Predicting Sustainable Consumption Behavior from HEXACO Traits and Climate Worry: A Bayesian Modelling Approach

Abstract

Addressing climate change requires deeper insight into the psychological drivers of pro-environmental behavior. This study investigates how personality traits, climate-related emotions, and demographic factors can predict sustainable consumption and climate action participation using a Bayesian regression approach. Drawing from the HEXACO personality model and key emotional predictors—Climate Change Worry (CCW) and environmental empathy (EE)—we analyzed data from 604 adults in Greece to assess both private and public climate-related behaviors. This research is novel in its integrative approach, combining dispositional traits and affective states within a Bayesian analytical framework to simultaneously predict both sustainable consumption and climate action. Bayesian model testing highlighted education as the most powerful and reliable predictor of sustainable consumption, with increasing levels—namely Doctoral education—linked to more environmentally responsible action. CCW produced small but reliable effects, supporting hypotheses that moderate emotional concern will lead to sustainable behavior when linked to efficacy belief. The majority of HEXACO traits, e.g., Honesty–Humility and Conscientiousness, produced limited predictive power. This indicates in this case that structural and emotional considerations were stronger than dispositional personality traits. For climate action involvement, Bayesian logistic models found no considerable evidence of any predictor, corroborating the perspective that public participation in high effort action is most likely to rely on contextual enablers instead of internal sentiments or attributes. A significant interaction effect between education and gender also indicated that the sustainability effect of education is moderated by sociocultural identity. Methodologically, this research demonstrates the strengths of Bayesian analysis in sustainability science to make sensitive inference and model comparison possible. The results highlight the importance of affect-related structural variables in behavioral models and have applied implications for theory-informed and targeted climate education and communication interventions to enable different populations to act sustainably.

1. Introduction

As scientific agreement about the growing impacts of climate change increases, knowledge of the psychological foundation of sustainable consumption has grown more critical. The behavior of individuals continues to be a key driver of both environmental deterioration and possible mitigation, and researchers have thus explored the dispositional, affective, and social–cognitive determinants of pro-environmental intentions and behavior (Vlasceanu et al., 2024). Though conventional models like the Theory of Planned Behavior (Venkatesh, 2000) and the Value–Belief–Norm Theory have created sound foundations for explaining sustainability-related decisions, there is a growing demand for more integrated models that also entail personality variables along with affective reactions (Vlasceanu et al., 2024; Hidalgo-Crespo et al., 2023).

There is already evidence of the predictive potential of personality in sustainability (Becht et al., 2024) through the ACO model (Ashton & Lee, 2009). These personality dimensions, which are linked to moral concern, responsibility, and self-regulation, have been shown to be consistently related to pro-environmental attitudes and behaviors (Ashton & Lee, 2009; Peng-Li et al., 2021). Meta-analytic results highlight Honesty–Humility as an especially strong predictor of sustainable behavior, accounting for more explanatory power than most Big Five equivalents (Soutter et al., 2020). Further, research with structural models and mediation analyses established that Honesty–Humility and Openness to Experience shape sustainable behavior not just directly, but also indirectly through moral emotions like guilt and moral anger (Peng-Li et al., 2021; Soutter et al., 2020).

Theoretically, Honesty–Humility is an orientation based on fairness, sincerity, and non-exploitation—values that are quite compatible with sustainable consumption values such as equity and regard for future generations (Peng-Li et al., 2021; Soutter et al., 2020). High scorers on the trait are less materialistic and more prosocial and ethical in their concerns, and are therefore extremely sensitive to the moral salience of the climate question. Conscientiousness, encompassing self-restraint, future planning, and conscientiousness, can foster sustainable behavior due to its link with personal responsibility and long-term goal orientation—both of which are required when combating global climate change (Ashton & Lee, 2009; Peng-Li et al., 2021). Openness to Experience, as expressed in curiosity, intellectual flexibility, and appreciation for beauty, can enhance one’s receptivity to environmental cues and interest in abstract issues such as global warming or environmental systems. These factors also influence emotional sensitivity to climate-related emotions, such as concern and empathy, offering pathways by which personality tendencies can direct action (Ashton & Lee, 2009; Peng-Li et al., 2021). Coupling character traits, therefore, offers a theoretically informed approach to the mechanisms by which individual dispositions intersect with affective response in predicting sustainability performance.

Simultaneously, affective predictors like Climate Change Worry (CCW), environmental empathy, and eco-guilt have also been significant motivators of environmental action (Panno et al., 2021; Shipley & Van Riper, 2022). Climate emotions are not one-dimensional and, depending on the intensity and context, can even mobilize debilitate action (Becht et al., 2024). Although there is some evidence to indicate that moderate climate anxiety facilitates pro-environmental behavior (Qin et al., 2024), other researchers point to the moderating role played by environmental efficacy, future self-continuity, and hope (Leite et al., 2023). Pride and guilt are among the emotions identified as immediate motivators or mediators between moral awareness and commitment to behavior (Shipley & Van Riper, 2022).

In spite of such encouraging advances, integrative models that concurrently consider both dispositional and emotional predictors and dispositional traits are uncommon. Furthermore, analytical methods applied to investigate these relations conventionally depended on frequentist approaches that could mask detail in effect sizes, model fitting, and parameter uncertainty. In contrast, Bayesian methods provide a more generally acceptable methodological option in psychological science, enabling more informative inferences through posterior distributions, model averaging, and Bayes factor comparisons (Bergh et al., 2021). Bayesian regression techniques have been used to test non-linear effects, mediation, and parameter uncertainty in climate psychology research in a more transparent way (Becht et al., 2024).

To this end, the current research answers three gaps in the literature: the lack of HEXACO-based trait–emotion models in the research on sustainable behavior, the requirement for methodological progress by Bayesian inference, and the requirement for more behaviorally fine-grained predictors like climate action engagement. Using a sample of the general population comprising 604 participants, we examine how Honesty–Humility, Conscientiousness, climate concern, and affective states predict sustainable consumption behavior and climate action using Bayesian linear and logistic regression analyses performed in JASP. Here in the present study, the term “sustainable consumption” is used to refer particularly to environmentally sound consumer actions—i.e., buying green products, reducing waste, and mindful purchasing. While proximal to the broader construct of “pro-environmental behavior,” which can involve activism as well as conservation behaviors, our particular focus is on consumption-directed decisions made within everyday contexts.

By placing this study at the juncture of personality, emotion, and sustainable behavior, we respond to a pertinent gap in the literature for integrative frameworks that incorporate stable dispositional characteristics and situational emotional reactions. Although past studies have concentrated on investigating personality or emotions related to climate independently, hardly any studies have investigated their joint predictive power for sustainable consumption and climate action involvement. Second, there is hardly any empirical research on the role of climate concern as a potential mediator—i.e., are individuals who score high on Honesty–Humility environmentally virtuous because they are more concerned or because they are guilty about climate change as well? Finally, the use of Bayesian techniques to this field is the exception rather than the norm, especially in country contexts like Greece, and therefore there are still unsolved issues on how these sophisticated methods can help us better understand the psychological processes behind sustainability behavior.

Theoretically, this study contributes to an integrative model, which incorporates moral personality traits and climate-related emotions in the explanation of individual differences in sustainable consumption. This integration combines two convergent strands of environmental psychology research to offer a more intricate explanation of how values and emotions jointly shape people’s behavioral responses to climate change. Methodologically, this study presents Bayesian regression models—through JASP and default priors—as a credible alternative to conventional analyses with the ability to offer a richer interpretation of uncertainty and a more lenient model evaluation. Practically, the results can guide climate education and advocacy in specifying the personality types that are most receptive to emotional appeal and in indicating emotional appeals (e.g., concern, sympathy) that best inspire behavioral change. By integrating trait-based and emotion-based predictors within a firm Bayesian framework, this study contributes to empirical knowledge as well as on-the-ground endeavors to foster sustainable behavior. Practically, the results can direct climate advocacy and education by determining the most emotionally receptive personality profiles to emotional appeals and by targeting the emotional motivators (e.g., anxiety, empathy) that are most effective at inspiring behavioral change. By integrating trait-based and emotion-based predictors within a powerful Bayesian framework, this study contributes to scientific knowledge and on-the-ground action towards sustainable behavior.

The findings of this research highlight the general dominance of education and concern for climate change in driving sustainable consumption, with Bayesian analysis offering strong support for these predictors. Although affective engagement—especially concern for climate change—exerted small but valid effects, all of the personality traits like Honesty–Humility and Conscientiousness had a minimal contribution to the predictive models. Climate action engagement, on the other hand, was not consistently accounted for by any psychosocial or demographic variable. Particularly, gender and education interacted significantly such that the effect of education on sustainable consumption differed significantly by gender, implying crucial sociocultural processes. Collectively, these results illustrate the complex interplay of structural, affective, and demographic variables in sustainability participation.

The remainder of this article is structured as follows: Section 2 discusses the literature on personality traits, emotional reactions triggered by climate, and their functions in sustainable behavior. Section 3 describes the conceptual model and methodology, including measurement and sampling details. Section 3.1 lays out the conceptual model and justification, describing systematically the hypothesized variable interrelations. Section 4 presents the statistical analysis plan and Bayesian modeling strategy. Section 5 reports the empirical findings for both sustainable consumption and climate action participation. Section 6 offers a detailed discussion of theoretical, practical, and methodological implications, while Section 7 concludes with limitations and directions for future research.

2. Literature Review

2.1. Personality Traits and Sustainable Consumption Behaviors

Current research has increasingly focused on the function of personality traits as antecedents of sustainable consumption behavior and eco-friendly values (Martin et al., 2022). The traditional dependence on the Big Five model has yielded a beneficial but limited understanding of the psychological foundation of environmental engagement. Individual studies and meta-analyses show that Openness to Experience and Conscientiousness, among others, are positively but modestly correlated with environmentally friendly behaviors like recycling, energy saving, and green consumerism (Hidalgo-Crespo et al., 2023; Soutter et al., 2020). Yet, with the addition of Honesty–Humility (H–H) in the HEXACO framework, it has been possible to offer a more moral-based account of ecological concern by including a dimension that involves fairness, honesty, and anti-greed values—qualities of substantial relevance to environmental responsibility (Kesenheimer & Greitemeyer, 2021; Soutter et al., 2020).

A number of studies have stated the explanatory ability of H–H and Openness on behavioral and attitudinal results. Panno et al. (2021) and Kesenheimer and Greitemeyer (2021) showed the way the traits directly and indirectly predict pro-environmental behavior (PEB) via mediators such as pro-environmental attitudes and moral emotions, including moral anger. Particularly noteworthy is the synergistic role played by personality, climate emotions, and social media engagement, as identified by Balaskas (2024) in demonstrating how HEXACO facets combine with contextual setting—online posts and emotional framing—to influence behavioral intention. Education level and gender also played moderator roles, with females and more educated individuals being more sensitive to eco-guilt and environmental tagging. This type of evidence is a reflection of the value of bringing personality theories into media and emotional predictors in explaining the multifactoriality of sustainable consumption behavior.

Meta-analytic evidence also offers additional support for the strength of H–H and Openness in large and heterogeneous samples. Soutter et al. (2020) and Cipriani et al. (2024) found that these dimensions consistently yielded medium effect sizes (r ≈ 0.20–0.25) for attitudes and behaviors, overshadowing Agreeableness and Conscientiousness’s predictive power. In addition, Cipriani’s meta-analysis built on this finding by demonstrating that reduced Openness and a greater endorsement of authoritarian attitudes (e.g., Social Dominance Orientation, Right-Wing Authoritarianism) were both firmly linked to climate change denial. These trends indicate that an ecologically conscious personality is not politically neutral, but instead situated within general sociopolitical and cultural orientations, which can either stimulate or repress its behavior (Cipriani et al., 2024).

