Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Feb 1.
Published in final edited form as: Health Commun. 2024 May 6;40(2):332–344. doi: 10.1080/10410236.2024.2344883

Modification Mechanisms of Descriptive Norm Perceptions Toward Vaping: The Role of Behavior Prevalence and Group Size in An Online Setting

Jiaying Liu 1, Rui Shi 2, Robert C Hornik 3
PMCID: PMC11538375  NIHMSID: NIHMS1990960  PMID: 38711233

Abstract

We often rely on descriptive norm perceptions as a mental shortcut for decision making. However, less is known about how such perceptions are shaped and modified by our experiences in day-to-day life. The interactive nature of the current media environment offers opportunities for individuals to access others’ health behavior choices through online user-generated content. Within a setting of online comment boards, the current study examined the descriptive norm perception modification process toward vaping with a large-scale experiment that systematically varied levels of exposure to online commenters’ vaping behavior choice indications. Findings revealed a significant positive effect of behavior prevalence on descriptive norm perceptions, which in turn were positively associated with vaping intention. This set of results was observed only when a sufficient total amount of comment exposures was ensured. The study provided empirical evidence for the underlying mechanism of the “quasi-statistical sense,” which helps people draw conclusions about behavior prevalence and may influence their behavioral decision making. Theoretical and practical implications are discussed.

Keywords: social influence, descriptive norms, online comments, exposure dosage, quasi-statistical sense, e-cigarette use

One of the most striking regularities of human cognition is the tendency of human minds to engage in decision-making through mental shortcuts (Fiske & Taylor, 1991). A mental shortcut people frequently resort to is social consensus information, i.e., the opinions or behavioral decisions of the majority in similar situations (Festinger, 1954). The underlying assumption is that the chance of the majority being wrong is acceptably low. Such information can manifest either in the forms of injunctive norm perceptions, i.e., what most others approve, or descriptive norm perceptions, i.e., what most others actually do (Cialdini et al., 1990; Lapinski & Rimal, 2005). While often related, injunctive norm perceptions are motivated by a desire for social acceptance, whereas descriptive norm perceptions are driven by the motivation to gain a more accurate interpretation of social reality in order to adapt (Bergquist & Nilsson, 2019; Cialdini, 2005). The latter is seen as forming more automatically since individuals are often unable to identify others’ actions as causal for their behaviors (Nisbett & Wilson, 1977). Despite evidence supporting the powerful influence of descriptive norm perceptions on behaviors, how they form and evolve remains insufficiently understood (M. Chung et al., 2022).

The current study attempts to enrich this line of inquiry by elucidating the underlying descriptive norm perception modification process with a large-scale experiment that systematically varies exposure dosage to behavior cues embedded in online user-generated comments. The study is situated in the context of electronic cigarette (e-cigarette) use, also known as “vaping” behavior. Although some evidence has suggested that e-cigarettes could serve as a quitting tool for combustible cigarette users (Kasza et al., 2021), vaping is also linked to potential health risks, including cardiovascular and respiratory problems, chemical poisoning, neurodevelopmental issues, and other long-term health implications that are not yet fully understood (Echeagaray et al., 2022; López-Ojeda & Hurley, 2024; Xie et al., 2020). Recent surveillance data indicates that as the percentage of cigarette smoking adults in the U.S. has reached an all-time low, e-cigarettes are gaining increasing traction (Cornelius et al., 2023). The growing popularity of vaping is alarming and may derail decades of tobacco control success.

Descriptive norm perception is likely important in shaping people’s vaping intention and behavior. Longitudinal cohort data has revealed a significant positive association between the number of close friends vaping and subsequent experimentation with e-cigarette use among established smokers (Yong et al., 2023). There is also evidence that overestimation of vaping prevalence can increase one’s vaping curiosity and susceptibility (Oh et al., 2022). Although many previous studies have examined the associations between vaping norms and individuals’ vaping decisions (e.g., Aleyan et al., 2019; Yong et al., 2023), very few have investigated how descriptive norm perceptions toward vaping are formed and evolved. Understanding such processes is particularly important in online settings, which have become increasingly popular and influential for health information dissemination (Huo et al., 2019).

Descriptive Norm Perception Modifications through Online Comments

Previously, individuals needed summary information of behavior prevalence or observable behaviors in their physical surroundings to acquire descriptive norm information (Tankard & Paluck, 2016). However, the interactive nature of the new media environment offers opportunities for individuals to access others’ opinions and behavior choices through online user-generated contents (Neubaum & Krämer, 2017; Walther & Jang, 2012). A multitude of studies have demonstrated how online comments, a most typical type of user-generated content, successfully sway viewers’ perceptions (Anderson et al., 2014; Lee & Tandoc, 2017; Shi et al., 2014; Walther et al., 2010). Most studies have focused on the effects of evaluative contents (e.g., Walther et al., 2010) or language use (e.g., Anderson et al., 2014) in the comments. Although an emerging body of research has begun to examine how comments may affect norm perceptions (e.g., J. E. Chung, 2018; Liu & Shi, 2019), systematic investigations have remained thin.

Online commentaries contain scattered individual behavior cues, which means a single online comment typically does not provide an explicit summary of the prevalence of a behavior as reports or news often do. But when taken together, these individual behavioral cues (such as mentions of personal engagement or non-engagement with a behavior) can enable individuals to form an overall impression of the behavior choice distribution among the commenters. It may be that individuals come to a comment board with pre-existing views; however repeated exposure to information about commenters’ behavior may modify their estimates of behavioral norms.

The spiral of silence theory (Noelle‐Neumann, 1974) may cast light on understanding the process of descriptive norm assessment. According to the theory, people have evolved a quasi-statistical sense for gauging opinion or behavior choice distribution (Scheufle & Moy, 2000). This process happens automatically, without conscious numerical calculation (Petersen, 2012). To test this idea in the context of norms conveyed by online comments, Liu and Shi (2019) manipulated behavior choice distributions as indicated in comments in the setting of a news website. The study confirmed that participants were able to piece together all the scattered and subtle behavior cue information and correctly gauged the dominant behavior choice, which influenced their estimation of the behavior prevalence in the United States.

Behavior Prevalence and Group Size

In experimental settings, no consensus has been reached regarding the best manipulation to influence individuals’ perceptions of a descriptive norm. Past research has operationalized static descriptive norm as 1) the percentage of people engaging in a behavior (e.g., “69% of all customers buying seafood in our shop yesterday chose MSC/ASC”, Richter et al., 2018), 2) a vague description of the proportion of people engaging in a behavior (e.g., “Most students here choose to eat vegetables at lunch”, Guichard et al., 2021), 3) the total number of people performing a behavior (e.g., “Every day more than 150 students have a tossed salad for lunch here”, Mollen et al., 2013), or 4) the average performance of other people (e.g., “The average number of steps taken by the participants was 8000 yesterday”, Wally & Cameron, 2017). There has not been any direct comparison of these manipulation methods, and sometimes they were considered interchangeable in terms of inducing descriptive norm perception changes.

Behavior prevalence.

Since descriptive norm perception is, by definition, people’s perception of behavior prevalence, it is reasonable to expect that people’s descriptive norm perception correlates with the actual behavior prevalence they are exposed to. For descriptive norm perceptions to automatically come to mind at the time of decision making, chronic accessibility to perceived behavior choice distribution needs to be stably forged in one’s mental shortcuts. Such chronic accessibility can be achieved through repeated incidents of experience, including vicarious experiences from media exposure (Fiske & Taylor, 1991; Higgins, 1996; Rhodes et al., 2008). Sufficient exposure to others’ behavior provides reinforcement opportunities, leading to an increased sense of familiarity, ease of processing, and heightened salience of the majority choice (Cialdini et al., 1990).

