Designing a Short Disaster Risk Reduction Course for Primary Schools: An Experimental Intervention and Comprehensive Evaluation in Hue City, Vietnam

Abstract

Disaster risk reduction (DRR) education is considered increasingly necessary, particularly for children. DRR educational interventions aim to enhance knowledge and attitudes related to self-protective capacity. However, comparative studies on students in areas prone to different disasters and comprehensive criteria covering both knowledge and attitudes toward behavior remain limited. A short DRR course was developed for primary schools across three regions (mountainous, low-lying, and coastal) in Hue City, one of Vietnam’s most vulnerable areas to extreme weather events. This study aimed to comprehensively evaluate student performance by applying Bloom’s taxonomy and treatment-control pre-post-follow-up design with panel analysis methods. From December 2022 to September 2023, three surveys, involving 517 students each, were conducted in six schools (three schools received the course and surveys, while the other three only participated in surveys). The intervention revealed similarities and differences between the groups. The course positively impacted on some elements of knowledge and preparedness intentions in students from low-lying and mountainous regions (including ethnic minorities). Higher-grade students in the mountainous region showed improvement in intentions, but not in attitudes toward self-protection. No gender differences in intentions were found. Although limited overall improvements, the study’s various methods, approaches and continuous assessment can be applied globally to design, implement, and assess DRR education courses effectively.

1. Introduction

Since the 1970s, a significant increase in both the frequency and intensity of disasters has been observed, particularly for weather-related events such as storms and floods. This growing complexity is partly attributed to the well-documented increase in greenhouse gas emissions and subsequent global warming [1,2,3].

Vietnam, a Southeast Asian nation, ranks among the countries most significantly affected by global climate change and is vulnerable to severe weather conditions [4,5]. Notably, Vietnam was among the top six countries most severely affected by extreme weather events between 1999 and 2018, according to the 2020 Global Climate Risk Index [1,4,6]. This susceptibility is particularly concerning in Vietnam’s coastal areas, which stretch over 3260 km and house approximately 70% of the population [7,8]. These regions face a growing threat from climate change-induced phenomena, such as storms, floods, and rising sea levels. Hue City is one of the most disaster-prone areas in Vietnam [9,10].

Children are known to be the most vulnerable group to disasters because they depend on safety and security from adults, and their psychological consequences can be significant in the long term [11]. Children, particularly those in developing countries, are vulnerable to the negative impacts of disasters on their physical and mental health, as well as educational opportunities. Over the last 20 years, children have constituted one-third of the population affected by disasters in Vietnam [12]. Therefore, education is considered crucial for disaster risk reduction (DRR). The critical role of DRR education within educational systems has been further emphasized in the Sendai Framework for Disaster Risk Reduction for 2015–2030. This Action Framework prioritizes disaster risk understanding, governance strengthening, DRR investment, and preparedness enhancement, aligning with the Action Framework adopted by Vietnam and 186 other nations [13,14]. Engaging in educational programs can foster a deeper understanding of disaster risks, ultimately ensuring appropriate responses during emergencies and contributing to the long-term resilience of communities [15,16]. In this context, prioritizing DRR education targeted toward elementary school students is essential. While many educational programs refer to “natural disasters”, this study followed the perspective of the UNDRR’s “No Natural Disasters” campaign [17], emphasizing that disasters are caused by the intersection of natural hazards and human vulnerability. Instead, the terms “disaster” or “disaster event” were used.

Bonifacio et al. highlighted the importance of considering context-specific issues related to the location of schools, such as coastal regions, in education [18]. This implies that it is crucial to consider the challenges in the specific areas where schools are located. For example, schools in coastal regions may face coastal erosion in addition to storms and floods, which is different from other regions. This suggestion was addressed in this study. This study was conducted in Hue City in mountainous, lowland, and coastal regions to consider different types of disasters. Owing to its complex climate, characterized by heavy rainfall and seasonal monsoons, and unique topography with various regions and coastal locations, Hue City faces the annual impact of most disasters. Moreover, the city is most vulnerable to climate change [19,20]. Flooding often disrupts educational activities by damaging school facilities (see Section 3).

In Hue City, the absence of parents from home often necessitates children to act independently during emergencies to save their lives. This vulnerability is further exacerbated by inadequate disaster prevention measures [21,22]. This study highlights the self-protective capacity suggested by Johnson et al. [23] when designing DRR programs. Therefore, it is essential to equip children with comprehensive knowledge and help them internalize the momentum of saving lives.

This study designed a DRR course for primary school students. The intervention aimed to provide knowledge and help students enhance their self-protective capabilities to save their lives. This study comprehensively evaluated the program by applying taxonomies of education and a field experiment measurement technique. Furthermore, the study explored the similarities and differences in the effects of the program in three topologically different regions. Recommendations were provided to improve the teaching methods for children in diverse areas.

2. Present Practice and Literature Review

2.1. DRR Education in Vietnam and Hue City

The Department of Education and Training is responsible for organizing campaigns to disseminate knowledge and skills on disaster prevention, mitigation, and response to climate change in schools according to the Plan to Implement Decision No.987/QD-TTG dated 9 July 2020, by the Prime Minister to strengthen the party’s leadership in the prevention, response, and overcoming the consequences of disasters (Plan No. 185/KH-UBND) [24]. In Hue City, raising awareness and understanding of climate change for students through extracurricular education and developing solutions to integrate climate change into education at all levels of the city is part of Task 3, Section 1 of the Action Plan to respond to climate change in City for the period 2021–2030, Vision 2050 (Decision No. 1720/QD-UBND) [19,25].

Disaster prevention topics are primarily integrated into teaching activities and extracurricular programs, such as knowledge contests, drawings, and dramas at the middle school level [26]. Many initiatives, with support from domestic and foreign organizations, have been conducted to train teachers and enhance the knowledge and skills of students at all levels. Tong et al. [26] presented various educational programs and projects related to DRR education in Vietnam and Hue City. In particular, the Ministry of Education and Training (MOET), focuses on integrating disaster preparedness into its curriculum and on developing educational materials for children, teachers, and community leaders. Inter-provincial dramas, quizzes, painting competitions, and safety drills were held. Booklets, teaching, and training materials were created in “Introducing Disaster Preparedness in Primary Schools” by the Vietnam Red Cross Society (VNRC) in 2001 [26]; “Integrated disaster preparedness in Thua Thien Hue Province” by VNRC in 2010; “Integration of DRR into schools” by Development Workshop France (DWF) in 2008 [26] to strengthen infrastructure and build safe new schools; and “Capacity building for DRR education for schools in Central Vietnam” by Sustainable Environment and Ecological Development Society (SEEDS) Asia from 2011 to 2013 to enhance the knowledge of DRR for teachers and governmental officials. These frameworks and initiatives demonstrate that the city is prioritizing DRR education. However, these practices are limited to elementary education.

Although these activities have attempted to enhance knowledge through competition, learning kits, and self-protective capacity-building activities, appropriate evaluation measures are absent. Without a proper evaluation, it is difficult to determine whether the knowledge, skills, and teaching methods provided by these programs are adequate.

2.2. Literature Review

2.2.1. Concepts of DRR Education

Many previous DRR education courses have emphasized disseminating knowledge (disaster causes, nature, and consequences) and skills (disaster prevention and mitigation measures) [27,28,29]. However, studies on DRR behaviors have revealed the ineffectiveness of these programs and the importance of fostering self-protective capacities to respond promptly to disaster risks [23]. According to Katada and Kanai, “attitude-oriented disaster prevention education” is the most effective approach for addressing this issue. This differs from “threatening disaster prevention education” and “knowledge-oriented disaster prevention education”, as it emphasizes children’s endogenous disaster prevention actions and prioritizes protecting lives. In Kamaishi City, Iwate Prefecture, “disaster education with a positive attitude” was identified as a crucial factor in the successful voluntary evacuation of children during a tsunami [30]. Thus, the DRR education course of this study was designed not only to supply basic knowledge about disasters, but also to foster self-protective capacities.

2.2.2. Evaluation Criteria for DRR Education

In DRR education, the most commonly used evaluation criteria are assessing knowledge about disaster events and protective actions during the emergency or preparedness phases. Most scholars have used knowledge-based outcome indicators to answer knowledge-related questions. Ronan and Johnston evaluated knowledge of response-related protective behaviors using correct responses for one hazard out of three to six available responses in children aged 7–13 years [31]. Similarly, Finnis et al. used an evaluation method for knowledge of harm mitigation and response behavior of 13–18-year-old children based on selecting correct responses. These were multiple-choice questions [28]. Similarly, Soffer et al. used this question format to evaluate the theoretical and practical knowledge of earthquakes in children aged 10–12 years [32].

Bloom’s taxonomy of cognitive domains has been applied in several studies. For example, Le et al. modified Bloom’s taxonomy within the cognitive domain to evaluate the knowledge of fire students after a new training course on landslide rescue. Student knowledge was evaluated using six open-ended questions for each hierarchical level, including “Remembering”, “Understanding”, “Applying”, “Analyzing”, “Evaluating”, and “Creating”, and concisely answering each question [33]. Molan et al. used the cognitive domain of Bloom’s taxonomy to evaluate children’s performance in bushfire safety after being educated using the designed learning tool. He divided the cognitive domains into lower and higher thinking skills and used a simple scoring system to assign one point to one correct response to assess student performance [34].

In addition to knowledge, Ronan and Johnston evaluated hazard awareness and risk perception regarding hazard occurrence, the risk of injury, and psychological issues related to fear or upset when discussing hazards. Students were instructed to answer a series of questions on a 3-point or 7-point scale [31].

Katada and Kanai evaluated self-protective capacities through attitude changes toward earthquakes following disaster prevention education. The study examined children’s responses to an actual disaster. For the eight questions, students selected one from a list of seven answer options, ranging from completely agree to completely disagree. The effects of education were evaluated by incorporating quantitative and qualitative evaluations by teachers [30].

Comprehensive evaluation criteria covering the knowledge and self-protective capacity domains have yet to be developed. Although the actual disaster response is a decisive criterion for evaluating self-protective capabilities, it is rarely observable. Thus, new comprehensive evaluation criteria that can be applied to ordinary times in DRR education are required.

2.2.3. Evaluation Framework for DRR Education

One-shot surveys and before–after comparisons are frequently used for DRR education evaluations. A one-shot survey was used in a single drill for tsunami evacuation and disaster education [35] and in a practical disaster prevention drill for disaster medical education [36]. In one group, before–after comparisons, as Johnson et al. [23] mentioned, were applied in a hypermedia system to evaluate earthquake education [37], using game techniques to test the communication of disaster information and knowledge to children [38] and to promote disaster awareness in multicultural societies [39]. However, these methods have a fundamental problem: they cannot distinguish the effects of educational attempts from the influence of external events that occur during the education period. A formal method for addressing this issue is through a randomized controlled trial. Typically, this method randomly designates control and treatment schools and conducts an education program only for treatment schools. The effect of education was measured as the difference in student performance between the treatment and control schools. However, owing to geographical, legal, and other limitations, it is often impossible to randomly select treatment and control schools. Thus, intentionally selecting a control school similar to the treatment school and comparing their performances is the second-best option [40]. Comparing treatment and control groups is uncommon in DRR education evaluations [23]. However, there are some examples in DRR education, such as earthquake preparedness and protective behavior [32,41,42,43] and volcanic risk [44]. Similar approaches have also been applied in other educational areas, such as waste separation education in primary schools [45,46], stunting prevention training among children in high schools [47], and an online pilot short course [48]. This evaluation framework was applied to our study.

2.2.4. Methods for DRR Education

Kagawa and Selby [49] and Nyberg et al. [50] mentioned several methods of teaching DRR ideas. These are interactive learning (group discussions with pairs or groups; brainstorming), inquiry learning (case study research), surrogate experiential learning (films, games, plays), field experiential learning (field trips, mapping, surveys), and action learning (campaigns, projects, street theater).

