Social Feedback Promotes Positive Social Sharing, Trust, and Closeness

Abstract

Positive social sharing (PSS) is an interpersonal emotion regulation strategy that enhances positive affect and social belonging, particularly when met with positive social feedback. Despite the ubiquity of PSS both in-person and online, what drives this behavior is not well understood. We hypothesized that positive social feedback serves as a reward that reinforces sharing behavior and strengthens social bonds. Participants made trial-by-trial choices about whether to share social media photos with peers who returned positive (‘likes’) or negative (‘dislikes’) feedback. Unbeknownst to participants, peer conditions were manipulated to yield varying amounts of positive and negative feedback. Social bonding was subsequently measured using a trust game and subjective closeness ratings. Participants shared more with peers who provided greater rates of positive feedback. This effect generalized to trust decisions and subjective feelings of closeness and varied individually as a function of interpersonal emotion regulation in daily life.

Sharing positive experiences with others is a common part of everyday life. Whether it be a promotion at work, a tropical family vacation, or funny animal videos, people have a tendency to share life’s positive moments with others. In fact, people share approximately 60-80% of positive events that occur throughout their day (Gable et al., 2004). Today, sharing positive life events is perpetuated even further by the ubiquity of social media and other social technologies (Choi & Toma, 2014). From a psychological perspective, this positive social sharing (or ‘capitalization’; Langston, 1994) is thought of as an interpersonal emotion regulation strategy that amplifies positive affect, strengthens feelings of social belonging, and contributes to life satisfaction and subjective well-being (Arewasikporn et al., 2019; Gable et al., 2004; Gable & Reis, 2010; Lambert et al., 2013; Rimé, 2009; Zaki & Williams, 2013). These benefits also hold true online, where sharing on social media is associated with enhanced positive affect and perceptions of social bonding and well-being (Choi & Toma, 2014; Kim & Kim, 2017; Kross et al., 2021).

One potential mechanism that could be driving positive social sharing is positive feedback from others, such as support, praise, or a smile. Indeed, self-report and daily-diary studies demonstrate that positive feedback from social partners amplifies the emotional benefits of social sharing (Gable et al., 2004; Lambert et al., 2013) and the perceived quality of close relationships (Canevello & Crocker, 2010; Donato et al., 2014; Gable et al., 2004, 2006). Online, positive feedback is given through ‘likes,’ ‘upvotes,’ and comments from others, and has also been linked to adaptive socioemotional outcomes, including heightened perceptions of social connection and reduced feelings of loneliness (Sas et al., 2009; Seo et al., 2016). Like other positive outcomes, such as food and money, positive feedback (both in-person and online) engages the brain’s reward circuitry, leading to positive affect and social connection (Bhanji & Delgado, 2014; Fareri et al., 2012; Flores et al., 2018; Sherman et al., 2016, 2018) and social reward-seeking behaviors (Farmer et al., 2019; Hackel et al., 2020; Jones et al., 2011; Lindström et al., 2021; Sip et al., 2015).

The effect of feedback on social behavior is thought to occur through reinforcement learning (RL) mechanisms, through which prediction error signals in the brain’s reward centers guide moment-by-moment decision making (Behrens, Hunt, & Rushworth, 2009; Bhanji & Delgado, 2014; Jones et al., 2011; Ruff & Fehr, 2014). During a social interaction, people learn from and form expectations about a social partner based on the positive and negative feedback they receive from them, and this learning in turn guides future decisions and social behavior (Fareri, Chang, & Delgado, 2020). One way to investigate these RL mechanisms is through computational models that track predictors of moment-by-moment decision making in social contexts (Chang et al., 2010; Fareri et al., 2012, 2015; Jones et al., 2011; Lindström et al., 2021). For example, Fareri et al. (2012) employed a computational RL model that tracks the extent to which the valence of a social outcome (i.e., positive or negative) modulates participants’ overall expected value of future social outcomes (Loss-Gain model; Frank et al., 2007). One advantage of this model is that it accounts for differences in the rate at which people learn from positive versus negative feedback, thereby representing the extent to which people separately weigh this feedback when forming impressions about a social partner and making decisions about future social interactions. Positive social sharing may similarly rely on social feedback and RL mechanisms to guide sharing decisions, leading people to share more with social partners who are likely to provide positive feedback in return.

Feedback also plays an integral role in strengthening social bonds. When determining whether a social partner is trustworthy—that is, whether someone is likely to reciprocate positively in a social exchange—people generally draw on feedback from that partner to guide trust decisions (Brudner et al., 2021; Fareri, 2019). Positive feedback perpetuates trust through classic reinforcement learning mechanisms. For example, King-Casas et al. (2005) found that partner reciprocity in an economic exchange (Trust Game; Berg et al., 1995; Delgado et al., 2005) elicited prediction error signals in the brain that correlated with subsequent ‘intention to trust’ decisions. Computational models employed in previous work have also shown that positive feedback takes on a social reward ‘bonus’ in the brain when received from close others playing the Trust Game (Fareri et al., 2015) and elicits a ‘social tie’ signal during cooperative interactions, reflecting dynamic changes in social bonding during moment-by-moment feedback from novel partners (Bault et al., 2015, 2017). Feedback during initial social interactions can also influence later trust decisions. For example, people are more likely to trust a social partner after receiving honest vs. dishonest advice from them (Bellucci et al., 2019) or after being included vs. excluded by them in a group activity (Fareri et al., 2012). Receiving positive feedback during social sharing may therefore strengthen trust and social bonds among both familiar and novel social partners.

