Risk Perception and Disaster Information Verification Among Chinese University Students Exposed to Simulated Social Media Posts

Summary

This randomized factorial experiment evaluated how source cues, information veracity, and emotional tone in simulated disaster-related social media posts affect Chinese university students' risk perception, credibility judgments, verification intentions, sharing intentions, and task-based verification choices.

Abstract

Disaster-related information increasingly circulates through social media during emergencies, but limited experimental evidence has examined how university students evaluate and verify such messages. This randomized 2 × 2 × 2 factorial experiment tested the effects of source cue, information veracity, and emotional tone in simulated disaster-related social media posts among Chinese university students. A total of 426 valid responses were analyzed. Misleading information and high-threat wording increased perceived risk, whereas official-source cues increased perceived credibility and sharing intention. Verification intention was more sensitive to misleading content and urgent wording, but task-based verification behavior showed a clearer association with source cues than with veracity or tone. The binary verification accuracy model provided weaker support than the continuous behavioral score. These findings suggest that risk perception, credibility judgments, intentions, and verification choices may diverge in the context of disaster misinformation. Campus risk communication should therefore emphasize concrete verification routines rather than relying only on credibility cues or general reminders to check information before sharing.

Introduction

Social media has become an important channel for disaster-related communication because it allows rapid dissemination of alerts, eyewitness updates, resource information, and public responses during emergencies. At the same time, the speed and openness of social media create conditions in which inaccurate or unverified disaster information can circulate before formal correction occurs. Recent work on social media data for disaster risk management has noted that social platforms can support preparedness and response, but their usefulness is limited by concerns about data accuracy, misinformation, privacy, and verification¹.

Disaster misinformation is particularly consequential because it can affect how people assess threats and choose protective actions under uncertainty. False or exaggerated claims during emergencies may increase confusion, intensify anxiety, misdirect attention, and reduce trust in official communication. A recent review of disaster-related misinformation reported that inaccurate social media content may contribute to panic, resource misallocation, and weakened emergency response when users rely on unverified information during fast-moving events². These risks make disaster misinformation different from ordinary online rumors. In emergency contexts, even short delays in verification may affect public understanding and response.

Existing misinformation research has shown that social media misinformation is not limited to one topic or one platform. It often emerges where information demand is high, evidence is incomplete, and users are motivated to share quickly. A review of social media misinformation identified disaster, health, and political communication as key domains in which false information can create social harm³. Disaster settings are especially vulnerable because users may seek immediate guidance while official information is still being updated. Under these conditions, people may rely on visible cues such as the apparent source, emotional wording, repost volume, and perceived plausibility.

Research on disaster misinformation has also emphasized the role of correction and verification. Studies of disaster rumors and their corrections on social media show that false claims may spread through networked sharing before corrections reach the same audience⁴. This creates a practical challenge for emergency communication: users must not only recognize doubtful content, but also know how to check it before forwarding. For university students, this challenge is relevant because they are frequent users of social media and often receive emergency-related information through peer groups, campus networks, and short-form platforms.

Chinese university students are a meaningful population for examining this issue. Recent evidence from Chinese college students indicates that they are aware of misinformation on social media, but they may still feel uncertain when judging its credibility and may experience emotional discomfort when exposed to false or doubtful information⁵. This means that students are not simply passive recipients of online content. They evaluate, interpret, and sometimes redistribute information, yet their verification choices may depend on how the message is framed and where it appears to come from.

Risk perception is central to this process. In emergency settings, perceived risk can influence whether individuals seek information, verify claims, avoid hazards, or warn others. A recent meta-analysis found a positive association between risk perception and emergency information-seeking behavior, although the relationship was not large and varied across emergency types⁶. This suggests that risk perception may increase attention to disaster information, but it may not automatically produce accurate verification behavior. Therefore, studies need to distinguish perceived risk from actual information-checking actions.

The theoretical basis for measuring risk perception comes from the broader risk perception tradition, which views risk judgment as a subjective evaluation of hazard severity, probability, controllability, uncertainty, and potential consequences⁷. In a disaster-related social media context, these judgments may be shaped not only by the factual content of the message but also by emotional tone and source cues. A high-threat post may increase perceived severity even when the content is uncertain. An official-looking source may increase credibility even when users have not independently verified the claim.

Source credibility is another important factor in online information evaluation. Prior research on fake news and information verification has shown that trust, media credibility, and social ties can shape users' willingness to verify or share information⁸. In disaster communication, official emergency management sources may carry stronger authority than peer-shared posts. However, the effect of source cues is not always straightforward. A credible source may increase trust and sharing, but it may also reduce suspicion if users assume that official-looking information is already verified. This makes it necessary to test whether official-source cues increase verification behavior, reduce it, or operate differently across perceived credibility and verification intention.

