Using the theory of planned behavior to identify key beliefs underlying flood‐related adaptive behaviors in the province of Québec, Canada

Given that flooding episodes are occurring at a greater rate due to climate change, individuals must adopt certain adaptation behaviors to prevent or mitigate the anticipated or negative impact of such events. However, few studies have assessed if and how households and individuals have actually taken action in this regard. Because some individual beliefs can be linked to facilitating factors and barriers to action, a better understanding of the adoption of adaptive behaviors requires a combined analysis of individual psychosocial factors. The purpose of this study was to develop a better understanding of the reasons underlying the adoption of behaviors related to structural adaptation to flooding by people living in or near flood‐prone areas in the Province of Québec (Canada). Results of a series of structural equation modeling showed that behavioral, normative and control beliefs were all significant predictors of the respondents' intention to adopt structural flood protective behaviors, with normative beliefs being the strongest. By identifying the best psychosocial predictors of the adoption of such behaviors, the results of this study provide valuable insights regarding the most effective factors to be used in public health messages to promote the adoption of behaviors related to structural adaptation to flooding.


| INTRODUCTION
Flood events are a growing concern worldwide, becoming more frequent and severe as a result of infrastructure development and climate change (Hirabayashi et al., 2013;Kundzewicz et al., 2014;Milly et al., 2002). Natural disasters such as flooding are the most common in Canada and are also the most expensive, with financial impact growing rapidly (Natural Resources Canada, 2017). From 1983 through 2008, Canadian insurance losses averaged CAN $405 million per year, whereas those losses rose to CAN1.8 billion per year for the period from 2009 to 2013 (Moudrak et al., 2015). Recent spring floods in Fort McMurray, in the Province of Alberta caused more than CAN$520 million in damage (Malbeuf, 2020). Similarly, in the Province of Québec, floods caused damage worth more than CAN$2 billion over the course of the 20th century. The series of flash floods in the Saguenay region in 1996 itself caused around CAN$1 billion in damage (Buttle et al., 2016). Important flooding also occurred in the springs of 2017 and 2019. In 2017, the damage caused by floods to southern Quebec municipalities was estimated at more than CAN$376 million, while affecting 293 municipalities in 15 regions, flooding approximately 5400 residences and forcing the evacuation of more than 4000 people as well as the closure of several roads (MAMH, 2020). As for the spring floods of 2019, they affected more than 250 Quebec municipalities (Saint-Arnaud, 2019), once again flooding thousands of homes, forcing the evacuation of more than 10,000 people and causing the closure of several roads. In the province of Quebec, flooding can affect 80% of municipalities located along a river (Natural Resources Canada, 2006).
Additionally, the potential impacts of flooding on physical and mental health are numerous (Zhong et al., 2018;Jermacane et al., 2018;Fernandez et al., 2015). For instance, a total of 18 deaths were reported during the 22 major flooding events that occurred in Canada between 2000(Guha-Sapir et al., 2015, on top of multiple injuries, diseases and mental health issues. The results of a large-scale survey carried out in certain regions of the province of Quebec highlighted that, almost a year after the floods, the people who were flooded and those who were disturbed by the floods were significantly more likely to have mental health problems (post-traumatic stress, psychological distress, an anxiety disorder such as a phobia, obsessive compulsive disorder or panic disorder, or a mood disorder such as depression, bipolarity, mania or dysthymia) than those who were not exposed to flooding (Généreux et al., 2020).