Here, we investigated whether positive feedback reinforces social sharing and enhances social bonding in a popular and ever-expanding setting—social media. To simulate naturalistic sharing conditions while also maintaining experimental control, we designed a sharing task inspired by paradigms that involve iterative feedback (e.g., Fareri et al., 2015; Will et al., 2017) and mimic real-world social media sharing environments (e.g., Baek et al., 2017; Sherman et al., 2016, 2018). We ultimately chose to simulate Instagram for this design because: 1) participants can provide many pictures, which maximizes statistical power and allows participants to form impressions of social partners through repeated interactions; 2) pictures shared on Instagram often capture personal life events and evoke positive emotions (De Paola et al., 2020; Vermeulen et al., 2018; Waterloo et al., 2018); and 3) it embodies a naturalistic sharing medium that is highly popular among emerging adults, which was the primary age range of our sample (Perrin & Anderson, 2019). In the sharing task, participants made trial-by-trial decisions about whether to share their positive Instagram pictures (and neutral control pictures) with three anonymous peers. Peers then provided positive and negative feedback on shared pictures through ‘likes’ and ‘dislikes,’ simulating the type of feedback one may encounter online. Unbeknownst to the participants, this feedback was experimentally manipulated to create three distinct feedback conditions. This design allowed us to capture changes in sharing over the course of the task as a function of how much positive and negative feedback participants received from each peer.

Our first pre-registered hypothesis was that, if feedback reinforces social sharing, then participants should share more with peers who provide more positive versus negative feedback. To further interrogate this hypothesis, we also employed an exploratory computational model adapted from the loss-gain model used in Fareri et al. (2012). This model tested the hypothesis that trial-by-trial sharing decisions are driven by reinforcement learning mechanisms in response to peer feedback, and that the value of positive and negative feedback may be weighted distinctly during this decision-making process. We also assessed whether positive feedback received during social sharing subsequently fosters social bonding by having participants complete a short Trust Game and closeness ratings following the sharing task.

The extent to which feedback affects positive social sharing behavior may vary at the individual level as well. Indeed, evidence suggests that people who more readily pursue interpersonal emotion regulation behaviors and believe that these behaviors are effective for improving their emotional wellbeing are more likely to exhibit affiliative behaviors and benefit from social support following an emotional event, as well as develop more supportive relationships within their social network (Williams et al., 2018). Given that positive social sharing is thought to be a response-dependent regulation strategy (Gable et al., 2004; Zaki & Williams, 2013), it is possible that reinforcement learning through feedback may be especially effective at promoting positive social sharing and social bonding in people who already engage in and perceive the efficacy of broader interpersonal emotion regulation strategies in their every-day lives. To examine this possibility, we added a parameter to our exploratory computational model that tracks the extent to which sensitivity to feedback interacts with individual differences in interpersonal emotion regulation to guide social sharing behavior.

Method

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study.

Participants

Participants were Rutgers University-Newark students recruited through the psychology subject pool to participate in a two-day study about college students’ social media use. Eligibility was determined based on whether they had an active Instagram account with a minimum of 40 previously posted photos (excluding videos, Stories, and Reels). A power analysis set to yield greater than 80% power determined that a sample size of 40 participants was sufficient to detect a significant within-subjects effect (p < .05) with moderate effect sizes based on pilot data (i.e., h² = .54, d = 1.10, r = −.47). To account for pre-registered exclusions, we collected data from 66 participants.

Participants were excluded if they (i) participated in Day 1 of the study but were deemed ineligible to participate in Day 2 due to insufficient or inappropriate photos based on instructed criteria (see Procedure: Day 1 below; n = 5); (ii) were lost to follow-up (n = 4); (iii) voluntarily withdrew from the study (n = 4); (iv) indicated prior to debrief that they did not believe the cover story (n = 3); (v) demonstrated misunderstanding of task instructions (n = 1); or (vi) were excluded due to experimenter error (n = 1). Four additional participants were excluded during data analysis because they made no response on at least half of trials per condition on any given run of the sharing task. Therefore, data from 43 participants were included in the final analyses (28 female, ages 18-32 y.o., M = 20.89 y.o., SD = 2.76). All subjects provided informed consent and were compensated with course credit and a $1.00 bonus payment. This study was approved by the Rutgers University Institutional Review Board.

Procedure

Day 1.

After providing informed consent, participants completed a battery of questionnaires on a lab computer. Participants were then asked to log in to their Instagram account and submit 35-40 screenshots of their personal photos, which would be used in the sharing task on Day 2. Participants were instructed to only provide photos that elicited positive emotions (e.g., should not include an ex-partner that they no longer have positive feelings towards), contained original content (e.g., should not be re-posts, ‘memes’, or quotes from other accounts), and did not contain any inappropriate or illegal content. Participants were also asked to rate each photo on how it made them feel on a Likert scale ranging from −4 (very negative) to 4 (very positive) and provide a brief caption for the photo (e.g., “On a vacation with my family”). All photos provided by participants were destroyed at the end of the session.

