More than 70% of Americans use social media platforms to post personal information, engage with posted content, and connect with others [1-4]. Adolescents and young adults were among the earliest internet and social media adopters and continue to use these websites at high levels [2,5,6]. Web-based challenges, or social media challenges, are popular phenomena, especially among adolescents and young adults, perhaps because of their frequent use of social networks. In these challenges, participants record themselves engaging in specific activities and share their experiences through social media platforms [6,7]. These challenges are ubiquitous and can be found on many social media platforms, including YouTube, Instagram, Facebook, and WhatsApp [8,9]. Although the activities involved in web-based challenges can vary from fun to fatal [10-13], they can generally be classified into 2 categories: (1) minimal harm challenges, which in some cases support a philanthropic cause such as the amyotrophic lateral sclerosis (ALS) ice bucket challenge [14], or (2) harmful challenges, which entail self-injurious behavior such as the cinnamon challenge [15]. Although the ALS ice bucket challenge has faced criticism (eg, safety concerns and waste of water), it is the most successful and influential fund-raising event to date [14]. In addition to raising more than US $115 million for ALS research [16], it is also credited for increasing public awareness about the disease [17].

In contrast, the cinnamon challenge involves swallowing a teaspoon of ground cinnamon without drinking any liquid for 60 seconds. The problem is that cinnamon does not dissolve or biodegrade in the lungs, as evidenced by animal-based laboratory studies, which experienced symptoms ranging from mild multifocal granulomatous inflammation to alveolar lipoproteinosis and alveolar cell hyperplasia [15,18-20]. The consequences are just as serious for humans because swallowing a large amount of cinnamon can cause pulmonary inflammation, allergic and irritant reactions, and even more serious situations, such as hypersensitivity-induced asthma attacks, which can be fatal [15]. However, none of these potentially fatal consequences have stopped adolescents and young adults from participating in the cinnamon challenge. As of 2013, there are more than 51,100 public YouTube clips of someone accepting this challenge, with some videos garnering more than 19 million views globally [15].

Given the significant amount of controversy concerning these web-based challenges, there is little research on the factors that lead individuals to participate in such challenges. For example, the extant literature on self-harm focuses primarily on a single challenge and its effect on public health and safety [15,21-23] or on how viewing content showing self-harm could lead to intentional self-harm by modeling the behavior of those we observe [24-26]. Furthermore, the literature on adolescent web-based risk focuses on the effects of engaging in web-based sexual and aggressive risk exposure [27,28]. To our knowledge, no quantitative research has comprehensively investigated the phenomenon of web-based challenges and why adolescents and young adults engage in these activities.

In this study, quantitative data were collected to explore adolescents’ and young adults’ exposure to web-based challenges and the determinants of their engagement with them through direct participation. The integrated behavioral model (IBM) [29] was used to investigate its generalizability to these behaviors on the web. It is important to reassess that and other existing behavioral theories concerning behaviors on the web because what may be true about traditional human behaviors may not apply to web environments [30].

As seen in Figure 1, IBM suggests that the intention to perform a behavior is driven by 3 factors: attitude, perceived norms, and personal agency regarding behavior. Attitude, defined as an individual preference for a certain behavioral performance, is composed of 2 dimensions: experiential attitude and instrumental attitude [31-33]. Experiential attitude is an individual’s emotional reaction to a behavior. For example, an individual with a positive emotional response toward a specific social media challenge is more likely to engage in it than an individual with a negative emotional response. Instrumental attitude is cognitively based, meaning that it is affected by a person’s beliefs about the outcomes of the behavior depending on the value of those outcomes.

Perceived norms regarding behavior and the social pressure to perform it are composed of injunctive and descriptive norms. Injunctive norms refer to the normative beliefs about others’ opinions toward participating in a challenge and the motivation to comply (if others approve or disapprove of the behavior). Descriptive norms refer to common patterns of behavior that lead to the expectations of people behaving according to that pattern.

Personal agency consists of 2 constructs: perceived control and self-efficacy. Perceived control refers to personal beliefs about the degree of control over performing the behavior. These beliefs are based on individual perceptions of how environmental factors will make the performance of the behavior difficult or easy. Self-efficacy is the individual’s certainty in their ability to perform the behavior in addition to their belief that they can overcome each prohibitive condition or obstacle [29].

The purpose of this study is to use IBM quantitatively to enhance our understanding of how each belief in IBM contributes to adolescents’ and young adults’ willingness to participate in web-based challenges. Another purpose of this study is to discern which beliefs are more influential than others. The findings from this study can be used to guide the development of interventions to reduce participation in harmful social media challenges among adolescent and young adult populations. Specifically, this research addressed the following research question: what is the effect, if any, of attitudes, perceived norms, and personal agency beliefs on adolescents’ and young adults’ willingness to participate in the cinnamon challenge and the ALS ice bucket challenge?

To explore our research question, we applied the IBM developed by Montano and Kasprzyk [29] depicted in Figure 1 to our hypotheses listed in Textbox 1.

