Determinants of Adapting to the Consequences of Climate Change in the Peruvian Highlands: The Role of General and Behavior-Specific Evaluations, Experiences, and Expectations

Abstract

Progressive climate change (CC) forces people—particularly in the Global South—to adapt to its consequences, some of which include droughts, flooding, and new diseases. This study investigates the determinants of behaviors for adapting to these threats in a population from the region of Cusco (Peru). Data were gathered via a cross-sectional interview-based survey in 2016, using random-route sampling. For up to 542 cases, we regressed a scale combining performed behaviors and intentions on psychological constructs, for the entire and sub-samples (n > 179, allowing to detect an R² of 10% with a power of 80% at p = 0.05). Behavior-specific evaluations—particularly perceived feasibility (β = 0.355), descriptive norms (β = 0.267), and cost-benefit evaluations (β = 0.235)—can explain most of the variance (44% with a total R² = 61%). Furthermore, trust in specific sources (β = 0.106), general trust (β = 0.098), and negative attitudes toward nature (β = 0.077) are positively related to adaptation, particularly regarding public behaviors (supporting community projects and policies). However, evaluations directly related to CC, such as risk perception (β = 0.010) or how much a behavior helps prevent damage (adaptation efficacy, β = −0.042)), do not explain adaptation, except for an effect of adaptation efficacy on changing daily behaviors. Experiences with and expectations of CC consequences are mostly unrelated to adaptation. However, worries about such events are correlated with adaptation (r between 0.097 and 0.360). We conclude that, to promote adaptation behaviors in this region, the focus should be on the characteristics of the behavior performance (e.g., its costs or feasibility), not on the expected risks of extreme events because of CC.

1. Introduction

Because of massive greenhouse gas emissions, global climate change (CC) is inevitable [1]. To reduce the impact of CC, much research has focused on individual behaviors that reduce the emission of greenhouse gasses, particularly in the Global North [2]. However, because some extent of CC is unavoidable [1], people around the world face adverse consequences of CC and, hence, must adapt to it at all levels of society [3]. Psychological research focuses mainly on the level of individuals [4]. Here, possible steps to reduce climate-related damage may include moving valuable objects to upper floors in areas that are prone to flooding or investing in a private water tank if droughts become a problem. Individual adaptation is particularly important in the Global South, where governments often lack the resources or interest to protect against the adverse consequences of CC. Here, actions at the community level can also become important (e.g., building a reservoir for a village without governmental support). Furthermore, individuals need to support policies that help adapt to the consequences of CC. Because of the importance of individual decisions and behaviors related to adapting to CC, these processes have attracted increasing attention [5]. However, only a few studies have investigated populations in the Global South [6,7], where individual adaptation to CC is even more important.

The present paper addresses this research gap by investigating a population in the Peruvian highlands. This population differs from the commonly investigated populations in the lowlands of the Global North in several ways. In addition to differences in education and cultural background, the population of the study region is more threatened by CC than most populations investigated in the Global North [8]. In the studied area, many people depend on agriculture or tourism, on which the climate has a critical influence. In addition, infrastructures, such as roads or water and energy supplies, are more vulnerable to droughts or excessive rainfall [9]. The most interesting characteristic of the investigated region is the relatively frequent occurrence of climate-related events, which are becoming more frequent and severe in parts of the world where once they were not very common. Hence, investigating this region allows a glimpse into the future of regions that have not yet suffered much of an impact from CC.

The present study aims to understand what drives adaptation to CC in the target population; understanding this can help inform campaigns that promote such behaviors. For this purpose, we propose a concept for specifying adaptation behaviors and combining constructs from various research traditions to explore a wide range of potential determinants. Their influence is investigated via regression and correlational analyses.

1.1. Adapting to the Consequences of CC

The present study focuses on adaptation behaviors, which are preventive measures that reduce the potentially harmful effects of CC. A problem with investigating adaptation behaviors is their wide variety. Specific adaptation behaviors look very different depending on the problem (e.g., droughts vs. flooding), and for each problem, many countermeasures exist (e.g., using less water during a drought vs. storing water before a drought). Not only can each specific behavior be evaluated differently, but some behaviors might not even be relevant for many people in a sample. This leads to the dilemma that evaluations of such specific behaviors may often be highly valid and reliable but only for a small part of the population; hence, the results might not be well generalizable. In turn, more general questions about adapting to CC consequences can help in addressing all persons of a population and generating generalizable results, but it remains unclear to what participants of a study are actually referring to.

For the current study, we chose a new approach by assessing adaptation behavior at the level between the abovementioned extremes. We assumed that classes of behaviors can be specified such that the behaviors are more similar to each other and more different from the behaviors of other classes. We distinguish (1) performing or changing individual behaviors, which often require changes in daily routines (e.g., taking fewer and/or shorter showers to save water); (2) investments, which are also behaviors at the individual or household level but are performed only seldomly and, thus, do not affect daily routines (e.g., buying a water tank to cover shortages of water supply); (3) measures at the community level, for example, the inhabitants of a village cooperating in constructing a water reservoir; and (4) support communal, regional, or national policies to address the adverse consequences of CC (e.g., regulations regarding the use of water for agriculture and electricity production). This system allows for generalization over several specific behaviors and any survey participant should be able to think of some behaviors of each class relevant to them and, thus, will understand what is meant with each behavior class. However, the abstraction from specific behaviors by evaluating behavior classes impedes the investigation of affective or symbolic (e.g., importance of a behavior for demonstrating status) aspects because such aspects depend on specific characteristics of the behaviors.

Besides these behavior classes, we distinguished three CC consequences to which a behavior prepares: (1) droughts, which are the most imminent and widely discussed threat to the region; (2) flooding, which is the CC consequence most frequently investigated in the Global North but also common in the investigated region; and (3) new diseases and pests that are not common to a region. The latter CC consequence is an important threat to the region but is more abstract than the other two consequences.

The proposed system allows two approaches for analyses. First, general tendencies can be estimated by running analyses over all data without distinguishing the types of CC consequences or classes of behaviors. This approach abstracts the idiosyncrasies of CC consequences and behavior classes; therefore, the results should be better generalizable. Here, the proposed system serves to improve the data quality by better specifying the evaluation objects (i.e., behaviors). Second, analyses that distinguish between different CC consequences or behavior classes can be run. This allows the investigation of differences between consequences and behaviors. In the current study, both approaches were used.

1.2. Determinants of Adaptation Behaviors

Many behavior change campaigns apply at least one of the following three approaches: convincing people of the advantageous characteristics of the target behavior (e.g., saving money), targeting more general evaluations not directly related to the behavior (e.g., trust in a brand), and referring to or creating (e.g., storytelling) experiences and expectations that promote the target behavior. Although all three approaches can be combined, they are related to rather distinct research traditions. Therefore, in the following, we present the potential determinants of adaptation behaviors as grouped by these three research traditions.

We start with the first approach: explaining behaviors based on how people evaluate the behavior performance. Most commonly used general behavior models in psychology take this perspective, such as the theory of planned behavior (TPB; [10,11]), the health action process approach (HAPA; [12]), and the norm activation model (NAM; [13]). Most of these models distinguish between evaluations of the consequences of the behavior and considerations about being able to perform and/or control the behavior. The most common construct representing evaluations of consequences is attitudes, which have instrumental (i.e., cost-benefit evaluations) and affective (i.e., how enjoyable or unpleasant a behavior is) aspects [14]. Because this investigation refers to behavior classes instead of single behaviors, affective aspects could not be considered. The latter varies much more between specific behaviors than instrumental evaluations. For example, using less water for showering or washing dishes might have similar instrumental evaluations. However, the affective evaluation might differ a lot. Within instrumental attitudes, many of the costs and benefits of a behavior might be considered. Of particular interest for the present study is the subjective effectiveness of the behaviors for reducing the respective aversive consequences of CC, which we call perceived adaptation efficacy [5]. Another important consequence-evaluating construct in many behavior theories—and considered as determinant of adaptation behaviors [7]—is norms. Norms are often divided into descriptive (i.e., what is “normal”, what most people are doing) and injunctive (i.e., what is the right thing to do, what is positively evaluated by others) norms [15]. The positive evaluations of a behavior do not necessarily lead to enacting it, which leads to the intention-behavior gap. Situational barriers can impede the performance of an intended behavior. Detailed considerations of such barriers and the coping strategies used to overcome them substantially increase the explicative power of behavior models. However, in the current study, we investigate classes of behaviors and are limited to a rather general evaluation of the feasibility of the performing behaviors of an entire class. To summarize the theoretical basis of the current study as derived from psychological behavior models, we expect that (1) cost-benefit evaluations, in general, and (2) perceived adaptation efficacy, in particular, (3) injunctive and (4) descriptive norms, and (5) a general evaluation of the feasibility of performing a class of behaviors are positively related to adaptation to CC consequences.

