Introduction

Beauty has been playing an increasingly important role in human life. Extra social and economic opportunities and resources that are brought by individuals’ higher level of facial attractiveness is denoted as the beauty premium. Researchers have demonstrated the existence of beauty premium phenomenon in social exchange scenarios across different industries and cultural backgrounds (Hamermesh 2011). Previous studies detect the beauty premium effect in labour market by sending resumes to companies posting job advertisements (Busetta and Fiorillo 2013; Lopez et al. 2013; Ruffle and Shtudiner 2015). It has been found that facial attractiveness had a robust positive effect on response rate, and such effect was more prominent for female applicants than male applicants (Busetta and Fiorillo 2013). In labour markets, more attractive individuals are often associated with higher salaries (Hamermesh and Biddle 1994), while in social interactions, they tend to receive higher social evaluations and more favorable conditions (Mobius and Rosenblat 2006). Beauty premium holds significant economic implications, as it is on par with racial and gender disparities in income within the U.S. labour market (Hamermesh 2011). The beauty premium intensifies unfairness (Jones and Price 2017), undermining social equity. Therefore, it is quite necessary to weaken and even eliminate the beauty premium.

The beauty premium is an innate human behavioural bias that leads to decision-making errors in experimental economic games. Psychology research typically examines bias formation through three perspectives: motivational, cognitive, and sampling (Ma 2019). To control decision-making biases, these perspectives should be considered. However, the motivational and cognitive approaches focus on internal mechanisms that are difficult for managers to influence effectively. Additionally, previous research on factors like human capital and financial resources (Gu and Ji 2019) still targets individual intrinsic factors, which are challenging for managers to manipulate effectively to mitigate bias.

Most research on decision-making biases has concentrated on internal motivational and cognitive limitations within individuals. But they rarely addressed the interplay between individuals and their environments, as well as the impact of this interaction on decision biases (Fiedler and Juslin 2005). One of the main characteristics of real-life economic decisions is that decisions are made by groups rather than by isolated individuals (e.g. households, executive boards or policy committees) (Carbone et al. 2019). Within group decision-making context, views or preferences of decision-makers can be swayed by their peers, markedly affecting the results of the decision-making process (Tan 2021; Yu et al. 2021). Group decision-making is widely regarded as a mechanism for facilitating information exchange and collective processes. It plays a crucial role in shaping individual choices. Unlike isolated individual decisions, decisions made in group context are deeply embedded within a network of social relationships and influences, reflecting the broader principle that “no person is an island” (Baumeister and Leary 1995). Research on social decision-making has underscored the prevalence of group settings and their critical role in shaping both individual and collective biases (Sassenberg et al. 2014). This focus has led to the development of a preliminary framework that categorizes the impact of group decision-making on biases into three main outcomes: reduce, exacerbate or have no effect (Kertzer et al. 2022).

First, group decision-making may reduce individual biases by facilitating information sharing and discussion, which often leads to more objective decisions (Mesmer-Magnus and DeChurch 2009). This occurs because group discussions often encourage a broader consideration of multiple factors, shifting the focus from appearance-based judgments to more comprehensive and relevant information. Thus, groups can act as corrective mechanisms, particularly in fairness and equality-driven contexts (Kerr and Tindale 2004; Reimer et al. 2010).

Second, social influence within group setting may amplify the bias. When group members share similar stereotypes or inclinations, group discussions may amplify these biases, reinforcing pre-existing beliefs (Chen and Sun 2016; Stasser and Titus 1985). This mutual reinforcement may occur through various mechanisms such as social validation, conformity, or peer pressure, which can perpetuate and even exacerbate appearance-based biases (Stasser and Titus 1985).

Third, in some cases, group decision-making may have little to no effect on individual cognitive biases or preconceptions, particularly when group members are not engaged in meaningful dialogue (Al-Badawi 2024; Hsieh et al. 2020; Kertzer et al. 2022). This highlights the limitations of social influence in overcoming certain types of deeply rooted cognitive biases.

This framework provides valuable theoretical insights into how group dynamics regulate biases based on the nature of group interactions. However, prior research on the beauty premium effect has predominantly focused on individual decision-making, with no study investigating its impact within group decision-making context. This study aims to systematically explore the manifestations and mechanisms of appearance bias in group decision-making, addressing the key question: Does group decision-making affect the beauty premium, and if so, does it mitigate or amplify the beauty premium?

Literature review

Beauty premium

The relationship between beauty and labour market outcomes has been widely validated by studies conducted across different countries, including Britain (Harper 2000), China (Hamermesh et al. 2002), Sweden (Rooth 2009), Argentina (Mobius and Rosenblat 2006) as well as Canada and America (Fletcher 2009; Roszell et al. 1989). Comparable beauty premiums have been identified among participants in large household surveys in the United States, Canada, and the United Kingdom (Hamermesh and Biddle 1994; Harper 2000), as well as among young high school graduates (Fletcher 2009), Dutch advertising executives (Pfann et al. 2000), and restaurant servers (Parrett 2015). These studies consistently demonstrate that facial attractiveness significantly impacts employment opportunities, wage levels and occupational prestige. While cultural values play a significant role in shaping the degree of beauty premium. For instance, China and Germany show particularly high beauty premium, ranging from 17.9% to 20%. Meanwhile, the most significant penalty for plainness is observed in Britain and Australia, with wage reductions of 10.9% to 14.9% (Liu and Sierminska 2014; Peng et al. 2020).

Gender also plays a crucial role in how the beauty premium is distributed. In labour markets, female attractiveness often leads to higher wages and better employment prospects (Liu and Sierminska 2014), particularly in industries where client interaction and public perception are important. Researchers found that attractive women are more likely to secure positions in customer-facing roles, such as hospitality and retail, where appearance can directly influence business outcomes (Hamermesh and Biddle 1994; Maestripieri et al. 2017; Mobius and Rosenblat 2006). Moreover, study suggests that women’s attractiveness is often perceived as a marker of their social skills, friendliness and competence (Eagly et al. 1991). Men, on the other hand, tend to benefit less from their physical appearance, except in certain fields like politics or professional sports (Johnston 2010). This gender disparity highlights the intersection of beauty and gender norms, where women’s appearance is often more scrutinized and valued in economic and social contexts.