Questionnaires.

To explore whether effects of feedback sharing behavior and social bonding vary as a function of individual interpersonal emotion regulation, participants completed the 16-item Interpersonal Emotion Regulation Questionnaire (IERQ; Williams et al., 2018), which measures tendency to pursue, and perceptions of efficacy of, positive and negative interpersonal emotion regulation strategies. A list of additional individual differences measures and subsequent results can be found in the Supplemental Materials.

Stimulus selection.

Following completion of Day 1, one of two experimenters reviewed the photos from each participant and recorded the number of ‘likes’ each photo received, the number of people in each photo, and the date on which each photo was posted. Photos were then sorted based on the number of ‘likes’ they received (M = 182.13 likes, SD = 140.48) and the emotional rating given by the participant (M = 3.75, SD = 0.47). The top 24 most-liked and highest-rated photos from each participant were selected for the sharing task (see Supplemental Materials for more details). Finally, photos were cropped and arranged into a standardized display format with the caption provided by the participant.

In addition to the participants’ own photos, the task also included 12 unfamiliar, neutral photos taken from public Instagram accounts. These photos depicted relatively non-emotional social scenes, such as people having a conversation, reading, commuting, and sitting in a park. Searches for these photos on Instagram’s public feed included terms such as, “neutral,” “normal,” “boring,” and “people.” Neutral photos were used as a control condition to confirm whether sharing was driven in part by positive emotions. As a manipulation check, we observed that participants perceived their own pictures as more positive (M = 3.75, SD = 0.47) than the neutral pictures (M = 0.46, SD = 0.65), t(84) = −26.97, p < .001, 95% CI [−3.54, −3.06].

Day 2.

Participants returned to the lab between 1 and 16 days after their Day 1 session (M = 6.93 days, SD = 3.01 days). In this session, participants were told a cover story that they would be interacting in real-time with other participants who, for confidentiality reasons, were sitting in separate testing rooms in the lab. In reality, and unbeknownst to the participants, there were no other peers, and they were not in fact interacting with anyone else. After receiving the cover story, participants completed the sharing task followed by a trust game and rated how close they felt to each peer at the beginning and end of the experiment. Prior to debriefing, participants were asked to provide written answers to two open-ended questions to probe whether they picked up on the fictional nature of the cover story (“From your perspective, what did you think was the goal of this research?” and “Did anything stand out to you during this experiment?”). Participants were then debriefed and explicitly asked whether they believed the cover story. To avoid unnecessarily excluding participants whose responses to this explicit question may have been biased by the debriefing process, we only excluded participants from analyses if they expressed disbelief in the open-ended questions, prior to being debriefed.

Sharing task.

The goal of the sharing task was to examine whether feedback from peers influenced decisions to share positive events. At the beginning of the task, participants were virtually introduced to the three anonymous, same-sex peers. As part of the cover story, participants were informed that they and their peers would be represented throughout the task using a randomly assigned emoji tied to their name. This procedure was implemented to avoid biasing participants’ prior expectations about the peers due to i) potential confounding influences associated with using pictures of real people (e.g., race, gender, facial expression, etc.; Todorov et al., 2015); and ii) the possibility that participants would interpret each emoji as representing something meaningful about each peer. To confirm that there were no initial biases toward one peer over another based on these initial impressions, we asked participants to rate how close (defined to participants as how “socially and emotionally connected”) they felt to each peer using the Inclusion of Others in Self scale (IOS; Aron et al., 2001). As expected, there were no group-level differences in closeness ratings across the three peer conditions prior to the beginning of the task, F(2) = 0.04, p = .960.

Participants were told that they were randomly selected to play the role of the ‘sharer’ and that the other three peers would provide feedback on photos that they shared (Figure 1b). On each trial, participants were randomly presented one of their own positive photos (or an unfamiliar neutral photo) for 5s, followed by a jittered fixation cross for 1-3s. Then, participants chose whether to share that photo with one of the peers, whose name and emoji were displayed on the screen. Participants were instructed to press the left and right arrow keys within 3s to indicate whether they wanted to share or not share the photo with that peer. This choice was followed by a jittered waiting period of 6-8s, during which participants believed that the peer was providing feedback on their photo (if they shared) or that the computer was going through a waiting period (if they did not share). Participants then viewed feedback from the peer (in the form of a ‘thumbs-up’ for positive feedback or a ‘thumbs-down’ for negative feedback) or a neutral symbol indicating that no feedback was given (on no-share trials). Participants were instructed that a thumbs-up from a peer indicated that the peer would give them a ‘like’ on their photo if it were shared on Instagram, whereas a thumbs-down indicated that the peer would not give them a ‘like.’

Figure 1.

