Bystander intervention trainings and messages are two of the main ways universities in the United States attempt to address the sexual violence crisis in college campuses [1,2]. Although it is common for bystander intervention messaging (here, bystander intervention is defined as the ability to identify situations in which a third party may assist someone in a risky and act to intervene [3]) to use bystander models as a theoretical framework [4,5], the social norms approach is also used to address sexual violence interventions [6]. For the social norms approach to be successful, it is necessary that the population of the message experiences pluralistic ignorance, where there is a difference between their own behavior and their estimate of peer behavior [6]. The social norms approach suggests that people who believe that their peers are engaging in a behavior are more likely to engage in that behavior themselves [6], whereas bystander models, in particular bystander barriers, suggest that the belief that others will take responsibility removes their own responsibility to engage in bystander intervention [7]. Many trainings and programs exist that focus on stimulating bystander intervention behavior [8,9], but less is known about the role of supporting media messages. In addition, it is necessary to investigate how these 2 opposing perspectives may be perceived as messaging mechanisms for college students.

There is a sexual assault crisis on college campuses. Estimates indicate that approximately 1 in 4 college women are sexually assaulted during their time in college [1], which is high compared with the rate of 1 in 5 among the general population of women in the United States [2]. In addition, certain populations seem to be more at risk, such as first-year students [10] and LGBTQ+ (lesbian, gay, bisexual, transgender, queer+) students [1]. Universities across the country have tried to curb the problem by using a series of prevention efforts against sexual violence, with one of the most popular ones being bystander intervention [11].

Bystander intervention training and campaigns, including in-person and online training, are effective ways of increasing intervention intentions among college students in multiple scenarios [5,12]. An example of these campaigns is Green Dot, which features a focus on motivational speech and in-person training [12]. Over 500 universities in the United States have adopted Green Dot [13]. It is worth noting, however, that the bystander effect has been brought into question [14,15]. This can be in part due to its questionable origins in the Kitty Genovese issue [16]. Even still, this event led to an eventual refinement of what it means to intervene as a bystander, and the trainings developed from there proved to be successful [4,5,12,17,18].

Using Instagram as a platform for public health concerns has been studied mostly in an exploratory manner [19]. However, some studies suggest that the issue is not necessarily reach but the ability of the entities governing these accounts to communicate with their publics [20]. In addition, such messaging developed for Instagram can be easily adapted for other channels, facilitating eventual scalability efforts.

Instagram reels, in particular, are more popular than regular Instagram posts and stories because they have better reach and can be found more easily than regular posts [21]. They also have more editing features that can be used to make messages engaging for the audience [21]. Instagram is also highly used among college-aged populations, more so than other platforms [22], such as Twitter (now X), Facebook, and Snapchat. Thus, the Instagram reel-inspired format can be an optimal channel for short and targeted messaging.

In this study, a formative survey was conducted to assess if pluralistic ignorance exists in a college population in relation to bystander intervention behavior in sexual assault situations. This step needed to be established first to ensure a social norms–based message would be feasible. In the presence of pluralistic ignorance, 3 theoretically based Instagram reel-based video messages with short scripts messages were developed and piloted in terms of credibility, intended audience, perceived message effectiveness, and a manipulation check of the conditions. One condition on bystander behavior barriers, focusing on showing participants only the actual behavioral data reported in the survey, was called the message readjustment condition in this study; another condition was based on the social norms approach, which was called the norm reinforcement condition in this study, as it focused on showing participants only the normative information from the data collected in the survey; and 1 combined both theoretical frameworks, showing both the normative information and the behavioral information with the intent of highlighting the discrepancy between the 2. The goal of this study is to assess the perceived message effectiveness, credibility, and intended audience reach of these messages pertaining to bystander intervention.

In the first semester of 2023 in a large, public, American university, data were collected from 195 undergraduate students aged 18 to 29 years (mean 20.52, SD 2.22 years), most of whom identified as women (118/195, 60.5%), cisgender (184/195, 94.4%), heterosexual (123/195, 63.1%), White (139/195, 71.6%), not Hispanic (110/195, 56.7%), and middle class (124/195, 63.6%). Participants were recruited from a randomized list of 2000 emails from the university’s registrar’s office, representative of the university as whole.