Another approach to explain behaviors is to start with more general evaluations. The research related to CC often focuses on risk perception. In the literature, there have been two interpretations of this concept. Some researchers focused on the threats from CC as abstract phenomena [16]. In these studies, no reference to any specific consequence of CC has usually been made. The interpretation closer to common research on risk perception assesses the severity and probability of the events related to CC (e.g., flooding), often without any reference to CC [5,7]. The present study considers both interpretations, where the former is called risk perception and the latter expectations. An important construct in the research tradition of risk perception and communication is trust. A meta-analysis by [17] revealed that, particularly, trust in scientist and environmental groups is correlated with CC-related behaviors, whereas more general measures of trust have smaller effects. In the current study, we include general measures of trust and measures of trust in specific groups, organizations, and institutions. Furthermore, a wide range of general evaluations exist that could be related to adaptation but are rarely investigated in such a context. Considering that many consequences of CC are natural events, the perception of, or attitude toward, nature could be a relevant factor. Most studies investigating attitudes toward nature have focused on (more) natural places compared to more civilized (e.g., urbanized) environments [18]. However, for the present investigation, we were interested in the attitude toward nature as something “good” or “bad”, independent of a comparison to cultivated or urban environments. The literature coming the closest to this concept focuses on the ambivalent emotional impacts and meanings of the natural environment experience [19,20]. In the current study, we have followed the approach used by [21], who assessed attitudes toward green spaces in two dimensions (one positive and one negative). However, in our study, instead of attitudes toward green spaces, we assessed the attitudes toward nature, that is, whether people perceive nature as kind or threatening. Other general evaluations that might be related to adaptation behaviors could include general optimism/pessimism, how much people feel that they can influence development, or who they think is responsible for doing something about the threat of CC. Because these constructs were not related to adapting to CC consequences in our explorations, we do not further explain them.

The concept of risk perception is often referred to as analytic processing, from which experiential processing is distinguished [22,23,24]. The idea is that experiences can determine decisions based on the emotions they provoke, which can override analytical considerations. Because of this expected strong effect on decisions, experiences with climate-related events and their effects are often investigated [25,26]. These experiences can be seen from a more “objective” perspective (e.g., number of times an event, such as flooding, has been experienced) or a more “psychological” perspective (e.g., how much one still feels intimidated from the experiences made). The connection between experiences and behaviors is, however, complex [27], and it might be rather the expectations about future events than experiences of past events that drive behaviors. Such expectations are one way to conceptualize risk perception and have been used in some studies [5,7]. These authors followed the scientific definition of risk and distinguished the severity and probability of events. For nonscientists, severity and probability might, however, be difficult to grasp. Therefore, we assessed the expected changes in the intensity and frequency of events and the felt certainty of these estimates. These expectations represent a cognitive, almost “rational” form of expectation. A more emotional form of (negative) expectation involves worrying about certain events. Although worries related to the rather abstract phenomenon of CC have received much attention [28], in the current study, we focused on worries about specific events independent of their cause, which, in our case, was CC.

The goal of the current study was to identify the determinants of adaptation behaviors for a population in the region of Cusco (Peru). For modeling behavior, we combined a retrospective (i.e., what behaviors the participants already performed) with a prospective (i.e., in what direction the participants intended to change their behaviors) perspective: a person who has positive evaluations might score high on performed behavior but not on the intention to change because the behavior is already high. In turn, even with positive evaluations, performed behavior can be low, but a person might want to increase behavior in the future. Therefore, only by combining both perspectives can correlations between evaluations and behavior be expected for all participants. This behavioral disposition, which combines the reported behavior and intentions to change the behavior, is abbreviated as B+I (behavior + intention). This construct was regressed on the behavior determinants introduced above or correlated with experiences and expectations.

2. Materials and Methods

2.1. Sample

Data were gathered in the region of Cusco, Peru. A detailed description of the study region and sample can be found in [29], of which the map of the study region is presented in Figure 1. Most participants (61.9%) came from the city of Cusco (approx. 400,000 inhabitants). The participants from Cusco are highly urbanized and many are less exposed to the effects of CC than the rest of the sample. With 16.5% of the participants, the second most important provider of participants is the city of Urubamba (approx. 14,000 inhabitants). Urubamba is also an urban area but is much more exposed to flooding and droughts than Cusco. The remainder of the sample lived in rural areas around Cusco (Izcuchaca and Huacarpay: 12.5%) and Urubamba (San Isidro de Chicón: 9.1%). These communities differ from their urban neighbors in that they are economically less well-off and more vulnerable to the consequences of CC.

Data were gathered via computer-assisted structured face-to-face interviews in Spanish with trained local interviewers from May 2016 to January 2017. For sampling members of the general public, a random route procedure [30] was used: the interviewers started from locations specified on a map and went in all directions, asking in every second house whether someone was willing to participate. According to population statistics for the Department of Cusco [31], all segments of the population have been well represented. However, people in the age category 20–29 years old and those with a university degree were overrepresented (difference in relative proportion: 11% and 26%, respectively), whereas people with no formal or primary education level (difference: 8% and 12%) and who learned Quechua as their first language (difference: 23%) were underrepresented. Of the 3609 people approached, 2059 (57.0%) agreed to start the interview. Of these, 255 participants did not speak Spanish and, therefore, answered a much shorter questionnaire in Quechua, and six stated that they lived in Lima. These 261 participants were excluded from the analysis.

To ensure data quality, the interviewers received extensive training and were briefed daily. Furthermore, we used item discrimination to identify participants with poor data quality. Several batteries assessed two different constructs, of which we expected different evaluations for most participants, even though the wording of the items looked similar. One example is the battery assessing the risk perception of CC, where items assessing possible threats were interlaced with items assessing possible advantages. Although theoretically possible, it is very improbable that a person would evaluate threats and advantages equally. Therefore, we interpreted too similar answers for both constructs as an indication of low data quality and excluded these participants from the analyses.

The final sample comprised 1298 respondents, of which 53.6% were women, 46.4% were men, and the mean age was 35 years old (s = 14.8). Each respondent only answered to a part of the entire questionnaire, because the survey included a wide range of topics, and answering all the items would have required approximately four hours. Therefore, each participant answered only a part of the items so that the interviews typically took 50–90 min. Six versions with overlapping sets of topics were created and randomly assigned to the participants. All analyses presented here were based on three questionnaire versions that included items related to adaptation to CC consequences. A total of 640 participants received one of these versions, of which 213 received items related to droughts, 212 received items related to flooding, and 215 received items related to new diseases. Furthermore, only 224 participants answered items both on adaptation and on experiences, and of the participants answering items on adaptation, only a part received items on expectations referring to the same CC consequences. This led to a rather complex sample structure, which is presented in Figure 2.

For each analysis performed with the entire database, separate power analyses were performed. For the analyses in this paper, the requirement was 182 cases for the regression analyses of each of the subsamples. With this number of cases, an R² of 10% can be detected in a 15-parameter model with a power of 80% and p = 0.05. With perfectly distributed cases across all questionnaire versions and no missing data, a total sample size of 1091 cases would be needed.

2.2. Procedure and Measures

All data used for the analyses, except the number of experienced events and structural factors, were assessed with 5-point (in the case of unipolar items) or 9-point (in the case of bipolar items) Likert scales. During the interviews, the participants could use a printed scale to indicate their answers (the extreme and neutral values were labeled verbally). The interviewers explained how to use the scale before asking the first question. To facilitate comparability and interpretation of the unstandardized predictors in the regression models, unipolar variables were normalized to the range from 0 to 1, and bipolar items were normalized in a range from −1 to 1. The wording of the items used in the questionnaire and the answer options are available in the Supplementary Materials.

2.2.1. Constructs Related to Adaptation Behaviors

The items referring to adaptation behaviors were organized into blocks, with each concerning a different CC consequence (droughts, flooding, and new diseases, with the latter referring to insect-born human diseases, such as Zika). Only one randomly selected block was presented to each participant. Within these blocks, the items were grouped by behavior class so that the participants assessed all evaluations of one behavior class and then all evaluations of the next class. Before starting with the questions about a behavior class, the interviewers explained the CC consequences and presented, for each class, examples of adaptation behaviors. All the constructs were assessed with one-item scales. For some analyses, the items of the four behavior classes were combined into four-item scales, to which the scale characteristics presented below are referred.

