Two general frameworks have been widely used to estimate and characterize risk perception levels in the context of human-made and natural hazards: the Psychometric Paradigm and Cultural Theory (described in the following sections). Both frameworks typically rely on data collection through questionnaires or workshops, using point-scale systems such as Likert scales or 5-point questions to quantify participant responses (Marris et al., 1998). These approaches are particularly useful when the objective is to compare risk perception across individuals or social groups and to relate perception patterns to decision-making and behavioural responses in emergency contexts. This research proposes employing both approaches to achieve the specified objectives of calculating relative VRP levels of populations highly exposed to volcanic activity from Villarrica volcano (Sect. 1.2) using a psychometric approach, and the characterization of social trends within different VRP groups based on a cultural theory perspective (Sect. 2.3). In this context, VRP levels are understood as analytical and comparative constructs rather than absolute or fixed categories, allowing the identification of gradients and profiles of perception within the studied population.

Villarrica is a stratovolcano located in the Southern Volcanic Zone (SVZ) at 39.42° S, 71.93° W, active since the Late Pleistocene and situated between Villarrica and Calafquen lakes (Fig. 1). It is primarily a basaltic to basaltic-andesite volcano and ranks among the most active volcanic systems in South America (Lara and Clavero, 2004). The estimated proximal hazard zone includes potential impacts from lava flows, lahars, and minor pyroclastic flows, covering an area of approximately 1500 km2 (Lara et al., 2021, 2011). The volcanic edifice spans about 400 km2, reaching an elevation of 2847 m above sea level and rising 2450 m above its base (Lara and Clavero, 2004). In this study, the names Pucón and Villarrica refer both to urban localities and to their corresponding municipalities. When referring to rural areas, the analysis includes sparsely populated zones within the municipalities of Pucón, Villarrica, and Panguipulli. Pucón, with approximately 17 000 inhabitants, and Coñaripe, with about 1500 inhabitants according to the 2017 Census, are the closest urban localities to the Villarrica summit and have been severely impacted by historic lahars from the volcano (Fig. 1). In addition, around 10 000 people live in surrounding rural areas, where valley floors are particularly exposed due to the channelization of lahar flows along river networks. Urban localities such as Pucón, Coñaripe, Villarrica, and Lican Ray experience significant seasonal population increases during the austral summer, driven by tourism and recreational activities associated with the volcanic landscape and the ski center near the summit. This combination of permanent residents, rural populations, and transient visitors has led to increased land use pressure and expanding exposure in areas affected by volcanic hazards, particularly lahars, reinforcing the relevance of analyzing volcanic risk perception in this dynamic social and territorial context.

About 160 eruptions have been recorded at Villarrica Volcano over the past 2000 years, with approximately 130 occurring during the last 500 years (Fig. 2; GVP, 2025). Most of these events were effusive eruptions with a Volcanic Explosivity Index (VEI) of 2 (Dzierma and Wehrmann, 2010; Sepúlveda, 2004; GVP, 2025). At least one eruption with a VEI greater than 2 is estimated to occur approximately every 22 years (Van Daele, 2014). Villarrica has also demonstrated the potential for VEI 3 eruptions, such as those in 1948, 1963, and 2015, and for VEI 4 eruptions, such as those inferred for 670 CE and 1230 CE (GVP, 2025). Most eruptions have exhibited Hawaiian to Strombolian eruptive styles, with associated hazards including lahars, lava flows, tephra fallout, and small-volume pyroclastic flows (Sepúlveda, 2004; Moreno and Clavero, 2006). The most significant eruptions resulting in casualties occurred in 1948, 1949, 1963, and 1971, with a cumulative death toll of approximately 60 people, all due to lahar impacts (Sepúlveda, 2004; Moreno, 2006). The VEI 3 eruption of March 2015 involved powerful lava fountains, pyroclastic density currents near the vent, and lahars, prompting preventive evacuations (Romero et al., 2023). Although no severe impacts were recorded, subsequent studies highlighted mental health effects among emergency response volunteers, emphasizing the need for psychosocial interventions during and after volcanic crises (Espinoza et al., 2019). Following the 2015 eruption, Villarrica has remained intermittently active, with periods of elevated unrest and a prolonged yellow alert status between 2022 and 2024, highlighting the persistence of volcanic hazard in the area and reinforcing the relevance of ongoing preparedness and volcanic risk perception among exposed populations (Romero et al., 2023; GVP, 2025).