Table 1 demonstrates that there was a discrepancy between peer perception and actual behavior of participants. The means were higher among the normative statements (bottom 3) as compared with their actual behavior items counterparts, indicating that participants said they more strongly agreed with the normative statements. This indicates that there was an opportunity for a social norms message to be developed, as pluralistic ignorance was identified.

Messages were developed based on the items tested in the formative research survey step. A total of 12 messages were developed, 3 per condition. The experimental conditions and the theoretical basis for each can be found on Table 2.

Data were collected in October of 2023. This data collection was exempt from institutional review board (IRB) review for similar reasons as the formative survey, where the university’s IRB reviewed the proposal and indicated that a full-board review was not necessary considering the minimal risks. To be eligible to participate, participants had to be currently in the United States, be an undergraduate student at the university being studied, and be between the ages of 18 and 29 years. Participants were recruited through a pool of students taking communication courses and were compensated with extra credit. Other recruitment methods were attempted, such as social media recruitment, but they yielded very few participants. These participants were compensated with Starbucks gift cards. These 3 conditions were added as screeners to ensure only participants who conformed to them participated in the pilot. This type of pilot testing strengthens the validity and reliability of the intervention, and it has been used in similar efforts [24,25].

Participants saw 5 of 12 Instagram reel-based short video messages developed in random order (3 per condition: norm readjustment, norm reinforcement, combined condition, and control). The following instructions were provided to the participants before they started the pilot: “Next, you will see 5 short videos. You will answer some questions about those videos and then some overall questions. Please make sure you can hear the videos as you watch them. Please click on the button on the bottom right to continue.” Participants were also provided with the following description of bystander intervention: “Definition of bystander intervention: Bystander Intervention is recognizing a potentially risky situation or interaction and choosing to respond in a way that could positively affect the outcome. For the purposes of this study, we are mostly referring to intervening in situations that could escalate to sexual violence.” As these messages had to be highly tailored, the messages referred to the students with the name of the university’s mascot (eg, if the university’s mascot was a tiger, the participants were called tigers). These references were removed from this paper to ensure the anonymity of the participants and an unbiased peer review process.

A total of 107 individuals participated in the pilot, but 81 participants responded to demographics items. Participants’ ages ranged from 18 to 24 years (mean 20.16, SD 1.33 years). Most identified as women (52/81, 64.2%) and cisgender (80/81, 98.8%). Most identified as heterosexual (60/81, 74.1%), but a considerable portion of the sample either chose a sexual minority (15/81, 18.4%) or chose not to answer (6/81, 7.4%). Most participants were non-Hispanic/Latine (44/81, 54.3%) and were Caucasian/White (61/81, 75.3%). Most participants also self-identified as middle class (58, 71.6%).

The messages were designed on Canva, using available music and visuals. The script was created based on the bystander items described in this paper. The scripts are approximately 10‐15 words from each other in terms of length. Please refer to Textbox 1 for a complete breakdown of each initial message, the condition that it belongs to, and the number by which the message is referred to in the Results section. Figure 1 shows an example of the message.

To evaluate the qualitative open-ended questions, thematic analysis was used [28,29]. Thematic analysis was chosen instead of other potential qualitative methods to allow for a more organic identification of themes and discussion among coders. The coders were the author and a colleague in the same discipline as the author. The first step in Braun and Clarke’s [28,29] thematic analysis is to become familiarized with the data. The coders read all the responses before starting to code them. After that, the responses were prepped in a spreadsheet format. The coders highlighted cells based on how they may represent a theme. If a segment may be referring to 2 or more themes, coders were to add a column next to the column where the segments were and highlight the cell next to the original segment. This procedure repeated itself for each question (likes, dislikes, and purpose) and per message (9 messages total). Although 12 messages were tested, the qualitative coding only happened for the messages that were related to bystander intervention and not the control messages (messages 10‐12). Thus, there is a total of 24 units to analyze, where each has between 28‐37 segments as responses. A total of 930 segments were coded. Then, we used 1 set of messages to train, where we coded the qualitative responses related to message 1 while we were together, but the coding happened separately. During the training, we established the scope of the themes, and we established that for the priming question, we were coding for 2 themes: participants understood the message, and participants did not understand the message. After coding the segments associated with message 1, we found that we agreed in most instances on the development of themes and how we approached the priming segments. We used the segments associated with messages 1‐4 to establish the most common themes and used those themes to code the rest of the segments in messages 5‐9. Thus, approximately 44% of the data were used to establish the themes, and those themes were reinforced using the rest of the data. We then moved on to code the segments associated with messages 2‐9 by ourselves, and compared the coding of the coder and my coding to assess if there was a need for us to meet again and discuss any major disagreements. Minor disagreements were addressed by email, phone calls, and text messages.