Facial attractiveness not only affects physical perceptions of appearance but also influences how individuals are judged in terms of social status. People tend to associate attractive individuals with higher social status, wealth and success, which in turn can amplify the beauty premium. This is particularly true in hierarchical societies, where appearance can serve as a visible indicator of one’s position within the social and economic structure (Holland et al. 2017). This interaction between beauty and social status suggests that the beauty premium is not merely a result of facial attractiveness but is also shaped by broader socio-economic hierarchies (Mattan et al. 2017).

Social influence

Group decision-making is widely applied in organisational settings, encompassing tasks such as personnel selection, policy development and negotiations (Hackman and Katz 2010; Kouchaki et al. 2015). Group decision-making significantly shapes individual behaviour by exerting social influence, which operates to affect how individuals perceive, evaluate, and ultimately decide. Social influence arises from interactions among group members and often manifests as social pressure, where individuals feel compelled to align their decisions with group norms or the opinions of influential members (Cialdini and Goldstein 2004). Such influence often drives individuals to adjust their attitudes, beliefs, or behaviours to align with those around them (Flache et al. 2017). Preferences are dynamic and may shift in response to changes in personal or social situations (Gintis 2018). This dynamic can lead to decisions that differ from those made independently, highlighting the powerful role of interpersonal relationships in shaping behaviour within groups.

Social psychological research highlights two primary mechanisms through which social groups impact their members: informational influence through cognitive processes and normative influence through social processes (Crano and Hannula-Bral 1994). For cognitive questions, groups focus on finding the “correct” answer, where informational influence prevails. For judgmental questions, groups define what is “moral” or “preferred”, where normative influence dominates (Kaplan and Miller 1987). This study focuses on the issue of individual preferences, thus falling under the category of judgmental issues. Its mechanism operates through normative influence.

Normative influence stems from individuals’ desire to meet others’ expectations, driving them to adhere to explicit or implicit group norms. This enhances cohesion and fosters a sense of belonging. In contrast, informational influence relies on the perceived truthfulness and quality of information. It is effective in clear, data-rich contexts but lacking the emotional connectivity that makes normative influence more stable for promoting group cohesion (Kaplan and Miller 1987). In uncertain or information-scarce situations, normative influence helps members adapt quickly, reducing conflicts by promoting adherence to group norms. Informational influence, on the other hand, encourages fact-based dissent, which can lead to disagreements and even opposition during intense conflicts (Kaplan and Miller 1987; Rugs and Kaplan 1993; Yanovitzky and Rimal 2006). Normative influence typically drives long-term group conformity as members internalize and integrate group norms into their personal values. In contrast, informational influence has more immediate and short-term effects, with individuals adjusting behaviours based on new information but showing lower persistence in group alignment over time (McLeod 2023).

This study aims to explore how a group environment, leveraging normative influence, affects individual decision-making bias incurred by beauty.

Experimental design

This study aims to identify strategies to reduce inequitable resource allocation caused by facial attractiveness in organisational management. Individuals with higher facial attractiveness often experience a resource accumulation spiral, gaining more workplace assets. Thus, the study focuses on resource distribution and fairness perception.

Trust games are commonly used to induce reciprocal behaviours among participants (Wilson and Eckel 2006). We designed a modified trust game which involves three stages: investment, anticipation, and outcome evaluation. At investment stage, participants as trustors make investment choices. At anticipation stage, the summary of investment is presented and participants wait for the return from the trustee. At outcome evaluation stage, returns from trustee are presented, and participants need to express their feelings (satisfied or not) through a button press.

The experimental design includes two different decision-making environments: individual decision-making and group decision-making. In individual decision-making setting, participants, acting as trustors, make their own decisions alone. In contrast, group decision-making setting involves three participants forming a group and acting as a collective trustor, where the investment for each trial is the average of the amount invested by each group member. This study compares the beauty premium effect during both investment and outcome evaluation stages under individual and group settings, to examine if group environment can influence the beauty premium effect.

In groups, individuals can adjust their initial beliefs based on the social influence of other members’ opinions. Drawing from previous study (Ma et al. 2022) that found a linear relationship between attractiveness and individual fairness preference, this study proposed that attractiveness affects participants’ perceptions of unfairness, influencing their satisfaction judgment. Specifically, in the context of group decision-making, it is anticipated that the influence of facial attractiveness on individual preference will weaken, consequently lessening decision biases associated with attractiveness. In this study, we primarily focus on two decision-making stages of the trust game: the investment stage (where participants choose how much to invest to the trustee) and the outcome evaluation stage (where participants express their feelings towards the amount returned by the trustee). To enhance the validity of our experimental manipulation, we conducted informal verbal inquiries (see Supplementary Information) at two key stages: pre-experiment and post-experiment.

In group context, individual behaviour is influenced by group normative influence, which may reduce the emphasis on attractiveness in their investment decisions. Therefore, this study posits the following hypothesis:

H1a: The beauty premium effect in group decision-making environment is smaller than that in individual decision-making environment, i.e., the coefficient of the influence of attractiveness on investment amount is smaller in group decision-making context.

To further understand the role of decision-making context, it is important to consider how the effect of decision-making environment operates across varying levels of facial attractiveness. Previous studies have commonly categorized attractiveness into three levels, high, medium, and low, reflecting gradients in social perception (Zhou et al. 2021). These levels represent a spectrum of attractiveness effects, with highly attractive individuals typically experiencing the strongest beauty premium, while medium and low levels of attractiveness may elicit less yet still significant biases. Understanding whether the effect of decision-making context operates uniformly across these levels can provide deeper insights into how social contexts influence the relationship between facial attractiveness and decision-making outcomes. Therefore, hypothesis 1b can be posited as follow:

H1b: The effect of decision-making context is consistent across high, medium, and low levels of facial attractiveness.

