Deconstructing Temporal Stages of Prosocial and Antisocial Risky Decision-making in Adolescence Free

Risk-taking is a prominent aspect of adolescent behavior. A recent neurodevelopmental model suggests that this trait could influence prosocial and antisocial decision-making, proposing a new category known as prosocial and antisocial risk-taking. The primary objective of this study was to examine the electrophysiological underpinnings of prosocial and antisocial risk-taking in adolescence, a developmental period characterized by elevated risky, prosocial, and antisocial decisions. To this end, 32 adolescents aged 13–19 years completed a modified dictator game to choose between three options, representing prosocial and antisocial risk-taking constructs and a risk-free fair one. At the behavioral level, adolescents favored antisocial risky decisions over prosocial risky ones. ERP results at the electrophysiological level in the response selection stage demonstrated that decision preceding negativity was more negative-going before making prosocial risky decisions than other decisions. During the feedback evaluation stage, feedback-related negativity was the least negative after selecting the antisocial risky option and receiving successful feedback. However, choosing the fair option and receiving neutral feedback resulted in the most negative feedback-related negativity. Moreover, P300 showed the most positive mean amplitude following the selection of the antisocial risky option and facing successful feedback, with the lowest positive amplitude observed after choosing the fair option and encountering neutral feedback. These results underscore the distinct electrophysiological underpinnings associated with prosocial and antisocial decisions involving risks.

People do not necessarily share the same interests throughout their social lives. Consequently, they sometimes have to inevitably act in favor of or against each other. However, there are instances, such as offering financial aid or providing emotional support, where people willingly extend help to or benefit others but not themselves. Such actions are referred to as prosocial behavior (Crone & Achterberg, 2022; Penner, Dovidio, Piliavin, & Schroeder, 2005; Denham, 1986). Conversely, individuals may also engage in behaviors that involve taking possession of others' resources, intentionally violating their personal and property rights or humiliating others, all categorized as antisocial behavior (Baumeister & Lobbestael, 2011; Calkins & Keane, 2009). Prosocial decisions often involve forgoing self-interest for the benefit of others, yielding intrinsic rewards (Luo, 2018; Moll et al., 2006). Conversely, in antisocial decisions, the self-oriented behavior that leads to harm to others and benefit to self is excessively rewarding (Byrd, Loeber, & Pardini, 2014; Quay, 1993).

Given their significance, extensive research in social cognition has sought to explore the causes and consequences of these behaviors. However, until recently, much of this research relied on traditional bargaining games such as dictator and ultimatum games, where participants make choices regarding prosocial, antisocial, or fair behaviors with deterministic outcomes (Miraghaie et al., 2022; Peterburs et al., 2017; Feng, Luo, & Krueger, 2015; Wu, Zhou, van Dijk, Leliveld, & Zhou, 2011). However, real-world prosocial and antisocial decisions do not always result in predetermined outcomes, as they may involve risks (Gross et al., 2021; Do, Moreira, & Telzer, 2017). In the real world, the decision to help or hurt others could happen at the expense of personal costs. For instance, intervening to help a peer facing humiliation may come with potential costs, such as risking one's reputation and experiencing humiliation. Do and colleagues (2017) coined the term “prosocial risk-taking” (PSRT) to describe such decisions, where individuals engage in behaviors that benefit others with the risk of experiencing personal cost. Meanwhile, antisocial risk-taking (ASRT) refers to decisions that involve causing harm to others while putting oneself at risk (Do et al., 2017). For example, when engaging in the act of theft, individuals expose themselves to the risk of facing legal consequences. Despite the cooccurrence of risk with both prosocial and antisocial decisions, little research has examined these decision-making processes. Thus, the current study aims to explore the neural and behavioral underpinnings of prosocial and antisocial risky decision-making.

Developmental factors play a crucial role in understanding these decisions. Due to asynchrony in developmental time courses between the affective and cognitive control brain systems, adolescence is a developmental period with a heightened level of risk-taking (Willoughby, Good, Adachi, Hamza, & Tavernier, 2013; Van Leijenhorst et al., 2010; Gardner & Steinberg, 2005) as well as prosocial (Do, McCormick, & Telzer, 2019) and antisocial (van den Bos et al., 2014; Moffitt, 1993) behavior. Critical changes in neural circuits associated with reward processing, inhibitory control system, and social processing are thought to be related to these changes in adolescence (Blakemore & Mills, 2014; Pfeifer & Peake, 2012; Van Leijenhorst et al., 2010; Casey, Getz, & Galvan, 2008; Lebel, Walker, Leemans, Phillips, & Beaulieu, 2008; Galvan et al., 2006; Ernst et al., 2005). Reward-related underpinnings like ventromedial prefrontal cortex, ventral striatum (Van Leijenhorst et al., 2010), and accumbens have been to exhibit increased activity before the sufficient development of control system, namely, lateral prefrontal cortex and dorsal anterior cerebral cortex. This is indicated as a possible reason for an inclination toward risky and rewarding choices (Van Leijenhorst et al., 2010; Casey et al., 2008; Lebel et al., 2008; Galvan et al., 2006; Ernst et al., 2005). Concerning social processing, structural and functional changes in neural substrates necessary for other's perceptions, like the medial part of the prefrontal cortex and TPJ, have been reported (Blakemore & Mills, 2014; Pfeifer & Peake, 2012). Moreover, overlapping circuits for risky and prosocial decisions were also found in adolescents (Telzer, 2016). Therefore, this critical period presents a valuable and appropriate time window for studying the compound construct of prosocial risky decisions and antisocial risky decisions along with each other.

On the basis of these neural changes, Do and colleagues (2017) suggest an increased inclination toward PSRT during adolescence. Although PSRT is not studied enough in adolescence, evidence from adulthood shows that risk-taking tendency has more weight than prosocial attitudes in delineating prosocial risky decision-making (Gross et al., 2021), mainly when the risk is low cost, and the decision is being made in the presence of peers (Liu, Xiao, Pi, Tan, & Zhan, 2023). On the other hand, antisocial risky decision-making as a single unit has not been experimentally studied in either adolescents or adults in the fields of psychology and developmental neuroscience. Yet, evidence has revealed an association between risky decision-making and antisocial behavior (Egan & Bull, 2020; Brindle, Bowles, & Freeman, 2019; Tieskens, Buil, Koot, Krabbendam, & van Lier, 2018; Aklin, Lejuez, Zvolensky, Kahler, & Gwadz, 2005).