Timeline of experimental tasks. A) Prior to the start of the task, participants rated each peer on how close they felt to each of them using the Inclusion of Other in Self Scale (IOS; Aron et al., 2001). B) In the sharing task, participants were first presented a positive or neutral photo and then chose whether to share that photo with each peer. If they chose to share, then the participant received positive or negative feedback from that peer. Photo used for illustrative purposes only and was not submitted by participants. Photo obtained from Thought Catalog on Unsplash. C) Participants completed a trust game in which they chose whether to keep a smaller, safer endowment or share a larger, riskier investment with each peer. D) Participants once again rated how close they felt to each peer at the end of the session.

As mentioned, the peers in this task were fictional and all feedback was experimentally generated. This design allowed us to examine sharing behavior as a function of how much positive and negative feedback each peer provided, where one peer provided primarily positive feedback (Pos condition; M = 82.74%, SD = 0.77% positive feedback), one peer provided primarily negative feedback (Neg condition; M = 82.77%, SD = 0.92% negative feedback), and one peer provided approximately equal amounts of positive and negative feedback (Neu condition; M = 41.48%, SD = 1.26% positive feedback and M = 58.52%, SD = 1.26% negative feedback). We confirmed that positive feedback was given significantly more in the Pos condition vs. the Neu and Neg conditions (ps < .001) and negative feedback was given significantly more in the Neg condition vs. the Pos and Neu conditions (ps < .001). There were 108 trials across three task runs (36 trials per run) presented using PsychoPy2 (v.3.2.3; Pierce et al., 2019). Peer identity (name, emoji) and condition (Pos, Neu, Neg) were counterbalanced across participants and conditions were randomly and equally presented across all three runs. Participants also completed a six-trial practice run prior to the start of the task, which was not included in data analyses.

Trust game.

Following the sharing task, participants completed a short economic game (trust game; Berg et al., 1995; Delgado et al., 2005) with the three peers (Figure 1c). There were 12 trials in this task (4 trials per condition, one at each investment magnitude). On each trial, participants were assigned the role of the investor, where they were presented an initial monetary endowment ($1.00-$4.00) that they could choose to either keep for themselves or invest with one of the three other peers. If participants chose to invest, then the initial endowment would triple in value (e.g., $1.00 endowment would become $3.00) with the understanding that the peer would then get to choose whether to split the tripled investment with the participant (e.g., resulting in a total payout of $1.50 each) or keep the entire investment for themselves (leaving the participant with no monetary payout at all). Thus, choices to invest in this task served as a behavioral measure of how much participants trusted each of the peers, which was examined as a function of condition (Pos, Neu, Neg) and investment magnitude ($3-$12). Participants were not shown trial-by-trial outcomes, but instead were told that one trial would be selected at random, and the outcome of that trial would result in their final bonus payment. Thus, participants were encouraged to make trust decisions on each trial independently of all other trials.

The primary hypotheses for this experiment were pre-registered and can be accessed on our Open Science Framework (OSF) repository at this link. Due to logistical constraints, pre-registration of our hypotheses occurred after data collection began but before any data were analyzed. De-identified data and data analysis scripts, as well as supplemental materials, are also available in our repository at this link.

Results

Statistical analyses were conducted in R (v.4.0.3; R Core Team, 2020). Deidentified study data and analysis code, as well as additional supplementary materials, can be accessed through our Open Science Framework repository (Brudner et al., 2021).

Picture valence influences sharing behavior

Consistent with prior work (Gable et al., 2004; Lambert et al., 2013; Langston, 1994; Rimé, 2009), we hypothesized that if sharing is driven in part by the positive affect elicited by the event itself, then participants should share the positive pictures more than the neutral control pictures, regardless of feedback condition. A paired samples t test revealed that participants shared positive pictures significantly more often (M = 71.7% of trials, SD = 24%) than neutral pictures (M = 55.7% of trials, SD = 32%), t(84) = 2.62, p = .01, 95% CI [−0.28, −0.04], There were no sex differences in overall sharing behavior (M_males = 0.69, SD_males = 0.31, M_females = 0.73, SD_females = 0.30), t(235.39) = 1.29, p = .200, 95% CI [−0.023, 0.108].

Social feedback modulates sharing behavior

To test our main hypothesis that feedback reinforces positive social sharing, we examined share choices as a function of feedback condition across the three runs of the task. Given our primary interest in studying the effects of feedback on positive social sharing behavior and given that participants shared more of the positive pictures than the neutral pictures, we restricted our analysis to only trials in which participants shared positive pictures. However, we report a summary of results including all trials (positive and neutral) in the Supplemental Materials. Additionally, this analysis deviated slightly from our pre-registered plan, which proposed to analyze sharing only in the first and last run. After pre-registering, we felt that the exclusion of run 2 trials from analyses was inappropriate and that including all three runs would better capture true effects (108 vs. 72 trials). Results from the originally pre-registered analysis plan follow a similar (though marginal) pattern and can be found in the Supplemental Materials.