Fact-checking research further suggests that verification behavior should not be treated as a simple extension of credibility judgment. Social media fact-checking intention has been linked to news literacy and trust in information sources⁹. However, intention to verify does not always mean that users know which verification actions are reliable. In disaster contexts, clicking comments, asking an unspecified friend, or forwarding first and checking later may appear to users as forms of response, but these actions do not provide the same reliability as checking official emergency channels or recognized news outlets.

This distinction is consistent with broader misinformation research showing that belief, perceived accuracy, and sharing decisions may diverge. People may share content for reasons other than accuracy, including social relevance, urgency, familiarity, or perceived usefulness¹⁰. Therefore, a study of disaster information verification should measure both self-reported verification intention and task-based verification behavior. Without this distinction, it is difficult to know whether students merely express willingness to check information or actually select reliable verification actions.

Although existing reviews have mapped misinformation mechanisms and detection challenges, there remains a need for controlled experimental evidence on disaster-related social media posts among Chinese university students. General reviews of fake news, disinformation, and misinformation have highlighted the difficulty of identifying false information, as misleading content often resembles truthful content and requires contextual verification¹¹. However, less is known about how source cues, information veracity, and emotional tone jointly shape students' risk perception, credibility judgments, verification intentions, sharing intentions, and task-based verification behavior in a disaster context.

To address this gap, the present study used a randomized 2 × 2 × 2 factorial experiment. Chinese university students were exposed to one simulated disaster-related social media post that varied by source cue, information veracity, and emotional tone. The study examined whether these manipulated factors affected risk perception, perceived credibility, verification intention, sharing intention, and verification behavior. By separating psychological intention from task-based verification behavior, the study aimed to provide a more precise account of how university students evaluate and respond to disaster-related information on social media.

The experiment was based on the assumption that disaster-message features can influence several response layers that do not necessarily move together. Source cues were expected to shape perceived credibility most strongly because verified or official-looking accounts provide an institutional authority signal. Information veracity was expected to influence risk perception and credibility because misleading claims often contain uncertainty, exaggeration, or unsupported threat statements. Emotional tone was expected to increase risk perception and verification intention by increasing perceived urgency. However, task-based verification behavior was expected to depend not only on perceived risk or credibility but also on whether participants selected concrete, reliable checking actions. Accordingly, the study tested the following hypotheses: H1, official-source cues will increase perceived credibility compared with peer-source cues; H2, misleading information will increase risk perception and reduce perceived credibility compared with accurate information; H3, high-threat wording will increase risk perception and verification intention compared with neutral wording; H4, source cue, information veracity, and emotional tone may interact, such that misleading high-threat posts attributed to peer sources will produce the highest risk perception; and H5, verification intention and task-based verification behavior will show only partial convergence because intention ratings and action choices reflect different response processes.

Protocol

Study design

Before any recruitment or data collection, the study protocol was reviewed and approved by the Research Ethics Committee of the University of Malaya (approval ID: UUM20259910). All procedures involving human participants were conducted in accordance with institutional requirements and the Declaration of Helsinki. Participants provided electronic informed consent before entering the questionnaire, and no personally identifiable information was exported for analysis.

This study used a randomized between-subjects online experiment to examine risk perception and disaster information verification among Chinese university students after exposure to simulated social media posts. Following stimulus-based misinformation research¹², the design manipulated predefined information cues rather than relying only on retrospective self-reported exposure. The experiment used a 2 × 2 × 2 factorial structure with three factors: source cue, information veracity, and emotional tone. The official-source + misleading-information cells represented fictional official-looking or impersonated official-style posts created solely for the experiment; they did not represent a real emergency management agency issuing false information.

The study procedure is shown in Figure 1. Participants completed eligibility screening, electronic informed consent, baseline measures, random assignment, exposure to one simulated post, post-exposure assessment, verification task, debriefing, and data-quality screening. Each participant viewed only one post to avoid learning effects across conditions. The factorial structure and randomization cells are presented in Table 1.

Participants first completed eligibility screening and electronic informed consent, followed by baseline measurement, individual-level random assignment to one of eight experimental cells, exposure to one simulated disaster-related social media post, post-exposure assessment, a task-based verification assessment, immediate debriefing, data-quality screening, and final statistical analysis. The workflow shows only procedural and screening steps and does not present empirical results.

