Investigating the Mechanisms by which Prevalent Online Community Behaviors Influence Responses to Misinformation: Do Perceived Norms Really Act as a Mediator?

This study addresses two currently open questions about how behaviors of online community members influence others’ responses to misinformation. First, in contrast to prior work, it directly measures norm perception to address whether (1) norm perception actually acts as a mediator, (2) others’ behaviors directly influence individuals’ responses to misinformation, (3) both direct and mediated effects occur. Second, it investigates norm perceptions about a behavior that is not readily observable in online communities, but is prone to misinformation, specifically, vaccination. To do so, it experimentally manipulates the prevalence of communicating about vaccination (an unobservable behavior) within an online community. The results demonstrate no evidence of a direct effect—the causal relationship between prevalence of communicating a behavior and intentions to respond to misinformation only occurs via norm perception as a mediator. The paper highlights implications of these findings for designing community-centered interventions to influence perceived norms, thereby mitigating misinformation spread and impacts.


Introduction
Vaccine misinformation is a growing societal problem, with signifcant global consequences.The ramifcations of vaccine misinformation transcend individual and public health harms, and manifest in broader efects.Examples include impacting social interactions, eroding trust in science, and diminishing confdence in authoritative entities [40,48,66,89], among other efects.For instance, the spread of misinformation about COVID-19 vaccines not only compounded the complexities of the pandemic and made vaccine coverage more challenging across the globe [73,108], it even inspired incidents of violent attacks [84,96].
Numerous approaches have been developed and investigated to address the spread of misinformation, both health related [e.g., 1,3,6,22,56,90,104,112] and otherwise, from fact-checkers that aim at disputing false information [3,6], to reducing the visibility of misleading content [27,104], to interventions designed to improve individuals' media related literacy skills [19,52,90,111].However, as argued by Aghajari et al. [10], despite the signifcant role of online communities on individuals' response to misinformation and the way misinformation has its broader impacts at the community level [11,21,36,42,46,47,100,106], less is known about the mechanisms by which online communities infuence individuals' responses to misinformation and how to design interventions around those mechanisms.
The behavior of community members online and their response to the world's events play a signifcant role in the spread and impacts of false and misleading content surrounding those events [21,36,42,46,47,100,106].The role of community's behavior extends beyond infuencing individuals' behaviors.Rather, it also infuences perceptions about what people in the community do, what they approve of, and how they expect other members to behave.Such perceptions are referred to as perceived norms [86,87].Perceived norms in turn infuence people's behaviors [67,87,105].
Within the context of misinformation, prior studies argue that perceived norms infuence how individuals identify pieces of false and misleading content, as well as the way they respond to, including potentially counteracting such content (e.g., fag it, provide corrective information, or share it) [36,57].However, these prior studies did not directly measure perceived norms.Thus, it remains unclear whether others' behavior directly infuences individuals' responses to false and misleading content, whether this infuence occurs via perceived norms as a mediating factor, or whether both direct and mediated efects occur.Furthermore, diferent types of perceived norms (e.g., perceptions about what people in a community do, i.e., descriptive norms vs. perceptions of what people in a community approve of, i.e., injunctive norms) [30,32,34,87]) may diferently infuence responses to false and misleading content.Prior work provides little expectation about which of these types of perceived norms are likely to have weaker or stronger efects.Indeed, expanding knowledge around these mechanisms and their impacts on individuals' responses to misinformation can provide valuable insights for designing interventions centered around the infuence of perceived norms to help addressing misinformation within online communities.
To investigate potential causal relationships between perceived norms and individuals' responses to false and misleading content, it is important both to manipulate and to measure perceived norms, as well as individuals' responses to such content, within a controlled, experimental setting.Recent research provides valuable insights into how manipulating prevalence of a behavior can be utilized to infuence perceived norms in an experimental context [11,67].For instance, previous studies suggest that both in online and ofine settings, when a behavior is commonly exhibited within a community, it is perceived to be normative as well [11,67,86].In particular, in the context of misinformation, Aghajari et al. [11] demonstrate that the prevalence of sharing false and misleading content greatly impacts on perceived norms of an online community.
That said, individuals' behavior may not always be readily observable from their online activities [e.g., 20,99], which can pose challenges when attempting to manipulate its prevalence.For example, in the context of vaccination, it may not be directly observable from people's online activities whether or not they are vaccinated.Consequently, it is less clear how to manipulate perceptions of norms in these less observable contexts, making it challenging to examine the relationships between perceived norms and responses to misinformation in these contexts.
Thus, the study presented here goes beyond prior work to make two novel contributions: ascertaining whether perceived norms are a mediating factor between the prevalence of a behavior and the intent to respond to misinformation about that behavior, and examining norm perception around a behavior not directly observable within an online community.Specifcally, motivated by the aforementioned prior work on the relationship between the prevalence of a behavior and norm perceptions [11,67], this study examines whether and how the prevalence of explicitly communicating about a behavior within an online community, in contexts wherein individuals' behaviors are not readily observable online, might similarly impact perceived norms.To do so, with a focus on COVID-19 vaccination as a testbed, it manipulates the ratio of community members who explicitly communicate having received their COVID-19 vaccines.To examine this hypothesis and to ensure this manipulation of perceived norms works successfully, in contrast to prior work, this study directly measures participants' perceptions of diferent types of norms about vaccination in a controlled setting.In addition, it examines whether and how the prevalence of explicitly communicating about a behavior (e.g., being vaccinated) might directly infuence participants' intentions to respond to misinformation (i.e., via fagging it or providing corrective information) about that behavior.Furthermore, measuring diferent types of perceived norms associated with an event (e.g., vaccination), this work investigates whether the efects of explicitly communicating a behavior on intention to respond to misinformation actually occurs via mechanisms of norm perceptions.
To examine the aforementioned hypotheses, this work conducts a between-subject, controlled experiment.To manipulate the prevalence of explicitly communicating a behavior that is not otherwise observable online, with a focus on COVID-19 vaccination as a testbed, it controls the ratio of community members who adopt a vaccinated profle picture frame on their profle and communicate having received their vaccines (See Figure 3).
Key fndings suggest that in the context of vaccination, wherein individuals' behaviors are not readily observable based on their online interactions, the prevalence of communicating about being vaccinated within a community signifcantly infuences perceptions of norms.These perceptions include both perceptions around both what people within a community do (i.e., descriptive norms) and what people in a community approve of (i.e., injunctive norms).The prevalence of communicating about this behavior, however, did not directly infuence participants' intention to respond to misinformation about vaccines.Rather, its efect is mediated by the mechanisms of norm perceptions.Put precisely, the prevalence of communicating about being vaccinated had its impact by infuencing perceptions of descriptive norms and injunctive norms, and perceived descriptive norms then had a direct impact on participants' intention to respond to misinformation about vaccines, either via fagging it or providing corrective information.
The paper concludes by discussing the implications of the fndings for leveraging the role of perceived norms to enhance online communities' resilience to misinformation.Specifcally, it ofers implications both in terms of (a) designing community-centered interventions around the role of perceived norms to address misinformation by leveraging the efects of perceived norms on individuals' responses to misinformation, and in terms of (b) designing community-centered tools that enable individuals to infuence how their community's norms are perceived, particularly in contexts wherein their behavior is less observable within an online community.