Among various tools available, EEG offers valuable insights into the neural mechanisms behind both prosocial/antisocial and risky decision-making. Its remarkable temporal resolution enables the examination of the rapid dynamics inherent in decision-making processes. However, like other forms of decision-making, these processes can be divided into distinct stages: the response selection stage, during which individuals make a decision, and the feedback processing, where individuals face the outcome of their choices (Cui, Chen, Wang, Shum, & Chan, 2013; Ernst & Paulus, 2005; Paulus, 2005). Each of these phases occurs quickly, in a fraction of a second, making them amenable to investigation through high temporal resolution techniques such as EEG and ERPs.

During the response selection stage, the decision preceding negativity (DPN) emerges as a pivotal component, characterized by a negative and slow potential belonging to the readiness potential (RP) and stimulus preceding negativity (SPN) families, typically observed 500–1000 msec before making a decision in the frontocentral regions (Giustiniani, Gabriel, Nicolier, Monnin, & Haffen, 2015; Cui et al., 2013; Bianchin & Angrilli, 2011). Although RP is mainly related to motor preparation (Schurger, Hu, Pak, & Roskies, 2021), SPN is associated with any stimulus (Tanovic & Joormann, 2019). However, DPN is specifically defined as the negativity preceding economic decisions with outcomes rather than general events (Guo, Song, Liu, Xu, & Shen, 2019; Giustiniani et al., 2015; Bianchin & Angrilli, 2011). Despite these distinctions, all of these components—RP, SPN, and DPN—belong to the family of slow cortical potentials, characterized by their negative polarity (Böcker, Baas, Kenemans, & Verbaten, 2001). The amplitude of the DPN preceding risky decisions reflects the prospective advantageousness or disadvantageousness of the decision outcomes (Guo et al., 2019; Giustiniani et al., 2015; Cui et al., 2013). Specifically, the amplitude of this component is more negative-going before selecting disadvantageous options associated with long-term losses, contrasting the less negative amplitude preceding advantageous choices linked to long-term benefits (Guo et al., 2019; Dong, Du, & Qi, 2016; Giustiniani et al., 2015; Cui et al., 2013; Bianchin & Angrilli, 2011). Moreover, when the option to pass the disadvantageous choice is available and selected, the amplitude of DPN increases, compared with when one decides to play and accept the risk associated with the disadvantageous choice (Cui et al., 2013). This suggests that when the specified option seems risky, DPN indexes the emotional unease in deciders, leading them to refrain from choosing it.

Two well-known ERP components, the feedback-related negativity (FRN) and the P300, have been widely examined in the feedback processing phase. The FRN is mainly observed in frontocentral regions approximately 250 msec after the feedback presentation (Gehring & Willoughby, 2002; Miltner, Braun, & Coles, 1997). Previous studies have shown that the amplitude of the FRN component is associated with feedback-based learning (Cui et al., 2013; Schuermann, Kathmann, Stiglmayr, Renneberg, & Endrass, 2011), outcome unexpectedness (Schuermann et al., 2011; Christie & Tata, 2009; Holroyd & Coles, 2002), and risk processing (Yi, Mei, Li, Liu, & Zheng, 2018; Schuermann et al., 2011; Gehring & Willoughby, 2002). Notably, the amplitude of FRN exhibits sensitivity to feedback valence, being more pronounced (i.e., more negative-going) following unfavorable outcomes or losses compared with favorable feedback or gains (Peterburs, Kobza, & Bellebaum, 2016; Leng & Zhou, 2014; Cui et al., 2013; Bianchin & Angrilli, 2011; Schuermann et al., 2011; Christie & Tata, 2009; Wu & Zhou, 2009; Gehring & Willoughby, 2002). Moreover, affective and motivational factors are crucial in modulating the FRN response. The motivational significance of the outcome, reflecting activation within the dopaminergic/reward system, influences FRN amplitude (Gehring & Willoughby, 2002; Holroyd & Coles, 2002). According to Holroyd and Coles (2002), the dopaminergic system, in conjunction with the ACC, mediates FRN, enabling individuals to choose among available options based on positive and negative consequences (Holroyd & Coles, 2002). According to this model, variations in dopamine levels, contingent upon the direction of prediction errors, inversely impact FRN amplitude. Elevated dopamine levels lead to decreased FRN amplitude following positive prediction errors, whereas reduced dopamine levels correlate with increased FRN amplitude following negative prediction errors (Holroyd & Coles, 2002).

The P300 is a positive potential elicited approximately 300–600 msec after feedback or stimulus presentation in the centroparietal regions and serves as an index for attention allocation based on the motivational salience of the stimulus (Nieuwenhuis, Aston-Jones, & Cohen, 2005; Picton, 1992). The amplitude of the P300 increases with the motivational significance of the feedback (Nieuwenhuis et al., 2005). Devoting more attention to stimuli or outcomes also produces a greater (more positive-going) P300 amplitude (Zhang et al., 2013; Donchin, 1981). In the realm of decision-making, risky decisions evoke a greater P300 compared with nonrisky decisions (He, Guan, Zhao, & Cao, 2013; Schuermann, Endrass, & Kathmann, 2012; Oberg, Christie, & Tata, 2011; Christie & Tata, 2009), indicative of heightened motivation toward risky options (Polezzi, Sartori, Rumiati, Vidotto, & Daum, 2010). Furthermore, the magnitude of this component is influenced by the recipient of the economic decision being made, with a larger P300 reported in the self-related conditions compared with the other-related ones (Liu, Hu, Shi, & Mai, 2020). This underscores the role of social context in shaping neural responses during decision-making processes, with individuals exhibiting heightened attentional engagement when decisions directly impact themselves.

The present study aims to investigate the electrophysiological correlates (i.e., DPN, FRN, and P300) of various stages of prosocial and antisocial risky decision-making in adolescents. Specifically, we focus on the response selection and feedback processing stages. To this end, we designed a modified dictator game based on the underlying concepts of prosocial and antisocial risky decisions, as well as a fair decision. Because antisocial and prosocial risky decisions have only recently been introduced, the precise alterations in the abovementioned electrophysiological components following these decisions or their outcomes were not entirely predictable. However, in the response selection stage, we hypothesized that the DPN amplitude would exhibit greater negativity for antisocial and prosocial risky decisions than for fair decisions. Moreover, given the lower self-outcome associated with prosocial risky decisions and their inherent emphasis on social communication, we further expect to observe a heightened DPN amplitude for prosocial risky decisions compared with antisocial ones. Moving to the feedback-processing phase, we expected different neural responses to prosocial and antisocial risky choices. We anticipated that the motivational significance of feedback would impact the FRN. Indeed, the outcomes of risky options (i.e., ASRT and PSRT) are more motivationally salient and rewarding than the risk-free (i.e., fair) option; therefore, they would elicit a smaller FRN. In addition, negative feedback was anticipated to evoke an increased FRN amplitude compared with positive feedback. The P300, serving as an indicator of outcome salience and representation updating, was expected to exhibit a larger (more positive) amplitude following feedback from both PSRT and ASRT options compared with neutral feedback following fair options. In addition, successful feedback outcomes were predicted to elicit a larger P300 compared with unsuccessful outcomes. However, we did not have specific expectations regarding the differences between prosocial and antisocial risky decisions regarding P300 amplitude.