In the current study context, “behavior prevalence” is conceptualized as the proportion of online comments that contain behavior cues indicating the vaper identity of the commenter (or someone s/he knows); presumably, the higher the proportion of vaper-norm comments on the comment board, the more likely the viewers will perceive a higher prevalence of e-cigarette use. Given that online comment boards in the real world are rarely unanimous in opinions and preferences expressed, viewers are likely to be exposed to a mix of behavior cues with opposite normative directions. Some comment boards could attract comments from mainly vapers and thus develop a vaper-dominant norm, while other boards may attract more non-vapers to form a dominant non-vaper norm. This conceptualization echoes the use of the other-total ratio in previous social influence studies, which have advocated that the ideal exposure metric should consider both majority and minority influences to describe how increases in majority size may raise awareness of the heightened unpopularity of the minority position (Mullen, 1983; Stasser & Davis, 1981). By taking into account both vaper and non-vaper norm influence, we aim to examine if increases in the percentage of vaper-norm comments are linked to increases in descriptive norm perceptions about vaping. We thus put forth the following hypothesis:

  • H1: Vaping behavior prevalence, operationalized as the percentage of vaper-norm comments out of the total number of comments on the online comment board, is positively associated with descriptive norm perceptions on vaping.

Group size.

Relatedly, group size, or the size of the overall evidence pool, also has implications for the salience of the majority category (Higgins, 1996; Tversky & Kahneman, 1973). Group size in this study is defined as the total number of comments on the comment board. The widely-known Asch conformity studies could shed some light on the effects of group size (Asch, 1951, 1955, 1956). In an in-person group setting, all confederates were instructed to unanimously select an obviously incorrect answer. Asch observed that with a group size of two (i.e., two confederates), participants conformed about 14% of the time; when the group size increased to three, the conformity rate jumped to 32%. The results suggested that there may be a positive association between group size and willingness to conform, provided that all group members remain in agreement without any expression of opposition or dissenting opinions.

Latané and Wolf in their social impact theory (1981) modeled the social influence process as a negatively accelerating curve such that each additional group member holding the majority opinion will have a smaller impact than that of the preceding member. For both Asch and Latané’s work, unanimity in the group is a prerequisite. Essentially, they were examining the impact of the group size when the behavior prevalence was kept constant at 100%.

Mullen (1983) adopted a self-attention theory approach and suggested that individuals within a group will become more attentive to the alignment of their own behavior with the perceived standard of appropriate behavior, when the relative size of their subgroup decreases. Mullen proposed a term called the Other-Total Ratio to describe the influence of the group as the proportion of the majority out of the group total (in our case: influence = number of vaper comments / number of total comments). Mullen’s model is the most relevant to the current study as it considers group influence when opinions are not necessarily unanimous in the group. The Other-Total Ratio perspective would predict that mere increases in overall group size would not affect personal judgment as long as the behavior prevalence remains consistent.

Therefore, in the current study, we propose to examine the potential role of group size, both as a main source of influence and as a moderating influence that may qualify the effect of behavior prevalence of the target behavior choice. Given the mixed findings from prior research regarding the main effect of group size, we propose the following research question:

  • RQ1: How does group size, operationalized as the total number of comments, affect people’s descriptive norm perceptions on vaping?

Regarding the potential moderating role of overall group size, a larger information pool provides greater exposure to the target normative direction and potentially affords more credibility. However, when the group allows information on different normative directions, a larger group size also means more exposure to comments supporting the other side, thus potentially diluting the salience of the majority behavioral prevalence. We thus propose:

  • RQ2: How does behavior prevalence affect descriptive norm perceptions at different levels of overall group size?

The Behavioral Implications

While it is established in the literature that descriptive norm perceptions are crucial for intention and behavior changes (Ajzen, 1991; Fishbein & Ajzen, 1975), most research has operationalized them as summary prevalence information (Rhodes et al., 2020; Tankard & Paluck, 2016). Although the theoretical accounts support the expectation that descriptive norm perceptions formed through accumulative exposure to individual behavior cues would have a positive effect on intentions, direct empirical evidence of the behavioral implications is lacking.

Additionally, social influence scholars have underscored the positive implications of behavioral changes facilitated by private acceptance, where individuals internalize the opinions or behavioral choices of others, leading to true modification of their own beliefs and perceptions (Cialdini & Goldstein, 2004). It has been suggested that private acceptance is likely to occur when one can make a choice in the absence of the influencing group or when they are allowed to respond anonymously (Newcomb, 1967). Our anonymous, non-coercive online comment board setting may hold the potential to foster genuine internalization of norms. Thus, our last hypothesis concerns whether normative information in the comments can meaningfully translate into norm-congruent intentions through descriptive norm perception changes:

  • H2: Descriptive norm perceptions mediate the association between behavior prevalence and vaping intention.

Method

Study Design and Procedures

Considering that comments do not appear in isolation but are often juxtaposed with their associated media productions, the experiment was situated in an online news website mock-up, where participants were exposed to a news article about e-cigarettes first and then directed to view different comment boards. To systematically examine behavior prevalence effects, we used two treatment factors to generate the experimental conditions: the total number of comments (i.e., group size manipulation) ranging from 1–20 (20 levels), and the number of vaper-norm comments1 ranging from 0–20 (21 levels). Due to the constraint that the number of vaper-norm comments cannot exceed the total number of comments on a single comment board, a full crossing of the two factors was not possible. For instance, if the total number of comments was set to N=5, the number of vaper-norm comments could only take the values from 0 to 5 instead of covering the entire range of its 21 levels. This partial factorial design led to a total of 230 conditions, as demonstrated by the non-vacant cells in Table 1. The focal independent variable in this study, behavior prevalence (i.e., percentage of vaper-norm comments), was then calculated as the ratio of the two treatment factors (number of vaper-norm comments / total number of comments) for each condition, as shown by the percentages presented in Table 1.

Table 1.

All Experimental Conditions (N = 230) Varying Total Number of Comments and Number of Vaper-Norm Comments

Total#╲Vaper-norm# 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 0.0% 100.0%
2 0.0% 50.0% 100.0%
3 0.0% 33.3% 66.7% 100.0%
4 0.0% 25.0% 50.0% 75.0% 100.0%
5 0.0% 20.0% 40.0% 60.0% 80.0% 100.0%
6 0.0% 16.7% 33.3% 50.0% 66.7% 83.3% 100.0%
7 0.0% 14.3% 28.6% 42.9% 57.1% 71.4% 85.7% 100.0%
8 0.0% 12.5% 25.0% 37.5% 50.0% 62.5% 75.0% 87.5% 100.0%
9 0.0% 11.1% 22.2% 33.3% 44.4% 55.6% 66.7% 77.8% 88.9% 100.0%
10 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0% 90.0% 100.0%
11 0.0% 9.1% 18.2% 27.3% 36.4% 45.5% 54.5% 63.6% 72.7% 81.8% 90.9% 100.0%
12 0.0% 8.3% 16.7% 25.0% 33.3% 41.7% 50.0% 58.3% 66.7% 75.0% 83.3% 91.7% 100.0%
13 0.0% 7.7% 15.4% 23.1% 30.8% 38.5% 46.2% 53.8% 61.5% 69.2% 76.9% 84.6% 92.3% 100.0%
14 0.0% 7.1% 14.3% 21.4% 28.6% 35.7% 42.9% 50.0% 57.1% 64.3% 71.4% 78.6% 85.7% 92.9% 100.0%
15 0.0% 6.7% 13.3% 20.0% 26.7% 33.3% 40.0% 46.7% 53.3% 60.0% 66.7% 73.3% 80.0% 86.7% 93.3% 100.0%
16 0.0% 6.3% 12.5% 18.8% 25.0% 31.3% 37.5% 43.8% 50.0% 56.3% 62.5% 68.8% 75.0% 81.3% 87.5% 93.8% 100.0%
17 0.0% 5.9% 11.8% 17.6% 23.5% 29.4% 35.3% 41.2% 47.1% 52.9% 58.8% 64.7% 70.6% 76.5% 82.4% 88.2% 94.1% 100.0%
18 0.0% 5.6% 11.1% 16.7% 22.2% 27.8% 33.3% 38.9% 44.4% 50.0% 55.6% 61.1% 66.7% 72.2% 77.8% 83.3% 88.9% 94.4% 100.0%
19 0.0% 5.3% 10.5% 15.8% 21.1% 26.3% 31.6% 36.8% 42.1% 47.4% 52.6% 57.9% 63.2% 68.4% 73.7% 78.9% 84.2% 89.5% 94.7% 100.0%
20 0.0% 5.0% 10.0% 15.0% 20.0% 25.0% 30.0% 35.0% 40.0% 45.0% 50.0% 55.0% 60.0% 65.0% 70.0% 75.0% 80.0% 85.0% 90.0% 95.0% 100.0%
Open in a new tab