Based on Hypothesis 1a and 1b, we are supposed to further identify the potential reason why the beauty premium effect is mitigated in group decision-making environment.

Research on group dynamics suggests that decision-making contexts, particularly group settings, often reshape how individual biases manifest (Kerr and Tindale 2004; Mannion and Thompson 2014). Specifically, group contexts are characterized by normative influences that encourage alignment with collective values and reduce reliance on personal preferences (Garcia et al. 2021; Kaplan and Miller 1987). In other words, the normative influence in group context serves as a form of social pressure that negatively moderates preference for attractiveness, resulting in a significant reduction of the beauty premium effect on investment behaviour.

In this way, we propose Hypothesis 2 that this weakening effect works by moderating individuals’ intrinsic preference for attractiveness, rather than directly affecting the investment decision itself.

H2: The decision-making context moderates the relationship between attractiveness and investment behaviour.

This hypothesis underscores the role of normative influence in group setting, which weakens the beauty premium by encouraging alignment with collective value over individual preference. To gain deeper insights into the moderating mechanism of group normative influence, it is essential to investigate the dynamic interplay between group norms and individual preference (see Supplementary Information).

The group normative influence triggered in this experimental environment can be specifically defined as the difference between an individual’s investment amount and group’s average investment amount, also known as the group deviation. In other words, the average investment amount of group members can be considered as the group norm, and the farther an individual’s investment amount deviates from this group norm, the greater the perceived group pressure might be. Since the investment amount is decided by individuals at the first stage of each trial (the investment stage) without knowing current trial’s group average investment amount (shown at second stage), we use the group deviation in preceding trial as the group normative influence for the current trial. We believe that this group influence can affect an individual’s investment decision in group decision-making context. Therefore, the following hypothesis is proposed:

H3: Group normative influence (quantified by group deviation in the preceding trial) can affect an individual’s investment amount in group decision-making environment.

Besides making decisions at investment stage, participants are also required to assess their feeling towards the return amount at outcome evaluation stage. Based on the results of previous study (Ma et al. 2022), we assume that participants’ satisfaction judgments during outcome evaluation stage will also be influenced by trustee’s attractiveness, exhibiting the beauty premium. In group decision-making, an individual’s satisfaction evaluation is shaped not only by factors like physical attractiveness and return rate, but also by the group. Group influence may dilute the effect of these factors, meaning satisfaction evaluations are a product of group influence and other factors.

Based on these reasons, the following hypotheses could be proposed:

H4: The decision-making environment can affect individuals’ satisfaction evaluations.

H4a: The group decision-making environment reduces the influence of physical attractiveness on satisfaction evaluation.

Social influence often manifests through the interplay of group norms, shared accountability, and collective expectations, which together drive individuals to adopt more cautious approaches in their decision-making processes. This phenomenon is well-explained by group dynamics theory (Lewin 1947), which highlights how the structure and interactions within a group create a framework that guides individual actions, promoting alignment with group objectives and careful deliberation to maintain social cohesion. Furthermore, social comparison theory (Festinger 1957) highlights how individuals in group context compare their decisions and evaluations to those of others to ensure alignment and avoid deviations that might lead to social disapproval. These mechanisms collectively foster alignment with group norms, encourage careful evaluations to avoid discord, creating an environment that promotes caution in group context. Based on these reasons, the following hypothesis could be proposed:

H4b: Individuals’ satisfaction evaluation process is more cautious in group decision-making environment.

Materials and methods

Participants

This study recruited 286 participants (95 males) via advertisement posted on the online bulletin board system of Zhejiang University and communication tools like WeChat. All participants were Chinese and native Mandarin speakers with normal or corrected-to-normal vision. This study was approved by the Ethics Committee of the NeuroManagement Laboratory at Zhejiang University. Informed consent was obtained from all participants. Participants were notified that money allocation within the game would be realized. After the experiment, participants were given a basic compensation of CNY 30, along with an extra bonus of a trial randomly selected from where participants were satisfied (in individual decision-making setting) or the majority of the group members were satisfied (in group decision-making setting).

Material

For the selection of photo stimuli, we created a face image database to ensure the scientific validity and feasibility of this study. Specific steps for creating the facial image library are shown in Supplementary Information.

Procedure

Individual decision-making version of the trust game

This study employed a modified version of the trust game to explore the beauty premium effect in both individual and group decision-making environments.

At the start of each trial, participants were shown a photo of the trustee. The photos included 120 real-person images and three black-and-white silhouettes (one distinctly male, one distinctly female, and one gender-neutral) used in the first three trials as practice. All real-person photos were randomly distributed across the 120 trials.

Figure 1 shows that participants play as trustors, with trustee’s actions determined by the experimental program. At the stage of Investment, participants start with CNY 10 and decide how much to allocate to the trustee. The amount then triples as earnings, from which the trustee returns a part to the participant and keeps the rest. Participants’ total earnings in each trial are CNY 10 minus the invested amount and plus the trustee’s return. At the stage of Evaluation, participants express their feeling, i.e., satisfied or unsatisfied, towards the returns by clicking the corresponding button. At the end of the experiment, a trial with which participants or the majority of participants were “satisfied” is chosen randomly, determining their extra compensation for participation.

Fig. 1: Experimental procedure for the online individual trust game.
figure 1

In each trial, participants are initially presented with the trustee’s photo, informed they have an initial CNY 10, and required to choose their investment amount for the trustee using a slider (0–10). Subsequently, the program display the current investment and its earnings. Lastly, participants see the amount returned by the trustee and provide feedback on their satisfaction by pressing a button.

Full size image

To determine the return amount from trustee to participant, we drew insights from previous Trust Game (TG) experiments (see Supplementary Information).