Thirty-two healthy right-handed adolescents aged 13–19 years (15 female participants; mean_age = 15.91 ± 2.16 years) were recruited using an online advertisement through social media (e.g., Instagram, Telegram, and WhatsApp). The outcome of a priori analysis using G*Power Version 3.1.9.7 (Faul, Erdfelder, Lang, & Buchner, 2007) indicated that a total sample size of 28 participants was sufficiently powered at 80% to detect a medium effect size in repeated-measures ANOVA with three measurements (f = 0.25). The inclusion criteria consisted of no history of neurological or psychiatric disorders, surgery, or severe head trauma, which was assessed by a medical doctor through an in-house standardized form in the National Brain Mapping Laboratory, where the experiment was performed, and the data were collected. Participants who did not consume caffeine, alcohol, or drugs for at least 24 hr before the experiment and had at least 7 hr of sleep the night before the experiment were included. All participants had normal or corrected-to-normal vision. Three participants were excluded from the final analysis due to excessive signal artifacts, leaving the data of 29 participants (15 male participants) to analyze. Participants and their parents provided written informed consent to participate and were compensated with 100 Tomans ($2). The ethics committee of Shahid Beheshti University approved the study.

Given the absence of an Iranian adolescent face database, our initial image selection was drawn from the Iranian Emotional Face database (Heydari, Sheybani, & Yoonessi, 2023; https://e-face.ir/). This database comprised 37 colorful face photographs depicting young Iranian adults (23 male and 14 female participants). To create adolescent representations, the photographs in the E-face database were transformed using the Faceapp application. The attractiveness of target faces could possibly influence participants' decisions. Research has shown that individuals tend to make more prosocial decisions when encountering more attractive receivers, as evidenced by greater allocations (Kleinhans & Nicholls, 2023; Rosenblat, 2008) or offers (Lucas & Koff, 2013; Andreoni & Petrie, 2008) in games like The Dictator and Ultimatum games. Thus, to control the effect of the receiver attractiveness, 21 participants (mean_age = 27.57 years, SD = 8.05 years; 9 male participants) evaluated the attractiveness of each face in a pilot investigation. They rated each of the photographs on a scale ranging from 0 to 10 based on how attractive they found the faces. Subsequently, 18 photographs (9 male and 9 female participants) with a mean attractiveness of 0.75 more and less than the median, which was 5, were selected for the next step. Finally, 12 photographs (6 male and 6 female participants) that did not exhibit statistically significant differences from one another were chosen.

As in the traditional Dictator Game, the task involves two players: the decider and the receiver. In each round, players are endowed with a fixed amount of money, and deciders must choose whether to keep the fair allocation between themselves and the receiver or to alter it. Participants were always assigned the role of the decider. They were informed that they would interact with six different receivers who had previously participated in the experiment on prior days in the laboratory. Moreover, participants were informed that their image would be presented to the next six participants as a receiver. However, no physical receiver was present in the game, and six different preselected face stimuli represented receivers (see the Stimuli Selection section).

As depicted in Figure 1, each trial began with a white fixation cross appearing in the center of the screen against a black background for 200 msec. Subsequently, participants were presented with the image of a same-gender receiver for 500 msec. The task was coded, so the same receiver never appeared consecutively across two trials. After a blank screen jittered interval of 700–1000 msec, individuals were presented with three options, each representing a particular type of decision, remaining on the screen for a maximum of 2000 msec as the decision time. Each option was denominated in coins; each player's share was 50 coins per trial. The central option represented a fair and risk-free decision, ensuring an equal distribution of 50 coins to both players upon selection. In contrast, risky options were presented on the right and left sides, introducing a 50% probability of success or failure. On the right-hand side, the prosocial risky option was presented, where, in the case of success, the receiver received 100 coins and the decider received 45 coins. In comparison, failure resulted in the receiver obtaining 50 coins and the decider receiving 30 coins. This option showcased a propensity for PSRT, as the participant took the risk of losing 15 coins to benefit the other player with 50 more coins. Conversely, the left-hand option mirrored the nature of its counterpart. However, upon success, the decider received 100 coins, with no allocation to the receiver. The failure led to the decider obtaining no more than 25 coins, and the receiver received 50 coins. Opting for this alternative represented ASRT, as the deciders risk their share to maximize their personal gain, leaving nothing for the receiver. Participants were informed that each player is allocated a fair share of 50 coins, and by choosing either the prosocial or antisocial risky option, they would intentionally alter this fair share. Specifically, they were told that one of the choices (ASRT) would violate the receiver's share and the other choice (PSRT) would benefit the receiver, whereas the last choice (fair) would lead to having equal shares.

If a decision was not made within the designated timeframe (i.e., no response), no coin was awarded to the participant, and the other player received their share, which was 50 coins. Participants had to utilize the right, down, and left arrow keys to select the PSRT, fair, and ASRT options, respectively. Whether the participant pressed one of the keys on the keyboard that corresponded to the options or did not press any of the mentioned keys for a maximum period of 2000 msec, all the options disappeared. Following a jittered interval of 800–1100 msec, feedback was provided for 1000 msec, categorized as success (indicated by a checkmark), failure (indicated by an x mark), and neutral (presented by an empty circle). There was an equitable probability for success and failure feedback, each presented with a chance of 50% on each trial (see Figure 2B for the feedback distribution). Neutral feedback also appeared either after choosing the fair option or after the decision period of 2 sec had passed. In addition, on the right and left sides of the check, x or neutral marks indicated the number of coins belonging to the decider and receiver, respectively. Finally, an intertrial interval of 1000 msec separated the trials.