Related Work
This section reviews prior research on the concept of social norms.It distinguishes two distinct types of social norms, i.e., collective norms and perceived norms [85][86][87].It highlights the signifcance of perceived norms compared to collective norms in infuencing individuals' behavior and opinions in a variety of subjects [e.g., 28,34,65,67,85,87,105], including their impacts on individuals' behaviors within the context of misinformation [36,43,57].
This section proceeds by discussing one of the mechanisms by which individuals perceive norms in an online setting, i.e., observing the prevalence of a behavior within a community [11,67].Prior work demonstrates the role of this mechanism in contexts where individuals' behaviors are observable online.This section considers how prevalence of communicating about a behavior may similarly infuence norm perceptions in contexts wherein individuals' behaviors are less observable online, such as in the contexts of vaccination.It follows by discussing the motivations behind how the prevalence of communicating about a behavior in such less observable contexts, either directly or via mechanisms of perceived norms, can potentially impact individuals' intention to respond to surrounding false and misleading content.[59].However, prior work has shown that it is perceived norms, perceptions of others' behaviors [30,33,59,71,72,87], that infuence an individual's opinions and behaviors.For example, Perkins and Berkowitz [80] demonstrate that students often held exaggerated perceptions about the prevalence of drinking.Such perceptions, and not the actual norms of drinking, infuence some individuals to rationalize their own excessive drinking habits [59].
Recent works suggests perceived norms particularly infuence individuals' behaviors regarding misinformation, including the way they identify and respond to misinformation [14,36,43,57,78,97,113]. Section 2.4 reviews these prior studies and examines whether perceived norms have a direct efect on individuals' behaviors in misinformation context.

Prevalence of a Communication a Behavior in an Online Community and Impacts on Perceived Norms
Recent experimental studies, in two diferent contexts of privacy and misinformation, demonstrate that the prevalence of a behavior within an online community infuence perceived norms [11,67].
In fact, the prevalence of a behavior within an online community infuences not only perceptions of what others do (i.e., perceived descriptive norms), but also perceptions of what they approve of (i.e., injunctive norms), and how they expect others to behave (i.e., subjective norms) [e.g., 11,67].For example, Masur et al. [67] suggest observing common behaviors of others regarding privacy practices infuences people's perceptions of all the three aforementioned types of perceived norms.In another case, Aghajari et al. [11] show that observing prevalence of posting false and misleading content can signifcantly infuence the three types of perceived norms.These prior studies have focused on contexts wherein individuals' behavior is observable based on their online activities.However, individuals' behaviors do not always occur online and, thus, would not be readily observable within an online community.Examples of these behaviors include whether or not people have been vaccinated, and if they follow guidance by health authorities, such as masking during the COVID-19 pandemic.The work presented in this paper hypothesizes that in these less readily observable contexts, the prevalence of explicitly communicating about a behavior within an online community might similarly impact diferent types of perceived norms.Specifcally, with a focus on vaccines, this work posits that: H1: Seeing a greater number of community members who explicitly communicate an otherwise unobservable behavior within an online community (e.g., having been vaccinated) will result in higher participants' perceptions of norms about the behavior within that community (e.g., norms of vaccination).This includes (H1.a) descriptive, (H1.b) injunctive, and (H1.c) subjective norms.

