Protection motivation theory (PMT), developed by Rogers (1975), explains how individuals are motivated to protect themselves from threats by assessing perceived severity, vulnerability, response efficacy, and self-efficacy (Sutton, 2001). Rooted in expectancy-value theory, PMT emphasizes both threat and coping appraisals while acknowledging the influence of environmental and cognitive factors, including response costs and stimulus duration. In agricultural research, PMT has been widely applied to study farmers' adaptation to climate change. For instance, Bagagnan et al. (2019) found that both risk and coping appraisals positively influenced adaptation behaviors among Gambian farmers. Similarly, Martin et al. (2007) and Bryan et al. (2009) reported that self-efficacy, perceived severity, and financial barriers shape adaptation decisions. Purwanti et al. (2023) observed that perceived risk and adaptation assessment significantly influenced Indonesian farmers' adaptation intentions, a finding echoed by Hasibuan et al. (2023), Singh (2020), and Abid et al. (2019) in South Asia. Additionally, Villamor et al. (2023) showed that New Zealand forest growers with high threat perception and self-efficacy were more likely to adopt protective measures. In Ethiopia, Regasa and Abera (2019) demonstrated that severity, susceptibility, efficacy beliefs, and perceived costs significantly predicted farmers' climate responses. Although PMT has been extensively applied in agricultural and forestry contexts for studying climate adaptation, its application in apiculture remains largely unexplored in existing research.

The theory of planned behavior (TPB), proposed by Ajzen (1991), is widely used to analyze farmers' behavior (Zhang and Wang, 2024; Dessart et al., 2019). TPB identifies attitudes, subjective norms, and perceived behavioral control as key predictors of behavioral intentions. Subjective norms involve perceived social expectations, perceived behavioral control reflects the ease or difficulty of performing the behavior, and attitudes relate to positive or negative feelings about the action. Behavioral intention indicates the willingness to perform the behavior, which is linked to actual behavior. Studies applying TPB to climate-related behaviors show mixed results: Asrari et al. (2022) found that attitudes and subjective norms positively influenced resilience intentions, but perceived control had no significant effect. Ofoegbu and Speranza (2017) reported strong positive effects of attitudes and subjective norms on sustainable forest management intentions, with perceived control showing a negative correlation. Meanwhile, Hounnou et al. (2023) observed that all three TPB factors positively influenced farmers' intentions to use climate information in decision-making.

A growing body of research has explored the determinants of climate change adaptation intentions using the protection motivation theory (PMT), either independently or in combination with other theoretical frameworks. For instance, Luu et al. (2019) used an extended protection motivation theory (PMT) to study farmers' adaptation intentions in Vietnam's Red River Delta, finding that perceived risks to health, finances, production, social ties, and mental wellbeing increased adaptation intentions, while denial and fatalism reduced them. In a comparative study, Kim et al. (2013) combined the Theory of Reasoned Action (TRA) and PMT to show that attitudes, perceived severity, response efficacy, and self-efficacy predict pro-environmental intentions among American and Korean students. Lin et al. (2024) found that vulnerability, self-efficacy, response efficacy, and subjective norms significantly affect forest tourists' attitudes and revisit intentions, while severity had no significant effect. Supporting the value of integrated models, Badsar and Karami (2021) demonstrated that combining TPB and PMT better explains farmers' sustainable behavior, with PMT variables playing a crucial role. Similarly, Cai et al. (2024) highlighted that climate experience strongly influences adaptation behavior within an integrated TPB-PMT framework. Collectively, these studies support PMT's relevance for understanding climate adaptation intentions and suggest that integrating TPB improves explanatory power. Nevertheless, while research on climate adaptation in agriculture is expanding, quantitative studies specifically examining the factors influencing beekeepers' adaptation intentions remain scarce, emphasizing the need for further empirical investigation in this sector.

A review of studies using PMT, TPB, or both reveals ongoing debate about whether self-efficacy (S-E) and perceived behavioral control (PBC) are distinct. Some research treats them interchangeably (Lin et al., 2024; Droms and Craciun, 2014), while others clearly differentiate the two (Yap and Lee, 2013; Badsar et al., 2023; Rhodes and Courneya, 2003). Manstead and Eekelen (1998) define PBC as the perceived ease or difficulty of performing a behavior, whereas self-efficacy reflects belief in one's ability to achieve desired outcomes, viewing challenges as opportunities rather than threats. Tavousi et al. (2009) further support this distinction, defining S-E as the assessment of task difficulty and PBC as perceived control over behavior. Nonetheless, Lin et al. (2024) note that TPB and PMT treat PBC and S-E as similar constructs, highlighting their conceptual overlap.

