Introduction

Trust as a social lubricant reconciles interactions between individuals, groups, and organizations (Adolphs et al., 1998; Bellucci et al., 2017; Krueger & Meyer-Lindenberg, 2019; Lount & Pettit, 2012). However, not everyone deserves to be trusted, and earning trust is often difficult. In the sphere of interpersonal relationships, personal characteristics can shake people’s decisions that are dependent on a sense of trust existing between participants. Social status as a typical instantiation of power dimension (competence, dominance, hierarchy, etc.) (Fiske, 2012; Tavares et al., 2015; Wu et al., 2014), is an essential social characteristic of people that attracts researchers to explore the relationship between social status and trust. Previous studies have paid active attention to one dimension, that is, how the social status of the person who trusts others (the “trustor”) influences their trust tendency (Dahlhaus & Schlösser, 2021). However, the degree to which social status, as the characteristic of the person trusted (the “trustee”), affects others’ trust-related decisions is limited. This issue should not be ignored because it is the antecedent to some significant phenomena, such as leadership and organizational performance. It is no exaggeration to say that this issue may be frequently encountered in the daily working life between employees and superiors (Dahlhaus & Schlösser, 2021). From the employees' perspectives, their degree of trust in superiors may be related to their productivity. Needing urgently to seek insights into the above issue, in this study, we filled this gap by exploring the impact that a trustee’s social status has on trust. We (1) characterized the effect of social hierarchy on trust-related decisions in different trustworthiness situations and (2) addressed its underlying cause by seeking insights into the computational processes guiding value-based decision-making and trustworthiness learning. These two objects of this study were intended to illustrate how social status, as a social prior of a trustee, affects the way of trust-related processing as well as how it integrates into this processing.

Modern behavioral economic models have made enormous strides in understanding social dilemmas by incorporating social prior information into consideration of social decision-making, which can interpret social influence in cognition (Delgado et al., 2005; Stanley et al., 2011; Yu et al., 2014). In the context of trust-related behaviors, social priors are factors worth considering that have already been confirmed by previous research, such as the impact of affiliation relationships (e.g., friends and strangers (Fareri et al., 2015)), impressions (e.g., moral character Bellucci et al., 2019; Fouragnan et al., 2013)), social categorization (e.g., ethnicity (Cañadas et al., 2015)), and characteristics of appearance (e.g. facial features and physical attractiveness (Yu et al., 2014)). It is not difficult to recognize from these studies the potential connection between trust and some social factors that are linked to benevolence, integrity, and affective commitment (Colquitt et al., 2007). For the issue of trustees' social status as a social prior affecting others' level of trust, one study examined this using another variable (the promise) and found that participants place more trust in the promises made by high-status individuals (Blue et al., 2018, 2020). Do people trust individuals of different social statuses asymmetrically in a situation where they process social status and trust directly? We assume that the advantages accruing to superiors from their social status may benefit them to gain trust rather than inferiors, which may relate to a factor that has a close link with trust——affiliation. Affiliation is an important dimension of relationships (Fiske, 2012; Tavares et al., 2015; Wu et al., 2014) and can effectively negate the interpersonal risk of trust. Based on the social distance theory of power, asymmetric dependence between individuals with asymmetric control over valued resources produces many phenomena that are related to the differences between different power-holders (Magee & Smith, 2013). More explicitly, the lower-ranking person who demands resources becomes more dependent on the higher-ranking person who holds the resources for desired outcomes. Differences in control over valued resources between people lead to differences in the degree to which they depend on each other, as well as different degrees of attraction of others’ motivation for affiliation. This asymmetric dependence was recognized by the power-as-control model (Fiske., 1993) and as a basic assumption introduced by the social distance theory of power (Magee & Smith, 2013). In terms of our research, people’s motivation for affiliation may target superiors rather than inferiors, which can be achieved by building trust. To address this issue, the first aim of this study was to reveal the effect of trustees’ social status on trust-related decisions.

In addition to social priors, trust-related decisions also depend on new evidence or, to put it another way, the trustee’s current trust-related performance (Bellucci et al., 2017). Previous studies examined three types of trust that emerge as individuals interact with another person: calculus-based trust, knowledge-based trust, and identification-based trust (Krueger & Meyer-Lindenberg, 2019; Lewicki & Bunker, 1995). From the perspective of the “trustor”, this process reflects trust-fostering. In the early stage, trustors adventure in ambiguous situations as well as internal computing for trustees’ feedback, then along with trust-related interactions, knowledge of their trustworthiness obtained with information accumulation, and, lastly, rewarding or unrewarding identification built at last. This processing implies that there is a link between a specific person and trust-related benefits, which allows an understanding of interpersonal trust to expand to a view of reinforcement learning (Delgado et al., 2005; Fareri et al., 2012; Fareri et al., 2015). In terms of the current issue under investigation, it is essential that the social prior of a specific person’s social status is integrated into trust-related decisions. Thus, the other aim of this study was to examine the underlying mechanism of trustworthiness learning by analyzing the integrated impact that knowledge of the social status and trustee feedback have on trust-based decision-making.