The number of coins was determined based on several factors. First, to avoid the contamination of self and other-related risk, and following the antisocial and prosocial risky decision definitions provided earlier, it was decided that the decider would always bear the risk. This meant that in cases of unsuccessful prosocial or antisocial decisions, the receiver would not experience any change and would continue to receive 50 coins. Conversely, in cases of success, the outcome had to be maximized: 100 coins for the decider in a successful antisocial risky decision and 100 coins for the receiver in a successful prosocial risky decision. To maintain consistency with previous studies, it was also important to ensure that the difference between the expected value of the prosocial and antisocial decisions, compared with the fair decision, remained the same. This led to the following equation: (100 + 50)/2 − 50 = 50 − (successful prosocial outcome for the receiver + 50)/2, which resulted in a successful prosocial outcome of 0 coins for the receiver. On the other hand, the expected return for the decider was modeled by the equation: (100 + unsuccessful antisocial risky outcome) − 50 = 50 − (successful prosocial risky outcome + unsuccessful prosocial risky outcome). Although this equation could yield various solutions, the goal was to keep the successful prosocial risky outcome as high as possible but less than 50 coins. This approach ensures that participants incur a cost when choosing the prosocial option—aligning with the definition of prosocial decisions—while still making the expected return appealing enough for consideration. This led to a successful prosocial risky outcome of 45 coins, which, in turn, yielded an unsuccessful prosocial risky outcome of 30 coins. This final configuration created a balance between the expected returns of the prosocial and antisocial decisions in relation to the fair decision (50 coins). The expected return for the decider was (45 + 30)/2 = 37.5 coins for prosocial decisions and (100 + 25)/2 = 62.5 coins for antisocial decisions. For the receiver, the expected returns were (100 + 50)/2 = 75 coins for prosocial decisions and (0 + 50)/2 = 25 coins for antisocial decisions.

The task consisted of at least 480 trials divided into six blocks of 80 trials each. If an option was not selected within 2 sec, one trial was added to the total number of trials. To have enough trials in different conditions for ERP analysis and to avoid the excessive choice of one particular option, whenever any of the three options was chosen 160 times, that option was blurred and could no longer be selected. When a block was completed after 80 trials, the total number of coins earned by the participant appeared on the screen, and a 60-sec rest period was allowed between blocks.

On the day before the experiment, participants were instructed to send a photograph of themselves in a predetermined manner. They were told they would randomly play against six other teenagers who had previously participated in the experiment or would participate later. They were also told that they would play the role of the decider in the experimental session, and their photograph would be used as the receiver for the other six participants (see Modified Dictator Game section).

The experiment was held in an electrically shielded, sound-attenuated room. Upon arrival, participants were instructed to participate in a computerized task. In this task, they would randomly encounter six adolescents within their age range who either already participated in the experiment or would participate later but had sent us their photographs. They were told that they would play with each of the six receivers multiple times in random order, and they could decide about each receiver in each round as they preferred. In addition, they were instructed that they would take on the receiver role for another six adolescents who would participate in the experiment on subsequent days. To this end, they were informed that their photographs would be shown to future participants. Participants were informed that they would be randomly awarded the number of coins received in some trials after the end of the project. Plus, they became aware that half of the money they would receive was based on their decisions, whereas the other half would be the average of the six times they would act as the receiver. Participants then sat on a comfortable chair approximately 60 cm away from a 22-in. screen with a refresh rate of 60 Hz. Afterward, the EEG electrodes were attached, and participants played 10 training trials to get acquainted with the experiment, followed by the main experiment. The experiment lasted approximately 50 min. After completing the task, all participants were debriefed and informed about the true nature of the experiment. They were asked whether they believed the cover story. None reported suspecting any form of deception.

The EEG data were recorded using a g.HIamp (g.tec company) device with Ag/AgCl electrodes, and 64 channels was placed in the 10–20 EEG system. The sampling frequency was set to 512 Hz, and A2 and FPz were used as reference and ground, respectively. The EOG electrodes were placed at the external canthi of each eye and below the right eye. Impedances were maintained below 10 kΩ. EEGLAB toolbox (Delorme & Makeig, 2004) was used to preprocess the EEG data. The raw signal was bandpass filtered in the 0.5- to 30-Hz range. Waveforms were first inspected visually; then, the Clean Rawdata plugin (Pernet, Martínez-Cancino, Truong, Makeig, & Delorme, 2021) was used to clean data for artifacts. Then, the data were rereferenced to the average of all electrodes. Next, an independent component analysis was performed, and sources related to eye movement, head movement, muscle activity, heartbeat artifacts, and channel noise were excluded from the data. Then, based on different stages of decision-making, different epochs were extracted.

To analyze the DPN component, a slow negative potential within the RP family (Bianchin & Angrilli, 2011), epochs were extracted from 800 msec before the response (i.e., pressing the response keys) to 800 msec after the response (Cui et al., 2013)—moreover, the time interval from 600 to 800 msec after the response was used as the baseline (Cui et al., 2013). Epochs with voltages outside the 75 μV- and −75-μV range were excluded from averaging (Liu et al., 2020). ERPs were separately averaged across PSRT (mean remaining epochs: 104.63), fair (mean remaining epochs: 102.17), and ASRT (mean remaining epochs: 111.10) trials. Regarding the topographic maps (see Figure A1), the literature (Cui et al., 2013; Bianchin & Angrilli, 2011), and the average RT of all participants (0.51 sec), Fz, FCz, and Cz electrodes in the −500- to 0-msec time window (i.e., 500 msec preceding the response) were selected to conduct a DPN analysis.

The components that were considered for analysis at this stage were P300 and FRN. Epochs were extracted in two ways based on decision conditions (ASRT, PSRT, and fair) as well as feedback conditions (ASRT failure; ASRT_F, ASRT successful; ASRT_S, fair, PSRT failure; PSRT_F, and PSRT successful; PSRT_S) from 200 msec preceding the appearance of feedback to 800 msec after displaying the feedback. Baseline corrections were made using the 200-msec interval before feedback onset. Epochs with a mean voltage greater than 75 μV or less than −75 μV were excluded from averaging. Epochs were once averaged across PSRT (mean remaining epochs: 109.87), fair (mean remaining epochs: 110.80), and ASRT (mean remaining epochs: 114.23) options (decision conditions) and then averaged across ASRT_F (mean remaining epochs: 56.60), ASRT_S (mean remaining epochs: 57.63), fair (mean remaining epochs: 55.47), PSRT_F (mean remaining epochs: 56.17), and PSRT_S (mean remaining epochs: 53.70) feedbacks (feedback conditions). According to previous studies (Li et al., 2020; Kardos, Tóth, Boha, File, & Molnár, 2017; Carlson, Aknin, & Liotti, 2016; Leng & Zhou, 2014) and the topographical distributions, P300 was analyzed across Pz, CPz, and Cz channels in the period of 300–450 msec. In line with previous studies (Liu et al., 2020; Zhong et al., 2020; Kardos et al., 2017), FRN was also calculated across Fz, FCz, and Cz electrodes in the 220- to 300-msec time range based on the visual inspection of the grand average waveforms.