This study aims to assess the core frameworks of protection motivation theory (PMT) and theory of planned behavior (TPB) models, along with an extended PMT model, to examine the determinants affecting beekeepers' intentions to implement resilience strategies in response to the recognized impacts of climate change. This research highlights the differences and relative significance of self-efficacy and behavioral control variables, as well as the impact of risk perception on attitudes toward resilience practices, as detailed in the extended version of the model. The empirical focus is placed on the beekeeping sector in Tunisia, acknowledging that adaptation strategies may differ significantly depending on the specific context. Furthermore, the study evaluates the added value of extending the PMT model through the inclusion of supplementary variables that may enhance its explanatory power.

The first six hypotheses formulated are based on the protection motivation theory (PMT) model (Figure 1). Risk perception is a fundamental predictor of the behavioral coping strategies that individuals use to deal with risk (Lee and Lin, 2022). The protection motivation model suggests that two pathways explain human behavior: threat appraisal and coping appraisal (Badsar et al., 2023; Chen, 2020). Threat appraisal involves assessing how individuals perceive a specific behavioral threat, specifically the perceived severity and vulnerability of climate change issues. On the other hand, perceived severity reflects the extent to which individuals emotionally evaluate the seriousness of potential negative consequences associated with a given threat (Badsar and Karami, 2021). The perception of severity pertains to how farmers evaluate the effects of climatic changes. Perceived vulnerability refers to an individual's assessment of the likelihood of experiencing climate change. Based on the above literature review, both constructs, “perceived severity” and “perceived vulnerability,” positively and significantly influence the behavioral intention to adopt resilience and adaptation strategies to face climate change. Therefore, it could be argued that:

Hypothesis (H1). The perceived risk of climate change positively and significantly influences the behavioral intention to adopt resilience and adaptation strategies to face climate change.

Hypothesis (H2). The perception of the occurrence of climate change positively and significantly influences the behavioral intention to adopt resilience and adaptation strategies to face climate change.

Lin et al. (2024) confirmed that response assessment significantly influences individuals' risk moderation behavior, especially regarding climate change risks. Response assessment includes factors such as response efficacy, self-efficacy, and response cost.

Self-efficacy reflects an individual's confidence in their ability to address specific problems, which affects their decision to engage in risk-reducing behaviors (Shafiei and Maleksaeidi, 2020). In this study, self-efficacy refers to farmers' perceived ability to implement adaptation and resilience strategies. Response efficacy pertains to the perceived effectiveness of specific risk prevention behaviors, which relates to how well farmers are adopting resilience practices. Response cost represents the expense associated with implementing recommended behaviors, influencing the intention to adopt resilience strategies in response to climate change. Based on this, the following hypotheses are established:

Hypothesis (H3). Response cost negatively impacts farmers' intention to adopt resilience strategies to address climate change.

Hypothesis (H4). Response efficacy (copying appraisal) positively impacts farmers' intention to adopt resilience strategies to address climate change.

Hypothesis (H5). Self-efficacy positively impacts farmers' intention to adopt resilience strategies to address climate change.

Behavioral intention is defined as the motivational drive that compels an individual to undertake a specific behavior, according to Arendt et al. (2013). Similarly, Badsar et al. (2023) identified farmers' behavioral intentions as a major predictor of their sustainable environmental practices. However, Hou and Hou (2019), Fishbein and Ajzen (1975), and Josef (2013) pointed out that behavioral intention determines actual behavior. Based on this, the following hypothesis is proposed:

Hypothesis (H6). Behavioral intention to adopt resilience and adaptation strategies positively and significantly influences the adoption of these strategies by beekeepers.

The next three hypotheses formulated are based on the Theory of Planned Behavior (TPB) (Figure 2). According to the TPB, perceived behavioral control is shaped by beliefs regarding factors that either facilitate or hinder the execution of a behavior (Asrari et al., 2022; Hounnou et al., 2023). This control significantly impacts disaster management. Thus, the following hypothesis is proposed:

Hypothesis (H7). Positive perceived behavioral control has a significant and positive influence on the intention to adopt resilience and adaptation practices to address climate change.