Prevalence of Communicating a Behavior and Intention to Respond to Related Misinformation
As discussed, the behavior of individuals regarding an event, particularly when exhibited commonly, greatly infuences the behaviors of others towards the same event [36,57,67,87,95].In the context of misinformation, for example, Colliander [36] demonstrates that exposure to critical comments made by others regarding fake news articles reduces the chance that people will share the fake news.In another example, Koo et al. [57] demonstrate that observing other people within a community actively correcting misinformation can encourage individuals to join this efort as well.
To examine how the behavior of community members might infuence others' responses to false and misleading content within a community, previous studies primarily focused on behavior exhibited in an online community and, thus, directly observable based on online activities.Examples include the way community members share false and misleading content and/or respond to such content shared in their community (e.g., fagging, commenting) [e.g., 36,57].However, as discussed in Section 2.2, individuals' behaviors do not always occur online and, thus, would not be readily observable within an online community.We hypothesize that in these less observable contexts the prevalence of communicating about a behavior may play a similar role.
H2: The prevalence of explicitly communicating about a behavior within an online community, specifcally in contexts where the behavior is not readily observable from individuals' online activities (e.g., vaccination), infuences participants' intentions to respond to false and misleading content related to that behavior within the community.

Figure 1:
We hypothesize that the prevalence of communicating about a behavior (e.g., being vaccinated) impacts diferent types of perceived norms (i.e., descriptive, injunctive, and subjective).In addition, we examine if communicating about a behavior has a direct (i.e., H2) or an indirect impact, mediated by mechanisms of perceived norms (i.e., H3), on intention to respond to misinformation about that behavior.

Perceived Norms and Intentions to Respond to Misinformation
Prior work provides insights around the potential role of perceived norms in how people identify and respond to false and misleading content, including countering such content (e.g., fag it, or correct it) [14,36,43,57].For example, Colliander [36] demonstrates that if people see others state that a piece of content is fake, they are less likely to share it.Colliander [36] argues that this infuence of others' responses occurs via perceived norms.In another experimental study, Andı and Akesson [14] design an intervention based on descriptive norms, wherein participants are shown a message that suggests most people think twice before sharing the news.They show that displaying this normative information can reduces the chance that people will share fake articles, and argue that this infuence occurs due to norm conformity to descriptive norms.Neither of the aforementioned studies, however, directly measured participants' perceptions of the community's norms.Thus, it is less clear whether the observed efects of their manipulation on intentions to share fake news actually occurred via the mechanism of perceived norm.Koo et al. [57], instead of manipulating perceived norms and investigating their efects, directly ask participants about perceptions of norms of correcting misinformation.Specifcally, they ask, "If a typical American has posted information that was made-up, how likely is it that they will correct it?"Their analysis reveals that perceived norms around self-correcting infuence individuals to self-correct.However, while the study measures participants' perceptions of norms, its focus is primarily on norms of self-correction and perceived norms within the United States as a whole, rather than norms specifc to a particular online community.
Motivated by these prior work, we posit that there exists a casual relationship between perceived norms about an event (e.g., vaccination) in an online community and participants' intention to respond to misinformation related to that event.In contrast to prior work, the work presented in this paper examines this hypothesis while directly measuring participants' perceptions of norms.

Procedure
The participants were presented with a consent form that provided detailed information regarding the study's requirements, duration, and compensation.Following their voluntary consent to participate, they were directed to a social media site named "Clikbrite" for one minute.Prior to proceeding to Clikbrite, participants were instructed to scrolling through the newsfeed to view others' posts, reply to those posts, like others' posts, or fag a post for the platform to review for potentially harmful content.
The feed consisted of 30 posts and their respective comments.These posts are created using a social media platform for experimental studies, called Truman [38].The Truman platform is designed and developed to allow researchers to create a controlled, yet realistic social media simulation, and to facilitate abilities to conduct experimental research about interactions on social media (See DiFranzo et al. [38] for more detail on the Truman platform).Thus, to gain experimental control required for this study while maintaining experimental realism, this platform is employed to conduct the presented study.
The content of the posts and their corresponding responses were generated collaboratively by researchers from Lehigh University and University of North Carolina at Chapel Hill.The posts were about everyday life (e.g., posting food, photographs of nature, and sports activities), while controlling not to include any political opinions in them, in addition to the study's manipulated variable (Described in Section 3.2).The authors pilot-tested the posts iteratively and refned the posts to ensure they are realistic and resembled a realistic experimental setting.
After the experiment, participants were given a post-survey.First, the participants were asked about their perceptions of the community's norms around vaccination.Next, they were asked about their intentions to address vaccine misinformation.This specifc order is implemented to ensure participants' intentions to respond to vaccine misinformation are informed by their perceptions about the community's norms.Note that a reverse ordering of these questions, i.e., asking participants about their intentions to respond to misinformation and then their perceptions of the community's norms may infuence the results, but this order was enforced by design and due to specifc direction of the relationship between perceived norms and its infuence on intention to respond to vaccine that this work examines.Finally, participants were debriefed.Specifcally, they were informed about the study's goal and how the actors and posts involved were created by researchers and were not genuine content from a real online community.Participants had the option to withdraw their data at this point.The study procedures were approved by the IRB number 21-2969 at University of North Carolina at Chapel Hill.

Experimental Design
This study employs a three-condition, between-subject design.It manipulates the prevalence of community members who explicitly communicate having received COVID-19 vaccines in each experimental condition.To do so, it utilizes an existing design element, a picture frame adopted on some of the community members' profle, by which they communicate about the reception of COVID-19 vaccines (Shown in Figure 3, and further explained below).
Participants are randomly assigned to one of the three experimental conditions, each exploring a feed with either 5%, 15%, or 30% of the community members communicating about being vaccinated, using the design element explained below.These proportions are informed by prior work [11,67].In a related study on misinformation, Aghajari et al. [11] manipulated 5%, 30%, or 60% of the posts in their experiment to examine their efects on norm perceptions.However, since they found there was not much diference between conditions with 30% and 60% manipulated posts compared to conditions with 5% and 30% , and considering how using 60% manipulated posts in our experiment would greatly impact experimental realism, we instead used 5%, 15%, or 30% ratios.With 30 posts in the feed, we rounded the number of posts to the nearest whole numbers corresponding to these percentages, resulting in 2, 5, and 9 posts, respectively.Even though these ratios are not exactly 5, 15, and 30, we refer to them as the 5, 15, and 30 conditions for simplicity.