We designed a survey consisting of twenty-eight questions addressing various aspects of risk perception and individuals' attitudes toward volcanic emergencies. The questionnaire design follows the psychometric paradigm introduced in Sect. 1.1, operationalized through closed-ended questions suitable for rapid application during field campaigns. Instead of applying multi-point Likert scales, responses were collected using binary (yes/no) and multiple-choice formats. This approach was selected to reduce ambiguity, facilitate comprehension across diverse social and educational backgrounds, and ensure consistency during in situ data collection. For the VRP analysis, responses were recoded into dichotomous scores. A response consistent with a theoretically favourable VRP condition was assigned a value of one (1), while other responses were assigned a value of zero (0). The complete set of survey questions, their wording, associated VRP factors, and scoring criteria are presented in Table 1.

We considered the proximal volcanic hazard of lahars only. This is because lahar flows have been the most impactful volcanic process from Villarrica eruptions. While we do not neglect the potential occurrence of ashfall impacts, we consider lahars more relevant for the application of VRP surveys among populations highly exposed to locally known and experienced volcanic hazards. The selected risk perception factors included knowledge, internal trust, external trust, and attachment to the place, which have been widely studied in volcanology (Sect. 1). The survey questionnaire also included questions to determine cultural groups (Sect. 2.3). The survey was administered during two fieldwork campaigns conducted by the Chilean National Geological and Mining Service (SERNAGEOMIN) personnel. The first campaign, in 2016, targeted 19 census zones of rural areas, while the second, in 2017, focused on the two primary urban areas at risk: Pucón and Coñaripe. In this study, the summit of Villarrica Volcano marks the tripartite political-administrative boundary between the districts of Villarrica, Pucón, and Panguipulli in Chile (Fig. 3). The distinction between districts in the VRP results is crucial for supporting emergency management, as volcanic emergency plans in Chile, as in many other regions, are designed at the district level. Efforts were made to represent the sparsely populated rural areas of all exposed districts as accurately as possible. However, some higher valley areas near the volcano's summit were inaccessible due to road closures and poor weather conditions.

A total of 412 questionnaires were completed, with the sample geographically distributed in proportion to the population based on the available census data at the time (Table 2). Non-resident transient populations were surveyed in urban localities, and responses related to attachment to the place were considered methodologically irrelevant for non-resident respondents and therefore excluded from the analysis. This resulted in two distinct data groups: residents (n=205), who responded to all four selected VRP factors, and non-residents (n=207), for whom attachment to the place was not considered. VRP calculations were applied independently to each group. Most questionnaires were collected in the urban locality of Pucón (229), where more than half of the respondents (150) corresponded to non-resident transient populations. In the urban locality of Coñaripe, 78 questionnaires were completed, with non-residents also representing the majority of respondents (57). In rural areas, only residents (105) were surveyed, through direct household interviews in sparsely populated zones located within the municipalities of Pucón, Villarrica, and Panguipulli (Fig. 3). All samples were calculated with a 95 % confidence level and a standard error of less than or equal to 0.1, representing the target population based on the available census data used for each zone at the time of survey application (Table 2).

Each survey question was assigned a value of one (1) when the response aligned with a theoretically optimal VRP condition (Table 1). A total of 28 survey questions were used to compute the VRP scores, distributed across four factors: knowledge (10 questions), internal trust (5 questions), external trust (6 questions), and attachment to place (7 questions). For the question addressing respondents' expected reaction in the event of an eruption, responses were dichotomized to capture risk awareness rather than emotional intensity: indifference was coded as 0, whereas all other responses indicating concern or affective engagement were coded as 1. Demographic and socio-cultural variables were excluded from the VRP score calculation and used exclusively for sample characterization and subgroup analyses. For non-resident respondents, attachment-to-place questions were omitted, and VRP scores were computed using the remaining 21 questions. Each factor score was calculated as the average of its associated questions. To ensure comparability across dimensions and avoid scale-related bias, factor scores were standardized using z-scores before clustering. The standardized scores were subsequently classified into five clusters using a k-means analysis (Eq. 1). The selection of five clusters was guided by the objective of producing an interpretable ordinal VRP scale (very high, high, moderate, low, and very low), consistent with classifications commonly used in risk perception and disaster risk studies. Raw factor scores, standardized z-scores, and final VRP level classifications for all respondents are provided in the Supplement (Table S2).