Two different two-way repeated-measures ANOVAs were conducted to investigate the differences between ERP components in different risky decisions. First, two-way repeated-measures ANOVAs were conducted with a 3 (Decisions: ASRT, PSRT, and fair) × 3 (Fz, FCz, and Cz for DPN and FRN; Pz, Cpz, and Cz for P300) within-subject factor design and DPN, P300, and FRN components as the dependent variables. Then, feedback effects were included, and the same analyses were done for P300 and FRN, this time with five levels of feedback (ASRT_F, ASRT_S, fair, PSRT_F, and PSRT_S). Because DPN is not a feedback-related component, those analyses were limited to FRN and P300. If sphericity was violated, the Greenhouse–Geisser correction was used, and corrected p values (uncorrected degrees of freedom) were reported in the Results section. Pairwise t tests were used as post hoc tests, and p values were corrected using the Benjamini & Hochberg FDR (FDR-bh) method (Benjamini & Hochberg, 1995). In addition, as part of a sensitivity analysis, we repeated all these analyses while controlling for age and gender. Moreover, to investigate the influence of facial stimuli on participants' decisions, we first assessed differences in option selection and RT across facial stimuli. Next, we applied two separate random effects models with option selection count and RT as dependent variables. Option type, age, and gender were included as fixed effects, whereas facial stimuli and participant ID were treated as random effects. All analyses were done using Python programming language and IBM SPSS Statistics Version 26 software. Effect sizes were calculated using the partial eta squared (η_p²) and the eta squared (η²).

During the first half of trials (240 first trials, in which participants could freely decide between all options), participants chose the three options at a comparable rate (see Figure 2A), with the antisocial risky decision as the most frequent (90.59 ± 26.34), the prosocial risky decision as the least frequent (64.07 ± 24.29), and the fair decision in between (83.45 ± 29.71). A one-way repeated-measures ANOVA with the number of selections as the dependent variable and Options (ASRT, fair, and PSRT) as the within-subject factor showed a significant main effect of Options, F(2, 56) = 5.04, p = .010, η_p² = 0.153. Post hoc pairwise t test results demonstrated a significant difference between the number of choosing ASRT and PSRT options, t(28) = 3.46, p = .005, η² = .215. In contrast, this significance was not observed when comparing fair option with antisocial, t(28) = −0.76, p = .45, and prosocial, t(28) = 2.21, p = .054, alternatives. In contrast, the RT was relatively equal when choosing fair (0.50 sec ± 0.29 sec) and ASRT (0.50 sec ± 0.29 sec) options, slightly quicker than selecting PSRT (0.52 sec ± 0.29 sec) options. The result of a one-way repeated-measures ANOVA with RT as the dependent variable and Options as a within-subject factor indicated that there was no significant difference between the RT of choosing ASRT, fair, and PSRT options, F(2, 56) = 2.02, p = .14.

Descriptive statistics of DPN, FRN, and P300 components at all selected electrodes (Fz, FCz, and Cz for DPN and FRN; Pz, Cpz, and Cz for P300) concerning feedback and decision conditions are reported in Table 1. Detailed results regarding the repeated-measures ANOVA and post hoc pairwise t tests along with bar charts (Figure A2) representing the mean and the standard error of ERP components mean amplitudes across decision and feedback conditions at three electrodes (Fz, FCz, and Cz for DPN and FRN; Pz, Cpz, and Cz for P300) were provided in Appendix A.

The mean DPN amplitude averaged for all three options (ASRT, fair, and PSRT) reached a maximum negative deflection in a time window of −500 to 0 msec over Fz, FCz, and Cz channels, as depicted in Figure 3. A two-way repeated-measures ANOVA (options: ASRT, fair, and PSRT × electrodes: Fz, FCz, and Cz) was used to examine the difference between the mean DPN amplitude across three options (Table A1). A significant difference was found between the mean DPN across options, F(2, 56) = 5.82, p = .006, η_p² = .172. Pairwise t tests (Table A2) revealed that the DPN component is significantly more negative-going before making the prosocial risky (−3.01 ± 1.95 μV) option compared with both the fair (−2.50 ± 1.95 μV), t(28) = −3.20, p = .007, η² = .023, and the antisocial risky (−2.41 ± 2.04 μV), t(28) = −3.05, p = .007, η² = .031, options. However, there was no significant difference between choosing the antisocial risky option and the fair one, t(28) = 0.40, p = .70.

The mean FRN amplitudes were measured at the same electrodes as the DPN in a time window of 220–300 msec post-feedback (Figure 4A–C). A 5 (Feedback Conditions: ASRT_F, ASRT_S, fair, PSRT_F, and PSRT_S) × 3 (Electrodes: Fz, FCz, and Cz) repeated-measures ANOVA (Table A3) showed a significant main effect of Feedback Conditions, F(4, 112) = 4.07, p = .009, η_p² = .127. Pairwise t tests (Table A4) results revealed that facing neutral feedback following making the fair decision evoked the most negative FRN (−4.58 ± 2.74 μV) compared with receiving successful feedback after deciding to choose ASRT (−3.46 ± 3.44 μV), t(28) = −3.13, p = .020, η² = .041, and PSRT (−3.82 ± 2.68 μV), t(28) = −2.97, p = .020, η² = .027 options. Moreover, a larger FRN is elicited for failure feedbacks (−4.33 ± 3.12 μV) compared with successful ones (−3.46 ± 3.44 μV), following antisocial risky choices, t(28) = −3.04, p = .020, η² = .024. However, no significant difference was found between successful and failure feedback following the prosocial risky option, t(28) = 1.29, p = .30.

The P300 mean amplitudes were measured at Cz, CPz, and Pz electrodes at 300–450 msec postfeedback (Figure 4C–E). A two-way repeated-measures ANOVA (Feedback Conditions: ASRT_F, ASRT_S, fair, PSRT_F, and PSRT_S × Electrodes: Pz, CPz, and Cz) revealed a significant main effect of Feedback Conditions, F(4, 112) = 18.63, p < .001, η_p² = .400, demonstrating that receiving successful feedback after choosing the ASRT option (2.80 ± 4.18 μV) elicited the largest P300 amplitude while observing the neutral feedback after choosing the fair option (0.17 ± 3.82 μV) evoked the smallest P300 (see Table A5 for detailed statistics). The pairwise t tests (Table A6) revealed significant differences between all feedback conditions (all ps < .05; see Table 2) except the difference between the failure ASRT and the successful PSRT feedbacks, t(28) = 0.52, p = .61.