We approach the above question by investigating it from the viewpoint of computational modeling. In classical reinforcement learning, individuals make decisions depending on the expected value, which is updated based on prediction error in a trial-by-trial way (Konovalov et al., 2018; Lockwood & Klein-Flügge, 2021). Prediction error (PE) measures the difference between expectation and the actual result, and, with the constraint of learning rate, the knowledge gained from feedback was quantified (Konovalov et al., 2018; Lockwood & Klein-Flügge, 2021). This approach helps characterize how different motives and contexts influence behaviors by specifying the process by which individuals transfer relevant experimental variables into the expected value (Konovalov et al., 2018). The expected value is an essential foundation for decision-making and is related to the objective value, which, in the context of trust in reciprocity, is the reciprocity profit; however, human pursuits typically go beyond this. Decisions are also made by comparing choice options in terms of their subjective values (Yu et al., 2021), which include some individual preferences (such as “tastiness” Maier et al., 2020; Sullivan et al., 2014)) and universally accepted abstract social values (i.e., morality, equity, affect (Fareri et al., 2015; Ming et al., 2008; Zhong et al., 2016)). Social status is a powerful social prior that indicates a consensual collection of beliefs about the relative value each member brings to the group (Berger et al., 1980; Kumaran et al., 2012; Magee & Galinsky, 2008; Polman et al., 2013), and it impacts many social behaviors and interactions, including fairness perception (Hu et al., 2014, 2016), empathy (Feng et al., 2015, 2016), resource allocation (Albrecht et al., 2013; Ball et al., 2001), and morality (Piff et al., 2012). This widespread influence suggests that there is a possibility that the impact of social status is an independent component that is not determined by other factors. Evidence gained from coordinate- and connectivity-based meta-analysis has revealed the possibility that guidance by the reward processing network has close links with the modulation of behavior in relation to the social hierarchy (Li et al., 2021). The social distance theory of power also begins with the assumption that differences in dependence between power-holders arise from differences in control over valuable resources (Magee & Smith, 2013). These accounts, whether derived from behavioral and neural experimental studies or theoretical discourse, demonstrate the close association between social hierarchy and values. It follows that, as to our research, the relationship between social status and trust may be driven by the social value that comes with social status. Thus, based on previous findings, we hypothesized that the social value of social status is another component of an individual’s expected value that is independent of reciprocity profits. We tested this hypothesis in a setting where participants made trust-based decisions in different trustworthiness situations, which allowed us to explore the stability of the impact of social status on trust. We formalized reinforcement models and tested the social value hypothesis by comparing models that highlight the social value term with other models based on other factors relative to learning. The proof of this hypothesis can provide an explicit explanation for the underlying mechanism of the effect of social status that is reconciled with the social distance theory of power.

In general, in our study, how the trustee’s social status and performance integrated and guided trust-related decisions triggered a surge of interest in our further exploration. On the one hand, we tested the stability and universality of the effect of social status in different performance-specific situations. To tease out the dynamic effect of social status, the accessibility of a trustee’s performance was gradually manipulated in three experiments using the trust game paradigm. Experiment 1 measured initial trustworthiness based solely on social status in a one-shot game. Experiment 2 preprogrammed three partners with different social statuses but with the same performance (reciprocity rate at 50%). Experiment 3 examined how a trustee’s performance modulates the social status effect by carrying out two sub-experiments that manipulated the accessibility of their performance from easy to difficult. On the other hand, we employed the approach of computational modeling, which integrated these two main factors (i.e., prior information about a trustee’s social statuses and their current performance of trustworthiness) into complete cognitive processing. Using this approach, which places the effect of social status into trustworthiness learning processing, we specified the frame that individuals transfer partners’ social status and trustworthiness into internal cognitive processing for making trust-based decisions. The uniqueness of the current study related above contributes to identifying the effect of the trustee’s social status on trust-related processing, explaining the underlying mechanism as well as providing quantitative evidence for relevant theories and phenomena.

Experiment

Experiment 1

Experiment 1 was designed to test the baseline of trust when individuals interact with partners bearing different social statuses. We conducted a one-shot trust game without providing feedback about a partner’s reciprocity. Participants had no opportunity to assess their partner’s trustworthiness after that point.

Methods

Participants

Participants were recruited through posters, forums, and networks. If they were interested in the study, they could contact researchers for relevant information and decide whether to participate in the experiment. The selection criteria for this study were: aged 18 ~ 50; in good health; and had not majored in psychology. Before the experiment, researchers interviewed participants to obtain the above mentioned information in order to decide whether to recruit that participant to this study. Participant recruitment for all experiments in this study was conducted as above. In the first experiment, in order to explore the baseline of individuals’ trust in partners with different social statuses using a larger and more diverse sample, one hundred and thirty-one participants were recruited in order to complete the experiment online. Nineteen participants were excluded from the analysis because they failed to complete the task. The final analysis was carried out on the remaining 112 participants (69 females, aged 26.92 ± 7.77). Participants provided informed consent and were compensated for their participation.

Stimuli

One hundred and fifty grayscale pictures of forward-looking male faces with neutral expressions were selected from the CAS-PEAL face database (Gao et al., 2008) as alternative facial stimuli. In order to eliminate the effect of facial features, another sample of participants (N = 30) rated these pictures for facial attractiveness and dominance on two 7-point Likert scales (1 = not at all, 7 = a lot). Forty pictures with neutral attractiveness (M = 3.14, SD = 0.34) and neutral dominance (M = 3.20, SD = 0.28) were selected as stimuli for Experiment 1.

Experimental design and procedure

Based on previous studies that investigated social status, an introduction about the definition of social hierarchy was provided to participants (Magee & Galinsky, 2008; Qu et al., 2017). We used the definition of social status that appears in the MacArthur Scale of subjective socioeconomic status, which refers to an individual’s rank in a social system in terms of their wealth, occupational prestige, and education (Adler et al., 2000; Kraus et al., 2011; Piff et al., 2010). They were told that high-status individuals were “those who are the best off, have the most money, most education, and best jobs,” whereas low-status people were “those who are the worst off, have the least money, least education, and worst jobs or no job” (Piff et al., 2012). After receiving their instructions, participants engaged in a 40-trial one-shot trust game in which they interacted with one partner a single time during each trial. Participants were informed that they would play as Player A (i.e., “trustors”) while interacting with participants from earlier sessions of this experiment, who played as Player B (i.e., “trustees”), and their pictures, along with information about their social status, were provided. Half of the 40 facial stimuli were randomly shown to participants with the caption “high-status” printed under the picture, and the other facial stimuli were displayed with the caption “low-status”.