Note: X-axis: total number of comments; Y-axis: number of vaper-norm comments. Percentage (i.e., behavior prevalence) shown in each cell was calculated by dividing the number of vaper-norm comments by the total number of comments in that condition. Number and percentage of non-vaper-norm comments are linear transformations of that of the vaper-norm ones in each condition (e.g., 20% vaper-norm condition = 80% non-vaper-norm condition).

Considering the subtlety of our norm manipulation with individual behavior cues, the overall effect size is expected to be small. We thus used η2 =.01 or Cohen’s f of about 0.10 (Cohen, 1988; Rosenthal, 1991) for the power analysis using G*Power 3.1 (Faul et al., 2009). A minimum total sample size of 1,130 was required to achieve the statistical power of .80 in detecting this effect. We thus randomly assigned about n=5 eligible participants to each of the 230 conditions, with the aim of recruiting at least N=1,150 participants to allow a reliable point estimate for each independent variable. By implementing true randomization (i.e., randomization elements were not forced to be displayed the exact same number of times), the number of participants for each condition ranged from 4 to 6. Our final sample for analysis was N=1,159.

The current study used online Qualtrics-based surveys distributed through Amazon Mechanical Turk (MTurk). We recruited high-quality Mturk workers and implemented screening procedures to ensure the credibility and reliability of responses (Online Supplement Appendix A). Following the completion of eligibility questions and the disclosure of their ever e-cigarette use status (i.e., whether they have ever used an e-cigarette, even one or two puffs), all participants were initially presented with the same short news article about e-cigarettes. They were then directed to read comments that varied in exposure to individual behavior cues based on the conditions they were assigned to. Their descriptive norm perceptions, intentions, and demographics (including age, gender, race/ethnicity, and education) were assessed afterwards. All study protocols were approved by the Institutional Review Boards at the authors’ institution.

Participants

A total of 1,159 eligible U.S. adults were recruited through MTurk. More than half of the participants were female (53.06%), and the mean age was 38.06 (SD = 12.62 ranging from 18 to 87. About half of the sample had finished high school or had some associate degree (48.58%), and 30.88% had finished college. The majority of the participants were White (77.31%), 7.94% African American, 5.69% Asian/Pacific Islander, 5.87% Hispanic/Latino, and 3.19% Others. Most of the participants had heard of e-cigarettes before the study date (97.41%), among which a sizeable portion had ever used an e-cigarette, including one or two puffs (42.62%). Given the widespread knowledge and prior e-cigarette usage among our sample, it is reasonable to assume that participants already held pre-existing descriptive norm perceptions about vaping before the study. Our experimental objective was to induce modifications to these existing perceptions. It is worth noting that we refrained from measuring participants’ pre-existing norm perceptions to avoid priming or making them overly sensitive to normative information during the experimental manipulation.

Stimulus Materials

News article.

The news stimulus was created by modifying articles from mainstream news websites such as New York Times and Huffington Post. It was used as a cover story for the comment manipulation and was viewed by all subjects. It was modified to contain no normative information, and the valence tone towards e-cigarette use was balanced (see Online Supplement Figure S1). We conducted a manipulation check among 72 MTurk participants independently from the main study sample (but with the same eligibility criteria and sample characteristics) and asked them to only read the news article. The results confirmed that participants rated the news as balanced (M = 3.03, SD = 0.78) and not significantly different from the balanced/neutral point (i.e., 3) on the 5-point Likert scale (t(71) = 0.38, p = 0.71).

Comments.

Comments stimuli were also collected from the same online news websites. To examine the net effect of norm manipulations, we controlled for the valence factor by design, i.e., we constructed the e-cigarette themes discussed in the comments to have an equal presence of positive and negative valence across all treatment conditions. We first collected and modified a total of 20 comments, with half holding a positive valence tone toward e-cigarette use (e.g., vaping is less harmful) and the other half holding a negative tone (e.g., e-juice contains carcinogens). Each comment tapped into a unique aspect or theme about e-cigarette use. Pre-tests of the 20 comments with an independent sample confirmed that the valence tone was interpreted as intended.2 Next, for each of these original comments, we created two versions (i.e., vaper norm and non-vaper norm) by adding in individual behavior cues accordingly, while keeping the rest of the content the same (see Figure 1 for a set of example comment stimuli; the complete comment stimuli pool can be found in Online Supplement Table S1). We varied the ways in which the individual behavior cues were expressed or conveyed across comments, including indications of who was engaging in the behavior (commenters themselves or others), position in the comments (beginning, middle, or end), as well as different ways of indicating the (non-) vaping status (e.g., “I’m a vaper”, “None of my friends vapes”, or “I tried several flavors”, etc.).

Figure 1.

Figure 1

Open in a new tab

Visual Display of the Three Versions of a Set of Example Comment Stimuli

Note. (A) no-norm comment (original); (B) vaper-norm comment; (C) non-vaper-norm comment.

We then developed a comment allocation algorithm to randomly select and order comments for each participant according to the condition they were assigned to. The algorithm aimed at keeping a balanced valence tone while showing each participant a unique set of comments. The allocation algorithm is used instead of fixed sets of comments to avoid the issue of case-category confound (Jackson, 1992).

Measures

Descriptive norm perceptions about vaping in the U.S.

Informed by prior findings that norms delivered by anonymous online commenters are more likely to affect distal descriptive norm perceptions (Liu & Shi, 2019), participants were asked to indicate how much they agree or disagree with a five-point Likert-type scale ranging from “1 – strongly disagree” to “5 – strongly agree” on the following statements: (a) “In the U.S., many people vape or use e-cigarettes”; (b) “Vaping or using e-cigarettes is not very common in the U.S.”; (c) “A high percentage of the population in the U.S. vape or use e-cigarettes”; (d) “Vaping or using e-cigarettes is not at all popular in the U.S.” After reverse coding the second and the fourth items, higher scores indicated higher perceived vaping descriptive norms in the U.S. population. The average score of the four items served as the focal outcome variable in the analyses (Cronbach’s α = .85).

Vaping intention.

The vaping intention variable was assessed by a standard intention measure (Fishbein & Ajzen, 2009) asking participants to indicate on a four-point Likert-type scale, ranging from “1 – Definitely will not” to “4 – Definitely will”, about how likely they will vape or use an e-cigarette, even one or two puffs, at any time in the next six months.

