BYSTANDER APATHY

From Observers to Interveners: Examining Bystander Intentions in Facebook Live Videos

Eun Hwa Junga and Nisha Suresh Bhojwanib

aSchool of Media & Advertising, Kookmin University, Seoul, South Korea; bHonest Financial Technologies, Jakarta, Indonesia

ABSTRACT This study investigated the bystander effect by examining intentions to intervene in simulated Facebook Live scenarios. A 4 (number of other viewers: 0 vs. 4 vs. 49 vs. 224) x 2 (level of urgency: non-urgent vs. urgent video) between-subjects experimental design was conducted with 210 par- ticipants to examine how the number of bystanders and the perceived urgency of a situation influence self-reported bystander intentions. With individual empathy levels controlled, an ANCOVA test revealed no significant main effect of the number of other viewers on the intention to intervene, but a significant difference was found between the 0 and 224 viewer conditions. Urgent situations elicited higher intervention intentions compared to non-urgent ones, without an interaction effect with viewer number. With the number of other viewers and urgency level as separate independent variables, mediation analysis found patterns differing from the traditional five-step bystander intervention model. These findings enhance our understanding how bystander intentions manifest in social media contexts and highlight the influence of perceived danger and audience size. The implications for online bystander behavior and future research directions are discussed.

KEYWORDS Bystander; Facebook live video; social impact; social influence; bystander intervention model

1. Introduction

In December 2015, Facebook launched “Facebook Live,” a feature allowing users to broadcast live content via their smartphone applications to friends, followers, and the broader Facebook community (Facebook, 2017). Notably, these live broadcasts remain accessible for viewing after their initial airing and are subsequently published to the owner’s page or profile (Facebook, 2017). However, there has been a disconcerting proliferation of violent incidents broadcasted on Facebook Live, often accompanied by reports of a sub- stantial viewership but limited intervention. A news analysis found that at least 45 instances of violence have been broad- cast via Live since its initial launch (Kantrowitz, 2017). These incidents include shootings, rapes, murders, child abuse, torture, and suicides.

For example, a man in Cleveland broadcasted a series of videos, including one where he shot an elderly man in the head. According to Facebook’s timeline, the account was disabled 20 minutes after the shooting video was reported, but that video was not reported for almost two hours (Lecher, 2017). This case illustrates that Facebook’s modera- tors acted fairly quickly and the delay came from viewers who did not report the video. However, a number of factors could cause viewers’ hesitance in reporting or their delayed reactions. These include concerns about their ability to pro- vide assistance, uncertainty regarding the outcomes of reporting violent content (i.e., whether it would lead to help

for someone or merely content removal), and sometimes, users may not even be aware that a reporting system is available (Lecher, 2017).

The cases also highlight instances where several people were aware of the situation but chose not to take action. One specific phenomenon related to social influence is the bystander effect, which describes the inverse relationship between the probability of receiving help and the number of bystanders present (Latan�e & Darley, 1970). This effect is underpinned by the social impact theory, which elucidates the factors that can impact social influence (Latan�e, 1981). Hudson and Bruckman (2004) explored the bystander effect in online educational environments and observed distinct behavioral changes compared to traditional classroom set- tings. They found that participants exhibited reduced fear of making mistakes, a greater sense of freedom in choosing to act or not, and overall heightened disinhibition. These changes were attributed to the unique characteristics of online environments, such as anonymity and the lack of immediate social cues, which differ significantly from face- to-face interactions. Furthermore, increased anonymity often fosters deindividuation and diminishes the sense of personal responsibility (McKenna & Bargh, 2000), which can reduce individuals’ willingness to intervene.

This study aims to delve into the bystander effect within the unique context of live videos on Facebook. While these videos make up a small portion of all the live content that is

CONTACT Eun Hwa Jung [email protected] School of Media & Advertising, Kookmin University, 77 Jeongneung-Ro, Seongbuk-Gu, Seoul 02707, South Korea � 2024 Taylor & Francis Group, LLC

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 2025, VOL. 41, NO. 14, 8655–8665 https://doi.org/10.1080/10447318.2024.2413292

being broadcasted on the platform, the severity of the inci- dents they capture necessitates a thorough examination. Notably, viewers spend at least three times more time watching live videos compared to non-live content (Facebook, 2016). According to a recent report (Bagadiya, 2024), platforms like Facebook continue to see high engage- ment rates with live streams, underscoring the enduring appeal and effectiveness of live streaming as a powerful tool for engaging audiences on social media platforms. Even though bystanders are not physically present and can remain anonymous, the potentially unlimited audience in the online world suggests that their role is significant (Holfeld, 2014). The specific focus of this research will revolve around the diffusion of responsibility, the bystander intervention model, and factors influencing social influence, such as group size. A deeper comprehension of the processes is essential before suggesting strategies to enhance bystander intervention within this distinctive setting.

2. Literature Review

2.1. The bystander effect and its theoretical frameworks

The bystander effect refers to the phenomenon where an individual’s likelihood of offering help decreases as the num- ber of passive bystanders increases (Darley & Latan�e, 1968). A contributing factor to this lack of intervention is the con- cept of diffusion of responsibility, wherein the presence of others shifts some of the responsibility for providing assist- ance onto them. This phenomenon holds that individuals are not non-interveners because of their personalities, but because of the situational forces around them that make them hesitant to intervene (Darley & Latan�e, 1968). Systematic reviews of bystander literature, examining various variables and experimental scenarios, have consistently affirmed the robustness of the social inhibition of helping (Fischer et al., 2011). Before an individual decides to inter- vene in a situation, they must first notice the event, interpret it as an emergency, accept personal responsibility for action, and believe they possess the necessary skills and resources (Latan�e & Darley, 1970). A negative decision, whether impli- cit or explicit, at any point in this sequence will result in a failure to help.

Group size influences individuals in various ways, impacting the likelihood and speed of their willingness to offer help. Three psychological processes can manifest when individuals find themselves in the presence of others, poten- tially hindering the completion of the five-step bystander intervention model illustrated in Figure 1.

Firstly, audience inhibition arises from the fear of nega- tive evaluation by others (Latan�e & Nida, 1981). This fear may lead individuals to worry about making mistakes, acting

inadequately, or misinterpreting the situation. With more people present, the risk of judgment from others escalates. Secondly, social influence refers to the tendency to rely on others’ behavior when defining an ambiguous situation (Latan�e & Nida, 1981). Observing the inaction of others may cause an individual to perceive the situation as less crit- ical or assume that inaction is the expected response. Lastly, the diffusion of responsibility reduces the psychological cost associated with non-intervention (Latan�e & Nida, 1981). The awareness of others’ presence shifts some of the responsibil- ity for providing help onto them, diminishing each individu- al’s motivation to take action.