Attitudes toward behavior are shaped by perceived benefits or drawbacks, influencing emotions and ultimately behavioral intentions (Lin et al., 2024). Cai et al. (2024) also confirmed that positive attitudes significantly affect herdsmen's motivation to adopt climate adaptation strategies. Based on this, we propose the following hypothesis:

Hypothesis (H8). Attitudes toward coping strategies have a significant and positive impact on beekeepers' intention to adopt these resilience and adaptation strategies and practices to face climate change.

Subjective norms refer to an individual's understanding of what others consider important and the expectations surrounding their behavior (Asrari et al., 2022; Ofoegbu and Speranza, 2017). These norms have a positive, direct, and significant influence on the intention to adopt resilient behaviors, while perceived behavioral control did not demonstrate a significant effect. Therefore, we propose the following hypothesis:

Hypothesis (H9). Subjective norms have a significant and positive impact on beekeepers' intention to adopt resilience and adaptation strategies and practices to face climate change.

In response to climate change, research has increasingly emphasized the relationship between climate risk perception and pro-environmental attitudes. Shen et al. (2024) demonstrate that climate risk perception has a significant positive impact on pro-environmental attitudes, highlighting a strong correlation between the two constructs. Similarly, Bradley et al. (2020) suggest that risk perception can indirectly forecast environmental behaviors. Moreover, Carducci et al. (2021) found that attitudes toward pro-environmental behaviors are positively associated with risk perception. Therefore, we advance the following hypothesis (Figure 3):

Hypothesis (H10). Risk perception has a significant and positive impact on attitudes toward adopting pro-environmental practices and developing coping strategies to address climate change.

Building on the work of Yap and Lee (2013), Trafimow et al. (2002), Ajzen (2002), Rhodes and Courneya (2003), and Manstead and Eekelen (1998), self-efficacy is identified as an internal factor, while perceived behavioral control (PBC) encompasses both internal and external influences. Ajzen (2002), along with Bandura (1997) and Armitage and Conner (2001), emphasizes that self-efficacy is a core component of PBC. Although commonly treated as a subdimension within PBC, self-efficacy can also be modeled as an independent (exogenous) latent variable that significantly influences PBC. Conceptually, PBC consists of two elements: self-efficacy, reflecting an individual's belief in their capability to perform a behavior, and controllability, referring to the perceived influence of external constraints or enablers. In this framework, self-efficacy has a direct impact on perceived behavioral control, which subsequently shapes behavioral intention and, ultimately, actual behavior. This how Li et al. (2023) explored, through SEM modeling, the effects of cotton farmers' self-efficacy, access to resources, and technological assistance on their perceived behavioral control regarding the adoption of smart agricultural technologies. Therefore, we assert that:

Hypothesis (H11). Self-efficacy positively impacts farmers perceived behavioral control to face climate change.

Wang et al. (2019) and Badsar et al. (2023) found that integrating the theory of planned behavior (TPB) and protection motivation theory (PMT) enhances the explanatory power of behavioral intention models in agriculture. Their results showed that the combined model provides a better fit and higher R² values compared to using either theory alone. Similarly, Xiang and Guo (2023) confirmed the superiority of the integrated approach in predicting farmers' adoption of green pest and disease control techniques. Consequently, we can conclude that:

Hypothesis (H12). The integrated model combining PMT and TPB has better predictability than the separate models.

The research was conducted in Tunisia. According to a report from the World Bank (2024), Tunisia is considered highly vulnerable to climate change and is expected to experience adverse impacts from increased temperatures, increased aridity, reduced precipitation, and rising sea levels. Furthermore, socio-economic and environmental implications will particularly affect water resources, agriculture, and livestock. In this context, climate change is threatening the beekeeping sector and honey production. The beekeeping sector in Tunisia contributes 0.1% to the national GDP and 1% to the agricultural GDP. Over the years, beekeeping in Tunisia has grown in significance. Sixty-five percent of beekeepers operate with fewer than 50 hives. Only 21.1% of beekeepers, those with more than 100 hives, are considered professionals (Jmal et al., 2024). In 2021, the average annual honey production was estimated at 2,360 tons, marking a 57% increase since 2011. However, fluctuations in the number of hives and honey production have been observed in recent years due to climatic conditions and decreasing honey resources [Office d'Elevage et de Pâturage (OEP), 2023]. In Tunisia, beekeepers estimate that colony losses can exceed 40%, with ~20% occurring during periods of extreme heat and another 20% during colder temperatures (Debbabi et al., 2024).

