Introduction
Disasters from natural hazards increasingly threaten lives, economies, and ecosystems globally, affecting all regions. From hurricanes along the Atlantic coastlines to earthquakes in the Pacific Ring of Fire, and flooding in South Asia to wildfires in Australia, these events transcend geographical and political boundaries.Reference Adger, Hughes and Folke1, Reference Burleson2 Climate change amplifies these hazards, intensifying their frequency and severity, posing significant challenges worldwide.Reference Kelman, Gaillard and Mercer3, Reference Companion and Chaiken4 Each disaster highlights the urgent need for resilience and preparedness, driving societies and governments to prioritize public safety and sustainable development.Reference Dept5 Building resilience has become a key goal for disaster-prone countries, with global organizations like the United Nations and the World Bank advocating preparedness to protect lives, economies, and community welfare.Reference Aitsi-Selmi, Egawa and Sasaki6, Reference Costella and Ivaschenko7
While disaster preparedness is essential, vulnerable groups—such as the elderly, individuals with disabilities, and economically disadvantaged communities—face universal barriers to accessing resources and information, including socioeconomic challenges, physical limitations, and weak social networks.Reference Sapirstein8–Reference Kapucu and Sadiq12 Research highlights how factors like financial constraints, restricted mobility, and lack of community support hinder preparedness, especially for socially vulnerable populations.Reference Nishiyama and Glauberman13–Reference Onuma, Shin and Managi15 Additionally, variables like past disaster experiences, trust in government, and community cohesion influence readiness behaviors.Reference Lee16 However, there is a pressing need for more research on tailoring strategies to the unique cultural contexts of vulnerable groups, as current literature has paid limited attention to how deeply rooted cultural practices influence preparedness. This leaves a critical gap in understanding the mechanisms through which culture shapes disaster-related behaviors.Reference Pastrana-Huguet, Casado-Claro and Gavari-Starkie17
In Japan, “bosai culture”—a set of community norms, rituals, and educational programs centered on disaster prevention—has been disseminated nationwide through school curricula, local government campaigns, and neighborhood disaster-prevention associations. These grassroots organizations, present in nearly every municipality, organize regular drills and maintain disaster stockpiles, embedding preparedness into daily life. Yet, the specific pathways by which bosai culture translates into individual and household actions remain underexplored.Reference Pastrana-Huguet, Casado-Claro and Gavari-Starkie17 Japan’s unique combination of frequent natural disasters and a deeply ingrained disaster awareness culture has fostered one of the world’s most advanced disaster management systems, exemplified by the “Disaster Prevention 4.0” framework, which empowers communities through education and active participation.Reference Nakai and Nakano18, Reference Lee19 Despite this progress, socially vulnerable populations continue to face significant challenges due to limited resources, physical capacities, or support networks, highlighting critical gaps in inclusivity within Japan’s disaster resilience framework.Reference Adu-Gyamfi and Shaw14, Reference Bhandari and Takahashi20 This study explores these gaps, offering actionable strategies to improve preparedness for vulnerable groups in Japan and globally.Reference Okuyama and Inaba21
This study aimed to examine the factors that shape disaster awareness and preparedness behaviors in Japan, with a particular focus on sociodemographic and behavioral characteristics influencing evacuation readiness and risk perception.
Materials and Methods
Settings and Participants
This study used data from the 2023 Japan COVID-19 and Society Internet Survey (JACSIS), a nationally representative project. Earlier JACSIS studies focused on vaccine passport perceptions (2021) and the effects of social restrictions on health outcomes.Reference Tanaka, Tabuchi and Katanoda22, Reference Hara, Asami and Sakano23
This research shifts focus to disaster preparedness and evacuation awareness, examining how vulnerabilities like nonregular employment, economic instability, and caregiving affect these areas. Participants provided informed consent online, following ethical and rigorous data collection practices common in large-scale studies.
