Dynamics of natural hazard risk awareness: Panel analysis insights from Switzerland

The analysis of risk awareness should be the initial stage in integrated natural hazard risk management to promote appropriate and effective measures for mitigating risks and strengthening social resilience inside the multi-risk framework. Nevertheless, earlier studies focused on cross-sectional data and overlooked the changes in risk awareness levels and associated independent variables with time. This study analyzes for the first time a balanced nationwide panel dataset of 1612 respondent-year observations from Switzerland (period 2015 – 2021, including the epidemic of COVID-19) to examine and compare the effects of potential independent variables on the four dimensions of natural hazard risk awareness (NHRA), ranging from the broadest dimension of Relevance to higher dimensions of Perceived Probability of an event, Perceived Threat to life and valuables, and Perceived Situational Threat . The analysis in this study incorporates multiple methods of Random-Effect Model (RE), Generalized Linear Model (GLM), and mediation analysis. Results show that NHRA increased in Switzerland to different extents (up to 23.24%) depending on the dimension. Event memory, perceived information impact and reported individual informed level appeared to be the most consistent independent variables positively influencing panel NHRA. Among these, perceived information impact as an important indicator of risk communication, was also found to serve as a mediator from risk preparedness to risk awareness. By encouraging residents to engage in “ Begin Doing Before Thinking ” (BDBT) programs to leverage subliminal effects and self-reflection, this study proposes that behavior-cognition feedback loops may facilitate a virtuous cycle. Our promising observations provide recommendations for an effective awareness-rising strategy design and suggest extensive insights from potential short-interval panel analysis in the future.


Introduction
Natural hazard risk awareness (NHRA) plays a critical role in promoting effective risk management and social resilience.However, traditional approaches often consider risks in isolation, neglecting the interconnected nature of various hazards.The multi-risk framework emphasizes the need for integrated risk management strategies that account for the interplay between different types of risks in a defined timeframe, including natural, technological, biological, and socioeconomic risks.A comprehensive multi-risk assessment must consider Hazard (physical phenomena that can cause damage), Exposure (elements at risk, such as buildings and infrastructure), and Vulnerability (susceptibility to damage, considering economic, social, and environmental factors) (Field et al., 2012;Gallina et al., 2016;Hochrainer--Stigler et al., 2023).As climate change advances, inducing glacial melting and permafrost thawing, the potential for natural hazard events increases (Braunschweiger and Ingold, 2023).This situation, combined with expanding infrastructures in high-risk regions and a growing but vulnerable population, especially during and post the COVID-19 pandemic, demands an adaptable framework for risk management to address the challenges featured by high uncertainties and considerable spatio-temporal variability (Walker et al., 2014).The evolution of risk management strategies further underscores the importance of risk awareness, which embodies the layperson's attitudes towards hazards and risks (Bonaiuto et al., 2016).Since public and expert perspectives on natural hazard risks often diverge (Veland and Aven, 2013), understanding public risk awareness is critical for pinpointing areas requiring emphasis within the integrated risk management process, and thus should be a preliminary step in natural hazard risk management (Bradford et al., 2012;Wickes et al., 2015).Nonetheless, current risk awareness research is marked by four primary deficiencies: conceptual clarity, factor exploration, mechanism elucidation, and dataset quality.
Risk awareness, as a concept, traces its origins to the work of Starr in the 1960s (Starr, 1969).It has been studied extensively to understand the relationship between individuals and their surrounding environmental threats (Bronfman et al., 2016;Wachinger et al., 2013).The concept is applied equivalently to risk perception in most research (Maidl et al., 2021;Rohrmann and Renn, 2000) while the latter one is a more objective and quantifiable concept that encompasses the knowledge of potential risks and hazards.Risk awareness is more related to how individuals subjectively interpret and evaluate those risks and are influenced by emotions and cultural independent variables (Lechowska, 2018).Although NHRA is a multi-faceted construct, much of the existing research has predominantly examined it from a unidimensional perspective (Wachinger et al., 2013).This approach fails to capture the complexity and nuance inherent in the concept, and thus limits the comprehensiveness of understanding and management.
Numerous independent variables have been identified to influence NHRA, however, their consistency remains inadequately explored and summarized.The independent variables are broadly categorized into three classes: risk characteristics such as severity and frequency, individual demographics encompassing gender, age, education, marital status, household size, occupation, and financial situation (Chauvin et al., 2007;Lechowska, 2018), and interactions of these elements such as information dissemination, local cultural context and personal event experiences (Kahan et al., 2011).Notably, the impact of personal experiences on risk awareness differs among different geographical regions and weakens over time (Brilly and Polic, 2005), particularly if the negative outcomes were considered minor or unremarkable given that personal involvement and related emotional response from an event could play a significant role on sustained effects (Bubeck et al., 2012;Ho et al., 2008).Personal experiences related to climate change and major social events like the COVID-19 pandemic may influence people's cognitive processes concerning natural hazards among all their far-reaching socioeconomic consequences (Schneiderbauer et al., 2021).Moreover, components of political culture, like social integration and civic involvement, have been identified as important independent variables in shaping NHRA (Kuhlicke et al., 2011;Maidl et al., 2021).High trust levels, influenced by media and government protective measures, could potentially diminish risk awareness (Bronfman et al., 2020;Siegrist and Cvetkovich, 2000).Attitudes towards the controllability of natural hazards, characterized by self-efficacy and fate-belief, also impact NRHA in a positive and negative way, respectively (Mertens et al., 2018;Mızrak and Turan, 2022).However, we can hardly discover robust evidence of their validity in the current research throughout time, at various geographic dimensions as well as among various population groupings.The recognition of universal, potential, and prominent independent variables of NHRA across variations is key to customizing risk awareness strategies for diverse contexts.
Research has indicated the strong relationships among risk awareness, risk preparedness, and risk communication, yet leaving the influence mechanisms unclarified and underappreciated.Mediated by outcome expectancy, self-efficacy and action coping, risk awareness has been broadly proven to positively influence the intention to prepare (Ban et al., 2019;Martin et al., 2009).But this finding has also been challenged as augmented risk awareness may also hinder mitigation measures in certain instances (Paul and Bhuiyan, 2010;Shapira et al., 2018).On the other side, the potential inverse influence of risk preparedness on risk awareness remains largely unexplored.A few research articles suggest current preparedness levels could reduce risk awareness, subsequently impacting willingness for further preparedness (Bubeck et al., 2012;Qing et al., 2021).Another important element, risk communication, which encompasses interactive exchanges of information among interested parties, is vital for promoting risk awareness (Adu-Gyamfi and Shaw, 2021;Rollason et al., 2018).Information can be communicated through various sensory channels and in different formats, and the sources we collect information from can affect our judgment, emotions, and behavior.For instance, the dual sensory channel utilization in television, employing both visual and auditory cues, may evoke stronger responses than radio's single auditory channel (Loewenstein et al., 2001;Xie et al., 2011).People-centered communication regarding natural hazards tends to be more impactful than top-down government approaches (Haer et al., 2016).Incorporating both one-way information transfer and two-way dialogue among the public, stakeholders, and decision-makers (Renn, 2017), with a tailored strategy is thus advisable for efficient risk communication and NHRA management (Kuhlicke et al., 2011;Maidl et al., 2021).Unfortunately, comprehensive explorations of the relationship among NHRA, risk preparedness and risk communication are still scarce and await further research for explicit implications.
The cross-sectional data is insufficient for factor stability verification and mechanism exploration, while panel data with their capacity to track changes over time could provide robustness to these studies.Several studies have investigated the temporal stability of risk attitudes and trust preference with cross-sectional data related to a major event while seldom paying attention to the NHRA regularly (Cassar et al., 2017).Single measurements can cloud understanding of directional mechanisms, an issue that could potentially be addressed by employing repeated or longitudinal observations.However, making comparison of results across different studies is difficult because heterogeneity in research questions and measurement approaches poses a challenge.There are two solutions: either the research community strives for consistency in methodologies to enable more meaningful comparisons and a more nuanced understanding of NHRA, or more panel surveys shall be conducted, which is, however, hard to be financed within a single project.
This study for the first time provides nationwide panel data to examine and compare the effects of these independent variables on NHRA within the broader context of multi-risk.We use the results of two nationwide surveys on NHRA in Switzerland in 2015 and 2021 to construct impact mechanisms and search for potential solutions for raising awareness and integrated risk management.Specifically, the research questions to be addressed are.
(1) Did NHRA in Switzerland change during the period 2015-2021?(2) What independent variables have long-term influences on individual NHRA? (3) What mechanisms effectively induce change in NHRA? (4) How can capacity-building practitioners take advantage of these insights?
The assumptions underpinning this study are as follows.
( factor between other independent variables and NHRA, particularly between risk preparedness and risk awareness.
These assumptions align with our research questions and provide a robust foundation for understanding the dynamics of NHRA in Switzerland within the context of multi-risk.By addressing these assumptions, we aim to deliver actionable insights for policy makers to enhance integrated natural hazard risk management.
X. Sun et al.