Group decision-making version of the trust game

The group version of the trust game closely resembles the individual version, with the primary difference being that participants form small groups, creating a collective decision-making environment (see Fig. 2). Initially, participants enter their personal details and wait for other group members to join an online session, where they are organised into groups of three for the trust game. Name codes of all three members remain visible on screen during the whole game. In each trial, the group views the same trustee, and each member decides their own investment amount privately. The average amount of three members is deemed as the final investment transferred to the trustee. The screen then shows the average investment amount and the investment earning, which is always three times of the investment. Finally, trustee’s return is presented and participants need to express their own feeling by pressing corresponding buttons. This experimental design is expected to foster the sense of group identity among participants. During the experiment, participants share both risks and rewards with their fellow group members.

Fig. 2: Procedure of the online group decision-making version of the trust game.
figure 2

Group version of the trust game closely resembles the individual version, with the key distinctions being a preliminary group formation stage and the determination of investment amount. Before the experiment commences, participants log into the system and submit details like their name initials in Pinyin. The experiment begins once all group members have logged in, with the system indicating their group number. Throughout the experiment, the initials of all group members are visible. For each trial, trustee’s initial capital is the average of the investment made by all group members. These features collectively establish a group decision-making context.

Full size image

More details of the experimental programs were shown in Supplementary Information.

Data analysis

This study collected valid data from 172 participants in the individual version. We then proceeded to filter this data, excluding 6 participants who met the specified exclusion criteria (see Supplementary Information).

For the group version, we gathered data from 38 groups, comprising 114 participants in total. Utilizing the same exclusion criteria as in individual version, no participant was excluded.

ANOVA analysis

In order to analyse the effect of different levels of attractiveness on behaviour, we divided the trustees’ photos into three types according to the attractiveness rating. To ensure an equivalent number of photos in each level while maintaining clear distinctions between categories, the following classification criteria were applied: Low (rating ≤ 44, total of 40 photos, Mean = 35.91 ± 3.87), Middle (44 < rating < 56, total of 42 photos, Mean = 50.10 ± 1.59), and High (rating ≥ 56, total of 41 photos, Mean = 67.41 ± 5.57). Then, a 2 Environment (individual, group) × 3 BeautyLevel (low, middle, high) ANOVA test was applied on investment amount, satisfaction level, and response time for satisfaction evaluation respectively. Environment is a between subject factor, and BeautyLevel is a within subject factor.

Regression analysis

At investment stage, the magnitude of beauty premium is measured as the slope of a linear regression equation, using trustee’s attractiveness as independent variable and participant’s investment amount as dependent variable. In the experiment setting, investment amount varies from 0 to 10, and attractiveness score varies from 0 to 100. To normalize the regression coefficients, attractiveness score was scaled down to a 10-point system.

Linear regression analyses were separately conducted on the investment amount in individual and group decision-making environment (independent variable: attractiveness; dependent variable: investment amount). Differences in regression coefficient between two environments were then compared using SUEST method (Lian and Liao 2017).

This study aimed to examine how individual and group decision-making contexts affect the influence of facial attractiveness by analysing participants’ satisfaction after receiving return outcome. Given that satisfaction was measured as a binary outcome (satisfied or not satisfied), this study used logistic regression to analyse satisfaction as dependent variable.

Moderation model for examining the impact of decision environment

In this study, we hypothesize that group decision-making environment attenuates the beauty premium effect. Specifically, we propose that individuals in group decision-making are more influenced by social preferences, such as group identification and conformity, which reduces their focus on attractiveness. This hypothesis is based on motivation crowding-out theory (Frey and Jegen 2001; Wollbrant et al. 2022), which posits that external group pressures or social norms suppress individuals’ intrinsic preferences, causing them to prioritize group values and social goals over personal preferences in group decision environment. Research on motivation crowding-out effect often employs moderation models to analyse how external factors (e.g., group identification) inhibit individual preferences (Xu et al. 2022). Thus, we apply a moderation model in this study (see details in Supplementary Information), using decision-making context (individual or group) as a moderating variable to test the influence of group environment on the beauty premium.

Additionally, to capture the dynamic characteristics of individual decision-making more comprehensively, we incorporate a lagged term (investment_lag) into the model, representing the investment amount of the previous trial. The mixed-effect model is defined as follows:

$$\begin{array}{l}{Investment}=\alpha +{\beta }_{1}\times {Attractiveness}+{\beta }_{2}\times {Environment}+{\beta }_{3}\times {Attractiveness}\\\qquad\qquad\qquad \times {Environment}+{\beta }_{4}\times {Investment}{\_}{lag}+{u}_{j}+\varepsilon \end{array}$$

Here, Environment is a dummy variable (0 = individual, 1 = group). If the interaction coefficient β3 is negative and significant, it indicates that the group context has a negative moderating effect on the beauty premium, suppressing individual preferences for attractiveness.

A Bootstrap sampling method was employed to mitigate the impact of unequal sample sizes (see details in Supplementary Information).

Group-informed dynamic investment model (GDIM)

If prior analyses reveal that group decision-making context markedly mitigates beauty premium effect at investment stage, it becomes crucial to delve into the factors driving this decrease. Unlike simple decision inertia (Jung et al. 2019), individual behaviour in group decision-making context is more adaptive and interactive (see details in Supplementary Information). The investment amounts from the previous trial serve as a dynamic starting point, adjusted by group normative opinions to optimize investment behaviour. To dynamically examine the beauty premium, we devised a decision model equation that uses information from previous trial to predict each trial’s investment amount:

$$\begin{array}{l}{Investment}=m\times {Previous\; Investment}+a\times {Group\; Deviation}\\\qquad\qquad\qquad +b\times {Attractiveness}+c\times {Previous\; return\; rate}\end{array}$$

Here, Group deviation is denoted as the value of participant’s own investment amount in previous trial minus the mean investment amount of group members in that trial: group deviation = own investment - mean investment. When the coefficient of group_deviation is positive, it suggests that participants tend to deviate from the group average. Conversely, a negative coefficient indicates a tendency for participants to align more closely with the group average.

Since the range of return rate is 0 to 1, in order to guarantee all factors with the same scale, we re-scale the values of all the other variables proportionally so that they fall within the range of 0 to 1.