This study adopted interactive learning, including group discussions on disaster risk topics. Surrogate experiential learning, such as minigames and learning from overseas disaster experiences through a foreign DRR video for schoolchildren and talks by returnees from other countries, was also incorporated.

3. Survey Area

Hue City is located in the north-central coastal region of Vietnam (see Figure 1). Before 1 January 2025, it was known as Thua Thien Hue Province, but it was reclassified as a city under Resolution No.175/2024/QH15 [51]. Data from 2023 indicate that the city covers 5033 square kilometers and is home to a population of approximately 1.16 million people. This population is relatively balanced, with slightly more women (585,114) than men (581,433). In terms of distribution, urban areas house approximately 616,235 residents, while rural areas house 550,312 residents. The average population density for the city is 235.8 people per square kilometer [20,52].

Hue City is hot and humid with a monsoon climate. It receives high rainfall, averaging over 2700 mm and concentrated heavily between September and February. This can be even higher in some areas. Humidity is consistently high, ranging from 85% to 86%. Southwest and northeast monsoons affect the city [53].

Heavy rain and storms have frequently affected the city. For instance, prolonged heavy rains from October to November 2010 suspended learning for approximately 130,000 students across 330 schools. Similarly, the major flood in November 2017 inundated over 40% of schools, with some low-lying regions experiencing flood depths between 0.5 and 1.5 m. This pattern continued in late 2018, with heavy rains causing localized flooding and the closure of more than 20 schools. In 2020, the Department of Education and Training was forced to suspend school activities at all levels owing to the impact of severe storms and floods. This decision stemmed from the widespread damage and degradation of infrastructure in low-lying regions, including schools and medical facilities. Reports indicate that the city experienced damage to 225 classroom roofs, leaks in 1425 rooms, collapsed or tilted fencing stretching 4714 square meters, and the uprooting of 3442 trees [20].

The topography of Hue City can be broadly divided into three distinct zones: delta, coastal and lagoon, and mountainous regions, as cited in Lee et al. [54]. Each region experiences a different type of disaster event. The delta region, which is a low-lying region, is particularly prone to annual flooding [54,55]. Coastal and lagoon regions are highly vulnerable to typhoons, floods, and rising sea levels because of their proximity to the coast [54,56,57]. Mountainous regions experience strong winds, heavy rains, and rising water levels that can lead to flash floods and landslides [55].

Our survey covered three regions in Hue City: mountainous, low-lying, and coastal, each with distinct types of disasters. Based on the synthetic report of the Climatic Assessment [20] and meeting from the local government, we selected one district from each region to represent the geographical location and the most significant impact of specific types of disaster event in these regions. We selected Nam Dong District from the mountainous region and Quang Dien District from the low-lying region. Finally, Phu Vang District, representing the coastal region, is affected by coastal erosion in addition to the above-mentioned regional hazard types.

DRR initiatives have been implemented in these regions. In 2017, approximately 1000 students in Phu Vang District participated in the “Schools of Son Tinh” campaign to enhance their knowledge and skills to save their lives [5,58,59]. According to the internal reports from Department of Education and Training of Quang Dien District (collected by the author on 9 December 2022), in Quang Dien District, approximately 18–19 primary schools organized painting and knowledge contests for students, following the activity for improving awareness and skills on environmental protection, climate change, and disaster risk reduction for teachers and students in primary and secondary schools. These activities reveal that the government and other organizations consider promoting DRR in the education sector in these regions. Although some primary and junior high schools within Hue City have implemented disaster prevention initiatives, including extracurricular activities such as painting and knowledge contests according to the internal reports from Department of Education and Training of Quang Dien District (collected by the author on 9 December 2022) and Phu Thuan 1 Primary School (collected by the author on 14 December 2022), disaster preparedness knowledge and behavior across various topographical regions have not been assessed.

4. Methodology

4.1. Research Framework

This study adopted a treatment-control pre-post-follow-up design [60]. Three treatment schools, one in each study region, participated in the DRR education course and questionnaire surveys, whereas three control schools from the three regions participated only in the questionnaire surveys. Assessing participants immediately after a program is a common method for evaluating knowledge gain [61]. However, in most studies, the long-term impact of these educational interventions was not evaluated [61,62] (Nolan et al. [63] cited). Therefore, an additional follow-up survey, Survey 3, was conducted in this study. Three surveys were conducted in April, May, and September 2023 to assess students’ knowledge and self-protective capabilities regarding disasters (See Figure 2).

4.2. Selection of Schools and Participants

Preparation for school selection began in December 2022 by collecting related data. This included reviewing reports and meeting minutes from school administrators and local governments (e.g., school year summary reports, Thua Thien Hue Province Statistical Yearbook 2021, Synthetic Report of the Climatic Assessment of Thua Thien Hue Province). These documents provided general information about the survey area, including prevalent disaster events, their impacts, and existing disaster education activities. Following data collection, permission was obtained from the local governments to approach schools across the three regions. Initial meetings and negotiations with eight schools led to agreements with six schools participating in the study. Among the four candidate schools in the low-lying region, two could not participate in our study because of conflicts in school schedules. Two schools in each region were selected so that the socioeconomic status, place of residence, and educational curricula were similar. These schools were then assigned as either treatment or control schools. The locations of six schools are shown in Figure 3.

Younger students (ages 9–10 years) are more receptive to positive attitude changes than older students (ages 11–13 years) [64]. Thus, 3rd and 4th graders were selected for this educational attempt. The selection of these grades also enabled three waves of follow-up surveys before they graduated to 5th grade. Each school has two to four classes in the 3rd and 4th grades. Depending on the school, two to four classes were selected, accounting for 40–100% of students in these grades participating in the study. In the control school in the mountainous region, one 3rd-grade class and one 4th-grade class were selected; these were all available. Accordingly, in each region, six–eight classes from grades 3 and 4 were selected to participate in the survey. Each class comprised an average of 26 students.

Table 1. Number of students and classes that participated in this study.

Region	School Type	Number of Classes	Number of Students
			Survey 1	Survey 2	Survey 3
Mountainous (N = 392)	Treatment	4	86 (F:39, M:47)	76(F:30, M:46)	83 (F:36, M:47)
	Control	2	53 (F:26, M:27)	55 (F:21, M:34)	39 (F:19, M:20)
Low-lying (N = 471)	Treatment	4	102 (F:42, M:60)	104 (F:43, M:61)	83 (F:36, M:47)
	Control	3	68 (F:25, M:43)	69 (F:24, M:45)	45 (F:17, M:28)
Coastal (N = 690)	Treatment	4	104 (F:51, M:53)	100 (F:50, M:50)	98 (F:50, M:48)
	Control	4	128 (F:60, M:68)	130 (F:62, M:68)	130 (F:61, M:69)
Total (N = 1553)	-	21	541	534	478

summarizes the number of participants.

4.3. Experimental DRR Education Course

A short DRR course was developed for primary school students to help them save lives proactively. The intervention aimed to empower children with the necessary knowledge and skills. It focuses on age-appropriate, engaging, and interactive methods for educating children on disaster preparedness, response, and risk reduction.

Table A1. DRR course lesson.

No.	Lesson	Content	Target	Objectives	Activities	Teaching Methods
1	Flooding/Storm Definition Chief reason	Chief hazards in the local region (flooding/storm and chief reasons)	To help students understand the chief hazards and the reasons for the living place	- Students can define chief hazards in the living place - Students can understand the reasons of flooding/storm	- List some disasters in the living place - Select types of chief disasters from pictures - Answer the questions about the reasons of flooding/storm	Questioning Picture Using the pictorial representation by placing an X on A4 paper
2	Impacts	Impacts of flooding/storm on the people, animals, and other aspects	To help students understand the impacts of flooding in the living place	- Students can describe the signs of disasters - Students can determine the impacts of flood on themselves	- Recognize the signs of disaster - Discuss the impacts of flooding/storm on the student’s life	Questioning Game to recognize Group discussion
3	Solution: Skills and changing attitude	Suitable activities and attitude when flooding/storm occurs	To help students identify the most important actions when a flood occurs	- Students can determine the suitable solutions when flooding occurs - Students can realize the importance of positive attitude in disaster prevention	- Discuss actions to perform before, during, and after flooding/storm - Answer through the video, what and how you can respond in the thunderstorm and heavy rain situation	Group discussion Video of Japanese students Questioning
4	Solution: Skills and changing attitude (cont.)	Increasing the activity in sharing information and evacuating to save others	To help students compose a way to share information with others in evacuation	- Students can determine the benefits of sharing information to others - Students can create other ways to help others in evacuation	- Listen to the talks by people who used to live abroad - Answer what information to share and how - Thinking about other ways and answer	Questioning Experienced people’s sharing Message Game: Should and Should not Write the message

presents the contents, target, objectives, activities and teaching methods of the four lessons. Each lesson lasted 35 min and was conducted during one school period. Educational activities were conducted in April and May 2023. School administrators arranged the time for our activities in early May so as not to affect the school activities and final examinations. The teaching material used to help students’ understanding is available as Supplemental Material for this paper.

The lessons covered causes, impacts, and solutions, such as appropriate response behavior and evacuation. In addition to the storm and flood situations that were considered in all regions, we added coastal erosion in the coastal region and landslides and flash floods in the mountainous region.

The use of a DRR educational video from another country was the core of Lesson 3. Visual tools are considered effective for educating children in various fields [18]. Shiwaku and Shaw used videos for DRR education during a large flood [65], and Kamil et al. used videos for geographic literacy to improve disaster knowledge [66]. Therefore, in this course, in addition to using pictures and games in Lesson 3, we presented a short drama about how Japanese students risked but saved their lives in sudden heavy rains and thunderstorms. As the video was in Japanese, the authors of this paper added subtitles in Vietnamese. Showing a DRR educational video featuring foreign situations to local children is a new attempt at DRR education.

Lesson 4 featured talks with Vietnamese returnees from Japan who had lived in Japan and experienced a disaster or DRR education. Listening to guests talk about experiences beyond student thoughts draws student attention. Mort et al. [67] invited knowledgeable older people, a social activist from the neighborhood, and a voluntary firefighter, who were considered “local experts”, to share their experiences of past disaster events and what children should do in case of a disaster.

Based on the approach of encouraging positive attitudes in fostering self-protective capacities, as discussed by Katada and Kanai [30] and the role of voluntary local people, as mentioned by Mort et al. [67], Vietnamese returnees from Japan were invited to participate in the lesson. Each returnee is a resident who understands the disasters in the survey area and can provide advice and encouragement to strengthen students’ self-protective capacities. Students found trusting and accepting their guidance easier due to their closeness to the community. In the lesson, meeting the returnees to share their experiences with local children was an exciting and notable moment in this study. The returnees talked either in class or on video for 15 min. Regarding the mountainous region, a woman shared her experiences in Aichi Prefecture, asking students to stay in a safe place and protect themselves during earthquakes and storms. In the coastal region, another woman shared via video recording her experiences with storms, earthquakes, and volcanic eruptions in Kagoshima and Tochigi, emphasizing the need for preparation and staying in a safe place. In the low-lying region, a man returning from Nagoya spoke about earthquake safety, urging the people to find cover to protect themselves.

Except for the part involving the returnees, the first author taught all lessons with enthusiastic support from school administrators and teachers. In the low-lying and mountainous regions, students had two periods per week, whereas in the coastal region, they had only one period per week. Additionally, unlike in other regions, the teaching schedule in the coastal region was interrupted for three weeks between the first and last three lessons because of conflicts with the school’s schedule.

4.4. Evaluation Criteria and Measurement Questions

Our comprehensive evaluation criteria included knowledge, attitude, and self-protective capabilities for dealing with disasters; students’ feelings about the learning process; and students’ satisfaction.

4.4.1. Knowledge and Attitude Criteria

The knowledge criteria were developed based on the six levels of revised Bloom’s taxonomy (remembering, understanding, applying, analyzing, evaluating, and creating) within the cognitive domain [68].

Table 2. Knowledge level criteria.