Disaster Response Policy in Japan
Japan’s disaster response policy, shaped by experience with major natural disasters, integrates early warning systems, legal frameworks, and community support structures to strengthen national resilience and response capabilities.Reference Kawane, Ozaki and Suresh24, Reference Nakai and Nakano25 Community-based disaster prevention organizations play a key role in enhancing local preparedness, particularly in resource-limited areas.Reference Ochiai, Asada and Kobayashi26
However, significant disparities remain in preparedness and response across regions. Rural and remote areas struggle with information dissemination and policy implementation, creating gaps in resources and training. Vulnerable groups, including the elderly and foreign residents, face barriers to accessing essential preparedness information.Reference Adu-Gyamfi and Shaw14, Reference Prabhakar, Iwata and Shaw27 Strengthening community-centered approaches and local collaboration is crucial to ensuring inclusivity and improving resilience nationwide.Reference Cho28
Outcome Variable
The study analyzed disaster preparedness behaviors among 28,481 participants, excluding 4519 responses due to incompleteness or low quality. Six binary variables captured preparedness actions, such as securing furniture, stockpiling supplies, preparing emergency kits, confirming family members’ safety, and identifying evacuation routes inside and outside the home (coded as 1 for “yes” and 0 for “no”).
Tetrachoric correlations were computed to construct a correlation matrix suitable for exploratory factor analysis, as the selected variables were binary and assumed to reflect underlying continuous traits. This method estimates the association between dichotomous items more accurately than Pearson correlations in such contexts. To assess the adequacy of the data, we conducted the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test of sphericity. The KMO test indicated sufficient sampling adequacy, while Bartlett’s test confirmed significant inter-item correlations, validating the use of factor analysis on these variables. Factors were extracted using parallel analysis and a Scree plot, followed by Weighted Least Squares (WLS) estimation with oblimin rotation. Model fit was assessed using Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residuals (SRMR). Variables with factor loadings ≥0.4 were considered significant contributors to the factor structure.
Additionally, we generated histograms (Supplementary Material 1) for WLS1 and WLS2 to visualize the distribution of these awareness levels across respondents. The histograms (Figure S1) reveal a notable clustering at both extremes, with a particularly high concentration at the left end, indicating a significant portion of respondents with low scores in both evacuation and disaster preparedness awareness. This bimodal distribution suggests distinct groups within the population, with many individuals either highly prepared or significantly lacking in preparedness actions.
Exposure Variable
This study analyzed how sociodemographic and health factors influence disaster awareness, emphasizing social vulnerability. Key variables included gender, age, and household size, which affect preparedness, with older adults and larger households facing unique challenges.Reference Fatemi, Ardalan and Aguirre29 Marital status and the presence of vulnerable members, such as children, elderly, or disabled individuals, further impact preparedness.Reference Cong, Feng and Chen30
Economic stability was captured through occupation and income, as nonregular workers and economically disadvantaged groups often lack resources for preparedness.Reference Howard, Agllias and Bevis31 Education and insurance were included due to their links to risk perception and financial readiness.Reference Boon32 Caregiving responsibilities and health factors, such as healthcare visits and health status, also play a role.Reference Fatemi, Ardalan and Aguirre33 Behavioral factors like smoking and alcohol use, along with past disaster experience, significantly influence preparedness.Reference Azmi, Hashim and Zamhury34, Reference Sun, Daniels and Spencer35
Geographical and social elements, such as region and community affinity, were included because high-risk areas and strong community ties often lead to greater readiness.Reference Jang36 For example, DeYoung et al. highlighted the role of age, income, and community ties in disaster preparedness among Cambodian and Laotian Americans.Reference DeYoung, Lewis and Seponski37
Variables were categorized as follows: gender (male/female), age (years), household size (members), marital status (married/unmarried), and vulnerable household members (yes/no). Occupation types ranged from management to nonlaborers. Education levels spanned secondary to graduate school. Insurance types included National Health Insurance, Employee’s Health Insurance, and public assistance. Health indicators included visit frequency, self-reported health, and checkup participation. Alcohol and smoking status were categorized as drinker/nondrinker and smoker/nonsmoker. Income was classified as low, moderate, high, or unknown, while region and community affinity captured geographical and emotional ties.