To accurately characterize this dynamic updating process, we introduced a weighting coefficient, m, into our model to quantify the degree to which individuals rely on their previous investment amounts. By estimating coefficient a, the model reveals the role of group normative influence in shaping individual decision adjustments. Attractiveness refers to the attractiveness rating of the trustee in current trial. Previous return rate refers to the return rate given by the trustee to the trustor in previous trial (Return rate = Returned amount ÷ Investment earnings).

Considering that this model emphasizes the critical role of group normative influence in shaping individual investment decisions dynamically, we named it as the group-informed dynamic investment model (GDIM). Due to the inefficiency of iterating through all possible combinations of influence factor weights (Osman et al. 2018), this study employed a classic heuristic optimisation method, the genetic algorithm, to obtain the optimal solution (optimal parameter combination) and to effectively capture this dynamic updating feature (see details in Supplementary Information).

Drift diffusion model

To further examine the dynamic cognitive process of satisfaction judgement in two different decision-making environments, we employed the drift diffusion model (DDM) to analyse cognitive processing of participants during satisfaction judgment (see Fig. 3). In this study, participants’ satisfaction judgments are based on gradual information accumulation. The drift rate (v) reflects the rate of this accumulation, and a decision is made when the accumulated information reaches a boundary. The decision boundary (a) includes two thresholds: “satisfied” and “unsatisfied.” The decision bias parameter (z) indicates participants’ initial preference for one option, and the non-decision time (t) represents time spent on visual stimulus processing and button pressing, excluding evidence accumulation and decision-making.

Fig. 3: Schematic diagram of drift diffusion model.
figure 3

The DDM formalizes the decision-making process as evidence accumulation toward one of two boundaries. Once the evidence accumulation reaches a boundary, the individual makes the corresponding choice. According to the DDM, the variation in an individual’s reaction time (RT) for decisions is primarily due to the different amounts of time required for evidence to accumulate to the response boundary.

Full size image

We used the hierarchical DDM (hDDM) method (O’Callaghan et al. 2017) and toolbox to implement DDM analysis (http://ski.clps.brown.edu/hddm_docs/).

Since the outcome evaluation stage is identical in both individual and group decision-making settings (each asks, “The other party returns Y yuan, are you satisfied?”) and the method of responding via button press remains the same, the “non-decision time” can be assumed to be consistent across both scenarios. Thus, in current modelling, we assume the non-decision time (t) for trust game is the same in both individual and group contexts.

For other three parameters in drift diffusion model (a, v, z), depending on whether each parameter varies with different contexts (individual decision-making, group decision-making), eight different models can be established. With three parameters, each having two possibilities (same or different values in the two decision-making environments), there are 2 × 2 × 2 = 8 models. We estimate parameters for each of these eight models, then compare their fitness indices to identify the best model and obtain corresponding parameters for subsequent analysis. The specific model fitting process of DDM is shown in Supplementary Information.

Results

ANOVA of environment and BeautyLevel

For investment amount, the main effect of Environment is significant, F(1, 278) = 9.335, p < 0.01, the investment amount in individual environment is significantly higher than in group environment. The main effect of BeautyLevel is significant, F(2, 556) = 346.972, p < 0.001, post hoc pairwise comparison reveals that high level is significantly higher than middle level and both of them are significantly higher than low level. The interaction between Environment and BeautyLevel is not significant, F(2, 556) = 1.508, p > 0.05, which supports the hypothesis H1b.

For satisfaction, the main effect of Environment is significant, F(1, 278) = 77.944, p < 0.001, the satisfaction in individual environment is significantly higher than group environment. The main effect of BeautyLevel is significant, F(2, 556) = 64.044, p < 0.001, post hoc pairwise comparison reveals that high level is significantly higher than middle level and both of them are significantly higher than low level. The interaction between Environment and BeautyLevel is significant, F(2, 556) = 4.947, p < 0.01. Simple effect analysis reveals that individual environment is significantly higher than group environment at every level of beauty (ps < 0.001). The results support the hypothesis H4.

For response time of satisfaction, the main effect of Environment is not significant, F(1, 278) = 0.056, p > 0.05. The main effect of BeautyLevel is significant, F(2, 556) = 5.069, p < 0.01, post hoc pairwise comparison reveals that the middle level is significantly higher than low level and there are no other significant differences. The interaction between Environment and BeautyLevel is significant, F(2, 556) = 4.459, p < 0.05. Simple effect analysis reveals that individual environment is significantly higher than group environment at low level of beauty, but no significant differences were observed at middle or high level (see Table 1, Fig. 4).

Table 1 Results of ANOVA on behaviour results.
Full size table
Fig. 4: ANOVA results of environment and BeautyLevel on behaviour results.
figure 4

The figures represent the effects of two factors on investment amount (upper left), satisfaction (upper right) and response time for satisfaction (down left). The x-axis indicates the categories of decision environment or beauty. The y-axis represents the value of the corresponding dependent variable.

Full size image

Regression results of facial attractiveness and investment amount

Regression equation of the beauty premium effect in individual environment is as follows (see Table 2):

$${\rm{Investment}}=-1.931+0.535\times {\rm{Beauty}}$$
Table 2 Coefficients of beauty premium regression equation under individual and group decision environments.
Full size table

The validity test result of regression equation is: F(1, 118) = 870.40, p < 0.001, R2 = 0.838, Root_MSE = 0.322.

The regression equation of beauty premium effect under group environment is (see Table 2):

$${\rm{Investment}}=-1.542+0.492\times {\rm{Beauty}}$$

The validity test result of regression equation is: F(1, 118) = 824.12, p < 0.001, R2 = 0.829, Root_MSE = 0.307.

The coefficient is significantly lower in group decision-making environment than in individual decision-making environment (χ2 = 6.17, p = 0.013), which supports the H1a.

Logistic regression results of attractiveness on satisfaction

This study proposes a logistic regression model to quantify the impact of attractiveness and return rate on satisfaction, including an interaction term between attractiveness and return rate to account for potential moderating effects (see Table 3).