Data Analysis

Considering that the focal independent variable in this study, behavior prevalence, was constructed by various combinations of the two higher-level treatment factors (the total number of comments and the number of vaper-norm comments), participants’ responses were expected to be influenced by both the fixed effect of behavior prevalence as well as the random effects resulting from exposure to different numbers of vaper-norm comments and total comments (e.g., obtaining a 20% estimation of behavior prevalence from a board with 5 comments compared to one with 20 comments may lead to different ultimate effects on descriptive norm perceptions). Therefore, given the multilevel structure of the dataset determined by our study design, where descriptive norm perception and intention measures (Level 1) were nested within the two treatment factors (Level 2), multilevel modeling analyses (also known as linear mixed effects modeling analyses) were performed to account for influences occurring at different levels and the interdependence of these measures using the lmerTest package in R (Kuznetsova et al., 2017). Specifically, to examine H1 and RQ1, the analytical model specified the fixed effects of the grand-mean-centered behavior prevalence and group size while adjusting for participants’ ever e-cigarette use status, age, gender, race/ethnicity, and education, with descriptive norm perceptions specified as the outcome variable. The total number of comments and the number of vaper-norm comments were also specified as random effects, with slopes allowed to vary across clusters to account for non-independence from these two sources.

To examine RQ2, we added the fixed effect of a grand-mean centered interaction term between behavior prevalence and group size to the equation specified for H1. In the event that a significant interaction effect was observed, the Johnson-Neyman (J-N) technique would be employed, through the use of R package interactions (Long, 2022), to further probe the interaction effect, identify regions of significance and the boundary values where the continuous moderator (i.e., group size) elicited statistically significant slopes (Johnson & Neyman, 1936). Compared to the traditional “pick-a-point” or “simple slopes” method where the significance of the independent variable (IV)’s effect is examined at some arbitrarily chosen, fixed points (Aiken & West, 1991), the J-N technique examines the IV’s effect on the entire range of a moderator and empirically determines how the effect of an IV on a dependent variable (DV) varies from being significant or not based on different values of the moderator (Bauer & Curran, 2005).

To test H2, we conducted multilevel structural equation modeling (MSEM) analysis, which weds the ability to deal with nested data structures with the strengths of SEM. A Bayes estimator (using default noninformative priors and the Gibbs sampler, two independent Markov chain Monte Carlo [MCMC] chains, the min/max number of total iterations = 20,000/50,000, burn-in = the first half of iterations, thinning = 1) was required to be employed in the specified MSEM model to carry out the mediation analysis, where behavior prevalence (Level 2) was modeled as the IV, descriptive norm perceptions (Level 1) as the mediator, and intention for e-cigarette use as the DV (Level 1), following a “2–1-1” design based on the classification of Preacher and colleagues (2010).

The MSEM approach allows the separation of the indirect effect into “within” and “between” components and returns unbiased estimates. According to Preacher et al. (2010), when the IV is a Level 2 predictor, it can predict only between-group level variability in the DV. Therefore, in a multilevel “2–1-1” mediational model, the question of interest is not simply whether the mediator variable mediates the relationship between the IV and DV, but whether, and to what degree, the between-group level variability in the mediator variable (i.e., descriptive norm perceptions in this study) serves as a mediator of the between-group level effect of the IV (i.e., behavior prevalence) on the between-group level component of DV (i.e., vaping intention). In other words, the structure of the “2–1-1” design assumes that the mediation effect is inherently functioning at the between-group level of analysis (Level 2). Therefore, in the current study, we focused on examining the between-group level result of the indirect effect to answer H2.

Similar to SEM models utilizing other estimation methods, model fit was first assessed. In the Bayesian context, posterior predictive checking would be used to assess the predictive accuracy of the model, yielding a posterior predictive p value, defined as the proportion of Chi-square values obtained in the simulated data that exceed that of the actual data. A p-value around .50 indicates an excellent-fitting model (Gelman et al., 2004). Empirical and simulation research considered the use of p values of .10 or .05 to be acceptable (Cain & Zhang, 2019; Muthén & Asparouhov, 2012). To interpret hypothesis testing results, the 95% credibility intervals were used to indicate the 95% probability that the posterior point estimate would lie between the lower and upper values of the interval. If the interval does not include zero, the null hypothesis is rejected, and the presence of an effect is inferred (Muthen, 2010). Mplus (version 8.8) was employed to analyze the MSEM model and provide indirect effect estimates.

Results

H1 predicted that behavior prevalence, operationalized as the percentage of vaper-norm comments out of the total number of comments, is positively associated with descriptive norm perceptions of vaping. The intraclass correlation coefficient (ICC), which refers to the ratio of the amount of between-group variance in relation to the total variance, indicated that about 8.67% of the variance was accounted for by the between-group variation. The ICC score suggested that there was a nonignorable amount of dependency in descriptive norm perception values as a function of Level-2 clustering (i.e., by the total number of comments and the number of vaper-norm comments). Thus, the use of linear mixed effects modeling to test our hypotheses and research question was warranted. The results suggested that after accounting for the Level-2 clustering effects and controlling for the fixed effects of group size, demographic and behavioral variables, behavior prevalence was found to be significantly and positively associated with descriptive norm perceptions (γ = 0.17, t(19.57) = 5.41, p < .001, 95% CI [0.11, 0.22]). In line with established conventions for estimating effect size in multilevel models, we calculated the f2, which quantifies the proportion of variance explained by the effect relative to the proportion of unexplained outcome variance (Aiken & West, 1991; Lorah, 2018). The resulting f2 value of 0.023 indicated a small effect size. Figure 2 illustrates that as the levels of behavior prevalence (on the X-axis) increased, the predictive values of descriptive norm perceptions (on the Y-axis) also increased, despite the relatively small effect size. H1 was supported. For RQ1, however, no main effect of group size was observed (γ = −0.02, t(31.11) = −0.63, p = .54, 95% CI [-0.07, 0.04]). See Model 1 in Table 2 for detailed results.

Figure 2.

Figure 2

Open in a new tab

Main Effects of Behavior Prevalence on Descriptive Norm Perceptions

Note. The above figure demonstrates the fitted line of the main effects of behavior prevalence on descriptive norm perceptions, accompanied by aggregated observed values of descriptive norm perceptions represented as dots. To facilitate stable estimation and clear demonstration, on the X-axis, individual behavior prevalence values were grouped in intervals with 10% increment. The observed values (dots) were averaged descriptive norm perception values within each interval.

Table 2.

Multilevel Modeling Results

Descriptive Norm Perceptions Toward Vaping
Fixed Effect Model 1 (Main effects) Model 2 (Interaction)
Focal variables γ p γ p
Behavior prevalence 0.17 <.001*** 0.18 <.001***
Group size −0.02 .535 −0.03 .263
Behavior prevalence × Group Size 0.07 .009**
Control variables
E-cig ever use status (ref. = No) 0.34 <.001*** 0.35 <.001***
Age −0.17 <.001*** −0.17 <.001***
Gender (ref. = Male) 0.12 <.001*** 0.12 <.001***
Race (ref. = Non-Hispanic White)
 Hispanic 0.04 .111 0.04 .106
 Non-Hispanic African American −0.02 .419 −0.02 .422
 Non-Hispanic Asian/Pacific Islander 0.01 .761 0.01 .744
 Non-Hispanic multiple races −0.03 .894 −0.02 .938
Education −0.13 <.001*** −0.14 <.001***
Random Effect Variance component SD Variance component SD
Group size variance (intercept) 0.011 0.105 0.010 0.109
Group size variance (slope) 0.019 0.136 0.015 0.141
No. of vaper norm variance (intercept) 0.000 0.000 0.002 0.000
No. of vaper norm variance (slope) 0.001 0.029 0.002 0.031
Total Variance Explained (R 2 ) 0.2289 0.2342
Open in a new tab

Note. γ represents standardized multilevel modeling regression coefficient. SD = standard deviation. Behavior prevalence and group size variables were grand mean centered before entering the regression analyses.