During each trial of the game, participants were endowed with ten yuan (10¥), and they were asked to decide how much they would like to share with the person present on the screen. A bar under the picture indicated the transfer values “0¥,” “2¥,” “4¥,” “6¥,” “8¥,” and “10¥” with their corresponding keys underneath displayed until participants responded. Participants were told that a choice to invest money in someone resulted in a quadrupling of the money transferred to the trustee who had already made a decision to either share half of the money with them or to keep it all. No feedback was provided to the participants, and the next trial started began one second (1 s) later; however, participants were informed that one randomly selected trial would be realized at the end of the experiment and that they might obtain a reward based on their choices and their partner’s reciprocity in this selected trial.

After participants had completed the trust game and rested, they performed a trustworthiness rating task. They viewed the partner’s picture with the caption about their social status and made a trustworthiness rating for each partner on a 9-point Likert scale (1 = not at all trustworthy, 9 = extremely trustworthy).

Measurement instruments

The presentation of the task and the recording of behavioral responses were performed using DiggMind software which can implement experimental tasks online. During the experiment, participants were required to concentrate on the task in a quiet setting.

Results

To examine the effect of social status on the baseline of trust, a paired samples t-test was conducted on the share amounts with the partners’ social statuses serving as within-subject variables (superior vs. inferior) (Fig. 1A). The results indicated a significant main effect of social status, t(111) = 4.113, p < 0.001, Cohen’s d = 0.389. In Experiment 1, during which participants played a one-shot trust game without receiving any feedback, they invested more in high-status partners (M = 3.702, SD = 1.823) than in low-status partners (M = 3.237, SD = 1.713). With respect to reaction time, we conducted the same paired samples t-test with the partners’ social statuses serving as within-subject variables, and no significant main effect was found, t(111) = -1.122, p > 0.05, Cohen’s d =  − 0.106.

Fig. 1
figure 1

Trust game decisions in Experiment 1 and 2. (A) Experiment 1: Share amount on average across the experiment condition of social status (± s.e.m); (B) Experiment 2: Percentage of trials in which participants shared in each block on average across the experiment condition of social status (± s.e.m). * p < 0.001

Full size image

Using a paired samples t-test, in subjective trustworthiness rating, no significant differences between high-status partners (M = 4.40, SD = 1.05) and low-status partners (M = 4.49, SD = 1.17) were found, t(111) =  − 1.122, p > 0.05, Cohen’s d =  − 0.106.

Discussion

In order to explore the initial trust that is gained from social status, Experiment 1 used a one-shot trust game, and no trust-related feedback was provided to participants. Without the possibility of learning partners’ trust-related performance, participants invested more in high-status partners than in low-status partners, demonstrating their higher degree of trust in the former group.

In Experiment 1, the participants’ investment amounts reflected a superior bias in the trust game, even when participants had no information about their partner’s trustworthiness. This may suggest that participants held a more positive belief that predicts more reciprocity in the future; however, both groups received similar trustworthiness ratings, which indicates that this type of positive belief extends beyond trustworthiness in the case of individuals who are deemed socially superior. We suggest that partners with high social statuses were endowed with an abstract social value. In other words, building relationships through trust with individuals of superior status is much more valuable than forming connections with people with a lower social status.

If participants are provided with their partner’s feedback, will this type of superior bias still emerge? More specifically, do people still view individuals who have a higher social status as valuable when sufficient information about their trustworthiness is available? Experiment 2 used a repeated trust game in which partners’ feedback was provided as a means of determining whether the superior bias still exists in this type of context and verifying the account of higher social value on the superior.

Experiment 2

Experiment 2 was designed to replicate and extend the findings of superior bias in baseline trust to a dynamic interaction context in which partners with different social statuses had the same neutral trustworthiness. In Experiment 2, we conducted a repeated trust game that provided an opportunity to gain more insights through computational modeling. Based on this approach, we tested the hypothesis that the superiors’ high social value led to them earning more trust from others.

Methods

Participant

In the current study, fifty-seven healthy participants (31 females, aged 20 ± 1.37 years) were recruited in Experiment 2. The procedure for participant recruitment in Experiment 2 was the same as that used in Experiment 1. Informed consent was given by all participants, and they were compensated for their participation after they had completed the experiment.

Stimuli

A total of eight facial stimuli (four female facial stimuli) were taken from the Chicago face database (Ma et al., 2015). To control the possible effect of gender, each participant was randomly shown three facial stimuli of the same gender in the task. The combination of facial stimuli and social status was counterbalanced across participants to control for potential confounding factors, such as facial features.

Experimental design and procedure

Following arrival at the laboratory, participants were informed about the procedure of the experiment and were asked to complete a consent form. Because the experiment involved the manipulation of partners’ social statuses, a fictitious experimental project was introduced in order to foster a feeling of reliability in the participants. In the introduction, the experimenter presented a fictitious “Decision Information Collection Project (DIC Project)” and invited them to participate in it. They were told that if they volunteered, they would become someone’s partner in another experiment (i.e., the decisions that they made in this task would be presented as data when they were paired with another participant). In addition to this, they were required to provide a photo for the experimental program. They were informed by the experimenter that participation in the DIC Project was a voluntary process, and participants could choose whether to join or not and that they were not required to become involved. Participants were also told that the current experiment did not have a direct relation to the DIC Project. To ensure that participants believed in the authenticity of this project, their level of participation willingness was assessed by the experimenter at the end of the task and they were told that if they decided to participate in the DIC Project, they would receive 5¥ as compensation.

