The role of trust in the social heuristics hypothesis

According to the social heuristics hypothesis, people intuitively cooperate or defect depending on which behavior is beneficial in their interactions. If cooperation is beneficial, people intuitively cooperate, but if defection is beneficial, they intuitively defect. However, deliberation promotes defection. Here, we tested two novel predictions regarding the role of trust in the social heuristics hypothesis. First, whether trust promotes intuitive cooperation. Second, whether preferring to think intuitively or deliberatively moderates the effect of trust on cooperation. In addition, we examined whether deciding intuitively promotes cooperation, compared to deciding deliberatively. To evaluate these predictions, we conducted a lab study in Colombia and an online study in the United Kingdom (N = 1,066; one study was pre-registered). Unexpectedly, higher trust failed to promote intuitive cooperation, though higher trust promoted cooperation. In addition, preferring to think intuitively or deliberatively failed to moderate the effect of trust on cooperation, although preferring to think intuitively increased cooperation. Moreover, deciding intuitively failed to promote cooperation, and equivalence testing confirmed that this null result was explained by the absence of an effect, rather than a lack of statistical power (equivalence bounds: d = -0.26 and 0.26). An intuitive cooperation effect emerged when non-compliant participants were excluded, but this effect could be due to selection biases. Taken together, most results failed to support the social heuristics hypothesis. We conclude by discussing implications, future directions, and limitations. The materials, data, and code are available on the Open Science Framework (https://osf.io/939jv/).


Study 3
In Study 3, we assessed whether self-report or behavioral trust measures were more predictive of intuitive cooperation. Participants played a one-shot anonymous public goods game under different time constraints, and we measured trust using the trust game [1,2], the Propensity to Trust Survey [3], and the World Value Survey trust question. We expected higher scores on the trust measures to be associated with more contributions under time pressure than under time delay, though we had no expectation regarding which measure would a better predictor. We also tested whether contributions were greater under time pressure than under time delay.

Methods
113 participants participated in this study in exchange for $8,000 Colombian Pesos and gained an additional amount in the public goods game and the trust game. We aimed for a sample size of 150 participants, but ended up with less due to resource and time constraints. We applied the same exclusion criteria as in previous studies and, additionally, excluded 13 participants that participated in Study 1. Our main analyses were restricted to player one in the trust game and, thus, we ended up with a sample of 57 participants (32 men, 24 women, and 1 other, M age = 21.91, SD = 2.89). Note that the smaller sample (compared to the expected sample) reduced our power considerably and therefore our findings should be replicated with larger samples before drawing any strong conclusions.
We conducted this study in the lab (at a public university in Colombia) to target a population with little experience with similar studies. The study was conducted in Spanish.
In the study, participants played a public goods game under different time constraint manipulations (time pressure or time delay), participated in a trust game (behavioral trust measure), and answered the Propensity to Trust Survey and the World Value Survey trust question (self-report trust measures). The public goods game implementation was identical to Study 1, except that participants were given a $4,000 Colombian Pesos endowment and there was no random payment. Participants were exposed to all trust measures, but were assigned to be either player one or two in the trust game to avoid deception (though we restricted main analyses to player one). We randomized the order of the games (public goods game and trust game) to examine order effects [4]. After the games, participants answered the Propensity to Trust Survey [3], the World Value Survey trust question, and other questions: experience with economic games and research participation more generally [5], social capital [6], Perceived Awareness of the Research Hypothesis scale and other questions about demand effects [5,7], demographics (including education given that we did not recruit only undergraduates), how many of the people in the room participants knew, and whether they participated in Study 1. We now explain the trust measures.
The trust game is a behavioral measure of trust [1,2]. In the game, there are two players in different rooms and the game has two stages. In the first stage, player one must decide how much of a show-up fee to send to player two (both players know that the money that player one sends to player two will be tripled), in the second stage, player two receives the tripled amount and decides how much of his or her money to keep (which in our setup is the amount sent by player one tripled plus the show-up fee) and how much to return to player one. Given space limitations, we ran separate sessions with players one and two and paid participants at the end of the week. The game involved real monetary incentives, with each player assigned $4,000 Colombian Pesos as a show-up fee. Trust was measured as the amount sent by player one to player two. Participants had to type the amount they wanted to send/return in a box. The trust game was anonymous and participants were not able to see the screens of other participants. The game was done on a computer, but players two were informed about the amount received in a separate sheet. We avoided random payment schemes, given that a trust game meta-analysis showed that they reduce the amount sent by player one [2]. We included four comprehension questions (text entered) and asked them why they made their decision (an open-ended question).
The Propensity to Trust Survey is a validated 21-item self-report instrument composed of two scales that measure individual differences in trust and trustworthiness.
Research has demonstrated that the trust and trustworthiness scales are reliable, related to several of the Big Five personality traits, and predict the behavior of people in a trust game better than the Big Five scales [3]. We administered the trust scale (7 items), included 14 Big Five filler items [8] to conceal the purpose of the questionnaire, and randomized the order of the questions. Trust was measured by computing an average with the items of the trust scale after reverse scoring them and responses were measured on a 6-point scale from 1 very inaccurate to 6 very accurate.
The World Value Survey trust question was the one used in previous research on intuitive cooperation [9]. The question asked respondents "To what extent do you feel you can trust other people that you interact with in your daily life?" and responses were measured on a 10-point scale ranging from 1 very little to 10 very much. We also included five interpersonal trust questions for exploratory purposes, 13 filler items to conceal the purpose of the questionnaire, and randomized the order of the questions. The questions were taken from the World Value Survey.