Cluster centroids represent the relative position of respondents within the multidimensional factor space, defined as the mean standardized scores of all factors for each VRP level. Clusters were ordered according to their centroid scores, from higher to lower values, and interpreted as very high, high, moderate, low, and very low VRP levels. The resulting classification reflects relative VRP conditions within the analyzed sample, where higher centroid scores indicate more favourable VRP conditions based on current disaster risk science understanding. Cluster centroids represent the mean position of each group in the multidimensional standardized factor space used in the k-means clustering procedure. Consequently, when projected onto a single factor, centroid values do not necessarily coincide with the median of the univariate score distributions shown in the boxplots. This difference reflects within-cluster heterogeneity, distributional asymmetry, and the fact that clustering is performed simultaneously across all factors rather than independently for each dimension. Therefore, centroid positions should be interpreted as reference values summarizing the relative contribution of each factor to the overall VRP classification, rather than as measures of central tendency within the individual score distributions. Cluster centroids for each VRP level are reported in the Supplement (Table S1). 1J=∑j=1k∑i∈Cj‖zi-cj‖2 Where J is the sum of squared Euclidean distances minimized by the k-means algorithm, zi is the vector of standardized (z-score) VRP factor scores for the case i, cj is the centroid (mean vector) of cluster j in the standardized factor space, Cj is the set of cases assigned to cluster j, and k is the number of clusters.

In this research, the primary classification of cultural groups distinguishes between two main population samples: residents and non-residents. This distinction is important for understanding relative influences on VRP, as non-residents or transient visitors are generally believed to have lower awareness levels due to limited knowledge of local volcanic emergency plans, lack of preparedness, and a greater tendency to approach volcanic summits (Wulan Mei et al., 2020; Bird et al., 2010; Nomura et al., 2004). Unlike rural areas, where only permanent residents were surveyed, urban areas such as Pucón and Coñaripe included both residents and transient populations (Figs. 4 and 5). This classification also incorporates social variables commonly used in risk assessments of individuals and communities exposed to volcanic activity. The selected variables include age, educational level, gender, native identity, religion, language, occupation, and economic activity, with classifications and known vulnerable groups described below.

K-means calculations result in a relatively balanced distribution of VRP levels across the 205 resident and 207 non-resident questionnaires, with each VRP category representing approximately 20 % of the sample, as expected given the clustering approach (Fig. 4). Among residents, the highest concentration corresponds to the high VRP level (25.9 %), whereas the lowest proportion is observed in the very high VRP level (17.1 %). Low and very low VRP levels represent 20.0 % and 19.5 % of the resident sample, respectively. Among non-residents, VRP levels are more evenly distributed. Moderate VRP represents the largest group (20.8 %), followed by very low VRP (20.3 %), while very high VRP shows the lowest proportion (18.8 %). Overall, residents tend to display slightly greater variability across VRP levels than non-residents, although both groups exhibit comparable distributions (Fig. 4).

VRP levels vary across districts within the study area. Given the high mobility and transient nature of non-resident populations, their spatial patterns require a separate analytical approach. Therefore, this section focuses exclusively on residents, who exhibit greater spatial stability. In the Pucón district, residents show a predominance of very high (26.2 %), high (21.5 %), and moderate (28.0 %) VRP levels, while low and very low VRP each account for 12.1 % (Fig. 5). In contrast, the Villarrica district concentrates the highest proportions of low (43.6 %) and very low (41.0 %) VRP levels among all districts. High VRP represents 12.1 % of the sample, very high VRP only 2.6 %, and no cases are recorded in the moderate category (Fig. 5). In Panguipulli, high VRP dominates (42.4 %), followed by very low (20.3 %) and low (16.9 %) VRP, while moderate and very high VRP each represent 10.2 % of the sample. Spatially, higher VRP levels are more frequently observed in urban areas, particularly in Pucón town, where very high, high, and moderate VRP levels are more prevalent (Fig. 6). In contrast, rural zones, especially in Villarrica and surrounding districts, concentrate larger proportions of low and very low VRP levels. These patterns suggest that urban contexts are associated with higher VRP among residents, although this interpretation should be considered descriptive rather than causal.

