People Copy Success More Than Failure in Social Learning

Abstract

Social learning is a fundamental mechanism for efficiently transferring and coordinating norms, skills, and sophisticated cultural information to individuals. However, the psychological mechanisms underlying social learning remain unclear. To investigate this, we recruited adult participants (N = 103), who observed a model’s performance in a two-choice reward-searching task. Two cues were used to determine the reward, with both cues possessing an alternative signal that had a specific rule for finding the reward. Although the model succeeded with one cue but failed with another, both possessed equal information, which enabled the participants to find the reward. Participants were more likely to use the cue linked to the model’s success than the model’s failure when asked to solve the task by themselves. This “copy success” bias reflects the psychological process underlying social learning suggested by previous evolutionary theories and provides practical suggestions for efficient learning processes.

Introduction

Imagine you have relocated to a new community lacking prior knowledge. You are unfamiliar with the language spoken by the community, their daily dietary habits, and how they interact. To effectively integrate into this new environment, an efficient strategy would be to observe the daily behaviors of community members, gather knowledge and skills from these observations, and incorporate cultural norms into their behaviors. Learning from others (social learning) emerges early in human development (D’Entremont et al., 1997; Meltzoff & Moore, 1977) and has been considered an essential building block for efficiently transferring and coordinating norms, skills, and sophisticated cultural information among individuals (Bandura, 1971; Hecht et al., 2013; K. Laland & Janik, 2006; Rendell et al., 2011).

An intriguing question that has attracted researchers’ attention is how people learn from others. Consistent with the theoretical assumption that natural selection favors adaptive social learning strategies (Boyd & Richerson, 1988; Henrich & McElreath, 2003; K. N. Laland, 2004), studies have found that people do not learn information from others indiscriminately. Instead, people acquire more knowledge from specific types of information compared to others. For instance, people selectively learn information based on how common it is (the “conformity rules”; Kendal et al., 2009; Wakano & Aoki, 2007). People acquire social information by considering their own pay-off, the pay-off to demonstrators, or the difference between the two (the “pay-off-based rules”; Schlag, 1998, 1999). People selectively copy previously higher-performing demonstrators than previously poorly-performing demonstrators, irrespective of their own performance (the “copy-successful-individual rule”; Mesoudi, 2008; Mesoudi & O’Brien, 2008; Nakawake & Kobayashi, 2022). These findings shed some light on the subject; however, additional research is needed to fully comprehend social learning mechanisms, as the process varies greatly depending on the context.

This study investigated people’s social learning strategies from a new perspective that categorizes the consequences of other people’s behaviors as “success” and “failure”—a behavioral consequence-based strategy (rather than the individual-based strategy proposed by the “copy-successful-individual rule”). Success is often characterized by the accomplishment of a desired outcome, such as correctly answering a question or choosing a box containing a reward, while failure is defined as the opposite. Successes provide information on effective strategies and techniques that can be adapted for future tasks (Andrieux & Proteau, 2016; Bandura, 1977; Scully & Newell, 1985). Conversely, failures offer information regarding potential pitfalls and mistakes that should be avoided in future tasks (Ali, 1981; Alves et al., 2017; Bledow et al., 2017; Eskreis-Winkler & Fishbach, 2022). As both the successes and failures of others contain valuable information, the optimal strategy for individuals is to learn effectively from both. Moreover, as learning from the failures of others does not require the observer’s personal experience of the failures, making it a risk-free process (Haruki, 1977), adopting this strategy is a reasonable approach. However, daily examples and theoretical reasoning suggest that it is more difficult for individuals to learn from the failures of others compared to learning from their successes.

According to Eskreis-Winkler and Fishbach (2020), on YouTube, there are approximately two videos with titles including the word “success” (∼25 million) for every one video with titles including “failure” (∼10.9 million). Moreover, “contrary to the common belief that newspapers sell negative news, since 1851, the New York Times has published twice as many articles about ‘success’ (∼596,000) versus ‘failure’ (∼370,000)” (Eskreis-Winkler & Fishbach, 2020). This implies that people are more likely to pay attention to and be willing to gain information from the successes of others than their failures (Chou & Edge, 2012; O’Brien et al., 2018). One reason may be that the successes of others provide more valuable information than their failures, leading to people’s preference for learning from the former. For instance, by observing the correct method of solving a math problem, one can gain valuable insights to solve the problem independently. However, if one observes someone using an incorrect method to solve the problem, it may not necessarily be helpful in finding a solution. The successes of others highlight approaches that lead directly to desired ends, whereas others’ failures are more oblique. This is because failures provide information on what we should avoid rather than what we need to do to succeed. In other words, failures alone are not sufficient to achieve the desired ends; it would be more efficient to choose to learn from the successes of others than from their failures.