Study period
During the study period between the first survey poll in 2015 and the second poll in 2021, a number of substantial changes have occurred in the Swiss context that are likely to have had effects on residents' NHRA.
There were four natural hazard events that occurred from 2015 to 2021: a century rockfall and subsequent debris flow in summer 2017, a catastrophic avalanche winter 2017/2018, an exceptionally dry period in summer 2018 and a severe flood due to the thunderstorms in summer 2015.
COVID-19, commonly known as the coronavirus pandemic, is a worldwide pandemic of coronavirus disease during 2019-2022.In addition to its prominent effects on physical health in terms of respiratory illness and neurological problems (Mungmunpuntipantip and Wiwanitkit, 2022), the stress that people are subjected to has threatened their mental health due to increased anxiety and fear (Amran, 2022;Ghosh et al., 2020).This might alter their risk awareness of other facets of social life (Auliyah and Santoso, 2022;Dubey et al., 2020;Zhu and Deng, 2020), and further touch NHRA.
The six-year from 2015 to 2021 has also been recorded as the warmest on record (WMO, 2022).Switzerland's average temperature rose 2.1 • C between 1864 and 2021, which is 2-3 times higher than the global average rate of warming (MeteoSwiss, 2022).The intensification of winter precipitation along with a transition from snowfall to rain, may potentially intervene in hydropower production and raise the danger of flooding (IEA, 2022), and thus profoundly affect the surrounding risk to Swiss residents and their related risk awareness.

Survey description
This study utilized panel data from two partly identical surveys conducted in Switzerland, six years apart in 2015 and 2021, and included respondents from diverse cultural and geographical regions indicated by the three Swiss official languages (German, French, and Italian).The questionnaires were designed based on previous research on flood risk awareness and preparedness among homeowners in Zurich (Maidl and Buchecker, 2015), and adapted to the context of hazard risk management in Switzerland.Both surveys covered various natural hazards, but the second one in 2021 included some additional questions on climate change and the 2019 Corona pandemic.The 2015 questionnaire was given by postal mail to a random sample of the Swiss population (N = 10,000) from Swiss Federal Statistical Office (FSO) with 20% of response rate (n = 2137), while the 2021 survey was conducted both via paper-based postal surveys and online surveys to the 2015 respondents and new random sample of 4000 Swiss residents (N = 6137, n = 1643).851 respondents from our 2015 pool returned in 2021, but this study was concentrated on 806 individuals for whom matching records from both 2015 and 2021 were available, allowing for a longitudinal data examination.Respondent's ages vary from 21 to 95 with a mean age of 64, which is 51.7% higher than the average age of the Swiss population (42 years old).With a ratio of 1.04:1, the distribution of male and female respondents is identical to Swiss population.Highly educated respondents covered 57.2% of all the participants and are over-represented compared to the general population, regarded as typical for paper-based postal surveys (Table 1).

Sample distribution: who tends to re-participate?
Approximately one-third of the 2000 participants (n = 806, Table S1) who had responded in 2015 participated in the survey again in 2021 with matched records, and are referred to as "panel participants", forming the primary focus of our study.Comparing the characteristics of panel participants with those who only participated once can help determine the independent variables impacting respondents' willingness to participate, interpret the rationale behind the survey results and facilitate the tracking of them for future surveys to optimize resource use and improve survey efficiency (Table 1).In the subsequent data analysis and Result section, only the panel participants were considered to control for individual heterogeneity and ensure the reliability and robustness of our results.
Significant differences were found between the two samples of participants.The panel participants were significantly older (Mean = 59.20-64.04,p < 0.001).A detailed analysis showed that participants aged 60-89 (45.93-62.70%)who had retired yet still engaged in reading and writing were more likely to participate again (Table S2).Additionally, those with stronger event memory were also more likely to participate in the surveys again (p < 0.001).