The theory of social impact (SIT) originated from research on bystander intervention. This theory asserts that an individual’s responses are influenced by the real, implied, or imagined presence or actions of other individuals (Latan�e, 1981). SIT is based on three key principles.

Firstly, it posits that the impact on a target individual is a multiplicative function of the strength, immediacy, and the number of other people (Latan�e, 1981). In simpler terms, the combined effect of these factors influences the impact on the individual. Secondly, SIT suggests that the impact on the target individual follows a power function, with the contribution of the Nth person being less than that of the (N − 1)th person (Latan�e, 1981). In essence, the impact of the first person is greater than the hundredth, indicating a diminishing effect as the number of people increases. In line with the diffusion of responsibility, the third principle sug- gests that when multiple people are present, the influence each person has is reduced. This means the overall impact is divided among all the people present. The influence of each person decreases according to an inverse power function, considering three factors: the strength (importance), imme- diacy (closeness), and the number of people present. In sim- pler terms, the more people there are, the less impact each individual has.

Building upon Latan�e’s work, the social influence model (SIM) aimed to incorporate both majority and minority influence processes into a unified theoretical framework while addressing various limitations and inconsistencies found in the SIT (Tanford & Penrod, 1984). There are two significant distinctions between the SIT and the SIM. The SIT posits that the initial influence source has the most impact, and each additional source augments influence, albeit to a lesser extent than its predecessor. In contrast, the SIM predicts that the second and third influence sources should have a greater impact than the first (Tanford & Penrod, 1984). In addition, while the SIT does not impose a limit on the amount of influence attainable and suggests that influence would continue to grow with the addition of more sources, albeit at a decreasing rate, the SIM forecasts

Figure 1. The bystander intervention model (Latan�e & Darley, 1970).

8656 E. H. JUNG AND N. S. BHOJWANI

that influence will eventually plateau once total influence is reached, essentially capping the potential for further influ- ence. Consequently, increasing the group size beyond four targets will not lead to a greater social influence (Tanford & Penrod, 1984).

Therefore, the pattern with which responsibility diffuses differs. SIT suggests that the inclusion of each additional bystander will stimulate further diffusion while SIM demon- strates that members of dyads will experience less diffusion of responsibility than implied by SIT as the responsibility would not continue to diffuse once the group size increases beyond four members. Forsyth et al. (2002) conducted a study in which two-, four-, six-, and eight-membered groups were asked to allocate 100 responsibility points among their members after completing a group task. Their findings pro- vided evidence consistent with SIT, showing that responsi- bility diffused in proportion to group size (Forsyth et al., 2002).

2.2. The bystander effect in CMC

The bystander effect, which has previously been observed in offline settings, has also been demonstrated in online con- texts. Early research in this domain examined electronic helping behavior and interactions within chat groups and via email. The surge in cyberbullying incidents and the heightened attention they garner have necessitated a deeper exploration of the online bystander phenomenon. When viewed through the lens of cyberbullying, the role of cyber- bystanders becomes critical—they can either assist victims, bolster bullies, or implicitly condone cyberbullying through passivity.

Contemporary studies, including some involving social networking sites, provide pertinent insights into cyberby- stander behaviors. These are relevant to the current investi- gation despite not directly exploring cyberbullying. For instance, Obermaier et al. (2016) simulated a cyberbullying event via comments in a mock university Facebook group. Their results did not indicate a direct bystander effect but highlighted an indirect effect, influenced by one’s sense of responsibility. In light of the bystander phenomenon in Computer-Mediated Communication (CMC), this research aims to delve into two primary online bystander determi- nants: the number of people in the group and the urgency of the incident.

2.2.1. The number of other people in the group Research has shown that the number of bystanders can sig- nificantly influence response times and intervention behav- iors in online environments. Markey (2000) investigated the bystander effect in online chat groups by analyzing 4,833 participants across 400 Yahoo! Chat groups. The study found that larger group sizes led to longer response times to help requests. However, this effect was significantly reduced when help was requested by naming a specific individual, emphasizing the role of personal responsibility. Voelpel et al. (2008) examined the bystander effect in online forums

through a quantitative analysis of 333 Yahoo! Groups with a total of 197,419 members. They found the bystander effect was more pronounced in smaller forums (fewer than 250 members), while response rates increased in larger forums. These studies highlight how group size affects bystander behavior in online environments and underscore the impor- tance of personal responsibility and group dynamics in influencing intervention rates. It is important to note that Markey (2000) and Voelpel et al. (2008) did not explicitly test the diffusion of responsibility, as their findings were correlational.

Supporting these findings, Barron and Yechiam (2002) studied the effect of the number of recipients in email requests on response rates and helpfulness. Using an experi- mental design, they found that sending email requests to one address at a time, rather than to multiple recipients, increased response rates and helpfulness. This suggests that perceiving additional recipients as capable bystanders can lead to the diffusion of responsibility.

While the bystander effect is theoretically rooted in SIT, a few studies exploring virtual diffusion of responsibility have found results that align more with SIM. These studies showed no classic or true bystander effect; participants’ intentions to intervene did not linearly decrease with an increasing number of bystanders. For example, Blair et al. (2005) found no significant difference in response distribu- tion between 1-person and 2-person conditions or between the 15-person and 50-person group conditions. A significant difference emerged only between groups with no or few others (collapsing the data from 1- and 2-person groups) and many others (collapsing the data from 15- and 50-per- son groups) (Blair et al., 2005). This indicates that a virtual diffusion of responsibility does occur, but its effect is not linearly related to group size.

A study comparing 24 bystanders with 5025 found no significant effect on behavioral intention (Obermaier et al., 2016). In addition, the feeling of responsibility decreased when a very small number (2 bystanders) was compared to a medium or large number of bystanders (24 or 5025) (Obermaier et al., 2016). Thus, this study uncovered a non- linear relationship between the number of bystanders and the feeling of responsibility. Similarly, Machackova et al. (2015) also found a non-linear bystander effect. When par- ticipants were the sole witness to a cyberbullying incident, they were more supportive than when there were more wit- nesses. However, no difference was found between the mod- erate number and many bystander groups (Machackova et al., 2015). Martin and North (2015) findings were consist- ent with those found in Blair and colleagues’ (2005) study. Therefore, research consistently suggests that in a virtual context, SIM offers a better explanation for the diffusion of responsibility than SIT, as the relationship between the number of bystanders and intervention intentions is not straightforward or linear.

2.2.2. Urgent situations The bystander effect has been reproduced in various situa- tions, leading to the identification of its limiting conditions.