Another reason might be that people tend to have a bias toward learning more from the successes of others than from their failures, even when they have gained equally valuable information for achieving goals from both experiences. This hypothesis gains support when considering that while the failures of others might contribute to a punitive role, providing information to a person that the behavior eliciting undesirable outcomes should be avoided, the success of others might function as a reward, reinforcing the learning of that behavior (Cheyne, 1972; Hillix & Marx, 1960; Thorndike, 1932).

However, no study has directly investigated whether people learn more from others’ successes than failures. Understanding such behavioral bias would be the first step in elucidating the underlying psychological mechanism of social learning and providing practical suggestions to achieve ideally efficient learning processes. Subsequently, in the current study, we developed a new experimental paradigm to investigate whether people are more likely to employ the cues associated with the successes of others than their failures, even when they gain sufficient information for achieving goals from both experiences. Specifically, participants observed the model’s performance on a two-choice reward-searching task. Two cues were used to determine the reward, and the model used different cues in different trials. Both cues possessed an alternative signal with a specific rule for finding the reward. The model succeeded with one cue but failed with another. Note that even if the cue was linked to the model’s failure, participants could use inclusive logic to understand the use of the signals and further find the reward. We investigated whether the participants favored one cue over another when asked to solve the task themselves.

This paradigm also helps evaluate the validity of the two possible reasons mentioned above. Given that both the success and failure of the model possessed sufficient information to enable the participants’ success, if people are more likely to learn from the successes of others than their failures because the successes of others possess more valuable information, then, in the current experiment, the participants would not have biases toward learning from others’ successes and failures. Alternatively, if the reason was not relevant to the asymmetric information of others’ success and failure but to the form of the demonstration (e.g., reinforcement of observing the success of others), then, even if participants understand the use of both cues, they are more likely to employ the cue associated with the success of the model than the one associated with the failure.

Method

Transparency and Openness

The methods of the current study, such as sample size, stopping rule, and experimental procedure, were registered prior to the experiment at https://doi.org/10.17605/OSF.IO/S57EP. All data are available at https://doi.org/10.6084/m9.figshare.22303588. Data were analyzed using R, version 4.1.2 (R Core Team, 2021).

Participants

The final sample included 103 adults (32 women, 70 men, and one other sex; M_age = 49.81 years, SD = 14.40, range = 18–76 years). A power analysis before data collection demonstrated that a sample size > 90 was required to test whether the number of participants making a specific judgment was different from a null hypothesized proportion of 50% by a binomial test (α = .05, 1−β = .80, expecting that 65% of participants in our sample would make a specific judgment. Specifically, our subjective expectations predicted that 65% of participants would choose to adopt cues linked to success; Meng et al., 2022). The experiment was stopped after we were confident that the sample size exceeded the required number.

An additional 118 adults (50 women, 67 men, and one other sex, M_age = 51.76 years, SD = 14.87, range = 18–87) participated in the study but were excluded from the final sample because they gave at least one incorrect answer to the follow-up questions. The sample was then included in an exploratory analysis.

The participants were recruited via a major Japanese Internet survey company (Cross Marketing Inc., Tokyo, Japan; The remuneration amounts to approximately 50 Japanese yen). Before accessing the survey, the participants read an informed consent statement stating that participation was voluntary and could stop at any time (Figure 1). By clicking an “agree” button, participants were informed that they were providing consent to complete the survey. This study was approved by the Ethics Committee of Osaka University (No. R4-057) and conducted following the ethical standards of the Japanese Psychological Association.

Figure 1.

Flowchart outlining the process for participants to respond to the survey.

Task

Overview

Using an online survey, participants watched videos and answered questions. Specifically, the participants watched videos in which a person acted to achieve a task goal by applying different approaches; one approach consequently led to successful goal achievement, and the other consequently led to unsuccessful goal achievement (learning phase; Figure 2). The participants were asked to perform the same task. Whether the participants adopted the model’s “successful approach” or “unsuccessful approach” was investigated (testing phase). Finally, we tested whether participants understood the use of these approaches (follow-up). The detailed stimuli and procedures are described below.

