跨日常生活领域热相关健康行为多维量表的开发与验证
Development and validation of a multidimensional scale for heat-related health behavior across domains of everyday life
研究开发并验证了一个跨日常生活领域的热相关健康行为多维量表。该量表整合多领域评估与时间—意向视角,用于区分反应性与预期性行为,并面向一般人群。作者指出,现有工具多局限于特定场景或人群,或混淆知识、意图与行动。
Heat is no longer a future threat for health; it is already shaping everyday life in Europe. Days with temperatures exceeding 30 °C are no longer exceptional events but have become a recurrent phenomenon and are projected to increase further as global warming continues1,2,3. Heat is not only an environmental challenge but a pervasive stressor that affects individual well-being and performance in daily life4,5,6. At the societal level, its consequences are substantial: heat contributes to rising healthcare costs4,7 economic losses8,9, and tens of thousands of excess deaths each year across Europe10,11,12. While efforts to limit further temperature increases remain essential, current projections suggest that, at best, warming can be slowed rather than prevented1,13.
Against this background, adapting to heat becomes a central public health challenge. In addition to structural and policy-level responses, this places increasing importance on individual behavior14. Individual health behaviors in general have been in the focus of intervention research for a long time; they have been applied to numerous domains (e.g. physical activity, dietary behavior, alcohol consumption, sexual health) and studied using a variety of methods15,16,17. In addition, they were linked to cognitive constructs such as knowledge and literacy and their determinants18. In this specific case, heat-related health behavior (HRHB) refers to actions undertaken by individuals in everyday life to prevent, reduce, or cope with adverse health effects of heat. The role of research and science in promoting specifically HRHB can be to propose and evaluate intervention strategies. All of those intervention strategies must be based on reliable tools that, on the one hand, allow for an assessment of the current situation and, on the other hand, are suitable for the development and evaluation of corresponding measures19. When reviewing the existing research on instruments for assessing HRHB, one encounters several instruments, all of which, however, have certain limitations. Apart from the fact that some instruments focus on specific settings such as work20 or groups of people such as homeless people21 and are therefore not transferable, other instruments involve an intertwining of knowledge, intention, action, and axial structures in the sense of a theoretical framework, which is not theoretically substantiated22,23,24. This makes it difficult to use these instruments as a basis for informed HRHB intervention development.
From a socio-ecological perspective, behavior is shaped by interactions between individuals and their physical and social environments, which differ across life domains25. This is why health behavior is typically embedded in specific domains of everyday life, such as nutrition, mobility, and daily activities. This domain-specific organization of behavior is well established in health research and reflects the contextual nature of behavioral practices, as described before. However, environmental stressors such as heat do not affect a single domain in isolation, but rather cut across multiple domains such as diet or activities simultaneously. Accordingly, a comprehensive assessment of HRHB requires a domain-spanning approach that captures behavioral responses across different areas of everyday life, but still offers the opportunity to distinguish between the domains.
In addition, a temporal component should also be taken into account in this case. When considering the explanation of health behavior based on established theories and models, it becomes clear that various temporal perspectives always play a role. For example, behavior can be understood as “coping” and a short-term adaptive response to an acute stressor26,27. While this understanding could generally be applied to the context of heat-related health stress, it mostly neglects humans’ capacity for anticipation. This capacity is accounted for in other health models that go beyond a simple stress-response logic like the Health Action Process Approach28. Furthermore, from the perspective of current public health research, a long-term, anticipatory, strategic behavior is particularly important and worthwhile, as it promises the greatest impact (compared to short- and medium-term adaptations)29,30,31. Therefore, based on the described models and current findings, health behavior should also be conceptualized along a temporal–intentional continuum, ranging from reactive responses to immediate stressors to anticipatory and strategic forms of behavior.
From a measurement perspective, this conceptualizes HRHB as a multidimensional construct comprising related but distinct behavioral tendencies. Within each behavioral dimension, individual items are assumed to represent manifestations of a broader domain-specific or temporal behavioral tendency and are therefore modeled reflectively. At the same time, we do not assume that all heat-protective behaviors are interchangeable manifestations of a single homogeneous construct. Rather, the overarching concept of HRHB comprises several distinct dimensions that may vary in relevance across individuals and contexts. This distinction is particularly important given the context-dependent nature of health behavior. To the best of the authors’ knowledge, no scale currently combines the assessment of heat-related behavior across multiple everyday-life domains with a temporal-intentional perspective while being suitable for use in the general population. Integrating these perspectives may provide information beyond the overall frequency of protective behavior by identifying both where in everyday life adaptation occurs and whether it is primarily reactive or anticipatory. Such differentiation may, in turn, help identify domain- and timing-specific intervention needs and inform more targeted public health strategies.
Therefore, the aim of this study is to develop and psychometrically evaluate a new scale (HRHB-Scale) for this purpose and to discuss its suitability for the development of interventions. Based on the theoretical foundations and empirical findings described above, the initial scale development followed a two-dimensional conceptual framework comprising domain-specific adaptations (diet, housing, clothing, activity, change of location) and temporal-intentional levels (reactive, adaptive, strategic). The present study examines the extent to which this theoretically proposed organization is empirically supported and which dimensions emerge as statistically robust and practically meaningful, and how HRHB relates to the established constructs of heat-related health knowledge (HRHK) and heat-related health literacy (HRHL). Given that knowledge and literacy may facilitate protective action but do not necessarily translate into behavior, we expected HRHB to be positively associated with both constructs while capturing a distinct behavioral dimension. We further examined whether HRHB contributes to explaining heat-related health symptoms beyond sociodemographic characteristics, HRHK, and HRHL. To this end, an initial 26-item version of the HRHB scale was evaluated in a large population-based sample in Germany. The validation followed a split-sample approach, combining exploratory factor analysis in one half of the sample with confirmatory testing in the other, followed by analyses of reliability and associations with heat-related health literacy, knowledge, and symptoms. Detailed information on scale development, measures, and statistical procedures is provided in the Methods.