*

p < .05,

**

p < .01,

***

p < .001.

To examine RQ2, a grand-mean centered interaction term between behavior prevalence and group size was added to the equation. The results suggested that the interaction between behavior prevalence and group size was significant (γ = 0.07, t(42.65) = 2.70, p = .010, 95% CI [0.02, 0.13]; Model 2, Table 2). To further decompose the interaction, the Johnson-Neyman technique was employed to probe the region of significance and identify threshold values of the continuous moderator variable (i.e., group size) for which the effect of behavior prevalence on descriptive norm perceptions becomes or ceases to be significant. Figure 3 shows the conditional effect of behavior prevalence on descriptive norm perceptions as a function of group sizes. The simple regression line representing the variation in slopes as function of group size is shown along with its 95% confidence bands. The region of significance is shown in dark grey, while the region of non-significance is shown in light grey. The empirically derived results suggested that when the value of group size is outside the interval [-33.08, 6.29], the slope of behavior prevalence effect on descriptive norm perceptions is significant at p < .05 level. Since the observed values of group size ranged from 1 to 20, this analysis identified the value of 6.29 (as indicated by the vertical dotted line in Figure 3) as the threshold value of the moderator, such that the impact of behavior prevalence on norm perceptions relationship became significant when the group size was larger than 6.29 (i.e., when individuals read seven or more comments in total). On the contrary, at lower levels of group size, the relationship remained non-significant.

Figure 3.

Figure 3

Open in a new tab

Johnson-Neyman Interaction Plot on Effects of Behavior Prevalence on Norm Perceptions at Different Group Sizes

H2 proposed to examine whether descriptive norm perceptions would mediate the association between behavior prevalence and vaping intention. We first performed a linear mixed effects modeling to examine if behavior prevalence directly influenced vaping intention. The result suggested that the direct effect of behavior prevalence on vaping intention was not significant (γ = 0.04, t(21.55) = 1.44, p = .16, 95% CI [-0.01, 0.09]). Next, we fit an MSEM model with a Bayes estimator to examine the mediation hypothesis. The model demonstrated an acceptable fit, with a posterior predictive p value of .116. The unstandardized regression parameters representing the median of the posterior distribution, as well as the 95% credibility intervals (CrIs) surrounding the estimate were produced for the MSEM path analysis. The results suggested that the residual direct association between behavior prevalence and vaping intention was not significant (B = 0.12, posterior SD = 0.08, 95% CrI [-0.04, 0.26], one-tailed p = .073), but the indirect pathway via descriptive norm perceptions was significant, B = 0.13, posterior SD = 0.03, 95% CrI [0.07, 0.19], one-tailed p < .001. H2 was supported. The positive indirect effect estimate confirmed a positive mediational pathway between behavior prevalence and vaping intention via descriptive norm perceptions. It is worth noting that since the descriptive norm perception and intention measures were both collected immediately after comments exposure, their relationship should be interpreted as an observational cross-sectional association. In other words, the positive indirect effect does not preclude rival possibilities, such as intention being causally influential on descriptive norm perceptions, or unmeasured third variables accounting for the perception-intention association.

Discussion

By integrating the disjointed literature on public opinion, conformity, and social norms, this study systematically examined how exposure to others’ behavior choices on online comment boards may affect individuals’ descriptive norm perceptions and intentions toward e-cigarette use. The findings confirmed that individuals could make inferences and projections about behavior prevalence in a direction consistent with the individual behavior cue distributions on the comment boards, such that increases in behavior prevalence (i.e., being exposed to more vaper norm comments) led to significant increases in their descriptive norm perceptions about e-cigarette use in the real world (H1). Although the main effect of group size (i.e., the total number of comments) on descriptive norm perception was not significant (RQ1), we observed that having an adequate group size facilitates the emergence of significant effects of behavior prevalence on normative perception changes (RQ2). Specifically, our empirical findings demonstrated that for a significant impact of behavior prevalence to emerge, the comment board needed to contain more than six comments. The results were also consistent with the hypothesis that descriptive norm perceptions mediate the relationship between behavior prevalence and vaping intentions (H2). These findings provide important theoretical and practical implications and set the stage for future research in this realm.

Theoretical Implications

The public opinion literature informed us that human beings have a quasi-statistical sense that collects opinion climate information from their surroundings to help decide their best moves in social situations (Noelle-Neumann, 1974; Scheufle & Moy, 2000). Nevertheless, little is known about how it operates to help people draw conclusions about the majority choices when they encounter mixed individual behavior cues with different behavioral implications (Petersen, 2012). While previous studies typically relied on summary statistics to convey descriptive norms, we utilized individual behavior cues mentioned by online commenters to manipulate observed behavior prevalence. This approach presents normative information in a more subtle manner that better reflects actual scenarios encountered in everyday life. Individuals typically gather information about other people’s health or risk behavior choices from various sources in their physical and media environments during their daily routines.

Findings from the current study provide important evidence that individuals’ quasi-statistical sense correctly gauges the prevalence of the target behavior (i.e., vaping in our case) through exposure to individual behavior cues embedded in online comments across a spectrum of behavioral prevalence and group size combinations. We also observed that the positive effect of behavior prevalence on descriptive norm perceptions started to become significant only when sufficient group size was in place. This finding may indicate that the size of the overall evidence pool has bearings on the perceived validity of the majority’s position. When reference groups are larger, they provide greater credibility and endorsement to the majority position, as its prominence is perceived to be established by the wisdom of a larger crowd. Additionally, in line with Rimal and Storey’s (2020) discussion on “normative volume,” i.e., the absolute number of individuals contributing to the formation of norm perceptions, a larger perceived normative volume exerts increased pressure on individuals to align their own normative endorsement accordingly. Thus, when keeping behavior prevalence constant, having an adequate size of the evidence pool is crucial for people’s “quasi-statistical sense” to accurately gauge behavior prevalence and produce meaningful changes in descriptive norm perceptions.

Additionally, while previous literature studying social influence in offline group settings (e.g., Asch, 1956) assumed the minority stance of the target participant and used a prevailing or unanimous majority to stimulate social pressure, our study provides novel evidence that with no default initial stance assumed, no coerciveness imposed (facilitated by the anonymous online comment setting), and across a continuum of behavior prevalence (i.e., no overwhelming predominance of one normative direction was in force), people can still form norm estimations that correspond to various combinations of behavioral prevalence and group size with accuracy.

It is worth noting that the anonymous online comment board setting of this study may have facilitated the descriptive norm perception modification process. Specifically, as suggested by conformity research, people are much more likely to conform when they perceive that choices are being made by different and unconnected social entities, compared to when they repeatedly see the same behavior choice from the same source, such as a single person or group (Weaver et al., 2007). In other words, people are more vulnerable to social influence when they are aware that the same choice has been made by five different, unrelated individuals, compared to when they witness the same choice being made by a single person on five occasions. Even though the quantity of exposure is equivalent, the degree of influence is markedly different. Given that the anonymous online comment board setting has an implicit assumption that the commenters are distinct, unconnected strangers, seeing the same behavior choice attested to repeatedly from these independent sources is likely to enhance the influence of that choice. In this sense, exposure to behavior cues in the current study does not equate to a “mere exposure” concept which puts emphasis on the quantity of exposures that often gives rise to processing fluency (Bornstein, 1989); rather, it focuses on exposures accumulatively obtained from distinct sources, which operate under a qualitatively different mechanism to exert social influence. Wilder (1977) argued that we could count on sturdy increases in behavior prevalence effects by simply adding majority voices only under conditions when majorities are seen as unrelated individuals having arrived independently at the same decision. This resonates with Asch’s (1951) statement that “consensus is valid only to the extent to which each individual asserts his own relation to facts and retains his individuality.” The unique format of online comments facilitates perceptions of such individuality for each of these social annotations.