Figure 2.

Structure of the experiment.

Learning Phase

The learning sequence included eight video clips, each with a duration of 12 s. Each video started with one male model presented on the screen, which was visible from the hip to the shoulder (the face of the model was not included in the stimuli). Two opaque white boxes were placed equidistant from the front of the model. Each box had two buttons of different colors (i.e., blue or gray) on different sides (i.e., left side and right side) of the box (e.g., a blue button on the right side and a gray button on the left side; counterbalanced between participants). The two boxes had the same buttons, with four buttons in total. For two buttons of the same color (but on different boxes), only one glowed when pressed. For instance, if the blue button on the right box lit up, the blue button on the left would not. A red ball was placed in each box as a reward for the task.

In each video, the model chose one box from two to obtain a ball. First, the model pushed two buttons in one color (only one of the buttons was illuminated). Then, the model chose to approach either the box with the lighting-up button or the box without the lighting-up button to try to get the ball. Consequently, the model got or missed the ball (corresponding sounds such as positive beeps and buzzing sounds also emphasized whether the model successfully achieved the goal).

Two types of events (videos) were presented. In the “successful” event, the model consequently got the ball by choosing the box that contained it. In the “unsuccessful” event, the model consequently missed the ball by choosing the empty box. The colors of the buttons pushed by the model differed between the “successful” and the “unsuccessful” events. The judgments of the model (i.e., whether to choose the lighting-up box or the no-lighting-up box) also differed between the two types of events.

Four successive videos of the learning sequence (i.e., the first four videos or the last four videos) repeatedly demonstrated a “successful” event in which the position of the buttons was counterbalanced across the videos. The other four successive videos of the learning sequence (i.e., the last or first four videos) repeatedly illustrated an “unsuccessful” event in which the position of the buttons was counterbalanced across the videos.

To counterbalance the factors such as the order of the events (i.e., whether the “successful” or the “unsuccessful” events were presented first) and the combinations of the events and the color of the buttons, we prepared eight learning sequences. Each participant was randomly assigned to watch the learning sequence.

At the beginning of the learning phase, the text instructions explained what the model tried to do in the video. The participants were then asked to learn from the model’s performance by watching the learning sequence. They could watch the learning sequence as many times as they wanted. After the participants finished the step, a second text instructed them that they would be asked to perform the same task later. Before that, they had one more chance to learn from the model’s performance by watching the learning sequence again. The learning sequence was then presented again, and the participants watched it multiple times.

Test

The participants were asked to complete the task they watched during the learning phase (the task was shown in the pictures). First, as the cue for choosing a box to approach, they were asked whether they would press the blue buttons or the gray buttons (color-choice question; note that each color corresponded to the “successful” or the “unsuccessful” event presented in the learning phase). To make the task sequence natural, the participants were then asked to answer in which box they thought they could or could not find the ball if the button on the right (or left; counterbalanced across participants) box lit up when pressed (prediction).

It is possible that participants’ judgments merely mirror a potential association between the models’ task outcome and the participants’ desirable task outcome. For instance, participants who aim to find the ball may adopt the model’s “successful” approach. Therefore, we conducted two types of tests wherein participants were asked to choose the box they believed the ball was in or not in. Each participant was randomly assigned to one of the test types.

Follow-up

Finally, two questions tested whether the participants understood the relationship between the responses of the buttons (i.e., whether it lit up/did not light up) and the outcomes (i.e., the position of the box in which they could find the ball). For each question, the participants were asked to answer whether they believed they could find the ball in a specific box (i.e., the box on the right or left side) if a specific button (i.e., the button on the right or left box) lit when they pressed the two buttons in that color. The button colors mentioned in the two questions differed; the color corresponding to the “successful” event was used in one question, and the color corresponding to the “unsuccessful” event was used in the other question. Factors such as the order of the questions (i.e., whether the blue button or the gray button was mentioned in the first or second question) and the position of the target box (i.e., whether the participants needed to predict the right box or the left box) were counterbalanced across the participants.

