Risk perception is an individual's subjective perception and risk evaluation, including the likelihood of risk occurrence, potential losses, and the importance of these losses to the individual (Slovic, 1987). In the emergency management field, risk perception refers to an individual's perception and assessment of various potential hazards (such as natural disasters, public health events, etc.), which affects their level of concern regarding risks and their motivation to implement response measures (Kellens et al., 2012; Li et al., 2020, 2024). For example, in the face of an epidemic, individuals' perception of infection risk affects their willingness and effort to obtain relevant information.

Risk perception includes dimensions such as the likelihood of risk occurrence, severity of risk consequences, and an individual's controllability of risk. These dimensions are interrelated and constitute an individual's overall risk perception. For example, individuals may believe that the likelihood of an earthquake occurring in a certain location is low; however, if an earthquake occurs, the consequences are serious, which can also affect individuals' overall perception of earthquake risk (Li et al., 2020).

Various factors—including an individual's personal experience, knowledge level, and sociocultural background—influence risk perception. Additionally, risk perception dynamically adjusts to changes in time and context. For example, individuals who have experienced earthquakes may have a deeper understanding of earthquake risks, fostering greater alertness and positivity toward subsequent emergency preparedness. Moreover, external information—such as media reports and expert advice—can affect individuals' risk perception (Li et al., 2020; Slovic, 1987).

The core theoretical framework of risk perception involves individuals' subjective cognition of the possibility, severity, and controllability of risk events (Slovic, 1987). Its theoretical development is primarily based on Protection Motivation Theory (PMT) and the Theory of Planned Behavior (TPB). Notably, PMT posits that individuals develop protective motivation upon perceiving a threat, which in turn drives adaptive behaviors, such as information seeking (Rogers, 1983). Further, TPB emphasizes the influence of subjective norms and perceived behavioral control on behavioral intention, and risk perception—as an attitude variable—indirectly affects information seeking behavior through behavioral intention (Ajzen, 1991).

Emergency information-seeking behavior refers to individuals actively seeking information associated with risks when facing emergency situations, including information on the nature of the emergency, possible impacts, and response measures (Carson et al., 2023; Kellens et al., 2012). This behavior is intended to help individuals better understand risks, reduce uncertainty, and make wiser decisions to protect their own safety and that of others.

Emergency information seeking behavior is purposeful and proactive. After perceiving risks, individuals selectively search for and obtain relevant information based on their needs and goals. Simultaneously, various psychological factors—such as individual cognition, emotion, and motivation—influence this behavior. For example, in the early stages of a viral outbreak, people may actively search for information regarding the transmission routes, symptoms, and prevention methods of the virus to satisfy their need to understand and respond to the epidemic (Carson et al., 2023; Kellens et al., 2012).

Information seeking in emergency situations is the process whereby individuals actively acquire information to reduce uncertainty and satisfy their decision-making needs (Kellens et al., 2012). Information Needs Theory suggests that when risk perception exceeds a threshold, individuals' subjective need for information increases, prompting them to invest more resources in information searches. Cost-Benefit Theory suggests that information seeking behavior is the result of individuals balancing time cost, cognitive load, and information value. High risk perceptions may enhance individuals' evaluations of information value, thereby increasing their willingness to search (Carson et al., 2023; Kellens et al., 2012). These theories suggest potential mechanisms driving emergency seeking behavior.

Each theoretical framework focuses on different interpretations of the risk–behavior relationship. Notably, PMT emphasizes the dual pathway regulation of threat assessment and response effectiveness (Rogers, 1983), which may overestimate the mediating role of cognitive assessment, while the Risk Information Seeking and Processing Model (RISP) focuses on the interactive effects of emotional responses and information sufficiency (Griffin et al., 2008). A meta-analysis found that studies using the TPB were more likely to identify the moderating effects of subjective norms (with an average effect value reduction of 0.07–0.12), indicating that the choice of theory may systematically affect the interpretation of results.