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 8657

An area of interest is the distinction between emergencies and non-emergencies. In a study by Shotland & Huston, 1979, solicitors requested a ride, providing explanations that depicted either an emergent or non-emergent situation. Findings revealed a higher likelihood of assistance in emer- gency scenarios, underscoring the role of perceived victim need in bystander decision-making. In another study, Harari et al. (1985) presented male bystanders with a stimulated rape scenario in a natural setting, comparing group versus individual bystander response. They observed an elevated helping rate from group bystanders, deducing that such con- texts diminished the diffusion of responsibility (Harari et al., 1985).

Contrary to numerous studies focusing on non-violent incidents, Fischer et al. (2006) investigated bystander response in an emergency elicited by a violent crime, posing potential harm to both the victim and the bystander. The bystander effect sustained in low-potential danger scenarios, but it dissipated in highly dangerous emergencies. This shift was attributed to the clearer recognition of emergencies and an augmented acceptance of the costs of providing assist- ance. Furthermore, there was no significant impact of partic- ipants’ sex and age or the bystander’s sex on the results, suggesting that these demographic factors did not influence the likelihood of intervention in high-danger situations.

In the realm of cyberbullying, incident severity was observed to positively influence behavioral intention. Obermaier et al. (2016) noted that extreme cyberbullying instances bolstered participants’ intervention intentions, although these intentions dwindled in the presence of numerous bystanders. A study by Bastiaensens et al. (2014), using a scenario of a harassing incident on Facebook, dem- onstrated that 13-year-old participants exposed to a more severe harassing incident were significantly more inclined to assist the victim. In conclusion, these diverse studies collect- ively highlight the nuanced influences on bystander inter- vention in urgent situations, spanning both physical and virtual contexts.

2.2.3. Empathy Empathy is a notable factor associated with bystander inter- vention. Simulating emergency and non-emergency situa- tions through video editing software, it was found that individuals with higher empathic concern responded more quickly to emergencies in the absence of other bystanders (Hortensius et al., 2016). Specifically on Facebook, individu- als with high empathy levels were more likely to intervene in a live discussion where two confederates created a bully- victim dynamic (Freis & Gurung, 2013). Additionally, Nickerson et al. (2014) investigated the factors influencing bystander intervention in bullying and sexual harassment incidents among secondary school students and found that empathy was significantly associated with higher interven- tion rates in both bullying and sexual harassment scenarios. Their findings support the importance of empathy in moti- vating bystander intervention.

Both cognitive and affective empathy types were correlated with support for cyberbullying victims. Cognitive empathy

involves understanding another person’s perspective, while affective empathy refers to sharing and responding to anoth- er’s emotions. However, only affective empathy predicted a supportive bystander response (Machackova & Pfetsch, 2016). This study, involving participants aged 12-18, found no age or gender effect on outcomes (Machackova & Pfetsch, 2016). Higher empathic concern in adolescents increased the likeli- hood of helping cyberbullying victims, whereas lower levels corresponded with joining the bullying or remaining passive (Van Cleemput et al., 2014).

Given the significant role of empathy in CMC, it is fac- tored into the current study as a control variable. The main effect of the number of other viewers, the level of urgency, and their interaction effect are examined to assess whether they operate similarly in this new context. Accordingly, the following hypotheses are proposed:

H1: Controlling for empathy, as the number of other viewers watching increases, the intention to intervene decreases.

H2: Controlling for empathy, the more urgent the situation, the greater the intention to intervene.

H3: Controlling for empathy, the intention to intervene will increase as the number of other viewers decreases and the level of urgency increases.

2.3. Mediating variables in the bystander intervention model

As for the bystander intervention model, a few studies have tested the sequential steps leading to the intervention deci- sion. Within two prosocial contexts, evidence confirmed that the bystander intervention model could be delineated into four sequential steps leading to behavior implementation (Anker & Feeley, 2011). This research additionally demon- strated that these stages might interrelate non-chronologically for certain behaviors (Anker & Feeley, 2011). Nickerson et al. (2014) conducted a confirmatory factor analysis and struc- tural equation modeling on a 16-item measure for bullying and sexual harassment based on the five steps proposed by Latan�e and Darley (1970). They discovered a five-factor struc- ture representing the five steps, wherein each step was influ- enced by the preceding one. Examining the first step (noticing the incident) with chatroom technology, results con- firmed that university students’ notice of a cyberbullying inci- dent significantly predicted their intervention, be it direct or indirect (Dillon & Bushman, 2015). Obermaier et al. (2016) examined three of the five steps and found that for severe cyberbullying incidents, the behavioral intention was medi- ated by the assessment of the situation as an emergency, fol- lowed by an increased feeling of responsibility.

From the literature reviewed, it is evident that the results are not consistent. Such failure in replication might be ascribed to asynchronous communication media or the over- whelming number of bystanders in online interactions (Allison & Bussey, 2016). In online contexts, bystanders may only witness the incident after the fact, leading them to believe the situation is resolved and assistance is no longer required. Individuals might also find it challenging to com- prehend the implications of their social interactions amidst

8658 E. H. JUNG AND N. S. BHOJWANI

the presence of numerous other individuals. Alternatively, the disparate findings might be attributed to varying num- bers of bystanders involved in each study (Allison & Bussey, 2016). While the bystander effect is markedly robust in face- to-face contexts, its emergence in CMC and the variables affecting it offline are not yet fully understood, given the unique characteristics of CMC.

In light of these considerations, the stages outlined in the bystander intervention model were examined to discern their influence on reaching an intervention decision, which proposed the following hypotheses.

H4: Emergency assessment, feeling of personal responsibility, and knowing how to act will mediate the relationship between the number of other viewers and the intention to intervene.

H5: Emergency assessment, feeling of personal responsibility, and knowing how to act will mediate the relationship between the level of urgency and the intention to intervene.

3. Method

An online experiment was conducted employing a 4 (0 vs. 4 vs. 49 vs. 224 other viewers) X 2 (non-urgent vs. urgent video) between-subjects design. This study design received prior approval from the Institutional Review Board (IRB). A total of 210 undergraduate and postgraduate students in Singapore, all of whom were current Facebook users, were recruited and randomly assigned to one of four experimental conditions using Qualtrics survey software (see Table 1). To ensure our study had sufficient statistical power, an a priori power analysis was conducted using G�Power (Faul et al., 2007). This analysis was aimed at detecting a medium effect size (f¼ 0.25) with a significance level of 0.05 and a power of 0.80 for our 4 (number of other viewers: 0 vs. 4 vs. 49 vs. 224) x 2 (level of urgency: non-urgent vs. urgent video) between-subjects design. Given the numerator degrees of freedom were set at 3, the analysis determined that a min- imum of 179 participants would be required. With 210 par- ticipants, our study not only meets but exceeds this requirement, thereby ensuring robust statistical power for our analyses.