Analysis

To address our main question of whether the participants were more likely to adopt the approaches that led to others’ successes/failures than the others, we tested whether the mean of the color-choice question was statistically different from chance (i.e., 0.5) by applying a binomial test. Because each color corresponded to either the “successful” or the “unsuccessful” event presented in the learning phase, the answer to the color-choice question was coded as “successful color” (as 1) or “unsuccessful color” (as 0).

Furthermore, according to the registered analysis plan, we conducted two exploratory analyses, including the 103 final samples and 118 participants excluded from the analysis (see Participants). The prediction questions were only used to make the task sequence natural; therefore, analysis regarding this question was not planned in our pre-registration.

Results

Main Analysis With the Final Sample

For the color-choice question, the participants were more likely to choose to press the buttons corresponding to the “successful” event they watched previously (66 participants; 64.1%) than the buttons corresponding to the “unsuccessful” event (37 participants; 35.9%; binomial test p = .006). Post-hoc generalized linear model approaches were applied to test whether participants’ judgments were influenced by their gender and age, the type of test, the order of the success/failure demonstrations, or the pattern of the surveys. These were set as fixed factors, and the model was set with a binomial error and logit link functions (1 = “successful color,” 0 = “unsuccessful color”). No significant effects were found for any factor (ps > .06).

Exploratory Analyses

For the registered exploratory analyses, first, we investigated whether the participants adopted the “successful” approaches of the model as a priority, irrespective of whether or not they understood the relationships between the responses of the buttons and the outcomes (i.e., the “rules of the task”). The results demonstrated that among the 221 participants, 130 (58.8%) chose to press the buttons corresponding to the “successful” event they had watched previously, and 91 (41.2%) chose to press the buttons corresponding to the “unsuccessful” event; there was a significant difference between them (binomial test p = .010).

Second, we investigated whether the understanding of the relationships between the responses of the buttons and the outcomes differentiated between the “successful” events and the “unsuccessful” events. In other words, whether participants gained information more efficiently from the “successful” events than from the “unsuccessful” events. The results demonstrated that among the 221 participants, 169 (76.5%) and 126 (57.0%) gave correct answers to the follow-up questions of the “successful” events and the “unsuccessful” events, respectively (both numbers included the participants who answered both questions correctly). The binomial test revealed a significant difference between the numbers (p = .014). To further investigate whether learning efficiency influences the approaches the participants adopted, we examined the participants who gave correct answers only to either of the follow-up questions (Table 1). A Chi-squared test demonstrated that the approaches the participants adopted did not relate to which type of follow-up questions they answered correctly (χ²(1) = 0.86, p = .098).

Table 1.

Participant Distribution Based on Correct Answers and Adopted Approaches in Follow-Up Questions.

	Adopted “successful” approaches	Adopted “unsuccessful” approaches
Correctly answered only the “Successful” follow-up questions	38 (42.7%)	28 (31.5%)
Correctly answered only the “Unsuccessful” follow-up questions	10 (11.2%)	13 (14.6%)

Finally, although the generalized linear model demonstrated the non-significant effect of the type of test on participants’ judgments, we additionally confirmed the results using Chi-square test (χ²(1) = 2.95, p = .086; Table 2).

Table 2.

Number (Percentage) of Participants by Task Type and Approaches Adopted.

	Adopted “successful” approaches	Adopted “unsuccessful” approaches
Task with goal to find the ball	38 (36.9%)	14 (13.6%)
Task with goal to miss the ball	28 (27.2%)	23 (22.3%)

Discussion

Social learning contributes significantly to human culture (Boyd & Richerson, 2005). Although many theoretical arguments and empirical investigations have demonstrated that humans are experts in social learning from a very early stage of ontogeny and evolution (D’Entremont et al., 1997; K. Laland & Janik, 2006), more investigations are needed to draw a fuller picture of how people learn from others. The current experiment revealed a “copy success” bias in which people are more likely to employ cues associated with the successes of others than their failures, and this bias occurred even when people gained sufficient information for achieving goals from both experiences. In the experiment, the participants observed a model successfully achieving a goal using one of the two cues and failed to achieve the goal using the other cue. The participants were then asked to perform a task in which they could use either cue. The results demonstrated that, although participants comprehended the use of both cues (i.e., understood how to achieve the goal by using both cues), they were more likely to choose the cue that the model previously used in successful performances than the cue that the model previously used in unsuccessful performances.