Risk perception and emergency information-seeking behavior have a close relationship. The higher an individual's risk perception, the stronger their emergency information-seeking behavior (Kellens et al., 2012). This is because when individuals perceive higher risk, they feel greater pressure and uncertainty, which prompts them to actively seek relevant information to reduce this pressure and uncertainty to protect themselves. Additionally, emergency information-seeking behavior is a goal-directed behavior whereby individuals actively obtain the “essence of the event, scope of impact, and response measures” to reduce uncertainty and protect their own and others' safety (Kellens et al., 2012). The occurrence of this behavior requires simultaneous satisfaction of the “motivation drive” (risk perception triggering), “conditional support” (behavioral control ability), and “emotional regulation” (emotional and information sufficiency perception), which correspond to the core concerns of PMT, TPB, and RISP, respectively, and constitute the integrated framework's behavioral endpoint (Li et al., 2020; Tang and Li, 2025).

Numerous studies have demonstrated that risk perception is an important factor affecting emergency information-seeking behavior. For example, in the field of natural disasters, studies have found that the higher people's perception of earthquake risk, the greater their likelihood of actively obtaining earthquake warning information and emergency knowledge before an earthquake occurs (David et al., 2024; Piracha, 2021). During public health events, people's perceptions of the epidemic risk can also affect their attention to and search frequency for epidemic-related information (Li and Li, 2024; Park et al., 2024; Zhao and Cai, 2009). Some studies have suggested that this relationship may be moderated by various factors such as individual differences, situational factors, and information dissemination channels. For example, individuals of different ages and educational backgrounds may have different perceptions of risk and information-seeking behavior, and the relationship between risk perception and information-seeking behavior may vary across diverse sociocultural contexts (Li and Yang, 2025). Therefore, an in-depth exploration of the relationship between risk perception and emergency information seeking behavior, as well as the moderating factors that affect this relationship, is of great significance for improving emergency management and enhancing the public's emergency response capabilities.

When conducting and reporting our system review, we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009) and conducted a systematic search with the participation of professional research librarians to identify relevant literature. Firstly, on May 2th, 2025, we searched the following these international electronic databases (Web of Science, Scopus, Elsevier, Wiley, ProQuest, CNKI, and Google Scholar) were searched to identify potentially relevant published and unpublished studies (Barends and Rousseau, 2018). We combined keywords and Boolean search strings (such as “^*”, “NOT”, “OR”, and “AND”) for systematic search: for example, in the Web of Sciences database, we used TS=(“risk perception^*” OR “perceived risk^*” OR “risk awareness”) AND TS=(“emergency information^*” OR “crisis^* information^*” OR “disaster information^*”) AND TS=( seek^* OR search^* OR retriev^* OR find^* OR use^*) AND TS=(behavio^* OR intention^* OR willingness^*) for retrieval, where TS is the subject word. Secondly, to minimize the impact of publication bias and obtain as the maximum number of potentially interesting studies as possible, we have also included unpublished conference papers, research proposals, and papers in the aforementioned database. Finally, we use the snowball method was used to review the reference sections of the literature obtained to identify other potential sources. A total of 632 studies were reviewed, and those that met the systematic review criteria were included. The search process is illustrated in Figure 2.

The inclusion criteria for constructing the dataset were as follows: (a) this study included any types of research design articles that evaluate the risk perception and emergency information seeking intention or behavior; (b) the study was a quantitative study, with the outcome indicators being emergency information seeking intention or behavior; and (c) this study reported the correlation coefficients [or other convertible coefficients, such as d, r², and standard error (SE)] and the sample size. We excluded studies that (a) did not investigate the risk perception and emergency information seeking intention or behavior; (b) duplicated results or studies; or (c) lacked full-text information sources such as comments, meeting minutes, or editorials.

To verify the accuracy of the coding, a coding team was established comprising three researchers with health and management knowledge as well as methodological experience. The coding process was completed by two researchers who independently coded each included study. First, through frequent discussions between the coding teams, the two researchers reached an agreement on coding standards and began the first round of experimental coding, covering many features that the coding team believed might be related to post-hoc analyses. These include, but are not limited to, author, publication year, region, sample size, demographic characteristics (gender, age, and education level), risk perception, region, theoretical background, measure method, and the main findings. Second, the two researchers independently encoded all the relevant research features based on the revised coding scheme. Third, two researchers cross-checked and verified that the coder's coding consistency reached 85% and then independently screened the title and abstract to determine which studies could be included. Articles that could not be identified by their titles and abstracts were screened for full-text. When the two researchers had different opinions, a third researcher decided whether to include or exclude the study. The two researchers independently extracted the data from the included studies and merged the results into a final data table. During the encoding and extraction processes, differences and disagreements were resolved through discussion and/or by a third reviewer. All excluded articles and the reasons for exclusion were explained.