Outcome variable (B+I): The outcome variable was the average of the reported adaptation behavior and intention to change this behavior. Adaptation behavior was assessed with one item for each behavior class (e.g., for investments, we asked “To what extent are you making such investments?”), offering five answer options ranging from “does nothing” to “does as much as possible”. The same held for intention, where the item for investments stated, “How much are you intending to increase, decrease, or maintain making such small investments, in the future?” with nine answer options ranging from “intention to reduce strongly” to “intention not to change” to “intention to increase strongly”. The scale combining the data for all four behavior classes had good internal consistency (Cronbach’s alpha = 0.795).

Perceived adaptation efficacy: For each behavior class, we asked how effective the behaviors of this class are perceived in addressing the respective CC consequence (e.g., for investments: “What do you think, how effective is it to make such small investments?”). The five answer options ranged from “not effective at all” to “completely effective”. The scale of the four items from the different behavior classes had rather low internal consistency (Cronbach’s alpha = 0.633) that needs to be considered in the interpretation.

Cost-benefit evaluations: The participants indicated whether they thought it was worthwhile to perform the respective behavior (e.g., for investments: “Considering all costs and benefits, do you think it is worthwhile to make such small investments?”). Nine answer options were provided ranging from “much more costly” to “equally costly as beneficial” to “much more beneficial”. When aggregated into a scale, the four items from the different behavior classes had sufficient internal consistency (Cronbach’s alpha = 0.699).

Descriptive norm: These items asked how many relevant others were performing the respective behaviors (e.g., for investments: “How many people who are important to you make such small investments?”). The five answer options ranged from “nobody” to “all”. Combining all four measures led to a scale with sufficient internal consistency (Cronbach’s alpha = 0.699).

Injunctive norm: The participants were asked to anticipate how relevant others would react if the participants performed the respective behavior (e.g., for investments: “If you would make such small investments, would people who are important to you think rather good or bad of you?”). The nine answer options ranged from “totally bad” to “neither bad nor good” to “totally good”. The combination of the four items had sufficient internal consistency (Cronbach’s alpha = 0.690).

Perceived feasibility: The participants indicated how feasible they thought it was to perform the respective behavior. The five answer options ranged from “not possible at all” to “totally possible”. The internal consistency of the combined scale was sufficient (Cronbach’s alpha = 0.730).

2.2.2. General Evaluations

General evaluations were assessed in different parts of the questionnaire, and all participants answered these items.

Risk perception of CC: This construct was assessed according to [16] via eight items comprising how concerned the participants were about CC, how often they worried about CC consequences, the subjective likeliness and seriousness of adverse effects of CC for the participant, society/Peru, and the natural environment (e.g., “How serious of a threat do you think that climate change is to the natural environment?”). The items had five answer options ranging from “not at all” (e.g., “not preoccupied at all”, “not probable at all”) to “very” (e.g., “very preoccupied”) or “almost certain”. The internal consistency of the scale was good (Cronbach’s alpha = 0.816).

Trust in specific sources: Trust was assessed as general trust (see below) and the trust in, or perceived trustworthiness of, various groups, organizations, and institutions. On the one hand, we asked how much the participants trusted (1) their friends, (2) strangers in their village/quarter, representatives of (3) the government, (4) NGOs and interest groups, (5) their religious community, and (6) commercial and industrial enterprises. On the other hand, we asked how much the participants thought that the following sources of information tell the truth: (1) their friends, (2) the press, (3) radio and television, (4) the government, (5) religious leaders, (6) nonprofit organizations, (7) environmental interest groups, and (8) scientists. All the items had five answer options ranging from “don’t trust at all” to “totally trust”. When aggregated into a scale, the 14 items had very good internal consistency (Cronbach’s alpha = 0.907). The detailed descriptive statistics of these items are compiled in Appendix A,

Table A1. Descriptive statistics for the items assessing trust in groups and institutions, calculated for all cases, including cases that did not answer to items on adaptation to CC consequences.

Variable		Range	N	Mean	SE	Std. Dev.
How much do you trust the following people?
	Your friends	[0, 4]	1245	2.092	0.036	1.260
	Strangers in your village/quarter	[0, 4]	1245	0.822	0.027	0.965
	Representatives of the government	[0, 4]	1236	0.731	0.029	1.026
	Representatives of NGOs and interest groups	[0, 4]	1239	1.092	0.032	1.132
	Representatives of your religious community	[0, 4]	1238	1.359	0.037	1.308
How much do you think that the following sources of information say the truth?
	Your friends	[0, 4]	1236	1.870	0.034	1.206
	The press	[0, 4]	1239	1.051	0.029	1.025
	Radio and television	[0, 4]	1239	1.065	0.030	1.048
	The government	[0, 4]	1240	0.742	0.028	0.995
	Religious leaders	[0, 4]	1234	1.211	0.034	1.211
	Non-profit organizations	[0, 4]	1220	1.217	0.032	1.130
	Environmental interest groups	[0, 4]	1228	1.726	0.035	1.225
	Scientists	[0, 4]	1220	2.117	0.035	1.224

Notes: N = number of cases; SE = standard error; Std. dev. = standard deviation; DV = dependent variable.

.

General trust: To assess this construct, we asked (1) how safe the participants felt at night, (2) whether something of value lost on the street would be returned, (3) whether there were many people the participants can trust, and (4) whether they were confident that, if needed, even a stranger would help. Each item had five response options ranging from “do not agree at all” to “totally agree”. The heterogeneity of the items led to a lower internal consistency (Cronbach’s alpha = 0.597) that needs to be considered in the interpretation.

Attitude toward nature: This construct was assessed with a two-factorial ad hoc scale comprising positive and negative attitudes toward nature. For positive attitudes, we used five items relating nature to positive adjectives (e.g., “Nature is good”) or relations (e.g., “We depend on nature”). Following the same principle, the five items for assessing negative attitudes toward nature used negative adjectives (e.g., “Nature is cruel”) or relations (e.g., “We are at the mercy of nature”). The two relational items reduced the reliability of the scale; therefore, only the three items based on adjectives were used to construct a scale. All the items had five answer options ranging from “do not agree at all” to “totally agree”. The internal consistency of both scales was sufficient, with Cronbach’s alphas of 0.705 and 0.719, respectively.

2.2.3. Experiences and Expectations

Of the 640 participants who received the questionnaire versions with items related to adaptation behaviors, only 224 (35.0%) were also assigned to questionnaire versions that comprised a rather large section on experiences related to weather events and CC-related phenomena. A short introduction explained that the questions referred to the following weather- or climate-related events: (1) droughts and water shortages, (2) severe and unusual flooding, and (3) diseases and pests that have not been common in the region.

Number of respective events experienced: The participants indicated how many times they had experienced the respective events in the past five years in the local area (0 = never experienced, 1 = experienced once, 2 = experienced twice, 3 = experienced three times, 4 = experienced four or more times). Note that this was the only variable not normalized to 1.

How much still intimidated with their respective experiences: Those participants who answered that they had experienced the respective event at least once were asked how much they still felt intimidated about these experiences (“How intimidated are you, even still, about these experiences?”). The five answer options ranged from “not at all intimidated” to “very intimidated”. Because all the participants who did not report having experienced the respective event had missing values for this variable, we also constructed a variable where such missing values were set to 0 (assuming that the lack of experience meant not being intimidated). All analyses, including intimidation, were calculated with both variables.

Expectations were assessed in all questionnaire versions, but only half of the items were presented to the participants, to limit the length of the interviews. The items were grouped by topics, and a predefined selection of items was randomly assigned to each participant. In this way, of the 640 participants who answered questions about adaptation behaviors, we also assessed expectations about drought of 104 (16.3%) participants, about flooding of 80 (12.5%), and about diseases in 86 (13.4%).

Expected changes: The participants indicated how they thought the frequency and duration/severity of droughts, flooding, and diseases would change in the future. For example, for flooding, we asked “What do you think, how much will the frequency of severe and unusual flooding decrease or increase in the future?” and “What do you think, how much will the magnitude of flooding decrease or increase in the future?” (both items had nine response options ranging from “will strongly decrease” to “there will be no change at all” to “will strongly increase”). For diseases, we only asked about expected changes in frequency.

Certainty about changes: We also asked about how certain participants felt about their expectations (“How certain are you that these changes will occur/that no change at all will occur?”), with five answer options ranging from “not certain at all” to “totally certain”).

Worry about respective events: Worries were directly assessed with the following question: “How worried are you about the following natural events? Droughts, flooding, and diseases for people, livestock, and crops”, with five answer options ranging from “not at all preoccupied” to “extremely preoccupied”.