Experimental Manipulation:
To manipulate the prevalence of community members who explicitly communicate about having received COVID-19 vaccines, this work utilizes a vaccinated picture frame on the profle of some of the community members.Profle picture frame was frst introduced in 2015 by Facebook [51].Many people have used this feature to express their stance regarding a variety of events-from supports for people of colors, to support for same sex marriage, to support people of Ukraine during Russia's attack on Ukraine -(See Figure 2 for some examples of profle picture frames on Facebook).By adopting this feature, not only these people increase the visibility of the associated event [44], but also make their otherwise unobservable behavior and/or opinions readily observable to others.Figure 3b shows the picture frame created for and employed in this experiment.This picture frame is informed by and is similar to the existing profle picture frames on social media platforms to enhance the experimental realism (See Figure 3).To ensure that participants notice the stimuli, akin to the mechanism employed on Facebook, a notifcation of updated profle pictures using vaccinated frames are featured on the newsfeed shown during the experiment.

Recruitment and Participants
Data was collected during Jan 2023.712 participants were recruited from Cloud Research to participate in the study.Participants were compensated $3.00 for their time.We conducted a power analysis for a medium efect size ( = 0.15) with 0.95 power and determined we should need 690 participants.This efect size is adopted from the efect size stated in Cohen [35], where efect values of 0.35, 0.15, and 0.02 represent large, medium, and small efect sizes, respectively.Thus, our sample met the required size based on our power analysis.
After removing the responses of participants who did not fnish all the steps of the experiment, the data of 636 participants was used for our analysis.
The participants reported a mean age of 22.18 ( = 1.92).The restricted age range within this sample is a result of the data being collected as part of a broader study that specifcally targets health communication among younger adults in an online context.The authors explored carefully factors from that larger study to ensure they do not have any relationship with the content of this study's manipulation.However, the authors acknowledge the limitation of the sample in terms of its constrained age range (i.e., the younger adults), and recognize its potential impact on the generalizability of the fndings.Section 7.3 further addresses this limitation and highlights the need for future research to examine whether the observed results hold across a broader age range.

Perceptions of Norms
To measure participants' perceptions of norms within the community that they observed, the scale developed by Park and Smith [77] is adapted, and updated to ft the focus of this study.This scale includes three items for descriptive norms (i.e., "The majority of people on clikbrite have taken a COVID-19 vaccine.","The majority of people on clikbrite believe that the COVID-19 vaccine is safe.","The majority of people on clikbrite are hesitant to take the COVID-19 vaccine."),three items for injunctive norms (i.e., "The majority of people on clikbrite approve of getting the COVID-19 vaccine.", "The majority of people on clikbrite endorse getting COVID-19 vaccines.", "The majority of people on clikbrite support taking COVID-19 vaccines."), and two items for subjective norms (i.e., "The majority of people on clikbrite feel obligated to take the COVID-19 vaccine.","It is expected of users on clikbrite that they get their COVID-19 vaccine." ).A 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree) is used to measure these metrics [58].In our data, this scale demonstrates a high degree of reliability for each type of perceived norms, that is descriptive norms ( = 0.73, = 4.60, = 1.01), injunctive norms ( = 0.93, = 4.77, = 1.17), and subjective norms ( = 0.77, = 4.29, = 1.12).

Intention to Respond to Vaccine Misinformation
To measure the intention of responding to vaccine misinformation, a 2-item scale is used (i.e., "If someone posted content on clikbrite that included information about vaccines I think is untrue, I would fag that post for reporting.", "If someone posted content on clikbrite that included information about vaccines I think is untrue, I would provide corrective information, such as in a comment.").These items are informed by prior work, which emphasizes on providing corrective information and fagging it as practices employed to address misinformation [36,55].Both items use a 7-point scale ranging from 1 (strongly disagree) to 7 (strongly agree).Confrmatory Factor Analysis (CFA) provides evidence that these two items contribute to a single underlying latent construct, i.e., the participants intention to respond to misinformation, with approximately similar magnitudes (i.e., 0.68, and 0.70, < 0.001).The standardized factor loadings for the items are greater than 0.60, which underscore the strong association of these items with the latent construct [45].We then use the single factor resulting from these loadings as the outcome variable in the subsequent analyses.

Control Variables
Given that prior research demonstrates demographic variables can have an impact on norm perceptions [17,29,60], in this experiment demographic variables are measured as control variables.In addition, participants' COVID-19 vaccination status was asked (i.e., "What is your COVID-19 vaccine status?"), to investigate potential impacts of vaccination status on perceived norms of vaccinations.The response options were as follows: "no vaccines", " Partially vaccinated (e.g., one dose of mRNA)", "Fully vaccinated (e.g., two doses of mRNA, or one-dose J&J)", "Fully vaccinated and received a booster dose").The distribution of COVID-19 vaccination status among participants, as self-reported by themselves, was as follows: 39.90% of the participants fully vaccinated and received a booster dose (n=265), 31.17%fully vaccinated (n = 207), 23.76% not received any vaccines (n = 158), and 5.12% were partially vaccinated (n=34).