Although some traits like Conscientiousness and Agreeableness have shown weaker yet positive relationships with sustainability behavior, their influence is mediated by factors like environmental norms or feelings of obligation (Hidalgo-Crespo et al., 2023). Traits such as Neuroticism or Emotionality have more conflicting or varied relationships. For instance, Sijtsma et al. (2023) reported zero predictive validity for HEXACO traits for trust behavior in adolescents, with indications that trait effects may be attenuated by developmental stage or context sensitivities. This is a reference to a possible developmental moderation of personality effects and points to the need to control for age, context, and affective states in ecological action modeling.

Recent empirical research has broadened the agenda of sustainable consumption research by incorporating various contexts of behavior, motivational theories, and policy instruments. Bahja and Hancer (2021) showed that eco-guilt, even though classically linked to ecologically responsible action, has a direct and strong effect on environmentally responsible tourist behavior (EFTB), but not on repeat intentions—indicating an intricate affect–behavior relationship mediated by context. Conversely, Zhang et al. (2025) used a multi-agent model to model the adoption of sustainable behavior within social networks and found that policy incentives like subsidies and information campaigns are more effective in promoting pro-environmental action than green labeling. This systems perspective at the macro level is the opposite finding to that of Zheng et al. (2023), at the individual level, where environmental awareness and green self-efficacy were strong predictors of green purchase intentions of green food, while competitive awareness reduced perceived control. These results emphasize the need to segregate motivational paths, as further underscored by Ye et al. (2024)’s experimental study of greenwashing: scarcity appeals will not be effective if viewed as manipulative, underscoring the mediating effect of trust and impression management motives. Horani and Dong (2023), in studying the sustainable purchasing of phones, also established that perceived sustainable value mediates between customer expectation and behavior intention, moderated by price sensitivity.

In sum, the literature indicates a clear pattern: Honesty–Humility and Openness to Experience are the most consistent and strongest predictors of pro-environmental behavior and climate action intentions, especially when placed within a model that controls for emotional, moral, and contextual effects. This research extends these foundations by using a Bayesian approach to analysis to investigate HEXACO personality traits, climate concern, and emotion-based moderators as predictors of sustainable consumption behavior. In doing so, it seeks to enhance theoretical models of personality-based environmental action and provide applied understanding to inform targeted climate promotion and educational outreach. Moreover, these articles demonstrate a converging focus on affective, cognitive, and contextual influences on green behavior with different assumptions about agency and scope. While Bahja and Hancer (2021) and Zheng et al. (2023) are most concerned with self-efficacy and network processes, Ye et al. (2024) and Horani and Dong (2023) focus on perceived credibility and value congruence. Notably, these works extend the scope from personality and determine the appropriateness of affective and situational constructs—drawing on the rationale for the current study to integrate dispositional and emotional determinants under a Bayesian model. Yet the relative lack of theoretical anchorage within trait models in these works serves as a reminder of the absence of the integration of stable personality dispositions and behavioral outcomes—an issue that this research bridges directly.

2.2. The Influence of Worry and Empathy on Sustainable Behavior

Theoretical and empirical developments in the last several years increasingly identify emotional processes—particularly Climate Change Worry (CCW) and ecological empathy—as the key to explaining sustainable behavior and climate action (Soutar & Wand, 2022). Despite the traditional dominance of cognitive and dispositional approaches to the research agenda, the emerging literature highlights that emotional involvement represents the motivational impetus by which environmental values are translated into behavioral expression (Zhang et al., 2025; Zheng et al., 2023).

Climate Change Worry, as a persistent worry or preoccupation with the anticipated impacts of climate change, represents a fundamental psychological construct with observable behavioral implications. Stewart (2021) developed the Climate Change Worry Scale (CCWS) to quantify the affective component of this worry, establishing its internal consistency, test–retest reliability, and validity. Empirical findings indicate that moderate levels of CCW typically enable participation in private- and public-sphere pro-environmental behavior (Vlasceanu et al., 2024; Becht et al., 2024). For example, Becht et al. found small yet reliable linear associations between CCW and climate action in adolescents in three large-sample studies. These results are the opposite to expectations of increased climate worry necessarily resulting in “eco-paralysis.” Nevertheless, curvilinear effects or reversals contingent on context have been identified within other research, where too much worry sometimes undermines action via perceived helplessness or affective overload (Qin et al., 2024; Zheng et al., 2023).

To grasp how climate concern is acted on in sustainable ways, one must take into account the psychological processes involved. Climate Change Concern (CCW) serves as an energizing stimulus when it is moderate and coupled with feeling efficacious or in control. Extended Parallel Process Model and Protection Motivation Theory state that concern is followed by action only if one feels that he or she can make a difference—that is, perceived efficacy goes hand in hand with perceived threat. In its absence, concern is followed by denial, avoidance, or burnout (Qin et al., 2024; Zheng et al., 2023). In contrast, environmental empathy is exercised by way of moral concern and affective identification with nature or others who are hurt and can draw upon prosocial norms and guilt-activated moral obligations. CCW and empathy hence both exercise control but in different processes: CCW elicits reflective coping or action in the future when handled positively, but empathy elicits restorative or preventive action by way of emotional identification and value consonance (Vlasceanu et al., 2024; Becht et al., 2024). Most importantly, differences between people in terms of self-efficacy, hope, and emotion regulation affect these channels and then condition whether or not feelings become stimuli or burdens for pro-environmental engagement.

Meanwhile, climate anxiety is not homogeneous. Ágoston et al. (2022) showed that CCW is diverse and includes related states like eco-anxiety, eco-guilt, and eco-grief with differential behavioral consequences. Qualitative data showed that people who felt eco-grief or eco-guilt participated in compensatory sustainable behavior, that is, these emotions were internalized as moral obligations. Likewise, Leite et al. (2023) found that hope and despair mediate the effect of CCW on sustainable behavior: although CCW is a firm predictor of pro-environmental behavior, this connection is strengthened when supplemented by hope stemming from efficacy beliefs and weakened by despair or denial.

These results corroborate wider research hypothesizing environmental empathy—the emotional identification with or moral care for nature and its actualities—which is an auxiliary aptitude to CCW (Soutar & Wand, 2022). Environmental empathy has been shown to correlate with a higher intention to defend ecosystems, give money to environmental organizations, and endorse climate policies (Balaskas, 2024). It not only operates as a short-term action trigger but also as a bridge between personality traits (e.g., Honesty–Humility) and behavior, through moral emotions such as eco-guilt and moral outrage (Panno et al., 2021). People high in empathy might react more to environmental harm and thus be more willing to take restorative or preventative action. Moral anger has especially been demonstrated to predict activism, acting as a kind of collective indignation that stimulates social action (Hornsey & Pearson, 2024).

Most significantly, research highlights that emotions need to be presented in a positive framework in order to ensure long-term commitment. Though negative moral emotions such as worry and guilt can elicit responses of urgency, over-reliance on distress-based communication can lead to burnout or avoidance. Research by Ojala et al. (2021) and Fischer et al. (2017) highlights the importance of balancing emotional appeals with the synthesis of moderate worry- and efficacy-based optimism in order to maintain behavior change. For instance, teenagers with greater green self-efficacy and future self-continuity were more prone to translate CCW into significant action (Qin et al., 2024), corroborating the argument that self-efficacy plays the role of an essential moderator within affective–behavioral models.

Contextual and social cues also influence the behavioral relevance of emotional reactions. Interpersonal encouragement and regular discussion of climate matters were found by Latkin et al. (2025) to considerably increase the effect of CCW on activism. In addition, empirical research by Vlasceanu et al. (2024) indicates that climate knowledge moderates the behavioral influence of CCW, whereas the personal experience of natural disasters does not. This has implications for communication campaigns in that worry is more likely to lead to action when it is based on correct knowledge and social support.

Cumulatively, these studies weave a consistent narrative: climate emotions, and especially worry, empathy, guilt, and moral anger, are proximal psychological precursors of climate action and sustainable consumption. Situated within a conducive social and cognitive context—such as self-efficacy, environmental concern, and emotion regulation—these emotions activate effective change (Kesenheimer & Greitemeyer, 2021; Bahja & Hancer, 2021). But emotional saturation or despair can suppress action, and so it is necessary to examine not only the occurrence of affect, but also its valence, intensity, and regulation.

Extending on this basis, the present research examines how sympathy for the environment and concern about climate change, in interactions with personality and demographic moderators, shape sustainable consumer behavior. Using a Bayesian design, the research also examines whether these affective variables have linear or curvilinear effects and how they interact with individual differences to impact outcomes. In so doing, it adds to the emergent literature that conceives of emotional involvement as an active, multidimensional driver of ecological action.

2.3. Bayesian Methods in Sustainability Research

Bayesian statistical methodologies have progressed significantly in environmental and sustainability science owing to their inferential power, flexibility, and transparency (Wagenmakers et al., 2018). As compared to frequentist null hypothesis significance testing with p-values and large-sample approximations, Bayesian inference provides posterior probability distributions and allows for direct assessment of evidence in the form of Bayes factors. Such features have rendered Bayesian methods particularly appealing in fields like environmental psychology, where models tend to include complicated predictor frameworks, modest-to-small sample sizes, and theory-driven hypotheses requiring cumulative proof (Becht et al., 2024; Hornsey & Pearson, 2024; Ojala et al., 2021).

Increasingly, more studies use Bayesian regression to investigate the psychological underpinnings of sustainable behavior. For example, Becht et al. (2024) used curvilinear and linear Bayesian models to compare the psychological relationship between climate anxiety and pro-environmental behavior in teenagers. Applying Bayes factors (BFs) to compare models, they found moderate-to-strong support for a simple linear relationship in the majority of analyses (BFs ≈ 7–13), with limited evidence only for a curvilinear “eco-paralysis” effect. This modeling strategy provided a substitute for null hypothesis testing in the form of a formal comparison of theoretical hypotheses and greater interpretive clarity. Bayesian ANOVA and multiple regression were also used by Von Gal et al. (2024) to test predictors of climate action intention, wherein posterior inclusion probabilities and BFs were reported for all the variables. In this study, variables such as eco-consequences and eco-anger had inclusion BFs above 4.5, which was interpreted as “moderate-to-strong” evidence in favor of their predictive significance.

Bayesian methods also have been used for multilevel and geospatial models in sustainability science. Karimi-Malekabadi et al. (2024) used Bayesian geospatial modeling to examine regional moral values (e.g., fairness, purity) as predictors of environmental attitudes and home carbon emissions in 3000+ U.S. counties. Results indicate that moral values, even after adjusting for political orientation and regional covariates, are still strong predictors of sustainable action at the community level. In addition, Penker (2024) used Bayesian negative binomial multilevel models in a cross-national, cross-generational analysis of public-sphere environmental activism and found substantial period effects on environmental protests over time.