For the recruitment process, a survey link generated from Qualtrics was shared with friends and fellow peers via social media channels, such as WhatsApp and other social net- working sites (SNSs). To compensate participants for their time, a $5 promotional code for a booking and travel appli- cation was provided upon completion of the survey. Participants’ ages ranged from 18 to 30 years old (M¼ 20.71, SD¼ 1.96; where M stands for mean and SD stands for

standard deviation). The sample was predominantly Chinese (N¼ 170, 81%) and primarily female (N¼ 149, 71%).

3.1. Stimuli

The stimuli used in this study were edited videos embedded in the survey. Two publicly available videos depicting an urgent and non-urgent situation were selected from YouTube. In add- ition, four levels for the number of viewers (0 vs. 4 vs. 49 vs. 224 other viewers) were chosen to investigate whether results would align more with SIT or SIM. Voelpel et al. (2008) only found the bystander effect between small (0-99) and medium- sized (100-250) groups. As an exploratory study, they cau- tioned not to conclude that a group size of 250 or more would be the same turning point in all contexts (Voelpel et al., 2008). Indeed, Obermaier et al. (2016) did find an indirect effect when comparing two versus 5025 bystanders, however, Machackova et al. (2015) cautioned against using very large numbers citing realism. When checking participants’ percep- tion of the number of bystanders, the mean difference between 224 and 5025 was not as big as the mean difference between two and 224 and 24 and 224 bystanders (Obermaier et al., 2016). Thus, 224 is already perceived as quite a lot. Blair and colleagues (2005) found a significant difference between zero and one others versus 14 and 49. Similarly, Martin and North (2015) found a significant difference between zero and one vir- tual bystanders when compared against four and 14. Using zero, four, 49, and 224 in the current study reflected a middle ground between the statements posited by SIM and the large audiences available on SNSs.

Urgent situations are defined as those involving a threat or actual harm that increases with time, necessitating outside help (Shotland & Huston, 1979). In line with this, the urgent video depicted a violent fight between two men, with one appearing to be in need of medical attention at the end. The non-urgent video depicted an argument between a man and a taxi driver about an illegally parked car, culminating in the taxi driver hit- ting the man’s car and driving away. These two videos were edited to appear as though they were streaming live on Facebook and were trimmed to thirty seconds each to avoid showing more people entering the video frame.

3.2. Procedure

Upon accessing the online survey, participants encountered an information sheet. Consent was obtained by participants’ indication of agreement with the information presented. Declining terminated the session. Participants who agreed were directed to the first question where they were asked if they currently had a Facebook account. Those without an account were redirected to the end of the survey to ensure the relevance of the study, as familiarity with the platform was important for interpreting their behavior.

Subsequently, participants were randomly allocated to the experimental conditions and brought to a page with the video. The subsequent questions presented depended on which video they watched. Both sets of questions measured the same variables, but the statements were tailored to fit

Table 1. The number of participants in experimental conditions.

Urgency of the incident in the video

TotalNon-urgent Urgent

Number of other viewers 0 23 24 47 4 29 28 57

49 24 26 50 224 28 28 56

Total 104 106 210

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 8659

the situation participants had just been exposed to. Manipulation check questions remained consistent across both the urgent and non-urgent videos. Following this, questionnaire items, including a control variable and demo- graphic questions, were presented. Upon completing the questionnaire, participants were directed to a debriefing message and thanked for their participation.

3.3. Measures

Measures were employed to evaluate whether the manipula- tions of our independent variables were successful, to quantify variables corresponding to the bystander intervention model, and to identify an individual’s level of empathy. Demographic questions were also included in the questionnaire.

3.3.1. Manipulation check Participants were asked to freely recall the number of other viewers watching the “live” video and assess this number on a 5-point Likert scale by rating whether “none” or “very many” others had seen the video (M¼ 2.5, SD¼ 1.10). Higher scores represent a perception of more viewers. Participants also assessed the degree of urgency of the incident they were exposed to according to the following items: “not threatening” – “very threatening,” “not alarming” – “very alarming,” and “not dangerous” – “very dangerous,” using 5-point Likert scales (M¼ 3.63, SD ¼ .94, a ¼ .90). Higher scores indicate a greater perceived urgency of the incident.

3.3.2. Empathy Adopted from Davis (1980), seven items (e.g., “when I see someone being taken advantage of, I feel kind of protective toward them”) were used to ask participants to indicate how much each statement described them (1¼ does not describe me well, 5¼ describes me very well) (M¼ 3.85, SD ¼ .66, a ¼ .84). Higher scores represent a greater level of empathy.

3.3.3. Emergency assessment Emergency assessment is defined as identifying whether the situation presented was hazardous to the individuals in the video. Adopted from Obermaier et al. (2016), three items (e.g., This situation constitutes a threat for the man in the video) were used to measure participants’ level of agreement on a 5-point Likert scale (1¼ Strongly disagree, 5¼ Strongly agree) (M¼ 4.14, SD ¼ .74, a ¼ .87). Higher scores indicate a higher perceived level of emergency.

3.3.4. Feeling of responsibility A feeling of responsibility refers to the participant accepting that they had a role to play as a bystander. Participants were asked to indicate their level of agreement to three statements (e.g., “I highly feel personally responsible to support the man in the video”) on a 5-point Likert scale (1¼ Strongly disagree, 5¼ Strongly agree) (M¼ 2.65, SD ¼ .90, a ¼ .81). Higher scores represent a stronger feeling of responsibility.

3.3.5. Knowing how to act Knowing how to act is operationally defined as the partici- pant acknowledging that they have the relevant skills and resources needed for intervention. Adopted from Nickerson et al. (2014), three items were used to measure knowing how to act (e.g., “I have the skills to support the victim in the video”) on a 5-point Likert scale (1¼ Strongly disagree, 5¼ Strongly agree) (M¼ 2.43, SD ¼ .95, a ¼ .83). Higher scores represent a greater perceived ability to act.

3.3.6. Intention to intervene Adopted by Obermaier et al. (2016), participants were asked to indicate their intention to intervene using a questionnaire item “I would intervene” on a 5-point Likert scale (1¼ Strongly dis- agree, 5¼ Strongly agree) (M¼ 2.53, SD¼ 1.17). Higher scores indicate a greater intention to intervene.