Data Analysis
The efects related to our hypothesis were investigated using structural equation modeling (SEM) [49,103], for two main reasons.First, the dependent variables of this study (i.e., diferent types of perceived norms) are highly correlated.SEM accounts for correlations between these variables.Second, it is likely that the efects of the prevalence of a communicating about a behavior (e.g., having been vaccinated) on intentions to address misinformation occur through mechanisms of perceived norms.SEM helps to estimate the hypothesized structure of the relationships among prevalence, perceived norms, and intentions to respond to vaccine misinformation.The SEM analysis is conducted in R using Lavaan package [91].Figure 4 depicts the results of this analysis.The multi-dimensional model ftted the data well ( 2 (43) = 182.508,< 0.001, = 0.965, = 0.947; = 0.027, = 0.043).Tukey's WSD (Wholly Signifcant Diference) post-hoc tests using simultaneous 95% confdence intervals is used for multiple comparisons between experimental groups [82].Specifcally, we used the function tukeySEM in R to conduct Tukey's WSD post-hoc analysis [69].Demographics are included in the models as between-subjects covariates and are reported when they demonstrate statistical signifcance.
While SEM is chosen for the aforementioned reasons, the Lavaan package in R that is used to conduct the SEM analysis does not support analysis of interactions for the latent variables1 [7,91].Therefore, the potential interaction efects are examined using multivariate analysis of variance (MANOVA).Tukey's honestly significant diference (HSD) tests using simultaneous 95% confdence intervals are used for investigating the details of the interaction effects.Due to the hypothesized structure of the relationships among prevalence, perceived norms, and the intention to respond to misinformation, however, the potential interaction efects of the manipulated variables are only investigated for the perceived norms.

Results
This section outlines the results of the SEM analysis, including the impacts of prevalence of communicating about a behavior on diferent types of perceived norms (Section 6.1).It also investigate whether this prevalence directly infuences participants' intention to respond to misinformation regarding the behavior (Section 6.2), and/or if this infuence occurs via mechanisms of perceived norms (Section 6.3).In addition, the section addresses the efects of participants' vaccination status on these dependent variables (Section 6.4).

Efects of the Prevalence of Communicating
about a Behavior on Perceived Norms

Perceived Descriptive Norms
In line with our prediction in H1.a, the prevalence of explicitly communicating about an otherwise unobservable behavior within an online community (i.e., being vaccinated) had a positive, and signifcant efect on perceived descriptive norms of vaccination ( = 0.30, < 0.001).Based on Tukey's WSD post-hoc analysis, there was a signifcant diference across all levels of the prevalence of communicating being vaccinated within the community (5%, 15%, and 30%).However, the diference in perceived descriptive norms was larger between conditions with 5% and 15% community members adopting vaccinated profle picture frames compared to the diference between conditions with 15% and 30% community members adopting vaccinated profle picture frames (0.42 vs. 0.24).These results suggest that a relatively small number of community members who take the initiative and explicitly communicate their behavior regarding vaccinations (e.g., by adopting a vaccinated profle pictures in this experiment) can infuence others' perceptions about what a community as a whole does about vaccination (i.e., descriptive norms).

Perceived Injunctive Norms
The prevalence of communicating about being vaccinated by members of the community greatly and signifcantly impacted perceived injunctive norms of vaccination ( = 0.26, < 0.001), supporting H1.b. Tukey's WSD Post-hoc tests demonstrate that the diference between perceived injunctive norms in conditions with 5% and 15% was greater than the diference between perceived injunctive norms between the conditions with 15% and 30% ratio of vaccinated profle picture frames in the community (0.75 vs. 0.26).This result suggests that, similar to descriptive norms, even minority of people in a community who explicitly communicate their reception of vaccines within their community can still greatly infuence perceptions about what the community as a whole approves of regarding vaccines (i.e., injunctive norms).
6.1.3Perceived Subjective Norms Perceived subjective norms were not statistically signifcantly afected by observing the prevalence of community members communicating about being vaccinated.More specifcally, the participants did not perceive that it is expected of users within the community to be vaccinated when more individuals in the community explicitly communicated about having received their vaccines (i.e., there is not sufcient evidence to support H1.c).This result could be observed due to how perceived subjective norms are tied to perceptions around how a community expects others to behave.The design element employed in this study, though, pertains more directly to what community members do, which is more relevant to descriptive norms.Infuencing perceptions around subjective norms may require alternate cues directly pertaining to a community's expectations of others (discussed in Section 7.3).

Efects of the Prevalence of Communicating about a Behavior on Intentions to Respond to Misinformation about the Behavior
The prevalence of communicating about being vaccinated within the community did not have a direct, signifcant impact on participants' intentions to address vaccine misinformation (i.e., there is insufcient evidence to support H2).More specifcally, observing more community members who adopted a vaccinated picture frame on their profle to communicate being vaccinated did not show signifcant impact on how participants would fag false and misleading content, or provide corrective information in response to such content.
Despite lack of observing its direct efects, the prevalence of communicating about a behavior had an indirect efect on intentions to respond to misinformation, which is mediated by perceived norms, specifcally perceived descriptive norms (Discussed in Section 6.3).