Data Analysis
This analysis began by examining 2 factor scores, WLS1 and WLS2, which served as the dependent variables. Initial examination revealed skewed distributions for both scores, indicating a departure from normality assumptions required for regression analysis. To address this, a Box-Cox transformation was applied to each variable to stabilize variance and improve normality. As WLS1 and WLS2 included negative values, a constant shift was added to ensure all values were positive, facilitating the proper application of the Box-Cox transformation. Log-likelihood evaluations indicated that a lambda value near zero was optimal, suggesting that a logarithmic transformation was appropriate for both variables.
Following the transformation and initial descriptive analysis of sample characteristics, an interesting pattern emerged: smaller values clustered at the left tail of the factor score distribution, suggesting a subgroup with distinctively low awareness and preparedness levels. Further analysis confirmed that the lowest 5% of individuals, determined by the quantile threshold for WLS1 (disaster awareness) and WLS2 (disaster preparedness) scores, and who consistently responded “no” to all 6 preparedness-related items (securing furniture, stockpiling food and water, preparing an emergency kit, confirming family safety, identifying indoor evacuation routes, and identifying outdoor evacuation routes), comprised the same demographic. This shared profile led us to identify them as the “low disaster awareness group.”
The study employed the Mann-Whitney U Test (Cliff’s delta) for continuous variables and the Chi-Square Test (Cramér’s V) for categorical variables to compare the “low awareness group” with the remaining participants. These tests identified distinct group characteristics, forming a foundation for further analysis using a generalized linear model (GLM). The GLM, with a Gaussian family, analyzed relationships between transformed WLS variables and independent predictors, incorporating quadratic terms for age and household size to account for non-linear trends. Variance inflation factors (VIFs) confirmed minimal multicollinearity, ensuring model stability.
A random forest model validated predictor importance via the %IncMSE metric, with reproducibility ensured using a 20% sample and set.seed(123). Logistic regression analyzed factors distinguishing the “low awareness group,” using untransformed predictors. Odds ratios highlighted demographic and socio-economic vulnerabilities contributing to low disaster awareness and preparedness.
Ethics Statements
This study received ethical approval from the Osaka International Cancer Institute (Approval Number: 20084-12) on March 28, 2024, and subsequently from the Tohoku University Graduate School of Medicine Ethics Committee (Approval Numbers: 2024-1-231 on June 27, 2024, and 2024-1-517 on October 22, 2024), in accordance with the 2020 ethical guidelines prior to the initiation of survey activities.
Results
Item analysis (Table S1) confirmed good discriminative validity for all variables, with item discrimination values between 0.51 and 0.65. The overall KMO value was 0.85 (Table S2), with individual values ranging from 0.81 to 0.93, and Bartlett’s test of sphericity (χ2 = 134,172.3, P < .001) validated the data’s suitability for factor analysis. Parallel analysis (Figure S2) identified 2 factors. Factor 1 included “confirming family members’ safety” and “identifying evacuation routes inside and outside the home,” while Factor 2 included “securing furniture,” “stockpiling supplies,” and “preparing an emergency kit,” explaining 76% of the variance. Model fit was strong, with a CFI of 0.99, RMSEA of 0.098, and SRMR of 0.011. All variables had loadings >0.4 (Table S2), and Cronbach’s α was 0.80 for Factor 1 and 0.72 for Factor 2. The factors correlated at 0.71, indicating reliability for subsequent analyses.
In Table 1, the comparison between the low awareness group (n = 8710) and others (n = 19,771) highlights key sociodemographic and health-related differences. The low awareness group is generally younger (median age 42.8 vs. 50.6 years; effect size 0.25), with slightly smaller households (2.4 vs. 2.5 members; effect size 0.09) and a higher proportion of males (53.0% vs. 47.8%; effect size 0.05). They are more likely to be single (55.1% vs. 38.8%; effect size 0.15) and regular employees (37.6% vs. 27.0%), with fewer nonworkers (21.7% vs. 33.7%; effect size 0.14). Educational disparities (effect size 0.05) show fewer university graduates, while a greater reliance on public assistance is noted (2.9% vs. 1.2%; effect size 0.10). Health-related differences include lower health screening participation (26.9% vs. 12.4%; effect size 0.18), poorer self-reported health (14.5% vs. 17.9%; effect size 0.11), and fewer hospital visits (67.3% vs. 54.7%; effect size 0.12). Behavioral factors reveal slightly higher smoking (effect size 0.02) and lower alcohol consumption (effect size 0.06). Geographically, the low awareness group has higher representation in Chugoku and Kyushu regions (effect size 0.08), while community attachment is significantly weaker (56.2% vs. 35.9%; effect size 0.19).