3.3.7. Demographic variables Participants were asked to indicate their age, gender, and race.

4. Results

4.1. Manipulation check

To ensure whether our manipulation was adequate, the relative assessment of the number of bystanders in the four conditions and participants’ perception of the urgency of the incident was tested. Results indicated a significant difference between the four groups, F(3, 206) ¼ 14.24, p < .001, suggesting that as the number of other viewers increased, participants reported an increasing number of others had seen the video. Participants perceived zero viewers with a mean score of 1.79 (SD ¼ 1.12), four viewers with a mean score of 2.39 (SD ¼ 0.92), 49 viewers with a mean score of 2.7 (SD ¼ 0.81), and 224 viewers with a mean score of 3.04 (SD ¼ 1.11). A signifi- cant difference in participants’ perception of the urgency of the incident was found between the two videos, t(208) ¼ −5.74, p < .001. The urgent video (M¼ 3.98, SD ¼ .77) was rated as more threatening, alarming, and dangerous than the non- urgent video (M¼ 3.29, SD¼ 0.97). Therefore, the manipula- tion was successful.

4.2. Testing hypotheses

For hypotheses 1 through 3, a 4� 2 ANCOVA was con- ducted with empathy as a covariate. Results showed no sig- nificant main effect of the number of other viewers on the intention to intervene [F(3, 202) ¼ 1.64, p ¼ .181, partial g2 ¼ .024]. Specifically, LSD post-hoc analysis showed a sig- nificant difference between 0 viewer (M¼ 2.83, SD¼ 1.05) and 224 viewers (M¼ 2.34, SD¼ 1.17) conditions in inten- tion to intervene (Mean difference ¼ .49, SE ¼ .23, p < .05), but the condition of 4 and 49 viewers were not statis- tically significant compared to the other three conditions. Therefore, H1 was partially supported. However, there was a significant main effect of the level of urgency on the inten- tion to intervene [F(1, 202) ¼ 14.4, p < .001, partial g2 ¼

8660 E. H. JUNG AND N. S. BHOJWANI

.067], supporting H2. The interaction effect between the number of other viewers and the level of urgency was not significant [F(3, 202) ¼ 1.58, p ¼ .19, partial g2 ¼ .023], not supporting H3.

To explore the mediating variables proposed by H4 and H5, PROCESS was used for mediation analysis with 5,000 bootstrap samples and 95% bias-corrected confidence inter- vals (Hayes, 2013). PROCESS allows mediators to be linked serially in a causal sequence and offers measures of effect sizes for indirect effects (Hayes, 2013). Multicategorical independent variables instead of dichotomous or continuous variables have been used to make an inference about direct, indirect, and total effects in a statistical mediation analysis (Hayes & Montoya, 2017; Hayes & Preacher, 2013). For both H4 and H5, a multiple mediator model with three mediators operating in serial (model number 6 on PROCESS) was employed (Hayes, 2013).

The prediction of H4 was that emergency assessment, feeling of responsibility, and knowing how to act would mediate the relationship between the number of other view- ers and the intention to intervene. The earlier pairwise com- parisons were used to determine which of the four groups would be used for mediation analysis. It showed the greatest statistically significant difference between 0 and 224 viewers. If no other viewers were watching the live video, participants intended to intervene more (M¼ 2.83, SD¼ 1.05) than when 224 other viewers were watching (M¼ 2.34, SD¼ 1.17). Therefore, cases belonging to the four and 49 other viewers conditions were filtered out, resulting in a sample size of 103.

As shown in Figure 2, the mediation analysis revealed significant relationships between the number of other view- ers and feeling of responsibility (b¼−.47, SE ¼ .17, t¼−2.83, p < .01), between feeling of responsibility and knowing how to act (b ¼ .54, SE ¼ .09, t¼ 5.98, p < .001), and between knowing how to act and intention to intervene (b ¼ .68, SE ¼ .12, t¼ 5.78, p < .001). The direct effect of the number of viewers on intention to intervene was mar- ginally significant (b¼−.36, SE ¼ .19, t¼−1.90, p < .06). However, there was not a significant indirect relationship between the number of other viewers and intention to inter- vene through three mediators (i.e., emergency assessment, feeling of responsibility, and knowing how to act), b¼−.02, 95% CI [-.06, −.006], because the confidence interval crosses zero. Instead, a significant indirect relationship was found: an indirect effect of the number of other viewers on the intention to intervene mediated by a feeling of responsibility and knowing how to act, b¼−.17, 95% CI [−.33, −.05]. Therefore, all three variables do not mediate the relationship between the number of other viewers and the intention to intervene, which partially supported H4.

H5 examined whether emergency assessment, feeling of responsibility, and knowing how to act would mediate the rela- tionship between the level of urgency and intention to inter- vene. Significant direct relationships emerged as indicated in Figure 3. The level of urgency significantly predicted emer- gency assessment (b ¼ .57, SE ¼ .09, t¼ 6.00, p < .001), but emergency assessment did not significantly relate to the feeling of responsibility. Feeling of responsibility significantly predicted

knowing how to act (b ¼ .62, SE ¼ .06, t¼ 10.15, p < .001) and intention to intervene (b ¼ .31, SE ¼ .08, t¼ 3.67, p < .001). Finally, knowing how to act predicted the intention to intervene (b ¼ .56, SE ¼ .08, t¼ 7.16, p < .001). The direct effect of the urgency level on intention to intervene was signifi- cant (b ¼ .57, SE ¼ .15, t¼ 3.65, p < .001). However, all the indirect relationships through mediators were not significant.

5. Discussion

Although the bystander effect is a robust social psychological phenomenon in face-to-face contexts, the results of this study elucidate the complex nature of the phenomenon within CMC. It was anticipated that groups in CMC would provide an adequate circumstance for the bystander effect to occur due to the possibility of larger groups who can be strangers to one another and whose presence is only implied – all of which are factors that strengthen the bystander effect (Fischer et al., 2011). While the bystander effect does occur, it seems that different patterns of results emerge within CMC.

Several studies have found results more aligned with SIM rather than SIT (Blair et al., 2005; Machackova et al., 2015; Martin & North, 2015; Obermaier et al., 2016; Voelpel et al., 2008). The direct effect of the number of other viewers (0 versus 224) on the intention to intervene, found in the serial mediation analysis, adds more credence to the notion that the bystander effect emerges only when comparing a very small versus a very large number of bystanders. Obermaier et al. (2016) found no effect on behavioral intention when comparing 24 versus 5025 bystanders but detected an indir- ect effect when comparing 2 versus 5025 bystanders.