Efects of Perceived Norms on the Intentions to Respond to Misinformation
The analysis shows participants perceptions of descriptive norms strongly and signifcantly predicted participants' intentions to address misinformation, either via fagging or correcting the false and misleading content ( = 0.57, < 0.05) (Supporting H3.a).More precisely, when participants perceived that the majority of the community members had taken their vaccine, and believed in the vaccine's safety, the participants were more likely to respond to vaccine misinformation, either via fagging or correcting it.Participants' intention to respond to misinformation, however, was not statistically signifcantly afected by their perceptions of others' approval of vaccination (i.e., injunctive norms) or perceptions of how others within the community were expected to receive their vaccines (i.e., subjective norms) (i.e., there is insufcient evidence to support H3.b or H3.c).These results could have occurred due to the specifc mechanisms by which each type of perceived norm impacts people's behaviors, as well as the specifc setting employed in this work, in terms of lack of participants' membership in the community.For instance, the infuence of descriptive norms occurs due to social conformity [86,87], which can occur even in settings where people do not have memberships within a community [e.g., 16,31,34], as in our study.However, the infuence of injunctive norms occurs via a desire for social approval and a tendency to avoid social disapproval [33,61].In the context of our study, without having membership in the community, our participants may not have felt a strong desire for social approval, and thus, were not infuenced by injunctive norms.

Efects of Vaccination Status on Perceived Norms of Vaccination and Intent to Respond to Vaccine Misinformation
Participants' vaccination status greatly and signifcantly infuenced their perceptions of descriptive ( = 0.17, < 0.001), and injunctive norms ( = 0.11, < 0.01).Put precisely, participants who reported to be fully or partially vaccinated also reported higher perceived descriptive and injunctive norms compared to the participants who were not vaccinated.These fndings, in line with prior work [102], suggests that individuals' own behavior can impact how they perceive norms.However, the impacts of participants' vaccination status on perceived subjective norms was not neither large nor statistically signifcant.In addition, the results did not indicate any statistically signifcant interaction efects between participants' vaccination status and the prevalence of communicating about being vaccinated on norm perceptions.Moreover, participants' vaccination status signifcantly predicted their intention to respond to misinformation about vaccines ( = 0.29, < 0.001).Employing Tukey's WSD post-hoc analysis, the results show participants who reported to be fully vaccinated, on average, also reported stronger intention to respond to misinformation about vaccines compared to those who had not receive any COVID-19 vaccines (3.84 vs. 4.73, p < 0.001).However, as discussed in Section 5, due to the limitation in Lavaan package [91] for analysis of interaction for latent variables, the potential interaction efects between vaccination status and norm perceptions on intention to respond to misinformation remains unexplored.

Discussion
This discussion frst compares results with those from prior work to describe the novel contributions ofered in this paper.It then considers the implications that the fndings from this study ofer about the role of perceived norms for enhancing online communities resilience to misinformation.Finally, it describes limitations of the presented study and suggests directions for future work.

Articulation of Contributions.
This work contributes to our understanding of how others' behaviors infuence individuals' response to misinformation.Specifcally, in contrast to prior work [14,36,57], by manipulating and directly measuring perceived norms, this work demonstrates the infuence of others' behavior on intention to respond to misinformation is an indirect efect, which is mediated by the mechanism of perceived (descriptive) norms.This fnding suggests perceived norms as an important community-oriented mechanism to design interventions around for addressing misinformation and its impacts (Discussed in Section 7.2).
This study also contributes to knowledge around the mechanisms of norm perceptions in contexts where individuals' behaviors are less readily observable online.It demonstrates that similar to how the prevalence of a behavior infuences perceptions about what is normative within an online community [11,67], in contexts where individuals' behaviors are not readily observable, the prevalence of communicating about a behavior infuences perceived norms.The results indicate that diferent levels of prevalence of explicitly communicating about a behavior (e.g., having been vaccinated) lead to diferent level of norm perceptions.However, the relationship between the prevalence of explicitly communicating being vaccinated within a community and perceptions of norms around vaccines is unlikely to follow a linear pattern.Specifcally, the disparity in perceived norms was greater when comparing a small prevalence (e.g., 5%) to a moderate prevalence (e.g., 15%) of community members who communicated being vaccinated, as opposed to the diference between a moderate prevalence (e.g., 15%) and a large prevalence (e.g., 30%) within the community.Therefore, even a minority of community members expressing that they have received their vaccine (e.g., 15%) can signifcantly infuence perceptions of normative behavior regarding vaccination.These results align with prior research that demonstrates people's perceptions of norms are often exaggerated, both in online and ofine settings [11,67,87].
While behaviors that directly pertain to misinformation (e.g., such as providing corrective information or fagging it) directly impact response to misinformation [e.g., 14,36], this study's results suggest that the efect of communicating about a behavior relevant to an event is unlikely to directly impact intentions to address misinformation.Rather, the efect of communicating about a behavior that pertains to an event is an indirect efect, which is mediated by the mechanisms of perceived norms, specifcally descriptive perceived norms.