Operationalization of key concepts
For the sake of clarity, we introduce the key concepts of the variables used in the study based on the needs of research questions and summarize them in Fig. 1.

Risk awareness
NHRA has been measured in many ways, among which the perception of probabilities in the neighborhood is the most notable measurement technique.Further, it is also often measured as perceived relevance and perceived threats to suffer personal damage financially or physically (Maidl et al., 2021).In corresponding studies, different dimensions have been proven strongly correlated (Miceli et al., 2008).
In this study, NHRA is condensed to four distinct dimensions according to the results of PCA, operationalized through a set of items measured on 4-point Likert scales comprising 27 items: (1) Relevance: general interests in the subject of natural hazards and their worry about related risks, (2) Perceived Probability: perceived probabilities of different hazard types in the respondents' region, (3) Perceived Threat: perceived personal threats that combine the threats of material loss and life danger and (4) Perceived Situational Threat: perceived threats under different situations, can be interpreted as the perceived riskiest environments and situations (at home, in free time, in transport or at work).We chose to keep the four subscales as separate dependent variables rather than integrating them into a single scale in order to reveal more information.Notably, we discovered that some independent variables were significant predictors in one subscale but not the others.

Natural hazard experience
The diverse approaches utilized in assessing past exposure to natural hazards can yield varied outcomes when investigating the association between such experience and NHRA (Demuth, 2018).The methods can include straightforward, single-item measures such as asking participants whether they have experienced a particular hazard, or more intricate multi-item scales that vary in their composition (Demuth et al., 2016;Sharma and Patt, 2012).
We measured various types of natural hazard experiences ranging from media-only knowledge, to experience as volunteer/professional, personal endangerment, or property loss.The utilization of recollections pertaining to natural hazards is employed as a means of delineating the influence of firsthand encounters via a binary-selection inquiry.In the 2021 questionnaire, we highlighted 4 natural hazard events that occurred between the two polls and their impacts to get informed about participants' changes in experience, structured with 4-point Likert scales.

Risk communication
The respondents were initially invited to choose the categories of information channels that are available for them from the "3P" set: probability in the neighborhood, possible effects, and protective measures against natural hazards.The 22 distinct sources of information on natural hazard risks were then presented for the individuals to evaluate their availability with a 6-point Likert scale from "none at all" (coded "1") to "much" (coded "6").Except for conventional weather forecasts and the utilization of mass media, the question set included additional communication means such as one-way communication (information campaigns, notifications from the authorities), dialogic communication, visiting websites, social media, as well as usage of printed material like books and leaflets.The perceived impact of information was also evaluated by respondents with 8 items from different dimensions.

Risk preparedness
What is assessed in research on risk preparedness varies greatly given a lack of coherence.The frequently used methods include the intention to undertake a specific action in the future (Botzen et al., 2009;Samaddar et al., 2014), and adopted behavior or realized protective actions (Lindell and Hwang, 2008;Miceli et al., 2008).The explicit distinction between the two options warrants closer scrutiny and can be combined in the surveys for comparisons (Njome et al., 2010) The diversity of results reflects the various circumstances of risk preparedness and the underlying dynamic nature of factor effects on risk preparedness while maintaining skepticism that stable variables can effectively explain preparedness's influence on awareness.
Operationalization of risk preparedness in this study consisted of 18 items that combined the intention and adopted protective measures.Respondents could answer on a 4-point Likert scale from "definitely no intention" (coded "1") to "definitely yes" (coded "4").An "already implemented" option was included and coded "5" in the 2021 questionnaire (in the 2015 questionnaire it is a 5-point Likert scale with "definitely yes" coded as "5" and "already implemented" externally  coded as "6").All the protection methods were later categorized into four dimensions through PCA: information gathering, social exchange, situational behavior, and construction measures.

Social capital: trust, social integration
The respondents were required to evaluate the statement "The responsible authorities ensure the best possible protection against natural hazards" to measure their trust in authorities.In this study, we also conducted an evaluation of social integration within the context of current social norms and individual participation in social activities.The assessment entailed quantifying active memberships in clubs, participation in community meetings, engagement in public discussions, and other relevant metrics.All the items were structured using a 4-point Likert scale to enable systematic analysis.The resulting sum score was utilized as their measure of social integration level.

Attitudes
We focused on two types of attitudes about dealing with natural hazards that have comparable results across two surveys, "self-efficacy matters" and "fate anyway", each measured with a 4-point Likert scale.

Experiences of COVID-19 and climate change
To investigate the specific subject, the 2021 poll examined whether the Corona pandemic altered their NHRA or impacted them financially.Climate change was also incorporated into the questionnaire to touch on this valuable international issue and broaden the scope of independent variables.

Analysis methods
Firstly, Principal Component Analysis (PCA) was utilized to minimize the multidimensional complexity of the data, enabling the creation of composite indices for further multivariate analysis.An array of statistical tests was conducted to maintain uniformity of variables across temporal and geographical scales, as well as among various population groups.These tests included the t-test, one-way Analysis of Variance (ANOVA), chi-square test, and the Wilcoxon test.To enhance the interpretation and understanding of the distribution and variations in dependent variables, visualizations were presented on Swiss maps.Pearson's correlation analysis was then employed to scrutinize the intrinsic relationships within dependent variables and other independent variables.Subsequently, a panel analysis was conducted to evaluate responses from a cohort of panel participants (n = 806), using the language as a proxy for geographical distributions.To investigate the stability and variations among the independent variables, several linear regression models were constructed and evaluated using different classification techniques.The study also took into account the potential mediation and moderation effects in our holistic exploration of effective risk awareness mechanisms.Mediator variables are used to explore the underlying processes or mechanisms linking independent variables to NHRA, while moderator variables are examined to understand how certain conditions influence the strength or direction of these mechanisms.We deployed four mediation analyses using diverse datasets (panel data with both years, 2015, 2021, and the delta of the two surveys) that allowed us to examine the consistency and directional shifts in mechanisms and reveal the significant mediation effects from risk communication.The mediation analysis was applied via bootstrapping techniques.For each of the 500 bootstrapped samples, standardized indirect effects were determined with a 95% confidence interval.All data analysis was conducted with R 4.1.1 and RStudio 2022.07.1.
In this study, we employed panel data analysis to assess the stability of various independent variables' impacts on the four NHRA dimensions.This technique is a robust statistical method also known as longitudinal or cross-sectional time-series analysis, leveraging data collected from the same subjects at multiple time points.It is particularly valuable in identifying and measuring changes over time and allows for the control of individual heterogeneity.While single measurements can provide insights, their reliability in differentiating between random and consistent impacts of a factor on NHRA is often questionable.Therefore, we incorporated repeated investigations into our methodology, enabling us to identify independent variables that consistently and significantly influence NHRA.To substantiate the robustness of our results, we employed a series of statistical tests within the framework of panel analysis, namely the F test, Hausman test, and Breusch-Pagan test.These tests were instrumental in guiding our selection of a random effects model, which was particularly suited for capturing unobserved variability inherent in our study.This model enabled us to not only scrutinize the direct effects of the variables in surveys but also to account for any underlying, unobserved influences in the random component of the model.