Results
Sample characteristics
The initial 26-item HRHB scale was evaluated in a population-based sample of adults in Germany. Data were collected through an online survey between 12 and 19 December 2025, resulting in a final sample of N = 3,351. The mean age was 51.2 years (SD = 17.64; range: 18–93). The sample comprised 52.4% women, 47.3% men, and 0.3% identifying as diverse. Participants were distributed across regions of Germany and age groups in accordance with predefined quota targets. With regard to health-related characteristics, the mean BMI was 27.24 (SD = 5.9). Sample characteristics regarding age, SES, and BMI are also presented in Table 1, regarding the distribution of gender, age groups, and regions are presented in S2 File (Table 5) in the supplementary materials. Assessed data was screened for univariate and multivariate outliers using standardized z-scores (|z| > 3) and Mahalanobis distance (p < .001), respectively32. Although a small proportion of cases was identified as outlying, these were retained, as they likely reflect valid population variability rather than data errors.
Item characteristics and distributional properties
The first version of the new HRHB-Scale consisted of 26 items. On the four-point response scale1,2,3,4, item means ranged from 2.15 to 3.70, with standard deviations between 0.53 and 1.06, indicating sufficient variability across items. Skewness values ranged from − 1.80 to 0.45 and kurtosis from − 1.05 to 3.39, suggesting moderate deviations from normality but remaining within acceptable ranges for Likert-type data33. Inter-item correlations ranged from approximately 0.20 to 0.70, indicating small to strong associations according to established guidelines34, with most correlations falling within the expected moderate range. Response distributions indicated ceiling effects for several items. The proportion of participants selecting the highest response category ranged from 15.0% to 73.7% across the initial item pool. For six items, more than 60% of participants selected the highest response category, with the highest proportions observed for Items 6 (73.7%), 7 (72.1%), 17 (67.7%), and 11 (67.2%). In contrast, other items showed substantially greater differentiation across response categories. One item (Item 25) showed comparatively lower endorsement and weaker correlations with other items, suggesting potential differences in behavioral relevance. Descriptive statistics for all items are provided in the S2 File (Table 6) in the supplementary materials.
Multidimensional structure of heat-related health behavior
The total sample was randomly split into two nearly equally sized subsamples for exploratory (n = 1676) and confirmatory factor analyses (n = 1675).
Exploratory factor analysis (EFA) initially suggested a complex structure, with parallel analysis indicating up to seven factors. However, closer inspection revealed that this solution was not interpretable in a meaningful way, as it included a single-item factor and several factors with overlapping content: One item (Item 25), reflecting long-term housing-related decision-making, formed a separate factor, showed weak or negative correlations with other items, and did not load meaningfully on any of the broader behavioral domains. Given both its statistical isolation and its conceptual focus on structural life circumstances rather than actionable behavior, this item was excluded from further analyses.
In addition, one reactive clothing-related item (Item 4) demonstrated inconsistent and cross-factor loadings and was more closely associated with general reactive coping tendencies than with the clothing domain, leading to its exclusion. After removing these items, the remaining factors were examined with regard to conceptual overlap, empirical proximity, and interpretability. Two factors representing closely related behavioral domains were therefore combined to achieve a more parsimonious and theoretically coherent structure, resulting in a five-factor solution (S2 File, Table 7): Diet across all temporal levels (Factor 1), housing across all temporal levels (Factor 2), clothing across all temporal levels (Factor 3), adaptive-strategic adjustment of activities and change of location (Factor 4), and reactive adjustment of activities and change of location (Factor 5). The final structure revealed a hybrid organization of HRHB. Factors 1–3 represented stable behavioral domains across temporal levels. In contrast, Factors 4 and 5, showed a stronger differentiation between reactive and anticipatory patterns. The distributions of the five HRHB subscales are shown in Fig. 1, illustrating differences in both central tendency and variability across behavioral dimensions. Descriptive statistics for the five subscales are provided in the S2 File, Supplementary Materials Table 9. Mean scores ranged from 3.24 to 3.50 on the four-point scale, indicating generally high endorsement across behavioral domains. The proportion of participants reaching the maximum possible subscale score ranged from 15.6% for adaptive-strategic activity/change of location to 40.4% for housing, indicating that the concentration of responses at the upper end persisted, to varying degrees, after item aggregation.
Overall, the five-factor solution was considered the most interpretable and conceptually meaningful representation of the data and was therefore retained for subsequent confirmatory factor analyses (CFA). CFA Model fit indices indicated a substantially improved fit compared to the one-factor model (CFI = 0.979 vs. 0.957; TLI = 0.976 vs. 0.953; RMSEA = 0.098 vs. 0.139; SRMR = 0.074 vs. 0.100). These findings support a multidimensional structure of HRHB and are consistent with a differentiated structure comprising domain-specific and behavior-specific components. Model fit indices for all tested CFA models are summarized in Table 2. Standardized factor loadings (S2 File, Table 7 in the supplementary materials) were consistently high across all factors, indicating a well-defined and stable factor structure.
Although the RMSEA value remained slightly above conventional cut-offs, this pattern is not uncommon in models with a relatively large number of ordinal indicators. Inspection of modification indices suggested some localized areas of misfit, such as potential correlations between conceptually related items. However, no extensive model modifications were implemented, as the aim was to retain a theoretically grounded and parsimonious factor structure rather than optimizing model fit in a purely data-driven manner. Taken together, the five-factor model was retained due to its superior fit and its conceptual interpretability.
In addition to the first-order factor structure, a higher-order factor model was tested to examine whether the identified subscales could be represented by a common overarching construct of HRHB. The higher-order model showed a comparable fit to the five-factor model (CFI = 0.978, TLI = 0.975, RMSEA = 0.100, SRMR = 0.078). The comparable fit of the higher-order model suggests that the five dimensions share sufficient common variance to justify the use of an overall HRHB score for specific analytical purposes. However, this should not be interpreted as implying that all individual behaviors are interchangeable manifestations of a single homogeneous latent construct.
Distribution of HRHB scores across the five behavioral dimensions. Violin plots show the distribution of subscale scores in the full analytic sample; boxplots indicate the median and interquartile range, and diamonds indicate mean values. Scores range from 1 to 4, with higher scores indicating more frequently reported heat-related health behavior.