Table 3 Logistic regression coefficients in individual and group decision-making environments.
Full size table

Logistic regression models in individual and group environments respectively are:

$${{\rm{Y}}}_{{\rm{Individual}}}={{\rm{b}}}_{1}\times {\rm{Return\_rate}}+{{\rm{b}}}_{2}\times {\rm{Beauty}}+{{\rm{b}}}_{3}\times {\rm{Return\_rate}}\times {\rm{Beauty}}+{{\rm{C}}}_{1}$$
$${{\rm{Y}}}_{{\rm{Group}}}={{\rm{m}}}_{1}\times {\rm{Return\_rate}}+{{\rm{m}}}_{2}\times {\rm{Beauty}}+{{\rm{m}}}_{3}\times {\rm{Return\_rate}}\times {\rm{Beauty}}+{{\rm{C}}}_{2}$$

Model fitting results in individual environment show that:

$${{\rm{b}}}_{1}=2.05,{{\rm{b}}}_{2}=-3.94,{{\rm{b}}}_{3}=9.82,{{\rm{C}}}_{1}=-0.3$$

The logistic regression model under individual environment can be expressed as:

$${{\rm{Y}}}_{{\rm{Individual}}}=2.05\times {{\rm{Return}}}_{{\rm{rate}}}-3.94\times {\rm{Beauty}}+9.82\times {{\rm{Return}}}_{{\rm{rate}}}\times {\rm{Beauty}}-0.3$$

Validity test results of logistic regression equation are as follows: χ2(3) = 1719.77, p < 0.001. The accuracy of the model’s judgment on the dependent variable (satisfaction/dissatisfaction) is 71.43%.

Model fitting results in group environment (see Table 3) show that:

$${{\rm{m}}}_{1}=5.57,{{\rm{m}}}_{2}=-2.35,{{\rm{m}}}_{3}=7.82,{\rm{C}}2=-3.72$$

The logistic regression model under group environment can be expressed as:

$${{\rm{Y}}}_{{\rm{Group}}}=5.57\times {{\rm{Return}}}_{{\rm{rate}}}-2.35\times {\rm{Beauty}}+7.82\times {{\rm{Return}}}_{{\rm{rate}}}\times {\rm{Beauty}}-3.72$$

Validity test results of logistic regression equation are as follows: χ2(3) = 3194.89, p < 0.001. The accuracy of the model’s judgment on the dependent variable (satisfaction/dissatisfaction) is 78.06%.

To compare the impact of various variables on satisfaction evaluations across individual and group decision-making contexts, a Bootstrap method (500 permutations) was used to analyze the significance of differences in logistic regression coefficients. Results show significant differences in coefficients for attractiveness (p < 0.05) and return rate (p < 0.001), but not for the interaction term of attractiveness × return rate (p > 0.05). The difference in the constant term is also significant (p < 0.001) (see Table 4).

Table 4 Comparison of coefficient differences of logistic regression between environments.
Full size table

Analysis of coefficient values, with b2 at −3.94 and m2 at −2.35, illustrates the negative effect of attractiveness on satisfaction in both individual and group decision-making contexts. This negative correlation suggests that higher attractiveness increases the likelihood of dissatisfaction, potentially due to elevated expectations. Participants might invest more in attractive individuals, expecting higher returns that often don’t materialize, leading to greater dissatisfaction. In contrast, lower expectations for less attractive individuals can result in unexpectedly positive outcomes. The smaller absolute value of m2 compared to b2 supports H4a, indicating that group decision-making environments mitigate the influence of attractiveness on satisfaction evaluation.

Moderation model results

Table 5 shows that the original mixed effect model results and those after bootstrap processing are highly consistent, confirming the robustness of the analysis and the reliability of the conclusions. The model indicates a significant negative moderating effect of the group decision-making context on the beauty premium. Specifically, attractiveness has a positive and significant effect on investment amount (coefficient = 0.547, p < 0.001), but the group context weakens this effect (interaction term Beauty × Environment: coefficient = −0.052, p = 0.021), supporting H2. This suggests that group norms suppress individual preferences for attractiveness.

Table 5 Results of mixed effect model and bootstrap summary.
Full size table

Furthermore, the significant coefficient of the lagged term investment_lag (coefficient = 0.352, p < 0.001) shows that individuals dynamically adjust their investment behaviour based on feedback from previous trials.

Results of the GDIM model

Model fitting was performed for the decision model equation that calculates each round’s investment amount based on information from previous rounds. The calculation equation is:

$${Investment}=m\times {Previous\; Investment}+a\times G{rou}p{Deviation}+b\times {Attractiveness}+c\times {Previous\; Return\; Rate}$$

The model fits well for each participant (mean MSE = 0.0087, see Table 6), with mean value of coefficients m = 0.1983, a = −0.8462, b = 0.5760, and c = 0.0882. This shows that group influence affects investment decisions, supporting the hypothesis that individuals adjust their investments dynamically based on prior feedback within group context. The significant Group Deviation term further supports H3, indicating that individuals adapt to group norms over trials to minimize deviation. The genetic algorithm performance results confirm the model’s high fitting performance (see Table 6, Supplementary Information).

Table 6 Summary of model fitting results from each participant in group environment.
Full size table

Results of drift diffusion model analysis

Drift diffusion model was used to analyze participants’ satisfaction evaluations and response time during the outcome evaluation stage. Results show that Model 1 has the smallest Deviance Information Criterion (DIC), indicating it is the best-fitting and most optimal model for the data. In Model 1, parameters a, v, and z vary across decision-making environments (individual vs. group), requiring separate estimation for each context (see Tables 7, 8).

Table 7 Model setup and fitting index.
Full size table
Table 8 Optimal model parameters.
Full size table

Table 8 shows the parameter estimates in the optimal model (Model 1). Significant tests reveal that in group environment, the decision boundary a (M = 2.4265 ± 0.0118) is larger than in individual environment (M = 1.8994 ± 0.0085, p < 0.001), supporting H4b. The drift rate v is higher in group environment (M = 0.6380 ± 0.0092) than in individual environment (M = 0.3991 ± 0.0093, p < 0.001), and the starting point z is lower in the group environment (M = 0.4827 ± 0.0025) than in the individual environment (M = 0.5658 ± 0.0025, p < 0.001). This suggests that in individual decisions, participants’ prior preference leans more towards the “satisfied” option. The Gelman-Rubin diagnostic was used to assess model parameter convergence, and results show high degree of consistency in sampling results across chains (see Supplementary Information).