Consistent with prior investigations of online social influence (e.g., Liu & Shi, 2019; Walther et al., 2010), we also observed that although the population of online commenters is nowhere near being representative of the overall population, viewers still projected their behavior distribution perceptions derived from the online comment boards to the general U.S. population. In other words, viewers may be relatively insensitive to the credibility or representativeness of the sources from which the behavior cues are collected. But in aggregate, these scattered pieces can together substantially modify people’s perceptions of the world.

Practical Implications

On the practical end, the study provides empirical support for the concern that people’s perceptions and behavioral decisions can be easily influenced by behavior indications online. It further reminds us of the importance for media practitioners and policymakers to closely monitor and prevent possible review fraud aimed at manipulating the public’s decision making. The lack of a direct relationship between behavior prevalence and intention, along with the discovery of a full mediational pathway through descriptive norm perceptions, may also provide insights for health interventionists. They can focus on factors that can effectively accelerate and precipitate the targeted normative inferential process to facilitate desirable behavior changes.

The current study also reveals the significant moderating role of group size in the effects of behavior prevalence on normative perception changes. Based on this finding, it is reasonable to expect that one is more likely to perceive that 40% of people vape when they observe 8 out of 20 people vaping, as opposed to 2 out of 5 people vaping. Although the behavior prevalence rate is the same, the overall group size critically determines the sensitivity of the “quasi-statistical” sense. This finding has important practical implications for creating effective approaches to monitor the development of undesirable descriptive norms online, as well as for devising more effective health programs. In terms of online surveillance, priority should be given to social influencers or networks that contain large clusters of user interactions, given the likelihood of these platforms serving as large “evidence pools” for norm modification. For vaping specifically, regulatory efforts may focus on limiting the detailed visual demonstration and textual depictions of the vaping behavior (i.e., behavior cues) in the promotional content disseminated by the tobacco industry to prevent inflated prevalence perceptions. In health education initiatives that utilize exemplars to illustrate recommended health behaviors, it is crucial to ensure sufficient total exposure to allow the norms to be modified toward the desired direction; and at the same time, diversify the exposure portfolio as much as possible (i.e., exposures are allocated to different sources) to increase effectiveness through enhanced social influence.

Limitations and Future Directions

Finally, we would like to acknowledge some limitations of the current study. First, while normative perception modification typically occurs over an extended period when people encounter an array of norm-relevant instances in the real world (Bicchieri & McNally, 2018), our study demonstrated the potential for the online social sphere to accelerate this process. However, one potential concern with this faster modification process is its potential short-lived nature. Therefore, longitudinal studies in real-world settings may be an important future direction to help investigate the sustainability of such constructed normative perception changes. Collecting intention measures at various time points can also help establish the causal direction between descriptive norm perception and intention changes. Second, while our behavior prevalence manipulation significantly influenced descriptive norm perceptions, its direct impact on intention changes was not observed. The anonymous comment board setting in our study may explain this discrepancy, as it fostered the perception that the comments represented a distant population, which aligned closely with the reference group for descriptive norm perceptions. However, when it comes to changes in personal intentions, one might not expect a strong direct impact of the comment manipulation, potentially because these comments were perceived as originating from a disparate and distal reference group. Future studies could explore how the alignment of reference groups implicated in the comment manipulation and outcome measures may impact the relationships observed. Third, considering that one’s tobacco use behaviors can impact their processing of normative information on vaping, future research would benefit from measuring and examining the potential moderating role of one’s prior and current vaping and smoking behaviors on the relationship between behavior prevalence and vaping intentions. Furthermore, the amount of exposure needed to bring in norm perception changes may vary by context and behavior. This is a single study; it will not generalize to every pair of exposure-norm relationships. The findings of the current study may be a special case, particularly considering that we were only able to examine a maximum of 20 as the overall comment group size and that we examined the norm modification process of only one health risk behavior. Therefore, the study finding is worth replication and re-examination under different situations, with different types of online environments (e.g., Facebook, Twitter, etc.), different levels of exposure, and when behaviors with different attributes are under investigation. Finally, in the realm of social norms literature, it has been consistently demonstrated that both descriptive and injunctive norms hold a significant sway over people’s behavioral choices, with well-established main and interactive effects on intention and behavior changes (Cialdini et al., 1990; Lapinski & Rimal, 2005; Rimal & Lapinski, 2015; Rimal & Real, 2003). However, the current study has exclusively focused on descriptive norms. Future research may explore how descriptive norm perceptions formed through exposure to accumulative individual behavior cues may impact injunctive norm perceptions and delve into their formation and modification mechanisms.

Conclusion

The current study is the first to systematically examine the effects of behavior prevalence, formed through exposure to accumulative individual behavior cues, on descriptive norm perceptions and behavioral intentions through a constructed social sphere. We provided empirical evidence that can help clarify how the “quasi-statistical sense” operates when individuals face an array of mixed individual behavior cues. We also observed evidence in support of the idea that comments-induced descriptive norm perceptions mediate between behavior prevalence and intentions. These findings offer important implications for theoretical development and indicate that online user-generated content may have practical consequences that can affect the collective modification of public perceptions of risky health behaviors.

Supplementary Material

Supp 1

Acknowledgments

Research reported in this publication was supported by the National Cancer Institute (NCI) of the National Institutes of Health (NIH) and FDA Center for Tobacco Products (CTP) under Award Number P50CA179546 to the University of Pennsylvania. Liu wishes to acknowledge support from the NIH and FDA (K01DA049292; R21DA056570). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or the Food and Drug Administration (FDA).

The authors extend their sincere appreciation to Celeste Condit for providing insightful feedback on a previous version of this manuscript. We also wish to acknowledge Yidi Wang for her diligent assistance with proofreading and ensuring the accuracy of our citations.

Footnotes

1

Number and percentage of non-vaper-norm comments are linear transformations of that of the vaper-norm ones in each condition (e.g., a 20% vaper-norm condition equates to an 80% non-vaper-norm condition).

2

N=273 participants from a communication department research subject pool at a large public university participated in the comments testing study. The results confirmed that, on a 7-point Likert scale (1“very negative”, 7“very positive”), the 10 comments intended to be perceived as positive (M = 4.83, SD = 0.45) were rated significantly more positive than the 10 comments intended to be perceived as negative (M = 2.33, SD = 0.45), t(272) = 59.20, p < .001.