As shown in Figure 5A–C, the mean amplitudes of 220–300 msec at Fz, FCz, and Cz were selected for FRN. A 3 (Decision Conditions: ASRT, fair, and PSRT) × 3 (Electrodes: Fz, FCz, and Cz) repeated-measures ANOVA (Table A7) revealed a significant main effect of Decision Conditions, F(2, 56) = 5.39, p = .013, η_p² = .161. Pairwise t test outputs (Table A8) showed that when participants received the feedback, FRN amplitude was more negative for choosing the fair option (−4.62 ± 2.62 μV) in comparison with the PSRT (−3.99 ± 2.73 μV), t(28) = −3.77, p = .002, η² = .018, and the ASRT (−3.89 ± 3.16 μV) alternatives, t(28) = −2.59, p = .023, η² = .021.

The mean P300 amplitudes were measured in a 300- to 450-msec time window after confronting feedback at Cz, Pz, and Cpz electrodes, as depicted in Figure 5C–E. A 3 (Decision Conditions: ASRT, fair, and PSRT) × 3 (Electrodes: Pz, CPz, and Cz) repeated-measures ANOVA (Table A9) demonstrated a significant main effect of Decision Conditions, F(2, 56) = 25.93, p < .001, η_p² = .481. Pairwise t tests (Table A10) suggested that when the feedback was displayed, the amplitude of the P300 component was larger after deciding to choose the ASRT option (2.38 ± 4.25 μV) compared with the PSRT (1.42 ± 3.77 μV), t(28) = 3.58, p = .001, η² = .025, as well as fair (0.16 ± 3.71 μV) alternatives, t(28) = 5.98, p < .001, η² = .135. P300 was also more positive after selecting prosocial risky compared with fair options significantly, t(28) = 4.54, p < .001, η² = .060.

To investigate the potential influence of age and gender on the results, sensitivity analyses were conducted at both behavioral (see Appendix B for more information) and neurophysiological (see Appendix C for more details) levels using mixed ANCOVAs. Across all analyses, the main effects of age and gender were statistically insignificant. The results remained consistent for behavioral data, DPN, and P300, with no interaction observed between the variables of interest and age or gender. For FRN in the feedback condition, while the previous results remained unchanged, a significant interaction was found between Age and Feedback, F(3.24, 84.28) = 0.4.28, p = .006 (Table C2). Despite this, none of the post hoc comparisons remained significant after correcting for multiple comparisons using the Tukey method (Table C6). In the decision condition for FRN, the results similarly remained consistent after adjusting for age and gender. Nonetheless, a significant interaction between Age and Condition was observed, F(1.68, 43.78) = 5.11, p = .014 (Table C4). Post hoc analysis revealed that the difference in FRN amplitude between fair and ASRT increased with age, β = 0.323, SE = 0.118, t(26) = 2.745, p = 0.027 (Table C7).

To examine whether facial stimuli affect participants' decisions, we performed two sensitivity analyses. We first investigated differences among facial stimuli in terms of option selection and RT. No significant differences were found in either the number of selections or RTs across facial stimuli (Tables D1–D4). In a second analysis, we performed two separate random effects models with the number of option selections and RT as dependent variables. The fixed effects included Option Type, Age, and Gender, whereas facial stimuli and participant ID were random effects. Consistent with previous analyses, participants showed a significant preference for choosing antisocial risky options over prosocial risky options, β = 4.42, p < 0.001. A significant difference was also found between fair option selection and prosocial risky option selection, with a higher frequency of fair option choices, β = 3.23, p < 0.001. In addition, fair decisions were made significantly faster than prosocial ones, accounting for the facial stimuli, β = 0.02, p = 0.027. Despite that, the variance and standard deviation for the photograph intercept were nearly zero when RT or the number of selections were the dependent variables (Table D5).

The current study examined prosocial and antisocial risky decision-making in adolescents both at behavioral and electrophysiological levels. More specifically, using the modified dictator game task, we studied the ERP components (i.e., DPN, FRN, and P300) elicited in different stages (i.e., response selection and feedback evaluation) of fair, prosocial, and antisocial risky decision-making processes. At the behavioral level, adolescents selected significantly more antisocial risky options over prosocial ones. ERP findings at the electrophysiological level and in the response selection phase demonstrated that DPN at the frontocentral regions was more negative-going before making prosocial risky decisions, compared with antisocial risky and fair decisions. In the feedback evaluation stage, FRN had the least negative mean amplitude after choosing the antisocial risky option and encountering successful feedback. However, choosing the fair option and facing the neutral feedback elicited the most negative FRN. In addition, P300 showed the most positive mean amplitude following making the antisocial risky decision and confronting the successful feedback. However, making the fair decision and encountering the neutral feedback resulted in the least positive P300. These findings highlight the different electrophysiological underpinnings of responses before and after making the prosocial and antisocial risky decisions.

Our behavioral results revealing a higher frequency of antisocial risky choices are in line with previous evidence for an increase in self-oriented behavior in adolescents (Güroğlu, van den Bos, & Crone, 2014; Crone, 2013) that might originate from heightened activation in neural substrates associated with enhancing self-gain, like dorsomedial prefrontal cortex (van den Bos, van Dijk, Westenberg, Rombouts, & Crone, 2011). Moreover, adolescents may be more vulnerable to antisocial decision-making compared with prosocial ones concerning the developmental trajectories of these decisions (Moffitt, 2007). There is a considerable rise in antisocial decisions during adolescence (Blonigen, 2010), whereas findings regarding prosocial decisions are mixed (Do et al., 2017; Eisenberg, Cumberland, Guthrie, Murphy, & Shepard, 2005). Although there is some evidence for increased prosocial giving during adolescence (Kwak & Huettel, 2016), it seems that what is more crucial than prosocially toward others is taking the perspective of others (Kilford, Garrett, & Blakemore, 2016). Perspective-taking was not evaluated among participants during the task, which represents a limitation of the study. Considering that the participants were unfamiliar with the receivers, the lack of perspective-taking might have influenced their decision-making (Güroğlu et al., 2014). This potential effect cannot be examined with the current data and should be addressed in future studies. These findings add to the previous ones by revealing that self-benefiting decisions are more prominent in adolescents' social decisions, even when they are antisocial and risky.

Our electrophysiological findings revealed that at the response selection stage, the mean amplitude of DPN preceding prosocial risky decisions was significantly higher (more negative) than before choosing antisocial risky and fair options. This finding presents several potential interpretations. At first glance, it aligns with existing literature indicating that DPN deflection tends to increase before disadvantageous outcomes (Guo et al., 2019; Dong et al., 2016; Giustiniani et al., 2015; Bianchin & Angrilli, 2011; Carlson, Zayas, & Guthormsen, 2009). In our modified dictator game task, opting for the PSRT option always results in obtaining fewer coins, suggesting a disadvantageous outcome for the individual and potentially explaining the greater DPN deflection.