2.3. Data Analyses

The core of our analyses was a linear regression model that we fitted in three different forms to the available data. First, we calculated scales from the four items of each construct, which assessed the values for different behavior classes. We fitted the model to all cases that answered questions related to adaptation behaviors (n = 640). This analysis led to general results concerning the relationships between the investigated constructs and B+I. The second analysis step consisted of fitting the model four times to the same data, here using one-item scales representing a behavior class (e.g., changing daily behaviors), each. This analysis allowed for the identification of differences between behavior classes. The third step used the same four-item scales as the first step, but the model was fitted to three different subsamples. Each subsample answered the items on adaptation related to a different CC consequence: in this case, 213 received items related to droughts, 212 to flooding, and 215 to new diseases. This analysis allowed the identification of differences in adaptation to different CC consequences.

In the final analysis, we investigated the relationships of experiences and expectations with B+I and risk perception. Because only a few participants answered items on adaptation and experiences and the three CC consequences were investigated seperately, the number of cases was too small to run regression analyses (n for droughts = 104; n for flooding = 80; n for diseases = 86). Therefore, we investigated these relationships via correlation analyses.

A priori, we assumed that all behavior-specific evaluations, risk perceptions, and experiences and expectations would be positively correlated with adaptation. We explored the effects of various general evaluations, that is, (1) trust, (2) attitudes toward nature, (3) optimism, (4) pessimism, (5) perceived influence on developments at different levels—from family to global, and (6) the perceived responsibility of different groups and institutions to reduce the negative effects of CC consequences. Furthermore, we investigated the effects of a wide range of structural factors (gender, age, education, income, number of children, household size, political position, religion, frequency of attending religious services, time living at the current location, and whether the participants were informed of living in a risk zone). Finally, we explored some items that indicated actual and desired forms of decision making. The descriptive statistics for these items and the results of an exploratory regression analysis are presented in Appendix B. Further descriptive statistics on how decisions are made at home are compiled in Appendix A,

Table A2. Answer frequencies of the item assessing the form of decision making at home, calculated for all cases, including cases that did not answer to items on adaptation to CC consequences.

Value	N	Missings	Frequency	Valid %
Which of the following statements describes best how decisions are made at your home?
The issue is discussed amongst the entire household, and all decide together (e.g., by vote).	1239	59	795	64.17
The issue is discussed amongst the entire household and some of the household decide.	1239	59	254	20.50
The issue is discussed amongst the entire household and only one person decides.	1239	59	74	5.97
The issue is discussed amongst some of the household, and they decide together.	1239	59	73	5.89
The issue is discussed amongst some of the household and only one person decides.	1239	59	29	2.34
The issue is not discussed with others in the household and one person decides.	1239	59	14	1.13

.

We will only present the results for a model comprising the constructs of which we expected effects—whether they had an effect or not—and constructs that showed effects, at least in some analyses and for some of its variables (e.g., we included all data-gathering regions, although some never showed effects). The investigated data are included in the database provided in the Supplementary Materials, and the syntax includes a simple approach to explore the effects of variables not included in the model. However, many tests were performed, and no significance correction for multiple tests was applied. Therefore, the effects found by exploration would require more confirmation by future studies.

We used an unweighted least-squares estimator with forced entry of all variables at once. Missing values were excluded listwise. These settings are not only common, but they also allow a direct interpretation of the results. The following preconditions for interpreting regression results were tested: heteroscedasticity (by visual evaluation of estimate-residual diagrams), non-normal distribution of residuals (by visual evaluation of residual distribution), autocorrelation (Durbin–Watson test), and multicollinearity (variance inflation factor, VIF). We did not test nonlinear relationships between the independent and dependent variables. The data analyzed and SPSS syntaxes used to produce the results are available in the Supplementary Materials.

3. Results

3.1. Explaining Adaptation Behaviors and Intentions

In this subsection, we present the results of the regression analyses that explain the variance of B+I. We distinguish among the structural factors, general evaluations, and behavior-specific evaluations. First, we analyze the relationship at a general level without distinguishing between different classes of adaptation behaviors or consequences of CC to which to adapt. Then, differences in the results because of different behavior classes or CC consequences are presented. All variables have been normalized to ranges of [0, 1] (unipolar) and [–1, 1] (bipolar).

3.1.1. General Results

The descriptive statistics for the averages of all behavior classes and with all cases are compiled in

Table 1. Descriptive statistics for the variables and cases used in the regression analyses, calculated for all behavior classes and all consequences of CC.

Variable		Range (Normalized)	N	Mean/ n ‘Yes’	SE/ % ‘Yes’	Std. Dev.
DV	B+I	[−1, 1]	589	0.378	0.010	0.236
Structural factors	Is woman	[0; 1]	636	341	53.6
	Age	[0, 1]	620	35	0.592	14.74
	Cusco	[0; 1]	640	396	61.9
	Chicón	[0; 1]	640	58	9.1
	Urubamba	[0; 1]	640	101	15.8
	Huacarpay or Izcuchaca	[0; 1]	640	85	13.3
General evaluations	Risk perception of CC	[0, 1]	632	0.754	0.006	0.155
	Trust in specific sources	[0, 1]	617	0.322	0.008	0.191
	General trust	[0, 1]	611	0.370	0.009	0.220
	Positive attitude towards nature	[0, 1]	612	0.889	0.006	0.147
	Negative attitude towards nature	[0, 1]	612	0.516	0.012	0.308
Behavior-specific evaluations	Perceived adaptation efficacy	[0, 1]	591	0.634	0.008	0.203
	Cost–benefit evaluations	[−1, 1]	587	0.388	0.016	0.376
	Descriptive Norm	[0, 1]	588	0.367	0.009	0.216
	Injunctive Norm	[−1, 1]	579	0.372	0.012	0.297
	Perceived feasibility	[0, 1]	588	0.606	0.009	0.225

Notes: N = number of cases; SE = standard error; Std. dev. = standard deviation; DV = dependent variable.

. B+I is relatively low (m = 0.378). Considering that B+I is a behavior measure modulated by intention, ideally, its values should be positive or only slightly negative because a behavior cannot be performed negatively. In fact, only 16 (4.8%) participants had negative scores, of which only 10 (2.7%) had even values < −0.10. This means that these participants indicated strong intentions to reduce their adaptation behaviors, despite performing such behaviors at a low level. In contrast to B+I, the risk perception of CC is high (m = 0.754); thus, campaigns might not be able to further increase it. The indicators of trust are, again, rather low (both < 0.4). Although negative attitudes toward nature have medium values (m = 0.516), positive attitudes have high values (m = 0.889). With respect to the behavior-specific evaluations, adaptation efficacy and feasibility are evaluated rather highly (both >0.6) but still have room for improvement. The evaluations regarding costs and benefits, and both norms are rather low (all <0.4).

The results of the overall regression of B+I on the selected predictors are presented in

Table 2. Results of the regression of B+I on structural factors and general and behavior-specific evaluations for all behavior classes and all consequences of CC.

Effects	Beta	B	SE	95% CI		p	Collin. Stat.
				LL	UL		Tol.	VIF
(Constant)		−0.062	0.053	−0.166	0.043	0.245
Is woman	−0.030	−0.014	0.013	−0.038	0.011	0.277	0.962	1.040
Age	0.062	0.098	0.045	0.010	0.185	0.029	0.885	1.130
Chicón	−0.009	−0.008	0.024	−0.054	0.039	0.754	0.842	1.188
Urubamba	−0.074	−0.047	0.018	−0.082	−0.011	0.010	0.873	1.145
Huacarpay or Izcuchaca	−0.118	−0.080	0.019	−0.118	−0.042	<0.001	0.871	1.148
Risk perception of CC	0.010	0.016	0.045	−0.072	0.103	0.728	0.824	1.213
Trust in specific sources	0.106	0.125	0.039	0.048	0.201	0.001	0.654	1.529
General trust	0.098	0.104	0.033	0.040	0.168	0.002	0.743	1.346
Positive attitude towards nature	−0.050	−0.080	0.046	−0.171	0.011	0.083	0.849	1.179
Negative attitude towards nature	0.077	0.058	0.022	0.015	0.101	0.008	0.852	1.174
Perceived adaptation efficacy	−0.042	−0.047	0.043	−0.132	0.037	0.273	0.485	2.063
Cost–benefit evaluations	0.235	0.148	0.023	0.103	0.193	<0.001	0.537	1.862
Descriptive Norm	0.267	0.284	0.035	0.215	0.353	<0.001	0.649	1.540
Injunctive Norm	0.121	0.094	0.026	0.043	0.144	<0.001	0.646	1.549
Perceived feasibility	0.355	0.365	0.042	0.283	0.448	<0.001	0.425	2.351

Notes: adj. R² = 0.615, F(15, 526) = 58.518, p < 0.001; n = 542. Beta = standardized B; B = estimate of the linear effect; SE = standard error; CI = confidence interval; LL = lower limit; UL = upper limit; p = probability of error; Collin. stat. = collinearity statistics; Tol. = tolerance; VIF = variance inflation factor.