Dependent variable distribution: natural hazard risk awareness (NHRA)
The repeated measurement of respondents' NHRA revealed substantial changes during the period from 2015 to 2021 with the four NHRA dimensions showing consistent directions of change but differing in magnitude.These trends are highlighted in the heatmap (Fig. 2), which displays the average NHRA values across diverse population segments among panel participants over the period.The NHRA dimension of Relevance, representing general interests and concerns, consistently showed the highest values compared to the other three dimensions across time.
In terms of temporal variations, the NHRA dimensions in 2021 have, at a minimum, preserved the levels observed in 2015.The NHRA dimensions of Relevance (2.77-2.89,p < 0.001) and Perceived Threat (1.85-2.28,p < 0.001) increased significantly, while the other two NHRA dimensions preserved the levels observed in 2015.The largest increase could be observed among the most elderly cohort (Age≥90,1.37-2.20)and Italian speakers (1.85-2.92),which deserves extra attention to identify driving factors and to tailor risk management strategies accordingly.

Relationships among NHRA and independent variables
All the dependent variables appeared to be positively correlated with each other but their relationships with independent variables were found to be specific (Fig. 3).Relevance, the broadest NHRA dimension, exhibited the strongest association with the independent variables, which suggested that it was the most susceptible to influence, followed by the dimension of Perceived Threat.Specifically, Relevance appeared to be most strongly interrelated to perceived information impact, i.e., how respondents reported that the information influenced them, the influence including whether they learned or were motivated to learn more about natural hazards, became more aware or more concerned about relevant risk, etc. Perceived Probability was mainly influenced by event memory; the more they remember about related local events, the higher they perceived probability of future events.Perceived Threat, on the other hand, was influenced by their fate-belief, with higher values indicating that respondents believe the risk of natural hazards is inevitable, thereby elevating their perceived threat to life and valuables.These findings highlight the key independent variables influencing NHRA, which we will explore further in subsequent panel analysis.
Our analysis revealed significant interrelationships among the independent variables, with a predominance of strong positive correlations, although negative associations were also noted.A particularly robust positive correlation was identified between social integration and two preparedness variables: social exchange and construction measures, as well as between social integration and two attitude variables: selfefficacy and attitude towards fate.The level of social integration could X.Sun et al.S3.
possibly impact their intention or adoption of relevant preparedness behavior, while also being impacted by residents' attitudes towards how to deal with risks.Accordingly and rather unexpectedly, a substantial positive correlation was observed between the two attitude variables themselves (p < 0.001).This suggested a noteworthy interplay between an individual's belief in their own capabilities and their acceptance of fate, which deserved further investigation.The Supplementary Fig. S1, showcasing a heatmap of the average values of various independent variables among panel participants from 2015 to 2021, highlighted an escalation in these two attitude variables over time.From a sociodemographic standpoint, it was discerned that males (p < 0.01), individuals from higher age groups (p < 0.05) and larger households (p < 0.01) tended to exhibit higher levels of social integration.Additionally, larger household sizes correlated with reduced perceived information impact, as the internal dynamics prevalent in larger households diminish the typical information dissemination effects in the context of natural hazards.

Panel analysis: influencing independent variables on NHRA
In the exploration of the relationships between independent variables and NHRA, our panel analysis revealed that NHRA was mainly influenced by perceived information impact, well-informed level, and event memory levels across various NHRA dimensions (Table 2b).A comparative evaluation of different panel analysis models underscored the superiority of the random-effect model over the fixed-effect and pooled models, as evidenced by the significance outcomes in the Hausman test and BP test (Table 2a).Notably, all the four NHRA dimensions appeared to be significantly and sustainably affected by perceived information impact (Info_impact, β = 0.104-0.289,p < 0.01), suggesting that improving the quality and reach of risk communication can substantially enhance NHRA.The three out of four NHRA dimensions, respectively excluding Relevance and Perceived Threat, were significantly influenced by participants' event memory and their levels of being well-informed.The extent of memory retention as direct experience, as well as the availability and accessibility of information as indirect experience, could largely impact NHRA generally and should be highlighted in risk management.
Furthermore, NHRA dimensions were broadly linked to risk preparedness, but had limited connection to attitudes and social capital.The NHRA dimension of Relevance was found to be positively associated with preparedness of information behavior (β = 0.109, p < 0.001), while Perceived Probability correlated with preparedness of construction measures (β = 0.053, p < 0.05).The dimension of Perceived Threat was found to be positively influenced by fate belief (β = 0.123, p < 0.001) and social integration (β = 0.018, p < 0.001).Contrary to expectations, however, self-efficacy did not demonstrate a significant impact on any NHRA dimension.These findings underscored the complex interplay between information dissemination, psychosocial beliefs, and social connectivity in shaping individual's NHRA.
Most of the independent variables that demonstrated a significant and consistent association with NHRA in the panel analysis also significantly influenced the changes in NHRA, thus reinforcing the establishment of a causal link between these variables and NHRA.To further explore the relationship between the independent variables and changes in NHRA, we computed the differences between the 2015 and 2021 data for the four NHRA dimensions and all independent variables, followed

Table 2
The comparison among three panel analysis models (a) and results of Random-effect model (b, n = 806 c 2).Completed version of results of this panel analysis model and a model using delta data between 2015 and 2021 can be found in Table S4 and Table S5 by a linear regression model to valid their connections.The results (Table S5) displayed high consistency with the findings from the panel analysis, with the majority of independent variables that were significantly associated with NHRA in the panel analysis maintaining their influence on the changes of NHRA.Notably, changes in the perceived information impact substantially affected changes in three NHRA dimensions: Relevance, Perceived Probability, and Perceived Threat.Similar effects were observed for independent variables such as event memory, attitude towards public measures, and fate-beliefs.Surprisingly, however, changes in the level of well-informed (a consistent and positive predictor of NHRA) did not significantly affect the changes in any of the NHRA dimensions.Although the overall results remain robust, the differences in outcomes between the two regressions suggest that while these independent variables influence NHRA status, they are not always the cause of changes in NHRA.