References

  1. Aiken LS, & West S (1991). Multiple regression: Testing and interpreting interactions. SAGE Publications. [Google Scholar]
  2. Ajzen I (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. 10.1016/0749-5978(91)90020-T [DOI] [Google Scholar]
  3. Aleyan S, East K, McNeill A, Cummings KM, Fong GT, Yong H-H, Thrasher JF, Borland R, & Hitchman SC (2019). Differences in norms towards the use of nicotine vaping products among adult smokers, former smokers and nicotine vaping product users: Cross-sectional findings from the 2016 ITC Four Country Smoking and Vaping Survey. Addiction, 114(S1), 97–106. 10.1111/add.14648 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Anderson AA, Brossard D, Scheufele DA, Xenos MA, & Ladwig P (2014). The “nasty effect:” Online incivility and risk perceptions of emerging technologies. Journal of Computer-Mediated Communication, 19(3), 373–387. 10.1111/jcc4.12009 [DOI] [Google Scholar]
  5. Asch SE (1951). Effects of group pressure upon the modification and distortion of judgments. In Guetzkow H (Ed.), Groups, leadership and men: Research in human relations (pp. 177–190). Carnegie Press. [Google Scholar]
  6. Asch SE (1955). Opinions and social pressure. Scientific American, 193(5), 31–35. 10.1038/scientificamerican1155-31 [DOI] [Google Scholar]
  7. Asch SE (1956). Studies of independence and conformity: I. A minority of one against a unanimous majority. Psychological Monographs: General and Applied, 70(9), 1–70. 10.1037/h0093718 [DOI] [Google Scholar]
  8. Bauer DJ, & Curran PJ (2005). Probing interactions in fixed and multilevel regression: Inferential and graphical techniques. Multivariate Behavioral Research, 40(3), 373–400. 10.1207/s15327906mbr4003_5 [DOI] [PubMed] [Google Scholar]
  9. Bergquist M, & Nilsson A (2019). The DOs and DON’Ts in social norms: A descriptive don’t‐norm increases conformity. Journal of Theoretical Social Psychology, 3(3), 158–166. 10.1002/jts5.43 [DOI] [Google Scholar]
  10. Bicchieri C, & McNally P (2018). Shrieking sirens: Schemata, scripts, and social norms. How change occurs. Social Philosophy and Policy, 35(1), 23–53. 10.1017/S0265052518000079 [DOI] [Google Scholar]
  11. Bornstein RF (1989). Exposure and affect: Overview and meta-analysis of research, 1968–1987. Psychological Bulletin, 106(2), 265–289. 10.1037/0033-2909.106.2.265 [DOI] [Google Scholar]
  12. Cain MK, & Zhang Z (2019). Fit for a Bayesian: An evaluation of PPP and DIC for structural equation modeling. Structural Equation Modeling: A Multidisciplinary Journal, 26(1), 39–50. 10.1080/10705511.2018.1490648 [DOI] [Google Scholar]
  13. Chung JE (2018). Peer influence of online comments in newspapers: Applying social norms and the social identification model of deindividuation effects (SIDE). Social Science Computer Review, 0894439318779000. 10.1177/0894439318779000 [DOI] [Google Scholar]
  14. Chung M, Jang Y, Knight Lapinski M, Kerr JM, Zhao J, Shupp R, & Peng T-Q (2022). I do, therefore I think it is normal: The causal effects of behavior on descriptive norm formation and evolution. Social Influence, 17(1), 17–35. 10.1080/15534510.2022.2052955 [DOI] [Google Scholar]
  15. Cialdini RB (2005). Basic social influence is underestimated. Psychological Inquiry, 16, 158–161. 10.1207/s15327965pli1604_03 [DOI] [Google Scholar]
  16. Cialdini RB, & Goldstein NJ (2004). Social influence: Compliance and conformity. Annual Review of Psychology, 55(1), 591–621. 10.1146/annurev.psych.55.090902.142015 [DOI] [PubMed] [Google Scholar]
  17. Cialdini RB, Reno RR, & Kallgren CA (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58(6), 1015–1026. 10.1037/0022-3514.58.6.1015 [DOI] [Google Scholar]
  18. Cohen J (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Routledge. [Google Scholar]
  19. Cornelius ME, Loretan CG, Jamal A, Davis Lynn BC, Mayer M, Alcantara IC, & Neff L (2023). Tobacco product use among adults – United States, 2021. MMWR. Morbidity and Mortality Weekly Report, 72. 10.15585/mmwr.mm7218a1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Echeagaray O, Savko C, Gallo A, & Sussman M (2022). Cardiovascular consequences of vaping. Current Opinion in Cardiology, 37(3), 227–235. 10.1097/HCO.0000000000000952 [DOI] [PubMed] [Google Scholar]
  21. Faul F, Erdfelder E, Buchner A, & Lang A-G (2009). Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behavior Research Methods, 41(4), 1149–1160. 10.3758/BRM.41.4.1149 [DOI] [PubMed] [Google Scholar]
  22. Festinger L (1954). A theory of social comparison processes. Human Relations, 7, 117–140. 10.1177/001872675400700202 [DOI] [Google Scholar]
  23. Fishbein M, & Ajzen I (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Addison Wesley. [Google Scholar]
  24. Fishbein M, & Ajzen I (2009). Predicting and changing behavior: The reasoned action approach. Psychology Press. 10.4324/9780203838020 [DOI] [Google Scholar]
  25. Fiske ST, & Taylor SE (1991). Social cognition (2nd ed.). McGraw-Hill Book Company. [Google Scholar]
  26. Gelman A, Carlin JB, Stern HS, & Rubin DB (2004). Bayesian data analysis (2nd ed.). Chapman & Hall. [Google Scholar]
  27. Guichard E, Autin F, Croizet J-C, & Jouffre S (2021). Increasing vegetables purchase with a descriptive-norm message: A cluster randomized controlled intervention in two university canteens. Appetite, 167, 105624. 10.1016/j.appet.2021.105624 [DOI] [PubMed] [Google Scholar]
  28. Higgins ET (1996). Knowledge activation: Accessibility, applicability, and salience. In Higgins ET & Kruglanski A (Eds.), Social psychology: Handbook of basic principles. Guilford. [Google Scholar]
  29. Huo J, Desai R, Hong Y-R, Turner K, Mainous AG, & Bian J (2019). Use of social media in health communication: Findings from the health information national trends survey 2013, 2014, and 2017. Cancer Control: Journal of the Moffitt Cancer Center, 26(1), 1073274819841442. 10.1177/1073274819841442 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Jackson S (1992). Message effects research: Principles of design and analysis. Guilford Press. [Google Scholar]
  31. Johnson PO, & Neyman J (1936). Tests of certain linear hypotheses and their application to some educational problems. Statistical Research Memoirs, 1, 57–93. [Google Scholar]
  32. Kasza KA, Edwards KC, Kimmel HL, Anesetti-Rothermel A, Cummings KM, Niaura RS, Sharma A, Ellis EM, Jackson R, Blanco C, Silveira ML, Hatsukami DK, & Hyland A (2021). Association of e-cigarette use with discontinuation of cigarette smoking among adult smokers who were initially never planning to quit. JAMA Network Open, 4(12), e2140880. 10.1001/jamanetworkopen.2021.40880 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Kuznetsova A, Brockhoff PB, & Christensen RHB (2017). LmerTest package: Tests in linear mixed effects models. Journal of Statistical Software, 82, 1–26. 10.18637/jss.v082.i13 [DOI] [Google Scholar]
  34. Lapinski MK, & Rimal RN (2005). An explication of social norms. Communication Theory, 15(2), 127–147. 10.1111/comt.12080 [DOI] [Google Scholar]
  35. Latané B, & Wolf S (1981). The social impact of majorities and minorities. Psychological Review, 88(5), 438. 10.1037/0033-295X.88.5.438 [DOI] [Google Scholar]
  36. Lee E-J, & Tandoc EC (2017). When news meets the audience: How audience feedback online affects news production and consumption. Human Communication Research, 43(4), 436–449. 10.1111/hcre.12123 [DOI] [Google Scholar]
  37. Liu J, & Shi R (2019). How do online comments affect perceived descriptive norms of e-cigarette use? The role of quasi-statistical sense, valence perceptions, and exposure dosage. Journal of Computer-Mediated Communication, 24(1), 1–20. 10.1093/jcmc/zmy021 [DOI] [Google Scholar]