Adaptation in human systems refers to "the process of adjustment to actual or expected climate and its effects, in order to moderate harm or exploit beneficial opportunities" (IPCC, 2022). Regarding floods, adaptation encompasses "reactive or proactive actions that individuals can take, alone or with others, to respond to the impacts of climate change, and to protect or benefit themselves, others, and/or the environment in the short and long term" (Carman & Zint, 2020, p. 10). In this view, actions or behaviors to adapt to floods "refer to actions that aim at preventing or minimizing anticipated or negative consequences of flood events. Such actions can be taken before, during or after an event" (Kuhlicke et al., 2020, p. 3) and are self-initiated household actions, in order "to avoid the direct physical and financial impacts imposed by climate-exacerbated hazards" (Wilson et al., 2020, p. 200). Numerous typologies of behavior that people can adopt in anticipation of a flood have been proposed (e.g., Dillenardt et al., 2022;Noll et al., 2022;van Valkengoed et al., 2022). One of these typologies, which distinguishes between structural measures (i.e., what is done on a house or land, such as waterproofing foundations) and nonstructural measures (i.e., what does not affect the house itself, such as drawing up an inventory of personal property) was used in several studies (e.g., Mondal et al., 2021;Noll et al., 2022;Oubennaceur et al., 2022;Valois et al., 2019;. However, few studies have assessed if and how households and individuals adopt and implement structural adaptation measures and behaviors while also aiming to identify the key psychosocial determinants of their adoption (Babcicky & Seebauer, 2019;Grothmann & Reusswig, 2006;Valois et al., 2019;. The current study focuses specifically on this second element. Monitoring of flood adaptation in the Province of Québec (Canada) began in 2015, namely with a study (Valois et al., 2019) conducted on Quebeckers living in or near a flood-prone area. This led to the development and the validation of five indices of adaptation to flooding according to the chronology of flood events: (a) pre-alert preventive behaviors, (b) post-alert behaviors, (c) behaviors during a flood not requiring evacuation, (d) behaviors during a flood requiring evacuation, and (e) post-flood behaviors. The same five indices were used in a subsequent study conducted in 2019 and 2020, which focused on the development of adaptation to flooding in the province over time . Although monitoring flood adaptation allows decision makers to better understand Quebeckers' susceptibility to flood events, it does not per se provide insights into how to improve the current situation and increase rates of adoption of adaptive behaviors. To better understand what motivates people to protect themselves and their homes from floods, research needs to be conducted on people's beliefs regarding flood adaptationrelated behaviors (Clayton et al., 2015;Gifford et al., 2011;Swim et al., 2011;. While it appears essential to monitor and evaluate progress in terms of adaptation, there is also a need to capture, in the monitoring and evaluation (M&E) of adaptation, the complexity of the process of adopting an adaptation response and further explore the psychosocial dimensions inherent in the adjustments or changes in behavior implied by adaptation.
The need to better understand the psychosocial processes by which adaptation occurs has led some authors to emphasize the importance of an individual perspective to adaptation (Carman & Zint, 2020;Meinel & Höferl, 2017) and autonomous behavioral adaptation (Kuhlicke et al., 2020;Wilson et al., 2020). Moreover, there is abundant literature showing that psychosocial factors or determinants like assessment of climate-related risks and impact, perceived adaptation efficacy and perceived self-efficacy (Clayton et al., 2015;Gifford et al., 2011;Grothmann & Patt, 2005) can influence an individual's motivation to adopt certain behaviors to protect themselves from the impact of a climate hazard. Conversely, low motivation to adapt or negative beliefs about adaptation are among the most important barriers to action (Grothmann et al., 2013). Because some individual beliefs can be linked to facilitating factors and barriers to action, a better understanding of the adoption of adaptive behaviors requires a combined analysis of these types of factors at the individual level (Meinel & Höferl, 2017).
In this regard, Ajzen et al. (2011) revealed how changing people's behaviors and lifestyles is more demanding than simply conveying information on a subject matter. A large number of factors, including psychological barriers, beliefs, personal values, political views, biases, attitudes towards a particular behavior and goals can limit change (Albright & Crow, 2015;Gifford, 2011;Gifford & Nilsson, 2014;Sterling, 2011;Stern, 2011;Yazar et al., 2021;Zografos et al., 2016). Various models have been used in attempts to identify what motivates people to change their behaviors, such as the Protection Motivation Theory (Rogers, 1983), the Protection Action Decision Model (Terpstra & Lindell, 2013), the Model of Private Proactive Adaptation to Climate Change (Grothmann & Patt, 2005) and the Health Beliefs model (Rosenstock, 1974).
Developed from the field of social psychology, the Theory of Planned Behavior (TPB) (Fishbein & Ajzen, 2010) is another model that has been frequently and successfully used in the past (de Leeuw et al., 2015;Deng et al., 2017;Oreg & Katz-Gerro, 2006;Valois, Bouchard, Aenishaenslin, et al., 2020;Zhang et al., 2020) to describe the psychological processes of people who must adopt new behaviors in order to adapt to climate change or make pro-environmental choices. Multiple meta-analyses [see, for instance, Steinmetz et al., 2016, Bamberg & Möser, 2007, Hines et al., 1987and Klöckner, 2013 have also demonstrated the strength of the TPB in explaining decision-making and crafting targeted behavior change interventions. As such, this social-cognitive model of human behavior can be successfully used to identify the beliefs that underlie people's willingness to adopt flood adaptation-related behaviors.