Most included studies are based on reported correlation coefficients that examine the relationship between risk perception and information behavior. The correlation coefficient is usually marked as “R” or “r”. However, other coefficients [such as R², d, or standard deviation (SD)] have also been used in some studies. In this case, it is necessary to convert these coefficients into r, as follows (Others can be find in Barends and Rousseau, 2018; Card, 2011; Li et al., 2020):

where m1 and m2 are the means of Groups 1 and 2, respectively, and s is the pooled SD across the two groups.

R² was the determinant coefficient, and n was the sample size.

W_i = n – 3, where Z is the Z-score when calculating Fisher's Z.

We conducted a bias risk assessment of the included emergency information seeking studies and the data were independently extracted by two researchers. Bias risk assessment is a measurement tool used to evaluate the quality of research and can provide a reference for the robustness and scalability of the research results. It comprises seven items, with the answer options of “yes” “no” and “unclear” for each item. This tool was adapted from previously developed bias risk tools for observational research (Bero et al., 2018; Higgins et al., 2024; Schünemann et al., 2013).

To determine the relationship more truthfully and reliably, we used Schmidt and Hunter's random effects model (2015) for analysis; Microsoft's Excel 2021 (version 16.76) was used to extract and organize data; R (version 4.3.1)'s “metafor” package and Comprehensive Meta Analysis Version 3 software (CMA 3.3.07; Biostat, USA) were used for meta-analysis, moderator analysis, publication bias analysis, etc. First, we used the correlation coefficients reported in each study, the internal consistency reliability coefficient (α), and then we calculated the true correlation coefficient (p) between risk perception and emergency information seeking behavior (p also known as the zero-order correlation coefficient, which is the correlation coefficient after weighted observation of the average sample size or the true correlation coefficient after measurement error correction). For the few studies that reported other effect values (such as d, mean, and SD), we converted them into a unified comprehensive effect value r. For studies without reported reliability, internal consistency reliability tends to overestimate the reliability of score-based standards (and, conversely, underestimate the correlation of corrections), α corrected the magnitude of the attenuation effect, as internal measurement errors are usually smaller than inter measurement errors (Lebreton and Senter, 2008). Therefore, we used the weighted average p of all studies that reported the reliability of this variable. For studies that included multiple and/or outcome measurements, we calculated a comprehensive r for inclusion in the overall meta-analysis (Barends and Rousseau, 2018). Subsequently, to better evaluate the impact of publication bias in the included studies, we used a combination of the Failsafe N (Rosenthal) and Egger regression analysis methods, as both methods have advantages and disadvantages in testing different samples. For example, the Egger regression analysis is prone to errors when the sample size is small. Therefore, it is unsuitable for testing small-sample studies. Failsafe N (Rosenthal) can compensate for this deficiency, which reflects the number of studies required to reverse the results of the meta-analysis and estimate the impact of potential unpublished negative results on positive meta-analysis results (Barends and Rousseau, 2018; Li et al., 2020). When the number of Failsafe N was much greater than the number of included studies, particularly when the number of studies exceeded 5 K + 10, where K is the number of included studies (Card, 2011), it is difficult to change the results of the meta-analysis. However, its judgment criteria are not the golden rule; thus, we combined the two to evaluate the impact of publication bias on the included studies. This is conducive to reflecting the robustness of the meta-analysis results regarding any systematic omissions in the published literature that yielded no significant results. Finally, the quality of the included studies was evaluated as the basis for the reliability and scalability of the research results.

