Effective public health messaging for university students: lessons learned to increase adherence to safety guidelines during a pandemic

The study addresses the problem of variable adherence to COVID-19 safety behaviors on university campuses, settings characterized by dense social interaction and students arriving from diverse geographic regions. In the U.S., campuses saw elevated transmission, underscoring the need to understand what drives compliance. Universities implemented testing, masking, distancing, and modified operations in Fall 2020, but questions remained about how best to foster protective behaviors. The authors examined how information sources, COVID-19 knowledge, and risk perception relate to personal and peers’ safety behaviors among students during the Fall 2020 semester. They hypothesized that greater COVID-19 knowledge and more accurate risk perception would each, independently and jointly, be associated with more protective behaviors. They predicted that reliance on official sources (e.g., CDC/WHO websites) and a large university course at Washington University in St. Louis (The Pandemic: Science and Society) would be associated with higher knowledge, more accurate risk perception, and safer behaviors.

Prior work links adherence to public health guidance with health literacy, perceived social norms, and a sense of social responsibility. Students often underestimate peers’ preventive behaviors; providing normative information can improve adherence. Educational interventions and courses can enhance knowledge and influence culture and safety practices on campuses, yielding spillover effects. Modeling studies highlighted frequent testing and mask use as effective but offered limited guidance on behavior change strategies. Research also shows information sources matter: social media and informal sources (family/friends) can be associated with lower knowledge and more misinformation, while official health organization sources are linked to more accurate information. Punitive approaches (e.g., strict enforcement) were used in some universities, but shifting social norms and providing reliable information are proposed as effective alternatives.

Design and setting: Cross-sectional online survey conducted at the end of the Fall 2020 semester (the first full semester since the pandemic began), centered on Washington University in St. Louis (WUSTL) and open more broadly. The context included a large, intensive 2‑credit online course, The Pandemic: Science and Society, offered in the three weeks prior to the Fall 2020 start, designed to convey reliable COVID-19 information and promote safe behaviors. Participants completed assignments that communicated course content to wider audiences. Participants: 1155 survey responses were recorded. Inclusion required completion of the questionnaire and consistency checks; 928 respondents met criteria and were analyzed. Demographics reported include university affiliation (majority from WUSTL), housing status (on-campus, off-campus near campus, or living at home), and exposure to the course (took, had some contact with content, or no exposure). Measures: - Personal safety score (pSafety): For multiple behavior items (e.g., mask use outside one’s bubble; attending large gatherings; indoor dining; time spent in crowded settings), response options were mapped to 0–4, with higher values indicating safer behavior (e.g., mask use 85–100% = 4; 70–84% = 3; 50–69% = 2; etc.). Scores were averaged per respondent (range 0–4). - Closest associates’ safety score (caSafety): Same scoring approach applied to respondent-reported behaviors of their 2–5 closest associates during Fall 2020 (range 0–4). This proxy was intended to reduce social desirability bias in self-reports. - COVID-19 knowledge (CKnow): Knowledge items about SARS-CoV-2/COVID-19 were assessed and summarized as a score; categorical summaries (e.g., 0–3 corresponding to correctness levels) were used in analysis. - Risk perception: Respondents rated risk for 10 activities as no/minimal risk vs somewhat/very risky (or unsure). “Unsure” received 0. Correct identification of risky vs non-risky activities (based on established transmission knowledge) yielded up to 10 points (e.g., large indoor gatherings = risky; outdoor walk/picnic or delivered groceries = not risky). - Information sources: Respondents selected their top three sources influencing their COVID-19-related behavior from categories including family; CDC/WHO/other health organization websites; television/radio/print news; friends; The Pandemic course; social media; official campus communications; government press briefings; other dashboards; healthcare provider; other university courses; other. Exclusion criteria: Incomplete surveys and conflicting answers (e.g., denial of course exposure followed by endorsement of course-derived content) were excluded. For top sources ranking, only ranks 1–3 were retained; values above 3 were removed. Statistical analysis: Within knowledge-score strata, distributions of primary information sources were compared via chi-square tests. Ordinal logistic regression (MASS::polr in R 4.0.3; Nagelkerke’s R² via DescTools) examined associations among CKnow, pSafety, caSafety, and risk perception, treating each as outcome in turn and testing interactions. Confidence intervals and AIC were reported; significance was assessed with 95% CIs and p-values (chi-square p < 0.05).