However, a deeper analysis suggests a nuanced perspective. If we solely consider the expected returns, one might anticipate that choosing the fair option would evoke a higher DPN compared with the antisocial risky option, which entails the highest expected return. The absence of a significant difference between antisocial risky and fair decisions in our study may be attributed to the deliberate design of our task, which integrates both risk and return factors simultaneously, sometimes in opposition. For instance, previous research has indicated higher DPN for risky options (Lu, Li, Li, & Li, 2023; Catena et al., 2012), implying that although the higher expected return associated with antisocial risky decisions might lower DPN, the elevated risk involved could conversely increase it. This complex interplay may contribute to our results' lack of distinction between fair and antisocial risky decisions. This may lead to the inability to find a difference between fair and antisocial risky decisions.

Notably, most prior studies have investigated DPN using the Iowa Gambling Task (Dong et al., 2016; Giustiniani et al., 2015; Cui et al., 2013; Bianchin & Angrilli, 2011), which involves a nonsocial situation where participants aim to maximize personal gain. However, interpreting DPN within the context of social interactions necessitates a distinct perspective. Studies examining SPN, another type of RP closely related to DPN, have demonstrated higher activation levels in social contexts compared with nonsocial ones, particularly when individuals anticipate judgment or feedback from others (van der Molen, Harrewijn, & Westenberg, 2018; van der Molen, Dekkers, Westenberg, van der Veen, & van der Molen, 2017; van der Molen et al., 2014). Furthermore, RPs are closely linked to affective and motivational valence regarding the anticipated outcome, with heightened activation observed when individuals are particularly motivated and want something (Giustiniani et al., 2020; Berridge, Robinson, & Aldridge, 2009) or when their needs are being addressed (Oumeziane, Schryer-Praga, & Foti, 2017). Thus, our findings may indicate that, in contrast to fair or antisocial risky decisions, individuals making prosocial risky choices are more attuned to their decisions and how they might be perceived by their peers, reflecting a heightened motivational orientation toward social evaluations.

Regarding the early phase of the feedback evaluation, FRN exhibited a more negative-going response when encountering failure feedback following antisocial risky decisions compared with successful feedback. However, no difference was observed between successful and unsuccessful feedback for prosocial risky decisions. This finding holds significant implications, given that greater FRN amplitude is typically associated with unfavorable feedback (Leng & Zhou, 2014; Cui et al., 2013; Bianchin & Angrilli, 2011; Schuermann et al., 2011; Pfabigan, Alexopoulos, Bauer, & Sailer, 2010; Christie & Tata, 2009; Wu & Zhou, 2009; Gehring & Willoughby, 2002). Our finding suggests that participants place considerable emotional significance on the outcomes of their decisions, particularly when opting for antisocial risky behaviors. This likely reflects greater activation of the dopaminergic reward system for the outcome of antisocial risky choices. However, when making prosocial risky decisions, participants might not discriminate between successful and unsuccessful outcomes as favorable and unfavorable that strongly, perhaps due to the more outcome-focused nature of antisocial behaviors. This observation may stem from the notion that antisocial behaviors are more outcome-oriented than prosocial behaviors.

Furthermore, we found no disparity in FRN amplitude between neutral feedback following fair decisions and unsuccessful outcomes of both prosocial and antisocial risky decisions. However, neutral feedback for fair decisions evoked higher FRN compared with successful outcomes of prosocial and antisocial risky decisions. Despite the predictability of neutral feedback following fair decisions, risk-taking and ambiguity appear to modulate FRN mean amplitude. Some studies have demonstrated a reduction in FRN amplitude with riskier decisions (Yi et al., 2018; Schuermann et al., 2011; Gehring & Willoughby, 2002). Two (ASRT and PSRT) out of three options in our task also entail risk. Thus, reduced FRN following antisocial and prosocial risky decisions compared with neutral feedback derived from risk-free fair decisions is in line with previous studies (Yi et al., 2018; Schuermann et al., 2011; Gehring & Willoughby, 2002).

In the later stages of the feedback evaluation, we observed a greater P300 response following success feedback in both prosocial and antisocial risky decisions than failure feedback. This finding corroborates previous research associating P300 with enhanced stimulus processing and the motivational significance of the outcomes (Kardos et al., 2017; Leng & Zhou, 2014; Kamarajan et al., 2010; Luu, Shane, Pratt, & Tucker, 2009; Wu & Zhou, 2009; Yeung, Holroyd, & Cohen, 2005). Furthermore, this discovery holds significance as it demonstrates the absence of a distinction between FRN responses to successful and unsuccessful prosocial risky feedback while revealing a notable disparity in the P300 component between these outcomes. One interpretation lies in the difference between these two components. Although FRN reflects a more early binary evaluation of the outcomes based on valence (positive vs. negative), P300 is linked to later stages of processing and reflects aspects such as outcome magnitude and salience, as well as attention, memory, and updating of mental representations (Polich, 2007; Hajcak, Moser, Holroyd, & Simons, 2006; Sato et al., 2005; Donchin, 1981). In addition, previous research has established a correlation between FRN and negative affect but not with positive ones, whereas the relationship is reversed for the P300 component (Sato et al., 2005). Thus, our findings may be attributed to the fact that when individuals make a prosocial risky decision, successful and unsuccessful feedback do not significantly differ in valence or at least neither induces negative affects. Consequently, we do not observe a difference in FRN amplitudes. However, they accurately discern and incorporate the difference between successful and unsuccessful feedback to update their representations regarding the relationship between the choice and its outcome.