After following these instructions, participants were told that they would play as Player A (i.e., “trustors”) while interacting with three participants from earlier sessions of the experiment who played as Player B (i.e., “trustees”). The partners’ photos and information about their social statuses were provided. Participants were told that their partners were randomly selected from the DIC Project, which collected the decisions that they made during the experiment to build a decision database. Participants were told that more than 500 people who played as Player B had already performed the same task and that the Player B database had already been established. At the same time, additional basic information related to comprehensive performance, including their highest educational degree attained, income level, and college entrance exam score, was collected, and a set of ability tests (including an intelligence test) were performed. All of the participants were ranked as inferior (marked as one star), intermediate (marked as two stars), or superior (marked as three stars) according to their comprehensive performance. We randomly selected three partners from the database. In the trust game, a partner’s photos and the number of stars corresponding to their social status were shown to participants (Fig. 2A).

Fig. 2
figure 2

Task schematic. (A) Partners in Experiment 2; (B) Trial procedure of trust game with a discrete selection reaction pattern; (C) Partners in Experiment 3; (D) Trial procedure of trust game with a contiguous selection reaction pattern

Full size image

After receiving the information about the social status of their partner, participants played the repeated trust game (Fig. 2B). During each trial of the task, the participant played the role of the trustor and interacted with one of the three partners who served as the trustee. Participants were given 5¥ before starting a trial, and could decide to either keep or share the money with the partner (trust stage). If the participant decided to share the money with their partner, the money was tripled (15¥) and given to the partner. At that time, their partner had the opportunity to decide whether to reciprocate with the participant (reciprocity stage). If the partner decided to reciprocate, the money was shared equally between the participant and their partner, and each of them received 7.5¥ (positive result). Otherwise, the trustee kept all of the 15¥, and the trustor obtained 0¥ (negative result). In the trust stage, if a participant decided to keep the money, they received 5¥, and this round of the game ended. After participants had made their decisions, they were presented with one of three possible types of feedback for two seconds (2 s), based on their responses: “You have kept the money,” “She/He has chosen to keep the money,” or “She/He has chosen to share the money.” The next trial began one second (1 s) later.

This task consisted of eight blocks, and each block included nine trials (72 trials in total, 24 trials for each partner). Unbeknownst to participants, the trustee’s decisions in the reciprocity stage were not controlled by other participants. Trustees had a preprogrammed reciprocity rate (i.e., the same neutral 50% return rate in which participants chose to invest). Before the experiment, participants were informed that one random trial would be selected at the end of the game and that they would obtain 0¥, 5¥, or 7.5¥.

Measurement instruments

The presentation of the task and the recording of behavioral responses were performed using E-Prime (version 2.0).

Model construction and estimation

Using a computational modeling approach within the context of a repeated trust game, we formalized and tested the social value hypothesis by comparing it with other models making differing predictions regarding possible factors relative to learning (Delgado et al., 2005; Fareri et al., 2012; Fareri et al., 2015). Model construction in this study was based on the reinforcement learning model, which included the decision process of pursuing benefit maximization and the updating process of decreasing the difference between outcome and expectation. We followed the Rescorla-Wagner updating rule (Sutton & Barto, 2018), which can be described mathematically in the following way: the expectation of future reward probability (i.e. the probability of reciprocation in this experiment, P(t+1), Eq. 1) is a function of current expectations (P(t)) and their discrepancy from the actual outcome that is experienced (γ(t), γ = 1 when partner reciprocates, γ = 0 when partner keeps), known as the prediction error (PE(t), Eq. 2), multiplied by a learning rate (\(\alpha\)) (Lockwood & Klein-Flügge, 2021). With the Eq. 3, the reciprocity rate converts into the expected value. Then, to calculate the probability of a participant investing with a given partner (IP, Eq. 4), the expected value (EVt) was calculated using a softmax function. The parameter β reflects whether a participant is more likely to behave in a more explorative or exploitative manner.

On the basis of the above learning processing, we constructed three models: the E2_SV model, the E2_RL model, and the E2_SV&RL model. In the E2_SV model, the expected value term included two types of values (Eq. 3): the monetary value (i.e., gain from the trust game) and the abstract social value (i.e., the effect of social status), which increases with the social status index (isocial status). The E2_LR model assumes that participants learn the trustworthiness of partners with different social statuses at different learning rates. The E2_SV&RL model considers the possibility that social status creates an effect through perceived social value and impacts learning rates. (see Supplementary Information for more detail).

To estimate the free parameters of each model for each participant, log-likelihood estimation was calculated using the maximizing function in Eq. 5, where j indexes the decision (share or keep), and n is the total number of trials.

For these three alternative models, we used the Akaike Information Criterion (AIC; Akaike, 1974), which applies a penalty scaled by the number of free parameters of a complicated model, to choose a more representative model. These estimations were conducted using custom MATLAB scripts.

