might be distrusted. Previously distrusted trustees who have not established trustworthiness and untrustworthy trustees (i.e., promise-breakers) are expected to make use of promises and messages in an attempt to affect investors’ decisions to trust. We also expect that when used and “working” to affect investors’ trust, signals conveying a trustworthy propensity will provide benefits to both investor and trustee on average.
3. Experimental Design and Procedures The experiment was conducted at Chapman University’s ESI laboratory. 458 participants (229 pairs) were recruited from a standard campus-wide subject pool for participation in an experiment that could last up to 45 minutes. Participants interacted with each other anonymously over a local computer network. The experiment, which lasted an average of 35 minutes total and did not involve deception, proceeded as follows. Upon arrival, participants in the experiment were told that they would receive $7 for participation, to be paid at the end of the experiment.
Participants then received instructions (see Appendix A) for a single trust game with (i) no indication of a subsequent game to follow and (ii) no promises that the experiment would end at conclusion of that game.
Participants were assigned to one of two roles: “Participant A” (investor), or “Participant B” (trustee). First, the trustee completed the following standardized statement (which we will refer to below as a promise) by selecting a natural number amount from 0 to 20: “I (Participant
B) promise to transfer back $___ of my income to you (Participant A) if you choose IN”. This statement was not binding, however. That is, the trustee was not obligated to transfer back the amount promised to the investor, and both trustee and investor knew this. The computer conveyed the trustee’s statement to the investor and then the investor chose either OUT or IN. If the investor chose OUT, she received $5 and the trustee $0. If the investor chose IN, then the trustee received $20 (the “income”), after which he selected a whole dollar amount from $0 to $20 to send back to the investor.
After the first trust game (Game 1) finished, participants were given instructions (see Appendix A) indicating that a second trust game (Game 2) would follow. In Game 2, participants were told they would remain in the same roles and interact with the same partner as in Game 1.
However, prior to Game 2, the trustee was given an opportunity to use a text box to send a oneway message to the investor. Trustees were told that “in these messages, no one is allowed to identify him or herself by name, number, gender, or appearance,” but that other than these restrictions, trustees could “say anything in the message.” If trustees wished not to send a message they were instructed to “simply click on the send button without having typed anything in the message box.” The computer conveyed the trustee’s message and subsequently the standardized promise to the investor, and then Game 2 proceeded. We specified that Game 2, which had the same rules as Game 1, was the last and final part of the experiment (i.e., there would be no subsequent games).3 There were 25 experimental sessions. Each session had between 10 and 24 participants.
The average experimental earnings were $18, ranging from a $0 to $40, plus $7 for arriving to the experiment on time and participating. No participant participated more than once, and no participant had prior experience with a similar game environment.
4.1. Game 1 We expect that trustees, aware of investor self-interest and motives for critical signal reception, would promise investors transfers of at least $6 (minimally higher than the payoff to the investor if he chooses OUT) but less than $20 (which would provide no benefit to the promise-maker). Two plausible focal points for promised return amounts are the midpoint of the $6-$19 range, $12.5 (though only whole dollar amounts like $12 or $13 could be chosen), and the even-split of $10. Wary that trustees’ may have less incentive to honor promises closer to $20 than to the even-split amount of $10, we also expect that investors should be more suspicious of the veracity of higher promises and therefore less likely to invest in higher promises. If the mind errs to caution, as we have suggested, and interprets the one-shot game as potentially repeatable, trustees who have been trusted should reciprocate enough to, at minimum, provide investors profitable returns on their investments. These predictions stand in stark contrast to the rational (non-cooperative) choice predictions that expect non-binding promises to have no effect on investors. According to rational choice theory, trustees who receive incomes should return nothing (despite what they may have promised) and, based on this, investors should always choose to not invest (regardless of the promise they received).
Figure 1 displays the aggregate distribution of investment and promise-keeping decisions in the experiment, while Figure 2 displays the distribution of promises made by trustees in Game
1. In Game 1, trustees on average promised to return $9.20 (SD=2.38) out of $20 and 83.4% (191/229) of investors chose IN.
After each trust game participants were also asked to fill out a 20 item survey in which they reported their emotional states consequent on their decisions, game interactions, and resulting outcomes. Analysis and discussion of the mediating roles of emotions are not included in this paper.