- Sample and exposure: Of 928 analyzed respondents, 83% were from WUSTL. Housing: 46% off-campus near campus; 36% on-campus; 14.5% living at home. Course exposure: 46.7% took The Pandemic course; 26.5% had some contact with its content; 25.4% had no exposure. - Top information sources (proportion selecting as a top-three source): Family 52.0%; CDC/WHO/other health organization websites 50.0%; television/radio/print news 47.4%; friends 38.6%; The Pandemic course 32.4%; social media 21.2%; official campus information 19.8%; government press briefings 14.3%; other COVID dashboards 13.8%; healthcare provider 7.3%; other university courses 4.0%; other 2.9%. - Knowledge and behaviors: Higher COVID-19 knowledge scores were associated with fewer attendances at large gatherings and indoor restaurant dining; similar patterns were observed for closest associates. Relationships between CKnow and pSafety/caSafety were significant but with low Nagelkerke’s R² in individual models. - Information sources and knowledge/behavior: Official health organizations and The Pandemic course were associated with higher knowledge and safer behavior scores than other sources. Among those listing CDC/WHO sites, a larger share scored ≥50% on knowledge compared to <50%. For The Pandemic course, 15.4% of those citing it as a top source answered all knowledge questions correctly versus 2.7% who answered all incorrectly. Among respondents with perfect knowledge (CKnow = 100%), 69% took the course, 13% had some contact, and 19% had no contact. Reliance on family/friends or social media was associated with lower knowledge and lower personal safety scores. Those relying on government press briefings tended to have lower personal and associate safety scores. - Risk perception: Most respondents (74.86%) had high risk-perception scores (9–10 correct of 10). Higher risk perception was associated with using official sources (e.g., CDC) but not with friends, TV, or social media. Safety behaviors were influenced by knowledge and accurate risk perception; models explained a meaningful portion of variance in personal safety (overall R² ≈ 0.33). Chi-square tests indicated distributions of information-source use differed significantly across knowledge and behavior score categories (p < 0.05).

Findings indicate that where students obtain COVID-19 information is linked to their knowledge, risk perception, and safety behaviors. Official health organization sources and a well-designed university course were associated with higher knowledge, more accurate risk appraisal, and more protective behaviors. Conversely, reliance on informal sources (family/friends) and social media was associated with lower knowledge and less safe behavior. These patterns align with broader literature on the roles of health literacy, social norms, and information quality. Given documented campus-to-community spillover in COVID-19 transmission, improving information environments on campuses can contribute to protecting both students and surrounding communities. The large-scale Pandemic course at WUSTL exemplifies a feasible educational intervention that models desired behaviors, communicates reliable information, and fosters a safety-oriented community—mechanisms known to strengthen adherence. The observed associations between higher risk perception and safer behaviors reinforce the importance of risk communication that is accurate without inducing fatigue or dread. Results suggest universities can leverage official sources, structured courses, and normative messaging to raise knowledge and shift behavior at scale.

The study shows that students’ preferred information sources are strongly associated with their COVID-19 knowledge, risk perception, and adherence to protective behaviors. Official health information (CDC/WHO) and a comprehensive university course were linked to better knowledge and safer practices, whereas informal and social media sources correlated with lower knowledge and fewer protective behaviors. Universities can meaningfully improve public health outcomes by providing reliable, accessible information and educational programming that normalizes and reinforces safety behaviors. Future work should: (1) test causal impacts of information interventions using longitudinal or experimental designs; (2) assess heterogeneity across student subgroups; (3) evaluate integration of official content into social media channels to counter misinformation; and (4) examine persistence of behavior change over time and spillover to surrounding communities.

- Self-reported behaviors are subject to recall bias and social desirability bias; inclusion of closest-associate behavior was an attempt to mitigate this. - Cross-sectional design with a single time point limits causal inference and assessment of temporal dynamics or message fatigue. - Potential measurement noise/typos in self-report and categorization of information sources; no detailed breakdown of specific news outlets/platforms. - Limited generalizability beyond university-affiliated populations; demographic subgroup analyses were limited. - Possible contemporaneous influences (e.g., variability in government communications during the study period) may have affected perceived reliability and behaviors.