  38. Long J (2022). Interactions: Comprehensive, user-friendly toolkit for probing interactions. (R package version 1.1.6) [Computer software]. https://cran.r-project.org/package=interactions
  39. López-Ojeda W, & Hurley RA (2024). Vaping and the brain: Effects of electronic cigarettes and e-liquid substances. The Journal of Neuropsychiatry and Clinical Neurosciences, 36(1), A5–5. 10.1176/appi.neuropsych.20230184 [DOI] [PubMed] [Google Scholar]
  40. Lorah J (2018). Effect size measures for multilevel models: Definition, interpretation, and TIMSS example. Large-Scale Assessments in Education, 6(1), 8. 10.1186/s40536-018-0061-2 [DOI] [Google Scholar]
  41. Mollen S, Rimal RN, Ruiter RAC, & Kok G (2013). Healthy and unhealthy social norms and food selection. Findings from a field-experiment. Appetite, 65, 83–89. 10.1016/j.appet.2013.01.020 [DOI] [PubMed] [Google Scholar]
  42. Mullen B (1983). Operationalizing the effect of the group on the individual: A self-attention perspective. Journal of Experimental Social Psychology, 19(4), 295–322. 10.1016/0022-1031(83)90025-2 [DOI] [Google Scholar]
  43. Muthen B (2010). Bayesian analysis in Mplus: A brief introduction (Version 3). Unpublished manuscript. Professor Emeritus at the Graduate School of Education & Information Studies, UCLA. Retrieved from https://www.statmodel.com/download/IntroBayesVersion%203.pdf [Google Scholar]
  44. Muthén B, & Asparouhov T (2012). Bayesian structural equation modeling: A more flexible representation of substantive theory. Psychological Methods, 17(3), 313–335. 10.1037/a0026802 [DOI] [PubMed] [Google Scholar]
  45. Neubaum G, & Krämer NC (2017). Opinion climates in social media: Blending mass and interpersonal communication. Human Communication Research, 43(4), 464–476. 10.1111/hcre.12118 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Newcomb T (1967). Persistence and change: Bennington College and its students after twenty-five years. Wiley. [Google Scholar]
  47. Nisbett RE, & Wilson TD (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84(3), 231–259. 10.1037/0033-295X.84.3.231 [DOI] [Google Scholar]
  48. Noelle‐Neumann E (1974). The spiral of silence a theory of public opinion. Journal of Communication, 24(2), 43–51. 10.1111/j.1460-2466.1974.tb00367.x [DOI] [Google Scholar]
  49. Oh H, Besecker M, Huh J, Zhou S, Luczak SE, & Pedersen ER (2022). Substance use descriptive norms and behaviors among US college students: Findings from the healthy minds study. Epidemiologia, 3(1), 42–48. 10.3390/epidemiologia3010005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Petersen T (2012). The enduring appeal of an unwieldy theory. International Journal of Public Opinion Research, 24(3), 263–268. 10.1093/ijpor/eds031 [DOI] [Google Scholar]
  51. Preacher KJ, Zyphur MJ, & Zhang Z (2010). A general multilevel SEM framework for assessing multilevel mediation. Psychological Methods, 15(3), 209–233. 10.1037/a0020141 [DOI] [PubMed] [Google Scholar]
  52. Rhodes N, Roskos-Ewoldsen DR, Edison A, & Bradford MB (2008). Attitude and norm accessibility affect processing of anti-smoking messages. Health Psychology, 27(3, Suppl), S224–S232. 10.1037/0278-6133.27.3(Suppl.).S224 [DOI] [PubMed] [Google Scholar]
  53. Rhodes N, Shulman HC, & McClaran N (2020). Changing norms: A meta-analytic integration of research on social norms appeals. Human Communication Research, 46(2–3), 161–191. 10.1093/hcr/hqz023 [DOI] [Google Scholar]
  54. Richter I, Thøgersen J, & Klöckner CA (2018). A social norms intervention going wrong: Boomerang effects from descriptive norms information. Sustainability, 10(8), Article 8. 10.3390/su10082848 [DOI] [Google Scholar]
  55. Rimal RN, & Lapinski MK (2015). A re-explication of social norms, ten years later. Communication Theory, 25(4), 393–409. 10.1111/comt.12080 [DOI] [Google Scholar]
  56. Rimal RN, & Real K (2003). Understanding the influence of perceived norms on behaviors. Communication Theory, 13(2), 184–203. 10.1111/j.1468-2885.2003.tb00288.x [DOI] [Google Scholar]
  57. Rimal RN, & Storey JD (2020). Construction of meaning during a pandemic: The forgotten role of social norms. Health Communication, 35(14), 1732–1734. 10.1080/10410236.2020.1838091 [DOI] [PubMed] [Google Scholar]
  58. Rosenthal R (1991). Meta-Analytic procedures for social research (Vol. 6). SAGE Publications, Inc., 10.4135/9781412984997 [DOI] [Google Scholar]
  59. Scheufle DA, & Moy P (2000). Twenty-five years of the spiral of silence: A conceptual review and empirical outlook. International Journal of Public Opinion Research, 12(1), 3–28. 10.1093/ijpor/12.1.3 [DOI] [Google Scholar]
  60. Shi R, Messaris P, & Cappella JN (2014). Effects of online comments on smokers’ perception of anti-smoking public service announcements. Journal of Computer-Mediated Communication, 19(4), 975–990. 10.1111/jcc4.12057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Stasser G, & Davis JH (1981). Group decision making and social influence: A social interaction sequence model. Psychological Review, 88(6), 523. 10.1037/0033-295X.88.6.523 [DOI] [Google Scholar]
  62. Tankard ME, & Paluck EL (2016). Norm perception as a vehicle for social change. Social Issues and Policy Review, 10(1), 181–211. 10.1111/sipr.12022 [DOI] [Google Scholar]
  63. Tversky A, & Kahneman D (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207–232. 10.1016/0010-0285(73)90033-9 [DOI] [Google Scholar]
  64. Wally CM, & Cameron LD (2017). A randomized-controlled trial of social norm interventions to increase physical activity. Annals of Behavioral Medicine: A Publication of the Society of Behavioral Medicine, 51(5), 642–651. 10.1007/s12160-017-9887-z [DOI] [PubMed] [Google Scholar]
  65. Walther JB, DeAndrea D, Kim J, & Anthony JC (2010). The influence of online comments on perceptions of antimarijuana public service announcements on YouTube. Human Communication Research, 36(4), 469–492. 10.1111/j.1468-2958.2010.01384.x [DOI] [Google Scholar]
  66. Walther JB, & Jang J (2012). Communication processes in participatory websites. Journal of Computer‐Mediated Communication, 18(1), 2–15. 10.1111/j.1083-6101.2012.01592.x [DOI] [Google Scholar]
  67. Weaver K, Garcia SM, Schwarz N, & Miller DT (2007). Inferring the popularity of an opinion from its familiarity: A repetitive voice can sound like a chorus. Journal of Personality and Social Psychology, 92(5), 821–833. 10.1037/0022-3514.92.5.821 [DOI] [PubMed] [Google Scholar]
  68. Wilder DA (1977). Perception of groups, size of opposition, and social influence. Journal of Experimental Social Psychology, 13(3), 253–268. 10.1016/0022-1031(77)90047-6 [DOI] [Google Scholar]
  69. Xie W, Kathuria H, Galiatsatos P, Blaha MJ, Hamburg NM, Robertson RM, Bhatnagar A, Benjamin EJ, & Stokes AC (2020). Association of electronic cigarette use with incident respiratory conditions among US adults from 2013 to 2018. JAMA Network Open, 3(11), e2020816. 10.1001/jamanetworkopen.2020.20816 [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Yong H-H, Chow R, East K, Thrasher JF, Hitchman SC, Borland R, Cummings KM, & Fong GT (2023). Do social norms for cigarette smoking and nicotine vaping product use predict trying nicotine vaping products and attempts to quit cigarette smoking amongst adult smokers? Findings from the 2016–2020 international tobacco control four country smoking and vaping surveys. Nicotine & Tobacco Research, 25(3), 505–513. 10.1093/ntr/ntac212 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supp 1
NIHMS1990960-supplement-Supp_1.docx (574.2KB, docx)

RESOURCES