The stimuli employed in this study aimed to address the limitations of asynchronous media prevalent in prior studies. Yet, this still did not result in a true bystander effect occur- ring, suggesting that other characteristics of CMC may have equally pronounced effects. The ability to be relatively anonymous on the Internet reduces accountability, leading to the depersonalization and deindividuation of the user (Okdie & Guadagno, 2008). Offline, it has been shown that when bystanders are anonymous or strangers, helping behaviors are significantly slower and less likely (Fischer et al., 2011; Levine & Crowther, 2008; Schwartz & Gottlieb, 1980; Solomon et al., 1982). Online, increased numbers of unknown cyberbystanders inhibited intervention (Brody & Vangelisti, 2015; Palasinski, 2012). In essence, online bystanders, being non-physical and potentially anonymous, limit the individual influencers that could spur them into action.

In the realm of cyberbullying, Allison and Bussey (2016) suggest that the presence of bystanders inconsistently affects witnesses’ responses. Obermaier et al. (2016) had to exclude 20-40% of their study sample because participants were unable to recall the number of bystanders with any degree of accuracy. Using participants who had previously wit- nessed cyberbullying incidents, individuals reported provid- ing more support to victims when they were “just a few” compared to when there was “a bit more” or “a lot” (Machackova et al., 2015). This suggest that online group

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 8661

size is not as influential as its offline counterpart. The possi- bility of endless audience sizes, compounded with their pre- dominant anonymity, might explain why CMC study outcomes tend more towards SIM than SIT.

Despite the statistically weak direct relationship between the number of other viewers and the intention to intervene, we found a significant indirect relationship between the number of other viewers and the intention to intervene mediated by a feeling of responsibility and knowing how to act. Our findings suggest that the diffusion of responsibility does not hinge on the awareness of others’ actions. When individuals feel responsible for an incident or a victim and possess knowledge of taking action, they are more likely to offer help.

The main effect of urgency level on the intention to intervene aligns with the findings of Shotland and Huston (1979). In the current study, the urgency of the situation displayed combined with the “live” manipulation might have signified to participants the victim’s need for help. The level of urgency did not interact with the number of other view- ers, contrasting previous literature that pinpointed the role of potentially dangerous emergencies in counteracting the bystander effect (Fischer et al., 2011).

Dangerous situations attenuate the bystander effect because they are perceived more clearly as emergencies. In addition, other bystanders reduce fear of intervention as they can provide social, physical, or psychological support

and dangerous emergencies are most effectively resolved by coordination and cooperation among a greater number of bystanders (Fischer et al., 2011). This is also supported by our mediation analysis of the level of urgency on intention to intervene, which is that urgency level had a direct rela- tionship with emergency assessment. This indicates that more urgent situations are less ambiguous and thus more clearly perceived as emergencies.

Although urgent situations imbue emergency assessment, it did not mediate the relationship between the level of urgency in the situation and the intention to intervene. The other two psychological factors in the bystander intervention model did not mediate the relationship. Participants might have been inclined to intervene in urgent situations due to the increasing harm over time but may have failed to develop a feeling of personal responsibility if the victim’s circumstances arose from their actions (e.g., fighting).

5.1. Theoretical and practical implications

This study deviates from prior paradigms of testing the bystander effect in CMC by specifying SNS features and employing more conventional stimuli instead of focusing on cyberbullying incidents. It aligns with studies demonstrating that responsibility does not diffuse in proportion to group size in CMC. Earlier studies that obtained a statistically sig- nificant difference collapsed groups with less than four

Figure 2. Serial mediation analysis of the number of other viewers on intention to intervene. �p < .05; ��p < .01; ���p < .001.

Figure 3. Serial mediation analysis of level of urgency on intention to intervene. �p < .05, ��p< .01, ���p< .001.

8662 E. H. JUNG AND N. S. BHOJWANI

bystanders and groups with more than four bystanders (Blair et al., 2005; Martin & North, 2015). However, the cur- rent study found that four viewers were not perceived as statistically different from 49 viewers. Thus, the finding sug- gests that groups sized four and beyond do not seem to be statistically different from one another. This implies that a threshold seems to exist in CMC before intervention rates stabilize. Especially, a victim is likely to receive assistance only if a minimal number of bystanders are present.

Several technological factors, such as the number of cues transferred, the size of the audience, and synchronous versus asynchronous communication, determine bystander behav- ior. Though “live” videos signified synchronous communica- tion (albeit one-way in this specific study), other pronounced CMC characteristics, such as anonymity and audience size, might sway the manifestation of the bystander effect online. Anonymity can foster greater disinhibition, but evidence suggests that group size is not as compelling an influencer online as it is offline (Allison & Bussey, 2016). This study found that the level of urgency was a stronger predictor of the intention to intervene than the number of other viewers, offering mixed implications for future vio- lence incidents on Facebook Live. A highly emergent situ- ation should theoretically drive intervention, but CMC attributes might curtail such effects.

In response to certain incidents, Facebook added 3,000 additional employees to its community operations team (Tsukayama, 2017). Yet, despite these additions, Facebook conceded they could not monitor everything (Tsukayama, 2017). The aspiration is to reduce the duration a video remains available once flagged, preventing viral spread. However, discerning video context remains challenging; wit- nessing a shooting differs from a perpetrator broadcasting their crimes. Even as Facebook leans on artificial intelligence to auto-detect inappropriate content, the community’s role is pivotal, especially in the context of discernment. If Facebook persists in its reporting system reliance, there is a clear need for user education about its utility and potential positive impacts.

5.2. Limitations and future research

Although the manipulation was successful, the means for both the urgent and non-urgent videos were above the scale midpoint. Using videos that did not fully encapsulate all the characteristics defining emergencies (those typically resulting from accidents and causing harm that escalates over time, according to Shotland & Huston, 1979) might have con- strained the mediating effect of the emergency assessment variable. The two types of situations used were also not comparable. Ideally, the situations would have the same onset, but the ending would vary to reflect either an urgent or non-urgent outcome.

Since this study employed stimulus resembling certain features of Facebook Live videos, future research could explore ways to incorporate and control for all the features (i.e., social interaction between viewers) and use the actual platform rather than an imitation. Generally, situations

where bystanders know each other mitigate the bystander effect more than situations where bystanders are complete strangers; this is true even in dangerous emergencies (Fischer et al., 2011). Integrating all the features would enable processes like social influence or audience inhibition to interact with the diffusion of responsibility.

Future research should develop context-specific measures for the five-step bystander intervention model, permitting more accurate measurement and deeper insight. The current measure was reliable; nonetheless, it did not yield substantial insight into how participants might intervene and what skills and resources they believe would aid them in such a situ- ation. Context-specific bystander behavior in the online environment, such as Facebook’s built-in reporting mechan- ism, could be incorporated into the measure to address these questions. Alternatively, qualitative research could supple- ment these results and explore the motivations behind non- intervention.