| THEORY OF PLANNED BEHAVIOR-TPB
The TPB is a theory of social psychology that is commonly used to explain human behavior (Ajzen, 1991;Wang et al., 2019;Zoellner et al., 2017). According to this theory, the decision to adopt a behavior is determined by the salient beliefs an individual have about adopting said behavior (see Figure 1).
More specifically, three broad classes of beliefs are distinguished: behavioral beliefs, normative beliefs and control beliefs. Behavioral beliefs are operationalized by measuring a person's beliefs about the likely outcomes of performing a particular behavior (outcome expectancies) weighted by the assessment of such outcomes (Fishbein & Ajzen, 2010, p. 18). It could refer to, for example, an advantage or a disadvantage of performing the behavior in question (e.g., making the structure of a dwelling more water-resistant would help maintain its value neighbors, government representatives) think they (the individuals) should do weighted by their motivation to comply with these referents. Finally, a control belief is defined as a person's subjective probability that a given facilitating factor will be present in the situation of interest, weighted with the factor's perceived power to facilitate performance of the behavior (Ajzen & Kruglanski, 2019). Financial help to improve the structure of a dwelling would be an example of a factor that may interfere with or facilitate performance of the behavior. The TPB also postulates that this perceived control, if it accurately reflects actual control, can moderate the effect of intention on behavior (see the dotted arrow in Figure 1). Intention should have a stronger effect on behavior when actual control is high rather than low.
People are more likely to act according to their intentions when they really have control over the adoption of the behavior (Fishbein & Ajzen, 2010). The purpose of this article was to develop a better understanding of the beliefs underlying the adoption of behaviors related to structural adaptation to flooding by people living in flood-prone areas in the Province of Québec (Canada). For that purpose, this study uses as dependent variable a validated pre-flood prevention-related behaviors index developed by Valois et al. (2019) and the key determining beliefs in people's motivation to adopt behaviors related to structural adaptation to flooding identified in a qualitative pilot study conducted in 2016 (Valois et al., 2017). To the best of our knowledge, no study aiming to identify the key psychosocial determinants of the adoption of behaviors related to structural adaptation to flooding used a validated measure of households and individuals adaptation actions as dependent variable. By allowing the identification of the key beliefs as psychosocial predictors of the adoption of behaviors aimed at making the structure of people's homes more water-resistant, the results of this study provide valuable insights regarding the most effective factors to be used in public health messages to promote the adoption of such behaviors.

| Preliminary study
Following Ajzen (1991), we conducted a pilot study before finalizing the questionnaire, with the goal to identify the key determining beliefs in people's motivation to adopt behaviors related to structural adaptation to flooding. The process broadened and validated the set of beliefs included in the main study. This preliminary study involved a total of 30 individuals randomly selected by a polling firm and who were surveyed in focus groups of five to eight individuals. We used an open-ended questionnaire to interview them regarding (a) the advantages and disadvantages of adopting various behaviors, (b) the individuals or groups of people who would approve or disapprove of the respondents performing these behaviors and (c) the factors that could facilitate or interfere with the adoption of these behaviors. Overall, the more frequently mentioned beliefs were selected and used to pinpoint the belief statements that would be investigated in the main survey. More details about the method used for the pilot study are provided in Valois et al. (2017).

| Main study
To identify the beliefs that influence the adoption of flood prevention-related behaviors in the Province of Québec (area of 1,667,441 km 2 ; population of 8.3 million people), a population-based study was conducted in 2019 using a cross-sectional design. The target population was identified according to the flood-prone areas designated by the Centre d'expertise hydrique du Québec (Quebec Water Expertise Center). We then selected households with their main residences located in or near a flood-prone area, resulting in a target population of 136,476 households.