E2_SV Model:

$${P}_{t+1}={P}_{t}+\alpha *{PE}_{t}$$
(1)
$${PE}_{t}={\gamma }_{t}-{P}_{t}$$
(2)
$${EV(S)}_{t}={P}_{t}*\left(m*7.5\right)+{i}_{social\ status}*\theta$$
(3)
$$IP=\frac{{e}^{\frac{{EV(S)}_{t}}{\beta }}}{\left({e}^{\frac{{EV(S)}_{t}}{\beta }}+{e}^{\frac{{EV(K)}_{t}}{\beta }}\right)}$$
(4)
$$LLE={\sum }_{t=1}^{n}{\text{log}}(IP,{j}_{t})$$
(5)

Results

Behavioral results

In order to explore the effect of social status on trust decisions over the course of the experiment with social status (superior vs. intermediate vs. inferior) and Block (block1-block8) serving as within-subject variables, a two-way repeated ANOVA was conducted on the percentages of decisions shared (Fig. 1B). The results showed a significant main effect of social status, F(2,112) = 7.524, p < 0.001, \({\eta }_{p}^{2}\)= 0.118. Post-hoc comparisons using Tukey correction demonstrated that participants were more likely to share with superior partners (M = 0.69, SD = 0.19) than with intermediate partners (M = 0.62, SD = 0.20, ptukey > 0.05) or inferior partners (M = 0.58, SD = 0.24, ptukey < 0.001), but there was no significant difference in share rate between intermediate and inferior partners (ptukey > 0.05). No significant main effect of Block or interaction of Block × social status was found ( all ps > 0.05).

We then conducted a one-way ANOVA with social status (superior vs. intermediate vs. inferior) within-subject variables on reaction time, and no significant difference was found (F(2,112) = 1.211, p > 0.05, \({\eta }_{p}^{2}\) = 0.021). Behavioral data analyses of all experiments were conducted by Jamovi (version 1.6.3).

Revelation in the computational model

The results of the model estimation and comparison are shown in Table 1. These results confirmed the social value hypothesis, as the E2_SV model fitted the participants’ data better than the E2_LR model (t(56) =  − 7.00, p < 0.001, Cohen’s d =  − 0.928) and the E2_SV&RL model (t(56) =  − 11.625, p < 0.001, Cohen’s d =  − 1.537). These findings suggested that participants learn about their partner’s trustworthiness through a trial-and-error approach based on their interactions in the trust game. Importantly, the superior bias in trust in which participants are more likely to invest money with a higher status partner is primarily driven by a social reward bonus that is independent of monetary reward.

Table 1 Model parameters in experiment 2
Full size table

Discussion

In Experiment 2, participants played the repeated trust game with three partners who had different social statuses that showed the same reciprocity rate. Here, a superior bias still emerged, even with abundant information about a trustee’s trustworthiness. As a means of interpreting the positive beliefs attached to superiors, we used a modeling approach to investigate the underlying process of trust behaviors biased by social status. Computational modeling revealed that social status can influence trust-related behaviors in a way that is represented as social value, independent of monetary reward. A possible explanation for this type of superior bias is that a partner’s trustworthiness is not easy to detect when the trustee’s neutral and same return rates are 50%. In Experiment 3, we manipulated the trustees’ reciprocity rates to make them more distinguishable.

Experiment 3

Experiment 3 included two sub-experiments with different learning difficulties. Experiment 3a used the repeated trust game in the same way as Experiment 2, which employed a binary selection reaction task. In the binary trust game, trustors were endowed with the same amount of money in each trial, which they could either keep or share with the trustee. In the binary version, they were expected to make a binomial forced-choice, which helped create a qualitative description for their discrete decision (i.e., keeping the money was regarded as a distrust decision and sharing the money with a trustee was regarded as a trust decision) (McCabe & Smith, 2000). Thus, in this task, the return rate, which was set higher or lower, was more detectable by participants. Experiment 3b conducted a repeated trust game using a contiguous selection reaction pattern. In the contiguous selection version, with an endowment by the experimenter, the trustor could pass any portion of the endowment to the trustee while keeping the rest or distrust them and not share any money (Evans & Krueger, 2009). Thus, the amount of money passed on by the trustor captured trust; so, the degree of trustworthiness was more difficult to detect in Experiment 3b.

Experiment 3a

Methods

Participants

We recruited 47 participants (26 females, aged 20.77 ± 1.89 years) for Experiment 3a. The procedure for participant recruitment in Experiment 3a was the same as that used in Experiment 1. Participants provided informed consent and were compensated for their participation.

Experimental design and procedure

Experiment 3a employed a 2 (social hierarchy: superior vs. inferior) × 2 (reciprocity rate: low vs. high) within-subject design. The manipulation of the partner’s social status was the same as that used in Experiment 2. The experimenter gave participants information about their partners and accounted for the meaning of the number of stars. Unlike Experiment 2, participants encountered two superiors marked with three stars and two inferiors marked with one star, intermediates did not participate in Experiment 3a (Fig. 2C).

Experiment 3a employed a repeated trust game using the same discrete selection reaction pattern adopted in Experiment 2 (Fig. 2B). The partner’s reciprocity rate was manipulated when participants chose to invest. One superior and one inferior had a high reciprocity rate (SH partner and IH partner), meaning they reciprocated 70% of all the trials when participants invested. Participants also encountered one superior and one inferior who had a low reciprocity rate (SL partner and IL partner), meaning they reciprocated only 30% of all the trials.

A total of 96 trials took place during this task and were evenly distributed across eight blocks. Twenty-four trials per partner condition were randomly administered across these blocks.

Measurement instruments

The presentation of the task and the recording of behavioral responses were performed using E-Prime (version 2.0).

Model construction and estimation

Based on the same model framework as used in Experiment 2, we formalized three models: the E3a_SV model, the E3a_RL model, and the E3a_SV&RL model. We used the same method as used in Experiment 2 to conduct the model fitting and parameter estimation (see Supplementary Information for more details).