First we evaluate the distributions of Game 1 promises associated with trusting and distrusting decisions, respectively, and how these investment decisions affected investor and trustee earnings. The distribution of promises in Figure 2 indicates that investors who chose IN received promises in the range of $6-$19 (99% of the time), with promises of $12 or $13 relatively uncommon (1% of the time), and the promise of $10 most common (more than 50% of the time). Investors who chose OUT received lower promises on average (i.e., M1=$8.61 (SD1=4.33) versus M2=$9.31 (SD2=1.75); Wilcoxon rank-sum test, p-value=0.01, n1=191, n2=38), and, compared to trusted promises, received either relatively low or relatively high promises overall. To confirm this observation, we estimate probit models (see Table 1, specifications 1 and 2), where the dependent variable is the investment decision in Game 1 and the independent variables are dummy variables for promises less than $9 and greater than $11, as well as the amounts of these promises. In specification (1), the dummy variables are negative and significant, indicating that investors are less likely to invest when promises are either relatively low or relatively high. Moreover, specification (2) indicates that, among promises lower than $9, there is a positive correlation between the amount promised and the probability of investment.
On the other hand, among promises higher than $11, there is a negative correlation between the amount promised and the probability of investment. In other words, promises closer to the evensplit of $10 elicit a higher rate of IN responses.
In Game 1, investment yielded greater payoffs than non-investment for both investors and trustees. Investors who chose IN received back $8.19 on average, which is more than their original endowment of $5 (Wilcoxon signed rank test, p-value0.01, n=191). Trusted trustees earned an average of $11.81, more than the $0 of distrusted trustees (Wilcoxon rank-sum test, pvalue0.01, n1=38, n2=191). The OLS estimation of specifications (3) and (4) in Table 1 indicates that the amount returned by trustee is correlated with the amount promised.
Specifically, specification (3) indicates that both low and high promises are followed by lower returned amounts. Moreover, specification (4) indicates that amount returned increases as it gets closer to the even-split promise of $10. These results support the rationale for why investors receiving especially high or low promises should tend to choose OUT.
For the investors who chose IN, the mean amount returned of $8.19 was significantly lower than the mean trusted promise of $9.31 (Wilcoxon signed rank test, p-value0.01, n1=n2=191). Despite mean returns being lower than promised, we find that promises tended to be veridical: 81.2% of trusted promises (155/191) were kept (i.e., the amount returned was equal to or greater than the promise), and 18.8% (36/191) were broken (i.e., the amount returned was less than the promise). Below we use the terms “promise-keepers” and “promise-breakers” to refer to trusted trustees who exactly met or exceeded their promised amounts, and who returned less than their promised amounts (whether the returns were monetarily profitable to the investors or not), respectively.
4.2. Game 2 While cheap signals are manipulated by trustees, affect investors, and provide net benefits to both investors and trustees in Game 1, facilitating profitable trust-based exchanges where previous reputations had not been established, Game 2 provides us a relatively different game environment in which to study cheap signals. In Game 2, reputations have been established for many investors and trustees – raising the question of whether the use of cheap signals will still matter where cues of trusting and trustworthy behavior (or its absence) are available.
In Game 2, trustees promised to return $9.79 on average, a higher amount than the mean of $9.20 promised in Game 1 (Wilcoxon signed rank test, p-value0.01, n=229). Game 2 promises resulted in 87.3% (200/229) of investors choosing IN, only slightly more than the 83.4% (191/229) of IN decisions made in Game 1 (Fisher's exact test, p-value=0.59, n=229).
Trustee reputation as established in Game 1 and the new promises issued in Game 2 affected investment decisions in Game 2. The estimation of probit models (specifications 1 and 2 in Table 2)4 indicates that both promise-breaking in Game 1 and promises lower than the even-split in Game 2 elicited less investment decisions in Game 2. Overall, the investments made in Game 2 paid off. Investors who chose IN received back $8.73 on average, which is more than the OUT payoff of $5 (Wilcoxon signed rank test, p-value0.01, n=200). Trusted trustees earned an average of $11.27, more than the $0 of distrusted trustees (Wilcoxon rank-sum test, pvalue0.01, n1=29, n2=200). The estimation of specifications (3) and (4) in Table 2 indicates that higher promises in Game 1, a greater extent of under-return relative to promise in Game 1, and uneven split promises in Game 2 all predict lower amounts returned in Game 2. Overall, and similar to Game 1, promises in Game 2 tended to be veridical; 75% of promises (150/200) were In estimation of Table 2, we have excluded variables corresponding to the amount promised in Game 1 (i.e., Promise19 × Promise1 and Promise111 × Promise1), since these variables are highly correlated with returns.
Nevertheless, even when these variables are included the estimates in Table 2 are very similar.
kept or exceeded, and 25.0% (50/200) were broken. In the sections below we further explore the effects of promises and messages on Game 2 investments and earnings across different “types” of dyads aggregated by Game 1 decisions.
4.2.1. Promise-Keepers For the subset of 155 Game 1 promise-keeping trustees, we observe higher average promises in Game 2. Figure 3 displays the distribution of Game 2 promises resulting in IN and OUT made by these promise-keepers. Overall, this set of trustees promised to return an average of $9.46 in Game 2, which is higher than their average promise of $9.02 in Game 1 (Wilcoxon signed rank test, p-value0.01, n1=n2=155).