A 3 (condition: Pos, Neu, Neg) x 3 (run: 1, 2, 3) analysis of variance (ANOVA) on participants’ average proportion of share choices revealed a significant main effect of condition, F(2) = 4.01, p = .019, η² = 0.02, and a marginally significant main effect of run, F(1) = 3.24, p = .073, η² = 0.01, but no significant condition x run interaction, F(2) = 0.18, p = .838, η² = 0.001. Follow-up pairwise comparisons revealed that the main effect of condition was driven by a significant difference in sharing in the Pos vs. Neg condition (p_adj = .016; Figure 2a). All other comparisons were nonsignificant (ps > .2).

Figure 2.

Participants’ share choices were influenced by how much positive and negative feedback each peer provided. A) Over the course of the sharing task, participants chose to share more with the peer that provided the highest amount of positive feedback, and less with the peer that provided the highest amount of negative feedback. B) Peer feedback predicted trial-by-trial share decisions, above and beyond participants’ subjective picture feeling ratings.

We also sought to examine whether feedback modulated trial-by-trial sharing behavior, above and beyond the characteristics of the pictures themselves. A hierarchical logistic regression on trial-level sharing choices (including both positive and neutral trials) revealed that feedback, particularly positive feedback, significantly predicted share choices above and beyond individual picture valence ratings (Table 1, Figure 2b). These results remained consistent even when including only positive trials and after accounting for the number of likes each picture initially received and the number of people present in the photo (Supplemental Materials, Table S1). Together, these results provide strong evidence that feedback influences social sharing such that sharing is reinforced by positive feedback but discouraged by negative feedback above and beyond other event characteristics.

Table 1.

Feedback condition predicts trial-by-trial sharing (positive and neutral trials) above and beyond picture valence rating

Predictor	β	SE	z
	Model 1
Intercept	.16**	.05	3.01
Valence Rating	.20***	.02	11.94
	Model 2 (model deviance = 18.49, p < .001)
Intercept	.13◇	.07	1.82
Valence Rating	.20***	.02	11.96
Pos Feedback Condition	.22**	.08	2.72
Neg Feedback Condition	−.12	.08	−1.52

^◇p < .1,

^*p < .05,

^**p < .01,

^***p < .001

Effects of feedback on sharing and social bonding vary as a function of individual differences in trait interpersonal emotion regulation tendencies

We also explored whether individual differences in trait interpersonal emotion regulation tracked with the effects of feedback on social sharing in this task. Positive social sharing is a response-dependent interpersonal emotion regulation strategy, and it is possible that people who readily engage in and benefit from these types of strategies in everyday life may be more sensitive to feedback in this laboratory setting. To explore this possibility, we first conducted Pearson’s R correlations between sharing (Pos minus Neg condition) and participants’ individual IERQ scores and subscale scores, which revealed no significant relationships (rs < .10, ps > .500). These results suggest that, at the aggregate level, trait IER was not associated with overall differences in sharing behavior as a function of feedback.

Despite these null findings at the aggregate level, prior research suggests that moment-by-moment changes in social behavior may also be driven by reinforcement learning mechanisms that update the expected value of future feedback based on positive and negative social outcomes (Bhanji & Delgado, 2014; Fareri et al., 2012, 2020; Jones et al., 2011; Lindström et al., 2021). If this is the case, then perhaps individual differences in trait IER modulate the rate at which individuals learn from social feedback during positive social sharing. To test this hypothesis, we implemented an exploratory computational reinforcement learning model adapted from past work (Loss-Gain (L-G) model; Fareri et al., 2012; Frank et al., 2007) in which we aimed to predict trial-by-trial likelihood of sharing as a function of participants’ expected value (EV) of potential feedback from each peer (see Supplemental Materials for model specifications). Participants’ EV was updated on each trial using a Rescorla-Wagner prediction error update rule; importantly, we estimated separate learning rates for experiences of positive (i.e., gain) and negative (i.e., loss) feedback. The updated EV was then fed into a softmax function to estimate the likelihood of sharing on each trial.

We also explored whether individual differences in every-day use of and benefit from interpersonal emotion regulation strategies interacted with feedback to drive trial-by-trial sharing in the task. We constructed and tested an additional novel model that was identical to the L-G model but weighted each participant’s initial expected probability values based on their IERQ score. This model (L-G + IERQ) measured whether the tendency to share in real life tracked with the motivation for positive feedback in this laboratory sharing environment. If seeking out positive feedback does indeed drive social sharing behavior (above and beyond initial positive affect), in both the current task as well as real life, then the L-G + IERQ model should best predict trial-by-trial sharing over the course of the task.

We compared the L-G and L-G + IERQ models to two other, pared-down models: one that only calculated EV but did not allow for trial-by-trial learning (EV model), and one that incorporated EV and the Rescorla-Wagner prediction error rule (Standard RL model) but did not consider type of feedback or individual differences. One participant was excluded from these analyses for demonstrating low rates of sharing overall (less than 10%, n = 1), as they likely did not receive enough feedback to properly learn from these outcomes. Thus, the final models included data from 42 participants.

Wilcoxon signed-rank tests demonstrated that the L-G + IERQ model fit the data significantly better than the L-G Model (z = −2.77, p = .006), the Standard RL model, (z = −5.12, p < .001), and the EV model, (z = −5.25, p < .001) (Figure 3). Learning rates from the L-G + IERQ model suggest that participants iteratively updated their beliefs as a function of trial-by-trial feedback, weighting positive (a = 0.83) relative to negative feedback (a = 0.06) more heavily when adapting their decisions to share photos with each peer (Table 2). While exploratory, these findings provide further evidence that positive feedback serves as a rewarding motivator for social sharing, especially in individuals who more readily engage in and benefit from social sharing outside of the laboratory setting.