Participants and recruitment

Participants were full-time undergraduate or postgraduate students enrolled in mainland Chinese universities. Eligibility criteria were age 18 years or older, current university enrollment, ability to read simplified Chinese, and regular use of at least one social media platform. Participants were excluded if they were younger than 18 years, not currently enrolled as university students, had joined the pilot test, submitted an incomplete questionnaire, failed the attention-check item, completed the questionnaire in an unrealistically short time, or showed patterned responses across scale items.

Recruitment was conducted through university notice groups, course announcement channels, and student association networks. The recruitment notice described the study as research on university students' evaluation of online disaster information. It did not disclose before participation that some posts contained misleading elements, because advance disclosure would change the verification task. The use of Chinese university students was justified because this population frequently encounters social media misinformation and reports uncertainty when judging online credibility¹³.

Sample size determination

The target sample size was set before data collection. The study aimed to retain at least 400 valid responses, with approximately 50 participants in each of the eight experimental cells. This target supported factorial comparisons, interaction testing, and sensitivity analyses after invalid responses were removed. Approximately 460 students were invited to allow for incomplete questionnaires, failed attention checks, and low-quality responses.

Given that associations between risk perception and emergency information-seeking behavior are usually modest in size¹⁴, the study retained a cell size large enough to stabilize main-effect and interaction estimates. Randomization balance across the eight cells was assessed using chi-square tests for categorical baseline variables and one-way ANOVA for continuous baseline variables before the primary models were fitted.

Power and sensitivity planning were defined before analysis. With a target retained sample of at least 400 participants and eight randomized cells, the study was adequately powered for small-to-moderate main effects in factorial models. The design was not powered to make strong confirmatory claims about three-way interactions, mediation, moderation, or logistic regression subgroup effects. Therefore, interaction, mediation, moderation, and binary verification-accuracy analyses were treated as secondary or exploratory and interpreted after the primary factorial models.

Experimental materials

Eight simulated social media posts were created for the 2 × 2 × 2 design. Each post described a plausible disaster-related situation relevant to university students, including heavy rainfall, urban flooding, temporary transport disruption, campus safety reminders, or short-term emergency preparedness. The posts used a screenshot-style layout similar to public-facing Chinese social media posts and contained a source label, headline, brief body text, time stamp, engagement indicators, and simple interface elements. No real disaster location, real victim, real emergency event, real official announcement, or identifiable public agency post was used.

The accurate versions contained internally consistent information and proportionate protective advice, such as checking official updates, avoiding flooded roads, and preparing basic necessities. The misleading versions used exaggerated threat claims, vague attribution, unsupported predictions, or unverifiable urgency statements¹⁵. The official-source condition used a fictional verified emergency-management-style label, whereas the peer-source condition used an ordinary forwarded student or peer account. The source names, locations, timestamps, engagement indicators, and interface details were fictional. Full text descriptions of all eight stimuli are provided in Table 2 so that readers can evaluate the independence of the source, veracity, and tone manipulations.

Pilot testing of stimuli

A pilot test was conducted with 30 university students who met the same eligibility criteria but were not included in the main experiment. Participants rated each post on realism, emotional intensity, source credibility, wording clarity, and perceived accuracy using 5-point scales. They also indicated whether any post appeared to refer to a real recent disaster event or a real government announcement.

Pilot testing was used to check realism and manipulation clarity rather than to make misleading posts obviously false, because misinformation may still be misjudged even under attentive processing conditions¹⁶. Stimuli were revised when wording was unclear, emotional contrast was too weak, threat wording was excessive, or content appeared too similar to a real event.

Three manipulation-check items were retained in the main questionnaire. Participants identified whether the post appeared to come from an official or non-official source, whether it appeared accurate or doubtful, and how emotionally threatening the wording appeared. Participants were not excluded only because they misjudged the veracity of a post, since misjudgment was part of the measured phenomenon. Exclusion was applied only when failed manipulation recognition occurred together with invalid response behavior, such as a failed attention check or patterned answers.

Experimental procedure

The experiment was administered through an online survey platform accessible by smartphone or computer from March 3, 2026, to March 21, 2026. After opening the link, participants viewed an electronic informed consent page describing the general purpose, voluntary participation, estimated completion time, confidentiality procedures, possible mild discomfort from disaster-related content, and the right to withdraw before submission. Participants confirmed that they were at least 18 years old and agreed to participate before entering the questionnaire.