Moreover, regardless of the outcome's success or failure, prosocial and antisocial feedback elicited a higher P300 than neutral feedback received after the fair decision. Notably, successful antisocial risky feedback triggered a greater P300 response than successful prosocial risky feedback, whereas unsuccessful antisocial risky feedback induced a heightened P300 compared with unsuccessful prosocial risky feedback. These findings can be interpreted according to the motivational significance of outcomes following different decision options in our modified dictator game. Given that the outcomes of making fair decisions were always constant, their motivational salience was lower compared with the outcomes of the other two options. Consequently, participants allocated less attention to the outcomes associated with the fair option. In addition, previous studies have reported larger P300 mean amplitude after making risky choices compared with risk-free ones (He et al., 2013; Schuermann et al., 2012; Oberg et al., 2011; Christie & Tata, 2009). Thus, the fair decision required less motivation and involved less risk, giving rise to the smallest P300 response. Regarding the difference between prosocial and antisocial feedback in the same situation (successful vs. successful, unsuccessful vs. unsuccessful), these findings may reflect prioritizing one's own gain over others' gain. Liu and colleagues (2020) demonstrated that individuals showed an increased P300 when making decisions for themselves compared with others. In our dictator game task, it could be inferred that adolescents perceived themselves as the primary target during antisocial risky decisions. In contrast, when making prosocial risky choices, the primary focus might be benefiting others. Consequently, participants weighted the additional obtained coins in the successful antisocial risky feedback more heavily than the confederate's additional outcome in successful prosocial risky feedback. Similarly, when the feedback was unsuccessful, participants incurred greater losses in antisocial risky decisions than in prosocial risky ones (25 vs. 20). On the other hand, as mentioned earlier, P300 is affected by allocating attentional resources based on the motivational significance of stimuli and the updating of mental representations (Nieuwenhuis et al., 2005; Picton, 1992). It is plausible that individuals update their mental representation toward prosocial risky decisions to a lesser extent than antisocial risky decisions or allocate fewer attentional resources to these outcomes. As a result, outcomes following antisocial risky choices may absorb more attention compared with outcomes following prosocial risky alternatives, leading to a greater P300 response. However, it is essential to note that these comparisons are valid only for the case of comparing successful antisocial to successful prosocial and unsuccessful antisocial to unsuccessful prosocial decisions. As our results show, there was no difference between unsuccessful antisocial risky and successful prosocial risky feedback, which emphasizes that antisocial risky decisions alone should not be considered as more salient compared with prosocial decisions.

In conclusion, this study represents the first investigation into prosocial and antisocial risk-taking in adolescents conducted simultaneously. Moreover, utilizing EEG signals, we, for the first time, explored the temporal dynamics of prosocial and antisocial risky decision-making within a social exchange context, revealing distinct electrophysiological indicators across various stages of social decision-making during adolescence. Although these findings offer valuable insights into understanding such decision types, they must be viewed in light of the limitations. First, limiting prosocial and antisocial risky decisions to a computer-based task may lack ecological validity. Future research can examine these decision types in field experiments to better capture the complexities of decision-making behaviors. Second, our study focused primarily on the economic and financial aspects of prosocial and antisocial risky decisions, neglecting potential nonmonetary aspects. This might also be one of the reasons that our participants chose antisocial risky decisions more than prosocial ones. It is possible that other types of prosocial risky decisions are more engaging for adolescents. Moreover, our study did not include other age groups or incorporate more exploratory behavioral and questionnaire data. Thus, a robust interpretation of this important issue cannot be provided. Future investigations should incorporate broader considerations to provide a more comprehensive understanding of these decision types. Third, the antisocial risky decisions may have been underestimated, as participants were not able to freely choose between all options throughout the entire task. However, during the first half of the task, on which the behavioral analysis is based, participants were selecting options freely without any restrictions. We found that by the beginning of block five (i.e., after completing 320 trials), 26 participants had the opportunity to choose the ASRT option freely. Nevertheless, in the final block, the majority of participants (17 out of 29) were no longer able to select the antisocial risky option. Another limitation is the imbalance in the number of coins across different conditions. Although incorporating risky prosocial and antisocial options within a task inevitably requires unbalanced coin distributions, future studies might investigate the differences between various options, including those that incur risk to both parties and nonrisky options. Next, the location of decision choices was not counterbalanced across participants, which may introduce a confounding effect due to the position of the options. To address this potential bias, we suggest that future studies counterbalance the position of options to minimize this effect. Another confounding variable could be the age of the participants. We found that FRN could be influenced by age because a significant interaction effect between age and feedback/decision conditions was observed. Yet, given the sample size and the age distribution within our study, this finding should be interpreted with caution, and future studies are needed to address this issue. One other limitation is linked to using more scales to consider broader aspects of participants' personalities and intentions. It is highly recommended to include post questionnaires regarding participants' purpose of choosing options and factors such as psychoticism, self-control, social conformity, and family environment (Morgado & da Luz Vale-Dias, 2016) in future studies to shed light on the underlying reasons why participants chose antisocial risky options more than prosocial ones. Last, possible gender effects were not reliable in the current study due to sample size limitations. Gender might be a more considerable variable that could impact decision-making behaviors, with research suggesting that risky and antisocial decisions are more prevalent among males, whereas females tend to exhibit more prosocial tendencies (Russell, Robins, & Odgers, 2014; Pursell, Laursen, Rubin, Booth-LaForce, & Rose-Krasnor, 2008; Byrnes, Miller, & Schafer, 1999). Therefore, future studies should strive to account for gender differences to gain a more nuanced understanding of decision-making dynamics in adolescents. By addressing these limitations and continuing to explore the complexities of prosocial and antisocial decision-making, we can further enhance our understanding of social behavior and contribute to developing interventions to promote positive decision-making outcomes among adolescents.

To examine the effects of gender and age on behavioral outcomes, we conducted two mixed ANCOVAs. The within-subject variable was the option (ASRT, fair, and PSRT), with age and gender included as the covariate. The dependent variables were RT (Table B1) and the number of options selected (Table B2). In case of significant results, pairwise comparisons were also reported (Table B3). In addition, descriptive statistics for RT and the number of options selected, grouped by options and gender, are reported in Table B4.

For each component (DPN, FRN, and P300), a 3 (decisions: ASRT, PSRT, and fair) × 3 (Fz, FCz, and Cz for DPN and FRN; Pz, Cpz, and Cz for P300) × 2 (gender: female, male) mixed ANCOVA was conducted with age as a covariate (Tables C1–C5). When the interaction effect of age and conditions was significant, pairwise comparisons (Tables C6–C7) and interaction plots (Figures C1 and C2) were reported. Descriptive statistics, grouped by gender, conditions, and electrodes, are reported in Table C8.

To investigate whether facial stimuli influenced participants' decisions, we conducted four 1-way ANOVAs with faces (the person they were playing with) as the independent variable. Because boys played with male receivers and girls played with female receivers, two separate ANOVAs were performed for each gender. The two dependent variables examined were RT and number of option selections). As shown in Tables D1–D2, facial stimuli did not significantly affect females' decisions, and the same absence of significant effect was observed for males (Tables D3–D4). In addition, to account for the random effects of facial stimuli, two linear mixed-effects models, with option (ASRT, Fair, and PSRT) as the main predictor and gender and age as covariates, were employed. Both RT and the number of option selections were treated as outcome variables (Table D5).