Level	Cognitive Domain	Task	Question
C1	Remembering	Selecting types of disasters	Which of the following types include disasters?
C2	Understanding	Discerning the correct causes and explaining	Are the following statements relating to the causes true or false? If incorrect, correct it in the explanation column
C3	Applying	Understanding how to prepare for disasters	When there is a likelihood of a storm or flood, in addition to watching the weather forecast, what else do you need to do to prevent storms and floods?
C4	Analyzing	Commenting about right actions after disaster	After the storm and flood passed, in front of a friend’s house there were leaves and garbage floating; the friend’s parents were sweeping rubbish and leaves. The friend played video games and did not help his parents sweep the trash. What do you think about the friend’s actions?
C5	Evaluating	Assessing response to heavy rains, thunderstorms situation	A and B were playing in the village’s football field, when it was windy and dark clouds appeared. It began to rain and it rained harder and harder. B said: Let us wait under this tree and call out to the adults. It is okay to rain like this (while it is raining heavily). A said: I should find a place to stay. Find the nearest safe and tall house for us to move there. We should not wait under the tree; it is both wet and dangerous if there is thunder. In your opinion, who is more reasonable (A or B) and please explain why?
C6	Creating	Using creative ways to propose solutions	According to you, what additional activities or games should the school have to help you better protect yourself when a disaster event occurs? (Encourage using creative ways to answer, such as a poem)

presents the six taxonomy levels and the questions for measuring the corresponding knowledge levels. We referred to studies by Vu [69], Noor et al. [70], and Le et al. [33] for creating these questions.

Students’ scores for each question in the cognitive domain were evaluated on a 0–3 scale. For C1 (Remembering), students were asked to select the type of disaster event using a multiple-choice question. Students received 3 points for a correct answer and 0 points for an incorrect answer. In C2 (Understanding), students were required to discern the correct causes and provide explanations. This question combined a yes/no component and an explanation. The maximum score was 3 points, with 1 point for a correct yes/no response and 2 points for a proper explanation. For C3 (Applying), understanding how to prepare for disasters, C4 (Analyzing), commenting on the correct actions after a disaster, and C5 (Evaluating), assessing responses to situations such as heavy rain and thunderstorms, the students were assessed through open-ended questions. The students scored up to 3 points. The question in C6 (Creating), which asked participants to propose solutions, was also open-ended. If students provided a suitable solution, they were awarded 1 point, and an additional 2 points were provided for a clear explanation. A lack of response or incorrect answers resulted in 0 points for all levels.

The affective domain is related to transformation in interest, attitudes, values, appreciation, and adequate adjustment [71].

Table 3. Affective domain criteria.

Level	Affective Domain	Question
A1	Receiving	I am willing to follow the pictures and videos in the lessons
A2	Responding	I am willing to answer the questions asked by the teacher I can actively participate in the group exercises
A3	Valuing	I feel confident in explaining the content learned
A4	Organizing	I can identify and handle different situations when a disaster occurs
A5	Characterizing	I can act to share what I learn with people around

presents the questions created based on the concept of the affective domain, including receiving, responding, valuing, organizing, and characterizing, as in Krathwohl et al. [72] and Bloom et al. [73]. This index was applied to the treatment schools in Survey 2. Students answered each question on a 5-point Likert scale (1 = strongly agree to 5 = strongly disagree) to evaluate their attitude.

Knowledge, or the cognitive domain in learning, has been emphasized and researched more than the affective domain [74]. The affective and cognitive domains can interact with each other [75]. In this study, the affective domain is associated with the cognitive domain. Various teaching methods were employed during the lessons to provide knowledge and motivate students’ self-protective capabilities. Each level in the affective domain corresponds to the content of the lessons and is related to the cognitive domain.

4.4.2. Disaster Preparedness Intention and Its Factors

Based on the Theory of Planned Behavior (Ajzen and Fishbein [76]), self-protective capabilities were measured through disaster preparedness intention. This theory identifies three factors affecting students’ intentions to save their lives. These are attitude toward behavior, perceived behavioral control, and subjective norms (see

Table 4. Disaster preparedness intention and its factors.

Factor	Question
Intention	I am willing to participate in school activities to reduce the impact of disasters.
	I will do my best to protect myself from a disaster.
	I am ready to move to a safe place before disaster strikes.
	I am ready to help my parents with disaster prevention.
	I am willing to share information about disaster prevention with my parents.
	I am ready to remind and help people when disasters occur.
Attitude toward behavior	Never playing outside in or near dangerous areas (such as floods, landslides/coastal erosion) is a good thing.
	I feel nervous if I must move to a safe place alone without an adult.
	Helping my parents arrange things in the house before disasters occur is useful.
	Always updating information about disasters is beneficial.
	Reminding my friends not to play in dangerous areas (floods, landslides/coastal erosion, etc.) is a good thing.
Perceived behavioral control	My school and my home are safe places where I can stay during a disaster.
	I remember easily the phone numbers of my parents and teachers to contact in case of an emergency.
	I can move to a safe place alone without an adult being with me easily.
Subjective norm	It is important to follow instructions from parents and teachers when responding to disasters.
	Everyone has responsibility toward disaster prevention.

). The theory assumes a positive effect of intention on actual behavior. However, this final linkage was not examined in this study because testing it requires information on students’ actual responses to disasters, and the field survey time of this study was not sufficiently long to observe them. Students were instructed to answer each item on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree). Four measurement questions of the four factors were adopted from Phan Hoang and Kato [46], Agboola et al. [77], and the CHEAKS (Children’s Environmental Attitudes and Social Knowledge Scale) questions, referred to Leeming et al. [78], Alp et al. [79], Treagust et al. [80], and Cruz and Manata [81].

4.4.3. Satisfaction and Motivation to Participate in the Program

Students’ overall satisfaction was assessed among the treatment school students using a 5-point Likert scale (1 = very dissatisfied to 5 = very satisfied). Additionally, students were asked to rate their preference for the nine teaching methods on a 5-point Likert scale (1 = really dislike to 5 = really like). An open-ended question was also provided for students to comment on ways to improve their teaching methods.

Based on the cognitive and affective domains, different levels of knowledge and attitudes formed the foundation for improving teaching methods, as noted by Bloom et al. [73]. The chief methods were applied at each level. If there was a positive change in the level, the relevant techniques were effective in improving the students’ performance. Akasah and Alias [82] highlighted how student responses to different teaching methods can affect their learning outcomes. Measuring the affective domain provides instructors with crucial information for enhancing their teaching methods.

4.4.4. Questionnaires

The questionnaire in Survey 1 comprised 43 questions. These questions addressed personal attributes, general perceptions of disasters, six knowledge items in the cognitive domain, students’ awareness of disasters, knowledge of disaster risk reduction strategies, and intentions toward disaster preparedness behaviors. Survey 2 included 35 questions, with general information and perception items excluded. Students in the treatment schools answered an extended version including 17 additional items (11 concerning student satisfaction and six based on Krathwohl and Bloom’s affective domain). The questionnaire in Survey 3 consisted of 38 questions. This version retained 16 questions assessing intentions toward disaster preparedness behaviors and included new items on students’ past behaviors and disaster-related activities within the previous three months. Students in the treatment schools answered one additional question (39 in total) related to memory of the lesson content. The reduction in the number of questions was a deliberate decision to focus on items considered most relevant and closely aligned with the core objectives of the study. The questions from the three surveys are summarized in

Table 5. A summary of the questions in three surveys.

Content	Survey 1	Survey 2	Survey 3	Treatment (T)/Control (C)	Total Questions	References
General information	✓	✓ (reduced)	✓ (reduced)	T, C	8	Created by authors
General perception of disasters (Intention to disaster preparedness (16 questions) see Table 4)	✓	✓ (reduced)	✓ (reduced)	T, C	29	Theory of Planned Behavior (Ajzen and Fishbein [76]), Phan Hoang and Kato [46], Agboola et al. [77], and the CHEAKS questions, referred from Leeming et al. [78], Alp et al. [79], Treagust et al. [80], and Cruz and Manata [81].
Knowledge related to disasters (Cognitive domain (6 questions) see Table 2)	✓	✓	✓	T, C	6	A revision of Bloom’s taxonomy of educational objectives (Anderson et al. [68], Vu [69], Noor et al. [70], and Le et al. [33]
Satisfaction and motivation (Affective domain (6 questions) see Table 3)	—	✓	—	T only	17	Krathwohl et al. [72] and Bloom et al. [73]
Memory of lesson	—	—	✓	T only	1	Created by authors
Disaster activities during vacation	—	—	✓	T, C	12	Created by authors

✓: The content was included in the survey.

, and the questionnaires are available as Supplemental Material.

4.5. Data Analysis

Our dataset of answers for the knowledge criteria and intention toward disaster preparedness behavior included three observations for treatment and control students. This panel structure allows the use of the difference-in-differences method, which removes external influences on student answers and achieves the unbiased effect of DRR education conducted in treatment schools [83]. Difference-in-differences estimation was performed using fixed effects regression models [84,85,86]. Using fixed effects controls for unobserved characteristics that remain constant over time and enhances the accuracy of statistical tests [84,86]. Unobserved characteristics regarding survey times and individual students were considered according to recommendations [83,84].

The following equations are based on equations (10.4), (10.19), and (10.20) in Chapter 10 of Stock and Watson [84] and Rossi and Villa [87].

y_it = β₁ (Treat_i × Survey2) + β₂ (Treat_i × Survey3) + δ₁Survey2 + δ₂Survey3 + α_i + u_it

(1)

y_it = β₁ (Treat_i × Male_i × Survey2) + β₂ (Treat_i × Male_i × Survey3) + β₃ (Treat_i × Survey2) + β₄ (Treat_i × Survey3)
+ β₅ (Male_i × Survey2) + β₆ (Male_i × Survey3) + δ₁Survey2 + δ₂Survey3 + α_i + u_it

(2)

y_it = β₁ (Treat_i × Grade4_i × Survey2) + β₂ (Treat_i × Grade4_i × Survey3) + β₃ (Treat_i × Survey2) + β₄ (Treat_i × Survey3)
+ β₅ (Grade4_i × Survey2) + β₆ (Grade4_i × Survey3) + δ₁Survey2 + δ₂Survey3 + α_i + u_it

(3)

where the following is true:

-

This equation features student i (i = 1, …, n) and survey time t (t = 1, 2, 3);

-

y_it is the value of the dependent variable for student i at survey time t, such as the scores of knowledge, and intention levels;

-

β₁, β₂, β₃, β₄, β₅, β₆, δ₁, δ₂ are unknown coefficients;

-

Dummy variables are specified for student i as follows:

+

If the student belongs to a treatment school, then Treat_i = 1, and otherwise 0;

+

If the student is male, then Male_i = 1, and otherwise 0;

+

If the student is in Grade 4, then Grade4_i = 1, and otherwise 0;

+

If t = 2, then Survey2 = 1, and otherwise 0; If t = 3, then Survey3 = 1, and otherwise 0.

-

α_i is the fixed effect for student i;

-

u_it is the random error term.

The dataset was unbalanced because the number of participants differed across the three surveys. The fixed effects model works with unbalanced panel data [88]. These models were estimated using the Stata SE 18 software.

5. Results

5.1. Student Answers Before the DRR Education

Table 6. Experience of disaster events χ².