Implications for Enhancing Community Resilience to Misinformation
The fndings from this study ofer valuable implications, both in terms of (a) designing community-centered interventions to address misinformation by leveraging the efects of perceived norms on individuals' responses to misinformation, and in terms of (b) designing community-centered tools that enable individuals to infuence how their community's norms are perceived, particularly in contexts where individuals' behaviors are less readily observable online.This study highlights the direct and signifcant efects of perceived norms on individuals' intentions to address false and misleading content.Therefore, in health related contexts and otherwise, researchers and platform designers can design around the role of perceived norms to infuence individuals to address misinformation, thus contributing to mitigating its spread and impacts.At least two strategies could be used to design such interventions.
First, design elements can directly target perceived norms about a community's response to misinformation itself.For instance, researchers can focus on the efect of descriptive norms (i.e., what others do) and highlight how a community as a whole actively responds to false and misleading content (e.g., fagging, reporting and correcting misinformation) to promote such practices.If these interventions work efectively, they might result in observing less misinformation within a community.While these interventions are benefcial in terms of addressing misinformation itself, in this case, there will be less opportunity to observe others' behaviors regarding misinformation, thus hard to infuence descriptive norms.Therefore, any intervention on descriptive norms cannot rely exclusively upon the presence of misinformation.Put diferently, it is important to also examine other ways of impacting descriptive norms without needing community members to observe misinformation and others' responses to it.Thus, as a second approach to complement these interventions, researchers can design interventions that target perceived norms about behavior(s) to which misinformation pertains.By employing these interventions, even when misinformation is less prevalent and people's behaviors regarding misinformation is not clear, community members can still infuence how norms of their community are perceived and how others might respond to misinformation within the community, as discussed in this paper.To design such interventions that target perceived norms about behavior(s), researchers can take inspiration from prior work, such as studies that designed and successfully employed interventions to mitigate alcohol consumption among college students [e.g., 68,79,81].
This study also shows how community members' initiative, to communicate about behaviors that are otherwise not readily observable, can signifcantly impact norm perception.Platform designers and academic researchers can thus consider potential design solutions-either by leveraging already existing afordances, such as the profle picture frames examined in this study, or by designing new low-efort, visual elements-to empower community members to communicate their otherwise unobservable behavior and infuence the perceived norms of their community.For example, during a health crisis, wherein false and misleading content may be more common, many people tend to look into how others are reacting to the situation [37,88].In such situations, behaviors of community members-e.g., whether they follow health authorities recommendations about masking, social distancing, or other practices-may be less readily observable online.Thus, nudging community members to share their behavior can help protect how their community's norms are perceived and, thereby, impact others' responses to misinformation surrounding the event.
Although this work focuses on communicating about an unobservable behavior (i.e., being vaccinated), communicating about an unobservable belief may similarly impact norm perceptions.For example, expressing belief in health authorities and their guidelines, or articulating trust in various scientifc fndings (e.g, climate change), which are both targeted by online misinformation [101,107,110], may similarly infuence perceptions about an online community's norms.The efects of these perceptions, in turn, can encourage more individuals to engage in addressing misinformation about these topics, as shown in this study.Therefore, future work can make valuable contributions by investigating the nuanced dynamics of how communicating about beliefs within an online community might infuence perceived norms and, in turn, individuals' responses to false and misleading content around those beliefs.
Indeed, these same mechanisms could be leveraged to engender inaccurate norm perceptions.For instance, both our results and prior work [11,67] show that actions by a small number of community members with a minority viewpoint (e.g., about antivaccination, excessive alcohol consumption, or explicit racism) can infuence overall norm perceptions.Put diferently, the mechanisms by which prosocial norms become amplifed are the same mechanisms by which antisocial norms can become amplifed.This point re-emphasizes that interventions must be designed, studied, and governed in ways that account for broader community-level effects [cf.10].Thus, just as researchers should examine communitycentered interventions to empower community members to infuence how their community's norms are perceived, they should also inform platform policies and moderation practices to mitigate potential deleterious efects of these interventions.For example, community moderators can monitor the implementation of these interventions, ensuring their utilization is exclusively directed towards encouraging prosocial behaviors.In addition, when addressing the misuse of these interventions for antisocial behaviors, moderators can underscore how such behaviors go against the community's norms, thereby infuencing perceptions about the community and its norms.Of course, similar to content moderation, these moderation practices can be efective only in communities wherein moderators hold good values and good faith [62,109].
Indeed, recent work has examined designing around the "wisdom of crowds" to account for the role of online communities when addressing misinformation as well [12,22,23,26,39].For example, Twitter's/X's Birdwatch [2], now refered to as "community notes", allows online users to identify tweets as misleading or not, write their fact-check review of tweets, and assess the quality of other users' fact-checks.While this intervention is shown to be efective in identifying misleading content on a large scale, it may also contribute to the broader impacts of misinformation at the community level [13,83].For example, Allen et al. [13] demonstrate that people's assessment of tweets is more infuenced by their partisanship than the content of the tweet, leading to increased polarization of online communities.To account for these broader impacts of misinformation at the community-level, it is thus important to supplement these approaches with community-centered interventions that extend beyond addressing individual pieces of false and misleading content [9,10].The work presented in this paper provides an illustrative example for such community-centered interventions.Specifcally, it demonstrates how interventions can be designed around a community's normative behaviors concerning an event or behavior targeted by misinformation, e.g., vaccination.