Although participants comprehended both cues, they demonstrated a greater inclination toward the one associated with the successful outcome of the model, as opposed to the cue associated with the model’s failure. This finding provides evidence supporting the notion that people’s tendency to learn more from others’ successes than their failures is not solely attributed to the fact that people gain more valuable information from others’ success than failure. Rather, it is influenced by the form of the demonstrations. It is possible that social learning tends to prioritize learning from the successes of others over their failures because the latter primarily function as a mechanism of punishment, providing information that certain behaviors should be avoided to prevent undesirable outcomes. Conversely, the success of others may serve as a form of reward that reinforces the learning of specific behaviors (Cheyne, 1972; Hillix & Marx, 1960; Thorndike, 1932). It is also plausible that individuals have become accustomed to the frequent phenomenon whereby the successes of others offer more valuable insights than their failures. As a result, people have developed an automatic tendency to prioritize acquiring knowledge from the successes of others while disregarding the potential gains from analyzing their failures.

A previous study using computer-based tasks demonstrated that participants selectively copied higher-performing demonstrators rather than poorly-performing demonstrators, irrespective of the demonstrators’ performance (Morgan et al., 2011). Moreover, simulation and experiments of cultural evolution have found that this kind of “copy-successful-individuals” strategy is significantly more adaptive than individual learning (Mesoudi, 2008; Mesoudi & O’Brien, 2008; Nakawake & Kobayashi, 2022). However, it is important to note that the “copy-successful-individuals” strategy involves a temporally sequential psychological process in learners. They need to monitor the performances of others, identify high-performing individuals, and then selectively learn from these individuals. In this case, high-performing individuals are inevitably more likely to possess valuable information to achieve goals than low-performing individuals. Furthermore, the “copy-successful-individuals” bias might be based on the respect-based response in learners toward demonstrators who have competence, experience, and knowledge, and thus gain admiration and prestige from the learners (Henrich & Gil-White, 2001; Horner et al., 2010). By contrast, the current study focused on the preference of social learning toward the successes of others and their failures in a situation where both convey an equal amount of valuable information to achieve goals and participants have digested the information. Additionally, because the models were identical across the demonstration of the successes and the failures, the “copy success” bias observed in the current study could hardly be explained from the perspectives of the informative asymmetry or prestige. Rather, the “copy success” bias might reflect a preference toward successes rather than failures in social learning at a relatively low-level and short-time cognitive process. Furthermore, in terms of the short time scale, the current findings could potentially elucidate the mechanisms involved in everyday social learning.

Studies focusing on asocial learning have demonstrated that compared to successes, it is harder for people to learn from failures (Eskreis-Winkler & Fishbach, 2019, 2022). In Eskreis-Winkler and Fishbach (2019), the participants answered binary-choice questions, after which they were told that they had answered correctly (success feedback) or incorrectly (failure feedback). Although both types of feedback conveyed the correct answer, the participants learned less from failed feedback than from successful feedback. The findings further indicated that failure undermines learning because it is ego-threatening, leading people to tune out. Consistent with this assumption, their results demonstrated that participants learned less from personal failures than from personal successes, yet they learned just as much from other people’s failures as from others’ successes.

Interestingly, even though the current study did not include ego threats among the participants, they learned less from failures than from the successes of others. Noteworthily, the investigated targets might have differed between previous studies and the current study. Eskreis-Winkler and Fishbach focused on participants’ performance in a memory test, whereas the current study focused on the strategy adopted by the model participants. The former and the latter might be considered the “input process” and “executive process” in learning, respectively. However, the exploratory analyses of the current study suggested that the “successful” events are more easily comprehended by the participants than the “unsuccessful” events. This implied that even in the “input process,” people seem to prefer successes over failures in social learning. This is consistent with the theoretical reasoning that learning from the failures of others may require more cognitive effort than learning from their successes (Eskreis-Winkler & Fishbach, 2020). Given that people are cognitive misers (Stanovich, 2009), they may have a hard time learning from others’ failures because it is difficult to extract valuable information from them. Future studies should investigate in more detail how people’s preferences are influenced by the interaction of social/asocial contexts and the “input/executive processes” of learning.