Reliability of the HRHB-Scale
Internal consistency was acceptable to good across the retained subscales. Cronbach’s alpha ranged from 0.62 to 0.85, and model-based omega ranged from 0.72 to 0.89. Reliability was highest for the clothing-related subscale and lowest for the reactive adjustment of activities and change of location subscale, which also comprised the smallest number of items. The total scale showed excellent internal consistency (alpha = 0.93; omega = 0.93) (see Table 3).
Behavior, knowledge, and health outcomes
Bivariate correlations indicated that the overall HRHB score was moderately positively associated with HRHL (r = .29) and HRHK (r = .27). Full correlation estimates, including 95% confidence intervals, p-values, and pairwise sample sizes, are provided in the S2 File (Table 8). The correlation between HRHB and heat-related symptoms was smaller in magnitude (r = .14). At the subscale level, more differentiated patterns emerged: All five HRHB subscales showed positive associations with HRHL (r = .19–0.28) and HRHK (r = .19–0.29). In contrast, correlations with heat-related symptoms varied across subscales. The strongest associations were observed for adaptive-strategic adjustments of activities and change of location (Factor 4) (r = .18) and housing (Factor 2) (r = .11), whereas correlations were weaker for diet (Factor 1) (r = .04), clothing (Factor 3) (r = .10), and reactive adjustments activities and change of location (Factor 5) (r = .08). HRHL and HRHK were both negatively associated with heat-related symptoms (r = − .15 and r = − .10, respectively), while their intercorrelation was small (r = .16).
Incremental validity
To examine the incremental validity of HRHB, hierarchical linear regression analyses were conducted (see Table 4). In the first model, sociodemographic variables (age, gender, and SES) explained 5.8% of the variance in heat-related symptoms (R² = 0.058). Adding HRHL and HRHK in the second model resulted in a significant increase in explained variance (ΔR² = 0.020, p < .001). In the third model, the HRHB total score was included, yielding a statistically significant but small incremental contribution (ΔR² = 0.054, Cohen’s f² = 0.062, p < .001), resulting in a total of 13.1% explained variance (R² = 0.131). In a subsequent model, the five HRHB subscales were entered simultaneously instead of the total score. This resulted in an even larger increase in explained variance compared to the model including HRHL and HRHK (ΔR² = 0.068, p < .001), with the full model explaining 14.5% of the variance in symptoms (R² = 0.145). At the level of individual predictors, adaptive-strategic adjustments of activities and change of location (Factor 4) (β = 0.25, p < .001) and, to a lesser extent, housing (Factor 2) (β = 0.05, p = .015) were significantly associated with heat-related symptoms, whereas the remaining subscales were not independently associated. In addition, HRHL (β = −0.17, p < .001) and HRHK (β = −0.02, p < .001) were negatively associated with symptoms, and significant effects of age, gender, and SES were observed.
These findings extend beyond instrument validation by indicating that behavioral adaptation contributes additional information to the statistical explanation of heat-related health symptoms at the population level.
Discussion
This study provides evidence that HRHB constitutes a distinct and measurable dimension of population-level heat adaptation, beyond established cognitive constructs such as knowledge and literacy. To investigate this, we developed a novel scale for assessing HRHB and validated the scale. During the validation process, several aspects emerged that offer important implications for the practice of intervention research, but at the same time clearly call for a nuanced discussion. The development of the instrument followed a theory-driven yet empirically grounded approach. A key methodological strength lies in the use of a split-sample design combining EFA and CFA, allowing for an initial identification of latent structures followed by an independent validation35. The analyses revealed five plausible factors which will be discussed later on a content level. Overall, the results support a multidimensional structure of HRHB, but only partially reproduce the originally proposed domain × temporal-intentional framework. Whereas diet, housing, and clothing were primarily differentiated by behavioral domain, temporal-intentional differentiation emerged specifically for activity and change-of-location behaviors.
At the item level, the distributional properties indicate pronounced ceiling effects, with many behaviors being reported at high frequencies. This may reflect socially desirable responding36. Particularly high endorsement was found for some relatively accessible protective behaviors, such as seeking shade or moving to a cooler location when feeling hot. While these distributions may reflect the genuinely widespread adoption of basic heat-protective behaviors, restricted variability at the upper end of the response scale limits the ability of these items to discriminate between individuals with already high levels of HRHB. This may also attenuate associations with related constructs and external criteria and should therefore be considered when interpreting the validity findings. At the same time, high endorsement does not necessarily indicate poor item quality in the present context, as capturing the prevalence of commonly adopted protective behaviors is itself relevant for population-level monitoring. Beyond their implications for discriminating between individuals with high levels of HRHB, the pronounced ceiling effects also raise concerns regarding the scale’s potential responsiveness to intervention-related changes. Although item aggregation increased variability at the subscale level, between 15.6% and 40.4% of participants already achieved the maximum possible score, leaving no room for further increases within these subscales. Moreover, a shift from “somewhat agree” to “agree” does not necessarily reflect a meaningful improvement in protective behavior, particularly for behaviors that depend on situational opportunities. High endorsement may also reflect the accessibility of certain behaviors or item formulations that make disagreement less likely, rather than social desirability alone. Consequently, the present findings do not establish whether the HRHB-Scale is sufficiently sensitive to detect meaningful intervention-related changes. Future longitudinal and intervention studies should examine its responsiveness and the interpretation of score changes, while potential refinements of highly endorsed items and response categories may help improve its sensitivity at the upper end of the scale.