	Mountainous Region			Low-Lying Region			Coastal Region
Disaster type	Treatment school	Control school	χ2(df), p	Treatment school	Control school	χ2(df), p	Treatment school	Control school	χ2(df), p
	(n = 86)	(n = 53)		(n = 102)	(n = 68)		(n = 104)	(n = 128)
Storms			χ2(1) = 2.12, p = 0.146			χ2(1) = 3.19, p = 0.074			χ2(1) = 3.08, p = 0.079
Yes	66 (76.7%)	46 (86.8%)		55 (53.9%)	46 (67.6%)		85 (81.7%)	92 (71.9%)
No	20 (23.3%)	7 (13.2%)		47 (46.1%)	22 (32.4%)		19 (18.3%)	36 (28.1%)
Floods			χ2(1) = 2.26, p = 0.132			χ2(1) = 3.40, p = 0.065			χ2(1) = 3.54, p = 0.060
Yes	28 (32.6%)	11 (20.8%)		66 (64.7%)	53 (77.9%)		24 (23.1%)	44 (34.4%)
No	58 (67.4%)	42 (79.2%)		36 (35.3%)	15 (22.1%)		80 (76.9%)	84 (65.6%)
Landslides			χ2(1) = 0.78, p = 0.376			χ2(1) = 2.74, p = 0.098
Yes	6 (7.0%)	6 (11.3%)		4 (3.9%)	7 (10.3%)
No	80 (93.0%)	47 (88.7%)		98 (96.1%)	61 (89.7%)
Coastal erosion									χ2(1) = 1.04, p = 0.307
Yes							8 (7.7%)	15 (11.7%)
No							96 (92.3%)	113 (88.3%)

summarizes students’ disaster event experiences across the three regions. Students from the treatment and control schools experienced storms, floods, and coastal erosion in the coastal region, as well as storms, floods, and landslides in the other two regions. Storms were the most commonly experienced disasters among students in mountainous (76.7% in treatment schools and 86.8% in control schools) and coastal regions (81.7% in treatment schools and 71.9% in control schools), while floods were the most frequently reported by students in the low-lying region (64.7% in treatment schools and 77.9% in control schools). There were no statistically significant differences in disaster types in the treatment and control pairs across three regions.

Table 7. Initial knowledge levels (Survey 1).

Region		Mountainous Region		Low-Lying Region		Coastal Region
Level of Bloom’s taxonomy	Mean (SD) and t test	Treatment school (n = 86)	Control school (n = 53)	Treatment school (n = 102)	Control school (n = 68)	Treatment school (n = 104)	Control school (n = 128)
Remembering	Mean (SD)	1.60 (1.51)	1.58 (1.51)	1.71 (1.49)	1.68 (1.50)	2.11 (1.38)	1.76 (1.48)
	t test	t(137) = 0.07, p = 0.940		t(168) = 0.13, p = 0.900		t(226) = 1.85, p = 0.066
Understanding	Mean (SD)	0.34 (0.52)	0.51 (0.54)	0.59 (0.62)	0.62 (0.60)	0.67 (0.82)	0.25 (0.45)
	t test	t(137) = −1.86, p = 0.065		t(168) = −0.31, p = 0.759		t(153) = 4.72, p < 0.001 ***
Applying	Mean (SD)	1.27 (0.68)	1.32 (0.80)	1.29 (0.73)	1.35 (0.71)	1.34 (0.83)	0.93 (0.75)
	t test	t(137) = −0.42, p = 0.675		t(168) = −0.52, p = 0.602		t(211) = 3.86 p < 0.001 ***
Analyzing	Mean (SD)	1.93 (1.19)	2.25 (1.09)	1.75 (1.22)	2.19 (1.15)	2.12 (1.05)	1.52 (1.22)
	t test	t(137) = −1.57, p = 0.119		t(168) = −2.39, p = 0.018 *		t(229) = 4.01, p < 0.001 ***
Evaluating	Mean (SD)	0.81 (0.74)	1.21 (0.95)	1.36 (1.13)	1.51 (0.74)	1.47 (0.75)	1.02 (0.90)
	t test	t(91) = −2.57, p = 0.012 *		t(168) = −1.06, p = 0.292		t(230) = 4.06, p < 0.001 ***
Creating	Mean (SD)	0.02 (0.15)	0.04 (0.19)	0.00 (0.00)	0.04 (0.21)	0.05 (0.21)	0.05 (0.23)
	t test	t(137) = −0.49, p = 0.623		t(67) = −1.76, p = 0.083		t(230) = −0.23, p = 0.822

* p < 0.05, *** p < 0.001; SD: Standard Deviation.

presents the initial knowledge levels of the students from Survey 1. For ideal comparisons, students from the treatment and control schools should have similar distributions of initial knowledge levels. The school pairs in the mountainous and low-lying regions revealed five similar means; thus, the relatively homogeneous students were divided into treatment and control schools. However, the treatment school outperformed the control schools in “Understanding”, “Applying”, “Analyzing”, and “Evaluating” levels in the coastal region. “Remembering” and “Creating” levels demonstrated no significant mean differences between the treatment and control pairs in all three regions.

Table 8. Initial in disaster preparedness intention (Survey 1).

Region		Mountainous Region		Low-Lying Region		Coastal Region
Item	Mean, SD and t test	Treatment school (n = 86)	Control school (n = 53)	Treatment school (n = 102)	Control school (n = 68)	Treatment school (n = 104)	Control school (n = 128)
Intention	Mean (SD)	4.40 (0.55)	4.53 (0.52)	4.61 (0.49)	4.59 (0.41)	4.55 (0.54)	4.45 (0.56)
	t test	t(137) = −1.31, p = 0.191		t(168) = 0.21, p = 0.837		t(230) = 1.44, p = 0.152
Attitude toward behavior	Mean (SD)	3.87 (0.74)	4.08 (0.51)	4.08 (0.51)	4.12 (0.58)	4.09 (0.52)	3.98 (0.66)
	t test	t(136) = −1.95, p = 0.054		t(168) = −0.44, p = 0.658		t(230) = 1.48, p = 0.141
Perceived behavioral control	Mean (SD)	3.14 (0.83)	3.64 (0.75)	3.73 (0.93)	3.59 (0.90)	3.42 (0.86)	3.35 (0.86)
	t test	t(137) = −3.57, p = 0.001 **		t(168) = 0.92, p = 0.360		t(230) = 0.68, p = 0.499
Subjective norm	Mean (SD)	4.62 (0.72)	4.67 (0.66)	4.64 (0.69)	4.60 (0.68)	4.50 (0.75)	4.36 (0.86)
	t test	t(137) = −0.39, p = 0.698		t(168) = 0.39, p = 0.699		t(229) = 1.38, p = 0.170

** p < 0.01, SD: Standard Deviation

presents the mean scores of disaster preparedness intention derived from Survey 1. There were no statistically significant differences; however, one exception was “Perceived behavioral control” in the mountainous region.

5.2. Effects on Knowledge and Attitudes

5.2.1. Knowledge Levels

Figure 4 presents the trends in knowledge level scores recorded in the low-lying region. The treatment schools demonstrated increasing trends in the mean scores across the three survey waves, except for the “Understanding” and “Applying” levels. For the control group, the mean scores remained relatively stable, as expected.

We then conducted a rigorous analysis to eliminate the influence of the initial score differences between the treatment and control schools and the external effects that occurred during our education and survey periods.

Table 9. Changes in knowledge levels estimated with fixed effects models.

Region	Specification	Level 1: Remembering		Level 2: Understanding		Level 3: Applying		Level 4: Analyzing		Level 5: Evaluating		Level 6: Creating
Mountainous (N = 392)	Treat × Survey2 Coeff. (Std. err.)	0.467	(0.301)	0.526 **	(0.159)	0.269	(0.146)	0.498 *	(0.249)	0.816 ***	(0.165)	0.011	(0.049)
	Treat × Survey3 Coeff. (Std. err.)	−0.010	(0.325)	0.054	(0.172)	−0.049	(0.157)	0.006	(0.268)	0.162	(0.178)	−0.030	(0.053)
	F test (144,243)	2.22 ***		2.45 ***		2.79 ***		1.91 ***		2.7 ***		1.25
Low-lying (N = 471)	Treat × Survey2 Coeff. (Std. err.)	0.183	(0.232)	0.813 ***	(0.163)	0.481 **	(0.137)	0.089	(0.206)	0.247	(0.171)	0.010	(0.032)
	Treat × Survey3 Coeff. (Std. err.)	0.893 **	(0.266)	0.506 **	(0.187)	0.264	(0.157)	0.613 **	(0.236)	0.407 *	(0.197)	−0.021	(0.037)
	F test (172,294)	2.72 ***		1.91 ***		2.24 ***		2.40 ***		1.79 ***		2.10 ***
Coastal (N = 690)	Treat × Survey2 Coeff. (Std. err.)	−0.051	(0.208)	0.060	(0.117)	0.028	(0.110)	−0.027	(0.165)	−0.141	(0.126)	0.014	(0.041)
	Treat × Survey3 Coeff. (Std. err.)	−0.056	(0.209)	0.119	(0.118)	−0.039	(0.111)	0.041	(0.165)	−0.219	(0.127)	0.044	(0.041)
	F test (241,444)	1.89 ***		2.89 ***		2.98 ***		3.08 ***		2.72 ***		1.69 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (1); Coeff. = Coefficient; Std. err. = Standard error.

presents the results of the fixed effects models. In Survey 2, students in the treatment schools increased their knowledge at the “Understanding” level in the low-lying (coefficient = 0.813, p < 0.001) and mountainous regions (coefficient = 0.526, p < 0.01). The low-lying region demonstrated an improvement in the “Applying” level (coefficient = 0.481, p < 0.01) and the mountainous region demonstrated it in the “Analyzing” (coefficient = 0.498, p < 0.05) and “Evaluating” (coefficient = 0.816, p < 0.001) levels.

In Survey 3, positive changes were recorded relative to Survey 1 in the “Remembering” (coefficient = 0.893, p < 0.01), “Understanding” (coefficient = 0.506, p < 0.01), “Analyzing” (coefficient = 0.613, p < 0.01), and “Evaluating” (coefficient = 0.407, p < 0.05) levels in the treatment school in the low-lying region. In contrast, other regions did not see statistically significant changes at the 5% level.

5.2.2. Attitudes

Figure 5 presents the affective domain indices after the DRR education in the treatment groups across the three regions. The highest positive responses were observed in the low-lying regions. In particular, the “agree” and “partly agree” responses at the “Receiving” level were highest in the low-lying regions (96%), followed by the coastal (92%) and mountainous regions (82.4%). The lowest rates were recorded at different levels for each region: “Valuing” level (89.1%) in the low-lying region, “Organizing” level (72.2%) in the mountainous region, and “Characterizing” level (76.8%) in the coastal region.

5.3. Effects on Disaster Preparedness Intention

For example, Figure 6 presents the trend in scores in the low-lying region. Surveys 1 and 2 indicated that the treatment school indicated a weak increasing trend regarding all four factors of intention in disaster preparedness activities. From Surveys 2 to 3, the mean scores increased in the treatment and control schools in terms of “intention” and “perceived behavioral control” factors. However, the mean scores decreased in both schools regarding the “attention to behavior” and “subjective norm” factors during this period.

Table 10. Change in disaster preparedness intention estimated with fixed effects models.

Region	Specification	Intention		Attitude Toward Behavior		Perceived Behavioral Control		Subjective Norm
Mountainous (N = 392)	Treat × Survey2 Coeff. (Std. err.)	0.099	(0.113)	0.034	(0.144)	0.470 *	(0.191)	−0.182	(0.188)
	Treat × Survey3 Coeff. (Std. err.)	0.098	(0.121)	0.214	(0.155)	0.524 *	(0.206)	−0.200	(0.202)
	F test (144, 243)	2.65 ***		1.48 **		1.52 **		1.51 **
Low-lying (N = 471)	Treat × Survey2 Coeff. (Std. err.)	0.113	(0.076)	0.245 *	(0.112)	0.076	(0.167)	−0.004	(0.116)
	Treat × Survey3 Coeff. (Std. err.)	0.119	(0.087)	−0.020	(0.128)	−0.288	(0.191)	0.122	(0.133)
	F test (172, 294)	2.06 ***		1.48 **		1.41 **		1.75 ***
Coastal (N = 690)	Treat × Survey2 Coeff. (Std. err.)	0.057	(0.073)	−0.096	(0.097)	0.043	(0.135)	−0.080	(0.123)
	Treat × Survey3 Coeff. (Std. err.)	−0.024	(0.074)	−0.080	(0.098)	−0.019	(0.135)	0.008	(0.124)
	F test (241, 444)	2.41 ***		1.42 ***		1.53 ***		1.74 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (1); Coeff. = Coefficient; Std. err. = Standard error.

presents the fixed effects model results for changes in disaster preparedness intention. The statistically significant effects of the DRR education course were minimal. Although Figure 6 presents a slight increase in intention to prepare for a disaster in the treatment school compared with the control school in the low-lying region, this difference was not statistically significant. The other two regions did not demonstrate statistically significant differences in intention. This statistical test recognized the improvement in the “Attitude toward behavior” between Surveys 1 and 2, with coefficient = 0.245 (p < 0.05) in the low-lying region. A similar trend was observed in the mountainous region with “Perceived Behavioral Control” after Survey 2 and Survey 3, with coefficients of 0.470 (p < 0.05) and 0.524 (p < 0.05), respectively.