This psychometric approach using K-means clustering provides insight into how each selected VRP factor contributes to the differentiation of VRP levels. The following sections describe the distribution of standardized factor scores for residents and non-residents, supported by boxplots and cluster centroid values.

This section describes the distribution of volcanic risk perception (VRP) within socio-cultural groups, distinguishing between residents and non-residents. Figures 9 and 10 illustrate the relative distribution of VRP levels across demographic, socio-economic, and cultural variables. Greater variability in VRP distributions is observed across age, educational level, native identity, employment status, and economic activity, whereas gender and religious affiliation display more stable and comparatively balanced profiles across VRP levels. Differences related to nationality and language are also evident, although the interpretation of these patterns is constrained by smaller subgroup sizes. The following subsections detail these distributions, highlighting similarities and contrasts between residents and non-residents without inferring statistical significance.

The Psychometric Paradigm and Cultural Theory approaches were combined in this research, where responses from a questionnaire allowed not only the assessment of relative risk perception levels but also an analysis of how selected factors influence them and how different cultural groups are characterized by these levels. Both approaches have been criticized for their low statistical explanatory power (e.g., Sjöberg, 1996; Marris et al., 1998). Marris et al. (1998) tested these approaches and found several cultural biases, although the psychometric approach explains about 32 % of the variation in risk perception for only one out of four specifically analyzed cultural groups. However, as suggested by other studies (e.g., Paton et al., 2010; Zeidler, 2015; Renn and Rohrmann, 2000; Jones et al., 2013), we do not disregard the complementary usefulness of Cultural Theory, particularly in characterizing the cultural patterns of the surveyed population. Rather than treating these frameworks as competing explanations, we argue that both approaches are complementary and, when used together, provide broader and more practical insights into volcanic risk perception. The psychometric paradigm is effective in capturing the multidimensional structure and relative positioning of key VRP factors, while Cultural Theory contributes to interpreting how groups commonly addressed in volcanic risk contexts are socially composed and differentiated. This combined approach allows the identification of distinct configurations of risk perception that would not be evident through the application of a single framework alone. Importantly, this integration facilitates a shift from interpreting risk perception solely in terms of individual factor scores toward understanding how combinations of knowledge, trust, and territorial attachment jointly shape differentiated risk perception profiles. Such an approach is particularly relevant for applied contexts, as both VRP levels and associated cultural patterns can inform civil protection agencies in designing more targeted and context-sensitive emergency management and communication strategies. In this study, Cultural Theory is not applied as a classificatory or predictive model, but rather as an interpretative lens to contextualize social and cultural patterns emerging from the VRP results. This approach follows recent applications that emphasize interpretation over typological assignment in disaster risk research.