Furthermore, considering the possibility that the “successful” events are more easily comprehended by the participants than the “unsuccessful” events, one might argue that in the current experiment, there was informational asymmetry between the two types of events. Specifically, successful events provide more direct and unambiguous information regarding the location of the ball than unsuccessful ones. In the latter case, additional cognitive operations are needed to deduce the rewarding box, and some ambiguity still remains. While we acknowledge this logic, it is noteworthy that the main argument of the current study is that the “copy success” bias occurs even when people gain sufficient information for achieving goals from both others’ success and failure, rather than whether the demonstration itself includes informational asymmetry between the success and failure. Designing an experiment in which success and failure convey completely identical efficient information is almost impossible. Imagine an experiment in which researchers ask participants to choose two correct boxes among three; the two boxes have rewards in them but the third one is empty. Prior to the participants’ judgments, a “success cue” helped the model actor find the correct box, and an “unsuccessful cue” helped the model actor find the empty box. Such an experiment could enhance the efficiency of the information provided by the “unsuccessful cue” compared to the “successful cue.” This is because the former would offer an efficient way to identify the two correct boxes simultaneously, making the “unsuccessful cue” carry more valuable information than the “successful cue.” However, in this case the informational asymmetry between the two cues is obviously large, making it challenging to interpret the findings as a bias.

Despite the above methodological limitation, our exploratory analyses suggested that learning efficiency (i.e., the extent to which participants understood the use of the cues based on the observation of the events) did not influence the approach adopted by the participants. When looking at the participants who only gave correct answers to either of the follow-up questions, in other words, who had asymmetrical information regarding the cues related to the “successful” and “unsuccessful” approaches, we found that regardless of whether participants had a good understanding of either of the approaches, it did not influence which cue the participants chose to use. This suggests that the “copy success” bias might occur at the “executive process” (e.g., conducting an action) in social learning rather than the “input process” (i.e., memorizing information). The exploratory analyses further suggested that, regardless of whether participants were asked to find or miss the ball, it did not influence the approach they adopted. This also supports the notion that participants’ choice of approach did not solely depend on which cue could provide more efficient information to achieve the goals. This is because if this was true, the participants would have chosen the “successful cue” when they needed to find the ball and the “unsuccessful cue” when they needed to miss the ball.

One might argue that the “copy success” bias was driven by an association between the models’ task outcome and the participants’ desirable task outcome. This assumption would predict that participants would be more likely to adopt the model’s “successful” approach than the “unsuccessful” approach when they were asked to find the ball, and they would be more likely to adopt the reversed approach when asked to miss the ball. However, our exploratory analyses demonstrated that the “copy success” bias was not influenced by whether participants were asked to choose the box they thought the ball was in or not. Therefore, the “copy success” bias could not merely be driven by an association between the models’ task outcome and the participants’ desirable outcome.

In the current study, participants could complete test tasks regardless of whether they adopted cues that were linked to the successes or failures of the model. Moreover, of particular significance was the fact that all participants included in the final sample were able to comprehend the use of both cues. This allowed the investigation of the existence of a “copy success” bias when participants understood information related to both cues associated with the successes and failures of others. Our predictions assumed the presence of this bias. Specifically, we predicted a distribution where, instead of a chance-level distribution where half of the participants copied cues linked to others’ successes and the other half copied cues linked to others’ failures, there would be a biased distribution with 65% of participants copying cues linked to successes and 35% copying cues linked to failures. Despite being based on researchers’ subjective expectations, the study’s results generally confirmed the accuracy of these predictions. Furthermore, the experiment employed a variety of stimulus patterns to counterbalance factors that could introduce noise (such as the position of the illuminated buttons). Therefore, the observed “copy success” bias is presumed to be a consequence of observing the model’s successes and failures rather than other factors. However, while the study demonstrated the effect size of the behavioral bias, the specific psychological processes individuals employed to selectively copy the model’s successes or failures remain unclear. While some considerations were given in the preceding text regarding reasons for prioritizing the copying of cues linked to successes, the psychological processes of individuals who prioritized copying cues linked to failures were not explored.