After screening, 50 publications were included, with a total sample size of 29,014 participants, most of whom were female (50.69%). These studies involved nine countries across three continents, covering both developing and developed countries, and demonstrated significant growth trends over the past 5 years (2021–2025). Specifically, the samples came from the following continents and countries: Asia (k = 36), North America (k = 7), Europe (k = 5), and Africa (k = 2); China (k = 32), United States (k = 7), South Korea (k = 3), South Africa (k = 2), Netherlands (k = 2), Spain (k = 1), India (k = 1), Belgium (k = 1), and Greece (k = 1). The vast majority of research came from developing countries (k = 35). Among the 50 included studies, from a theoretical perspective, most were based on TPB and the theory of reasoned action (k = 27), while grounded theory, PMT, Uncertainty Theory, and others were relatively rare. The research methods mainly included SEM (k = 37) and hierarchical regression (HR) (k = 13). Online sample collection methods (k = 23) included email, online questionnaires, and telephone interviews; offline methods (k = 19) included distributing questionnaires and conducting onsite interviews; and mixed collection involved combining online and offline methods (k = 8). Among the four types of emergency situations, most were related to health (k = 26), social safety (k = 15), natural disasters (k = 6), and accident emergencies (k = 3). Table 1 presents detailed information.

Table 2 summarizes the results of the meta-analysis of risk perception and emergency information-seeking behavior. A significant positive correlation was found between overall risk perception and emergency information-seeking behavior (r = 0.10, random effects model; r = 0.08, fixed effects model; k = 66, n = 43,717). The confidence interval did not include zero, indicating that risk perception significantly positively impacted emergency information-seeking behavior, which suggests that risk perception significantly influences emergency information-seeking behavior.

Specifically, regarding the correlation between risk perception and willingness to search for emergency information, the random effects model indicated a significant positive correlation between the two (confidence interval did not include zero), with a composite effect value of 0.08 (k = 19, n = 14862), which was slightly higher than the composite effect value under the fixed effects model (r = 0.04).

Concerning the correlation between risk perception and emergency information-seeking behavior, the random effects model indicated a significant positive correlation between the two (confidence interval did not include zero), with a comprehensive effect value of 0.12 (k = 19, n = 14,862), which was slightly higher than the comprehensive effect value under the fixed effects model (r = 0.11).

As Hunter and Schmidt (2004) highlighted, differences in research results may be attributable to statistical artifacts or potential regulatory factors, such as disaster type and demographic and methodological characteristics. Therefore, in addition to core hypothesis testing, we conducted an exploratory moderation factor analysis to evaluate whether our findings differed owing to the demographic and methodological features primarily studied in the database. Table 3 presents the results of the analysis wherein we tested two demographic characteristics, national development level (developed and developing countries) and male proportion (male proportion greater than or equal to 50% and male proportion less than 50%), as moderating factors for risk perception and the willingness to search for emergency information. Additionally, we considered emergency types [i.e., four types of disaster emergencies (natural disaster emergency, accident emergency, health emergency, and social security emergency)] and several methodological features [i.e., publication time, sample collection strategy (online or offline), and measurement methods (SEM and HR) as moderating variables]. Overall, publication time had a weak or no moderating effect (e.g., in the relationship between overall risk perception and overall emergency information-seeking behavior, risk perception and emergency information-seeking willingness, and risk perception and emergency information-seeking behavior). The regression analysis of publication time showed no impact (the estimated value was not significant). In the analysis of cultural background (national development level), developed countries had a slightly stronger moderating effect on the relationship between overall risk perception and overall emergency information-seeking behavior, as well as on the relationship between risk perception and emergency information-seeking intention, than developing countries. Notably, developing countries had a stronger moderating effect on the relationship between risk perception and emergency information-seeking intention than developed countries (0.12 vs. 0.07). The four types of emergencies had different regulatory effects. In the relationship between total risk perception and total emergency information-seeking behavior, health and natural disaster emergencies had a significant positive regulatory effect, whereas accidents and social security emergencies had a weak negative regulatory effect (−0.06 and −0.02, respectively). Noteworthily, in the relationship between risk perception and emergency information-seeking willingness, health, accident, and natural disaster emergencies had a significant positive regulatory effect, while social security emergencies had a significant negative regulatory effect. In the relationship between risk perception and emergency information-seeking behavior, health, natural disaster, and social security emergencies had a significant positive regulatory effect, whereas accident emergencies had a significant negative regulatory effect. Concerning the method of collecting sample population data, offline methods, such as field surveys, had a slightly stronger moderating effect on all relationships than online methods, such as email, phone, and online questionnaires. Regarding measurement method, the significant positive regulatory effect of SEM measurement was stronger than that of HR measurement. Concerning sample population characteristics, the significant positive regulatory effect of a male proportion greater than 50% (i.e., male-dominated) on all relationships was stronger than that of a male proportion less than 50% (i.e., female-dominated).