. The model fits the data well and explains 61.5% of the variance. The adjusted R² of only the structural variables is 4.1%. Adding the general evaluations increases the adjusted R² to 17.8%. Therefore, by far, the largest share of the explained variance (43.7%) comes from behavior-specific evaluations.

The strongest effects on B+I are perceived feasibility (B = 0.365) and descriptive norms (B = 0.284), but cost-benefit evaluations (B = 0.148) and trust in specific sources (B = 0.125) are also important variables. Although the effect of the injunctive norm is smaller (B = 0.094), it is highly statistically significant. Additionally, general trust has a statistically significant effect. Thus, adaptation behaviors are related mainly to behavior-specific evaluations and trust. Surprisingly—and contrary to our hypothesis—neither risk perception (B = 0.016) nor perceived adaptation efficacy (B = −0.047) have any effect on adaptation. This will be discussed later.

Finally, attitudes toward nature also have a weak effect, which, for the negative attitudes, even reach statistical significance. Thus, people with more negative and less positive attitudes toward nature adapt more to CC consequences. Furthermore, the participants from Urubamba, Huacarpay, and Izcuchaca present lower levels of B+I.

3.1.2. Comparing Different Classes of Adaptation Behaviors

To investigate the differences between the four classes of adaptation behaviors, we analyzed them separately. The descriptive statistics between the classes differ only minimally (

Table 3. Means (standard deviations, number of cases) of B+I and behavior-specific evaluations for the different classes of behaviors, calculated for all CC consequences.

Variable		Changing Daily Behaviors	Investments	Community Projects	Policy Support
DV	B+I	0.44 (0.31, 580)	0.36 (0.31, 578)	0.36 (0.31, 578)	0.34 (0.28, 573)
Behavior-specific evaluations	Perceived adaptation efficacy	0.64 (0.29, 582)	0.65 (0.28, 575)	0.65 (0.28, 575)	0.59 (0.32, 564)
	Cost–benefit evaluations	0.37 (0.55, 572)	0.32 (0.55, 571)	0.32 (0.55, 571)	0.4 (0.47, 563)
	Descriptive Norm	0.39 (0.28, 570)	0.33 (0.28, 563)	0.33 (0.28, 563)	0.34 (0.3, 559)
	Injunctive Norm	0.36 (0.42, 563)	0.34 (0.4, 560)	0.34 (0.4, 560)	0.35 (0.43, 553)
	Perceived feasibility	0.66 (0.29, 573)	0.57 (0.29, 563)	0.57 (0.29, 563)	0.59 (0.3, 558)

Notes: DV = dependent variable.

).

Additionally, regarding the effects, only a few differences are apparent (

Table 4. Parameter estimates (B) and their confidence intervals of the regression of B+I on the structural factors and general and behavior-specific evaluations for each behavior class separately.

Effects	Changing Daily Behaviors	Investments	Community Projects	Policy Support
(Constant)	−0.124 [−0.287, 0.038]	−0.020 [−0.192, 0.153]	−0.171 * [−0.322, −0.021]	−0.070 [−0.228, 0.088]
Is woman	−0.011 [−0.050, 0.027]	−0.019 [−0.059, 0.021]	−0.011 [−0.047, 0.025]	−0.035 [−0.073, 0.003]
Age	0.060 [−0.077, 0.198]	0.142 [−0.002, 0.286]	0.126 [−0.003, 0.255]	0.055 [−0.077, 0.188]
Chicón	−0.039 [−0.114, 0.036]	−0.004 [−0.080, 0.071]	0.011 [−0.059, 0.081]	−0.033 [−0.105, 0.038]
Urubamba	−0.035 [−0.091, 0.021]	−0.015 [−0.072, 0.043]	−0.052 [−0.104, 0.001]	−0.077 ** [−0.130, −0.023]
Huacarpay or Izcuchaca	−0.070 * [−0.130, −0.010]	−0.073 * [−0.136, −0.010]	−0.128 *** [−0.183, −0.072]	−0.110 *** [−0.169, −0.050]
Risk perception of CC	0.092 [−0.043, 0.227]	0.056 [−0.082, 0.195]	0.046 [−0.085, 0.177]	0.005 [−0.128, 0.138]
Trust in specific sources	0.024 [−0.094, 0.143]	0.172 ** [0.052, 0.293]	0.118 * [0.008, 0.229]	0.305 *** [0.191, 0.420]
General trust	0.091 [−0.008, 0.190]	0.052 [−0.053, 0.156]	0.155 ** [0.060, 0.250]	0.009 [−0.089, 0.108]
Positive attitude towards nature	−0.020 [−0.161, 0.122]	−0.162 * [−0.316, −0.007]	0.019 [−0.113, 0.150]	−0.004 [−0.143, 0.136]
Negative attitude towards nature	0.052 [−0.014, 0.119]	0.054 [−0.015, 0.124]	0.069 * [0.007, 0.131]	0.083 * [0.018, 0.148]
Perceived adaptation efficacy	0.162 *** [0.081, 0.243]	0.038 [−0.046, 0.122]	0.030 [−0.056, 0.115]	0.014 [−0.064, 0.093]
Cost–benefit evaluations	0.118 *** [0.075, 0.160]	0.060 ** [0.015, 0.106]	0.067 ** [0.017, 0.117]	0.071 ** [0.018, 0.124]
Descriptive Norm	0.246 *** [0.169, 0.323]	0.230 *** [0.148, 0.313]	0.252 *** [0.180, 0.324]	0.183 *** [0.116, 0.251]
Injunctive Norm	0.019 [−0.031, 0.069]	0.099 ** [0.039, 0.160]	0.095 ** [0.036, 0.154]	0.126 *** [0.073, 0.179]
Perceived feasibility	0.324 *** [0.240, 0.408]	0.365 *** [0.277, 0.452]	0.262 *** [0.179, 0.344]	0.263 *** [0.183, 0.344]
Adj. R2 (n)	0.498 *** (506)	0.456 *** (497)	0.480 *** (495)	0.471 *** (488)

Note: * p < 0.05, ** p < 0.01, *** p < 0.001.

). All four models explain approximately the same variance (R² is between 46% and 50%) so that the effects can be well compared between the models. Most importantly, perceived adaptation efficacy now has a significant effect but only for changing daily behaviors (B = 0.162). In addition, the positive attitude toward nature reaches statistical significance, but only for investments (B = −0.162). For trust, we find that, for individual behaviors (B = 0.024, for changing daily behaviors), trust in specific sources is less important, whereas its effects increase for public behaviors, such as policy support (B = 0.305), with general trust being significant only for community projects (B = 0.155). For community projects and policy support, the effect of perceived feasibility is also lower (B = 0.262 and 0.263, respectively). Finally, for changing daily behaviors, cost-benefit evaluations appear to be more important (B = 0.118), and injunctive norms are less important (B = 0.019).

3.1.3. Comparing Different Consequences of CC

In the last step of the regression analyses, we compared the results for the different consequences of CC: droughts, flooding, and the spread of new diseases. The sample was split into subsamples of participants who answered the adaptation items with reference to one of the three CC consequences. Therefore, these analyses were performed with considerably fewer cases (between 179 and 182). Additionally, the descriptive statistics of the three subsamples are almost identical (

Table 5. Means (standard deviations, number of cases) for the variables and cases used in the regression analyses, for the different consequences of CC, calculated over all classes of behaviors.

Variable		Droughts	Flooding	Diseases
DV	B+I	0.42 (0.24, 198)	0.35 (0.23, 198)	0.36 (0.24, 193)
Structural factors	Is woman	107 (51.2%) of 209	111 (52.4%) of 212	123 (57.2%) of 215
	Age	0.34 (0.14, 204)	0.36 (0.15, 207)	0.35 (0.14, 209)
	Cusco	128 (60.1%) of 213	136 (64.2%) of 212	132 (61.4%) of 215
	Chicón	19 (8.9%) of 213	20 (9.4%) of 212	19 (8.8%) of 215
	Urubamba	38 (17.8%) of 213	38 (17.9%) of 212	25 (11.6%) of 215
	Huacarpay or Izcuchaca	28 (13.1%) of 213	18 (8.5%) of 212	39 (18.1%) of 215
General evaluations	Risk perception of CC	0.76 (0.15, 207)	0.75 (0.15, 212)	0.75 (0.17, 213)
	Trust in specific sources	0.33 (0.2, 204)	0.33 (0.19, 205)	0.3 (0.18, 208)
	General trust	0.38 (0.23, 199)	0.37 (0.21, 204)	0.36 (0.22, 208)
	Positive attitude towards nature	0.88 (0.16, 201)	0.9 (0.15, 203)	0.89 (0.14, 208)
	Negative attitude towards nature	0.5 (0.31, 201)	0.54 (0.3, 203)	0.51 (0.31, 208)
Behavior-specific evaluations	Perceived adaptation efficacy	0.68 (0.21, 199)	0.6 (0.18, 198)	0.61 (0.2, 194)
	Cost–benefit evaluations	0.47 (0.38, 198)	0.36 (0.34, 197)	0.33 (0.39, 192)
	Descriptive Norm	0.38 (0.21, 198)	0.36 (0.23, 197)	0.36 (0.21, 193)
	Injunctive Norm	0.43 (0.28, 196)	0.35 (0.31, 194)	0.34 (0.29, 189)
	Perceived feasibility	0.64 (0.23, 198)	0.57 (0.21, 197)	0.61 (0.23, 193)

Notes: DV = dependent variable.