Results

Behavioral results

To explore the trustworthiness learning process with social status as prior information, a three-way repeated ANOVA of share rate was conducted with social status (superior vs. inferior), reciprocity (high vs. low), and Block (block1—block8) serving as within-subject variables (Fig. 3A). We found significant main effects for social status (F(1,46) = 14.233, p < 0.001, \({\eta }_{p}^{2}\) = 0.236) and reciprocity rate (F(1,46) = 46.539, p < 0.001, \({\eta }_{p}^{2}\) = 0.503). These results revealed that participants were more likely to share with superiors (M = 0.68, SD = 0.16) than inferiors (M = 0.58, SD = 0.19) and with partners who showed a high reciprocity rate (M = 0.73, SD = 0.20) compared with partners who had a low reciprocity rate (M = 0.52, SD = 0.17). The main effect of Block was significant (F(7,322) = 5.536, p < 0.001, \({\eta }_{p}^{2}\) =0.107), which demonstrated that share rates changed with learning processing. The interaction between Block and reciprocity rate was significant (F(7,322) = 8.759, p < 0.001, \({\eta }_{p}^{2}\)= 0.160). As partners’ return rates changed, participants’ share rates also changed, and they learned about different changes in return rates of partners throughout the experiment.

Fig. 3
figure 3

Trust game decisions in Experiment 3. (A) Experiment 3a: Percentage of trials that participants shared on average across the experiment conditional on the partner context (± s.e.m); (B) Experiment 3b: Share amount across the experiment conditional on the partner context (± s.e.m)

Full size image

Next, we conducted a two-way ANOVA with social status (superior vs. inferior) and reciprocity rate (high vs. low) serving as within-subject variables on reaction time, and no significant main effects or interactions were found ( all ps > 0.05).

Revelation in the computational model

The results of model estimation and comparison are shown in Table 2. These results of the AIC indicated that the E3a_SV model captured participants’ decisions better than the E3a_LR model (t(46) =  − 8.474, p < 0.001, Cohen’s d =  − 1.240) and the E3a_LR&SV model (t(46) =  − 6.771, p < 0.001, Cohen’s d =  − 0.988). This indicated that social value as an important internal factor associated with social status significantly influenced trust-related decision processing.

Table 2 Model parameters in experiment 3a
Full size table

Experiment 3b

Methods

Participants

A total of 58 participants (31 females, aged 20.48 ± 2.08 years) were recruited in Experiment 3b. The procedure for participant recruitment for Experiment 3b was the same as that used in Experiment 1. Participants provided informed consent and were compensated for their participation.

Experimental design and procedure

Experiment 3b employed a 2 (social hierarchy: superior vs. inferior) × 2 (reciprocity rate: low vs. high) within-subject design. The manipulation of the partner’s social status was the same as that used in Experiment 2. The experimenter gave the participants information about their partners and accounted for the meaning of the number of stars. As in Experiment 3a, participants encountered two superiors marked with three stars and two inferiors marked with one star (Fig. 2C).

The manipulation of the partner’s reciprocity rate was related to the setting of the task. Experiment 3b used a procedure similar to the one used in Experiment 2, except for the reaction pattern and partner reciprocity rate. Experiment 3b conducted a repeated trust game using a contiguous selection reaction pattern (Fig. 2D). In each round, participants were endowed with 100 points. They were instructed to select a portion of the points (from 0 to 100 points in increments of 10 points) to invest in the partner according to the corresponding keys displayed underneath their picture. Once they had made their selection, the invested points were tripled and transferred to the partner. The amount of their reciprocation depended on the investments of participants and the pre-set reciprocity rate. The median reciprocity rate was fixed at 33% (i.e., 33% of partners obtained points) for all four partners. The change rule of their reciprocity rate was different. One superior and one inferior who had higher trustworthiness (SH partner and IH partner) both had a higher rate of increase and a lower rate of decrease in their reciprocation. Their incremental step was 6%, and their reciprocity rate gradually reached 51%. Their decremental step was 3%, and the reciprocity rate gradually reached 24% (i.e., 24%, 27%, 30%, 33%, 39%, 45%, 51%). One superior and one inferior who had lower trustworthiness (SL partner and IL partner) both had a lower rate of increase and a higher rate of decrease in their reciprocation. Their incremental step was 3%, and the reciprocity rate gradually reached 42%. Their decremental step was 6%, and the reciprocity rate gradually reached 15% (i.e., 15%, 21%, 27%, 33%, 36%, 39%, 42%).

There were eight blocks of 28 trials n this experiment (224 trials. Seven types of reciprocity rates for each partner comprised one block, and their order was random. The trustworthiness of partners was more difficult to ascertain in Experiment 3b than it was in Experiment 3a.

Measurement instruments

The presentation of the task and recording of behavioral responses were performed using E-Prime (version 2.0).

Model construction and estimation

Based on the same model framework as Experiment 2 and 3a, we formalized three models: the E3b_SV model, the E3b_RL model, and the E3b_SV&RL model. The only difference was that we adapted these models to the contiguous-selection decision rule. The methods of model fitting and parameter estimation were the same as those used in Experiment 2 (see Supplementary Information for more details).