Influence of crisis on NHRA: COVID-19 and climate change
The crisis context of natural hazards, such as Corona pandemic and climate change, impacted the world in a broad way inside the multi-risk framework during the study period.The physical and psychological effects from individuals' related experiences could potentially influence their NHRA.Exploring the influence of these experiences can assist in understanding their current role for both NHRA and the change of NHRA in extensive contexts.In recognition of this, the 2021 survey introduced specific questions to measure the impacts from the Corona pandemic on individuals and concerns about climate change.
The regression analysis highlighted that NHRA can be subject to potential influences from both the Corona pandemic and climate change experiences, however, these influences appeared to be limited to NHRA dimensions.Table 3a verified that Perceived Probability seemed to inversely correlate with perceived economic impacts from Corona (p < 0.05).The NHRA dimensions of Relevance (β = 0.110, p < 0.001), Perceived Threat (β = 0.094, p < 0.001) and Perceived Situational Threat (β = 0.067, p < 0.01) were found to be associated with reported increased concern about natural hazards due to Corona.The complementarity of these two results demonstrates the breadth and pervasiveness of Corona's influence on NHRA.
Regarding the changes of NHRA observed between 2015 and 2021 (Table 3b), the level of concern respondents expressed towards Corona still contributed to changes in both Relevance (β = 0.055, p < 0.01) and Perceived Threat (β = 0.072, p < 0.01).However, the economic impacts from the Corona pandemic did not significantly influence NHRA.Also, respondents' beliefs regarding the role of climate change on natural hazards appeared to have an influence on their NHRA.The NHRA dimensions of Relevance and Perceived Situational Threat were found to be positively connected to the recognition of climate change's challenge to the Swiss natural hazard management framework (p < 0.05).Astonishingly, the impacts from climate change beliefs did not extend to more dimensions of NHRA (p > .05), in particular for Perceived Probability.

Risk communication: mediator from preparedness to awareness
To further explore the consistency of mediation mechanisms on NHRA with risk communication (perceived information impact) as a mediator, we investigated the interactions between the variables of interest in a final step.
Perceived information impact, representing the respondents' selfreported impacts of information on their NHRA, appeared to significantly bridge the gap between respondents' risk preparedness and risk awareness.The analysis revealed a consistent pattern across a spectrum of mediation mechanisms, encompassing all the four NHRA dimensions and four categories of risk preparedness, with the specific risk communication indicator: perceived information impact.In Fig. 4 we have chosen to show the mediating mechanism between the construction measures and the NHRA dimension of Relevance, as this result is representative and consistent in the 2021 data and the delta data between 2015 and 2021.The mediation analysis implied that while intended preparedness behavior by authorities or oneself might not directly influence NHRA, the perceived impact of relevant information from risk communication could facilitate it.For instance, residents' preparedness for construction measures, such as conducting structural protection measures on buildings or properties, creating an emergency concept for tenants, etc., could enhance an information-based understanding of natural hazards (β = 0.261, p < 0.001), which, in turn, heightened their related Relevance (β = 0.445, p < 0.001).Despite varying mediation results from different preparedness categories and risk awareness dimensions across datasets, a stable trend and significant robustness were maintained.
The analysis of delta data and moderation effects could help understand the changes in NHRA.We calculated the delta between 2015 and 2021 for all the variables that were included in the models, reran them and confirmed the mechanism's consistency.While the changes in construction preparedness did not have a direct correlative effect on the changes in the NHRA dimension of Relevance, this relationship was indirectly mediated through the impact of information dissemination.Furthermore, moderation effects could substantially shape this mediating process.For example, varying levels of trust in authorities could modulate how the changes in construction measures influence the change of individual's engagement with related information.This suggested that higher trust levels could enhance the positive promotion from the changes of construction measures to the changes of perceived information impact.Other independent variables such as fate-belief could also strengthen this mediation mechanism while event memory may hinder the mechanism through moderation effects.Recognizing and leveraging the interplay between these variables could be crucial for the enhancement of risk management practices.

Table 3
Regression estimates of the perceived Corona and climate change impacts a) on 2021 NHRA, b) on difference of NHRA between 2015 and 2021 data (n = 806), controlling for effects of socio-demographic variables.The independent variables refer to how much respondents agreed with some specific statements (Fig. 1).Completed version of results can be found in Table S6.

Discussion
Enhancing risk awareness is important for effective natural hazard management, with the potential of significantly increase residents' risk preparedness.Despite considerable efforts in risk communication for raising residents' risk awareness, the effectiveness and the factors that influence this effectiveness remain insufficiently explored.The literature suggests that the links between risk communication, risk awareness and risk preparedness are multifaceted and potentially non-linear.An important aspect of better understanding these relationships and dynamics is to examine whether these factors change over time.Given the pioneering role of natural hazard risk awareness (NHRA) in the risk management inside the multi-risk framework, this study aims to reveal its change among Swiss residents during a period marked by significant natural hazard events, the COVID-19 pandemic, and intensified public discourse on climate change drawing on analysis of a balanced nationwide panel dataset with 1612 (n = 806*2) respondent-year observations.To examine and compare the effects of potential independent variables on NHRA, multiple methods, including random-effect panel model, generalized linear models, and mediation analysis were applied.Our findings revealed that the levels of the four NHRA dimensions, Relevance, Perceived Probability, Perceived Threat and Perceived Situational Threat, have notably increased in Switzerland in 2021 compared to 2015.Among the four dimensions, Perceived Threat went through the most notable changes while Perceived Probability did not vary significantly.Relevance exhibited significantly higher values across spatiotemporal scales.It appeared to be also strongly correlated with independent variables and could therefore be a promising starting point for risk management and education efforts.From the perspective of independent variables, perceived information impact was found to have substantial effects both directly as an independent variable impacting risk awareness and indirectly as a mediator between risk preparedness and risk awareness.Trust in authority and fate attitudes turned out to potentially magnify the benefits of mediation.Further potential stable positive independent variables include personal event memory and individual informed levels.Climate change and Corona pandemic were found to be additional important factors affecting NHRA during 2015-2021, but their impacts were limited to certain NHRA dimensions.
This section discusses the main findings in more depth in the context of the current study to validate the insights of the study and to disclose recommendations for future natural hazard risk management.Our findings underscore the importance of integrated risk communication strategies that address multiple hazards simultaneously.For instance, the COVID-19 pandemic likely influenced public vulnerability and NHRA, highlighting the interconnected nature of risks.By framing NHRA within the multi-risk context, this study provides a more comprehensive understanding of risk awareness and offers valuable insights for designing effective awareness-raising strategies.