The responses to individual measurement questions were also examined. A measurement question of the perceived behavioral control factor, “I can move to a safe place alone without an adult being with me”, indicated a positive effect of the education program.

Table 11. Ability to protect themselves without adult appearance with fixed effects models.

I Can Move to a Safe Place on My Own Without an Adult Being with Me
Specification	Mountainous Region (N = 392)		Low-Lying Region (N = 471)		Coastal Region (N = 690)
Treat × Survey2 Coeff. (Std. err.)	0.726 *	(0.356)	−0.362	(0.308)	0.573 *	(0.262)
Treat × Survey3 Coeff. (Std. err.)	0.203	(0.384)	−0.293	(0.353)	0.329	(0.264)
	F test (144,243) = 1.87 ***		F test (172,294) = 1.87 ***		F test (241,444) = 1.64 ***

* p < 0.05, *** p < 0.001; Coeff. = Coefficient; Std. err. = Standard error.

presents that this effect was statistically significant in the mountainous (coefficient = 0.726, p < 0.05) and coastal regions (coefficient = 0.573, p < 0.05).

5.4. Gender and Grade Analysis in Knowledge, Attitude and Disaster Preparedness Intention

5.4.1. Gender Analysis

Gender analysis was conducted using fixed effects models to determine whether there were any differences between male and female students (see

Table A2. Gender analysis using fixed effects models in knowledge.

Region	Specification	Level 1: Remembering		Level 2: Understanding		Level 3: Applying		Level 4: Analyzing		Level 5: Evaluate		Level 6: Creating
Mountainous (N = 392)	Treat × Survey2 (Coeff. (Std. err.))	0.558	(0.379)	0.643 **	(0.201)	0.470 *	(0.183)	0.464	(0.314)	0.641 **	(0.208)	0.023	(0.062)
	Treat × Survey3 (Coeff. (Std. err.))	−0.230	(0.387)	0.049	(0.205)	−0.026	(0.186)	0.109	(0.32)	0.093	(0.212)	0.029	(0.063)
	Treat × Male × Survey2 (Coeff. (Std. err.))	−0.142	(0.39)	−0.195	(0.207)	−0.337	(0.188)	0.052	(0.323)	0.296	(0.214)	−0.023	(0.064)
	Treat × Male × Survey3 (Coeff. (Std. err.))	0.391	(0.373)	0.010	(0.198)	−0.041	(0.180)	−0.185	(0.309)	0.122	(0.204)	−0.105	(0.061)
	F test (144,243)	2.19 ***		2.39 ***		2.80 ***		1.87 ***		2.71 ***		1.23
Low-lying (N = 471)	Treat × Survey2 (Coeff. (Std. err.))	0.228	(0.291)	0.979 ***	(0.204)	0.773 ***	(0.169)	0.371	(0.257)	0.329	(0.214)	0.023	(0.041)
	Treat × Survey3 (Coeff. (Std. err.))	1.124 ***	(0.324)	0.596 **	(0.228)	0.478 *	(0.189)	0.844 **	(0.287)	0.625 **	(0.239)	−0.030	(0.045)
	Treat × Male × Survey2 (Coeff. (Std. err.))	−0.077	(0.298)	−0.281	(0.209)	−0.496 **	(0.173)	−0.480	(0.264)	−0.139	(0.220)	−0.023	(0.042)
	Treat × Male × Survey3 (Coeff. (Std. err.))	−0.406	(0.324)	−0.153	(0.227)	−0.368	(0.188)	−0.398	(0.286)	−0.382	(0.239)	0.017	(0.045)
	F test (172,294)	2.72 ***		1.87 ***		2.07 ***		2.39 ***		1.77 ***		2.09 ***
Coastal (N = 690)	Treat × Survey2 (Coeff. (Std. err.))	−0.132	(0.261)	0.119	(0.147)	0.063	(0.138)	−0.027	(0.206)	−0.052	(0.158)	0.035	(0.051)
	Treat × Survey3 (Coeff. (Std. err.))	−0.148	(0.261)	0.120	(0.147)	−0.069	(0.138)	−0.023	(0.206)	−0.181	(0.158)	0.063	(0.051)
	Treat × Male × Survey2 (Coeff. (Std. err.))	0.160	(0.310)	−0.117	(0.175)	−0.070	(0.164)	0	(0.245)	−0.176	(0.188)	−0.041	(0.061)
	Treat × Male × Survey3 (Coeff. (Std. err.))	0.183	(0.313)	0.001	(0.176)	0.062	(0.166)	0.128	(0.247)	−0.074	(0.190)	−0.038	(0.062)
	F test (241,444)	1.88 ***		2.83 ***		2.90 ***		3.07 ***		2.67 ***		1.68 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (2); Coeff. = Coefficient; Std. err. = Standard error.

). Two interaction variables (Treat × Male × Survey2 and Treat × Male × Survey3) were added to the original models. These variables represent male students at the treatment school in Surveys 2 and 3, respectively. Among the six cognitive levels, gender differences were only observed at the “Applying” level in the low-lying region. The results in Table 9 show that a pooled sample of the genders in treatment schools in low-lying regions on average improved significantly (coefficient = 0.481, p < 0.01).

Table 12. Gender analysis using fixed effects models in knowledge level 3: applying (subset of Table A2).

Level 3: Applying
Specification	Mountainous (N = 392)		Low-Lying (N = 471)		Coastal (N = 690)
Treat × Survey2 Coeff. (Std. err.)	0.470 *	(0.183)	0.773 ***	(0.169)	0.063	(0.138)
Treat × Survey3 Coeff. (Std. err.)	−0.026	(0.186)	0.478 *	(0.189)	−0.069	(0.138)
Treat × Male × Survey2 Coeff. (Std. err.)	−0.337	(0.188)	−0.496 **	(0.173)	−0.070	(0.164)
Treat × Male × Survey3 Coeff. (Std. err.)	−0.041	(0.180)	−0.368	(0.188)	0.062	(0.166)
	F test (144,243) = 2.80 ***		F test (172,294) = 2.07 ***		F test (241,444) = 2.90 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (2); Coeff. = Coefficient; Std. err. = Standard error.

further demonstrates that, in the low-lying region, female students showed significant positive gains in Survey 2 (coefficient = 0.773, p < 0.001) and outperformed male students (coefficient = −0.496, p < 0.01). However, no significant gender differences were found in treatment schools in the remaining two regions.

The results of gender differences in Survey 2 were also analyzed by comparing attitude levels across different regions, as shown in

Table A3. Gender analysis using Chi-square test in attitude.

	Mountainous Region			Low-Lying Region			Coastal Region
	Male	Female	χ2(df), p	Male	Female	χ2(df), p	Male	Female	χ2(df), p
5. Characterizing	n = 44	n = 29	χ2(3) = 0.41, p = 0.939	n = 58	n = 43	χ2(3) = 0.26, p = 0.967	n = 49	n = 50	χ2(4) = 3.37, p = 0.497
Disagree	3 (6.8%)	2 (6.9%)		1 (1.7%)	1 (2.3%)		1 (2%)	0 (0%)
Partly disagree	0 (0%)	0 (0%)		0 (0%)	0 (0%)		4 (8.2%)	1 (2%)
Neutral	11 (25%)	7 (24.1%)		4 (6.9%)	2 (4.7%)		7 (14.3%)	10 (20%)
Partly agree	26 (59.1%)	16 (55.2%)		8 (13.8%)	6 (14%)		13 (26.5%)	14 (28%)
Agree	44 (100%)	29 (100%)		45 (77.6%)	34 (79.1%)		24 (49%)	25 (50%)
4. Organizing	n = 44	n = 28	χ2(4) = 3.47, p = 0.482	n = 58	n = 42	χ2(4) = 2.4, p = 0.662	n = 50	n = 50	χ2(3) = 1.92, p = 0.588
Disagree	2 (4.5%)	4 (14.3%)		2 (3.4%)	0 (0%)		0 (0%)	1 (2%)
Partly disagree	3 (6.8%)	2 (7.1%)		1 (1.7%)	0 (0%)		0 (0%)	0 (0%)
Neutral	2 (4.5%)	4 (14.3%)		3 (5.2%)	3 (7.1%)		5 (10%)	8 (16%)
Partly agree	15 (36.6%)	8 (26.7%)		16 (27.6%)	7 (16.3%)		16 (32.7%)	15 (30%)
Agree	12 (27.3%)	9 (32.1%)		41 (70.7%)	30 (71.4%)		32 (64%)	30 (60%)
3. Valuing	n = 41	n = 30	χ2(3) = 3.80, p = 0.284	n = 58	n = 43	χ2(3) = 9.98, p = 0.019 *	n = 49	n = 50	χ2(3) = 0.92, p = 0.822
Disagree	0 (0%)	0 (0%)		0 (0%)	0 (0%)		0 (0%)	0 (0%)
Partly disagree	1 (2.4%)	3 (10%)		2 (3.4%)	1 (2.3%)		1 (2%)	2 (4%)
Neutral	4 (9.8%)	6 (20%)		8 (13.8%)	0 (0%)		8 (16.3%)	11 (22%)
Partly agree	15 (36.6%)	8 (26.7%)		16 (27.6%)	7 (16.3%)		16 (32.7%)	15 (30%)
Agree	21 (51.2%)	13 (43.3%)		32 (55.2%)	35 (81.4%)		24 (49%)	22 (44%)
2. Response	n = 86	n = 60	χ2(4) = 4.25, p = 0.373	n = 116	n = 86	χ2(4) = 1.73, p = 0.785	n = 97	n = 100	χ2(4) = 1.35, p = 0.852
Disagree	4 (4.7%)	0 (0%)		1 (0.9%)	0 (0%)		1 (1%)	0 (0%)
Partly disagree	3 (3.5%)	2 (3.3%)		1 (0.9%)	1 (1.2%)		1 (1%)	2 (2%)
Neutral	13 (15.1%)	8 (13.3%)		6 (5.2%)	3 (3.5%)		16 (16.5%)	17 (17%)
Partly agree	19 (22.1%)	10 (16.7%)		29 (25%)	18 (20.9%)		28 (28.9%)	28 (28%)
Agree	47 (54.7%)	40 (66.7%)		79 (68.1%)	64 (74.4%)		51 (52.6%)	53 (53%)
1. Receiving	n = 44	n = 30	χ2(4) = 4.05, p = 0.399	n = 58	n = 43	χ2(4) = 5.41, p = 0.247	n = 50	n = 50	χ2(2) = 1.74, p = 0.418
Disagree	2 (4.5%)	0 (0%)		1 (1.7%)	0 (0%)		0 (0%)	0 (0%)
Partly disagree	4 (9.1%)	1 (3.3%)		0 (0%)	1 (2.3%)		0 (0%)	0 (0%)
Neutral	2 (4.5%)	4 (13.3%)		1 (1.7%)	0 (0%)		3 (6%)	5 (10%)
Partly agree	13 (29.5%)	8 (26.7%)		14 (24.1%)	6 (14%)		22 (44%)	16 (32%)
Agree	23 (52.3%)	17 (56.7%)		41 (70.7%)	36 (83.7%)		25 (50%)	29 (58%)

* p < 0.05.