Data were collected based on a survey conducted between May and June 2024, involving a sample of 120 beekeepers across Tunisia. Given the absence of a recent national census of farmers, particularly beekeepers, and the limited availability of data on the structure of beekeeping operations and beekeeper characteristics, a purposive sampling method was used. This approach allowed for the inclusion of a wide range of beekeeper profiles (in terms of training, age, region, gender, main activity, years of experience, etc.) and types of operations (size, legal status, practices, etc.). A questionnaire was designed based on a literature review to collect both quantitative and qualitative data about beekeepers' strategies for addressing climate change and/or their intention to adopt such strategies, along with the factors influencing their choice of a particular resilience strategy. The questionnaire is organized into five sections: (1) s-economic variables of the beekeepers, (2) identification of beekeeping activities, (3) problems and threats in beekeeping, (4) perception of the impact of crises and risks encountered, and (5) strategies for resilience and adaptation to climate change, including the factors that determine the beekeeper's intention to adopt resilience strategies and their choice among these strategies¹. The questionnaire incorporates both closed-ended questions and five-point Likert scale questions to gauge attitudes toward climate change, resilience strategies, and the various factors influencing beekeepers' intention to adopt resilience strategies and their choice of strategies, based on the conceptual model and the literature review. Table 1 shows the specific questionnaire items used for the fifth section and their reference sources.

Descriptive data analysis was carried out using SPSS 26, while modeling with structural equations was performed using Smart PLS software (version 4.0.9.6). This study used the partial least squares structural equation model (PLS-SEM), a multivariate analysis technique used for estimating variance-based structural equation models. It is regarded as having a predictive approach to Structural Equation Modeling (SEM), as it emphasizes the explanation of different variations. Additionally, it is favored among analytical tools due to its minimal requirements for observations and measurements and its ability to identify potential relationships that may warrant further investigation in other experiments or studies (Xiang and Guo, 2023).

The measurement model was used to investigate the relationship between measurement indicators and constructs, while the structural model was utilized to analyze the relationships between exogenous and endogenous constructs.

The estimation of the measurement model involves three types of analyses (Roussel et al., 2005): the reliability of each construct (latent variable), the validity of the constructs, and the degree of fit of the measurement model (Hair et al., 2017, 2019, 2021, 2011).

According to Anto et al. (2023) and Mutyasira et al. (2018), reliability is interpreted based on the obtained alpha values: an alpha >0.90 indicates perfect reliability, between 0.70 and 0.90 indicates high reliability, between 0.50 and 0.70 indicates moderate reliability, and below 0.50 indicates low reliability. Additionally, unlike Cronbach's alpha, composite reliability does not assume that all indicators are equally reliable, making it more appropriate for PLS-SEM, which prioritizes indicators based on their reliability during model estimation. Composite reliability values between 0.60 and 0.70 are considered satisfactory in exploratory research, while values from 0.70 to 0.90 are deemed acceptable in more advanced stages of research (Anto et al., 2023).

In assessing construct validity, discriminant validity is commonly evaluated using two key criteria: the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT). Among these, the HTMT ratio is especially important, as relying solely on the Fornell–Larcker criterion may not adequately confirm the distinctiveness of the constructs. As suggested by Henseler et al. (2014), the HTMT value should not exceed the threshold of 0.85 to confirm discriminant validity.

The variance inflation factor (VIF) is commonly used to assess collinearity among formative indicators. Higher VIF values indicate greater collinearity, with values of 5 or above signaling potential collinearity issues among predictor constructs. However, according to Hair et al. (2017), the VIF should ideally be around 3 or less to ensure acceptable levels of multicollinearity.

The overall fit of a structural equation model can be evaluated using fit indices such as the standardized root mean square residual (SRMR) and the normed fit index (NFI). According to Hu and Bentler (1998), an SRMR value below 0.08 indicates a good fit, reflecting a small average difference between observed and predicted correlations. The NFI, recommended by Lohmöller (1989), measures how well the proposed model fits relative to a null model, with values >0.50 considered acceptable.

To assess and compare the performance of the PMT, TPB, and extended PMT models, several evaluation criteria can be applied, including the coefficient of determination (R²), predictive relevance (Q²), root-mean-square error (RMSE), and the Bayesian Information Criterion (BIC) (Hair et al., 2017, 2021; Sharma et al., 2019; Schwarz, 1978). The R² value reflects the proportion of variance in the dependent variable that is explained by the independent variables (Hair et al., 2017, 2019). A higher R² value indicates better predictive capacity. In turn, the Q² statistic evaluates the model's predictive relevance, with values greater than zero signifying that the model has meaningful predictive power (Hair et al., 2017, 2019; Cohen, 1988).