Discussion

This study examined the impact of group decision-making on the beauty premium, finding that group decisions generally mitigate appearance-related biases. This attenuation effect underscores the potential of group norms to reduce biases observed in individual context. By addressing the core research question, the study enhances our understanding of how group dynamics interact with physical attractiveness to shape decision outcomes.

Behavioural differences between decision environments

Regression analyses reveal that the impact of attractiveness is weaker in group decision-making than in individual setting. Group decision-making mitigates the beauty premium across all levels of attractiveness (high, medium, and low). This suggests that group dynamics, driven by collective deliberation and normative influence, promote fairer evaluations and reduce reliance on superficial attributes like physical appearance. These findings align with prior research indicating that group contexts prioritise shared goals over individual preferences (Tan, 2021). Using the drift diffusion model, this study examines cognitive processing differences between individual and group decision-making by analysing satisfaction evaluations and response time. Key findings include: (1) A higher decision boundary in group setting indicates a more cautious approach, driven by the strong link between individual and group rewards. Individual decisions directly affect the group’s success. This fosters a heightened sense of group identity and responsibility, prompting individuals to deliberate more thoroughly and align their decisions with collective goals. (2) An increased drift rate in group decision-making suggests quicker information processing, likely due to shared goals and social norms that filter irrelevant preferences. (3) In individual decision-making, a higher starting point reflects a predisposition towards satisfaction, driven by personal preferences. In contrast, group settings shift this starting point towards neutrality, reducing personal control and potentially increasing dissatisfaction due to shared decision-making.

Overall, group decision-making not only shifts social preferences but also alters cognitive processing, prioritising fairness and collective goals over individual biases.

The regulation mechanism of group decision-making

Based on our findings, we propose the following regulation mechanism of group decision-making on the beauty premium: Group formation assigns individuals a membership, fostering a sense of social identification. This identification motivates individuals to conform to group norms, prompting them to dynamically adjust their decisions across trials. As a result, their intrinsic preferences for facial attractiveness are crowded out by social influence, thereby mitigating the beauty premium effect. The details of this regulation mechanism are discussed in the following paragraphs.

Social identification

The differences between individual and group decision-making can be explained from the perspective of collaboration theory and social identification theory (Cialdini and Goldstein 2004; Kelman 1958). Social identification theory distinguishes between personal identity (based on individual attributes) and social identity (based on group membership) (Turner et al. 1979). These identities influence decision-making differently, often leading to distinct behaviours in individual versus group contexts. This explains why the “beauty premium” is lower in group decision-making: when individuals see themselves as group members, they struggle to maintain a distinct individual identity (Frey and Tropp 2006).

Due to the effects of reputation and non-anonymity, individuals desired to gain or to preserve a good social reputation as a cooperative and trustworthy person on decision-making in group context (Wagner et al. 2022). Besides, people also intend to make group-serving attributions for the groups they belong to (Baumeister and Leary 1995). Tajfel and Turner (2019) also proposed that people have the motivation to maintain a positive image of one’s groups. When individuals are brought into a group, group members exert more social influence on each other and they will elicit a stronger cooperative tendency as well as imply less bias.

Dynamic decision adjustment

The GDIM model fitting results reveal how individuals in group setting are influenced by their prior decisions. By incorporating a lagged term, the study shows that individuals adjust their decisions based on feedback from previous trials, highlighting the cumulative and indirect effects of group norms. The significant group deviation term indicates that normative influence is dynamic, with individuals progressively adapting to group norms across trials to minimize deviation from the group’s average investment. This reflects a feedback-driven adaptation process where individuals continuously align with group norms.

The dynamic adjustment process can be explained by collaboration and social influence theories (Cialdini and Goldstein 2004; Kelman 1958). Collaboration theory highlights mutual adaptation and coordination in group decision-making, with members adjusting strategies based on repeated interactions and observed group outcomes. Social influence theory emphasizes how social comparison and information sharing shape individual decisions. In group setting, members’ decisions are influenced by others’ behaviours and group outcomes, leading to gradual convergence. This reflects the deep and often unconscious impact of social dynamics on individual judgments.

Motivation crowding-out effect

These findings revealed a significant crowding-out effect, where social preferences within group setting suppress individual preferences for attractiveness, shifting the primary drivers of investment decisions from personal biases to social goals. This transition reflects how perceived social norms within the group encourage members to work collectively. Group members are guided less by individual preferences and more by collective interests, effectively mitigating the impact of attractiveness bias in decision-making. This study suggests that the external motivations formed by group members have certain impacts (crowding-out effect) on their intrinsic motivations, thereby reducing the influence of intrinsic motivations on individual decisions, leading to a reduction in the beauty premium effect. This regulation mechanism offers fresh empirical insights into how decision environments shape individual behavioural preferences dynamically.

Theoretical integration and implications

This study advances our theoretical understanding of group decision-making by emphasizing its dynamic, iterative nature, elucidating the underlying mechanisms of group influence, innovating experimental methodologies, and clarifying the cumulative and indirect effects of group norms on individual preferences.

Methods for controlling bias

Previous research has examined factors like human capital and financial resources to mitigate biases such as the beauty premium effect (Gu and Ji 2019). But these individual-focused approaches overlook the interaction between individuals and their environments, which is crucial for understanding decision biases (Fiedler and Juslin 2005). Our research advances this understanding by introducing group normative influence and examining how individuals dynamically adjust and optimize their decisions under this social pressure. This aligns with prior findings that individuals in group settings often experience social pressure, leading to decisions that differ from those made individually (Cialdini and Goldstein 2004). This perspective resonates with Flache et al. (2017), who emphasize that individuals often adjust their perspectives and behaviours to align with those around them. Our model’s results further support the principles of normative influence, where individuals conform to group norms to meet others’ expectations. Even subtle normative influences, as examined in our study, can produce significant outcomes.