Furthermore, while this study included viewer sizes up to 224, exploring even larger viewer sizes, such as thousands or millions, may yield new insights into the dynamics of bystander intervention in highly viral content scenarios. Future research could incrementally increase the viewer size levels to include much larger audiences, which would help to map out the effects of extremely large viewer sizes on bystander intervention more comprehensively. We also acknowledge that our study’s wide spacing between viewer categories (49 to 224) makes it difficult to determine the precise threshold for influencing bystander intentions. Future research should aim to investigate this threshold with more granular viewer categories to provide a more accurate understanding of how different group sizes impact bystander behavior.

Additionally, to strengthen the validity of our classifications of urgent and non-urgent situations, future research should consider conducting additional testing with a separate group of participants who did not take part in the formal experiment. This group will evaluate the urgency levels of the video clips to ensure there is indeed a significant difference. This additional step will help to validate our initial classifications and provide a robust basis for our conclusions.

Moreover, considering empathy’s significant role in influ- encing bystander behavior, future studies should directly hypothesize about its impact on bystander intentions. This approach would provide a more comprehensive understand- ing of how both cognitive and affective empathy contribute to intervention behaviors in various contexts.

Lastly, generalizing these results to other events (e.g., instances more violent than fighting) and people (e.g., non- students) should be done with caution. Future studies should aim to replicate and extend our findings using larger and more heterogeneous samples.

6. Conclusion

This study contributes to the existing literature as the first to examine the diffusion of responsibility using live videos. Investigating the bystander effect within the SNS context, it

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 8663

incorporates both situational (urgent incidents) and personal (empathy) variables commonly associated with offline bystander effect studies. Urgent situations did lead to height- ened intervention. By testing the bystander intervention model, it was found that feelings of responsibility and know- ing how to act mediate the relationship between the number of other viewers and the intention to intervene. In contrast, no mediators were found in the relationship between the level of urgency and the intention to intervene. The findings of this study suggest that the unique characteristics of com- puter-mediated communications influence the manifestation of the bystander effect online.

Disclosure statement

No potential conflict of interest was reported by the author(s).

ORCID

Eun Hwa Jung http://orcid.org/0000-0003-0698-9887

References

Allison, K., & Bussey, K. (2016). Cyber-bystanding in context: a review of the literature on witnesses’ responses to cyberbullying. Children and Youth Services Review, 65, 183–194. https://doi.org/10.1016/j. childyouth.2016.03.026

Anker, A., & Feeley, T. (2011). Are nonparticipants in prosocial behav- ior merely innocent bystanders? Health Communication, 26(1), 13– 24. https://doi.org/10.1080/10410236.2011.527618

Bagadiya, J. (2024). 38 Facebook statistics and facts for every marketer in 2024. https://www.socialpilot.co/blog/facebook-statistics

Barron, G., & Yechiam, E. (2002). Private e-mail requests and the dif- fusion of responsibility. Computers in Human Behavior, 18(5), 507– 520. https://doi.org/10.1016/s0747-5632(02)00007-9

Bastiaensens, S., Vandebosch, H., Poels, K., Van Cleemput, K., DeSmet, A., & De Bourdeaudhuij, I. (2014). Cyberbullying on social network sites. An experimental study into bystanders’ behavioural intentions to help the victim or reinforce the bully. Computers in Human Behavior, 31, 259–271. https://doi.org/10.1016/j.chb.2013.10.036

Blair, C., Foster Thompson, L., & Wuensch, K. (2005). Electronic help- ing behavior: the virtual presence of others makes a difference. Basic and Applied Social Psychology, 27(2), 171–178. https://doi.org/10. 1207/s15324834basp2702_8

Brody, N., & Vangelisti, A. (2015). Bystander intervention in cyberbul- lying. Communication Monographs, 83(1), 94–119. https://doi.org/10. 1080/03637751.2015.1044256

Darley, J., & Latan�e, B. (1968). Bystander intervention in emergencies: diffusion of responsibility. Journal of Personality and Social Psychology, 8(4), 377–383. https://doi.org/10.1037/h0025589

Davis, M. H. (1980). A multidimensional approach to individual differ- ences in empathy. JSAS Catalog of Selected Documents in Psychology, 10, 85.

Dillon, K., & Bushman, B. (2015). Unresponsive or un-noticed?: cyber- bystander intervention in an experimental cyberbullying context. Computers in Human Behavior, 45, 144–150. https://doi.org/10.1016/ j.chb.2014.12.009

Facebook (2016). News feed FYI: Taking into account live video when ranking feed. Newsroom. https://newsroom.fb.com/news/2016/03/ news-feed-fyi-taking-into-account-live-video-when-ranking-feed/

Facebook (2017). About Facebook live j Online streaming videos. Facebook Live. https://live.fb.com/about/

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G�power 3: A flexible statistical power analysis program for the social,

behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175–191. https://doi.org/10.3758/BF03193146

Fischer, P., Greitemeyer, T., Pollozek, F., & Frey, D. (2006). The unre- sponsive bystander: are bystanders more responsive in dangerous emergencies? European Journal of Social Psychology, 36(2), 267–278. https://doi.org/10.1002/ejsp.297

Fischer, P., Krueger, J. I., Greitemeyer, T., Vogrincic, C., Kastenm€uller, A., Frey, D., Heene, M., Wicher, M., & Kainbacher, M. (2011). The bystander-effect: a meta-analytic review on bystander intervention in dangerous and non-dangerous emergencies. Psychological Bulletin, 137(4), 517–537. https://doi.org/10.1037/a0023304

Forsyth, D., Zyzniewski, L., & Giammanco, C. (2002). Responsibility diffusion in cooperative collectives. Personality and Social Psychology Bulletin, 28(1), 54–65. https://doi.org/10.1177/0146167202281005

Freis, S., & Gurung, R. (2013). A Facebook analysis of helping behavior in online bullying. Psychology of Popular Media Culture, 2(1), 11–19. https://doi.org/10.1037/a0030239

Harari, H., Harari, O., & White, R. (1985). The reaction to rape by American male bystanders. The Journal of Social Psychology, 125(5), 653–658. https://doi.org/10.1080/00224545.1985.9712039

Hayes, A. (2013). Introduction to mediation, moderation, and condi- tional process analysis (1st ed.). Guilford Publications.