Apart from these application-specific uses, Bayesian modeling provides general methodological benefits in sustainability settings. First, it enables the inclusion of prior information—for instance, priors derived from meta-analytic estimates or earlier research—rendering analyses more theory-driven and sensitive (Bergh et al., 2021). This is especially beneficial for testing multifaceted, integrative models like personality–emotion–behavior chains in sustainability psychology. Second, Bayes factors provide a direct model comparison so that researchers can estimate how much more the data favor one theoretical model over another (e.g., emotion-only vs. personality-plus-emotion models) (Wagenmakers et al., 2018). Third, Bayesian regression is less affected by small-to-moderate sample sizes compared to frequentist methods and provides full posterior distributions, giving credible intervals that make statements about the true parameter uncertainty without relying on asymptotic assumptions.

For applied research, software packages such as JASP have made Bayesian techniques progressively more user-friendly. JASP supports the specification of user-defined priors, calculation of Bayes factors, and graphical examination of posterior distributions within an open-source environment that encourages transparency and replicability. Research by Reveco-Quiroz et al. (2022), demonstrative of this trend, applied Bayesian beta-regression and model-based variable selection to national survey sustainable consumption indexes, highlighting Bayesian modeling’s potential in enhancing prediction and inference in sustainability-oriented measures.

In spite of these advances, the use of Bayesian techniques on mainstream sustainability-related psychology remains relatively limited. There are encouraging applications, yet the literature is still evolving, and more work needs to be undertaken to institutionalize practices for prior selection, model comparison, and Bayes factor interpretation across research. Furthermore, relatively few studies have yet applied Bayesian approaches to mediation, moderation, or interaction modeling in environmental behavior models of this complexity, despite their appropriateness for these analyses. Our research fills this gap by using Bayesian regression modeling to estimate personality trait and climate emotion impacts on sustainable consumption simultaneously, thereby allowing us not only to examine direct relations but also to test the evidence for rival explanatory models. In this manner, we intend to push forward both the methodological and the substantive frontiers of psychological research on sustainability. To address these relationships, the following research questions were formulated:

RQ1: To what extent do HEXACO personality traits (e.g., Honesty–Humility), climate-related emotions (e.g., worry, empathy), and demographic variables predict sustainable consumption behavior in a Bayesian framework?

RQ2: What is the strength of the evidence for specific predictors—such as Honesty–Humility and Climate Change Worry—in explaining sustainable consumption behavior, based on Bayes Factors and posterior estimates?

RQ3: Do personality traits, climate worry, and environmental empathy predict the probability of engaging in climate action (yes/no), and how strong is this evidence under Bayesian logistic regression?

RQ4: Are there meaningful differences in sustainable consumption or emotional responses across gender, age, or education levels, and do these variables interact with psychological predictors?

RQ5: How do demographic factors such as gender and education interact in predicting sustainable consumption, and what is the strength of these interactions based on Bayesian model comparison?

3. Methodology

3.1. Conceptual Model and Rationale

The current research suggests an integrative psychological model that investigates how climate emotions and personality jointly forecast sustainable consumer behavior and climate action participation. Instead of repeating individual theories in a vacuum, this section brings the primary theoretical relationships established previously together and translates them into a testable framework. In particular, the model merges the most essential dimensions of the HEXACO model of personality—Honesty–Humility and Conscientiousness—with emotional reactions like Climate Change Worry (CCW) and environmental empathy, to study their unique and additive contribution to pro-environmental behavior. A second behavioral outcome of interest, climate action participation (binary), provides the opportunity for further insight into behavioral engagement over and above attitudinal measures. This model builds on current evidence in environmental and personality psychology but addresses key theoretical gaps in how stable personality factors integrate with dynamic emotional experiences in the prediction of climate-relevant behavior. This model relies on previous theoretical and empirical research and is intended to represent the influence of relatively stable dispositional variables and dynamic emotional states on climate-relevant behavior.

Whereas previous studies have investigated either personality traits or emotional reactions in isolation, their interplay has been less frequently considered. Accumulating evidence indicates that broad personality dimensions influence the emotional terrain on which individuals experience and react to environmental adversity. Personality traits like Honesty–Humility and Openness have been found to predispose individuals to feel intense moral emotions like eco-guilt and moral anger, which act as drivers of sustainable behavior (Kesenheimer & Greitemeyer, 2021; Soutter et al., 2020). For example, high Honesty–Humility can render a person more sensitive to environmental degradation and more likely to transcend self-interest for the collective good, whereas Conscientiousness can moderate how emotional upset, e.g., climate anxiety, is transduced into structured behavioral reactions instead of paralysis. On the other hand, people who are low in empathy or high in emotional instability might endure overwhelming climate anxiety but lack the psychological competence to convert it into action, and hence the emotion–behavior correlation deteriorates (Martin et al., 2022). Therefore, the model captures a trait–emotion interaction principle on which both the intensity and behavioral expression of climate-related emotions are influenced by individual differences.

The present research remedies this shortage by incorporating personality traits and climate emotions not just as main effect predictors but also as interacting factors. This provides the opportunity to test direct effects and moderated effects on sustainable consumption. By delineating direct and conditional paths, the model allows for the examination of whether personality has a facilitator or inhibitor effect on the influence of climate emotions on behavior, or the reverse.

Methodologically, this study uses a Bayesian regression approach to estimate model parameter strength and uncertainty. Bayesian approaches have numerous benefits, such as quantifying the evidence for null and alternative hypotheses, defining intricate predictor structures, and eliciting prior knowledge with data (Wagenmakers et al., 2018). This is especially useful for theory-driven research with various interrelating psychological constructs. Bayesian modeling also allows for formal model comparison via Bayes factors and fosters open, reproducible science. By employing this method, we become part of an increasing convention in environmental psychology to employ inferential techniques more appropriately matched to research on human behavior’s complexity and uncertainty (Wagenmakers et al., 2018; Karimi-Malekabadi et al., 2024).

In short, this conceptual model describes how personality (Honesty–Humility, Conscientiousness), emotional predictors (CCW, environmental empathy), and their interactions explain two important behavioral outcomes: sustainable consumption and participation in climate action. It responds to calls for more comprehensive, integrative frameworks that move beyond trait-based or emotion-based explanations in isolation, offering a nuanced account of how internal dispositions and emotional processes jointly shape sustainable action. By doing so, it informs both psychological theory and the design of behavior change interventions targeting environmentally significant behaviors. The conceptual model is illustrated in Figure 1.

3.2. Data Collection and Sampling

A quantitative cross-sectional design was used to investigate the interrelations of HEXACO personality dimensions, emotional climate reactions, and pro-environmental behaviors such as sustainable consumption and climate action involvement (Reveco-Quiroz et al., 2022; Von Gal et al., 2024). The design enabled the measurement of psychological and behavioral variables at one time point, which was appropriate for exploratory hypothesis testing without requiring a longitudinal follow-up (Penker, 2024; Berkhout et al., 2024).

Data were gathered using a standardized self-report questionnaire between June and October 2024, both online via Google Forms, and on paper, in an attempt to approach heterogeneous populations across all of Greece. The 75-item questionnaire (Appendix A) was translated from validated measures such as the HEXACO-60 (Ashton & Lee, 2009) and available scales of climate change concern, environmental empathy, and sustainable consumption (Huth et al., 2023; Kesmodel, 2018). All of the items were on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree). The questionnaire was divided into two sections: demographic information and psychological constructs.

The sampling method employed convenience and snowball techniques (Olsen & St George, 2004; Sandelowski, 2000). Initial participants were invited via environmental networks, university mailing lists, and professional websites, and then requested to pass on the survey to their networks. Participation was voluntary, unpaid, and anonymous, with inclusion criteria restricted to consent and age (18+). Online participants were contacted via email or social media invitations with the survey link, and offline paper-based data collection was organized through collaborating educators, NGO contacts, and local community facilitators.

Snowball sampling was employed to maximize coverage in hard-to-reach groups, such as young activists and those who resided in areas that were considered to be at climate-related risk. This served to capture diverging perspectives within the analysis. While this non-probability approach restricts generalizability, it is the conventional procedure in exploratory behavioral research for reaching diverse and otherwise inaccessible populations (Olsen & St George, 2004; Spector, 2019).

The total sample included 604 adult participants, well above the minimum for Bayesian multiple regression and generalized linear modeling with multiple predictors to ensure sufficient statistical power and model stability (Bergh et al., 2021; Ojala et al., 2021). Although the sample was skewed toward younger adults, participants over 40 were purposefully retained to reflect real-world diversity and to examine age-related patterns in sustainability behavior. Prior to analysis, responses with more than 20% missing data or patterns indicative of inattentive responding (e.g., straight-lining) were excluded. Data were inspected for completeness, normality, and outliers before analysis. A pilot study was carried out among a subsample of the participants to receive feedback on the clarity, linguistic adequacy, and cultural applicability of all items and changes were made accordingly.

Ethical principles were strictly adhered to. There was anonymous and voluntary participation, with the informed consent of all participants. This study was approved by the ethics review committee of the affiliated institution and was conducted in accordance with data protection law.

3.3. Measurement Scales

All of the constructs in this study were measured with well-validated self-report measures. Sustainable consumption (SC), the main dependent variable, was measured with a three-item scale based on Severo et al. (2021), emphasizing behavioral changes in terms of environmentally friendly purchases, reduction of waste, and climate-induced consumption change. Climate action participation (binary measure) was drawn from the items of Zeier and Wessa (2024), measuring the engagement of individuals in activism or environmentally adaptive measures for climate change mitigation. Climate Change Worry (CCW) was measured using a 10-item scale adapted from Stewart (2021), measuring concern regarding future climate impacts, media-seeking behaviors, and worry related to environmental hazards. Environmental empathy (EE) was measured using a four-item scale from Vlasceanu et al. (2024) and Wang et al. (2023), indicating affective concern and empathic identification with non-human life and nature. Personality traits were measured using the Honesty–Humility and Conscientiousness subscales of the HEXACO-60 questionnaire, in line with the theoretical framework of this study (Soutter et al., 2020). Demographic factors (age, gender, and education) were also considered in order to control the individual differences in behavioral outcomes. The answers were marked on five-point Likert scales from 1 (strongly disagree) to 5 (strongly agree). The reliability and validity of the scales were established through internal consistency measures (e.g., Cronbach’s α) and pilot testing for linguistic ease and cultural appropriateness in the Greek environment.

3.4. Sample Profile

The sample was also representative of a heterogeneous population in relation to gender, age, and education and consisted of 604 participants. The sample was gender-balanced as well, with males constituting 51.7% of the sample, whereas females constituted 48.3%. Looking at age, the most prominent group was that of individuals in the 26–30 years age bracket, comprising 39.2%, or N = 237, followed by the 31–40 years age bracket, comprising 25.0%, or N = 151, and the 18–25 years age bracket, comprising 23.3%, or N = 141. The other age brackets were the 41–59 age bracket, comprising 10.4% (N = 63), and the 60+ bracket, comprising 2.0% (N = 12). Education-wise, the highest percentage of the group had a Bachelor’s degree, at 36.1% (N = 218); a High School diploma, at 29.3% (N = 177); or a Master’s degree, at 26.7% (N = 161). The Doctoral category was composed of 3.8% PhD Candidates (N = 23) and 4.1% individuals with a Doctoral degree (N = 25). An examination of this distribution creates a diverse sample, which can be helpful in gaining insights along demographic lines that can affect the outcome of major variables in the study. An overview of the sample’s demographics can be seen in

Table 1. Sample profile.