Descriptive statistics for the low-disaster-awareness population and the rest of the low awareness population
a: Mann-Whitney U test (Cliff’s delta)
b: Chi-Square test (Cramér’s V)
Generalized Linear Model
In Table 2, the findings for WLS1 demonstrate significant non-linear relationships with both age (first-degree: Estimate = 10.99, P < .001; second-degree: Estimate = 3.31, P < .001) and household size (first-degree: Estimate = 4.34, P < .001; second-degree: Estimate = –2.07, P < .001). Several demographic factors showed positive associations with WLS1, including being female (Estimate = 0.03, P < .001), having a spouse present (Estimate = 0.04, P < .001), and higher educational attainment (university graduates: Estimate = 0.03, P < .001; postgraduate degrees: Estimate = 0.07, P < .001). Regarding employment status, general employees (Estimate = –0.02, P = .01) and non-regular workers (Estimate = –0.03, p = 0.01) showed negative associations, while student status had a significant positive effect (Estimate = 0.11, P < .001). Health-related factors such as good self-reported health status (Estimate = 0.12, P < .001) and regular health check-ups (Estimate = 0.08, P < .001) were positively associated with WLS1, while lacking health insurance had a negative impact (Estimate = –0.05, P = .001). Other factors showing positive associations included higher income levels (middle-income: Estimate = 0.04, P < .001; high-income: Estimate = 0.08, P < .001), community attachment (Estimate = 0.09, P < .001), and disaster experience (Estimate = 0.08, P < .001). Regional differences were also significant, with areas like Kanto (Estimate = 0.05, P < .001) showing positive effects while others like Chugoku (Estimate = –0.04, P = .01) demonstrated negative associations.
Overall generalized linear model (GLM) regression results
For WLS2, similar nonlinear relationships were observed with age (first-degree: Estimate = 10.46, P < .001; second-degree: Estimate = 3.66, P < .001) and household size (first-degree: Estimate = 6.27, P < .001; second-degree: Estimate = –3.77, P < .001). Being female (Estimate = 0.07, P < .001), having a spouse present (Estimate = 0.06, P < .001), and higher education levels (university graduates: Estimate = 0.08, P < .001; postgraduate degrees: Estimate = 0.12, P < .001) were positively associated with WLS2, while nonregular employment showed a negative association (Estimate = –0.04, P = .001). Health-related factors demonstrated strong correlations, with good self-reported health (Estimate = 0.13, P < .001), regular check-ups (Estimate = 0.13, P < .001), and hospital visits (Estimate = 0.03, P < .001) showing positive associations, while smoking had a negative effect (Estimate = –0.03, P = .001). Income levels, particularly high income (Estimate = 0.13, P < .001), showed strong positive associations with WLS2. Regional variations remained significant, with the Kanto area showing positive associations (Estimate = 0.11, P < .001) while both Chugoku (Estimate = –0.09, P < .001) and Kyushu regions (Estimate = –0.07, P < .001) demonstrated negative effects.
The variance inflation factors (VIFs) of WLS1 and WLS2 for all variables were below 10, indicating no severe multicollinearity.
Random Forest (Figure 1)
Key predictors of evacuation awareness (WLS1) and disaster preparedness awareness (WLS2) revealed both overlaps and differences. For WLS1, age was the strongest factor, followed by spouse, household size, and community attachment, with health status and caregiving also influential. Occupation, income, health screenings, and hospital visits had moderate effects, while sex, alcohol use, region, insurance, disaster experience, smoking, and vulnerable household members were less significant. For WLS2, household size and age were the top predictors, followed by spouse and region, emphasizing the roles of family structure, age, marital status, and geography. Health screenings, caregiving, community attachment, occupation, and education also contributed, with income and insurance playing secondary roles. Similar to WLS1, alcohol, smoking, disaster experience, and vulnerable household members had minimal impact.