Main differences: risk awareness levels among all variations
The changes of NHRA across time are worth attention.Our results confirmed Assumption 1: the four dimensions of NHRA grew in Switzerland during the period between 2015 and 2021.Among the four dimensions, Perceived Threat went through the most notable changes while Perceived Probability did not vary significantly.Perceived Threat and Relevance are more emotionally driven, reflecting individuals' subjective feelings of personal vulnerability and the anticipated impacts of risks, while Perceived Probability relies on cognitive mechanisms.Accordingly, situational changes seem to mainly affect emotions and thus rather result in a higher fear to risks (Perceived Threat) than Perceived Probability (Qing et al., 2021).So far, further empirical studies on how risk awareness changes over time in different dimensions that might support this assumption are unavailable and therefore, this field requires further investigation.Considering that the huge increase of Perceived Threat took place parallel to an increase of belief in fate, there will be an ambivalent interpretation of the findings.It highlights the potential success of recent risk management in heightening risk awareness that might enhance social capacity building, but at the price of a decreased sense of security to life that might in turn reduce residents' well-being.
The largest increase of NHRA was observed among Italian speakers (Figs. 2, 1.85-2.92),who are concentrated in the Ticino region.Spatial differences in NHRA appear to arise from the frequency and intensity of natural hazards, from severe events like the COVID-19 pandemic, and from cultural variations in how residents approach risk management policies.This is particularly evident in mountainous regions, where events have a pronounced impact on risk awareness (Schneiderbauer et al., 2021;Slinger et al., 2007).The challenging terrain in the Italian-speaking regions led to events like Bondo rockfalls and excessive snowfall, promoting protective behaviors against prevalent natural hazards and intensified awareness among residents (Pielmeier et al., 2023).Additionally, Ticino's higher COVID-19 infection and mortality rates up to May 2021 (when our repeated measurement was conducted) could have undermined residents' general resilience to natural hazards and increased their vulnerability, both physically and psychologically (FOPH, 2022;Repubblica e Cantone Ticino, 2023).Ticino also stands out with a much lower score of individualism, emphasizing collectivist values compared to other regions (Kaasa et al., 2014).Residents tend to be more in favor of government intervention, which may have contributed to their stricter adherence to policies and higher NHRA (Mazzonna, 2020).The evidence from Italian speakers highlights the Fig. 4. Results of mediation and moderation analysis (n = 806*2), 2021 (left) and the difference between 2015 and 2021 data (right).Background colors represent variable categories: Light blue for independent variables, light red for dependent variables, dark red for mediator (perceived information impact, risk communication), and dark blue for moderation variables (event memory, social capital, attitudes, socio-demographic independent variables).The estimates are extracted from linear regression models solely including related variables.
X. Sun et al. complex interaction between hazard, exposure, and vulnerability in shaping NHRA, emphasizing the need for tailored risk management and resilience strategies within the multirisk framework.The strategies should be adapted to the specific spatial and cultural contexts of regions, leveraging the community's preference for collective action and support for government intervention, rather than applying a uniform approach solely based on general trends.

Reliable independent variables: what influences NHRA
Three independent variables, self-reported memory of local hazard events (Event memory), the availability and accessibility of information channels (Well-informed), and perceived impact of used information (Perceived information impact), consistently and robustly influenced NHRA.
Interpreted as direct experiences in many studies, event memory has been shown to exert a major influence on raising respondents' NHRA, as verified in this study and confirming our Assumption 2. The causal pathways are more complicated than the straight relationship between event experience and risk awareness (Felgentreff, C, 2003;Terpstra, 2011).Personal experience's intensity and relevance determine the time period within which individuals are able to recollect prior events and are motivated to build up a higher level of NHRA, which is defined as "window of opportunity" (Wachinger et al., 2013).
While the firsthand experience of natural hazard events can effectively raise risk awareness, the majority of individuals are unlikely to encounter a catastrophic event during their lifetime and depend on references from indirect experience: risk communication (Becker et al., 2017).From our results, two independent variables regarding risk communication: well-informed level and perceived information impact, emerged as two best NHRA change determinants.Among those, perceived information impact was found to function as both a factor and a mediator between risk preparedness and risk awareness.The well-informed level represents forms of information that are most salient or accessible to an individual, and perceived information impact refers to how an individual interprets the effectiveness of the available information.Such interpretation involves incorporating information about a threat; the ability to generate, comprehend, and apply information is thus crucial in the emergence and change of individual risk awareness (Ferrer and Klein, 2015;Reyna et al., 2009).For example, empirical research indicates that individuals with high numeracy skills are less susceptible to bias and less likely to incorporate irrelevant information into assessments but significantly raise the application and impact of important information (Peters et al., 2006).
To enhance perceived information impact across residents, we can take inspiration from the Social Impact Theory in psychology.According to this theory, social impact, denoted as I, is determined by the product of three independent variables: the strength (S) or power of the source, the immediacy (i) or proximity of the source, and the number (N) of sources (Latané, 1981): From this perspective, increasing the number of information sources or exposure frequency can enhance the impact of information, particularly when the sources are highly credible and immediate.Despite the lack of substantial evidence for Wachinger's claim about media's ability to shape or even determine NHRA (Wachinger et al., 2013), most studies agree that information provided by the mass media (print, broadcast, digital, etc.) does have an effect, but only if respondents lack the direct experiences and are unable to independently verify claims about risks or benefits (Siegrist and Gutscher, 2006).
Regarding Assumption 2, our study also indicates that climate change and COVID-19 positively impacted NHRA, as indicated by the multi-risk framework, although not to the extent we initially hypothesized.Recognition of climate change's impacts on natural hazards significantly enhances NHRA's dimensions of Relevance and Perceived Situational Threat.Climate change can exacerbate natural hazards and impact human livelihoods (Anguelovski and Carmin, 2011), leading to the common assumption of a corresponding increase in NHRA (Lujala et al., 2015).Our findings, however, suggest that the impact of perceived climate change is confined to the basic dimension of Relevance about natural hazards and does not extend to higher dimensions like Perceived Probability and Perceived Threat.Distant cries from the earth have limited impacts on specific risk awareness, which urgently calls for expanding the reach of risk information and enhancing awareness among potential risk groups (Maidl and Buchecker, 2015).Regarding the Corona pandemic, the NHRA dimensions of Relevance and Perceived Threat are significantly and positively associated with the perceived connection between personal concern about natural hazard and their experience of Corona pandemic.Conversely, Perceived Probability inversely correlates with reported economic impact from the pandemic.The former result can be explained by an increased general anxiety stimulated by the Corona pandemic experience (Schwinger et al., 2020), which is also reflected by a high correlation between perceived Corona's economic impact level and people's fate beliefs.The later result is connected with the fact that, among numerous risk sources in the multi-risk environment, natural hazards are seldom prioritized in the risk competition.The desire to protect daily livelihoods generally takes precedence over awareness of other risks (Lavigne et al., 2008).Individuals are therefore less likely to focus on unpredictable natural hazard costs when faced with significant economic costs, regardless of their source, which would result in a decline in their NHRA.