The pilot was assessed using descriptive statistics only, as the goal of this stage of the messages is to check for credibility, perceived audience, manipulation checks, and perceived effects. Means and frequencies were used to make these assessments.

Participants were compensated with the opportunity to enter a drawing to receive a gift card by Tango Cards in the value of US$100 (three available cards), US$50 (six available cards), or US$25 (eight available cards) in the formative survey and with extra credit in a communication course of their choice for their participation in the pilot for most participants, with very few receiving Starbucks gift cards instead. Data collection happened through an online survey and was collected anonymously. The drawing information was completely unlinked from the survey responses. This study was exempt from review by the author’s institution. Exempt review means the protocol was reviewed by a member of the university’s IRB, but a full-board review was not deemed necessary, as risks were considered minimal.

In this study, the feasibility of a social norms message was assessed by the means of a formative survey study, demonstrating that there was indeed a discrepancy in terms of pluralistic ignorance. The presence of pluralistic ignorance allows for the development of social normative messaging. After establishing the presence of pluralistic ignorance, a message was piloted, and results indicated that the messages reached the intended audience, had the intended perceived effect, were perceived as credible, and the manipulation checks indicated the conditions were properly differentiated from each other while representing their purpose. The feasibility of messages focusing on bystander barriers and combining bystander barriers and the social norms approach was also explored. These messages were also highly accepted among the participants, as well as highly understood. Further research is necessary to explore which theoretical approach is ideal to reach this population in terms of the topic of bystander intervention in sexual assault situations. This was the first study to consider combining opposing approaches to bystander intervention messaging—one where it is suggested that people should consider the behavior of their peers [6] and one where they should take more responsibility for their own actions [7].

Although all 3 approaches showed appropriate levels of perceived message effect and credibility, it is necessary to underscore that the reinforcement condition and the combination condition were particularly successful. This suggests that the bystander barrier associated with taking responsibility may not be as strong of a factor in bystander intervention messaging as it is an attitude to be changed by messaging. The fact that the common denominator among the more successful messages is the presence of normative information is in line with the current literature surrounding sexual violence prevention in the form of media messages [30] as well as theoretical frameworks used to address sexual violence prevention [6,7], further building on it by adding a component that goes beyond norms only. However, more must be explored in terms of combined message by proper experimental designs.

The discrepancy between the qualitative and quantitative results of the pilot in terms of credibility highlighted the need to conduct mixed methods pilots to holistically assess messages that concern public health. Although the quantitative results indicated that credibility was appropriate as is, the qualitative data indicated that there were ways to improve credibility even more. The message should be adjusted before further evaluation of its effects. Possible reasons for this discrepancy could be that, although participants find the message credible overall, as in, the message had face validity to them, they would expect proper sourcing of this information. When asked quantitatively if the messages were credible, they may tend to say it is credible, but when they have the space to expand on that, they explain how this credibility could become more formal.

In terms of practical implications, this study shows the possibility for universities and other organizations that conduct bystander intervention training to provide their publics with messages that can underscore these trainings. It is unlikely that an Instagram reel-based message alone can change one’s behavior. However, media messages can serve to reinforce adopted norms [30-32]. This can, in turn, maintain a change in behavior from a more complex intervention that goes beyond media exposure.

This study, although novel, is not without limitations. The messages were highly tailored for the students of the university where the study was conducted. Although this is ideal for social normative messaging, it can interfere with the scalability of the study. Furthermore, more assessments using inferential statistics are necessary to establish the potential effects of these messages. These are the next steps in this line of research. In addition, the messages lacked proper sourcing, which participants identified particularly in the qualitative results as a flaw. Future developments of these messages should provide proper sourcing to enhance the credibility of this intervention.

Overall, this study explores how different theories can contribute in different ways to the same public health concern. The messages assessed here have proven to be adequate for the next stage of evaluation with an experimental design. However, this next step would not be as rigorous if the procedures discussed here had not been conducted first.