Residents and non-residents represent a fundamental distinction in this research, as both groups were surveyed using the same VRP questions, except for those related to place attachment. Although VRP levels were calculated independently for residents and non-residents, comparative interpretations focus on structural patterns and relative factor contributions rather than direct metric equivalence. This distinction is justified because transient visitors typically lack a local sense of community, including emotional attachment to place and sustained civic participation in emergency planning processes (Paton et al., 2008; Barberi et al., 2008; Ricci et al., 2013). For residents, the results indicate consistently positive scores across all VRP levels, which may reflect a strong sense of community cohesion when facing volcanic emergencies (Sect. 3.3.1). Only three VRP factors were evaluated for both residents and non-residents: knowledge, internal trust, and external trust. Among these, knowledge exhibited the greatest variation between the two groups. Residents scored lower, whereas non-residents showed more moderate scores (Sect. 3.3). This discrepancy may indicate that populations living in exposed areas do not necessarily possess higher levels of volcanic risk knowledge, a pattern widely documented in volcanic risk perception studies (e.g., Carlino et al., 2008; Ricci et al., 2013; Davis and Ricci, 2004). This finding suggests that long-term exposure and familiarity with volcanic environments may contribute to the normalization of risk, where everyday coexistence with volcanic activity reduces perceived threat and the motivation to seek or retain detailed hazard-related information. Regarding internal trust, non-residents displayed lower and predominantly negative VRP scores compared to residents (Sect. 3.3), suggesting lower reliance on local social networks and a possible tendency to overestimate their individual capacity to cope with volcanic hazards (Barberi et al., 2008). In contrast, external trust levels were similar across both groups, with predominantly positive scores across all VRP levels (Sect. 3.3). This result indicates that both residents and non-residents place comparable trust in scientific experts and authorities, increasing the likelihood of compliance with official recommendations and emergency protocols (Espluga et al., 2009; Njome et al., 2010; Haynes et al., 2008).

Some distinct cultural patterns associated with VRP levels were identified primarily among residents, whereas non-residents were more evenly distributed across VRP categories (Sect. 3.4). Among residents, age and educational level emerged as the most relevant differentiating factors. Adults and elderly individuals were more frequently concentrated in low and very low VRP levels, consistent with previous findings that identify elderly populations as potentially vulnerable groups in disaster contexts. However, as observed in similar studies, no consistent or linear relationship between age and risk perception can be firmly established (e.g., Barberi et al., 2008). Educational levels exhibited a stronger association with VRP levels (Sect. 3.4). Among residents, individuals with primary education were predominantly concentrated in low (61 %) and very low (35 %) VRP categories, although a notable proportion was also found in the high VRP category. In contrast, residents with secondary education were more frequently represented in moderate and very high VRP levels, suggesting that formal education may play a role in enhancing risk awareness and interpretation. This pattern may reflect greater exposure to scientific information, risk communication materials, or institutional narratives related to volcanic hazards.

Regarding occupation, residents outside the workforce, such as retired individuals or those engaged in unpaid domestic activities, were predominantly concentrated in low and very low VRP categories (Sect. 3.4). This group may face increased vulnerability due to limited access to information channels or reduced engagement with institutional preparedness initiatives. Conversely, students, primarily represented among non-residents, displayed VRP levels concentrated at both extremes, very high and very low, suggesting heterogeneous perceptions potentially influenced by unequal access to reliable information and digital media environments (e.g., Chester et al., 2008; Njome et al., 2010). Among economically active residents, VRP levels varied substantially depending on economic sector (Sect. 3.4). Occupations closely linked to local livelihoods, such as agriculture, handcrafting, and especially tourism, showed pronounced variability in VRP levels. Given the direct economic consequences that volcanic activity can impose on these sectors, such variability may reflect tensions between economic dependency, perceived threat, and risk acceptance. These findings highlight the importance of developing sector-specific risk communication strategies that acknowledge both economic vulnerability and perceptual diversity.

Although self-identified native populations were slightly overrepresented in low and very low VRP categories (Sect. 3.4), this result should be interpreted with caution. Similar patterns have been reported in other Indigenous contexts, where lower scores in standardized risk perception measures coexist with strong place-based knowledge, experiential understanding of hazards, and culturally embedded interpretations of risk that are not fully captured by survey-based instruments (Roder et al., 2016). Indigenous relationships with volcanic landscapes are often shaped by culturally embedded knowledge systems, cosmologies, and historical experience, which may not be fully captured through standardized survey instruments. Greater integration of Indigenous knowledge frameworks is therefore needed to understand better how volcanic risk is perceived and responded to within these communities (e.g., Hastangka and Suprapto, 2023; Niroa and Nakamura, 2022; Roder et al., 2016). No clear patterns were identified concerning gender identity or religion. While existing literature suggests that these factors may influence risk perception through worldview formation and belief systems (e.g., Cutter, 2017; Barberi et al., 2008; Eiser et al., 2015; Clarke, 2004; Breskaya and Zrinščak, 2024), their effects may be highly context-dependent and require more targeted qualitative or mixed-method approaches to be adequately captured.