The aim of scale refinement was not to maximize statistical homogeneity or minimize item count alone. Retaining only the highest-loading items would have produced a shorter instrument but could have reduced the content coverage of relevant heat-protective behaviors within each domain. Item retention therefore considered both psychometric performance and conceptual relevance. This is particularly important for HRHB, where different concrete behaviors within the same domain may represent distinct intervention targets despite sharing an underlying behavioral dimension. Accordingly, the present HRHB-Scale was developed as a comprehensive population-level measure rather than as an ultra-short screening instrument. Nevertheless, two items deviated from the overall pattern and were excluded from the final model. Item 25 showed weak and partly negative associations with other items, suggesting that it captures structurally constrained housing conditions rather than modifiable heat-related behavior. Consistent with psychometric principles, items that do not share sufficient variance with other indicators of a latent construct may weaken internal consistency and factorial validity. Item 4 demonstrated comparatively weak and inconsistent factor loadings, indicating a lack of clear alignment with a single latent dimension. Its content further suggests overlap with heterogeneous reactive coping behaviors rather than a distinct behavioral domain, supporting its removal to improve the coherence and interpretability of the scale. Item 9 also warrants consideration. Compared with most other items, it describes changing one’s surroundings at a relatively abstract level rather than specifying a concrete behavioral action. Although the item showed a substantial loading on factor 4, its broader wording may allow respondents to refer to heterogeneous behavioral strategies. Future refinements of the scale could therefore examine whether a more behaviorally specific formulation improves precision without narrowing the intended content of the item.
The CFA supported the five-factor structure identified in the exploratory analysis. Overall, the model showed an acceptable to good fit across most indices, with CFI and TLI indicating good model fit, while RMSEA values were slightly above conventional cut-offs.
This pattern suggests that the model captures the underlying structure of the data reasonably well, although some degree of model misfit cannot be excluded. CFA model fit indices should therefore be interpreted with some caution. The comparatively elevated RMSEA values are likely influenced by the large number of items and the use of ordinal data, both of which are known to affect the behavior and interpretation of fit indices in structural equation modeling37,38,39. Accordingly, greater emphasis was placed on incremental fit indices such as CFI and TLI, which consistently indicated a good model fit.
The magnitude of the validity coefficients warrants a nuanced interpretation. Associations with HRHL and HRHK were moderate (r ≈ .27–0.29), and thus below benchmarks commonly applied when convergent validity is assessed against alternative measures of the same construct. However, HRHL, HRHK and HRHB were explicitly conceptualized as related but non-equivalent constructs, representing competence, knowledge and behavior, respectively. Accordingly, moderate rather than strong correlations are theoretically plausible and support relatedness without suggesting redundancy. Associations with heat-related symptoms were smaller (r = .04–0.18 across subscales) and should therefore be considered limited evidence of criterion-related validity in isolation. Importantly, however, symptoms represent a complex outcome that may both influence and result from behavior, particularly in cross-sectional data. The incremental validity analyses further showed that HRHB contributed explanatory information beyond HRHL and HRHK. Taken together, the findings therefore support the nomological positioning of HRHB while providing more cautious evidence for convergent and criterion-related validity.
A plausible explanation for the positive relationship of HRHB and symptoms is that individuals who experience heat-related symptoms are more likely to engage in both reactive and anticipatory behaviors. This is in line with results on health behavior in general40,41, as proposed e.g. by the Health Belief Model on a theoretical level27. Importantly, the positive association between HRHB and heat-related symptoms does not necessarily contradict the negative associations observed for HRHL and HRHK. Although higher literacy and knowledge may support preventive action, the present data do not establish whether behavioral adaptation mediates their associations with symptoms. Moreover, reported protective behavior may be influenced by different and potentially opposing processes. Individuals with greater previous or current heat exposure, higher vulnerability, or more pronounced symptoms may have stronger reasons to adopt protective behaviors, including anticipatory adjustments. Such individuals may therefore report both more protective behavior and more symptoms, even if their behavioral adaptations are beneficial. Conversely, the negative associations of HRHL and HRHK with symptoms may reflect other pathways or differences in individual characteristics that are not captured by the present analyses. Consequently, the observed associations should not be interpreted as evidence of a single causal pathway from literacy and knowledge through behavior to reduced symptom burden.
The positive HRHB–symptom association also warrants further consideration at the subscale level, as the stronger associations observed for adaptive-strategic behaviors (Factor 4) point to a more nuanced interpretation. Rather than reflecting purely reactive coping, these findings may indicate experience-based anticipatory adaptation, whereby individuals who have previously experienced heat strain are more likely to adjust their behavior proactively in response to expected conditions. However, neither the temporal sequence of symptom experience and behavioral adaptation nor the potential protective effects of these behaviors can be established from the present cross-sectional data. At the same time, the association with housing-related behaviors (Factor 2) suggests that structural living conditions play a key role, as individuals exposed to less favorable environments may both experience greater heat strain and be more compelled to adopt coping strategies. Taken together, these findings suggest that HRHB reflects a combination of reactive and anticipatory processes, rather than being driven exclusively by immediate symptom perception.
Finally, the analyses provide evidence for incremental validity. Adding HRHB to models including sociodemographic variables, HRHL, and HRHK resulted in a statistically significant, albeit small, increase in explained variance (ΔR² = 0.054; Cohen’s f² = 0.062). A somewhat larger incremental contribution was observed when the five HRHB subscales were considered separately (ΔR² = 0.068; Cohen’s f² = 0.080). Although these effects are small according to conventional benchmarks, they indicate that HRHB captures additional information about heat-related symptoms beyond sociodemographic and cognitive factors. Given the multifactorial nature of heat-related health outcomes, this incremental contribution may nevertheless be relevant for population-level research and monitoring. Beyond its psychometric relevance, the multidimensional structure provides practical information that would be obscured by relying exclusively on an overall HRHB score. The subscales differed in their distributions (Fig. 1) and showed distinct patterns of association with heat-related symptoms. Thus, similar overall levels of HRHB may reflect substantially different behavioral profiles. Distinguishing these profiles may be particularly relevant for public health monitoring and intervention development, as deficits in one behavioral domain may require different strategies than deficits in another. Accordingly, the overall HRHB score may provide a useful summary measure for population-level monitoring, whereas the subscales offer greater diagnostic value for identifying domain-specific intervention needs. These findings underline the importance of explicitly capturing behavior as a distinct construct in the context of heat and health, beyond knowledge and literacy alone.