Participants first completed baseline measures, including demographic characteristics, daily social media use, prior disaster experience, perceived information literacy, and general trust in official emergency information. The platform then randomly assigned each participant to one of the eight experimental cells using embedded individual-level randomization. Participants viewed one simulated post and could not return to previous pages after completing the post-exposure section.

Immediately after exposure, participants completed measures of perceived risk, perceived credibility, emotional response, verification intention, and sharing intention. The verification task required concrete behavioral choices rather than only intention ratings, because young users' confidence in detecting misinformation does not necessarily translate into accurate identification in realistic digital environments¹⁷. Participants selected verification actions from a fixed list before proceeding to the debriefing page.

At the end of the questionnaire, all participants received an immediate debriefing. The page stated that the post was created for research purposes, noted that some versions contained misleading elements, and reminded participants to verify disaster-related claims through official emergency management channels before sharing. Participants were also instructed not to copy, save, or forward any stimulus material outside the study.

Measures

Risk perception was measured with five items covering perceived severity, likelihood, personal relevance, campus or community impact, and need for precaution. These items reflected the established view of risk perception as a subjective judgment involving perceived consequences, uncertainty, and controllability¹⁸. Responses were recorded on a 5-point Likert scale from 1 = strongly disagree to 5 = strongly agree. Higher scores indicated higher perceived risk.

Perceived credibility was measured with four items assessing whether the post appeared believable, accurate, trustworthy, and worth relying on. Verification intention was measured with four items assessing willingness to check official channels, search for supporting evidence, compare multiple sources, and delay forwarding before confirmation. Sharing intention was measured with three items assessing willingness to repost, forward to classmates, or remind others based on the post. Fact-checking intention was treated as a separate construct because prior social media research links it to news literacy and trust judgments¹⁹.

Information literacy was measured before exposure with six items covering source evaluation, cross-checking habits, recognition of misleading headlines, awareness of emotional framing, ability to distinguish official from non-official information, and willingness to consult authoritative sources. Prior disaster experience was measured with two items on personal experience of natural disasters or emergency disruption and previous online searches for disaster-related information. General trust in official emergency information was measured with three items.

Verification behavior was scored using a task-based multiple-response rule. Participants could select any of eight actions: checking an official emergency management website, searching a verified government account, comparing the claim with recognized news outlets, checking the university emergency notice page, asking an unspecified friend, reading comments only, forwarding first and checking later, or doing nothing²⁰. The first four actions were coded as reliable verification actions, and each received one point; the latter four actions were coded as passive, unreliable, or premature-sharing responses and received zero points. The continuous verification behavior score, therefore, ranged from 0 to 4. A binary verification accuracy variable was coded as 1 when a participant selected at least one reliable verification action and did not select forwarding before verification. Variable definitions, item numbers, coding rules, and score construction are provided in Table 3.

Data quality control

Data quality control was completed before statistical analysis. Of 460 submitted or initiated responses, 34 were excluded, and 426 were retained for analysis. The exclusion sequence was fixed in advance: age ineligibility, non-student status, incomplete questionnaire, pilot-test participation, failed attention check, unrealistically short completion time, and patterned response. Completion time was considered unrealistically short if it was less than one-third of the median completion time. Straight-line responses were removed only when they appeared together with failed attention checks or unrealistically short completion times.

Duplicate or suspicious submissions were screened using non-identifying survey indicators, including repeated response patterns and abnormal submission timing. No names, student identification numbers, phone numbers, social media account names, facial images, precise locations, IP addresses, or device identifiers were exported for analysis. Cases with missing values on primary outcome variables were excluded from the main analysis. Because the questionnaire required completion of core experimental items before submission, missingness in primary variables was expected to be minimal. The screening criteria and operational definitions are listed in Table 4.

Ethical considerations

The study used limited concealment only for the presence of misleading elements in some posts. It did not conceal the general topic, task type, voluntary nature, confidentiality procedures, or possible mild discomfort. This concealment was necessary to preserve the validity of the verification task and was followed by immediate debriefing. The disaster-related materials avoided real locations, real victims, real emergency cases, graphic images, and real government announcements because irresponsible handling of emergency misinformation can increase anxiety and interfere with appropriate public response²¹.

The anonymized dataset contained only research variables and randomly generated participant codes. Raw survey exports, screened analytic data, codebook files, stimulus materials, and statistical syntax were stored separately on a password-protected computer accessible only to the research team. No personally identifiable information was exported, analyzed, or shared.

Statistical analysis

Statistical analysis was conducted using IBM SPSS Statistics version 27.0 and R version 4.3. Continuous variables were summarized as means and standard deviations. Categorical variables were summarized as frequencies and percentages. Internal consistency of multi-item scales was assessed using Cronbach's alpha, with α ≥ 0.70 treated as acceptable for group-level analysis.