Limitations and Future Work Directions
This work has three primary limitations, which both suggest important directions for future work.The frst limitation pertains to the methodology employed to conduct the experiment.Specifcally, participants' perceptions of a community's norms are reported based on observing the community for a relatively short time, and with limited interaction within the community.While the experimental social media site ofered interactivity, enabling participants to engage with the content posted on the feed (e.g., through actions such as liking, fagging, and adding comments), participants were not able to interact with other users beyond interacting with their posts on the feed.For example, they could not check out other users' activity page, or share content beyond responding to posts shared by others.However, people's perceptions of community norms may be infuenced by their engagement with others within a community and perhaps be shaped over time [87].Future work is encouraged to further investigate the dynamics involved in norm perceptions in an even more ecologically valid experimental setting that incorporates extended participants' engagement and interaction.
Second, the sample in this study is limited to a specifc age range, i.e., younger adults.This limitation may impact the generalizability of the fndings to other age ranges.Thus, future work is required to examine whether the observed results hold across a broader age range and enhance understanding of these mechanisms in a more diverse population.That said, this study still ofers valuable insights into the role of perceived norms on individuals' responses to misinformation among younger adults, which are indeed the population who use social media the most [63,74].
Third, our study does not employ a manipulation check to explore if participants noticed the stimuli.That said, the results of our pilot test suggests the stimuli was clearly noticeable.Employing a manipulation check could further improve this study's validity.
This paper suggests a number of important research directions, which, while lie beyond the scope of the current study, are valuable for future work to explore.This study examines norm perceptions within a context where participants do not have a membership in the community.However, diferent factors within communities wherein people hold memberships in could potentially infuence the mechanisms of norm perceptions.Examples include the efects of groups individuals belong to and care about (i.e., reference groups) [102], salience of membership within a community [53,76], individuals' needs to belong [24], among others.For example, social media infuencers play can greatly afect individuals' behavior [25,64] and are therefore likely to carry an important role in how norms of an online community are perceived.This infuence can play a role on people's respond to surrounding misinformation as well.Thus, future work, perhaps using an observational study, is encouraged to examine the potential efects of such factors involved in norm perceptions and the way they might impact the relationship between norm perceptions and intention to respond to misinformation.
In addition, recent studies suggest opinions about Covid-19 vaccines vary along political positions [8,18,54].Thus, people with diferent political ideologies may diferently perceive norms of an online community.These perceptions can in turn infuence people's intentions to respond to vaccine misinformation.The conducted study did not measure participants' political ideologies and, thus, did not explore these potential relationships between political ideologies and intentions to respond to vaccine misinformation.Future work is encouraged to investigate potential role of political ideologies on the mechanisms of norm perceptions, and people's response to misinformation surrounding vaccines.
Furthermore, people's defnition of misinformation is subjective [15,75,93,94].Thus, rather than asking participants about their intent to respond to specifc posts, our study asked participants about their intention to respond to information that they think is untrue.Future work can make valuable contributions by accounting for factually correct and incorrect information about vaccines, and examine the infuence of perceived norms on people's responses to such content.
Moreover, the results of this study did not support the efects of the prevalence of communicating about a behavior on subjective norms.These results may be attributed to the way subjective norms are tied to perceptions around how a community expect others to behave, while the design element used in this study pertains more directly to what community members do themselves (i.e., more readily relevant to descriptive norms).Future work can examine other designs by which community members may communicate their expectations of others regarding vaccinations (e.g., "get vaccinated and save lives"), or other similarly unobservable behaviors, and investigate their efects on subjective norms.Understanding how to infuence subjective norms are particularly important in contexts of vaccination, as perceived subjective norms not only can potentially infuence responses to vaccine misinformation, but also directly impact individuals' intentions to get vaccinated as well [50].
Lastly, while this work focuses on vaccination, there are other contexts that are prone to misinformation and communicating about a behavior and/or beliefs can potentially infuence perceived norms.Examples include communicating about supporting marginalized communities, trusting in environmental conservation eforts, among others.Future work can make valuable contributions by exploring mechanisms of norm perceptions in these other contexts, and examining how the relationships between perceived norms and response to misinformation within those contexts.

Conclusion
There is a strong impetus to "do something" about misinformation, health related and otherwise, on social media.By contributing to the prerequisite evidence base that can inform online platform design and community governance, this paper helps address the question of what to do.Specifcally, given the great role of individuals' responses to misinformation on the spread and impacts of misinformation on online communities, this work expands knowledge around the factors that infuence these responses (e.g., to fag it or to provide corrective information).In particular, it demonstrates how norm perception about a behavior, particularly perceived descriptive norms, greatly impact the intent to respond to misinformation about that behavior.Drawing on these fndings, this work ofers design suggestions for leveraging the role of perceived norms to encourage individuals to address misinformation within their community and mitigating its spread and impacts.
This paper also contributes to our understanding about the mechanisms of norm perception in contexts where individuals' behaviors are less readily observable within an online community.It demonstrates that in these less observable contexts, some of which are prone to misinformation and its impacts (e.g., vaccination), community members can still greatly infuence perceptions about their community's norms by taking the initiative and explicitly communicating their behavior.Drawing upon the insights derived from these fndings, this work provides design suggestions for researchers and platform designers to explore community-centered interventions to help community members express their otherwise unobservable behaviors within their community and contribute to how their community's norms are perceived.These interventions can either leverage already existing afordances, such as the profle picture frames examined in this study, or design new ones.The paper also discusses how the deployment and governance of these interventions needs to be studied to ensure that they have the intended efects in terms of mitigating misinformation and its impact on online communities.

Figure 2 :
Figure 2: Examples of picture frames online users have employed to show support for a) the Black Lives Matter Movement, b) the LGBTQ+ community, and c) people of Ukraine during Russia's invasion.
With the development of COVID-19 vaccines, in collaboration with the US Department of Health and Human Services (HHS) and the Centers for Disease Control and Prevention (CDC), Facebook developed vaccine profle picture frame to encourage people get COVID-19 vaccines [4].Many people used this feature to show they have received their COVID-19 vaccines, and/or to show how they support vaccinations to encourage others to do so as well [4, 5, 41].

Figure 3 :
Figure 3: Examples of (a) an actual vaccinated profle picture frame employed on the Facebook platform, and (b) the vaccinated profle picture frame employed in this study.

Figure 4 :
Figure4: The results of SEM analysis indicate a substantial and statistically signifcant infuence of the prevalence of communicating about a behavior (i.e., being vaccinated) on perceived descriptive and injunctive norms of vaccinations.Although communicating about this behavior did not directly afect participants' intentions to respond to vaccine misinformation, it had an indirect efect, which is mediated with descriptive norms.In particular, it infuences perceived descriptive norms, and descriptive norms in turn, predicted participants' intentions to respond to false and misleading content regarding vaccinations.(N/S denotes non-signifcant relationships and number of stars shows level of signifcant as follows: * p<.05, *** p<.001.)