There are some limitations that should be taken into account in the future. First, the sample shows slight deviations from population benchmarks (e.g., age and BMI), which may limit generalizability. Second, the use of mandatory items eliminated missing data but may have introduced forced responses. Participants who were unsure or indifferent may have selected arbitrary or one of the two midpoint responses, potentially attenuating true variance and weakening associations between variables42. At the same time, this approach may have inflated agreement on socially desirable items, contributing to the observed ceiling effects and leading to a slight overestimation of the prevalence of heat-protective behaviors. Additionally, HRHB is inherently context-dependent, meaning that observed patterns may vary across environmental, socioeconomic, and cultural settings or may be dependent on seasonal effects. This is especially important as data collection took place in December, several months after the main heat season in Germany. Although the items were designed to capture typical heat-related behavior rather than behavior during a specific heat event, responses therefore relied partly on retrospective appraisal. This may have introduced recall bias and may have resulted in reports reflecting generalized perceptions of one’s typical behavior rather than situation-specific behavioral responses. Future studies should therefore examine the scale during periods of acute heat exposure and assess the stability of its psychometric properties across seasons.
Most importantly, the findings highlight a central implication for public health: knowledge and competence do not automatically translate into behavior and better health, as also supported by extensive research43,44,45. Although HRHB showed moderate associations with both HRHL and HRHK, these constructs are clearly not interchangeable. This underlines the need to explicitly target behavior in addition to cognitive components when designing interventions.
The results of this study indicate that not all behavioral domains are equally relevant in practice. While all five subscales were positively associated with heat-related health symptoms, two stood out with stronger associations. Housing-related behaviors (Factor 2) appear to reflect structurally constrained adaptation, whereby individuals exposed to unfavorable living conditions may be both more affected by heat46 and more likely to implement coping strategies. Similarly, adaptive-strategic adjustments of activities and change of location (Factor 4) may indicate experience-based anticipatory adaptation, in which prior exposure to heat strain leads to more proactive behavioral responses. However, the stronger correlation of these two subscales should also not be overemphasized as exclusive effects: behavioral responses to heat are generally triggered by perceived or experienced strain but differ in their intensity and variability across domains. Behaviors related to diet, clothing, and reactive activity (Factors 1, 3, and 5) are likely more habitual, less resource-intensive, and easier to implement, which may explain their comparatively weaker differentiation by symptom burden.
From an application perspective, the HRHB-Scale offers potential for population-level assessment of heat-related behavioral patterns and for informing the development of domain-specific interventions47. The overall score provides a summary measure of reported protective behavior, while the subscales allow for a more differentiated assessment of behavioral domains and temporal patterns. However, the pronounced ceiling effects and the lack of longitudinal evidence on responsiveness currently limit conclusions regarding its suitability for monitoring behavioral changes or evaluating intervention effects. Further validation is therefore needed before the scale can be recommended for these purposes. Some domains (e.g., nutrition, housing, clothing) operate relatively independently of timing, whereas others (activity and change of location) require a clear distinction between reactive and anticipatory behavior. Consequently, public health interventions should be tailored along both domain-specific and temporal dimensions. Behavioral recommendations related to diet and clothing may be likely to benefit from knowledge-based approaches and risk communication, as they are largely under individual control and relatively independent of situational context. In contrast, housing-related adaptation often requires structural or environmental interventions, as individual scope of action is limited by external constraints such as housing conditions or affordability. For activity and change of location, a distinction is necessary: reactive behaviors may be supported through real-time, situational guidance (e.g., heat warnings, micro interventions), whereas anticipatory behaviors require the development of planning and self-regulation skills to enable proactive adaptation to expected heat conditions.
Importantly, the implementation of heat-protective behavior is not solely determined by individual knowledge, motivation, or willingness to adapt, but also by the resources and opportunities available to individuals. Socioeconomic circumstances, housing conditions, occupational requirements, and access to cooler environments may substantially constrain individual adaptive capacity. This is particularly relevant from a health equity perspective, as individuals with greater heat exposure or vulnerability may simultaneously face greater barriers to implementing protective behaviors. The exclusion of Item 25, which addressed long-term housing-related decisions, illustrates this tension between individual behavioral adaptation and structurally constrained opportunities for action. Consequently, lower HRHB scores should not automatically be interpreted as a lack of individual willingness to adapt. Rather, they may also reflect unequal opportunities to engage in protective behavior. This distinction is essential when translating HRHB assessments into public health interventions: identifying behavioral deficits should not lead to placing responsibility solely on individuals, but should also inform structural measures that enable equitable access to heat protection.
Clearly, these intervention approaches need to be tested in detail in future research. Effective support for HRHB therefore requires not only targeted behavioral interventions, but also attention to the structural conditions that enable individuals to adopt protective behaviors.
Methods
Development of the HRHB-Scale
The HRHB-Scale aims to map both content domains and distinguish between temporal levels. The following temporal levels of HRHB were defined: (1) immediate responses to acute heat exposure (reactive); (2) anticipatory adjustments in response to short-term heat events (adaptive); (3) longer-term preparation for future heat exposure (strategic). The content domains included diet, clothing, housing, change of location, and activities. These domains were derived from existing research on heat-related health impacts and adaptation behaviors. Specifically, prior studies have identified common symptoms of heat stress and the contexts in which they occur6,48,49, as well as behavioral responses captured in related measurement instruments22,23,24. The domains were intended to be clearly distinguishable from one another, while also possessing high practical relevance and making a difference, particularly in the context of heat.
Item development followed an iterative process involving four experts from psychology, sports science, and health research. In a first step, an item pool of approx. 35 items was jointly developed and refined to ensure adequate coverage of both dimensions, resulting in a matrix structure combining domains and temporal levels. Further discussions narrowed it down to a total of 26 items (cf. S1 File, Table 1 in the supplementary materials). The matrix was not evenly filled across all cells. This was an intentional decision, reflecting the varying relevance and feasibility of behaviors across domains and temporal levels. For example, certain combinations—such as long-term strategic adaptations in diet—were considered neither meaningful nor realistic. Nevertheless, all domains and temporal levels are represented by multiple items.
A four-point Likert scale (“disagree,” “somewhat disagree,” “somewhat agree,” “agree”) was used to avoid a neutral midpoint and encourage respondents to express a directional tendency in their behavior. This approach has been recommended in survey methodology to reduce central tendency bias and improve discriminative power in attitude and behavior measurement50. The responses are coded on a scale of 1 to 4, and mean scores are calculated for the entire scale as well as for the different dimensions overall, individual domains, and temporal levels.