In this study, approximately 36% (37) of participants prioritized copying cues linked to failures. Subsequently, we explored the psychological processes to understand why these 37 participants prioritized copying cues linked to failures rather than cues linked to successes. First, we looked at the experimental conditions under which the majority of participants prioritized copying cues linked to failures. The 37 participants were divided into two groups: those who watched a video where the model consistently placed their hand in a box with an illuminated button (the “successful” event involved retrieving a ball from a box with a glowing button of a certain color, while the “unsuccessful” event involved missing the ball from a box with a different color glowing button; 21 participants), and those who watched a video where the model consistently placed their hand in a box with a non-illuminated button (16 participants). Each group was further divided based on whether they were instructed to choose the box they believed the ball was in or not in during the subsequent test. We found that 17/37 participants had experienced a specific experimental condition. Notably, (only) in this specific condition, among participants who experienced it, more individuals prioritized copying cues linked to failures than to successes (eight participants copied successful case clues). This specific experimental condition involved being instructed to find the empty box (i.e., the box the participants believed the ball was NOT in) during the test after watching a video where the model placed their hand in a box with an illuminated button (the “successful” event involved retrieving a ball, while the “unsuccessful” event involved missing the ball). Why were the participants inclined to copy cues linked to failure in this experimental condition? It might be that the “unsuccessful” event in this condition, where the model placed their hand in the illuminated button box but failed to find the ball, was perceived as unexpected, contrary to participants’ expectations (presumably, participants expected the ball to be in the box with the glowing button). Previous research suggests that unexpected events are more memorable than expected ones (e.g., Meng et al., 2019). Therefore, the information from the “unsuccessful” event, including the cues linked to failures (e.g., the color of the button used in the “unsuccessful” event), might have been strongly retained in participants’ memories. As participants in this state were instructed to find the empty box during the subsequent test, this instruction might have activated their information processing related to “finding the empty box.” Consequently, the information from the unexpected “unsuccessful” event, strongly retained in memory owing to the prior expectation violation, might have been invoked, leading to a tendency to prioritize copying cues inked to failures. At least for these 17 individuals, they might have prioritized copying cues linked to failures through such psychological processes. Conversely, if we can resolve the experimental limitations regarding the association between button flashing and the prediction of success or failure (i.e., illuminated = likely to find the ball, non-illuminated = likely to miss the ball), further research may possess greater sensitivity to detect the effect size of the “copy success” bias. These considerations suggest promising directions for future research.

There are several limitations to the current study that should inspire future investigations. First, to enhance the level of evidence and establish a more conclusive determination regarding whether the “copy success” bias was due to bias that occurred during the “input stage,” bias that occurred during the “executive stage,” or both, future studies could test whether the information at the “input stage” influences the “copy success” bias. For instance, by additionally measuring participants’ confidence in understanding the use of cues, one can investigate in a more detailed manner whether people are more likely to adopt cues they remember better rather than adopting cues associated with others’ previous successes. Second, the current study focused on only adult participants, so it is unclear when and how the “copy success” bias in social learning develops. Developmental studies have demonstrated that children learn from others by considering others’ previous behavior; they keep track of the history of potential informants and selectively gain new information from informants who are expected to have reliable sources (Birch et al., 2008; Koenig et al., 2004; Koenig & Harris, 2005). These sensitivities in early social learning imply that children would also have differentiated responses toward the successes of others and their failures, which is expected to be investigated in future studies using the simplified methods of the current study. Third, the findings should be interpreted with caution regarding the variety of domains of social learning and types of learned patterns, which may rely on different rules and mechanisms. For example, many social and cultural norms are based on negative information (what you should not do) rather than positive (e.g., Fehr & Fischbacher, 2004; Herrmann et al., 2008). In this case, information on others’ failures may be more valuable than success. Comparing social learning patterns across different types of situations and types of tasks would provide much more valuable information. In a similar vein, for contemporary psychology and social sciences, information about one specific regularity in one specific context may be seen as insufficient to draw any solid conclusions. Investigating how this type of “learning from success” interacts with other possible mechanisms (e.g., social status of the model; Henrich & Gil-White, 2001) could significantly increase the value of the investigations. While this aspect is not within the scope of the current study, it should be considered in future research endeavors.

The current study demonstrated that people are more likely to employ cues that are linked to the successes of others than to their failures, even if they gain valuable information for achieving goals from both experiences. Our daily lives are full of the successes and failures of others. While it would be ideal to prioritize learning approaches that provide the most valuable information for achieving desired outcomes, people tend to learn more from the successes of others than from their failures. The psychological bias demonstrated here is expected to encourage a new wave in elucidating social learning mechanisms and provide suggestions for efficient learning processes.