A survey-based study was used to investigate the application of IBM in the prediction of social media challenge behavior among adolescents and young adults. The developed survey included measures of the IBM constructs, similar to the studies reported in the literature [34-36]. The survey was pilot tested and modified accordingly. Finally, the survey was deployed to a larger sample to explore the reasons for participation in these challenges, retrospectively.

Binomial logistic regression was used to understand whether participation in a social media challenge (ie, either the cinnamon challenge or the ALS ice bucket challenge) can be predicted from people’s attitudes, perceived norms, and personal agency beliefs. Participation in a social media challenge is a dichotomous dependent variable (ie, 1=participated or 0=did not participate), justifying the use of the binomial logistic regression [40]. The binomial logistic regression analysis was performed using SPSS 24.0 to predict cinnamon challenge participation first with 7 predictors: age group, experiential attitude, instrumental attitude, injunctive norm, descriptive norm, perceived control, and self-efficacy. Four interaction predictors were also added to the model: experiential attitude by the value of experiential attitude, instrumental attitude by the value of instrumental attitude, the injunctive norm by the motivation to comply, and the descriptive norm by the motivation to comply. Only the interaction terms recommended by IBM were included in the analysis [29] to ensure greater power to detect significant findings within our multivariate analysis [41]. Participants who had completed the cinnamon challenge and those who did not participate in any challenge were included in this model (n=316).

A second binomial logistic regression model was used to predict ALS ice bucket challenge participation using similar predictors assessing the participants’ perception of the ALS ice bucket challenge. Participants who had completed the ALS ice bucket challenge and those who did not participate in any challenge were included in the second model. For each model, fit indices, McFadden pseudo R², effect size estimates, estimated regression coefficients and their significance, and corresponding odds ratios and their confidence intervals were calculated.

All the data were checked to confirm the independence of observations; the existence of a linear relationship between an independent variable and the logit transformation of the dependent variable; and the absence of any multicollinearity, significant outliers, high leverage points, and highly influential points [40].

The results from the direct logistic regression model predicting cinnamon challenge participation are presented in Table 3. A test of the full model with all predictors against a constant-only model was statistically significant (χ²₁₁=221.8; n=316; P<.001), indicating that the predictors, as a set, reliably distinguished between people who had participated in the cinnamon challenge and those who had not. The deviance in participating in the cinnamon challenge accounted for by these predictors was large, with R²_L=0.5. To test each predictor’s significance, each variable was removed from the model, and the change in χ² was examined to determine if the removal of a variable led to a worsening of the model fit [42,43]. Independent removal of 4 of the 11 predictors significantly harmed the model fit, specifically instrumental attitude (Δχ²₁=11.5; P<.001), injunctive norm (Δχ²₁=30.4; P<.001), descriptive norm (Δχ²₁=6.6; P=.01), and the interaction term injunctive norm by the motivation to comply (Δχ²₁=8.8; P=.003). Figures 3-5 illustrate the form of these relationships.

To interpret the significant interaction for injunctive norm by motivation to comply, simple slopes were calculated from the regression coefficients at the mean of motivation to comply and 1 SD above and below the mean of motivation to comply [42]. This analysis found the slope of injunctive norm and probability to participate at the mean of motivation to comply to be β=1.01, and at 1 SD above and below the mean of motivation to comply to be β=1.80 and β=0.23, respectively. Figure 6 illustrates the form of this interaction.

A similar approach was used to predict ALS ice bucket challenge participation, with the results presented in Table 4. A test of the full model with all predictors against a constant-only model was statistically significant (χ²₁₁=151.1; n=318; P<.001; R²_L=0.34), meaning that, as a set, the predictors reliably distinguished between people who had participated in the ALS ice bucket challenge and those who had not. Independent removal of 2 of the 11 predictors significantly harmed the model fit, specifically experiential attitude (Δχ²₁=20.4; P<.001) and descriptive norms (Δχ²₁=9.6; P=.003). Figures 7 and 8 illustrate the form of these relationships.

To our knowledge, this is the first study to quantitatively investigate the theoretical constructs for predicting participation in web-based challenges using data provided by actual participants. This study aimed to investigate the behavioral beliefs of people who have participated in these challenges and compare them with the beliefs of people who did not. Thus, we identified potential factors that were critical to the participants’ final decision. The results showed the attitude subconstructs, the perceived norm subconstructs, and the interaction between injunctive norm and motivation to comply to be good predictors of cinnamon challenge participation. In addition, the experiential attitude and the descriptive norm are good predictors of ALS ice bucket challenge participation.