Cognitive interviews with N = 4 individuals from groups vulnerable to heat stress (low socioeconomic status [SES], older adults, pregnant women, outdoor worker) conducted prior to the main survey revealed no indications of comprehension issues of the new HRHB-Scale.
Procedure and study population
Participant recruitment and data collection for the main survey were carried out with the support of an online market research panel provider. For large populations, sample sizes in the range of 1,500–2,000 are commonly considered sufficient to achieve a margin of error of approximately ± 2–3% at a 95% confidence level51,52. Since the sample was to be split for certain analytical steps (see below), the minimum sample size was set at 3,000, taking into account representative distribution ratios regarding place of residence in Germany and age group53.
Individuals were eligible to take part if they were at least 18 years old, lived in Germany at the time of the survey, and gave informed consent before participation. No further general inclusion or exclusion criteria were applied with regard to gender, education, or health status. In December 2025, a quantitative online questionnaire was distributed by the panel provider to members of its panel. Invitations to participate were issued via participants’ panel accounts, email notifications, or mobile applications. After agreeing to participate and providing informed consent, respondents completed the survey using their own devices, including desktop computers, laptops, or smartphones. The dataset made available to the research team did not contain any personally identifiable information and was therefore fully anonymized. The study was conducted in accordance with the ethical standards of the Declaration of Helsinki, and ethical approval was granted by the ethics committee of the Technical University of Munich (2024-142-NM-BA).
Additional measures
To assess the validity of the HRHB-Scale, data was collected using additional instruments. These are described in more detail below. All items and instruments used are reported in the S1 File (Tables 1, 2, 3 and 4) in the supplementary materials in their used German version and an English translation.
Heat-related health literacy
To assess HRHL, the validated heat-related health literacy scale (HRHLS)54 was used. The scale incorporates the core dimensions of health literacy (finding, understanding, appraising, and applying heat-related health information) and, in its validated form, consists of a total of 16 items (Cronbach’s alpha = 0.96). For analysis, responses were coded on a scale of 1 to 5 and subsequently used as a mean score. Higher HRHL is theoretically expected to facilitate protective behavior by supporting the identification, appraisal and application of heat-related health information, although literacy alone does not determine whether protective behavior is implemented18.
Heat-related health knowledge
HRHK was assessed using a set of 13 items designed to reflect established recommendations and evidence on heat-related health protection behaviors, drawing on the domains included in the HRHB-Scale. Each item was presented in a false/true format with an additional “don’t know” option. Responses were coded dichotomously, with correct answers scored as 1 and both incorrect and “don’t know” responses scored as 0. The correct answer was “true” for all items except items 2, 8, and 11, which are inverted. A higher sum score indicates that the person in question has a better understanding of what constitutes appropriate health-protective behavior in heat. As the instrument was intended to assess factual knowledge rather than a latent construct, no psychometric validation was performed.
Heat related health symptoms
In addition, participants were asked about the occurrence of heat-related symptoms. To this purpose, they were presented with a list of 14 possible symptoms and asked to rate, on a four-point scale (“never,” “rarely,” “often,” “almost always”), how frequently these symptoms occurred in the context of heat. The responses were coded from 1 to 4, and the mean value was subsequently calculated. The list of symptoms was compiled by the research team but is based on empirical findings regarding which symptoms classically indicate a health-related impairment49,55,56.
Sociodemographics
In addition to the constructs described above, several sociodemographic variables were collected, including participants’ age and gender. A SES index was constructed based on educational attainment, occupational position, and household income, following the conceptual approach of the Robert Koch Institute (GEDA 2019/2020-EHIS) as updated by Müters et al.57. Each component was first coded and subsequently z-standardized. The overall SES index was then calculated as the mean of the three standardized components, with higher values indicating a higher SES. Participants were asked to indicate their federal state of residence. In addition, the respondents were asked about their weight and height in order to calculate the body mass index (BMI) (kg/m²).
Data analysis
Data analyses were conducted using R Statistics 4.3.3. Data were screened for univariate outliers using standardized z-scores and for multivariate outliers using Mahalanobis distance32. Given the large sample size and the focus on population-level variability, no cases should be excluded solely based on statistical outlier criteria. For univariate outliers in continuous variables such as BMI, derived from self-reported weight and height, as well as age, plausibility checks were performed by inspecting minimum and maximum values. Implausible entries (e.g., extreme height or BMI values) were identified and discussed.
The validation of the HRHB-Scale followed a stepwise procedure35. First, descriptive item analyses were conducted to examine item characteristics, including response distributions, means, standard deviations, skewness, kurtosis, floor and ceiling effects, and inter-item correlations.
The structural validity of the HRHB-Scale was examined using a combination of exploratory and confirmatory approaches. Although the scale was initially developed based on a theoretically informed distinction between different domains and temporal levels of behavior, this structure had not yet been empirically established. Therefore, an EFA was conducted in a randomly selected half of the sample to investigate the underlying structure of the items. The number of factors was initially guided by parallel analysis. Given that parallel analysis can suggest complex solutions in item sets with heterogeneous content, factor retention was additionally informed by interpretability and conceptual coherence58. Following adjustments, conceptually overlapping factors were combined.
This refined model was subsequently tested using CFA in the second half of the sample. Models were estimated using the WLSMV estimator, which is appropriate for ordinal indicators and yields robust estimates for Likert-type data. Model fit was evaluated using the comparative fit index (CFI), Tucker–Lewis index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR)59,60,61,62. In addition, a higher-order factor model was tested to examine whether the identified subscales could be represented by a common overarching factor of HRHB.
Third, internal consistency was assessed for each retained subscale using Cronbach’s alpha and model-based omega derived from the confirmatory factor model. Given the ordinal response format and the established factor structure, omega was considered the primary reliability coefficient.