Figure 3.

Comparison of model fits for each computational model. The L-G + IERQ model was significantly better fitting of the data (as measured by lower AIC) than the L-G model, the Standard RL model, and the EV model.

Table 2.

Model parameters

Model	α(SE)		β(SE)	φ(SE)	AIC(SE)
Expected Value			0.95(0.03)		165.47(4.35)°
Standard RL	0.09(0.04)		0.96(0.02)		165.42(4.49)°
	Positive Feedback	Negative Feedback
L-G	0.90(0.04)	0.06(0.03)	0.91(0.03)		145.02(2.57)°
L-G + IERQ	0.83(0.06)	0.06(0.03)	0.65(0.06)	2.32(0.11)	144.53(2.44)**

L-G = Loss-Gain Model, L-G + IERQ = Loss-Gain model with initialized parameter for Interpersonal Emotion Regulation Questionnaire score.

^*p < .01,

^°comparison model.

Social feedback during positive social sharing influences trust decisions

We also hypothesized that feedback during the sharing task would generalize to trust decisions in a separate economic exchange. A one-way ANOVA revealed a significant effect of condition on trust decisions, F(2) = 13.18, p < .001, η² = 0.17 (Figure 4a). A follow-up Tukey’s HSD test revealed that this effect was driven by greater trust decisions in the Pos condition compared to the Neg (p_adj < .001) and Neu (p_adj = .006) conditions. These results remained significant even after including investment magnitude as a covariate, F(2) = 25.93, p < .001, η = 0.09.

Figure 4.

Peer feedback during social sharing influences trust decisions. A) In a subsequent trust game, participants were more trusting of the peer that provided the highest amount of positive feedback during the sharing task. B) The effects of feedback (Pos minus Neg peer condition) on sharing photos positively correlated with trust decisions in a subsequent trust game (Pos minus Neg peer condition).

Additionally, we examined whether the influence of feedback on sharing corresponded to the influence of feedback on trust at the individual level. A Pearson’s correlation revealed a significant positive association between sharing and trust decisions, r(41) = .51, p = .004, suggesting that participants who were more willing to share in the Pos vs. Neg condition were subsequently more trusting of the Pos vs. Neg peer (Figure 4b).

Finally, we also investigated whether the influence of feedback on trust varied as a function of individual differences in interpersonal emotion regulation. A Pearson’s correlation revealed a significant positive relationship between trust decisions and IERQ total scores, r(41) = .30, p = .050, suggesting that individuals who more readily engage in and benefit from interpersonal emotion regulation strategies in everyday life were more sensitive to feedback, in that they were more trusting of the peer in the Pos compared to Neg condition following the sharing task.

Social feedback during positive social sharing influences feelings of interpersonal closeness

To test the hypothesis that positive feedback during social sharing strengthens interpersonal closeness, we examined changes in self-reported closeness with each peer between the beginning and end of the study. A 3 (condition: Pos, Neu, Neg) x 2 (time: pre-task, post-task) ANOVA revealed significant main effects of condition, F(2) = 12.02, p < .001, η² = 0.09, and time, F(1) = 17.16, p < .001, η² = 0.06, as well as a significant condition x time interaction, F(2) = 13.63, p < .001, η² = .10 (Figure 5a). Follow-up pairwise comparisons revealed that the interaction was driven by an increase in closeness ratings over time in the Pos condition (p_adj < .001), but not the other two conditions (ps_adj > .3). As reported earlier, closeness ratings did not differ across the three conditions at the beginning of the task (ps_adj > .9). At the end of the task, however, closeness ratings were significantly greater in the Pos condition compared to the Neg (p_adj < .001) and Neu (p_adj < .001) conditions. There was no difference in post-task closeness ratings between the Neg and Neu conditions (p_adj = .363).

Figure 5.

Peer feedback during social sharing influences interpersonal closeness. A) Prior to the sharing task, participants reported equally close to all the peers. However, by the end of the study, subjective feelings of social closeness increased for the peer who provided the highest amounts of positive feedback. B) The effect of feedback on trust was positively correlated with post-task closeness ratings (Pos minus Neg condition).

We examined whether the influence of feedback on sharing and trust corresponded to the influence of feedback on closeness at the individual level. Pearson’s correlations revealed a marginal positive association between sharing and post-task closeness, r(41) = .28, p = .071, and a significant positive association between trust and post-task closeness, r(41) = .50, p < .001 (Figure 5b). This analysis excluded one outlier due to a Pos minus Neg closeness difference score more than 3 standard deviations away from the group mean. Including this outlier does not change the results. Together, these findings suggest that participants who were more sensitive to feedback during the sharing task and subsequently during the trust game, felt closer to peers who provided more positive vs. negative feedback at the end of the task.