. Based on the number and percentage of responses from males and females using a 5-point Likert scale, a Chi-square test was conducted to examine gender differences. No significant differences were found at any attitude level in the coastal and mountainous regions, as well as at most levels in the low-lying region. A gender difference was found only at Level 3, “Valuing”, in the low-lying region (χ²(3) = 9.98, p = 0.019) (see

Table 13. Gender analysis using Chi-square test in attitudes in level 3: valuing (subset of Table A3).

Low-Lying Region
Level 3: Valuing	Male	Female	χ2(df), p
	n = 58	n = 43	χ2(3) = 9.98, p = 0.019 *
Disagree	0 (0.0%)	0 (0.0%)
z-value	-	-
Partly disagree	2 (3.4%)	1 (2.3%)
z-value	z = 0.21	z = −0.25
Neutral	8 (13.8%)	0 (0.0%)
z-value	z = 1.59	z = −1.85
Partly agree	16 (27.6%)	7 (16.3%)
z-value	z = 0.77	z = −0.89
Agree	32 (55.2%)	35 (81.4%)
z-value	z = −1.04	z = 1.21

* p < 0.05.

). To identify which response option contributed to the difference, standardized residuals (z-values) were calculated and compared against the critical value of ±1.96, as noted by Filed [89] (pp. 743–744), and Sharpe [90]. Although none of the z-values exceeded this threshold, the most notable contribution to the difference came from the “Neutral” response: male students had a higher-than-expected frequency (z = 1.59), while female students had a lower-than-expected frequency (z = −1.85).

A comparison of male and female students’ disaster preparedness intention was also analyzed (see

Table A4. Gender analysis using fixed effects models in disaster preparedness intention.

Region	Specification	Intention		Attitude Toward Behavior		Perceived Behavioral Control		Subjective Norm
Mountainous (N = 392)	Treat × Survey2 (Coeff. (Std. err.))	0.059	(0.142)	−0.026	(0.182)	0.304	(0.240)	−0.416	(0.235)
	Treat × Survey3 (Coeff. (Std. err.))	0.150	(0.145)	0.201	(0.185)	0.331	(0.244)	−0.131	(0.240)
	Treat × Male × Survey2 (Coeff. (Std. err.))	0.064	(0.146)	0.101	(0.187)	0.332	(0.246)	0.386	(0.242)
	Treat × Male × Survey3 (Coeff. (Std. err.))	−0.093	(0.140)	0.022	(0.179)	0.400	(0.236)	−0.124	(0.231)
	F test (144,243)	2.64 ***		1.47 **		1.53 **		1.52 **
Low-lying (N = 471)	Treat × Survey2 (Coeff. (Std. err.))	0.169	(0.095)	0.274	(0.140)	0.171	(0.209)	0.101	(0.145)
	Treat × Survey3 (Coeff. (Std. err.))	0.092	(0.106)	−0.068	(0.157)	−0.371	(0.233)	0.262	(0.161)
	Treat × Male × Survey2 (Coeff. (Std. err.))	−0.096	(0.097)	−0.050	(0.144)	−0.162	(0.214)	−0.179	(0.148)
	Treat × Male × Survey3 (Coeff. (Std. err.))	0.049	(0.106)	0.086	(0.156)	0.150	(0.233)	−0.242	(0.161)
	F test (172,294)	2.05 ***		1.47 **		1.40 **		1.76 ***
Coastal (N = 690)	Treat × Survey2 (Coeff. (Std. err.))	−0.030	(0.092)	−0.060	(0.122)	0.091	(0.169)	−0.119	(0.154)
	Treat × Survey3 (Coeff. (Std. err.))	−0.106	(0.092)	−0.164	(0.122)	−0.058	(0.169)	−0.103	(0.154)
	Treat × Male × Survey2 (Coeff. (Std. err.))	0.171	(0.109)	−0.073	(0.145)	−0.096	(0.201)	0.075	(0.184)
	Treat × Male × Survey3 (Coeff. (Std. err.))	0.164	(0.110)	0.173	(0.146)	0.080	(0.202)	0.223	(0.185)
	F test (241,444)	2.42 ***		1.41 ***		1.53 ***		1.74 ***

** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (2); Coeff. = Coefficient; Std. err. = Standard error.

). Equation (2) was applied and calculated for all three regions. The results show no significant difference in the behavioral intention between male and female students across all three regions, as the p-values are greater than 0.05.

5.4.2. Grade Analysis

On average, a pooled sample of the two graders received positive impacts from the course, as shown in Table 9. When comparing grade differences, knowledge levels varied across the three regions (see

Table A5. Grade analysis using fixed effects models in knowledge.

Region	Specification	Level 1: Remembering		Level 2: Understanding		Level 3: Applying		Level 4: Analyzing		Level 5: Evaluate		Level 6: Creating
Mountainous (N = 392)	Treat × Survey2 (Coeff. (Std. err.))	0.505	(0.345)	0.806 ***	(0.179)	0.395 *	(0.166)	0.401	(0.285)	0.781 ***	(0.186)	0.021	(0.057)
	Treat × Survey3 (Coeff. (Std. err.))	0.175	(0.369)	0.219	(0.192)	−0.072	(0.177)	−0.133	(0.305)	−0.092	(0.199)	−0.008	(0.061)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	−0.065	(0.388)	−0.643 **	(0.202)	−0.305	(0.186)	0.212	(0.32)	0.047	(0.209)	−0.022	(0.064)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−0.394	(0.371)	−0.346	(0.193)	0.052	(0.178)	0.294	(0.307)	0.541 **	(0.200)	−0.046	(0.061)
	F test (144,243)	2.19 ***		2.58 ***		2.82 ***		1.90 ***		2.80 ***		1.24
Low-lying (N = 471)	Treat × Survey2 (Coeff. (Std. err.))	0.526	(0.276)	0.666 ***	(0.199)	0.338 *	(0.167)	0.063	(0.253)	0.578 **	(0.208)	0	(0.040)
	Treat × Survey3 (Coeff. (Std. err.))	1.659 ***	(0.310)	0.607 **	(0.223)	0.168	(0.188)	0.411	(0.283)	0.532 *	(0.233)	−0.029	(0.045)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	−0.613 *	(0.287)	0.264	(0.207)	0.257	(0.174)	0.046	(0.262)	−0.593 **	(0.216)	0.017	(0.041)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−1.402 ***	(0.312)	−0.190	(0.225)	0.174	(0.189)	0.371	(0.285)	−0.221	(0.235)	0.016	(0.045)
	F test (172,294)	2.81 ***		1.90 ***		2.18 ***		2.40 ***		1.83 ***		2.08 ***
Coastal (N = 690)	Treat × Survey2 (Coeff. (Std. err.))	−0.543 *	(0.275)	0.232	(0.156)	−0.109	(0.147)	−0.330	(0.218)	−0.348 *	(0.168)	0.063	(0.055)
	Treat × Survey3 (Coeff. (Std. err.))	−0.261	(0.274)	0.153	(0.155)	−0.165	(0.146)	−0.152	(0.217)	−0.291	(0.167)	0.019	(0.054)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	0.845 **	(0.312)	−0.295	(0.177)	0.235	(0.166)	0.521 *	(0.248)	0.356	(0.190)	−0.083	(0.062)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	0.351	(0.313)	−0.056	(0.177)	0.220	(0.167)	0.335	(0.248)	0.123	(0.191)	0.046	(0.062)
	F test (241,444)	1.88 ***		2.90 ***		2.97 ***		3.10 ***		2.72 ***		1.70 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (3); Coeff. = Coefficient; Std. err. = Standard error.

). Some significant parts of the results are shown in

Table 14. Grade analysis using fixed effects models at various knowledge levels (subset of Table A5).

Region	Specification	Level 2: Understanding		Level 5: Evaluating
Mountainous (N = 392)	Treat × Survey2 (Coeff. (Std. err.))	0.806 ***	(0.179)	0.781 ***	(0.186)
	Treat × Survey3 (Coeff. (Std. err.))	0.219	(0.192)	−0.092	(0.199)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	−0.643 **	(0.202)	0.047	(0.209)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−0.346	(0.193)	0.541 **	(0.200)
	F test (144,243)	2.58 ***		2.80 ***
Low-lying (N = 471)	Specification	Level 1: Remembering		Level 5: Evaluating
	Treat × Survey2 (Coeff. (Std. err.))	0.526	(0.276)	0.578 **	(0.208)
	Treat × Survey3 (Coeff. (Std. err.))	1.659 ***	(0.310)	0.532 *	(0.233)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	−0.613 *	(0.287)	−0.593 **	(0.216)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−1.402 ***	(0.312)	−0.221	(0.235)
	F test (172,294)	2.81 ***		1.83 ***
Coastal (N = 690)	Specification	Level 1: Remembering		Level 4: Analyzing
	Treat × Survey2 (Coeff. (Std. err.))	−0.543 *	(0.275)	−0.33	(0.218)
	Treat × Survey3 (Coeff. (Std. err.))	−0.261	(0.274)	−0.152	(0.217)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	0.845 **	(0.312)	0.521 *	(0.248)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	0.351	(0.313)	0.335	(0.248)
	F test (241,444)	1.88 ***		3.10 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (3); Coeff. = Coefficient; Std. err. = Standard error.

. In the mountainous region, a significant positive change was observed in Grade 3 after Survey 2, particularly at the “Understanding” and “Evaluating” levels, through the interaction variable Treat × Survey2. In Survey 2, Grade 4 students performed worse than Grade 3 at the “Understanding” level (coefficient = −0.643, p < 0.01), while an opposite trend was recorded in Survey 3 at the “Evaluating” level (coefficient = 0.541, p < 0.01).

In the low-lying region, Grade 3 students experienced positive impacts at the “Remembering” level in Survey 3 and at the “Evaluating” level across both surveys. Grade 4 students performed worse than Grade 3 at the “Remembering” level in Survey 2 (coefficient = −0.613, p < 0.05) and Survey 3 (coefficient = −1.402, p < 0.001), as well as at the “Evaluating” level in Survey 2 (coefficient = −0.593, p < 0.01).

An opposite pattern was found in the coastal region, where Grade 4 students outperformed Grade 3. Specifically, after Survey 2, Grade 4 students scored higher at the “Remembering” level (coefficient = 0.845, p < 0.01) and “Analyzing” level (coefficient = 0.521, p < 0.05), indicating that older students performed better than their younger ones.

Concerning attitude levels, grade differences were also analyzed across the regions (See the full results in

Table A6. Grade analysis using Chi-square test in attitude.