Before hypothesis testing, manipulation checks were examined. Independent-samples t tests compared perceived emotional intensity between neutral and high-threat conditions. Chi-square tests examined recognition of official versus peer source cues. Perceived accuracy ratings were compared between accurate and misleading posts. Randomization balance was examined across the eight experimental cells before outcome models were fitted.

The primary outcomes were risk perception, perceived credibility, verification intention, sharing intention, and the continuous task-based verification behavior score. The binary verification accuracy variable, interaction terms beyond planned two-way contrasts, mediation, moderation, and robustness models were treated as secondary or exploratory. Factorial ANOVA and general linear models were used for continuous primary outcomes. Two-way and three-way interaction terms were included to examine whether the effect of information veracity varied by source cue or emotional tone. Logistic regression was used for binary verification accuracy, with odds ratios and 95% confidence intervals reported.

Mediation and moderation models were treated as secondary analyses. Perceived credibility was tested as a mediator between source cue and verification intention. Risk perception was tested as a mediator between emotional tone and sharing intention. Information literacy was tested as a moderator of the association between perceived credibility and sharing intention. Because misinformation responses are shaped by cognitive evaluation, trust judgment, and sharing motivation rather than message accuracy alone²², these models were interpreted after the primary factorial models. Indirect effects were estimated using 5,000 bootstrap resamples.

Robustness checks were conducted in three steps. First, the primary models were repeated after excluding participants who failed any manipulation-check item. Second, models were adjusted for gender, education level, daily social media use, prior disaster experience, and baseline trust in official emergency information. Third, verification behavior was analyzed both as a continuous score and as a binary accuracy variable. Statistical significance for primary outcomes was set at p < 0.05. Secondary analyses were interpreted by effect direction, confidence intervals, and consistency with the primary models rather than by marginal p-values alone. Effect sizes were reported as partial eta squared, Cohen's d, standardized coefficients, or odds ratios according to model type.

Data management and reproducibility

The anonymized analytic dataset contained one row per participant and one column per variable. It included participant code, experimental condition, demographic variables, baseline measures, manipulation-check items, post-exposure outcomes, verification behavior indicators, composite scores, and exclusion flags. Participant codes were randomly generated and were not linked to student identity.

Composite scores were calculated by averaging valid items within each scale after confirming item direction. Higher scores consistently indicated higher levels of the measured construct. Reverse coding was applied only when required by item wording. The data-cleaning log documented all exclusions, recoding decisions, composite-score calculations, and model specifications. The reproducibility package included the anonymized analytic dataset, codebook, stimulus descriptions, screening log, and statistical syntax. Transparent documentation of stimulus design, randomization, exclusion rules, scoring procedures, and analysis syntax was maintained to support reproducibility in misinformation and risk communication research²³.

Results

Participant characteristics and experimental allocation

After eligibility screening, consent confirmation, completion checks, attention-check screening, and data-quality review, 426 valid responses were retained for analysis. The final sample included 171 male participants, 239 female participants, and 16 participants who preferred not to report gender. The mean age was 21.01 years (SD = 1.96). The sample included 388 undergraduate students and 38 postgraduate students. Partic...

Discussion

This randomized experiment examined how Chinese university students responded to simulated disaster-related social media posts that varied by source cue, information veracity, and emotional tone. The results showed three main patterns. First, misleading disaster information and high-threat wording increased perceived risk. Second, official-source cues increased perceived credibility and sharing intention. Third, task-based verification behavior was more strongly associated with source cues than with information veracity ...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
IBM SPSS Statistics	IBM Corp.	Version 27.0	Conduct descriptive statistics, reliability analysis, t tests, chi-square tests, factorial ANOVA, and general linear models
Online survey platform	Wenjuanxing/Questionnaire Star or equivalent online survey system	Online survey system; version not applicable	Administer eligibility screening, electronic informed consent, baseline questionnaire, random assignment, stimulus exposure, post-exposure measures, and verification task
R statistical software	R Foundation for Statistical Computing	Version 4.3	Conduct statistical analysis, robustness checks, logistic regression, and bootstrap-based secondary analyses
Statistical syntax/codebook files	In-house	SPSS/R syntax and codebook	Document variable coding, exclusion rules, composite-score construction, and model specifications
Stimulus screenshot-style interface templates	In-house	Fictional Chinese social media post layout	Present simulated disaster information in a realistic social media format without using real disaster events or real government announcements