Variable importance plot.
Overall, age, family structure, community ties, and health behaviors emerged as key drivers for both evacuation and disaster preparedness awareness.
Logistic Regression
The findings from the logistic regression analysis are consistent with results obtained from the GLM and Random Forest analyses. Key predictors of low disaster awareness and preparedness, such as age, household size, employment status, educational attainment, health-related factors, lifestyle behaviors, income, regional variations, and community attachment, showed similar associations across all models. Detailed results for the logistic regression analysis can be found in the supplementary materials (Table S3). Overall, the patterns indicate that social and demographic factors significantly influence disaster awareness and preparedness.
Discussion
This study used factor analysis to identify 2 dimensions of disaster awareness: disaster preparedness and evacuation awareness. It revealed a vulnerable subgroup characterized by low scores in both dimensions, typically younger individuals with economic instability, smaller households, and weaker community attachment.
A key contribution of this study is its nuanced dual-measurement approach, addressing limitations of previous binary classifications in studies like Tada et al. (2020), Amao et al. (2007), and Çelik (2023).Reference Tada, Iwamoto and Kubo38–Reference Çelik40 By examining both disaster and evacuation awareness, the study uncovered a strong correlation between the 2 dimensions, identifying a distinct subgroup with specific risk perceptions and behaviors. This methodological advancement highlights the importance of standardized tools for assessing disaster preparedness. Although the 2 awareness factors are strongly correlated, they likely reflect complementary but distinct behavioral domains. For instance, individuals may prepare emergency supplies but lack clarity on evacuation protocols, or vice versa. These behavioral mismatches may limit the effectiveness of preparedness efforts and indicate the need for integrated, multidimensional strategies in public health disaster campaigns.
Caregiving responsibilities significantly influenced both dimensions of awareness (WLS1 Estimate = 0.06, P < .001; WLS2 Estimate = 0.06, P < .001), suggesting that managing others’ vulnerabilities heightens disaster consciousness.Reference Ludin, Firdaus, Arbon, Heggy, Lee and Zekry41 This expands the focus of disaster research beyond individual traits to family dynamics, emphasizing how caregiving roles shape preparedness. Economic vulnerability, nonregular employment, and public assistance reliance were also linked to lower awareness levels, reinforcing global concerns about disaster preparedness inequities.Reference Nirupama, Popper and Quirke42, Reference Kyne43 Lifestyle factors like occupation type, caregiving, and hospital visit frequency showed distinct patterns of influence.Reference Doğulu, Ikizer and Karanci44 While community attachment positively impacted awareness, age and marital status emerged as stronger predictors, challenging the focus on community ties in conventional research.Reference Jang36, Reference Alshehri, Rezgui, Li, Brebbia and Beriatos45
The identified sociodemographic patterns underscore the importance of designing demographic-specific interventions. For example, nonregular workers—who often fall outside institutional safety nets—may benefit from workplace-based preparedness programs, while older adults might require offline communication tools and assistance in evacuation drills. Although Japan has a nationally coordinated disaster-management plan formulated by the Cabinet Office, evacuation orders and related warnings are issued by individual municipalities. This administrative arrangement means that local cultural norms, past disaster experience, and the capacity of municipal governments all shape residents’ preparedness. The lower readiness observed in regions such as Chugoku and Kyushu—though the precise reasons remain unclear—may reflect these contextual factors and the differing salience of major events like the 2011 Great East Japan Earthquake, whose lessons resonated most strongly in eastern Japan. Future research should therefore examine how such localized sociocultural and administrative conditions mediate preparedness behaviors and whether region-specific policies could prove more effective than uniform national measures.
To address these issues, policymakers may consider going beyond general awareness campaigns by integrating preparedness components into existing social welfare programs, which could better support economically vulnerable groups. While our findings do not establish causality, the observed associations suggest that such integration may offer benefits. Workplace initiatives targeting nonregular employees, along with community-based efforts leveraging social networks, can enhance preparedness for households with vulnerable members.Reference Howard, Agllias and Bevis31, Reference Kwon and Ryu46 These strategies aim to reduce inequalities and build inclusive systems for comprehensive disaster awareness and resilience. For example, local governments could consider embedding disaster preparedness checks into routine community health outreach services for recipients of public assistance, or codevelop evacuation guidance with employers of part-time or nonregular workers. Furthermore, while younger individuals and those without strong community ties were identified as having lower preparedness levels, they may play crucial roles during disasters due to their physical capacity and access to information. Therefore, involving these populations more actively in community-based disaster planning could not only reduce their isolation but also enhance overall resilience.