).

All regression models explain approximately the same amount of variance (R² between 60% and 63%) and, thus, the results can be compared (

Table 6. Parameter estimates of the regression of B+I on the structural factors and general and behavior-specific evaluations for each CC consequence separately.

Effects	Droughts	Flooding	Diseases
(Constant)	−0.069 [−0.252, 0.115]	−0.038 [−0.234, 0.158]	−0.089 [−0.274, 0.095]
Is woman	−0.021 [−0.064, 0.022]	−0.031 [−0.075, 0.012]	0.005 [−0.041, 0.050]
Age	0.054 [−0.099, 0.207]	0.115 [−0.032, 0.263]	0.111 [−0.053, 0.275]
Chicón	−0.047 [−0.128, 0.035]	0.007 [−0.073, 0.087]	0.018 [−0.069, 0.105]
Urubamba	−0.065 * [−0.124, −0.007]	−0.026 [−0.087, 0.034]	−0.036 [−0.110, 0.037]
Huacarpay or Izcuchaca	−0.107 ** [−0.172, −0.042]	−0.065 [−0.146, 0.016]	−0.062 [−0.124, 0.000]
Risk perception of CC	0.040 [−0.129, 0.209]	0.063 [−0.102, 0.228]	−0.036 [−0.175, 0.103]
Trust in specific sources	0.021 [−0.106, 0.149]	0.115 [−0.026, 0.256]	0.248 *** [0.102, 0.393]
General trust	0.100 [−0.009, 0.209]	0.085 [−0.031, 0.200]	0.147 * [0.029, 0.265]
Positive attitude towards nature	0.063 [−0.091, 0.216]	−0.165 * [−0.323, −0.007]	−0.125 [−0.303, 0.053]
Negative attitude towards nature	−0.013 [−0.090, 0.065]	0.033 [−0.041, 0.108]	0.131 *** [0.056, 0.207]
Perceived adaptation efficacy	−0.100 [−0.266, 0.066]	−0.078 [−0.222, 0.066]	0.001 [−0.151, 0.154]
Cost–benefit evaluations	0.168 *** [0.082, 0.254]	0.170 *** [0.085, 0.254]	0.113 ** [0.038, 0.188]
Descriptive Norm	0.251 *** [0.127, 0.374]	0.261 *** [0.142, 0.380]	0.312 *** [0.179, 0.444]
Injunctive Norm	0.064 [−0.028, 0.156]	0.121 ** [0.036, 0.205]	0.066 [−0.030, 0.161]
Perceived feasibility	0.412 *** [0.250, 0.574]	0.427 *** [0.283, 0.571]	0.316 *** [0.176, 0.456]
Adj. R2 (n)	0.600 *** (182)	0.610 *** (181)	0.631 *** (179)

Note: * p < 0.05, ** p < 0.01, *** p < 0.001.

). Some important differences can be found in the regression results. Most importantly, only for new diseases—the most abstract threat investigated—do general evaluations play an important role (B = 0.248, 0.147, 0.131, and −0.125 for trust in specific sources, general trust, negative attitudes toward nature, and positive attitudes toward nature, respectively) but not risk perception (B = −0.036). An exception is the (marginally) significant effect of positive attitudes toward nature for flooding (B = −0.165). In addition, descriptive norms are more important for adapting to new diseases (B = 0.312), whereas for adapting to other CC consequences, cost-benefit evaluations and perceived feasibility are more important (for cost-benefit evaluations, B = 0.168 and 0.170; for perceived feasibility, B = 0.412 and 0.427). Injunctive norms are only relevant for adapting to flooding (B = 0.121).

3.2. The Role of Experiences and Expectations for Adaptation to CC Consequences

Only a small number of participants answered both the items on adaptation behaviors and those on experiences and expectations. Therefore, it was not possible to run regression analyses, and we focused on the direct relationship between these variables and B+I (the assumed dependent variable). Moreover, we also examined the correlation with risk perception, which might mediate the effects of experiences and expectations on adaptation, for which more data are available.

The descriptive statistics of these variables are compiled in

Table 7. Means (standard deviations, number of cases) of variables related to experiences and expectations.

Variable	Droughts	Flooding	Diseases
Number of respective events experienced [0, 4]	1.17 (1.49, 200)	0.80 (1.19, 198)	0.74 (1.16, 222)
How much still intimidated (only if experienced)	0.67 (0.27, 99)	0.64 (0.29, 86)	0.60 (0.29, 84)
How much still intimidated (if not experienced set to 0)	0.33 (0.38, 203)	0.28 (0.37, 199)	0.24 (0.35, 212)
Expected change in frequency	0.48 (0.59, 175)	0.12 (0.64, 138)	0.17 (0.59, 171)
Certainty about expected change in frequency	0.67 (0.26, 175)	0.56 (0.25, 135)	0.56 (0.23, 171)
Expected change in duration/magnitude	0.64 (0.38, 173)	0.12 (0.64, 133)	Not assessed
Certainty about expected change in duration/magnitude	0.67 (0.24, 172)	0.57 (0.26, 134)	Not assessed
Worry about respective events	0.77 (0.24, 197)	0.63 (0.28, 207)	0.74 (0.24, 203)

Note: All variables have been normalized to a range of [0, 1], except number of experiences, which has a range of [0, 4].

. On average, the participants had experienced more than one drought (m = 1.17) and almost one flood (m = 0.80) or disease or pest (m = 0.74) within the past five years. These numbers highlight that droughts were the predominant problem in the study region. This is also reflected in expectations, where considerable increases in the frequency (m = 0.48) and duration (m = 0.64) of droughts are expected. These numbers are considerably smaller for flooding and disease (all m < 0.2). In contrast to these differences, how intimidated participants are still about the experienced events, the certainty of the expected changes, and worries are quite similar for the three CC consequences.

Table 8. Correlations between B+I and risk perception, respectively, and variables related to experiences and expectations of CC consequences.

	Droughts		Flooding		Diseases
Variable	B+I	Risk Perception	B+I	Risk Perception	B+I	Risk Perception
Number of respective events experienced [0, 4]	−0.096 [−0.339, 0.160] (n = 61)	0.142 * [0.003, 0.275] (n = 200)	0.029 [−0.217, 0.270] (n = 65)	−0.090 [−0.226, 0.051] (n = 198)	−0.235 [−0.448, 0.007] (n = 67)	0.047 [−0.090, 0.181] (n = 209)
How much still intimidated (only if experienced)	0.351 * [0.003, 0.616] (n = 33)	0.378 *** [0.193, 0.535] (n = 98)	0.050 [−0.324, 0.408] (n = 29)	0.280 ** [0.071, 0.463] (n = 86)	−0.155 [−0.523, 0.268] (n = 24)	0.193 [−0.024, 0.392] (n = 83)
How much are still intimidated (if not experienced set to 0)	−0.084 [−0.326, 0.170] (n = 62)	0.196 ** [0.059, 0.325] (n = 202)	0.102 [−0.144, 0.355] (n = 66)	0.058 [−0.082, 0.196] (n = 199)	−0.183 [−0.404, 0.061] (n = 67)	0.079 [−0.057, 0.212] (n = 210)
Expected change in frequency	−0.117 [−0.317, 0.094] (n = 89)	0.094 [−0.055, 0.239] (n = 175)	−0.132 [−0.355, 0.107] (n = 70)	0.106 [−0.063, 0.269] (n = 137)	0.030 [−0.185, 0.241] (n = 85)	0.172 * [0.021, 0.314] (n = 170)
Certainty about expected change in frequency	0.173 [−0.037, 0.368] (n = 89)	0.314 *** [0.173, 0.441] (n = 175)	0.051 [−0.190, 0.286] (n = 68)	0.180 * [0.011, 0.339] (n = 134)	0.096 [−0.118, 0.300] (n = 87)	0.210 ** [0.061, 0.349] (n = 170)
Expected change in duration/magnitude	0.025 [−0.185, 0.232] (n = 89)	0.067 [−0.083, 0.214] (n = 173)	0.014 [−0.226, 0.251] (n = 68)	0.110 [−0.063, 0.275] (n = 132)	Not assessed	Not assessed
Certainty about expected change in duration/magnitude	0.253 * [0.046, 0.437] (n = 89)	0.315 *** [0.173, 0.443] (n = 172)	0.055 [−0.187, 0.289] (n = 68)	0.242 ** [0.074, 0.395] (n = 133)	Not assessed	Not assessed
Worry about respective events	0.285 * [0.021, 0.508] (n = 56)	0.440 *** [0.318, 0.545] (n = 196)	0.097 [−0.141, 0.324] (n = 70)	0.398 *** [0.276, 0.506] (n = 207)	0.360 ** [0.116, 0.559] (n = 61)	0.401 *** [0.277, 0.510] (n = 201)