We used a stratified sample to reflect the geographical distribution by administrative region of the population in the flood-prone areas across the Province of Québec. A polling firm conducted a telephone survey with the respondents, targeting people who were 18 years old or over, who could speak either French or English and whose primary residence was either in a flood-prone area or within 150 m of a flood-prone area. Each interview lasted 30 min on average, with a response rate of 24.5% (n = 951). However, given the aim of our study, respondents also had to be homeowners, since renters would not be able to perform most of the behaviors. For this reason, renters (n = 188) were removed from the initial sample. The final sample included 763 homeowners, with 194 living within 150 m of a designated flood-prone area and 569 individuals living in a flood-prone area. The data were reweighted to keep the same proportion for each administrative region in the sample as in the target population (Deville & Särndal, 1992). Regions with fewer respondents in the final sample than in the quotas received higher weights to achieve a total regional weight equal to the quotas for each region.
Among the respondents, there were 331 men (43.4%) and 432 women (56.6%); the majority were in the 60-64 year old group (14.0%). Most respondents had a university degree (37.6%) and an annual household income above CAN$100,000 (20.8%). See Valois et al. (2019) for a more detailed description of the data collection strategy and Table 1 for more details about the characteristics of the sample.
The questionnaire administered assessed the constructs of the TPB: the structural flood protection-related behaviors (Valois et al., 2019), the respondents' intention to adopt these behaviors and their behavioral beliefs, normative beliefs and control beliefs towards the adoption of these behaviors. This questionnaire was pretested to verify its psychometric qualities (n = 65 respondents not included in the final sample). Using Item Response Theory (IRT) and other test theory analyses, we made sure that the question scales were appropriate and allowed discrimination across all levels for the latent trait measured. Latent constructs are used to quantify variables that are not measurable directly (e.g., beliefs). However, it is possible to estimate them through a statistical model including multiple variables that can be measured and are related to the latent construct.

| TPB constructs
Structural flood protection-related behaviors. We used the pre-flood prevention-related behaviors from the index developed by Valois et al. (2019) to measure adaptation to flood risks. This index was originally composed of 15 behaviors that were either structural or nonstructural. We differentiated structural behaviors, which consist of physical changes to a house or the land around the property from nonstructural behaviors, such as seeking information, owning a water pump, knowing how to cut off the water or electricity and making a list of belongings in case of flooding. Therefore, the adoption of pre-flood protection-related behaviors was measured using nine of the 15 behaviors from the original index that concern structural behaviors (See Table 2). Response options were yes, no, do not know and refusal. Item response theory analyses and classical test theory analyses indicated that the scale used to measure the structural flood protectionrelated behavior index was adequate: all items showed good discrimination across varying standardized levels of the trait measured and had option characteristic curves that showed good discrimination (see Valois et al., 2019), that is, each increasing option became more likely than the previous as the trait level increased.
Intention to adopt structural flood protection-related behaviors. This variable was measured with three questions on the respondents' intention to adopt these behaviors, their desire to do so and the likelihood of them doing so (Table 3). The measure had good internal consistency according to Cronbach's α.
Behavioral beliefs. Participants rated each of the behavioral beliefs in terms of its perceived likelihood and its subjective value (Table 3). According to an expectancy-value model (Fishbein & Ajzen, 2010), the perceived likelihood of each outcome was multiplied by its subjective value and the resulting products were summed.
Normative beliefs. Participants rated each of the normative beliefs in terms of its perceived subjective probability that important referents would expect them to perform the behavior and the motivation to comply to this expectation (Table 3). Then, perceived subjective expectation was multiplied by motivation to comply and the resulting products were summed (Ajzen & Fishbein, 1975).
Control beliefs. Participants rated each of the control beliefs in terms of its perceived subjective probability that a given facilitating factor will be present in the situation of interest and the factor's perceived power to facilitate performance of the behavior (Table 3). Then, perceived control belief was multiplied by the factor's perceived power and the resulting products were summed (Ajzen, 1985).

| Statistical analysis
We applied structural equation modeling (SEM) (Byrne, 2012) using Mplus Version 8 (Muthén & Muthén, 2017). Along with the linear mixed model/ hierarchical linear model, SEM "is the most widely recognized statistical solution in the social sciences" (Tarka, 2018, p. 322). According to Novikova et al. (2013), SEM offers three main advantages over traditional multivariate techniques: (1) SEM explicitly assesses the measurement error, as it offers estimates of error variance parameters for independent and dependent variables (Byrne, 2012). (2) SEM allows the description of latent (unobserved) traits, that are estimated from multiple observed variables, as well as the relations among T A B L E 2 Component of the index of structural flood protection-related behavior.