		Frequency (N)	Percentage
Gender	Male	312	51.7%
	Female	292	48.3%
Age	18–25	141	23.3%
	26–30	237	39.2%
	31–40	151	25.0%
	41–59	63	10.4%
	60+	12	2.0%
Education	High School	177	29.3%
	Bachelor’s Degree	218	36.1%
	Master’s Degree	161	26.7%
	PhD Candidate	23	3.8%
	Doctoral	25	4.1%

.

4. Statistical Analysis Plan

4.1. Bayesian Analysis Framework

In order to examine the psychological predictors of sustainable consumption behavior, we used a Bayesian multiple regression model in JASP (Version 0.19.3) (JASP Team, 2025). Bayesian analyses offer several advantages over conventional frequentist approaches, especially when dealing with the complex, theory-driven models common in sustainability psychology. Bayesian inference, in contrast to null hypothesis significance testing, yields complete posterior distributions of model parameters, allows researchers to quantify evidence both for and against hypotheses, and promotes model comparison with Bayes factors (Wagenmakers et al., 2018; Karimi-Malekabadi et al., 2024).

This analytical approach is in line with recent uses in sustainability science and environmental psychology. Here, for example, Becht et al. (2024) employed Bayesian model comparison to ascertain whether climate anxiety predicted adolescents’ pro-environmental behavior via linear or curvilinear effects, and they found the linear model to be more strongly supported in two of three studies (BFs ≈ 7–13). Also, Von Gal et al. (2024) conducted Bayesian regression and ANOVA to contrast the extent to which climate-related emotion and message frames forecasted behavior intentions and reported BF_incl values, posterior inclusion odds, and 95% credible intervals for all predictors.

The analysis involves Bayesian multiple linear regression with sustainable consumption (SC) as the dependent variable. The predictors were personality traits of the HEXACO model—Honesty–Humility and Conscientiousness—and emotion predictors (Climate Change Worry and environmental empathy) and control variables (sex, age and education). All predictors were mean-centered prior to analysis, in line with previous research (Becht et al., 2024).

4.2. Prior Distribution and Sensitivity Checks

Priors were specified based on conventional criteria. We used the Jeffreys–Zellner–Siow (JZS) prior for all regression coefficients, a Cauchy distribution with mean zero and a scale parameter r = 0.354, advocated for by Wagenmakers et al. (2018) and utilized by Von Gal et al. (2024). This option corresponded to a weakly informative prior around zero, with the anticipation that the majority of effects would be modest in size but with big effects being plausible—a reasonable compromise for psychological data. Model priors were a Beta(1,1) distribution, indicating a uniform inclusion probability for all candidate predictors. To evaluate sensitivity, analyses with wider priors (e.g., r = 0.5) were conducted to check the stability of parameter estimates and Bayes factors across different assumptions. These checks demonstrated little variation in model preference or posterior estimates, ensuring the validity of findings. Informative priors were not employed as this study was exploratory (Bergh et al., 2021; Ojala et al., 2021). This maximizes transparency and replicability and is consistent with recent guidelines in Bayesian psychological science against strong priors when theoretical predictions are uncertain.

4.3. Inference and Model Comparison

Inference was drawn by comparing the complete model and the null model using BF₁₀, and by examining the inclusion Bayes factor (BF_incl) for each predictor in order to determine its relative contribution. Interpretation relied on standard guidelines (Berkhout et al., 2024), where BF₁₀ values greater than 3 were taken to reflect moderate evidence for a model, with values greater than 10 providing strong support, and values less than 1/3 supporting the null hypothesis. In addition to Bayes factors, we reported posterior means and 95% credible intervals as an index of estimation uncertainty to allow an interpretable evidence strength. This is in accordance with current reporting practices in Bayesian psychology (Bergh et al., 2021; Wagenmakers et al., 2018; Karimi-Malekabadi et al., 2024).

4.4. Methodological Justification and Precedent

Our methodology is consistent with best practice in the field of sustainability research today, wherein Bayesian methods have grown more frequently and are used to estimate high-order uncertainty about complex psychological predictors of environmental behavior. To provide an example, Becht et al. (2024) used Bayesian regression to test whether climate anxiety predicts pro-environmental behavior among adolescents via linear or curvilinear processes, with BFs ranging from 7 to 13 as moderate-to-strong support for a linear process. Likewise, Von Gal et al. (2024) employed Bayesian model averaging and ANCOVA to examine the effect of climate messaging, reporting BF_incl values, posterior inclusion probabilities, and 95% credible intervals directly. Further uses illustrate the versatility of the approach: Penker (2024) used Bayesian multilevel analysis in analyzing generational trends in climate activism; Karimi-Malekabadi et al. (2024) performed Bayesian geospatial regression in predicting regional carbon footprints from moral values based on weakly informative priors; and Wang et al. (2023) used Bayesian multinomial logistic regression in an investigation of low-carbon shopping behavior. Sijtsma et al. (2023) also demonstrated the strength of Bayesian inference through its application in investigating HEXACO traits’ predictive ability for trust behavior and concluded a Bayes factor of approximately 0.02 as being strong evidence in support of the null. Cumulatively, these studies showcase Bayesian analysis’ methodological rigor and inferential benefits in modeling the presence as well as absence of effects in sustainability psychology (Becht et al., 2024; Ojala et al., 2021; Latkin et al., 2025).

5. Data Analysis and Results

A Bayesian Pearson correlation analysis was conducted to examine the associations among sustainable consumption (SC), HEXACO personality traits, climate-related emotions, and demographic variables. Pearson’s correlation coefficients, 95% credible intervals (CrIs), and Bayes factors (BF₁₀) were reported to assess the direction, strength, and evidential support for each relationship, as shown in

Table 2. Correlation table.

	SC	CCW	EE	HH	E	X	A	C	O	Gender
SC	-	[0.011, 0.169]	[0.002, −0.157]	[−0.008, 0.150]	[−0.100, −0.059]	[0.058, −0.101]	[0.044, −0.115]	[0.108, −0.051]	[0.014, −0.145]	[−0.025, −0.182]
CCW	(0.090, 0.599)	-	[0.070, −0.089]	[0.008, 0.124]	[−0.140, −0.019]	[0.058, −0.102]	[0.018, −0.141]	[0.077, −0.082]	[0.063, −0.097]	[0.143, −0.016]
EE	(−0.078, 0.320)	(−0.009, 0.052)	-	[−0.120, 0.039]	[−0.030, 0.129]	[0.048, −0.111]	[0.115, −0.044]	[0.108, −0.051]	[0.060, −0.099]	[0.145, −0.014]
HH	(0.071, 0.236)	(0.045, 0.094)	(−0.041, 0.084)	-	[−0.080, 0.080]	[0.077, −0.082]	[0.108, −0.051]	[0.073, −0.086]	[0.226, 0.071]	[0.022, −0.137]
E	(0.020, 0.058)	(0.061, 0.157)	(0.050, 0.107)	(−8.79 × 10−5, 0.051)	-	[0.035, −0.124]	[0.060, −0.099]	[0.118, −0.041]	[−0.138, −0.290]	[0.068, −0.091]
X	(−0.021, 0.058)	(−0.022, 0.059)	(−0.032, 0.069)	(−0.104, 1.316)	(−0.045, 0.094)	-	[0.145, −0.014]	[0.102, −0.057]	[0.104, −0.055]	[0.106, −0.053]
A	(−0.036, 0.075)	(−0.062, 0.163)	(0.036, 0.075)	(0.131, 9.139)	(−0.036, 0.075)	(−0.030, 0.067)	-	[−0.041, −0.198]	[0.106, −0.054]	[0.086, −0.074]
C	(0.029, 0.066)	(−0.002, 0.051)	(0.029, 0.066)	(−0.007, 0.052)	(0.029, 0.066)	(0.023, 0.059)	(−0.121, 4.256)	-	[0.106, −0.054]	[0.058, −0.101]
O	(−0.066, 0.189)	(−0.017, 0.056)	(−0.020, 0.057)	(0.150, 45.840) **	(−0.216, 85,617.653) ***	(0.025, 0.061)	(0.026, 0.063)	(0.055, 0.125)	-	[0.087, −0.072]
Gender	(−0.104, 1.346)	(0.064, 0.174)	(0.066, 0.191)	(−0.058, 0.139)	(−0.012, 0.053)	(0.027, 0.063)	(0.006, 0.052)	(−0.021, 0.059)	(0.008, 0.052)	-

Note: Below the diagonal are Pearson’s r with Bayes factor (BF₁₀) in parentheses. Above the diagonal are 95% credible intervals. Asterisks indicate evidence strength: BF₁₀ > 10 (*), >30 (**), >100 (***).

.

The relationship between Climate Change Worry (CCW) and sustainable consumption was positive and small (r = 0.09, BF₁₀ = 0.60), with the 95% CrI [0.011, 0.169] containing zero, offering anecdotal support for this association. Honesty–Humility (r = 0.071, BF₁₀ = 0.24) and Conscientiousness (r = 0.029, BF₁₀ = 0.08) also shared similarly small correlations with SC, and Bayes factors did not yield enough evidence to include them. Emotionality (r = 0.020, BF₁₀ = 0.06) and Extraversion (r = −0.021, BF₁₀ = 0.06) were not consistently related to SC.

There was no strong evidence for an association between SC and environmental empathy (r = −0.078, BF₁₀ = 0.32) or with the demographic variable gender (r = −0.104, BF₁₀ = 1.35). The 95% CrIs for all these correlations contained zero, and the Bayes factors were not greater than the threshold for moderate evidence (i.e., BF₁₀ > 3).

The most robust evidential support in the data was observed between Openness and Emotionality (r = −0.216, BF₁₀ = 85.618) and Openness and Honesty–Humility (r = 0.150, BF₁₀ = 45.84), neither of which had 95% CrIs containing zero. Both correlations did not involve the dependent variables but informed us about the latent trait structure and possible collinearity for regression modeling.

In total, the Bayesian correlation analysis was unable to discover any substantial correlations between sustainable consumption and any predictor with BF₁₀ ≥ 3. Nevertheless, the results are useful initial pointers for variable inclusion in later regression analyses.

5.1. Bayesian Multiple Linear Regression on Sustainable Consumption

In order to compare demographic and psychological predictors of sustainable consumption, a Bayesian model comparison method was used in JASP. Personality (Honesty–Humility, Conscientiousness, Emotionality, Extraversion, Agreeableness, Openness), climate emotions (Climate Change Worry [CCW], environmental empathy [EE]), and demographic (gender, age, education) variables were added as possible predictors. Bayesian linear regression models were estimated with the Jeffreys–Zellner–Siow (JZS) prior (r = 0.354), and model probabilities were assessed via Bayes adaptive sampling (BAS), comparing against the null model (education alone).

Table 3. Model comparisons for sustainable consumption.

Models	P(M)	P(M|Data)	BFM	BF10	R2
education	0.008	0.137	20.744	1.000	0.034
gender + education	0.002	0.049	33.885	1.789	0.043
CCW + gender + education	5.051 × 10−4	0.042	86.582	4.599	0.052
CCW + education	0.002	0.040	27.471	1.464	0.042
Openness + education	0.002	0.015	10.182	0.557	0.039
CCW + Openness + gender + education	2.525 × 10−4	0.015	59.466	3.248	0.057
EE + education	0.002	0.014	9.554	0.523	0.039
HonestyHumility + Conscientiousness + EE + CCW + Emotionality + Extraversion + Agreeableness + Openness + gender + education + age	0.083	0.014	0.155	0.009	0.071
HonestyHumility + education	0.002	0.014	9.227	0.505	0.039
CCW + gender + education + age	2.525 × 10−4	0.012	47.690	2.612	0.056

Note: Table reports Bayesian model comparison results predicting sustainable consumption (SC). P(M): prior model probability; P(M|data): posterior model probability given the data; BF_m: Bayes factor comparing each model to all other models; BF₁₀: Bayes factor comparing each model to the null model (education only); R²: explained variance.

shows the ten highest fitting models based on Bayes factors (BF₁₀) and posterior model probabilities. The most well-supported model was the one with education as the sole predictor of SC, with moderate evidence for the alternative compared to the null (BF₁₀ = 1.00) and posterior model probability of P(M|data) = 0.137. The inclusion of gender provided a modest increase in explanatory power (R² = 0.043), yet only anecdotal evidence for the leading model (BF₁₀ = 1.789).