Result

Behavioral result

A three-way repeated-measures ANOVA, with social status (superior vs. inferior), reciprocity rate (high vs. low), and Block (block1- block8) serving as within-subject variables was conducted on the share amount to investigate how social status and reciprocity rate affected participants’ share amounts over time (Fig. 3B). The results showed that share amount varied as a function of both social status (F(1,57) = 25.514, p < 0.001, \({\eta }_{p}^{2}\) = 0.309) and reciprocity rate (F(1,57) = 47.659, p < 0.001,\({\eta }_{p}^{2}\) = 0.455). Participants shared more with superiors (M = 51.38, SD = 16.83) than with inferiors (M = 40.22, SD = 20.18) and more with partners who had high reciprocity rates (M = 52.00, SD = 17.23) than with partners who had low reciprocity rates (M = 39.60, SD = 18.60). We also found a significant interaction between these two factors (F(1,57) = 4.581, p > 0.05, \({\eta }_{p}^{2}\) = 0.074; IH partner: M = 45.13, SD = 21.99; IL partner: M = 35.31, SD = 21.15; SH partner: M = 58.86, SD = 18.07; SL partner: M = 43.89, SD = 19.87). Post-hoc tests showed that when given the opportunity to share with two partners who both had low reciprocity rates, participants shared more with the SL partner than with the IL partner (ptukey < 0.01), and when it came to sharing with two partners who both had high reciprocity rates, participants shared more with the SH partner than with the IH partner (ptukey < 0.001), which demonstrated the effect of social status. On the other hand, the effect of reciprocity rates also came into play when participants were given the opportunity to share with two superiors and two inferiors. They ended up sharing more with the SH partner than with the SL partner (ptukey < 0.001), as well as sharing more with the IH partner than with the IL partner (ptukey < 0.001), but there was no significant difference between the IH partner and the SL partner (ptukey = 0.97). The main effect of Block was significant (F(7,399) = 3.880, p < 0.001,\({\eta }_{p}^{2}\) = 0.064), and the interaction of Block × reciprocity rate was also significant (F(7,399) = 14.456, p < 0.001, \({\eta }_{p}^{2}\) = 0.202). These results indicated that participants learned about their partner’s trustworthiness through feedback and adapted their trust decisions accordingly.

For the reaction time, we conducted a two-way ANOVA with social status (superior vs. inferior) and reciprocity rate (high vs. low) as within-subject variables on reaction time, and no significant main effect or interaction was found (all ps > 0.05).

Revelation in the computational model

The results of model estimation and comparison are shown in Table 3. It was found that the E3b_SV&RL model suited the participants’ data better than the E3b_LR model (t(57) =  − 10.689, p < 0.001, Cohen’s d =  − 1.404) and the E3b_SV model (t(57) =  − 2.915, p < 0.01, Cohen’s d =  − 0.383).

Table 3 Model parameters in experiment 3b
Full size table

Given that the E3b_SV&RL model fitted the participants’ data the best, we further investigated whether the learning rate was affected by social status. The result of the paired-sample t-test demonstrated that when participants interacted with superiors, their learning rate was significantly lower than when they interacted with inferiors (t(57) =  − 3.055, p < 0.01, Cohen’s d =  − 0.401). A higher learning rate indicated that the learner is more likely to strongly reassess their decisions based on recent feedbacks (Lockwood & Klein-Flügge, 2021). This result demonstrated that when participants interacted with superiors, they were affected less by the immediate reinforcers and were more likely to treat superiors in a consistent manner.

Discussion

With two sub-experiments, we further investigated how social status affects the process of trustworthiness learning. These two experiments both showed the effects of social status and reciprocity rates on participants’ trust decisions, while they also revealed different learning processes. In Experiment 3a, the effects of social status and reciprocity rate guided learning. When given information about partners’ trustworthiness that was more accessible, participants learned trustworthiness more quickly in the later blocks, and both partners shared a similar learning rate. We suggest that the short learning window does not provide additional evidence of the updating process. In Experiment 3b, we implemented the learning process by creating a situation in which differences between partners’ reciprocity rates were more difficult to detect. We found that in this situation, social status as a social prior dominated the learning process. In addition to revealing a superior bias in learning results, modeling results showed a lower learning rate when participants interacted with superiors.

General discussion

Trust stands at the dawn of human cooperation (Bellucci et al., 2017; Ferrin et al., 2007; Jones & George, 1998). Studying the relevant factors affecting trust is a specific interest for researchers who devote themselves to understanding the social human, because trust plays an essential role in interpersonal relationships. When the map of trust-fostering gradually takes shape and becomes clear, it becomes apparent that there is a gap in understanding how social status affects trust-related behaviors. In this study, we developed three experiments to explore the behavioral expressions induced by social status in trust-related decisions under different trustworthiness situations and revealed the potential reason behind these behaviors using a computational modeling approach that integrated social status into the trustworthiness evaluation process. In general, we found robust superior bias in trust-related behaviors and discovered that social value linked to social status is a powerful drive operating behind this bias.

Superior bias in trust-related behaviors

In this study, we examined the effect of social status on trust-related behaviors in different trustworthiness situations. In Experiment 1, which employed a one-shot trust game with no feedback of reciprocity information, we found that participants demonstrated more trust-related behaviors when interacting with superiors than they did when engaging with inferiors. In Experiment 2, when we controlled the trustworthiness of different social status partners by providing the same neutral reciprocity rate (50%) in a repeated trust game, the high-status partner gained more trust than the low-status partner. Consistent with the above experiments, this superior bias was extended to Experiment 3, which involved setting different levels of trustworthiness that matched the different social statuses of partners in an easily observable situation (Experiment 3a) and a difficult to observe situation (Experiment 3b). Given different trustworthiness situations, we found evidence of a stable superior bias when individuals had to decide to trust or distrust partners who held different social statuses, whether it was in a situation where no trustworthiness information was provided, where partners possessed the same levels of trustworthiness or where different social status partners matched the high and low level of trustworthiness. Furthermore, the results of Experiment 3b highlighted the role of social status in situations where it was difficult to judge a partner’s trustworthiness. In this type of situation, the impact of social status was so pronounced that participants had a similar rate of investment in both untrustworthy superiors and trustworthy inferiors. The findings of our study thus provided further evidence that the superiors have a privileged advantage when it comes to gaining the trust of others.