Behavior
Frequency Percent constructs. The psychometric properties of measures and the estimation of relations among constructs can be compensated for biases linked to random error and constructirrelevant variance (Tomarken & Waller, 2005). (3) SEM can be used as a conceptual or theoretical structure or model to test fully developed models against available data. Unlike other procedures (e.g., multiple regression), SEM offers "measures of global fit that can provide a summary evaluation of even complex models that involve a large number of linear equations" (…) instead of "minitests of model components that are conducted on an equation-by-equation basis" (Tomarken & Waller, 2005, p. 34).
First, we tested the TPB model in which the behavioral, normative and control beliefs influence the adoption of structural flood protection-related behaviors through intention. Second, from this model, we tested the presence of a moderating effect of control beliefs on the influence of the intention to adopt adaptive behaviors (Ajzen, 2012). Finally, in order to gain a better understanding of the most important beliefs, we tested the TPB model using the individual beliefs to identify which one of them more strongly predicts intentions. Comparative fit index (CFI), Tucker-Lewis index (TLI) and Root mean square error of approximation (RMSEA) were used to evaluate the fit of the models. Acceptable model fit is indicated by CFI and TLI values equal to or greater than 0.90 and less than 0.95 and RMSEA values between 0.05 and 0.08. Excellent model fit is indicated by CFI and TLI values equal to or greater than 0.95 and RMSEA values equal to or less than 0.05 (Hu & Bentler, 1999;Kline, 2015). Testing the moderating effect of the control beliefs on the influence of intentions regarding the behavior adoption index requires the use of models with latent variable interactions. Adding interactions to the models within Mplus prevents the computing of fit statistics such as CFI, TLI, and RMSEA (Little et al., 2006).

| RESULTS
As a preliminary analysis, we evaluated the correlation between the items of each latent construct. The correlation between the behavioral belief that adopting the behaviors would protect respondents' physical health and the behavioral belief that adopting the behaviors would protect the respondents' psychological health was r = 0.784. This was high enough to cause fitting issues for the SEM model. Thus, these constructs were combined by using the highest score on either of these items. A similar issue was detected in the normative beliefs. The belief that "experts" thought the respondents should adopt the behaviors was strongly correlated with two other beliefs: (i) the belief that "municipalities and government representatives" thought the respondents should adopt the behavior (r = 0.708); ii) the belief that "people having experienced a flood whom the respondents knew" thought they should adopt the behaviors (r = 0.624). Thus, this item was omitted from the analysis.
Then, a series of SEM analysis was performed. First, we tested the TPB model that included the variable "structural flood protection-related behaviors," "intentions to adopt structural flood protection-related behaviors," and the underlying behavioral, normative and control beliefs (see Figure 2). The fit of this model to the data ranged from acceptable (CFI = 0.934, TLI = 0.926) to excellent (RMSEA = 0.039). Results show that the model accounted for 45.8% of the variance in the respondents' intentions to adopt structural flood protection-related behaviors and 20.2% of the variance in actual behavioral adoption. Results revealed that each category of underlying beliefs were significant predictors of the intentions, with normative beliefs being the strongest (γ = 0.456, SE = 0.038, p < 0.01), followed by behavioral beliefs (γ = 0.254, SE = 0.038, p < 0.01) and control beliefs (γ = 0.233, SE = 0.045, p < 0.01).

Constructs
Items Response options Cronbach's α • The steps that would be involved or that the respondent has done to improve the structure of his/her home are simple • The laws and regulations would be or were easy to understand to help the respondent improve the structure of his/her home • The respondent would have or have had access to easily understandable information to improve the structure of his/her home • The respondent does not have or did not have physical constraints limiting his/her ability to improve the structure of his/her home As a second SEM analysis, we attempted to fit the same model with a moderating effect of the control beliefs over the effect of the intentions on the adoption of the behaviors. This moderating effect proved to be nonsignificant (γ = 0.000, SE = 0.053, p-value = 0.996). Thus, the model was not retained.