	Mountainous Region			Low-Lying Region			Coastal Region
	Grade 3	Grade 4	χ2(df), p	Grade 3	Grade 4	χ2(df), p	Grade 3	Grade 4	χ2(df), p
5. Characterizing	n = 42	n = 31	χ2(3) = 4.02, p = 0.259	n = 44	n = 57	χ2(3) = 3.18, p = 0.364	n = 42	n = 57	χ2(4) = 10.80, p = 0.029 *
Disagree	3 (7.1%)	2 (6.5%)		2 (4.5%)	0 (0%)		0 (0%)	1 (1.8%)
Partly disagree	0 (0%)	0 (0%)		0 (0%)	0 (0%)		0 (0%)	1 (1.8%)
Neutral	8 (19%)	10 (32.3%)		2 (4.5%)	0 (0%)		8 (19%)	9 (15.8%)
Partly agree	28 (66.7%)	14 (45.2%)		7 (15.9%)	7 (12.3%)		7 (16.7%)	20 (35.1%)
Agree	42 (100%)	31 (100%)		33 (75%)	46 (80.7%)		27 (64.3%)	22 (38.6%)
4. Organizing	n = 42	n = 30	χ2(4) = 5.17, p = 0.270	n = 44	n = 56	χ2(4) = 3.62, p = 0.46	n = 42	n = 58	χ2(3) = 4.07, p = 0.254
Disagree	6 (14.3%)	0 (0%)		0 (0%)	2 (3.6%)		0 (0%)	1 (2.4%)
Partly disagree	2 (4.8%)	3 (10%)		0 (0%)	1 (1.8%)		0 (0%)	0 (0%)
Neutral	5 (11.9%)	4 (13.3%)		4 (9.1%)	2 (3.6%)		10 (17.2%)	3 (7.1%)
Partly agree	10 (25.6%)	13 (40.6%)		8 (18.2%)	15 (26.3%)		12 (28.6%)	19 (33.3%)
Agree	17 (40.5%)	14 (46.7%)		31 (70.5%)	40 (71.4%)		33 (56.9%)	29 (69%)
3. Valuing	n = 39	n = 32	χ2(3) = 2.60, p = 0.456	n = 39	n = 32	χ2(3) = 5.39, p = 0.145	n = 42	n = 57	χ2(3) = 3.70, p = 0.295
Disagree	0 (0%)	0 (0%)		0 (0%)	0 (0%)		0 (0%)	3 (5.3%)
Partly disagree	3 (7.7%)	1 (3.1%)		1 (2.3%)	2 (3.5%)		0 (0%)	3 (5.3%)
Neutral	5 (12.8%)	5 (15.6%)		1 (2.3%)	2 (3.5%)		7 (16.7%)	12 (21.1%)
Partly agree	10 (25.6%)	13 (40.6%)		8 (18.2%)	15 (26.3%)		12 (28.6%)	19 (33.3%)
Agree	21 (53.8%)	13 (40.6%)		34 (77.3%)	33 (57.9%)		23 (54.8%)	23 (40.4%)
2. Response	n = 83	n = 63	χ2(4) = 13, p = 0.011 *	n = 88	n = 114	χ2(4) = 7.36, p = 0.118	n = 84	n = 113	χ2(4) = 4.22, p = 0.376
Disagree	2 (2.4%)	2 (3.2%)		1 (1.1%)	0 (0%)		0 (0%)	1 (0.9%)
Partly disagree	1 (1.2%)	4 (6.3%)		1 (1.1%)	0 (0%)		0 (0%)	1 (0.9%)
Neutral	7 (8.4%)	16 (25.4%)		7 (8%)	2 (1.8%)		13 (15.5%)	20 (17.7%)
Partly agree	19 (22.9%)	15 (23.8%)		16 (18.2%)	31 (27.2%)		22 (26.2%)	34 (30.1%)
Agree	54 (65.1%)	26 (41.3%)		63 (71.6%)	80 (70.2%)		49 (58.3%)	55 (48.7%)
1. Receiving	n = 42	n = 32	χ2 (4) = 0.51, p = 0.973	n = 44	n = 57	χ2(4) = 12.40, p = 0.014 *	n = 42	n = 58	χ2(2) = 7.31, p = 0.026 *
Disagree	1 (2.4%)	1 (3.1%)		1 (2.3%)	0 (0%)		0 (0%)	0 (0%)
Partly disagree	3 (7.1%)	2 (6.3%)		1 (2.3%)	0 (0%)		0 (0%)	0 (0%)
Neutral	3 (7.1%)	2 (6.3%)		2 (4.5%)	0 (0%)		0 (0%)	8 (13.8%)
Partly agree	11 (26.2%)	10 (31.3%)		3 (6.8%)	17 (29.8%)		15 (35.7%)	23 (39.7%)
Agree	24 (57.1%)	16 (50%)		37 (84.1%)	40 (70.2%)		27 (64.3%)	27 (46.6%)

* p < 0.05.

and part of them in

Table 15. Grade analysis using Chi-square test at various attitude levels (subset of Table A6).

Mountainous Region	Grade 3	Grade 4	χ2(df), p	Low-Lying Region	Grade 3	Grade 4	χ2(df), p
2. Response	n = 83	n = 63	χ2(4) = 13.00, p = 0.011 *	1. Receiving	n = 44	n = 57	χ2(4) = 12.40, p = 0.014 *
Disagree	2 (2.4%)	2 (3.2%)		Disagree	1 (2.3%)	0 (0.0%)
z-value	z = −0.18	z = 0.21		z-value	z = 0.86	z = −0.75
Partly disagree	1 (1.2%)	4 (6.3%)		Partly disagree	1 (2.3%)	0 (0.0%)
z-value	z = −1.09	z = 1.25		z-value	z = 0.86	z = −0.75
Neutral	7 (8.4%)	16 (25.4%)		Neutral	2 (4.5%)	0 (0.0%)
z-value	z = −1.68	z = 1.93		z-value	z = 1.21	z = −1.06
Partly agree	19 (22.9%)	15 (23.8%)		Partly agree	3 (6.8%)	17 (29.8%)
z-value	z = −0.07	z = 0.09		z-value	z = −1.94	z = 1.70
Agree	54 (65.1%)	26 (41.3%)		Agree	37 (84.1%)	40 (70.2%)
z-value	z = 1.26	z = −1.45		z-value	z = 0.6	z = −0.52
Coastal region	Grade 3	Grade 4	χ2(df), p	Coastal region	Grade 3	Grade 4	χ2(df), p
5. Characterizing	n = 42	n = 57	χ2(4) = 10.80, p = 0.029 *	1. Receiving	n = 42	n = 58	χ2(2) = 7.31, p = 0.026 *
Disagree	0 (0.0%)	1 (1.8%)		Disagree	0 (0.0%)	0 (0.0%)
z-value	z = −0.65	z = 0.56		z-value	-	-
Partly disagree	0 (0.0%)	1 (1.8%)		Partly disagree	0 (0.0%)	0 (0.0%)
z-value	z = −1.46	z = 1.25		z-value	-	-
Neutral	8 (19%)	9 (15.8%)		Neutral	0 (0.0%)	8 (13.8%)
z-value	z = 0.29	z = −0.25		z-value	z = −1.83	z = 1.56
Partly agree	7 (16.7%)	20 (35.1%)		Partly agree	15 (35.7%)	23 (39.7%)
z-value	z = −1.32	z = 1.13		z-value	z = −0.24	z = 0.2
Agree	27 (64.3%)	22 (38.6%)		Agree	27 (64.3%)	27 (46.6%)
z-value	z = 1.36	z = −1.17		z-value	z = 0.91	z = −0.77

* p < 0.05.

). The results of the Chi-square test indicate significant differences at the “Receiving” level (χ²(4) = 12.40, p = 0.014) among students in the low-lying region, and at the “Responding” level (χ²(4) = 13.00, p = 0.011) in the mountainous region. In the coastal region, significant differences were recorded at both the “Receiving” (χ²(2) = 7.31, p = 0.026) and “Characterizing” (χ²(4) = 10.80, p = 0.029) levels.

When considering standardized residuals (z-values), the findings suggest that older students tend to choose certain responses more frequently than younger ones. For instance, at the “Receiving” level, Grade 4 students in the low-lying region tend to respond “Partly agree” more than Grade 3 students (Grade 4: z = 1.70; Grade 3: z = −1.94). A similar pattern was found in the coastal region at the “Neutral” level (Grade 4: z = 1.56; Grade 3: z = −1.83). However, since these z-values do not exceed the threshold of ±1.96, the grade-level differences in response patterns are not considered statistically significant.

Table 16. Grade analysis using fixed effects models in disaster preparedness intention.

Region	Specification	Intention		Attitude Toward Behavior		Perceived Behavioral Control		Subjective Norm
Mountainous (N = 392)	Treat × Survey2 (Coeff. (Std. err.))	−0.090	(0.126)	0.037	(0.164)	0.715 ***	(0.217)	−0.402	(0.213)
	Treat × Survey3 (Coeff. (Std. err.))	−0.095	(0.135)	0.341	(0.176)	0.733 **	(0.233)	−0.250	(0.228)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	0.422 **	(0.142)	0.011	(0.184)	−0.496 *	(0.244)	0.517 *	(0.240)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	0.407 **	(0.136)	−0.272	(0.177)	−0.374	(0.234)	0.101	(0.230)
	F test (144,243)	2.82 ***		1.49 **		1.60 **		1.49 **
Low-lying (N = 471)	Treat × Survey2 (Coeff. (Std. err.))	0.072	(0.093)	0.136	(0.137)	−0.121	(0.202)	0.061	(0.142)
	Treat × Survey3 (Coeff. (Std. err.))	0.140	(0.104)	−0.007	(0.154)	−0.648 **	(0.227)	0.211	(0.159)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	0.074	(0.097)	0.194	(0.142)	0.352	(0.210)	−0.117	(0.147)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−0.039	(0.105)	−0.028	(0.155)	0.658 **	(0.229)	−0.161	(0.160)
	F test (172,294)	2.03 ***		1.40 **		1.47 **		1.75 ***
Coastal (N = 690)	Treat × Survey2 (Coeff. (Std. err.))	0.076	(0.098)	0.074	(0.129)	−0.058	(0.179)	0.052	(0.164)
	Treat × Survey3 (Coeff. (Std. err.))	−0.012	(0.097)	−0.079	(0.128)	−0.155	(0.178)	0.011	(0.163)
	Treat × Grade4 × Survey2 (Coeff. (Std. err.))	−0.034	(0.111)	−0.291 *	(0.146)	0.173	(0.204)	−0.227	(0.186)
	Treat × Grade4 × Survey3 (Coeff. (Std. err.))	−0.019	(0.111)	0.002	(0.147)	0.238	(0.204)	−0.003	(0.187)
	F test (241,444)	2.39 ***		1.43 ***		1.53 ***		1.74 ***

* p < 0.05, ** p < 0.01, *** p < 0.001 are marked for treatment dummy variables. Variables were explained in Equation (3); Coeff. = Coefficient; Std. err. = Standard error.

presents the results of disaster preparedness intention by grade level. The scores of various factors differ across the three regions. In the mountainous region, Grade 4 students scored higher in “Intention” after Survey 2 (coefficient = 0.422, p < 0.01) and Survey 3 (coefficient = 0.407, p < 0.01), and in “Subjective Norm” after Survey 2 (coefficient = 0.517, p < 0.05). A similar trend was observed in “Perceived Behavioral Control” in the low-lying region (coefficient = 0.658, p < 0.01). In contrast, Grade 3 students scored higher than Grade 4 students in “Attitude Toward Behavior” in the coastal region during Survey 2 (coefficient = −0.291, p < 0.05), observed in “Perceived Behavioral Control” in the mountainous region (coefficient = −0.496, p < 0.05).

5.5. Satisfaction and Motivation to Participate in the Course

According to Figure 7, over 95% of the students in the treatment groups were satisfied with the DRR program across the three regions, including 100% of the students in the low-lying region. Regarding teaching methods, the highest satisfaction was among students in low-lying regions (93.3%), followed by those in coastal (85.7%) and mountainous regions (73.4%). Specifically, “Look at the pictures and guess the type of disasters” was the most liked activity in the low-lying (96.8%) and mountainous regions (80.6%), whereas the game “Should or Should Not” was the most liked activity in the coastal region (94.9%). The least liked activity was “Placing an X far or close to yourself shows more or less fear of disasters”, regardless of the region. Regarding the new methods introduced in this study, watching videos of Japanese students and sharing returnees’ experiences were liked by approximately 85%, with the highest percentage being observed among students in low-lying areas (over 94%) (see Figure 8).

6. Discussion

6.1. Initial Knowledge Levels and Intention to Preparedness

The data in Table 6 show that the experience of common disaster events is similar between students in the treatment and control schools within each region. The Chi-square test results are not statistically significant (p > 0.05). Specifically, storms are the most commonly experienced disasters among students in mountainous and coastal regions, while floods are more frequent in low-lying regions. This can be explained by the similar level of socioeconomic development and disaster exposure across these regions. Overall, storms and floods are the most frequently experienced disasters among students in the study area.