Furthermore, construct validity was examined by testing associations of the HRHB-Scale with related constructs35. Positive associations with HRHL and HRHK were expected as indicators of convergent validity. Associations with heat-related health symptoms were examined as an indicator of criterion-related validity. This would be in line with empirical examined relationships between these construct in different domains of health18. To examine the incremental validity of HRHB, hierarchical linear regression analyses were conducted. In a first step, sociodemographic variables (age, gender, and socioeconomic status) were entered. In a second step, heat-related health literacy (HRHL) and heat-related health knowledge (HRHK) were added. In a third step, HRHB was included, first as an overall score and subsequently as separate subscale scores. This approach allows for the assessment of the additional variance explained by HRHB beyond established cognitive constructs34. Statistical significance was evaluated at p < .05, and effect sizes were interpreted based on changes in explained variance (ΔR²) and standardized regression coefficients.
References
IPCC. Global Warming of 1.5°C: IPCC Special Report on Impacts of Global Warming of 1.5°C Above Pre-industrial Levels in Context of Strengthening Response to Climate Change, Sustainable Development, and Efforts to Eradicate Poverty (Cambridge University Press, 2022).
Matzarakis, A. Klimawandel in Deutschland. In: (ed Schneider, S.) Gesundheitsrisiko Klimawandel: Neue Herausforderungen für Sport, Beruf und Alltag. 1. Göttingen: Hogrefe AG; 33–40. (2024).
German Weather Service. Klimastatusbericht für Deutschland Jahr 2024 [Climate status report for Germany year 2024]2025. Available from: https://www.dwd.de/DE/leistungen/klimastatusbericht/publikationen/ksb_2024.pdf?__blob=publicationFi
Mücke, H-G. Klimawandel und Hitze. In: (ed Schneider, S.) Gesundheitsrisiko Klimawandel: Neue Herausforderungen für Sport, Beruf und Alltag. 1. Göttingen: Hogrefe AG; 53–63. (2024).
Liu, J. et al. Heat exposure and cardiovascular health outcomes: a systematic review and meta-analysis. Lancet Planet. Health. 6 (6), e484–e95 (2022).
Cramer, M. N., Gagnon, D., Laitano, O. & Crandall, C. G. Human temperature regulation under heat stress in health, disease, and injury. Physiol. Rev. (2022).
World Health Organization. Climate chance 2023 [Available from: https://www.who.int/news-room/fact-sheets/detail/climate-change-andhealth
De Sario, M. et al. Occupational heat stress, heat-related effects and the related social and economic loss: a scoping literature review. Front. public. health. 11, 1173553 (2023).
Hassan, M. & Islam, A. Estimating the Economic Burden of Climate-Sensitive Diseases and Heat Exposure on Public Health Budgets in Rapidly Urbanizing Regions. Annals Appl. Sci. Sustainable Transformation. 16 (3), 1–22 (2026).
Ballester, J. et al. Heat-related mortality in Europe during the summer of 2022. Nat. Med. 29 (7), 1857–1866 (2023).
Gallo, E. et al. Heat-related mortality in Europe during 2023 and the role of adaptation in protecting health. Nat. Med. 30 (11), 3101–3105 (2024).
Garcia-Leon, D. et al. Temperature-related mortality burden and projected change in 1368 European regions: a modelling study. Lancet Public. Health. 9 (9), e644–e53 (2024).
Wolf, M., Ölmez, C. & Schönthaler, C. al. e. Klimawirkungs- und Risikoanalyse 2021 für Deutschland | Teilbericht 5: Risiken und Anpassung in den Clustern Wirtschaft und Gesundheit [Climate Impact and Risk Analysis 2021 for Germany | Sub-report 5: Risks and Adaptation in the Economy and Health Clusters]2021.
Li, T., Liu, Y., Du, P., Cai, W. & Guo, Y. A call for evidence-based adaptation: mitigating the rising global health burden of extreme heat (The Lancet Planetary Health, 2025).
Prochaska, J. J., Spring, B. & Nigg, C. R. Multiple health behavior change research: an introduction and overview. Prev. Med. 46 (3), 181–188 (2008).
Lippke, S., Nigg, C. R. & Maddock, J. E. Health-promoting and health-risk behaviors: theory-driven analyses of multiple health behavior change in three international samples. Int. J. Behav. Med. 19 (1), 1–13 (2012).
Perski, O. et al. Understanding health behaviours in context: A systematic review and meta-analysis of ecological momentary assessment studies of five key health behaviours. Health Psychol. Rev. 16 (4), 576–601 (2022).
Sorensen, K. et al. Health literacy and public health: a systematic review and integration of definitions and models. BMC Public. Health. 12, 80 (2012).
Michie, S., van Stralen, M. M. & West, R. The behaviour change wheel: a new method for characterising and designing behaviour change interventions. Implement. Sci. 6, 42 (2011).
Tetzlaff, E. J., Oddson, B. E., Janetos, K. M. T., Meade, R. D. & Kenny, G. P. An Evaluation of Health and Safety Personnel and Workers’ Occupational Heat Stress Knowledge in Ontario: Paper B. Am. J. Ind. Med. 68 (12), 1077–1087 (2025).
Robertson, B. W., Dow, K., Salinas, J. & Cutter, S. L. Heat risk perceptions and coping strategies of the unhoused. Int. J. Environ. Res. Public Health. 21 (6), 737 (2024).
Sayili, U. et al. The heat wave knowledge, awareness, practice and behavior scale: Scale development, validation and reliability. PLoS One. 17 (12), e0279259 (2022).
Jordan, A. et al. Association between self-reported protective behavior and heat-associated health complaints among patients with chronic diseases in primary care: results of the CLIMATE pilot cohort study. JMIR Public. Health Surveillance. 10, e58711 (2024).
Arsad, F. S., Hod, R., Ahmad, N., Baharom, M. & Tangang, F. The Malay-version knowledge, risk perception, attitude and practice questionnaire on heatwaves: development and construct validation. Int. J. Environ. Res. Public Health. 19 (4), 2279 (2022).
Sallis, J. F., Owen, N. & Fisher, E. Ecological models of health behavior. Health behavior: Theory, research, and practice. 5(43–64). (2015).
Lazarus, R. S. Coping theory and research: past, present, and future. Biopsychosoc. Sci. Med. 55 (3), 234–247 (1993).
Rosenstock, I. M., Strecher, V. J. & Becker, M. H. Social learning theory and the health belief model. Health. Educ. Q. 15 (2), 175–183 (1988).