Finally, we performed another SEM for which each item from the underlying behavioral, normative and control beliefs directly predicted intentions. In other words, six behavioral beliefs (six expectancy-value products), four normative beliefs and five control beliefs were tested in predicting the respondents' intentions to adopt structural flood prevention-related behaviors. The 15 beliefs explained 45.2% of the variance in intention to adopt structural flood prevention-related behaviors, the model providing an excellent fit: CFI = 1.00; TLI = 1.002; RMSEA = 0.00. Two behavioral beliefs were significant predictors of intentions: "adopting the behaviors would protect the respondents' physical health and psychological health" (γ = 0.097, SE = 0.041, p = 0.016); "adopting the behaviors would help maintain the value of the respondents' home" (γ = 0.126, SE = 0.043, p = 0.003). Furthermore, two of the four normative beliefs were significant predictors of intentions, notably the belief that the respondents' family and friends would think that they should adopt the behaviors (γ = 0.358, SE = 0.054, p = 0.001) and the belief that the people the respondents knew who had experienced a flood would think that they should adopt the behaviors (γ = 0.201, SE = 0.047, p = 0.001). Finally, two out of the five facilitating factors were significantly related to intentions: the belief that the steps involved in improving the house structure would be simple (γ = 0.176, SE = 0.042, p = 0.001); the belief that not having any physical constraints to their ability to improve the house structure would influence their control over the adoption of the behaviors (γ = 0.147, SE = 0.038, p = 0.001).

| DISCUSSION
The main goal of this study was to use the TPB to identify the beliefs of people living in flood-prone areas that are the most strongly associated with the adoption of adaptive behaviors aimed at making the structure of people's homes more water-resistant. Results from the study allowed to clearly identify beliefs that are significantly related to people's decision-making when it comes to flood preparedness. Notably, our results suggest that all three categories of beliefs from the TPB had an impact on the respondents' intention to adopt structural flood protection-related behaviors, with normative beliefs being the most important determinant. Results also show that none of the prevention-related behaviors comprising the index were adopted by more than half of the respondents. These results are similar to those we obtained in a previous study .
Although the results highlight the need for a higher adaptation rate by the population at risk, they also confirm the potentiality of targeted interventions based on the individual beliefs significantly related to the respondents' intention to adopt behaviors to improve the structure of their homes. To be effective, an intervention should target and modify the behavior-relevant beliefs that serve as determinants to adopting specific adaptation measures. This means that interventions aiming to foster the adoption of structural flood protection-related behaviors would be more effective if they were designed to reinforce supportive beliefs and counter negative ones (Fishbein & Ajzen, 2010, p. 332). Our results suggest that normative beliefs are the best determinants of the respondents' intention to adapt by improving the structure of their homes. Assessment of the key normative beliefs of Quebeckers living in flood-prone areas or within 150 m from a designated flood-prone area shows that the respondents who believed that "their family and friends would think they should adopt the behaviors" and that "the people they knew who had experienced a flood would think they should adopt the behaviors" are more likely to develop a positive intention to adopt adaptation behaviors and put such intention in action. Public communication strategies and policies would gain from emphasizing such influence. Therefore, public awareness messages featuring the testimonials of figures portraying neighbors, families and friends could be key in improving adaptation. Similar work based on the TPB has already proven effective in shaping people's beliefs and attitudes on speeding (Stead et al., 2005). Narratives with personal testimonials have also been used successfully for health campaigns against smoking (Kim et al., 2012). In this case, using personal testimonials from people most likely to influence other people's decisions (i.e., figures posing as friends and family or people who have experienced a flood) regarding structural behaviors could be helpful in crafting effective media campaigns. Alongside normative beliefs, control and behavioral beliefs may influence individuals' intention to adopt structural flood protection-related behaviors. Literature pertaining to the Protection Motivation Theory shows that coping appraisal (i.e., a cognitive process by which a person evaluates various actions that may reduce a perceived threat) has a strong influence on flood mitigation behavior (Babcicky & Seebauer, 2019). Coping appraisal includes constructs like response efficacy, self-efficacy and response costs, which are somewhat included in control and behavioral beliefs in the TPB. Babcicky and Seebauer (2019) found that response efficacy (i.e., believing that an action will be effective in reducing damage expected from a threat) and response costs (i.e., the financial resources, time and effort to implement an action) were significant predictors of structural protection. This is partly in line with our results. Moreover, assessment of the key behavioral beliefs show that the respondents who believed that adopting structural flood protection-related behaviors would produce outcomes like "protecting their physical health and psychological health" and "helping to maintain the value of their home" were more likely to develop a positive intention to adopt adaptation behaviors and put this intention in action. Other studies (e.g., Grothmann & Reusswig, 2006;Terpstra & Lindell;2013) had also shown that response efficacy was a good predictor of the intention to adopt actions to adapt to flooding. For instance, Terpstra and Lindell (2013) showed that flood hazard adjustments were significantly more likely to be adopted in the future by individuals who felt they would be more effective at protecting people and property and who thought they would be useful for other purposes than flood protection.