The analysis showed that attitude and perceived norm subconstructs (hypotheses 2, 5, and 6) are strong predictors of cinnamon challenge participation. This finding is consistent with other studies that used IBM to predict other behaviors such as condom use, which also found these 2 constructs to be the strongest predictors [34,44]. As seen in the Results section, the relationship between instrumental attitude, injunctive norm, and descriptive norm and probability of participating in the cinnamon challenge is proportional. The positive relationship between instrumental attitude and probability of participation indicates that the more people perceive enjoyment and rewards involved in the cinnamon challenge, the more willing they were to engage in the challenge. This shows that those people thought the challenge was easy, with minimal harmful consequences. In addition, the positive relationship between injunctive norms and the probability of participation shows that the more perceived attention paid to the challenge by the public, the higher the probability of participants engaging in the cinnamon challenge because they believe their videos will receive more views. In addition, our findings suggest that there is a positive relationship between descriptive norms and the probability of participating in the cinnamon challenge. This relationship means that the less attention participants received from people around them, warning them about participating in the challenge, the higher the chance they would engage in the challenge. Consequently, it appears that the more the peers were engaging in the challenge, the higher the likelihood that participants would engage in the challenge, as they may have believed it is a common behavior that is okay to do. These findings are similar to a previous study on criminal behavior, highlighting the significant role that culture plays in committing crime or violent behavior [45]. In other words, in a culture where crimes occur frequently, there is a higher chance of more people committing more crimes and violent behaviors in the future.

In addition, testing hypothesis 7 showed that there is a significant interaction between injunctive norms and motivation to comply. The interaction implies that there is a positive relationship between injunctive norms and the probability of participation in the cinnamon challenge. However, this relationship is stronger for those with low motivation to comply with scores. This means that people with low motivation to comply with predominant social norms are more likely to participate in the cinnamon challenge than those with high motivation to comply. This finding is different from most of the literature on human behavior, which suggests the opposite of our findings. This is mainly because of the negative nature of the behavior that this study investigates, which involves self-harm. For example, a person with low motivation to comply with their parents is more likely to commit a self-harm behavior than someone with high motivation to comply with their parents.

Analysis of the change in model fit after removing each of the significant predictors indicated that the injunctive norm explains most of the variability in the probability of cinnamon challenge participation, followed by instrumental attitude and descriptive norm. Thus, interventions to reduce participation in similar challenges in the future should focus on these constructs, with greater emphasis on the injunctive norm, as it is the stronger predictor. This could be done by having people adolescents trust send persuasive messages highlighting the consequences of challenge behavior and explaining why they should not engage in these activities [46,47]. In addition, as there is a significant interaction between injunctive norms and motivation to comply, intervention development should consider both of these factors simultaneously. Changing only 1 of these 2 factors may lead to an undesired or unintended effect on other’s impact on challenge participation. The intervention should specifically mention the disapproval of such behaviors from those around us, even those who say they do not comply or care about what others say.

Unlike cinnamon challenge participation, only the experiential attitude and descriptive norm significantly predicted ALS ice bucket challenge participation (hypotheses 2 and 6). In other words, adolescents and young adults primarily participated in this challenge for two reasons (1) enjoyment or popularity (getting more views and likes on social media) and (2) a sense of obligation due to the large number of participants completing the challenge, which made them feel that it’s the norm to do so [48-51]. We believe that other factors were not significant because of the positive nature of the challenge. For example, even people who did not participate in the challenge generally rated it as easy to perform and believed that they were capable of completing it. These beliefs may explain why perceived control and self-efficacy factors were not found to be significant predictors of ALS ice bucket challenge participation. In addition, as this challenge, in particular, was very popular, even people who chose not to participate generally indicated that everyone around them would approve of their participation. This explains why the injunctive norm factor was not found to be a significant predictor of ALS ice bucket challenge participation.

Among the significant predictors of ALS ice bucket challenge participation, experiential attitude explained the largest amount of variability, followed by the descriptive norm. These findings can help develop or market other philanthropic challenges by focusing on making them enjoyable with obvious direct rewards and emphasizing the attention given to them by the public. By developing a challenge that targets these beliefs more than the others in IBM, one could potentially create a philanthropic challenge that goes viral and leaves a health-promoting impact on society with minimal harmful consequences.

This study has several limitations. Only 2 challenges were used to represent all other similar challenges. This could limit the generalizability of the findings; hence, future work could investigate the applicability of these findings to other challenges. Moreover, this study was retrospective and cross-sectional in nature, making it difficult to draw conclusions about causal relationships between the predictors and outcomes. Future work could study the impact of the constructs in controlled settings by developing interventions and examining their effects on people’s willingness to participate in social media challenges.

A theoretical framework was used to guide the study design and to inform the development of theory-driven intervention efforts to change social media challenge participation intention and behavior. The cinnamon challenge was used to represent challenges with a harmful impact, and the ALS ice bucket challenge was used to represent positive-impact challenges. The constructs that were critical to the participants’ decision to participate were identified. This study provides a good theoretical model to understand the phenomenon of social media challenges. In addition, the findings provide information about which constructs should be the focus of intervention efforts. The content and thrust of those intervention efforts must be based on knowledge of how the specific items making up each construct apply specifically to social media (eg, the desire to get likes and affirmation and the social norms that are portrayed via media, videos, and images).