This study shows that group context shapes behaviour and preferences indirectly through normative influence and feedback, rather than altering individual preferences directly. This finding supports the theory of group normative influence, expands traditional frameworks by highlighting the indirect, cumulative impact of social factors over time, and underscores the role of feedback in driving individual adaptation. The results reveal a deeper understanding of how individuals change from personal biases to collective alignment through an iterative feedback loop, emphasizing the nuanced, indirect influence of group dynamics on personal preferences.

Theories on bias mitigation emphasize that interventions should also be examined for its scope (whether they address various biases broadly or target specific subsets) (Korteling et al. 2021; Wongvorachan et al. 2024). This study shows that group decision-making acts as a broad bias-reduction mechanism by suppressing appearance-related biases through collective norms that de-emphasize superficial preferences. These findings extend theories on group dynamics and collective rationality (Forsyth 2014) by highlighting the role of group norms in reducing biases.

Group decision-making environment

In this study, although participants had no direct communications, they were still affected by implicit normative influence. This influence, rooted in social identity and perceived group norms, subtly guided individuals to align their behaviours and judgments with the group. Over time, it fostered group conformity, with individuals adjusting their responses to match implicit group standards. This highlights how social identification can shape behaviour without overt interaction. As Kahneman (2013) noted, collecting independent judgments confidentially, rather than through open discussion, avoids such social influences skewing individual opinions.

Unlike traditional studies that focus on collective group dynamics and outcomes, our research emphasizes individual decision-making within groups, aligning more closely with real-life complexities. By focusing on individual performance, we reveal how group dynamics impact personal decision-making behaviour.

Dynamic decision modelling

This study demonstrates that group decision-making is not a static, one-off event but rather an evolving process where members progressively adjust and coordinate strategies through repeated interactions. Our model incorporates factors such as individual prior decisions, group deviance, attractiveness, and return rates to capture the dynamic investment decision-making process in group environment. These factors collectively influence whether individuals repeat prior decisions or adjust their behaviour in response to new information and group norms. Such results highlight the trade-offs individuals make between historical decisions and emerging environmental cues.

Our findings contribute to management theory by emphasizing that group decisions are shaped by continuous adaptation and alignment among members, expanding our understanding of decision-making as an iterative, interaction-based process.

Complexity of human behaviour

Group decision-making is often more rational and less biased than individual decision-making (Jans et al. 2011; Postmes et al. 2005; Sassenberg et al. 2014). Our study examines how normative social influence in group setting can reduce certain biases, such as the beauty premium. However, this finding may not apply universally, as the effectiveness of group decision-making in mitigating biases depends on the nature of the bias, the group’s openness and the group’s ability to recognize it (Kerr and Tindale 2004). For example, biases like framing effect or impulsive buying tendencies, which are less recognized, may not be effectively curbed in groups and can even be exacerbated through mutual reinforcement and emotional contagion (Bang and Frith 2017; Kertzer et al. 2022). In contrast, biases like the beauty premium, which are more explicitly recognized, can be mitigated through group influence and social validation.

Practical implications

Beyond theoretical contributions, the findings of this study hold practical significance for various applied fields, including organisational decision-making, policy design, and social psychology. By demonstrating that group decision-making environment can broadly suppress appearance-related biases, the study highlights the potential for group-based evaluative processes to promote fairness and equity. Previous studies indicate that genetic factors account for only about 20% of the variation in trusting behaviours, while environmental factors explain the remaining 80% (Ahern et al. 2014; Cesarini et al. 2008). By modifying environmental cues (creating a group decision-making context) and optimising information sampling methods (gathering insights from a broader range of team members), decision-making within management scenarios can be significantly improved. The GDIM model and its key parameter estimates provide practical value for designing behavioural interventions, especially in optimizing investment decisions and enhancing individual performance in group context. For instance, in organisations, group deliberation in hiring or promotion can reduce superficial biases like attractiveness, promoting merit-based outcomes.

The current approach facilitates the development of adaptable decision-support systems and group collaboration tools. By examining individual decision patterns within groups, it identifies key cognitive influences and focal points, enabling the design of targeted systems that leverage individual strengths and mitigate cognitive limitations in group settings. This enhances both individual and group decision quality.

Conclusion

This research validates that the beauty premium can be mitigated in group context, demonstrating the power of social forces over instinct. Through the experiments, we examined how social factors influence individual preferences and the role of group dynamics in human decision-making. Using mixed-effect modelling with a lagged term, we captured the dynamic impact of group context on the beauty premium, showing how individuals adapt to group norms over sequential decisions. The significant crowding-out effect of social preferences reduces emphasis on attractiveness, shifting investment behaviour from personal to social goals. Our findings highlight group norms’ ability to counteract biases and offer practical insights for promoting fairness through process design.

Limitations and future studies

This study provides valuable insights into how group decision-making context moderates the beauty premium. Nonetheless, it has several limitations that open avenues for further research.

A key limitation of our study is its focus on Chinese participants from a collectivist cultural background, which likely influenced group decision-making dynamics and the role of group norms in shaping individual evaluations. In collectivist cultures, individuals prioritize group harmony and consensus over personal preferences, amplifying the influence of group norms and encouraging conformity (Ding et al. 2024; Liu et al. 2019). This may reduce appearance-related biases as participants align their evaluations with group consensus. In contrast, individuals from individualistic cultures, which emphasize autonomy and self-expression, may show weaker group influence and greater independence in decision-making, potentially sustaining biases (Ding et al. 2024). Future research should explore these dynamics in individualistic cultures to enhance the generalizability of our findings. Besides, other additional factors such as individual emotional characteristics, cognitive differences, communication styles may affect decision-making and management optimisation. Future studies could explore these factors further and integrate them into the analytical framework to gain a more comprehensive understanding.