The inclusion of Climate Change Worry (CCW) alongside gender and education yielded a model with moderate evidence (BF₁₀ = 4.599) and higher explained variance (R² = 0.052), suggesting CCW contributes incrementally to the prediction of sustainable consumption. Models including broader HEXACO traits (e.g., Honesty–Humility, Conscientiousness, Openness) were worse than more parsimonious models, with substantially lower posterior probabilities and Bayes factors (e.g., BF₁₀ = 0.009 for the full model).

Generally, the data support a modest role for climate change concern and education in predicting sustainable consumption behavior. Somewhat unexpectedly, the model incorporating CCW, sex, and education had more than four times more support than the null model did, indicating that emotional concern regarding climate change, independent of demographics, is perhaps a valuable predictor of pro-environmental consumption.

Posterior estimates for the regression coefficients give additional information on the most probable predictors of sustainable consumption behavior (

Table 4. Posterior summaries for regression coefficients predicting sustainable consumption (SC).

Coefficient	P(incl)	P(excl)	P(incl|data)	P(excl|data)	BFinclusion	Mean	SD	Lower CI 95%	Upper CI 95%
Intercept	1.000	0.000	1.000	0.000	1.000	4.136	0.024	4.089	4.184
HH	0.500	0.500	0.349	0.651	0.537	0.020	0.035	0.000	0.110
C	0.500	0.500	0.184	0.816	0.225	0.003	0.016	−0.010	0.052
EE	0.500	0.500	0.330	0.670	0.492	−0.022	0.039	−0.120	0.001
CCW	0.500	0.500	0.561	0.439	1.280	0.056	0.060	0.000	0.166
E	0.500	0.500	0.167	0.833	0.201	3.882 × 10−4	0.015	−0.015	0.055
X	0.500	0.500	0.171	0.829	0.206	−0.002	0.013	−0.019	0.038
A	0.500	0.500	0.181	0.819	0.220	−0.003	0.014	−0.041	0.024
O	0.500	0.500	0.393	0.607	0.648	−0.025	0.038	−0.113	0.000
gender	0.500	0.500	0.589	0.411	1.434	−0.065	0.066	−0.181	0.000
education	0.500	0.500	0.999	0.001	785.186	0.103	0.024	0.057	0.149
age	0.500	0.500	0.310	0.690	0.449	−0.011	0.021	−0.064	1.176 × 10−4

Note: P(incl) and P(excl): prior inclusion/exclusion probabilities; P(incl|data) and P(excl|data): posterior inclusion/exclusion probabilities; BF inclusion: Bayes factor for including the predictor versus excluding it. Mean, SD, and 95% credible interval represent the posterior distribution of each coefficient. Coefficients with credible intervals not including zero or BF inclusion > 3 suggest meaningful effects.

). As can be gleaned from Table 2, education was the most robust and stable predictor with an inclusion Bayes factor (BF_inclusion) of 785.19, indicating conclusive evidence for its impact. The posterior mean for the education coefficient was 0.103 with a 95% credible interval that did not include zero [0.057, 0.149], indicating that greater educational attainment was credibly related to lower sustainable consumption scores.

Climate Change Worry (CCW) also showed moderate evidence for inclusion (BF_inclusion = 1.28), with a posterior mean of 0.056 and a credible interval narrowly excluding zero [0.000, 0.166]. This indicates that participants who are more worried about climate change are somewhat more likely to consume sustainably, although the evidence strength is still limited. Gender also had anecdotal support for inclusion (BF_inclusion = 1.434), a negative posterior mean (−0.065), and a credible interval just below zero [−0.181, 0.000], suggesting that males are less sustainable in their consumption than females, although this result is not reliable. The other predictors, Honesty–Humility, environmental empathy, and Openness, all had BF_inclusion statistic values between 0.5 and 0.6, indicating no strong evidence either way for including them. Credible intervals for the coefficients either included zero or had lower bounds close to zero, nullifying any case for credible effects. The remaining personality dimensions (Conscientiousness, Emotionality, Extraversion, Agreeableness) and age all had BF_inclusion values < 0.5 and broad credible intervals spanning zero, indicating evidence against their importance in this model.

Figure 2 plots marginal posterior inclusion probabilities for each of the predictors in the Bayesian linear regression model of sustainable consumption. The dashed vertical line indicates the prior inclusion probability of 0.50 under a uniform model prior (i.e., all predictors had equal a priori chances of being included). Bars extending to the right of this line suggest greater empirical support for inclusion based on the observed data.

Of all the predictors, education had the highest posterior inclusion probability—far exceeding the prior—showing definitive evidence of being in the model. It was followed closely by gender and climate change concern (CCW), which both recorded posterior inclusion probabilities greater than 0.50, offering moderate to strong evidence of their predictive utility.

Conversely, all three variables Emotionality, Extraversion, and Agreeableness also had inclusion probabilities far lower than the prior threshold, indicating that there was negligible or no evidence within the data that would include these predictors in the model. Conscientiousness, age, and environmental empathy (EE) also failed to pass the prior inclusion threshold regardless of theory relevance.

In total, this inclusion probability profile acknowledges the contribution of demographics (gender, education) and climate-related affective engagement (CCW) in describing sustainable consumption and disputes the role of numerous aspects of personality after considering model uncertainty.

5.2. Bayesian Logistic Regression on Climate Action Participation

We performed a Bayesian logistic regression to determine demographic and psychological predictors of climate action participation. Models were compared based on Bayes factors (BF₁₀), posterior model probabilities, and variance explained (R²).

The null model with no predictors received the highest posterior model probability (P(M|data) = 0.408) and was used as the reference model (BF₁₀ = 1.00). The full model with all the demographic and psychological predictors received little support (BF₁₀ = 0.057, P(M|data) = 0.023, R² = 0.004), demonstrating that the data favored the null model over any single predictor or combination (

Table 5. Model comparison for climate action participation.

Models	P(M)	P(M|data)	BFM	BF10	R2
null model	0.083	0.408	7.583	1.000	0.000
HonestyHumility + Emotionality + Extraversion + Agreeableness + Conscientiousness + Openness + EE + CCW + gender + education + age	0.083	0.023	0.262	0.057	0.004
Agreeableness	0.008	0.017	2.262	0.458	0.001
Extraversion	0.008	0.016	2.124	0.430	0.001
CCW	0.008	0.015	1.957	0.397	0.001
HonestyHumility	0.008	0.013	1.705	0.346	0.000
Emotionality	0.008	0.013	1.702	0.346	0.000
EE	0.008	0.013	1.678	0.341	0.000
Conscientiousness	0.008	0.013	1.675	0.340	0.000
gender	0.008	0.013	1.672	0.340	0.000

Note: Table summarizes Bayesian model comparison results for predicting the binary outcome of climate action participation (CAP). P(M) and P(M|data): prior and posterior model probabilities; BF_m: Bayes factor comparing each model to all others; BF₁₀: Bayes factor relative to the null model (intercept-only); R²: McFadden’s pseudo R².

).

Of the individual predictor models, models featuring Agreeableness (BF₁₀ = 0.458), Extraversion (BF₁₀ = 0.430), and Climate Change Worry (CCW) (BF₁₀ = 0.397) had the highest relative model evidence. These all fell short of BF₁₀ > 1, though, offering anecdotal-to-weak evidence against their value as predictors.

No model achieved moderate levels of evidence (BF₁₀ > 3) compared to the null. The results therefore indicate that, given current data and priors, there is no strong evidence that any individual psychological or demographic variable significantly adds to the prediction of climate action engagement over and above the null model.

Posterior summaries indicated that there was no predictor with strong evidence for inclusion, as all Bayes factors for inclusion (BF_incl) were less than the standard cut-off of 3, and all 95% credible intervals (CrIs) contained zero (

Table 6. Posterior summaries of coefficients from Bayesian logistic regression predicting climate action participation (CAP).

Coefficient	P(incl)	P(excl)	P(incl|data)	P(excl|data)	BFinclusion	Mean	SD	Lower CI 95%	Upper CI 95%
Intercept	1.000	0.000	1.000	0.000	1.000	0.122	0.776	−1.937	1.246
HH	0.500	0.500	0.221	0.779	0.284	0.003	0.032	−0.061	0.096
E	0.500	0.500	0.222	0.778	0.285	0.003	0.033	−0.052	0.112
X	0.500	0.500	0.237	0.763	0.310	0.007	0.031	−0.038	0.106
A	0.500	0.500	0.240	0.760	0.316	0.008	0.032	−0.063	0.115
C	0.500	0.500	0.220	0.780	0.282	−0.002	0.029	−0.079	0.076
O	0.500	0.500	0.219	0.781	0.281	5.141 × 10−4	0.031	−0.071	0.096
EE	0.500	0.500	0.221	0.779	0.283	−0.003	0.038	−0.105	0.091
CCW	0.500	0.500	0.231	0.769	0.300	0.008	0.044	−0.074	0.129
gender	0.500	0.500	0.220	0.780	0.282	0.002	0.042	−0.086	0.104
education	0.500	0.500	0.219	0.781	0.281	5.122 × 10−4	0.020	−0.049	0.060
age	0.500	0.500	0.219	0.781	0.281	−5.180 × 10−4	0.021	−0.054	0.055

Note: Table shows posterior summaries for each predictor’s effect on climate action participation (CAP). P(incl) and P(excl): prior inclusion/exclusion probabilities; P(incl|data) and P(excl|data): posterior probabilities; BF inclusion: Bayes factor for including vs. excluding the predictor. Mean, SD, and 95% credible interval reflect the posterior distribution of the log-odds coefficient. Predictors with BF inclusion > 3 or CIs that do not include 0 indicate notable effects.

). Agreeableness, for example, had the highest probability of inclusion (P(incl|data) = 0.240), yet the Bayes Factor for inclusion was modest (BF_incl = 0.316), and the posterior 95% CrI [−0.063, 0.115] contained zero, suggesting no credible effect. Climate Change Worry (CCW) had a posterior mean of 0.008 and a CrI of [−0.074, 0.129], again showing no significant contribution. The other predictors (e.g., Honesty–Humility, Emotionality, EE, gender, education, age) all had posterior inclusion probabilities near or below 0.22, and the accompanying CrIs included zero, offering anecdotal-to-weak support for inclusion.

These findings indicate that none of the variables in question singled out in this model situation offer strong evidence of a significant predictive relationship with involvement in awareness actions related to the climate.

As can be seen from Figure 3, all of the predictors featured posterior inclusion probabilities less than the prior threshold value of 0.50, which meant the data were not sufficient in their support to include them in the logistic regression model of climate action participation. The education, sex, and personality dimensions of Honesty–Humility and Climate Change Worry had moderate inclusion probabilities (~0.22–0.24), which were still far from the values typically accepted to indicate strong effects (i.e., >0.75). This graphical support supplements the numerical model comparisons, together favoring the null model and the lack of strong individual predictors in this instance.