For model comparison, the Bayesian Information Criterion (BIC) is valuable, as it balances model fit and predictive accuracy, especially in contexts with small sample sizes typical in PLS-SEM analyses. Sharma et al. (2021, 2019) support the use of BIC, noting that the model with the lowest BIC value is preferred for selection.

Table 2 shows that the beekeepers surveyed include women (27.5%) and men (72.5%). The variable “age” divides the sample into three categories, with a predominance of individuals aged 36–65 years, accounting for 60.8%. Those under 35 years of age represent 35.8%, and only 3.4% of surveyed beekeepers are over 65 years of age. Among the 120 beekeepers surveyed, 65.8% are married. The average household size is 3.46, with the most common household size being 2 people. There are four identified levels of education in our sample: secondary education accounts for 30%, university education for 41.7%, primary education for 25.8%, and illiteracy affects ~2.5%.

The survey results indicate that only 20% of beekeepers are members of a professional organization, such as an association or agricultural development group.

Environmental crises have profound and diverse effects on beekeeping, as indicated by beekeeper testimonies. Consistent with the findings of Makamane et al. (2025) in South Africa's Eastern Cape Province, a significant majority of respondents (93.3%) perceive variations in harvest seasons, while 92.5% report that global warming is impacting bee activities. Furthermore, 85.8% of beekeepers indicate that high temperatures reduce colony fertility, and 84.2% observe that higher temperatures lead to premature flowering. Additionally, 75% cite laying failures due to water shortages, a direct result of high temperatures. Pollen and nectar collection and colony development are affected for 76.7% of beekeepers, and 65.8% notice changes in bee habitats. Moreover, 60.8% report an increase in certain diseases, and 5.8% note a rise in invasive species.

These findings (Figure 4) underscore the urgent need for action to mitigate the effects of environmental crises on beekeeping. This includes promoting sustainable agricultural practices, protecting bee habitats, and enhancing colony resilience to climate change.

As suggested by Sok et al. (2020), we initially focused on analyzing the TPB and PMT models in their basic, original forms by examining the direct effects of various factors. This approach paved the way for the development of the extended PMT model, which was then tested for potential mediation effects between factors (Borau et al., 2015). This approach allowed us to demonstrate that the model extension enhances the understanding and fit of the behaviors exhibited by beekeepers. Specifically, starting with the original TPB model, we found that, unlike subjective norms, factors such as attitudes and perceived behavioral control significantly influenced the intention to adopt mitigation strategies. In turn, this intention significantly impacted the adoption of resilience strategies. In the original PMT model, three out of five explanatory variables had a substantial effect on the intention to adapt to climate change. The two factors that were not found to be significant were the perceived cost of adaptation and the perception of climate change risks. Furthermore, intention played a key role in influencing the adoption of adaptation strategies (Table 6).

In the extended PMT model with proposed mediation effects, the attitude variable mediated the relationship between risk perception and adaptation intent, whereas risk perception did not have a significant direct effect on adaptation intent in the original PMT model. Additionally, the direct effect of self-efficacy on adaptation intent, observed in the original PMT model, remained significant in the extended PMT model. Notably, self-efficacy was shown to have a significant impact on perceived behavioral control, aligning with our hypotheses. These findings emphasize the added value of the extended PMT model, even though the sample size was relatively small. In fact, as highlighted by Sok et al. (2020), larger sample sizes are necessary to better detect significant mediation and moderation effects in extended structural equation models.

In addition, as outlined in the methodological approach of our study, the three-model comparison is based on R² and BIC (see Table 7). The extended PMT model accounts for 66.6% of the variation in the beekeeper's intention to adopt resilient behaviors in response to climate change. This value significantly surpasses those of the other two models, indicating that the extended model is more explanatory. Consequently, the extended PMT model demonstrates the best in-sample predictive capability. Comparing BIC values, the results reveal that the integrated PMT model (BIC = −94.348) significantly outperforms both the PMT model (BIC = −86.819) and the TPB model (BIC = −81.159) in predicting beekeepers' intentions to adopt resilient strategies for climate change. Our hypothesis H12 is thus validated (Sharma et al., 2019, 2021).

The three models—extended PMT, PMT, and TPB—account for ~18% of the variance in resilient behaviors and adaptation strategy choices, demonstrating acceptable explanatory power close to the 20% threshold typically expected in behavioral studies (Hair et al., 2010). The BIC values for all three models are similar, indicating comparable predictive accuracy for beekeepers' resilient behaviors.