Publication bias (i.e., selective reporting) refers to any situation during the processes of data collection, analysis, interpretation, and publication that may result in a systematic deviation of the conclusions from true results, such as self-examination by authors to meet theoretical expectations or the tendency of journals to support significant results. To alleviate this concern, we employed Failsafe N (Rosenthal) and Egger regression analyses to test for publication bias, both of which are usually based on defining the distribution of the effect size for a given zero-order relationship. Table 1 presents specific information on the publication bias analysis of the included studies. Owing to the inclusion of studies that did not satisfy the conditions for two or more articles and the fact that the standard error was not zero, the analysis of publication bias was limited, and the results could not be obtained. Table 1 presents the results of the publication bias analysis. The Failsafe N (Rosenthal) value was significantly greater than 5K + 10 (where K represents the number of included studies), the intercept terms of the Egger regression analysis were mostly equal to or close to zero, and all p-values were non-significant. For example, for overall risk perception and emergency information-seeking intention, the Failsafe N was 1,998 and 780, respectively (significantly greater than 5K + 10, where K is 66 and 19, respectively), and the p-values were 0.46 and 0.11, respectively (both greater than or equal to 0.05). Figure 3 depicts their overall funnel plot, which indicates that the effect values of all included studies were almost uniformly distributed. The funnel plot was utilized to check whether there was the possibility of reporting bias in the study. Therefore, we can conclude that, although the current study exhibited publication bias, its impact on the results was negligible.

Table 4 presents the risk of bias assessment are presented in Table 4. The bias tools primarily included five aspects: recall description, blind methods, random generation, task concealment, and other biases. Two reviewers conducted the biased risk of bias. Overall, the low, high, and unknown risk rates were 80.4%, 11.3%, and 8.3%, respectively. The average score of the 42 articles was low (80.4%), while five articles had high scores (11.3%), and three articles had unclear ratings (8.3%).

This study breaks through the universality assumption of traditional risk perception theory and confirms “situational sensitivity” in relation to emergency information-seeking behavior, adding a typological analysis dimension to emergency communication theory. At a practical level, emergency management departments need to develop differentiated strategies for different types of disasters. For health emergencies and natural disasters, real-time and accurate information should be strengthened (such as short science videos and intelligent terminal warnings), and addressing information transparency and trust rebuilding (such as third-party source intervention and multi-party communication mechanisms) should be prioritized. In response to the active search needs of the public in developing countries, efforts should be invested to strengthen the construction of localized emergency information platforms; to address gender differences, more targeted communication channels should be designed (e.g., female-dominated social media accounts focusing on mental health and family protection information) (Ajzen, 1991; Batrancea et al., 2025; Kellens et al., 2012; Slovic, 1987).

This study's results carry key practical implications for improving the effectiveness of information dissemination and public emergency response capabilities. First, understanding the impact mechanism of risk perception on emergency information-seeking behavior can help government departments and relevant institutions release and disseminate risk information more specifically when working on emergency management. Increasing the public's risk perception level can stimulate their willingness to actively search for emergency information and enhance their attention to and demand for emergency information, thus improving information dissemination's effectiveness (Batrancea et al., 2022a,b; Kellens et al., 2012; Wen, 2019).

Second, this study revealed differences in the relationship between risk perception and information-seeking behavior in different types of disasters, suggesting that relevant departments should adopt different information dissemination strategies when dealing with different types of disasters. For example, to prevent and respond to natural disasters, various information dissemination channels should be utilized, accurate and timely warnings should be released in advance, public risk awareness education should be strengthened, and risk perception and emergency information-seeking capabilities for natural disasters should be improved. In public health incidents, it is necessary to pay greater attention to the transparency and credibility of information, the timely release of epidemic dynamics, prevention and control measures, and scientific knowledge to alleviate public panic, guide people to search for and use health information reasonably, and actively cooperate with epidemic prevention and control efforts (Shen et al., 2022; Slovic, 1987).