5.3. Group Comparisons and Differences

5.3.1. Gender Differences

Bayesian independent samples t-tests were used to examine gender differences in sustainable consumption (SC), Climate Change Worry (CCW), and environmental empathy (EE). For SC, the Bayes factor (BF₁₀ = 2.25) offered anecdotal-to-moderate evidence for a gender difference. For CCW (BF₁₀ = 0.30) and EE (BF₁₀ = 0.33), the evidence was against the alternative hypothesis, indicating there was no credible difference between groups for these variables (

Table 7. Bayesian independent samples t-test for gender differences.

	BF10	Error %
SC	2.249	0.009
CCW	0.302	0.064
EE	0.331	0.058

).

Bayesian inference provides moderate evidence for gender association with SC behavior differences. The strongest evidence for the alternative hypothesis (H₁: gender differences in SC) is at a prior width of 0.1583 and a maximum BF₁₀ = 4.515. With the default user prior (r = 0.707), the BF₁₀ = 2.249, providing moderate evidence for gender differences. Wider priors decrease the level of evidence to the anecdotal range. This finding is resilient to past assumptions to a moderate degree, though it is attenuated with broader priors. The result encourages the ongoing investigation of gender as a legitimate demographic moderator in models of sustainable behavior.

Visual inspection is consistent with the Bayesian analysis, which identifies a small but persistent trend for women to have more sustainable consumption scores (Figure 4 and Figure 5). There is still a considerable overlap of distributions, highlighting that gender differences, although present, are not pronounced.

5.3.2. The Effect of Education on Sustainable Consumption

In order to check if sustainable consumption (SC) varied across education levels, a Bayesian ANOVA was performed. The predictor model including education was very strongly preferred over the null model, BF₁₀ = 52.77, suggesting very strong evidence in favor of an effect of education on SC. The posterior model probability was P(M|data) = 0.981, whereas for the null it was P(M|data) = 0.019. The Bayes factor for inclusion (BFinclusion) for education was 52.77, which meant that the inclusion of education in the model was far more probable than exclusion. The model-averaged explained variance was limited, with an R² = 0.033, and 95% CI [0.008, 0.063]. The findings indicate that educational level is an appropriate predictor of sustainable consumption behavior among the sample, although the amount of explained variance is still small (

Table 8. Bayesian ANOVA results for the effect of education on sustainable consumption.

Section	Model/Effect	P(M)	P(M|data)	BFm	BF10	Error %	P(incl)	P(excl)	P(incl|data)	P(excl|data)	BF incl	Mean R2	95% CI
Model Comparison	education	0.500	0.981	52.767	1.000
Model Comparison	null model	0.500	0.019	0.019	0.019	0.014
Analysis of Effects—SC	education						0.500	0.500	0.981	0.019	52.767
Model-Averaged R2	R2											0.033	(0.008, 0.063)

Note: P(M) = prior model probability; P(M|data) = posterior model probability given data; BF_m = Bayes factor for each model against all others; BF₁₀ = Bayes factor compared to the null model; BFinclusion = inclusion Bayes factor for each effect; R² = model-averaged explained variance; 95% CI = 95% credible interval for R².

).

To investigate more closely the significant main effect of education on sustainable consumption (SC) which was discovered, Bayesian post hoc tests were conducted.

Table 9. Bayesian post hoc pairwise comparisons of education levels on sustainable consumption.

Post Hoc Comparisons—Education		Prior Odds	Posterior Odds	BF10,U	Error %
High School	Bachelor’s Degree	0.320	0.128	0.400	0.044
	Master’s Degree	0.320	1.104	3.457	0.006
	PhD Candidate	0.320	2.381	7.454	0.004
	Doctoral	0.320	21.542	67.421	4.900 × 10−8
Bachelor’s Degree	Master’s Degree	0.320	0.068	0.212	0.077
	PhD Candidate	0.320	0.589	1.843	0.008
	Doctoral	0.320	3.583	11.215	3.568 × 10−7
Master’s Degree	PhD Candidate	0.320	0.326	1.020	0.010
	Doctoral	0.320	1.713	5.362	0.005
PhD Candidate	Doctoral	0.320	0.102	0.320	0.007

Note: Posterior odds are corrected for multiple testing. Bayes factors (BF₁₀,_U) are uncorrected and interpreted as evidence in favor of a difference between groups. Values > 3 indicate moderate, >10 strong, and >30 very strong evidence for H₁. Prior: default Cauchy (0, r = 1/√2).

provides an overview of the pairwise comparisons with default Cauchy priors and multiple-testing-corrected posterior odds.

The results provided extremely strong evidence for differences between the Doctoral and a number of lower education levels. More specifically, the High School and Doctoral comparison returned a Bayes factor (BF₁₀) of 67.42, and Bachelor’s and Doctoral returned BF₁₀ = 11.22, both providing conclusive support for a group difference. Moderate evidence was also found for Master’s and Doctoral (BF₁₀ = 5.36), and High School and PhD Candidate (BF₁₀ = 7.45). On the other hand, other comparisons (e.g., High School vs. Bachelor’s, PhD Candidate vs. Doctoral) resulted in anecdotal or no data, with BF₁₀ < 3.

These results affirm that education at higher levels, i.e., Doctoral-level, is linked to more sustainable consumption, validating the hypothesis that education is a significant predictor of pro-environmental behaviors.

Figure 6 presents the posterior distributions of sustainable consumption (SC) across education levels. As is apparent from the plot, Doctoral degree holders have higher SC than High School graduates with non-overlapping 95% credible intervals for the presence of such a difference. These depicted differences are in line with the results of the Bayesian ANOVA and post hoc tests of extreme evidence for differences between Doctoral and lower levels of education (e.g., BF₁₀ = 67.42 for Doctoral vs. High School). The plot gives a graphical overview of the distribution of model-estimated group effects, confirming the finding that higher education level is linked to more sustainable consumption behavior.

5.3.3. Age Group Differences in Sustainable Consumption

We conducted a Bayesian ANOVA to determine if sustainable consumption (SC) varied as a function of age group (

Table 10. Bayesian ANOVA results for the effect of age on sustainable consumption.

Section	Model/Effect	P(M)	P(M|data)	BFm	BF10	Error %	P(incl)	P(excl)	P(incl|data)	P(excl|data)	BF incl	Mean R2	95% CI
Model Comparison	null model	0.500	0.873	6.852	1.000
Model Comparison	age	0.500	0.127	0.146	0.146	0.004
Analysis of Effects—SC	age						0.500	0.500	0.127	0.873	0.146
Model-Averaged R2	R2											0.002	(0.000, 0.023)

Note: P(M) = prior model probability; P(M|data) = posterior model probability given data; BF_m = Bayes factor for each model against all others; BF₁₀ = Bayes factor compared to the null model; BFinclusion = inclusion Bayes factor for each effect; R² = model-averaged explained variance; 95% CI = 95% credible interval for R².

). Model comparison revealed that the null model was very much favored over the age model, BF₁₀ = 0.146, with moderate evidence for the null hypothesis that SC does not vary by age. The posterior probability for the null model was P(M|data) = 0.873, whereas P(M|data) = 0.127 was the posterior probability for the age model.

Effect analysis also favored the marginal contribution of age with a Bayes factor for inclusion, BFinclusion = 0.146, and a posterior inclusion probability of merely 0.127. Model-averaged R² was 0.002 with a 95% credible interval of [0.000, 0.023], reflecting exceedingly trivial explained variance in SC based on age.

These findings offer moderate confirmation of the lack of age-based differences in sustainable consumption within the sample.

Post hoc Bayesian pairwise comparisons by age group revealed no significant pairwise differences in sustainable consumption (

Table 11. Bayesian post hoc comparisons across age groups for sustainable consumption.

Post Hoc Comparisons—Age		Prior Odds	Posterior Odds	BF10,U	Error %
18–25	26–30	0.320	0.369	1.154	0.017
	31–40	0.320	0.661	2.069	0.010
	41–59	0.320	0.065	0.205	0.050
	60+	0.320	0.154	0.481	0.005
26–30	31–40	0.320	0.041	0.129	0.120
	41–59	0.320	0.072	0.225	0.052
	60+	0.320	0.100	0.314	0.006
31–40	41–59	0.320	0.104	0.325	0.037
	60+	0.320	0.097	0.304	0.006
41–59	60+	0.320	0.124	0.389	0.005

Note: All comparisons use default Bayesian t-tests with a Cauchy (0, r = √2⁻¹) prior. BF₁₀,_U indicates uncorrected Bayes factors for each pairwise comparison. Values above 3 suggest moderate evidence for a difference.

). The most compelling contrast was between the age range 18–25 and 31–40 (BF₁₀ = 2.069), which provided anecdotal-to-moderate evidence for a difference in SC in the younger adults’ favor. All other pairwise comparisons resulted in Bayes factors less than 1, in favor of the null hypothesis. The implications are that age differences in sustainable consumption in this sample are perhaps small or variable.

These results suggest that while SC appears visually differentiated in some age categories (see Figure 7), the Bayesian evidence does not strongly support a systematic effect of age on sustainable consumption behavior.

5.4. Interaction Between Gender and Education on Sustainable Consumption

In order to test the interaction effect of gender and education on sustainable consumption (SC), a Bayesian ANCOVA with emotional engagement (EE), Climate Change Worry (CCW), and HEXACO personality traits as covariates was performed (

Table 12. Interaction effects of gender and education on sustainable consumption.

Section	Model/Effect	P(M)	P(M|data)	BFm	BF10	Error %
Model Comparison	CCW + gender × education	0.125	0.411	4.881	1.000
	EE + CCW + gender * education	0.125	0.270	2.594	0.658	2.446
	gender * education	0.125	0.270	1.655	0.465	2.571
	EE + gender * education	0.125	0.128	1.023	0.310	5.525
	CCW	0.125	1.549 × 10−5	1.085 × 10−4	3.771 × 10−5	2.736
	EE + CCW	0.125	1.206 × 10−5	8.443 × 10−5	2.936 × 10−5	2.411
	null model (incl. all covariates)	0.125	6.306 × 10−6	4.414 × 10−5	1.535 × 10−5	1.586
	EE	0.125	5.138 × 10−6	3.597 × 10−5	1.251 × 10−5	2.468

). Model comparison indicated that the model with the interaction term (gender × education) and CCW had the highest posterior model probability, P(M|data) = 0.411, and outperformed all the other models with a Bayes factor of BF₁₀ = 1.000. The model performed better than the null model (with the covariates only) and competing models without the interaction term.

In the analysis of effects, the gender by education interaction was strongly supported with a posterior inclusion probability of P(incl|data) = 1.000 and a Bayes factor for inclusion of BF_incl = 25,641.56, which shows strong evidence for including the interaction (

Table 13. Gender by education interaction effects.

Effect	P(incl)	P(excl)	P(incl|data)	P(excl|data)	BF incl	Mean R2	Lower 95% CI	Upper 95% CI
gender * education	0.500	0.500	1.000	3.900 × 10−5	25,641.556
EE	0.500	0.500	0.398	0.602	0.661
CCW	0.500	0.500	0.681	0.319	2.137
Model-Averaged R2	R2					0.123	0.076	0.176

). CCW also provided moderate evidence for inclusion (BF_incl = 2.14), but EE did not have substantial support (BF_incl = 0.66). Model-averaged predictions indicated a mean R² = 0.123 (95% CI [0.076, 0.176]), indicating that the predictors and interactions included in the models accounted for around 12.3% of SC variance.