Hayes, A., & Montoya, A. (2017). A tutorial on testing, visualizing, and probing an interaction involving a multicategorical variable in linear regression analysis. Communication Methods and Measures, 11(1), 1–30. https://doi.org/10.1080/19312458.2016.1271116

Hayes, A., & Preacher, K. (2013). Statistical mediation analysis with a multicategorical independent variable. The British Journal of Mathematical and Statistical Psychology, 67(3), 451–470. https://doi. org/10.1111/bmsp.12028

Holfeld, B. (2014). Perceptions and attributions of bystanders to cyber bullying. Computers in Human Behavior, 38, 1–7. https://doi.org/10. 1016/j.chb.2014.05.012

Hortensius, R., Schutter, D., & de Gelder, B. (2016). Personal distress and the influence of bystanders on responding to an emergency. Cognitive, Affective & Behavioral Neuroscience, 16(4), 672–688. https://doi.org/10.3758/s13415-016-0423-6

Hudson, J., & Bruckman, A. (2004). The bystander effect: a lens for understanding patterns of participation. Journal of the Learning Sciences, 13(2), 165–195. https://doi.org/10.1207/s15327809jls1302_2

Kantrowitz, A. (2017). Violence on Facebook live is worse than you thought. BuzzFeed. https://www.buzzfeed.com/alexkantrowitz/heres- how-bad-facebook-lives-violence-problem-is?utm_term=.sybxrL5g2#. sd6ej87oY

Latan�e, B. (1981). The psychology of social impact. American Psychologist, 36(4), 343–356. https://doi.org/10.1037/0003-066X.36.4.343

Latan�e, B., & Darley, J. (1970). The unresponsive bystander: Why doesn’t he help?. Appleton-Century-Crofts.

Latan�e, B., & Nida, S. (1981). Ten years of research on group size and helping. Psychological Bulletin, 89(2), 308–324. https://doi.org/10. 1037//0033-2909.89.2.308

Lecher, C. (2017). Don’t blame users when violence on Facebook goes unreported. The Verge. https://www.theverge.com/2017/5/1/15478210/ facebook-violence-video-unreported-users-blame

Levine, M., & Crowther, S. (2008). The responsive bystander: How social group membership and group size can encourage as well as inhibit bystander intervention. Journal of Personality and Social Psychology, 95(6), 1429–1439. https://doi.org/10.1037/a0012634

Machackova, H., & Pfetsch, J. (2016). Bystanders’ responses to offline bullying and cyberbullying: the role of empathy and normative beliefs about aggression. Scandinavian Journal of Psychology, 57(2), 169–176. https://doi.org/10.1111/sjop.12277

Machackova, H., Dedkova, L., & Mezulanikova, K. (2015). Brief report: the bystander effect in cyberbullying incidents. Journal of Adolescence, 43(1), 96–99. https://doi.org/10.1016/j.adolescence.2015. 05.010

Markey, P. (2000). Bystander intervention in computer-mediated com- munication. Computers in Human Behavior, 16(2), 183–188. https:// doi.org/10.1016/s0747-5632(99)00056-4

8664 E. H. JUNG AND N. S. BHOJWANI

Martin, K., & North, A. (2015). Diffusion of responsibility on social networking sites. Computers in Human Behavior, 44, 124–131. https://doi.org/10.1016/j.chb.2014.11.049

McKenna, K., & Bargh, J. (2000). Plan 9 from cyberspace: The implica- tions of the Internet for personality and social psychology. Personality and Social Psychology Review, 4(1), 57–75. https://doi. org/10.1207/S15327957PSPR0401_6

Nickerson, A., Aloe, A., Livingston, J., & Feeley, T. (2014). Measurement of the bystander intervention model for bullying and sexual harassment. Journal of Adolescence, 37(4), 391–400. https:// doi.org/10.1016/j.adolescence.2014.03.003

Obermaier, M., Fawzi, N., & Koch, T. (2016). Bystanding or standing by? how the number of bystanders affects the intention to intervene in cyberbullying. New Media & Society, 18(8), 1491–1507. https:// doi.org/10.1177/1461444814563519

Okdie, B., & Guadagno, R. (2008). Social influence and computer mediated communication. In S. Kelsey & K. St.Amant (Eds.), Handbook of research on computer mediated communication (pp. 477–491). IGI Global.

Palasinski, M. (2012). The roles of monitoring and cyberbystanders in reducing sexual abuse. Computers in Human Behavior, 28(6), 2014– 2022. https://doi.org/10.1016/j.chb.2012.05.020

Schwartz, S., & Gottlieb, A. (1980). Bystander anonymity and reactions to emergencies. Journal of Personality and Social Psychology, 39(3), 418–430. https://doi.org/10.1037//0022-3514.39.3.418

Shotland, R., & Huston, T. (1979). Emergencies: what are they and do they influence bystanders to intervene? Journal of Personality and Social Psychology, 37(10), 1822–1834. https://doi.org/10.1037//0022- 3514.37.10.1822

Solomon, L., Solomon, H., & Maiorca, J. (1982). The effects of bystand- er’s anonymity, situational ambiguity, and victim’s status on helping.

The Journal of Social Psychology, 117(2), 285–294. https://doi.org/10. 1080/00224545.1982.9713438

Tanford, S., & Penrod, S. (1984). social influence model: a formal inte- gration of research on majority and minority influence processes. Psychological Bulletin, 95(2), 189–225. https://doi.org/10.1037/0033- 2909.95.2.189

Tsukayama, H. (2017). Facebook adds 3,000 employees to screen for violence as it nears 2 billion users. The Washington Post. https:// www.washingtonpost.com/news/the-switch/wp/2017/05/03/facebook- is-adding-3000-workers-to-look-for-violence-on-facebook-live/?nore- direct=on&utm_term=.fd5498b8aaf2

Van Cleemput, K., Vandebosch, H., & Pabian, S. (2014). Personal char- acteristics and contextual factors that determine “helping,” “joining in,” and “doing nothing” when witnessing cyberbullying. Aggressive Behavior, 40(5), 383–396. https://doi.org/10.1002/ab.21534

Voelpel, S., Eckhoff, R., & F€orster, J. (2008). David against Goliath? group size and bystander effects in virtual knowledge sharing. Human Relations, 61(2), 271–295. https://doi.org/10.1177/ 0018726707087787

About the authors

Eun Hwa Jung is an associate professor in the School of Media and Advertising at Kookmin University. Her research explores the psycho- logical effects of emerging media technologies, including social media, robots, and AI-based platforms, on users' perceptions and behaviors.

Nisha Suresh Bhojwani is a data analyst at Honest Financial Technologies. Her research interests include consumer behavior, organ- izational and operational optimization, and engaging clients within the Asia-Pacific region.

INTERNATIONAL JOURNAL OF HUMAN–COMPUTER INTERACTION 8665

Copyright of International Journal of Human-Computer Interaction is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.