Although this study focused on Japan, many observed associations—such as those related to age, economic status, and household structure—are likely relevant across disaster-prone regions globally. While Japan’s “bosai” culture offers a unique context, the dual-factor framework and indicators used here are adaptable to other settings. The standardized methodology and diverse sample support the external validity of the findings, though cross-national research is needed to further validate and refine the approach.
Implications
The analysis underscores the need for disaster preparedness policies to prioritize vulnerable groups through targeted, data-driven strategies. Policymakers must move beyond one-size-fits-all approaches by integrating preparedness education into institutions and strengthening community networks. Focusing on predictors like age, economic stability, and social support can transform resilience into a collective societal effort rather than one limited by individual resources.
The assessment tool developed in this study refers to a dual-factor index combining disaster preparedness and evacuation awareness. With further validation, this approach can help identify regional and demographic disparities and inform more targeted interventions. By integrating these 2 key dimensions, the tool may support municipal-level efforts in planning and prioritizing disaster preparedness resources. This structured method contributes to improving the evidence base for resilience-building policies.
Additionally, our findings suggest that place attachment plays a nuanced role in shaping evacuation awareness. While community attachment may enhance preparedness, prior studies have indicated that excessive attachment can hinder evacuation decisions in times of crisis. Recognizing this dual nature of place attachment highlights the need for disaster communication strategies that both emphasize the temporary nature of evacuation and ensure a reassuring return process—thereby reducing perceived trade-offs between safety and belonging.
Limitation
This study employed GLM for their flexibility in handling non-normal outcomes. However, GLM still faced limitations, such as difficulties in fully capturing potential non-linear relationships and reduced explanatory power when key covariates (e.g., socioeconomic status, health behaviors) were not included. Future studies may address these issues by incorporating a broader set of predictors and exploring continuous variables to better leverage advanced modeling approaches for precise predictor-outcome relationships. Additionally, this study relied solely on quantitative data, which, while effective for identifying patterns, may overlook contextual nuances in how individuals perceive and respond to disaster risks. Incorporating qualitative methods such as interviews or focus groups in future research could reveal deeper motivations, cultural influences, and structural barriers, offering a more comprehensive basis for intervention design.
Limitations included selection bias, reducing generalizability, and the study’s observational design, which limits causal inference. Categorical variables may oversimplify relationships, and excluding covariates like community support and household dynamics reduced explanatory power. Using more detailed and representative data could address these issues in future research.
Conclusion
The study identifies a strong overlap between disaster preparedness and evacuation awareness, both heavily influenced by social vulnerabilities such as age, economic stability, household size, and community attachment. The analysis highlights that individuals facing socioeconomic disadvantages consistently exhibit lower awareness levels. This points to systemic issues requiring unified policies that integrate both preparedness and evacuation planning to address shared vulnerabilities. By focusing on these areas, targeted interventions can bridge awareness gaps and enhance resilience for high-risk populations. This nuanced understanding provides valuable insights for shaping inclusive disaster management frameworks globally.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/dmp.2025.10202.
Author contribution
Authors had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: All authors; Acquisition, analysis, or interpretation of data: All authors; Drafting of the manuscript: All authors; Critical revision of the manuscript for important intellectual content: All authors; Statistical analysis: Akemi Hara, Akihiko Ozaki, Michio Murakami; Administrative, technical, or material support: Takahiro Tabuchi; Supervision: All authors.
Competing interests
Dr. Ozaki received personal fees from Medical Network Systems Inc., Kyowa Kirin Company Limited, and Taiho Pharmaceutical Company Limited outside of the submitted work. Dr. Tanimoto received personal fees from Medical Network Systems Inc. and Bionics Company Limited outside of the submitted work. No other disclosures are reported.