Note: All variables have been normalized to a range of [0, 1], except number of experiences, which has a range of [0, 4]. * p < 0.05, ** p < 0.01, *** p < 0.001.

compiles the investigated correlations. The number of experiences is almost unrelated to B+I and risk perception, except for the correlation of experienced droughts with risk perception (r = 0.142) and the somewhat strange (marginally not statistically significant) negative correlation between experienced new diseases and B+I (r = −0.235). The correlations for how intimidated participants are still about the experienced events are considerably stronger, at least for risk perception. For B+I, a statistically significant correlation is observed only in the case of droughts (r = 0.351). This information was only assessed for those participants who reported having experienced the respective events at least once. The variables that consider the participants who have never experienced the respective event with 0 intimidation have similar (and, thus, mostly low) correlations as the number of experienced events. For diseases, no significant correlations are found.

The next group of variables assesses expectations about future developments with the respective CC consequences. Here, the most interesting tendency is that not the expected amount of change is correlated with B+I or risk perception (with the exception of the correlation of the expected increase in new diseases with risk perception, r = 0.172) but the perceived certainty that these changes will occur. Therefore, it is not how much the frequency or severity of events is expected to change but how certain people are about such changes happening. However, here, relevant correlations are only found for risk perception and, for B+I, only for droughts.

Finally, a general affective assessment of how much the participants worried about droughts, flooding, or new diseases shows rather strong correlations with risk perception (r between 0.359 and 0.440) and with droughts, along with for B+I (r = 0.285). For flooding, no relevant correlation is present between worries and B+I (r = 0.097).

4. Discussion

The present study aimed to identify important determinants of behaviors for adapting to CC consequences in a population not only vulnerable to the future effects of CC but that has also experienced extreme events that will become more frequent and severe because of CC. We presented various regression analyses that can explain the behavioral disposition B+I with structural factors, general evaluations, and behavior-specific evaluations. We started with a general model that combines the data of all classes of adaptation behaviors and consequences of CC and then compared the results for different behavior classes and consequences. Finally, we investigated how the experiences and expectations of CC consequences were related to B+I and risk perceptions.

4.1. General Discussion

This discussion is oriented by the groups of determinants investigated. We first discuss the behavior-specific evaluations because they had the greatest explanatory power (adding 43.7% explained variance in the general model), then the general evaluations, and, finally, the experiences and expectations. Structural factors have little explanatory power (4.1% explained variance in the general model). The participants from Urubamba, Huacarpay and Izcuchaca, present lower levels of B+I than the participants from Cusco, particularly for community projects and policy support for adapting to droughts. Furthermore, there is a weak positive effect of age, particularly for investments and community projects for adapting to flooding and new diseases. Considering that our sample comprised participants from a wide range of groups—from highly urbanized (almost ‘western’) inhabitants of Cusco to rural communities practicing subsistence farming—the minimal effect of structural factors indicates that our results can be generalized to many other populations.

The general model (i.e., with scales aggregating the data over all behavior classes and CC consequences) shows that behavior-specific evaluations—particularly perceived feasibility, descriptive norms, and cost-benefit evaluations—can explain most of the variance of B+I. Notably, the adaptation efficacy of the behaviors is unrelated to adaptation. Investigating each behavior class separately reveals only a few relevant differences between the classes. For changing daily behaviors, perceived adaptation efficacy has a substantial and statistically significant effect. This means that changing one’s daily behavior in the described forms is (also) motivated by reducing the damage from CC consequences, which might also be the reason for the greater effect of the cost-benefit evaluations. In contrast, for all other classes of adaptation behaviors, these considerations appear to not be relevant. The fact that, with very similar distributions for all subsamples, considerable effects could be found for one behavior class but not for the other classes rules out several methodological problems that could explain the lack of effect (e.g., the low internal consistency found for this scale, inadequately constructed items, lack of variance, nonlinear relationships). Therefore, we interpret this unrelatedness of adaptation efficacy to B+I (apart from changing one’s daily behavior) as adaptation behaviors are often being motivated not by reducing the negative impacts of CC consequences but as characteristics of the behaviors that are unrelated to CC.

Only few studies have investigated such behavior-specific evaluations as determinants of adaptation behavior quantitatively and most studies considered only adaptation efficacy. The only study we found that considered norms estimated a small but significant positive effect for this construct [7]. Studies that investigated adaptation efficacy as sole or together with one other behavior-specific evaluation, found strong positive effects. For example, ref. [32] found this for changing individual behaviors to deal with flooding in the Netherlands. This result is in line with our finding, as for this behavior class, our study also found a positive relation. Interestingly, ref. [32] found only a weak effect of self-efficacy, which might be due differences between European countries and a population in the Andean highlands. Another study from the Global South investigating agricultural adaptation in Sri Lanka [33] found only a weak marginally non-significant effect (for p = 0.05) for adaptation efficacy. A conclusion of this literature comparison could be that we need more investigation into behavior-specific evaluations as determinants of adaptation.

Furthermore, for the most abstract threat, that is, new diseases, descriptive norms are more important, whereas for adapting to the more concrete threats, cost-benefit evaluations and perceived feasibility are more important, with the former being more important for changing their daily behaviors. A possible interpretation of this result could be that it is easier to think about the costs and benefits for more concrete threats whereas, in the case of rather abstract threats, people are more guided by the behavior of other people. Injunctive norms affect only adapting to flooding. This could be related to governmental flood protection measures and risk prevention campaigns that have been led by nongovernmental organizations [34], which could make adaptation measures (e.g., building houses further away from rivers) seem more mandatory.

The results specific to each class of adaptation behaviors and each CC consequence might be different in other populations. While we do not have any a priori assumptions pertaining to how the found variance and tendencies should be different in other populations, the evaluations of each behavior class and CC consequence might depend strongly on the existing behavioral options and climatic conditions. For example, droughts are a very concrete threat for the investigated population but might be barely imaginable for others. Community behaviors might not be possible in some populations, while political behaviors might be limited in others. Such aspects need to be considered when the presented results are transferred to other populations and locations.

The general evaluations have revealed rather surprising effects. Trust is one of the most important drivers of adaptation, whereas risk perception—unlike our hypothesis—is unrelated to B+I. Trust is particularly important for policy support and community projects. In the latter case, general trust, which is not relevant for other behavior classes (maybe due to its low internal consistency), is also relevant. This replicates the finding [17] that trust is more important for public than for private behaviors. Furthermore, we find for the abstract threat of new diseases that trust is one of the most important factors, while for the more concrete threats of droughts and flooding, trust is unrelated to B+I. Considering that, for these behaviors, neither adaptation efficacy nor risk perception is related to B+I, trust might instead be required to see the advantages of the specific behaviors than to be more confident about CC having devastating effects.

Interestingly, attitudes toward nature also have relevant and statistically significant effects. Although negative attitudes toward nature have similar small effects for all behavior classes, positive attitudes toward nature have a medium negative effect only for investments. Furthermore, both attitudes toward nature are relevant for explaining adaptation to the abstract threat of new diseases (although the effect of positive attitudes toward nature is not statistically significant) but not for the more concrete threats of droughts and flooding—except for a significant negative effect of positive attitudes toward nature on flooding. Across all the models, a consistent effect can be observed: the more negative and the less positive the attitude toward nature was, the more they adapted to the consequences of CC. It makes sense that one would protect more against a threatening nature than against a kind nature. On a more general level, adaptation to more concrete threats is driven mostly by behavior-specific evaluations, whereas general evaluations—and descriptive norms—become more important as fewer people know, understand, or directly perceive the effects of CC consequences.