Results
We examined whether the cognitive process manipulation check succeeded and evaluated understanding of the games in the relevant sub-sample of trust game players one.
Decision times were right-skewed and we therefore applied a log 10 transformation (see Fig   1). Though the difference between log 10 decision times in the conditions was significant,

Fig 1. Distribution of decision times (left) and distribution of log10 decision times (right).
We used dominance analysis to compare the relative importance of the trust measures in predicting intuitive cooperation [10,11]. In the trust game, trust was measured as the amount sent from player one to player two (M = 2,612.35, SD = 1,280.77, n = 57); in the Propensity to Trust Survey, trust was measured using the trust scale (M = 3.63, SD = 0.88, n = 57; α = 0.74); and in the World Value Survey, trust was measured using a survey question (M = 7.39, SD = 1.61, n = 57). When predictor variables are orthogonal, determining the relative importance of those predictors is easy and straightforward (e.g., using squared bivariate correlations between each of the predictors and the criterion), but when predictor variables are correlated, estimation of relative importance is still possible, but it becomes more complex [12,13]. When predictors are correlated, the preferred statistical tools are dominance analysis [11] or relative weight analysis [12]. It has been shown that these two approaches yield virtually identical estimates of relative weights [12].
Specifically, dominance analysis estimates the importance of a predictor in multiple regression by evaluating its contribution to the R 2 across all possible subset regressions.
There are different forms of dominance ranging from complete dominance (strongest) to general dominance (weakest; note that if dominance is achieved in a stronger form, then it is also achieved in the weaker forms): 1) complete: additional contribution of one predictor over another needs to be satisfied in every subset model; 2) conditional: average additional contribution needs to be satisfied across all models of a fixed size, and 3) general: average additional contribution needs to be satisfied across all models [11]. We used residualized dominance analysis to test the predictive power of interactions, given that the standard version does not provide "clean" estimates of higher order effects [13]. Specifically, we compared the predictive power of the interactions between the trust measures and time pressure in predicting contributions. Dominance analysis was conducted using the yhat R package [14]. Survey trust question seems to be the better general predictor of intuitive cooperation, followed by the Propensity to Trust Survey, and the trust game, the average contribution of each predictor to the R 2 is quite small.
We selected the interaction that was the better predictor (World Value Survey × time pressure) and tested its significance (see Table 1    Whereas the t-test against the lower value suggests that we cannot reject effects as extreme or more extreme than -0.26, the t-test on the upper value indicates we can reject effects as extreme or more extreme than 0.26. The equivalence test based on Welch's t-test is therefore not significant, t(108.82) = 0.96, p = .17, implying that we cannot reject the null hypothesis that the effect is smaller than -0.26 or larger than 0.26 (see Fig 2).