Additionally, focusing on individual differences and the moderating effect of information dissemination channels can help further optimize the effectiveness of emergency information dissemination. More targeted information content and dissemination methods should be designed and provided to groups with different individual characteristics, such as age, educational background, and gender. For example, for the younger generation, social media and other online platforms can be utilized to publish emergency information in various forms that are easy to understand and disseminate, and traditional media channels and community promotions can be used to ensure that they obtain important emergency information in a timely manner. Simultaneously, it is necessary to strengthen collaborative cooperation between traditional and social media, leveraging their respective advantages; improve the quality and coverage of emergency information dissemination; better satisfy the information needs of the public in emergency situations; and enhance people's emergency response capabilities (Li and Yang, 2025; Slovic, 1987).

This study systematically explored risk perception's impact on emergency information-seeking behavior and its moderating factors through meta-analysis. This approach offers a comprehensive perspective and theoretical support for further understanding this relationship. In practical applications, the mechanism of risk perception and moderating factors—such as disaster type, individual differences, and information dissemination channels—should be fully considered to develop more scientific and effective emergency information dissemination strategies to improve the public's ability to respond, ensuring safety in emergency situations.

Although this study makes valuable contributions, some limitations persist. First, the literature reviewed was concentrated in certain countries and regions and cannot fully represent risk perception and emergency information-seeking behavior worldwide. Significant differences may exist in people's risk perceptions, information search habits, and responses to disasters in different cultural backgrounds, and the geographical limitations of existing research affect the generalizability of the results (Li et al., 2020).

Second, the literature contains diverse measurement tools and indicators for evaluating risk perception and emergency information-seeking behavior, resulting in heterogeneous results. Some studies utilized subjective perception measurement methods, whereas others concentrated on objective behavioral recording. Differences in measurement methods complicate cross-study comparisons and make integration complex (Barends and Rousseau, 2018; Hedges and Tipton, 2009).

Finally, this study has some limitations in the research design. Most included studies had a cross-sectional design, rendering it difficult to determine the causal relationship between risk perception and emergency information-seeking behavior. Cross-sectional studies can only reflect associations at specific time points and cannot capture the dynamic changes and causal directions of variables over time, nor can they provide in-depth explanations of the mechanisms underlying the moderating variables (Barends and Rousseau, 2018).

Future research should focus on the following directions: First, samples should be expanded to cover more countries, regions, and cultural groups to enhance the universality and cross-cultural applicability of the results. By comparing risk perception and emergency information-seeking behavior across different cultural backgrounds, we can gain a deeper understanding of the role of cultural factors and provide a foundation for developing cross-cultural emergency information dissemination strategies. Second, developing and validating standardized risk perception and emergency information-seeking behavior measurement tools will improve the consistency and comparability of results, reduce heterogeneity caused by differences in measurement methods, and enable more accurate comparisons between different studies and integration. Third, longitudinal research designs should be strengthened. A longitudinal research design should be utilized to track the changes in risk perception and emergency information-seeking behavior of the same group at different time points, which will help determine causal relationships and temporal evolution patterns, and increase the understanding of the dynamic mechanisms of the moderating variables.

Fourth, the mechanisms of moderating variables should be further explored. Moderating variables that impact the relationship between risk perception and emergency information-seeking behavior include individual differences, disaster types, and information dissemination channels. To explore the interaction between individual differences and specific types of disasters, it is important to examine the specific effects of information dissemination channels in different types of disasters and cultural backgrounds, and their interactions with risk perceptions. Additionally, attention should be paid to other potential moderating variables, such as social support and community environment, to more comprehensively elucidate the complex relationship between risk perception and emergency information-seeking behavior.

Fifth, multidisciplinary perspectives should be integrated. We need to encourage interdisciplinary collaborative research and explore risk perception and emergency information-seeking behavior from multiple perspectives by combining theories and methods from psychology, sociology, communication studies, information technology, and other disciplines. Psychological experimental methods and models can be employed to accurately measure the psychological mechanisms of risk perception, drawing on the theoretical framework of sociology to study the role of social networks and social capital. Communication studies can be utilized to study and evaluate the impact of media channels and information presentation methods, and we can explore the potential application of emerging technologies in emergency information dissemination and risk perception monitoring, considering the development trends of information technology (Li et al., 2020).