Only few studies considered trust as a determinant of adaptation. Besides the already mentioned [17], ref. [33] also found weak effects of trust to explain policy support. We did not find any study considering attitudes towards nature as determinant of adaptation to CC consequences. In contrast, several studies investigated risk perception as a determinant. Many of these studies found rather strong positive effects of risk perception [32,33,35,36], but none of these studies considered trust and behavior-specific evaluations together with risk perception. Also in our study, risk perception has a positive correlation with B+I (r = 0.221, p < 0.001), but considering other constructs for explaining adaptation, this effect disappears. Therefore, we conclude that risk perception can serve as an indicator for a readiness to adapt, but it appears to have no direct effect on adaptation behavior.

Regarding the general evaluations, more research is required. Not only are the presented investigations exploratory and require reconfirmation, but we also expect different results for other populations. For example, in the Global North, trust might be higher and attitudes towards nature might be more positive. The resulting lower variance in these constructs might reduce the effects. Furthermore, risk perception might be related to political opinions and, if the respective parties advocate or reject adaptation to CC, risk perception might be stronger (positively or negatively) correlated with B+I.

The last set of determinants are experiences and expectations. The participants have experienced many extreme events that could become more frequent and severe because of CC. Droughts are most frequently experienced, but on average, flooding and new diseases and pests are experienced almost once in five years. However, the number of experienced events is mostly unrelated to B+I or risk perception (with the exception of the risk perception of droughts). We have found an unexpected negative effect of the number of experienced new diseases on B+I. Although this result is marginally nonsignificant and needs further confirmation, a possible explanation could be that, having experienced certain diseases, the participants might feel safe because they have achieved immunity. The psychological effects of such events (i.e., how intimidated the participants still feel because of experiencing the respective events) show stronger correlations, at least for risk perception. Particularly for droughts, the effects are stronger, and even for B+I, a relevant (although marginally nonsignificant) correlation can be observed. In addition, when it comes to expectations, statistically significant correlations are found only for risk perception (except the change in duration of droughts that correlates with B+I) and only for the certainty that the frequency or severity changes (except the change in frequency of new diseases that is correlated with risk perception). As a general emotional reaction to threatening developments, worries are considerably correlated with B+I (except for flooding) and risk perception.

Some earlier studies investigated the effects of experiences on risk perception and adaptation. For example, ref. [25] shows weak effects of flooding experience on concern, which is close to risk perception in our study, and on adaptation to heat waves. Ref. [37] shows that not only experiences but also damage expectations are related to adaptation to flooding of homeowners. While these effects are small, they are still larger than the effects in the present study. The longitudinal study of [38] could give a clue about reasons for the absence of effects we found. The conclusion drawn from the complex study of [38] was that the first shock of experiencing a heat wave increases policy support for adapting to this threat, but a longer exposure did not lead to more support. Considering that in our sample only 22% of the participants had not experienced any of the three extreme events, it might not make any difference anymore, whether further events were experienced or not.

We found a somewhat strange correlation pattern. The number of events is unrelated to risk perception, but how intimidated the participants still are shows larger correlations. However, when cases with no events experienced are added (with intimidation = 0), these correlations mostly disappear (except for droughts). This could be interpreted in the form of a difference in “sensitivity” to events. People who react more sensitively to extreme events also show higher levels of risk perception. Because we do not know how sensitively people react without having experienced an event, correlations appear only if people have experienced such events. The conclusion could be that no causal relationship exists between experiencing extreme events and risk perception or B+I.

While further investigations are required, we would expect that the findings about no effects of the number of experiences and “cognitive” expectations should hold in other populations as well. We have no a priori assumptions that in other populations psychological factors, particularly emotional evaluations, should be less important than the ‘objective’ experiences and expectations. However, the various CC consequences might be perceived differently by other populations. Particularly recent events or immanent dangers might increase the relation of experiences and expectations with B+I.

4.2. Strengths and Limitations

The present study has shown that an investigation of generalized adaptation behaviors is fruitful. Assessing evaluations for classes of behaviors and then combining the evaluations over all classes and consequences led to sharp profiles of effects. Nevertheless, analyzing each class and consequence separately yielded additional results but with less power. A drawback of this procedure is that evaluations that refer to more specific behaviors, such as affective evaluations or symbolic (e.g., status) functions, cannot be assessed. These constructs might have considerably increased the explanatory power of the model, which, however, turned out to be sufficiently high to draw conclusions.

The data were gathered from a population of particular interest that has been neglected in previous research related to investigating adaptation to CC. However, when generalizing the results, the considerations mentioned in the general discussion need to be observed.

It is important to highlight the explorative character of the current investigation. The systematic test of a large number of variables allowed for the identification of several potential determinants of B+I—and the discarding of several other variables from such a role—that are not often mentioned related to adaptation to CC. However, such exploration meant running numerous tests without significance correction or preregistration. Although we reported only robust results that were found in a variety of models, it cannot be ruled out that some effects became significant—or nonsignificant—by chance. Therefore, it is important to replicate the results, particularly the effects of the general evaluations. Nevertheless, we are convinced that running such explorative investigations is essential to advance research. Only with such a procedure can unexpected effects be found so that we can avoid being stuck in always the same mental models. However, it is important to be cautious with these results until a series of further studies confirms the findings.

Finally, the limitations of the study design need to be mentioned. This was a cross-sectional correlational study based on self-reports mostly assessed with one-item scales. This means that no causal conclusions can be drawn and that, for behavior and experiences, more objective measures might be desirable. In addition, the number of cases for some analyses, particularly of experiences and expectations, was very small. This might make the results less robust, so a replication with larger samples would be desirable.

4.3. Practical Implications

Humans must inevitably address some of the adverse consequences of CC. Therefore, in addition to promoting mitigation behaviors to limit CC, adaptation behaviors need to be promoted. This is particularly important in regions of the Global South. Our study provides several implications for doing so.

The most important lesson is that references to CC might not be that important. The degree to which an adaptation behavior is performed appears to depend more on behavior-specific evaluations. However, only for changing everyday behaviors the evaluation that the behavior is effective in counteracting negative CC consequences is positively related to B+I. Hence, to promote adaptation behaviors, the focus should be on highlighting their positive characteristics beyond mere damage reduction and increasing their feasibility. The descriptive statistics indicate that, in the investigated population, cost-benefit and normative evaluations particularly have the potential to increase, whereas feasibility (and adaptation efficacy) is evaluated rather highly; however, there is still room for improvement. As an exception, changing individual behaviors can be promoted by highlighting how the behavior reduces the negative impacts of CC, but not through injunctive norms, which, for this behavior class, are unrelated to B+I.

Risk perception often focuses on changing CC-related behaviors [39,40]. Although risk perception should be particularly important in the case of adaptation behaviors, our results indicate that, in contrast, risk perception is unrelated to adapting to CC consequences. However, trust is an important factor for promoting adaptation behaviors, particularly for public or “political” behaviors (such as community projects and policy support) and for adapting to more abstract threats (such as new diseases). More generally, for more abstract and, thus, less understood or perceivable threats, descriptive norms and general evaluations are more important. Adapting to more concrete threats depends more on cost-benefit evaluations and feasibility.

A final result from the regression analyses is that people adapt more to CC consequences if they perceive nature as threatening. Although inducing a more negative perception of nature should not be too difficult, doing so might have several negative consequences. In particular, these measures may reduce the behaviors that protect the environment. Therefore, we do not recommend implementing such campaigns.

Finally, campaigns often focus on increasing risk perception by providing the “experiences” of the possible consequences of CC in some form [41,42]. According to our results, neither are experiences well suited for increasing risk perception (and much less B+I), nor is risk perception alone an important influence on B+I. Furthermore, in our sample, risk perception was already quite high, and there would not have been much room to increase this evaluation. Nevertheless, psychological reactions to experiences and the certainty about expected changes because of CC were related to risk perception. The fact that the expected changes are not related to B+I or risk perception would indicate that—at least for changing risk perception—it is more critical to highlight the certainty of changes than how much worse the situation might become. Using the experiences/expectations route, the most effective way to increase risk perception and B+I would be to induce worries about future events in the target population. However, this approach is ethically problematic because negative emotions are induced that might not all be removed by performing the respective behavior. Therefore, we discourage the use of this form of intervention.

5. Conclusions

The present study challenges many of the common ideas of how to promote adaptation behaviors. In particular, the role of experiences and risk perception appears to be less important than what is commonly assumed. Promotion campaigns need to consider the specific characteristics of the behaviors, and performing these behaviors needs to be attractive beyond their function to protect against adverse CC consequences. Furthermore, trust might play a relevant role in adapting to CC consequences, particularly for more abstract threats, such as adapting to new diseases, and public behaviors, such as supporting community projects and policies. Finally, perceiving nature as more threatening also increases adaptation. However, the effects of these general evaluations require further investigation.