Impacting mechanism: behavior-cognition feedback loop
While the relationship between risk awareness and risk preparedness remains controversial in the literature (Haynes et al., 2008;Wachinger et al., 2013), our results suggest that this relationship is characterized by a feedback loop.Residents' preparedness levels tend to enhance their information-based understanding of natural hazards and thus heighten their related NHRA, which in turn increases their risk preparedness.Traditional models often depict this relationship in an oversimplified and static way, based on empirical research corroborating that NHRA can motivate individuals to proactively avoid risk (Ban et al., 2019;Martin et al., 2009).Evidence of the reversed relationship from behavior to cognition, however, also exists.As the Theory of Planned Behavior (TPB) described, people's cognition can be supported or strengthened by a variety of actions given the reciprocal nature of psychological independent variables and behavior (Bamberg and Schmidt, 1998).
This mechanism has been substantiated in various fields.For example, the universal bus pass program showed that frequent bus use led to more accurate and positive perceptions of public transportation (Heath and Gifford, 2002).Similarly, a study on adolescent condom use found that individual's attitudes became more positive as the frequency of use increased (Ajzen et al., 1996).More recently, behavior-cognition feedback loops have also been observed in reactions to the COVID-19 pandemic, with citizens adjusting their awareness based on government responses, ultimately reshaping public reactions to a new level (Vigoda-Gadot et al., 2023).In the context of risk awareness, some studies suggest that individuals may improve their understanding of risks by preparing appropriately and accelerating the pace of post-disaster recovery by reducing unnecessary fears about the likelihood and consequences of future events (Ajzen, 2015;Qing et al., 2021).In this way, the adoption of risk preparedness can result in subsequent, appropriate reductions in risk awareness (Brewer et al., 2004).
In this study, behavior has been found to feed back into awareness in a positive way and shed light on risk management.As stated in Assumption 3, mediated by risk communication, risk preparedness could have positive influences on risk awareness.The feedback loops between behavior and cognition subtly influence each other and could be amplified by perceived information impact.According to studies on media influence, individuals select elements from media narratives and use their existing frames of reference to construct understanding and meaning.Therefore, people favor confirming their pre-existing beliefs when reading, watching or hearing media content, so that learning remains restricted (Breakwell, 2007).Promoting people's level of preparedness may break this barrier and help elevate their awareness levels by way of new experiences.Higher risk awareness can then be built upon the foundation of enhanced risk preparedness, achieving continuous improvement in risk management at the individual level (Chou et al., 2015).This dynamic interaction between risk preparedness and risk awareness, mediated by effective risk communication, has significant implications for how we understand and approach individual risk management strategies.By instructing residents to adopt a "Begin Doing Before Thinking" (BDBT) or "Do First & Think Later" approach, we could inspire residents to prepare for natural hazards through various communication channels and to improve risk awareness via experiential learning.The application of the Swiss Confederation's Common Information Platform for Natural Hazards (GIN) provides a good example for this virtuous cycle.This platform offers reliable hazard data in Switzerland (FOEN, F. O. for the E, 2021), and residents' habitual access to it is expected to be promoted by economic subsidy programs or within community training programs.Such repeated exposure could enhance users' risk awareness and in turn reinforce social resilience to future events through a better understanding of natural hazards.This program can thus amplify perceived information impact and heighten NHRA through subconscious habit formation over time.

Limitations and future research
The novel longitudinal research design implicated challenges that brought about certain limitations of the study.While our study provides valuable insights into the factors influencing NHRA, addressing some limitations in future research will further enhance our understanding and ability to promote effective natural hazard risk management.
The 2021 survey expanded on the previous 2015 survey, but some changes were made to enhance the questionnaire and led to incompatible data and limited comparisons.To improve the robustness and informativeness of panel data, future research should maintain consistency in the questionnaire as much as possible, thereby enhancing data comparability across different survey waves.
Our sampling method, while random, had a significant education level and age distribution bias compared to the Swiss population.This over-representation presents a limitation of our study and could affect the generalizability of our findings.Highly educated individuals may have greater access to risk information and resources, potentially leading to higher NHRA levels.Similarly, older individuals might have more experience with natural hazards, influencing their NHRA.To limit the impact of this bias, we conducted our analysis by always controlling for the effects of socio-demographic factors and made these effects transparent to readers.Future research should aim to recruit a more representative sample or oversample specific groups to simulate the population and validate these results.
The six-year interval was not originally planned but was a byproduct of the difficulties of conducting surveys during the COVID-19 pandemic.Continuous measurements, despite their cost and time-intensive nature, offer a more precise picture of trends and timing of changes.Future research should aim for more frequent data collection with shorter intervals to more accurately evaluate the effectiveness of risk communication strategies and better establish causal relationships between risk communication, preparedness, and NHRA.
The reliance on self-reported data for NHRA dimensions introduces the possibility of response bias.Future studies could benefit from incorporating objective measures of NHRA.For example, eye-tracking technology can be employed to study how individuals visually process natural hazard risk-related information.This method can reveal which parts of a warning or information graphic capture attention and for how long, offering insights into their immediate cognitive engagement with risk information and reflecting NHRA levels.
Additionally, the behavior-cognition feedback loop within the context of natural hazard risks and influence factors for the interpretation of exposed information presents an area requiring further exploration.Empirical validation of this feedback loop would be valuable for policy makers and enlightening for future research, potentially informing more effective risk communication strategies.Future research should continue to explore the multi-risk framework to better capture the dynamic interactions between various risk domains and their impact on public awareness and resilience.Integrative approaches, including objective measures and more frequent data collection, are essential to capture temporal dynamics and improve the robustness of findings.This will ultimately contribute to more resilient and adaptive risk management practices in the face of complex and interconnected hazards.