Our results also show that the belief that adopting structural flood protection-related behaviors involves reasonable expenses did not influence the intention to adopt. This is also similar to the results reported by Terpstra and Lindell (2013), which, contrary to their hypothesis, did not show a negative correlation between resource-related attributes like perceived cost and hazard adjustment adoption intentions. However, our result was slightly different from what was observed in our preliminary study (Valois et al., 2017), in which respondents identified financial constraints as a significant barrier to preventive adaptation and, at the same time, financial assistance as one of the most important facilitators. Results regarding control beliefs shed an interesting light regarding the role of cost in the intention to adopt, as efforts to adapt might not always involve financial resources. Indeed, the results show that the respondents who believed that "the steps that would be involved in improving the house structure would be simple" and that "not having any physical constraints to their ability to improve the house structure" are more likely to develop a positive intention to adopt adaptation-related behaviors and put their intention in action. This suggests that, in addition to promoting preventive behaviors and providing grants, incentives or tax credits to people living in flood-prone areas allowing them to implement some of these costly preventive measures, public outreach should also ensure that people living in flood-prone areas come to consider that the efforts required to adapt may not be as demanding or difficult as they initially thought. Offering personalized support, for example, could favor such considerations. This, however, raises the question as to who or what organization should be in charge of such support. In this respect, collective movements could play a key role, for instance, by setting up communities of practice.
Previous studies have highlighted the importance of considering the local contexts when studying flood vulnerability and the processes underlying people's decisionmaking with regard to flood protection Rufat et al., 2015). This study is particularly relevant to the context of Quebec. It limits potential generalizations regarding the results, but ensures that the beliefs highlighted are significant in terms of crafting effective public health outreach in Quebec. Another limitation is that it relies on self-reporting to measure adaptive behavior adoption. However, objective measurements would have been much more labor-intensive, considering the wide area covered by the study and the number of structural behaviors measured. This means that the participants may have exaggerated their behavior adoption rates, but some factors helped mitigate this potential problem, such as clear statements provided to the respondents that their responses were anonymous. In addition, self-reports are often used in other studies on pro-environmental behaviors (Bakhsh et al., 2018;Bichard & Kazmierczak, 2012;Chen & Liu, 2015), and the behavior adoption rates that we obtained are similar to those obtained in our other studies on flood prevention (Valois et al., 2019;. Finally, the cross-sectional nature of the study is another important limitation in bringing a causal interpretation to the observed associations.

| CONCLUSION
This study contributes to the development of a better understanding of the determinants of the adoption of behaviors related to structural adaptation to flooding. Public authorities working to convince people to take action to protect their homes from flood impacts could benefit from information pertaining to specific individuals' beliefs regarding such actions. While the transmission of information to the public on structural adaptation is important, the effectiveness of such communication effort hinges on the development of messages that will reinforce positive beliefs about structural adaptation to flooding or change negative beliefs about such action. More specifically, these messages could focus on what important referents think an individual should do, for example by presenting positive testimonials of figures portraying neighbors, family members and friends. Communication could also emphasize the positive consequences likely to occur following the adoption of the targeted behaviors (e.g., benefit to physical and mental health; home maintaining its value), as well as the idea that adapting may turn out to be less complex than anticipated with support being made available. The methodology outlined in this study could be replicated in the future to develop a further understanding of factors influencing individuals' choices of adaptation measures, which would help refine our knowledge of the different profiles of individuals to attain, and from there develop specific messages.