Schwarzer, R. Health action process approach (HAPA) as a theoretical framework to understand behavior change. Actualidades en Psicología. 30 (121), 119–130 (2016).
Prochaska, J. J. & Prochaska, J. O. A review of multiple health behavior change interventions for primary prevention. Am. J. Lifestyle Med. 5 (3), 208–221 (2011).
Squires, H., Kelly, M. P., Gilbert, N., Sniehotta, F. & Purshouse, R. C. The long-term effectiveness and cost‐effectiveness of public health interventions; how can we model behavior? A review. Health Econ. 32 (12), 2836–2854 (2023).
Sloboda, Z. & David, S. B. Commentary on the culture of prevention. Prev. Sci. 22 (1), 84–90 (2021).
Li, X., Deng, S., Li, L. & Jiang, Y. Outlier detection based on robust mahalanobis distance and its application. Open. J. Stat. 9 (1), 15–26 (2019).
Kline, R. B. Principles and practice of structural equation modeling (Guilford, 2023).
Cohen, J. Statistical power analysis for the behavioral sciences (routledge, 2013).
Kyriazos, T. A. Applied psychometrics: The 3-faced construct validation method, a routine for evaluating a factor structure. Psychology 9 (08), 2044–2072 (2018).
Tourangeau, R. & Yan, T. Sensitive questions in surveys. Psychol. Bull. 133 (5), 859 (2007).
DiStefano, C., McDaniel, H. L., Zhang, L., Shi, D. & Jiang, Z. Fitting large factor analysis models with ordinal data. Educ. Psychol. Meas. 79 (3), 417–436 (2019).
Shi, D., Lee, T. & Maydeu-Olivares, A. Understanding the model size effect on SEM fit indices. Educ. Psychol. Meas. 79 (2), 310–334 (2019).
Xia, Y., Yang, Y., RMSEA & CFI TLI in structural equation modeling with ordered categorical data: The story they tell depends on the estimation methods. Behav. Res. Methods. 51 (1), 409–428 (2019).
Davidson, K. W. & Scholz, U. Understanding and predicting health behaviour change: a contemporary view through the lenses of meta-reviews. Health Psychol. Rev. 14 (1), 1–5 (2020).
Johnson, B. T., Scott-Sheldon, L. A. & Carey, M. P. Meta-synthesis of health behavior change meta-analyses. Am. J. Public Health. 100 (11), 2193–2198 (2010).
Tourangeau, R., Rips, L. J. & Rasinski, K. The psychology of survey response. (2000).
Wicher, A. et al. Why Health Knowledge Does Not Translate into Health Behavior: Educational and Psychological Perspectives. J. Educ. Health Sport. 88, 68589 (2026).
Sligo, F. X. & Jameson, A. M. The knowledge—behavior gap in use of health information. J. Am. Soc. Inform. Sci. 51 (9), 858–869 (2000).
Osborn, C. Y., Paasche-Orlow, M. K., Bailey, S. C. & Wolf, M. S. The mechanisms linking health literacy to behavior and health status. Am. J. Health Behav. 35 (1), 118–128 (2011).
Kemen, J., Schäffer-Gemein, S., Grünewald, J. & Kistemann, T. Heat perception and coping strategies: a structured interview-based study of elderly people in Cologne. Ger. Int. J. Environ. Res. public. health. 18 (14), 7495 (2021).
CCSS & CoCoSS Behavioral Research to Improve the Public’s Health DoHP, Disease Prevention IoM. Promoting health: Intervention strategies from social and behavioral research. Am. J. Health Promotion. 15 (3), 149–166 (2001).
Seebass, K. Who is feeling the heat? vulnerabilities and exposures to heat stress—individual, social, and housing explanations. Nat. Cult. 12 (2), 137–161 (2017).
Ebi, K. L. et al. Hot weather and heat extremes: health risks. lancet 398 (10301), 698–708 (2021).
Krosnick, J. Question and questionnaire design. In: Marsden AP, Wright J, editors. Handbook of survey research2010. pp. 263–314.
Fowler, F. J. Jr Survey research methods (Sage, 2013).
Groves, R. M. et al. Survey methodology (Wiley, 2011).
Statistisches Bundesamt. Tabellen Bevölkerungsverteilung 2025 [Available from: https://www.destatis.de/DE/Themen/Gesellschaft-Umwelt/Bevoelkerung/Bevoelkerungsstand/Tabellen/bevoelkerung-altersgruppen-deutschland.html
Friedrich, J., Bachner, J., Blaschke, S., Mess, F. & Schmid-Ellinger, J. Coping with heat: exploring psychometric properties and relationships of a heat-related health literacy scale. Health Promot. Int. 41 (1), daag020 (2026).
Faurie, C., Varghese, B. M., Liu, J. & Bi, P. Association between high temperature and heatwaves with heat-related illnesses: a systematic review and meta-analysis. Sci. Total Environ. 852, 158332 (2022).
Thiel, J. et al. The spectrum of heat-related diseases-a meta-review. Int. J. Public. Health. 70, 1608592 (2025).
Müters, S., Michalski, N. & Hoebel, J. Aktualisierung der Berechnungsgrundlagen für den Index des sozioökonomischen Status in der Studie Gesundheit in Deutschland aktuell (GEDA) 2019/2020-EHIS. (2023).
Hayton, J. C., Allen, D. G. & Scarpello, V. Factor retention decisions in exploratory factor analysis: A tutorial on parallel analysis. Organizational Res. methods. 7 (2), 191–205 (2004).
Bentler, P. M. Comparative fit indexes in structural models. Psychol. Bull. 107 (2), 238 (1990).
Tucker, L. R. & Lewis, C. A reliability coefficient for maximum likelihood factor analysis. Psychometrika 38 (1), 1–10 (1973).
Steiger, J. H. Structural model evaluation and modification: An interval estimation approach. Multivar. Behav. Res. 25 (2), 173–180 (1990).
Byrne, B. M. Structural equation modeling with Mplus: Basic concepts, applications, and programming (routledge, 2013).
来源:Nature 健康护理主题 · nature.com