In order to more closely explain the two-way interaction influence of gender and education level on sustainable consumption (SC), Bayesian post hoc tests were carried out with independent samples t-tests with default Cauchy priors (r = 1/√2). The comparisons showed particular group differences, and Bayes factors (BF₁₀) were included to provide evidence strength for every comparison (

Table 14. Post hoc comparisons—gender and education.

Comparison	Prior Odds	Posterior Odds	BF10,u	Error %
GENDER
Male vs. Female	1.000	2.249	2.249	0.009
EDUCATION
High School vs. Bachelor’s Degree	0.320	0.128	0.400	0.044
High School vs. Master’s Degree	0.320	1.104	3.457	0.006
High School vs. PhD Candidate	0.320	2.381	7.454	0.004
High School vs. Doctoral	0.320	21.542	67.421	4.900 × 10−11
Bachelor’s Degree vs. Master’s Degree	0.320	0.068	0.212	0.077
Bachelor’s Degree vs. PhD Candidate	0.320	0.589	1.843	0.008
Bachelor’s Degree vs. Doctoral	0.320	3.583	11.215	3.568 × 10−7
Master’s Degree vs. PhD Candidate	0.320	0.326	1.020	0.010
Master’s Degree vs. Doctoral	0.320	1.113	3.562	0.005
PhD Candidate vs. Doctoral	0.320	0.102	0.320	0.007

Note: Posterior odds are corrected for multiple comparisons with a fixed prior probability of 0.5 for the null hypothesis across all contrasts. All comparisons are based on a default Bayesian t-test with a Cauchy prior (r = 1/√2). BF₁₀,u denotes the uncorrected Bayes factor.

).

The comparison between males and females provided modest support for a difference, where the males had higher scores for SC compared to the females (BF₁₀ = 2.249, error = 0.009). Post hoc tests for level of education revealed a number of strong-to-overwhelming effects. More specifically, SC was rated significantly higher by the Doctoral-level group compared to the High School-educated group (BF₁₀ = 67.421), and also compared to the Bachelor’s- and Master’s-level groups (BF₁₀ = 11.215, BF₁₀ = 3.562, respectively). Moderate-to-strong evidence was also found in support of higher SC in PhD Candidates versus individuals with just a High School degree (BF₁₀ = 7.454), and in individuals with a Master’s degree versus individuals with a High School degree (BF₁₀ = 3.457). Comparatively, contrasts between lower or neighboring levels of education provided anecdotal to poor evidence, i.e., High School and Bachelor’s degree (BF₁₀ = 0.400), Bachelor’s and Master’s degree (BF₁₀ = 0.212), Bachelor’s and PhD Candidate (BF₁₀ = 1.843), Master’s and PhD Candidate (BF₁₀ = 1.020), and PhD Candidate and Doctoral (BF₁₀ = 0.320). These findings suggest that SC behavior tends to increase notably at the highest levels of formal education

To supplement the statistical results and offer interpretability, the following are the plots of distributional trends and the interaction effects of the Bayesian analyses (Figure 8, Figure 9 and Figure 10). The plots give a clearer sense of how sustainable consumption (SC) differs by education levels and gender and how their interaction influences such differences. Visual inspection supports the main effects specified in the model, pointing out both the main effects and the subtle interaction between the categorical predictors.

Figure 8 illustrates the distribution of SC scores across education groups. The raincloud plot shows that individuals with Doctoral and PhD Candidate education tend to report higher SC compared to High School and Bachelor’s degree groups, in line with the Bayesian ANCOVA results indicating strong-to-extreme evidence for group differences (e.g., BF₁₀ = 67.421 for Doctoral vs. High School). The distribution shapes and median shifts highlight heterogeneity in sustainable consumption across education levels.

Figure 9 depicts the posterior distributions for each education level’s effect on SC. The density peaks for Doctoral and PhD Candidate groups are shifted leftward (lower log effect sizes), while High School and Master’s degree are more right-skewed, reflecting the observed post hoc comparisons. The largest divergence is observed between Doctoral and High School groups, consistent with the extreme evidence (BF₁₀ = 67.421), supporting education as a significant predictor of SC.

Figure 10 is an interaction plot of the effect of education by gender on SC scores. The patterns indicate a gender–education interaction in which women have a little more SC at the mid-level education (Master’s), and men have a sharper drop from High School to Doctoral. This pattern is in support of the inclusion of the gender × education interaction term in the Bayesian ANCOVA, in which the interaction term had overwhelming evidence (BF_incl = 25,641.556). This interaction partly explains the subtle variations in how level of education impacts sustainable consumption differently in each gender.

These findings support that the Doctoral-level-educated participants demonstrated consistently higher sustainable consumption behavior, validating the theory that higher education is associated with stronger pro-environmental behavior.

6. Discussion

This study investigated the sociodemographic and psychological predictors of sustainable consumption (SC) and climate action participation (CAP) using a Bayesian analytic framework. Drawing on HEXACO personality traits, climate emotions, and sociodemographic variables, we examined their collective and individual predictive value within the framework of Bayesian regression models. The findings present mixed but theoretically intriguing conclusions with implications for both environmental psychology scholarship and applied sustainability efforts.

6.1. Sustainable Consumption: The Dominance of Education and Climate Concern

Bayesian analysis strongly emphasized education as the most reliable and consistent predictor of sustainable consumption behavior. With a decisive inclusion Bayes factor (BF_incl = 785.19) and a posterior credible interval excluding zero, education was the overarching explanatory variable. These results are abundantly consistent with the leading literature suggesting that formal education is conducive to stronger environmental awareness, green values, and critical thinking abilities necessary for making decisions towards sustainability (Becht et al., 2024; Latkin et al., 2025). Interestingly, our research identified that Doctoral education level respondents had considerably greater sustainable consumption scores compared to High School-level respondents, consistent with research linking formal education with greater ecological awareness and longer temporal orientation (Kesenheimer & Greitemeyer, 2021; Leite et al., 2023; Latkin et al., 2025). Education can support cognitive abilities like systems thinking as well as future self-continuity—central psychological precursors to climate-concordant behavior (Zhang et al., 2025; Zheng et al., 2023; Latkin et al., 2025).

Alongside education, Climate Change Worry (CCW) showed moderate support as a predictor (BF_incl = 1.28). Although the effect size was modest, this result adds to the mounting evidence that affective involvement—specifically moderate, regulated worry—can inspire sustainable action (Ye et al., 2024; Fischer et al., 2017; Latkin et al., 2025). This is consistent with affective–motivational theories proposing that worry triggers action when combined with efficacy beliefs and cognitive clarity (Leite et al., 2023). The modest impact of CCW can be explained by emotional saturation, or heterogeneity in the translation of worry into action due to individual emotion regulation and social support within the setting (Zheng et al., 2023; Ye et al., 2024; Horani & Dong, 2023).

In contrast, HEXACO personality traits—Honesty–Humility and Conscientiousness in particular—did not come out as effective predictors in the Bayesian analysis. The null result contradicts a string of past studies that found medium effect sizes for Honesty–Humility in predicting sustainable behavior (Kesenheimer & Greitemeyer, 2021; Soutter et al., 2020). This difference can partly be accounted for by contextual and methodological considerations specific to this research. For example, the Greek sociocultural context, as characterized by its sociopolitical environment, educational process, and public discussion of sustainability, can shape the expression of moral personality traits in environmentally consequential action. The operationalization of sustainable consumption in this research also focused on self-reported everyday lifestyle practices subject to more determinant control by structural resources (e.g., availability, affordability, norms) than by stable personality dispositions.

One explanation is that structural factors such as education swamp trait effects in this particular cultural setting, or that trait effects are indirectly mediated through intervening affective reactions not included in direct regression specifications. It is also possible that the virtues of traits such as Honesty–Humility operate through emotional or motivational mediators such as guilt, concern, or moral identity—none of which were tested as mediators in this study. Another explanation is that the high uncertainty in our model for the trait effects indicates context-dependent moderation, whereby the development of moral and Conscientious traits is contingent on perceived social norms, efficacy beliefs, or environmental opportunities. Future research should use moderated or mediated modeling techniques to dissect such possible pathways. These findings demand more integrative and cross-cultural studies investigating the interaction of personality, context, and affect in determining sustainability behavior (Zheng et al., 2023; Kesmodel, 2018; Müller-Pérez et al., 2025).

6.2. Predictive Value of Personality and Emotional Traits for Climate Action

Bayesian logistic regression modeling showed that none of the sociodemographic and psychological variables theorized to be strong predictors of climate action participation (CAP) were found. The null model had the best posterior probability (P(M|data) = 0.408), and no rival model had a Bayes factor (BF₁₀) greater than 1, meaning that none of the personality dimensions, affect variables, or sociodemographic variables had explanatory value for activism-based behavior in this sample. This finding runs counter to previous research highlighting the importance of moral character—e.g., Honesty–Humility—and emotional reactions—e.g., empathy, guilt, or outrage—underpinning collective environmental behavior (Panno et al., 2021; Ye et al., 2024).

There are a number of possible explanations for this. For one, psychology’s behavioral specificity hypothesis contends that the determinants of low-cost, private-sphere behavior (e.g., green consumption) may not apply to high-cost, public-sphere behavior such as protest, giving, or activism (Hidalgo-Crespo et al., 2023; Soutter et al., 2020; Balaskas, 2024). Activism also involves riskier, more time-consuming, or more political involvement and therefore may be more under the control of social–political identity, perceived behavioral efficacy, or normative influence than dispositional qualities as such (Becht et al., 2024; Stewart, 2021). Within the Greek setting, structural facilitators of public environmental behavior—i.e., civic engagement opportunities, activist group visibility, and institutional trust—can be absent or in short supply, strangling the conversion of concern into action. Social disillusionment or feelings of political inefficacy can also introduce psychological barriers even among individuals who hold strong pro-environmental attitudes (Becht et al., 2024; Stewart, 2021).

Second, while climate feelings such as concern and compassion are theorized to instigate moral motivation (Shipley & Van Riper, 2022), their influence on behavior is contingent upon mediating variables—such as self-efficacy, social support, and perceived opportunity to act (Leite et al., 2023; Horani & Dong, 2023). When these enabling conditions are lacking, these emotions are likely to cause concern without mobilization, or may even promote despair and passivity (Parmentier et al., 2024). Subsequent studies should consider employing mixed-method designs, such as qualitative studies and collective efficacy scales, to enhance the measurement of these structural and contextual barriers. Finally, longitudinal follow-up on activism and social network outcomes might reveal lagged or cumulative effects of climate worry on activism.

In conclusion, though personality and emotional traits can influence environmental concerns and low-threshold action, our findings indicate their limited explanatory value for active climate action. This gap requires the development of more advanced theoretical frameworks and more nuanced methodological designs that are able to account for the complexity of climate activism.