Finally, we also investigated whether the influence of feedback on closeness varied as a function of individual differences in interpersonal emotion regulation. A Pearson’s correlation revealed no significant relationship between closeness and IERQ total scores, r(41) = .06, p = .719.

The influence of feedback on sharing indirectly affects post-task interpersonal closeness through trust formation

Given the aforementioned correlations, we explored whether there was a cumulative relationship between these three variables that would more broadly explain the influence of feedback on interpersonal bonding during social sharing. Specifically, we tested whether there was an indirect relationship between feedback-related sharing and post-task closeness through trust using a bootstrapping procedure in the ‘mediation’ package in R (Hayes, 2009; Tingly et al., 2014). The initial total effect between sharing and closeness was marginal (β = 2.49, p = .079). The coefficient between sharing and trust was significant (β = .98, p < .001), as was the coefficient between trust and closeness (β = 2.44, p = .004). Including trust in the model further weakened the direct effect of sharing on closeness (β = 0.12, p = .714). A bootstrapping analysis with 1,000 simulations confirmed that the indirect effect between sharing and closeness through trust was statistically significant (β = 2.38, 95% CI [0.64, 5.83], p < .001; Figure 6). These findings highlight how feedback received during positive social sharing indirectly affected participants’ subjective feelings of closeness toward the peers through the formation of trust.

Fig. 6.

Indirect effects model for feedback-related sharing, trust, and closeness. Sharing in the positive vs. negative condition was indirectly related to post-task ratings of interpersonal closeness through differences in trust decisions in the positive vs. negative condition.

Discussion

We investigated the factors that drive positive social sharing behavior and social bonding. Using a novel social media sharing paradigm, we demonstrated that feedback from anonymous peers influenced sharing behavior such that participants shared more often with peers who provided higher rates of positive versus negative feedback. An exploratory computational model suggested that participants’ decisions to share were best predicted by a combination of trial-by-trial feedback and individual differences in their engagement in interpersonal emotion regulation in everyday life. These effects of feedback on sharing also facilitated subsequent social bonding: participants were more trusting of and felt closer to peers who provided more positive versus negative feedback during the sharing task. An indirect-effects analysis highlighted how individual differences in sensitivity to peer feedback during social sharing were indirectly related to feelings of interpersonal closeness by way of trust. Together, these findings suggest that feedback received during positive social sharing on social media reinforces sharing behavior and fosters social bonds through mechanisms such as trust and closeness.

Our results support and extend a growing body of work investigating the relationship between positive emotions, social sharing, and feedback. Consistent with prior evidence that sharing behavior is driven in part by positive affect of the event itself (Gable et al., 2004; Lambert et al., 2013; Langston, 1994; Rimé, 2009), we found that participants’ subjective feelings toward the photos predicted sharing choices. Above and beyond this positive affect however, participants’ share choices were also contingent upon the amount of positive and negative feedback they received from each peer. These findings extend prior work in which constructive feedback (a form of feedback that reflects a positive evaluation and understanding from a social partner) influences socioemotional outcomes during sharing interactions (Gable et al., 2004, 2006; Lambert et al., 2013); and in which feedback influences decision making in other social contexts (Fareri et al., 2012).

Our findings also contribute to existing literature on motivation for sharing in online social settings. Specifically, we provide further evidence that sharing on social media involves self-related, value-based, and socially motivated decision-making processes (Baek et al., 2017; Ihm & Kim, 2018; Scholz et al., 2019, 2020; Vermeulen et al., 2018) and that social-media feedback has rewarding and reinforcing properties (Lindström et al., 2021; Sherman et al., 2016; 2018). Uniquely, our study presents these novel findings through direct manipulation of social feedback and measurement of real-time behavior, strengthening the evidence for these mechanistic processes and setting the foundation for further research in this domain.

Our study also provides new insights into how positive feedback during social sharing can foster social bonding. Here, we characterized social bonding following the sharing task as: i) how much the participants trusted each of the peers, as measured by the extent to which they chose to gamble with each peer during the Trust Game, and ii) how close participants felt toward each peer, as measured by subjective ratings. That participants were more trusting of and felt closer to peers who provided more positive versus negative feedback echoes prior evidence of greater relationship satisfaction, intimacy, and trust when positive social sharing is followed by constructive feedback (Gable et al., 2004, 2006). These findings also align with emerging evidence that interactions on social media, when met quickly with large amounts of positive feedback, fuel social bonding by generating feelings of social support and strengthening relationships that began offline (Carr et al., 2016; Kalpidou et al., 2011; Marengo et al., 2021; Neubaum & Krämer, 2015; Pouwels et al., 2021; Rousseau et al., 2019; Seo et al., 2016). Our work therefore contributes to a broader theoretical discussion surrounding the reciprocal associations between sharing, positive feedback, trust, closeness, and socioemotional well-being (Bault et al., 2015, 2017; Fareri et al., 2012, 2015; Peters et al., 2018; Poore et al., 2012; Rimé et al., 2020).