Within the context of learning processes, the repeated games in our study captured the learning process that revealed how social status and reciprocity rates were integrated when participants updated and synthesized information and made decisions about whether to trust their partners or not. Social priors, such as the impacts of moral impression, can hinder or bias the later updating of trial-and-error learning processing, which has already been demonstrated by social learning studies (Delgado et al., 2005; Fareri et al., 2012; Fouragnan et al., 2013). In this study, we discovered that this type of trustworthiness learning processing was also affected by social status. This finding emerged from Experiment 3, which included two sub-experiments with different learning difficulties. In a difficult to observe situation (Experiment 3b), we observed that participants employed a lower learning rate when they encountered superiors than when they interacted with inferiors. Learning rate is an important subject-specific parameter that quantifies updating by scaling prediction error (Lockwood & Klein-Flügge, 2021). This result suggested weakened updating for superiors. In other words, it proved that immediate monetary reward had less influence when the decision was related to superiors and that participants preconceived beliefs about superiors led them to treat this group of individuals in a more consistent manner. We can speculate that this low learning rate likely empowers social status to shake trust-related decisions longer and weakens the effect of immediate feedback, so that the trustworthy inferior and untrustworthy superior earn a similar degree of trust. We noticed that this phenomenon appeared in Experiment 3b only and not in Experiment 3a, which employed a simpler task. We argue that the reason for this might be connected to learning and cognitive resources; that is, trustors interacted with superiors in an unguarded way with a lower learning rate when cognitive resources were limited. Searching out clues about people who have already made relatively positive impressions, and concentrating on learning from feedback when it comes to assessing individuals with more unknown characteristics can be viewed as appropriate strategies for making a wise trust-based choice, but they can also sometimes act as a double-edged sword and result in oversights.

Above all, social status not only affected trust behaviors but also biased the process of updating trust learning; however, we have to admit that learning is a complex process that is not solely motivated by information updating. Other phases of learning may also be affected by prior knowledge of social statuses, such as information collection and integration, and these issues are worth further exploration in future studies.

Social value account for superiors’ halo

A question that motivated our research on superior bias was “Why do high-status people gain more trust?” Based on our empirical findings, we suggest that, though trust is mediated by others’ trustworthiness, it cannot account for superior bias since this phenomenon emerged across different trustworthiness situations (Experiment 1 to Experiment 3). We also did not find any evidence that links superiors with trustworthy impressions (Experiment 1). This proves that the inherent power of social status resting in trust-related behaviors and learning does not work through trustworthiness. Based on previous studies that demonstrated the effect of social status permeating many aspects of human social life (Becker, 1963; Kraus et al., 2019), we assumed that its impact is independent and is closely link to social value. Using a computational modeling approach, we formalized and tested the social value hypothesis by comparing it with other models that make differing predictions regarding possible factors. The findings of computational modeling corroborated the claim that the superior’s high social value contributed to their tendency to gain more trust from others.

Trust in reciprocity was motivated by two types of valuable rewards: the expectation of monetary profit as the extrinsic incentive and building effective relationships as the intrinsic incentive (Declerck et al., 2013; Krueger & Meyer-Lindenberg, 2019). Thus, the asymmetric dependence between different social statuses stressed in the social distance theory of power can be shown to be exhibited in trust-related behaviors. Our finding of superior bias in trust-related situations is in line with the social distance theory. It is the explicit expression that leads us to further discuss the implicit mechanism. When a superior and an inferior have the same reciprocity rate, the superior’s advantage in gaining trust is based on their social value. The high-status group often has competitive advantages and primary access to precious resources, which enhance their social value (Chiao et al., 2009; Fiske, 1992). On the one hand, strengthening cooperation and reducing competition with superiors are both actions that have a positive impact, and they can be realized through building mutual trust. On the other hand, dominance theory suggests that in the social hierarchy, individuals fight for higher positions and mobilize their resources for survival and development (Cheng et al., 2013; Cummins, 1999; Magee & Galinsky, 2008). These cooperative relationships aimed at high-status individuals are often strategic or automatic since this group can help in dire situations or when opportunities for gaining resources arise (Blue et al., 2020; Silk, 1992; Stevens et al., 2005).

It is also plausible that accounts of social value provide insight into the “halo effect” that accompanies high status (Benoit-Smullyan, 1944). An advantage in social status simultaneously brings about other positive outcomes, such as more positive appraisals (Kraus et al., 2019), greater attention (Feng et al., 2015), and greater tolerance (Becker, 1963). Accounting for social value has been proposed as a means of interpreting attitudes toward norm violation by different social status groups (Hollander, 1958; Polman et al., 2013). Previous studies have shown that people judge high-status wrongdoers less harshly than low-status individuals who commit the same types of acts (Bowles & Gelfand, 2009; Ungar, 1981). It is speculated that this tolerance of norm violations for high-status individual exists because there is a perception that their merits offset faults, and thus they retain value to the group; however, if the latter is exhausted and they fail the group, their preferential treatment diminishes (Blue et al., 2018; Wiggins et al., 1965). Our findings provide objective and quantifiable evidence based on computational modeling for the accounting of social value.

Limitations and future research

In this study, combining three experiments with computational modeling we characterized how the trustee’s social status influences trust-related behaviors. We found that higher status holds an additional social value independent of trust profit, resulting in superior bias. However, more evidence is needed to support this view and this study provided clues for further explorations linking the effect of social status with that of social value, such as research based on the neural mechanism and interpersonal synchronization as well as studies by manipulating the social value of social status.

Conclusions

Employing three experiments, we found that participants trusted superiors more than inferiors across different trustworthiness situations, and thus revealed stable superior bias. Using the repeated trust game combined with computational modeling, we observed and dissected the process of trust learning and demonstrated that high-level concepts and social status can bias the information updating of immediate feedback. Moreover, reinforcement learning models confirmed the accounting of social value and provided insight into why people tend to grant goodwill to high-status individuals. These findings deepen our understanding of the effect of social status on trust-related behaviors by demonstrating how social status modulates trust-related behaviors and trustworthiness-updating via internal social value.