6.3. Gender and Education: An Interaction Worth Noting

One of the most robust findings of this research was the unconditional interaction between education and gender in predicting sustainable consumption (SC), which was supported by a Bayes factor for inclusion of BF_incl = 25,641.56. This points to the fact that the interaction between the level of education and ecologically friendly behavior was not consistent across genders. More specifically, women had comparatively higher SC scores at the middle ranges of education (i.e., Master’s level), while men had a steeper score slope with increasing sustainability behaviors at only the highest level of education (i.e., Doctoral level). This pattern is suggestive of a moderating role of gender on the behavioral effect of education, possibly because of differing value internalization processes, environmental contact, or socialization practices. These findings are also consistent with a growing body of research linking gender identity and environmentalism, in general, pointing towards women, to a greater degree, exhibiting stronger ecological values, empathy towards nature, and sustainable lifestyles (Horani & Dong, 2023; Soutar & Wand, 2022). Social Role Theory (Balaskas, 2024) is one of several theoretical frameworks arguing that women, being socialized into communal and care roles, would be more inclined to take on behaviors viewed as nurturing or socially responsible—attitudes that would be comfortable in environmental stewardship.

Additionally, this discussion reflects existing research demonstrating that learning context does matter, specifically for women. For instance, research has shown that exposure to environmental material, interdisciplinary sustainability education, or participation in civic discourse within learning contexts effectively increases women’s environmental dedication (Cipriani et al., 2024; Zheng et al., 2023; Horani & Dong, 2023). This aligns with the Value–Belief–Norm Theory that pro-environmental behavior is established by internalizing personal norms activated by values—norms that can differentially be assisted by education according to gendered expectations and opportunities for critical reflection (Latkin et al., 2025; Müller-Pérez et al., 2025; Goodman, 1961; Severo et al., 2021).

Current results suggest the value of intersectional analysis within environmental psychology since demographic variables do not function independently but in conjunction with each other to produce behavioral results. Future studies should persist in an examination of such dynamics through the inclusion of the field of study, exposure to environmental education, and gender identity complexities, which might yield a truer explanation of how sustainability orientations are developed. Policy-wise, this finding testifies to the potential of educational interventions that are responsive to gendered learning styles and life experience.

6.4. Implications for Theory and Practice

6.4.1. Theoretical Contributions

This study informs theoretical advancements in sustainability psychology through providing support for a dual-pathway explanatory framework of sustainable consumption behavior. In particular, the results emphasize that structural (education) and affective (climate concern) factors possess combined predictive ability, while dispositional personality traits, Honesty–Humility and Conscientiousness, have limited explanatory utility in a Bayesian analysis. This contradicts the overall trait model assumption that pro-environmental behavior is being overwhelmed by stable individual differences (Soutter et al., 2020; Panno et al., 2021).

Therefore, this research is more consistent with Value–Belief–Norm Theory and its extensions to include affective predictors like guilt, concern, and empathy (Shipley & Van Riper, 2022; Ye et al., 2024; Fischer et al., 2017). The prima facie dominance of education and climate concern indicates that context- and value-based explanations can put forward stronger predictive models than personality theory for describing sustainability behavior—especially in cultures where structural variables predominate in behavioral streams. Importantly, these results indicate a boundary condition for dispositional models. For high-salience occasions like climate change, where there is heated public discussion and moral imperative, affect-prone judgments and knowledge schemas can dominate the more slowly emerging, dispositionally charged pathways. This is consistent with dual-process theories of pro-environmental action, wherein affective and situational information will bypass trait-compatible reasoning and produce behavioral responses more directly.

Additionally, the education-by-gender interaction which was discovered adds a novel theoretical consideration: sociodemographic moderators are also strong enablers of sustainability value expression and need to be included in integrative behavioral theories. In doing this, they add weight to demands for environmental psychology intersectional models that consider how identity factors (e.g., gender roles, education level) intersect to enable environmental action (Horani & Dong, 2023; Soutar & Wand, 2022). Lastly, the null results for climate action participation (CAP) highlight the need to distinguish between private-sphere (e.g., consumption) versus public-sphere (e.g., activism) activities. While private sustainable behaviors will be more sensitive to internal dispositions like education and concern, public action will be more sensitive to external facilitators like access, social support, and opportunity structures. This distinction offers a helpful lens for sharpening theories of behavior change in all environmental engagement domains.

6.4.2. Practical Implications for Educators and Policymakers

These results have practical implications for environmental education program designers, climate change communication campaigns, and policy interventions. Firstly, the robust predictive function of education supports the case for the incorporation of sustainability-focused curricula—especially at the tertiary education level. These curricula will need to have embedded components that foster systems thinking, future self-continuity, and the critical consideration of long-term environmental consequences. Furthermore, self-efficacy building and moral responsibility could be necessary in bridging the gap between knowledge and sustainable behavior. Integrating content that fosters critical thinking regarding ecological matters, systems thinking, and affective resonance (e.g., climate anxiety coupled with a sense of hope and efficacy) could strengthen behavioral outcomes. Second, because climate worry is a motivator of sustainable consumption, this implies that communications campaigns need to go beyond stating facts but should also inspire emotional engagement—for instance, through affective framing that is both urgent and efficacious in collective action terms (Leite et al., 2023; Ojala et al., 2021; Severo et al., 2021). Our results indicate that moderate concern for the climate combined with efficacy cues and hope are the most likely to encourage participation without disengagement through fear. Agency-directed, shared-responsibility, and action-capable messages possess the ability to translate worry into productive motivation. Constructive affective framing can motivate action without triggering eco-paralysis or denial. Third, the gender–education interaction result suggests that uniformity is suboptimal. Targeted interventions are needed to acknowledge how gendered roles and experiences of education condition attitudes towards sustainability and can foster outreach and engagement. That is, appealing to relational and community-oriented values in female-majority groups, or innovation and responsibility in male-majority technical groups, can increase receptivity.

Finally, the null results on climate action participation also imply that greater priority is warranted to minimize structural and perceived barriers to activism. This involves enhancing civic infrastructure, making more accessible opportunities for collective action, and enhancing the visibility of model examples of success. Policy intervention can be directed at maintaining open and inclusive avenues for civic participation, incorporating social norms and peer influence into behavior change models.

6.4.3. Methodological Contributions and Bayesian Advantages

This work is methodologically valuable for sustainability psychology due to demonstrating the use of Bayesian statistical modeling—a method with intrinsic benefits over conventional frequentist approaches. In areas of research with small effects, interrelated predictors, and theoretical pluralism, like environmental behavior, Bayesian inference offers a more tractable, transparent, and informative analytic approach (Bergh et al., 2021; Wagenmakers et al., 2018).

Relative to null hypothesis significance testing (NHST) based on ad hoc p-value thresholds providing weak evidence for the absence of effects, Bayesian approaches allow direct probability statements regarding hypotheses and model parameters. In the present study, the capacity to evaluate posterior distributions, credible intervals, and Bayes factors allowed for the sensitive interpretation of positive and null results. For example, the inclusion Bayes factor (BF_incl = 785.19) for education gave decisive evidence for its importance in the prediction of sustainable consumption, whereas BF_incl values < 1 for the majority of HEXACO traits gave plausible evidence for there being no effect. These inferential properties are particularly useful when investigating complicated, multivariate behavioral models such as those in which standard regression would distort parameter uncertainty or overfit because of collinearity. In addition, Bayesian model comparison and model-averaging techniques, as utilized in JASP, allowed for a principled comparison of rival explanatory systems—e.g., emotion-only vs. personality-and-emotion models. This is in line with recent arguments in psychology and behavioral science for cumulative theory testing, wherein evidence strength is tempered across models, not test-by-test (Ojala et al., 2021; Wagenmakers et al., 2018; Karimi-Malekabadi et al., 2024).

The application of default JZS priors maintained analytic conservatism while being within the reach of applied researchers who have no experience with subjective or expert-driven priors. Sensitivity analyses also demonstrated the conclusions’ robustness to other prior elicitation, which aligned with the criticism that Bayesian procedures can be too reliant on prior assumptions (Vlasceanu et al., 2024; Karimi-Malekabadi et al., 2024). Lastly, this research demonstrates how Bayesian approaches can promote replicability and transparency in sustainability science. The complete transparency of prior assumptions, model probabilities, and posterior estimates aligns with open science principles and enables other researchers to replicate, update, or extend models with new data—a necessary step toward evidence accumulation in climate psychology.

7. Conclusions

Combating climate change demands more than awareness—it demands specific interventions that elicit commitment to action. This research demonstrates that sustainable consumption is most strongly explained by structural and affective factors, i.e., education and worry about climate change, and personality traits, such as Honesty–Humility, have restricted explanatory potential. The robust gender by education interaction also highlights the requirement for shaping interventions around sociocultural identities. Our results validate integrative sustainability behavior models that emphasize education experience and affective commitment beyond trait prediction. Although climate concern moderately inspires sustainable behavior, climate action engagement seems to be governed by wider social and contextual processes, beyond individual dispositions or emotions. Methodologically, Bayesian analysis promotes transparency and accuracy in assessing predictor importance, particularly in small-effect and theoretically intricate models.

For all its strengths, this study is not without limitations. First, the cross-sectional design circumscribes causal inference and insight into dynamic behavioral change. Longitudinal and experimental designs would be an asset for subsequent research to determine the temporal order and examine intervention effects over time (Taherdoost, 2016). Second, the use of self-report measures runs the risk of biases like social desirability or recall error (Vehovar et al., 2016). While validated measures and pilot testing were used to provide maximum measurement reliability, objective behavioral data like ecological momentary assessments or observational measures would contribute to validity. Third, the reliance on snowball sampling limits the generalizability of the results. The method, nonetheless, allowed access to a diverse sample that would be hard to obtain using probability-based methods (Olsen & St George, 2004; Sandelowski, 2000).

Follow-up studies need to investigate indirect processes, for instance, the mediating role of moral emotions or cognitive appraisals, and contextual moderators such as media exposure, subjective norms, or political affiliation (Ojala et al., 2021). Null results for climate action participation (CAP) also deserve more scrutiny—especially in designs controlling for situational constraints, structural barriers, and motivational thresholds that vary between attitudinal and activist behavior. Methodologically, subsequent research should continue to apply Bayesian models to multilevel, dynamic, or person-centered models that will better capture trait–emotion–context interactions (Parmentier et al., 2024; Karimi-Malekabadi et al., 2024). The extension of the research to multicultural settings and social processes would thus add to the generalizability of the findings based on sensitivity to the reality that while environmental issues occur worldwide, their experiences are not necessarily uniform across environments and societies. While composite scores served to capture these complex constructs, it is believed that subscale analyses would provide fuller appreciations of their different dimensions. Fourth, while the study sampled adults across the full range of ages, the 40-plus age group was to some extent underrepresented. This restricts the generality of age-related findings, and future research should try stratified or quota sampling with the aim of achieving an improved balance between the age groups. Additionally, future studies would be advantaged by measuring climate-related content knowledge, since it can moderate or mediate the influence of dispositional traits and emotions on pro-environmental behavior. Understanding how knowledge interacts with personality and affective involvement could explain the behavioral heterogeneity among various populations.

In total, this research adds to sustainability psychology by virtue of the integration of dispositional and emotional predictors within a Bayesian model. The findings show the priority of education and climate concern in predicting sustainable consumption and contradict the presumed supremacy of stable personality traits as drivers of climate action.

Finally, this research confirms the value of developing emotionally intelligent, well-educated, and socially mobilized citizens as agents of environmental transformation. A grasp of the psychological and demographic contours of engagement with sustainability is crucial to shaping interventions that will echo across varied populations. To anchor this understanding, the current research provides both a methodological and theoretical underpinning for future work on activating personal involvement with collective climate responsibility—a foundation for a more engaged, participatory, and resilient society.