Conclusion
This study contributes to the understanding of natural hazard risk awareness (NHRA) by expanding previous research in terms of concept, data, independent variables, and mechanism.We investigated four dimensions of NHRA, Relevance, Perceived Probability, Perceived Threat and Perceived Situational Threat using a comprehensive and novel method that allowed to reveal the dynamics of risk awareness in a defined time period.Furthermore, this study provided the first evidence suggesting a positive interplay between risk preparedness and risk awareness based on a behavior-cognition feedback loop.Our study also adds to the growing body of literature on multi-risk research by emphasizing the interconnected nature of various hazards and their combined impact on NHRA and management.
Besides offering valuable insights for panel data research on NHRA, our study also provides key implications for the design of effective awareness-raising strategies and integrated risk management for policy makers.The identified consistent influencing factors of risk awareness should be considered and utilized in risk communication to enhance society's capacity to prepare for, respond to, and recover from the adverse effects of natural hazards.The effectiveness of event memory, influenced by the event's intensity and relevance as well as the authorities' ability to sustain these memories (Dow and Cutter, 2000), points to the significance of leveraging post-disaster periods.These "windows of opportunity" are critical for introducing and promoting innovative mitigation strategies in time.To bridge the gap in experiential learning due to the absence of direct hazard experiences, indirect communication should be designed resembling direct personal experience to increase psychological intimacy with information (Shaw et al., 2004), such as simulating floods or retelling hazard events.Through recalling or imitating personal experience, perceived information impact is expected to be enhanced, together with the connection between the individuals and the communicated risks.The observation of limited impact from climate change on NHRA reveals the complex, dual nature of climate change in the context of natural hazard risk communication.While it remains valuable to integrate climate change into communication, we should ensure a balanced manner, for example, without overwhelming or detracting from the primary focus on immediate risk mitigation and preparedness actions, to avoid counterproductive outcomes from overstating.
The relationship between people's education, their self-efficacy and their NHRA reveals an interesting dynamic.Education appeared to moderately reduce fear of natural hazard risks, but higher education levels were linked to increased interest and concern about natural hazards.This could be attributed to the link between education and selfconfidence, which when excessive, could lead to indifference to common risks (Bhandari and Deaves, 2006).Notably, self-efficacy is negatively correlated with the well-informed level in this study.This finding complements a study on the Chinese Wenchuan earthquake, which found that the more individuals perceived they understood a risk, the more they relied on personal judgment over authority advice (Yu et al., 2020).Such tendencies align with the concept of "critical trust", as introduced by Poortinga and Pidgeon, which encourages practical reliance on authorities coupled with healthy skepticism (Poortinga and Pidgeon, 2003).This emergent pattern of critical but non-destructive distrust among the public underscores a shift towards a more engaged and competent participatory democracy, where citizens are both critical and involved.Risk management should thus consider the potential for increased public participation and critical trust, and ensure that communication and engagement efforts are tailored to meet the expectations and capabilities of an increasingly informed and engaged public.By understanding the ways how people recognize and react to risks, based on their physical environment and cultural context, authorities can generate more effective, localized strategies for risk preparedness and mitigation.Tailoring risk management to align with community values and preferences, particularly in regions with a strong collective value like Ticino, could enhance adherence to safety regulations and improve overall resilience, leading to more robust and adaptive risk management frameworks.
Although a well-designed risk education program can be beneficial, it may only attract individuals who are already engaged in the subject matter, thus the community requires innovative ways to engage new audiences in this field.The core lies in directly linking their daily lives to the potential occurrences of natural hazard events to raise their interests.Apart from enhancing targeted information exposure, one promising approach can be a simulation game that enables users to simulate varying intensities of natural hazards, observe potential damage to their homes and daily lives, and explore mitigation strategies.As a complement to traditional strategies, advanced features such as illustrating the tangible impacts of global warming scenarios on the frequency and severity of natural hazards can bridge the knowledge gap on climate change awareness.This interactive tool would allow players to visualize the pervasive impacts of natural disasters, especially in the context of the escalating climate crisis, thus subtly implanting the severity and likelihood of these natural hazard events against the background of climate change into their cognition.

)
Assumption 1: NHRA in Switzerland increased during 2015-2021, a period marked by extraordinary climate change and the global COVID-19 pandemic.(2) Assumption 2: Risk communication, personal event experience, as well as recognition and experience related to climate change and major social events like the COVID-19 pandemic have significant long-term impacts on NHRA.(3) Assumption 3: Risk communication serves as a critical mediating
X.Sun et al.

Fig. 2 .
Fig. 2. Mean value heatmap of the four NHRA dimensions for socio-demographic independent variables and groups among panel participants (n = 806*2).Colorcoded squares represent NHRA values: blue for low, yellow for medium, red for high.Columns denote socio-demographic independent variables and groups.Numbers below the plot indicate survey years: 1 for 2015, 2 for 2021.

Fig. 3 .
Fig. 3. Dependent variable and independent variables' relationships among panel participants (n = 806*2): Correlation and GLM estimates.Squares are sized and colored to indicate correlation strength and direction (blue = positive, red = negative).GLM estimates presented as purple (positive) and green (negative) lines, with thickness indicating estimate magnitude.Significant relationships are denoted with solid lines.Model details can be found in TableS3.

Table 1
Summary demographics of Swiss population (Left, Year = 2015 and 2021) and the 2015 survey sample (Right, n = 2137).The 2015 survey sample is grouped by "Participation times", 1 = participated only in the 2015 survey, 2 = participated in both the 2015 and 2021 surveys (panel participants, our study subjects).The p values and significance levels (sig.)display the t-test results comparing the two groups of participants to identify the characteristics of individuals who are more likely to re-participate.
) Comparisons of Random-effect model, Fixed-effect model and Pooled model.