The use of an exploratory computational model further revealed that individual differences in social sharing behaviors in real life tracked with how participants learned from trial-by-trial feedback. Computational modeling has long been used to capture moment-by-moment changes in reward-seeking behavior as a function of moment-by-moment positive and negative outcomes (McClure et al., 2003; O’Doherty et al., 2003; Samson, Frank, & Fellous, 2010). In social contexts, prior research has demonstrated that reinforcement learning mechanisms modulate social behavior in response to positive and negative feedback (Bhanji & Delgado, 2014; Fareri et al., 2012, 2020; Jones et al., 2011; Lindström et al., 2021). Here, we found that adding IERQ scores improved the fit of a reinforcement learning model that separately weighed the influence of positive and negative feedback on prediction error calculations (L-G Model; Fareri et al., 2012), suggesting that social sharing tendencies outside of the laboratory may predict sensitivity to social feedback and in turn reinforce sharing behaviors. These findings align with and extend prior work highlighting the critical role of reinforcement learning in social contexts (Fareri et al., 2015; Farmer et al., 2019; Hackel et al., 2020; Jones et al., 2011; Kaczmarek et al., 2021; Lindström et al., 2021; Sip et al., 2015).

Unlike past research, which has primarily studied social sharing in the context of close relationships (e.g., Gable et al., 2004; Kaczmarek et al., 2021; Lambert et al., 2013), our study focuses on feedback from unfamiliar peers (but see Reis et al., 2010). Examining sharing and feedback dynamics among close social partners is certainly valuable given that people frequently interact with and receive psychological benefits from strong social ties (Campos et al., 2014; Gable et al., 2004). That said, the value of positive interactions with unfamiliar or weaker social ties should not be underestimated. Indeed, interactions with weak social ties capture attentional resources (Weng et al., 2017), have benefits for social capital and well-being (DeAndrea et al., 2012; Granovetter, 2021; Liu & Yeo, 2021; Sandstrom & Dunn, 2014; Sutcliffe et al., 2018; Tilston & Sandstrom, 2018; Walsworth et al., 2021), and serve as the impetus for developing closer relationships both in-person and online (Ostertag & Ortiz, 2017; Standlee, 2019). Broader and more public sharing and subsequent positive feedback may also strengthen weak social ties and form the foundation for deeper relationships in private spaces (Sas et al., 2009). Our findings in the context of this past research therefore suggest that positive feedback from novel peers during social sharing can have a substantial impact on social relationships and well-being, with important implications for promoting adaptive social behaviors and mitigating loneliness (Hawkley & Cacioppo, 2010; Zaki & Williams, 2013).

We recognize that our choice of stimuli and study design may limit the extent to which we can generalize our findings to other forms of sharing and feedback, including in-person sharing and comments versus likes. Yet, prior work suggests that, despite material differences in communication tools, emotional content is similarly shared in both on- and offline social settings (Derks et al., 2008), and social feedback typically acquired in-person (e.g., encouraging words; Smith, Sip, & Delgado, 2014) or online (e.g., social media ‘likes’; Sherman et al., 2016, 2018) activates overlapping reward networks in the brain. Thus, while research in this domain has historically focused on in-person, dyadic social sharing interactions with feedback from close partners (e.g., Gable et al., 2004; Langston, 1994), our work aligns with and adds to this narrative by highlighting an online form of communication that has become inherent to our global social fabric. Additionally, our use of a controlled experimental design and real-time behavioral measurements answers recent calls for further refined methodological approaches to studying social sharing (Rimé et al., 2020) and contributes to a literature that has largely, until now, relied upon self-reports, daily diaries, and retrospective behavioral observations.

Despite the promising implications of our study, we acknowledge that not all instances of social sharing and peer feedback, particularly on social media, has positive outcomes. For example, positive feedback reinforces moral outrage on social media, and, when combined with feedback-sensitive design algorithms, may distort social and political discourse in these public spaces (Brady et al., 2021). In vulnerable populations, such as adolescents, social media use and positive feedback may normalize and promote risky behaviors such as alcohol or cigarette use (Hendriks et al., 2018; Mohr et al., 2016; Sherman et al., 2016). Additionally, people with depression, social anxiety, low self-esteem, and certain attachment styles have all shown aberrant behavioral and psychological outcomes related to social sharing, feedback, and social media use (Blackwell et al., 2017; Forest & Wood, 2012; Kashdan et al., 2013; Sanchez et al., 2017; Shallcross et al., 2011; Smith & Reis, 2012). Thus, further work is needed to disentangle in what contexts and populations positive social sharing might be most beneficial, and to identify ways to promote adaptive social sharing in individuals who may not otherwise readily do so.

This study lays the groundwork for future investigations into the underlying mechanisms of positive social sharing and their role in promoting social bonding. While we offer preliminary insights into variations in who may be more likely to share and be more sensitive to social feedback, our findings are limited by sample size and demographics (young adults, primarily women) and would benefit from replications with larger, more representative samples. Additionally, while our paradigm was designed to balance naturalistic sharing with experimental control, future work is needed to confirm whether our findings generalize outside of the laboratory (e.g., Lindström et al., 2021; Wiese et al., 2011). Nevertheless, this work provides critical insights into the motivational factors that promote positive social sharing and contributes to a growing literature the psychosocial benefits of interpersonal emotion regulation.