In Survey 1, initial differences were observed at certain levels of the cognitive domain, as shown in Table 7, even though the groups were selected based on similar socioeconomic development and exposure to disasters. These differences may be influenced by unmeasured contextual factors, such as television, parental sharing, and variation in teachers’ emphasis on disaster-related content in textbooks (e.g., Vietnamese, Nature and Science, History and Geography), as noted by Mai et al. [91]. These factors could have contributed to differences in some levels of students’ initial knowledge and intentions. These results suggest that socioeconomic and environmental characteristics and disaster exposure may not fully account for all levels of cognitive development and intention, particularly in diverse populations.

6.2. Knowledge and Attitude Levels

The impact of the DRR course on students’ knowledge levels was diverse across knowledge levels and across the three regions (Table 9). The low-lying regions indicated many statistically significant improvements immediately after education (Survey 2) and three months later (Survey 3). Students in low-lying regions appear to have a better capacity to absorb knowledge than those in other regions, considering their higher satisfaction with the lessons and teaching methods observed in low-lying regions. Moreover, all levels improved after Survey 2, especially the “Understanding” level (Figure 4). A possible reason is that students came to recognize that disasters were caused by natural factors and human actions, whereas before the survey, they believed disasters were caused only by either nature or humans, not both.

Since the mountainous region demonstrated statistically significant improvement only in Survey 2, the effects of this DRR education were limited to the short term after the education. This may be because many students in mountainous regions are ethnic minorities with limited Vietnamese proficiency [92,93], which can make it difficult for them to retain information over the long term.

A possible reason for the absence of significant knowledge improvements in coastal regions is scheduling. In the low-lying and mountainous regions, students had more frequent and continuous classes compared with those in the coastal region (see Section 4.3). This may affect how well they absorb the content taught in class.

Regarding attitudes based on the affective domain in Figure 5, the receiving level is the highest across all regions. This indicates that the pictures and videos used in lessons are visually accessible and easy to follow. However, each group exhibits different small challenges, including difficulty in explaining content (“Valuing” level) in the low-lying region, handling diverse situations (“Organizing” level) in the mountainous region, and sharing information with others (“Characterizing” level) in the coastal region. These results may suggest the need for further studies to enhance students’ abilities to share, interact with those around them, and respond to various situations through drills, especially by strengthening connections with families and community levels [94].

6.3. Disaster Preparedness Intention

Although this DRR education program was designed to improve self-protective capabilities, its effects on students’ disaster preparedness intentions were limited. Table 10 indicates no statistically significant difference in the intentions between the treatment and control schools after DDR education in any regions. Only “Attitudes toward behavior” among the treatment students in the low-lying region and “Perceived behavioral control” in the mountainous region demonstrated a positive improvement compared with the control students. Particularly, most measurement items regarding “Attitudes toward behavior” in the low-lying region, except for “Helping my parents arrange things in the house before disasters occur is useful”, and all items under “Perceived behavioral control” in the mountainous region showed higher scores, indicating positive effects. These results partly reflect the effect of the course on behavioral and control beliefs. According to the Theory of Planned Behavior, “Attitudes toward behavior” and “Perceived behavioral control” are among the factors that increase the intention to behave to save lives.

Among the individual measurement questions, positive responses to “I can move to a safe place alone without an adult being with me” statistically significantly increased in the mountainous and coastal regions, suggesting greater confidence in acting independently. In the low-lying region, in contrast, students may have focused more on knowledge or attitude toward behavior rather than practical actions.

The results for most factors in the Theory of Planned Behavior were not statistically significant, possibly due to a “ceiling effect”, as scores were already high at the beginning. The mean pre-test scores were above 3.19 in the mountainous region, 3.61 in the low-lying region, and 3.22 in the coastal region. This issue has been previously noted by Cheng [95,96]. When participants’ scores are already high or near the maximum, it becomes difficult to detect meaningful differences without using a sensitive measurement tool [95,97]. Overall, there are no differences between students in the mountainous regions (Co Tu ethnic minority) and those in the low-lying and coastal regions (Kinh).

This DRR education course included several active learning lessons within a limited lecture period, resulting in a short allocation of time for each lesson. This may have hindered the effective development of students’ self-protective capabilities. Additionally, relying solely on classroom lectures may be insufficient to influence the attitudes of primary school students. An effective approach would be to increase lecture time and combine classroom instruction with practical drills outside the classroom, as suggested by Soffer et al. [32]. This combination may lead to the desired behavioral changes.

Additionally, students’ intentions can be strengthened by engaging with diverse media sources, such as discussing disaster-related topics with their families or exploring information online, as suggested by Shiwaku et al. [98].

6.4. Gender and Grade Differences in Knowledge, Attitude and Disaster Preparedness Intention

Gender differences were observed only at limited knowledge levels in the low-lying region, as shown in Table 12. Female students demonstrated a higher capacity than male students in applying disaster-related knowledge, which aligns with the findings of Rahman [99]. Since no significant differences between female and male students were identified in most knowledge levels in the low-lying region and across any knowledge levels in the mountainous and coastal regions, it can be inferred that both genders had equal opportunities to participate in and receive knowledge from the educational program.

Regarding disaster preparedness, the findings show no significant difference between male and female students across the three regions. This is probably because the scores for all students were similar across the regions.

The results of grade differences in knowledge levels are presented in Table 14. Prominent trends differ within each region. In the low-lying region, Grade 4 students performed worse than Grade 3 students in the “Remembering” and “Evaluating” levels. In contrast, the coastal region showed the opposite trend, with Grade 4 students outperforming Grade 3 in the “Remembering” and “Analyzing” levels. In the mountainous region, Grade 4 students scored lower at the “Understanding” level, while Grade 3 students scored higher at the “Evaluating” level. These inconsistent patterns between Grade 3 and Grade 4 across the regions may be influenced by contextual and regional factors, similar to the attitude-related findings. In addition, these results indicate that knowledge levels are related to age or grade level. However, higher grades do not always correspond to higher knowledge scores. This observation is also reflected in Rahman’s study [99], where younger students had higher knowledge scores than older ones, while Wei et al. [100] noted that students in higher grades scored lower than their younger ones because of the pressure of a heavy course load and limited time. In short, the findings suggest that the course suits both grades, given the positive impacts discussed previously.

Table 15 presents the results of differences in attitude levels between Grade 3 and Grade 4 students across three regions. Each region shows differences at different attitude levels. Although none of the responses on the 5-point Likert scale exceeded the threshold for statistical significance (z ± 1.96), a general trend suggests that older students (Grade 4) tended to select higher-frequency responses, as indicated by the highest positive z-values. This finding aligns with Rahman’s study, which shows that the upper grades demonstrate higher awareness, as also cited by Seddighi et al. [94,99]. While some grade-level differences were recorded, there was no consistent pattern across the three regions. Contextual or regional factors may influence these variations.

Table 16 reveals no consistent trend in grade-level differences across the three regions in disaster preparedness intention. Notable patterns show that older students (Grade 4) scored higher in “Intention” and “Subjective Norm” in the mountainous region, as well as in “Perceived Behavioral Control” in the low-lying region, compared with the younger (Grade 3) students. In contrast, in the coastal and mountainous regions, younger students (Grade 3) outperformed the older ones (Grade 4) in “Attitude toward behavior” and “Perceived behavioral control”, respectively. These results show a similarity with the findings by Rahman [99], when upper-grade students show higher awareness, but as mentioned by Bandecchi et al. [101], students do not clearly state whether older or younger students consistently demonstrate higher awareness. These findings again show that contextual and regional factors are potentially influential.

Knowledge, attitude, preparedness intention effects, and gender and grade differences were summarized in

Table 17. Summary of knowledge, attitude, preparedness intention effects, and gender and grade differences.

Region (Main hazards)	Knowledge		Attitude
	Immediately	After three months
Mountainous (Storms, flash floods, landslides)	3 levels	No improvement	High positive responses
	No gender difference	No gender difference	No gender difference
	1 level: Grade 4 lower	1 level: Grade 4 higher	Unclear grade difference
Low-lying (Storms, floods)	2 levels	4 levels	High positive responses
	1 level: Males lower	No gender difference	Unclear gender difference
	2 levels: Grade 4 lower	1 level: Grade 4 lower	Unclear grade difference
Coastal (Storms, coastal erosion, floods)	No improvement	No improvement	High positive responses
	No gender difference	No gender difference	No gender difference
	2 levels: Grade 4 higher	No grade difference	Unclear grade difference
Region (Main hazards)	Preparedness intention
	Immediately	After three months
Mountainous (Storms, flash floods, landslides)	1 factor	1 factor
	No gender difference	No gender difference
	2 factors: Grade 4 higher, 1 factor: Grade 4 lower	1 factor: Grade 4 higher
Low-lying (Storms, floods)	1 factor	No improvement
	No gender difference	No gender difference
	No grade difference	1 factor: Grade 4 higher
Coastal (Storms, coastal erosion, floods)	No improvement	No improvement
	No gender difference	No gender difference
	1 factor: Grade 4 lower	No grade difference

.

6.5. Satisfaction and Motivation for the Course

Preferred teaching methods differed across the three regions. The method “Look at the pictures and guess the type of disasters” was most preferred in the low-lying and mountainous regions, indicating that visual materials helped students engage with and be interested in the lesson. In addition, students’ satisfaction with the teaching methods in mountainous regions was lower compared with the other regions. This may be explained because most students in this region are ethnic minorities and have difficulty reading and understanding Vietnamese, as mentioned in some studies [92,93]. Therefore, they were more difficult when accessing the various teaching methods.

Newly introduced methods, such as watching videos of Japanese students and hearing about the experiences of returnees, received different satisfaction ratings. Although students in mountainous and low-lying regions preferred “Watch videos of Japanese students” more, “Foreign experience sharing by a local person about earthquakes, storms, floods” was favored in the coastal region. This result suggests that students generally favor watching videos. Combining subtitles and voice-overs may further improve the effectiveness of the video method, as mentioned by Fernández-Costales [102] and Marcus-Quinn [103]. In addition to pictures, comics can be used as educational media for disaster preparedness in primary schools, as suggested by Noviana et al. [104].

Figure 7 shows that among the three regions, students in the low-lying region were most clearly motivated to learn from this program. Students in the mountainous and coastal regions had lower motivation. Therefore, it is essential to use these methods effectively. Through the self-evaluation of satisfaction with teaching methods, scholars and practitioners can select suitable methods for DRR education.

7. Conclusions

A short DRR education program was implemented for primary school students in three different regions in a central area of Vietnam. Overall, there were partially positive effects on the students’ knowledge, attitudes, and intentions to protect themselves. Various teaching methods have affected students’ ability to absorb knowledge differently, particularly the less commonly used methods. These findings provide a basis for selecting suitable DRR education teaching methods for future research.

This study focused on primary school students’ self-protection abilities across different regions with distinct disasters, including ethnic minority groups, a factor often underrepresented in other research [23]. A major strength of this study was the design of a DRR education course using multiple teaching methods, especially the incorporation of internationally related approaches, which are less common in the context of DRR programs. Various methods enhance the evaluation of how DRR educational activities affect children, as noted by Johnson et al. [23,44]. The study used a treatment-control pre-post-follow-up design, a relatively uncommon approach. Comprehensive and advanced techniques were applied, including Bloom’s Taxonomy, the Theory of Planned Behavior, and panel data analysis using Difference-in-Differences and fixed effect models (often used in econometrics). In addition to the previous studies that suggested a focus on location and self-protection [18,23], this study also considered differences in gender, grade, and ethnicity. Although the intervention did not significantly improve all levels of knowledge, attitudes, and intentions related to disaster preparedness, the results provide valuable insights for designing, implementing, and evaluating DRR education courses. These findings offer a critical foundation for future research by highlighting the current study’s strengths and limitations.

Using various methods within a limited time with young students is considered a factor that led to limited improvement in their course performance. It may be more effective to design longer and more frequent sessions. Incorporating other approaches may enhance the intervention’s effectiveness and support long-term observation and assessment. Other vulnerable groups should be considered in future DRR training courses, such as students with disabilities, students from low-income families, and those without parental care. These approaches provide a foundation for future studies in other regions within Vietnam or extending to other countries.