Committing to behavior change: How to prevent dogs from visiting Atlantic Forest remnants?

Behavioral interventions have been applied to induce multiple pro‐environmental changes, yet testing their effects on biodiversity conservation is novel. For certain behaviors, a promising tool is changing the context within which people make decisions through nudges. One such behavior is dog restraining to prevent visits to natural areas, a worldwide source of negative impacts on wildlife. In this preliminary study, we experimentally tested the effects of two interventions (i.e., an informative video on conservation and an informative video plus commitment nudge) on changing dog management among rural owners. We then assessed interventions' effects on owners' behavior change (n = 52) and intention to keep the restraining practice in the future (n = 50). Both interventions promoted a momentary change in owners' behavior, but their effect sizes did not differ. Yet, neither treatment increased the owners' intention to maintain the restraining practices. Although nudges alone may not be the most appropriate solution to reduce the severe dog impacts on wildlife, they might be of practical relevance to conservation stakeholders who deal with threats that are not prone to mitigation through more traditional strategies (e.g., command and control).


| INTRODUCTION
As human activities account for the environmental and biodiversity crisis that led us into the Anthropocene (Malhi, 2017), authors have urged for studies on behavior change to guide more effective conservation strategies (e.g., Nielsen et al., 2021;Selinske et al., 2020).Although nongovernmental or other organizations (e.g., zoos and aquariums) have developed and implemented behavior-based conservation programs for some time (e.g., Maynard et al., 2020), studies that experimentally test the effects of interventions to promote behavioral changes for biodiversity conservation are just emerging (e.g., Byerly et al., 2019;Niemiec et al., 2021).Nevertheless, evidence from other fields shows that behavioral interventions may help induce environmentally benign behaviors, such as reducing residential water use (Ferraro & Price, 2013).Hence, behavioral interventions may contribute to engaging people in biodiversity conservation, particularly concerning changes in everyday, widespread behaviors such as managing pets that can freely roam and damage ecosystems (e.g., Linklater et al., 2019).
Dogs are an introduced species known to negatively affect wildlife worldwide and, thus, an important conservation concern (Gompper, 2014).In Brazil, studies have demonstrated that dogs interact with wildlife in lethal and sublethal ways, besides being part of zoonotic cycles.For instance, they have been identified as either predators or competitors of over 20 mammal species such as rodents, primates, and carnivores (Campos et al., 2007;Guedes et al., 2021;Oliveira et al., 2008).Furthermore, the interactions between dogs and wildlife have consequences for animal and human health, as studies show the ability of pathogens to cross species barriers.For example, in Atlantic Forest remnants, dogs that have access to protected areas are more likely to have detectable antibodies against the Rickettsia sp.bacteria responsible for the Brazilian Spotted Fever (Campos et al., 2017), a disease that killed 417 people in 5 years (2015-2020) (DATASUS, 2022).
Because of their close association with people, freeranging dogs rely on human subsidies such as food and veterinary care, which favor high population densities.In fact, in rural areas, as in large parts of the Atlantic Forest, free-ranging dogs are the most abundant carnivore (e.g., Guedes et al., 2021;Ribeiro et al., 2019), impacting forest remnants far from their homes.Approaches to mitigate the impacts of free-ranging dogs on wildlife have traditionally relied on removal, neutering, and, occasionally, euthanasia.Besides ethical concerns, studies investigating the effectiveness of these methods are inconclusive (Smith et al., 2019).Because such practices do not limit dogs' mobility (Garde et al., 2016) nor the introduction of new individuals (Villatoro et al., 2016), they may fail to control dog populations and invasions effectively.A few studies investigated the avoidance of dog-wildlife interactions through dog training (e.g., Dale et al., 2013).However, they did not include a human-behavior component to the interventions, such as adopting restraining practices to prevent dog visits to natural areas.
Therefore, understanding which strategies can effectively change owner's management practices to reduce dogs' negative impacts is warranted.Broadly, there are two ways of promoting human behavioral changes.The first assumes people are rational actors and they do not make systematic prediction errors when deciding.Economic incentives and educational or information-based approaches are examples of such cognitive strategies (Dolan et al., 2012).For instance, to manage invasive species, researchers have traditionally relied on knowledge-transfer interventions, assuming informed people will develop supportive attitudes and behaviors toward conservation (McLeod et al., 2015).However, humans respond not only to incentives and information but also to more subtle contextual cues.Therefore, the second way to promote behavior change focuses on altering the context in which people make decisions (i.e., choice architecture) in a simple, nonmandatory, and non-prohibitive way.Entitled nudges, these interventions rely on reducing decisions' cognitive or material costs (Thaler & Sunstein, 2008); for instance, commitment nudges.Commitments rely on the principle that individuals tend to act consistently with their promises to avoid unbalances between their opinions, beliefs, and behaviors (i.e., cognitive dissonance; Festinger, 1957).Commitments usually work better when made actively and publicly because they strengthen the need to act aligned with both our self-and perceived image (Cialdini, 2007).Such a strategy has been used in environmental psychology in fields like recycling, energy saving, and transportation alternatives (see the review by Lokhorst et al., 2013).For example, commitments have succeeded in encouraging towel reuse in hotels (e.g., Baca-Motes et al., 2013) more effectively than informational approaches.However, they are still rare in biodiversity conservation.
To the best of our knowledge, only two studies evaluated the effects of behavior-based interventions to prevent interactions between dogs and wildlife.In Australia, David et al. (2019) developed a social marketing campaign to engage owners in a 4-week dog training program to reduce koalas-dogs interactions that significantly increased wildlife aversion-related behaviors in dogs.Walsh (2021) had mixed results testing the effects of four messages (i.e., kiwi conservation, dog attacks to people, social comparison nudge, and dog loyalty) on dog registration among rural and urban owners in New Zealand.Although registration fees fund dog control activities, which impact kiwi conservation, they do not directly affect dogs' population dynamics and behavior.
In the Brazilian Atlantic Forest, observational evidence suggests that two owners' behaviors predict dog visits to forest remnants: (i) maintaining dogs less time in confinement and (ii) encouraging dogs to roam (e.g., to accompany owners during forest incursions) (Biffi, 2017, unpublished work).Therefore, we departed from these behaviors and evaluated a nudge-based approach to reduce dog visits to the forest.Our preliminary study tested and compared the effects of an informational intervention and the additional effect of commitment on owners' behavior.We hypothesized that a symbolic and public commitmentattaching a sticker to the property (see Figure 2; e.g., at the front gate)-would increase owners' likelihood of performing management practices to restrain dog movements and the intention to keep them in the long term, compared to the informational approach alone.

| Study area
We conducted our study in a rural region of the Southeastern Brazilian Atlantic Forest, specifically in the Joan opolis municipality (Figure 1).Joan opolis' territory area is 374.293km 2 with an estimated population of 13,453 inhabitants (IBGE, 2023).In the region, top predators such as the puma (Puma concolor) and the jaguar (Panthera onca) are rare or absent, respectively.Furthermore, forests are highly impacted by dogs, which are the most abundant carnivore in forest remnants (Ribeiro et al., 2019).Past research also highlights 53% of the dogs are uncontained at all times, and 41% leave their houses alone because most dwellings lack fencing (Biffi, 2017, unpublished work).

| Study design
We designed a field experiment to examine how owners' commitment to adopting dog-restraining practices influenced their subsequent behavior and intention to act consistently in the future.
We chose commitments based on a qualitative pretest performed with 16 dog owners (11 men and 5 women) in six rural neighborhoods in Joan opolis.In the pretest, we opted for a convenience sample (Trochim & Donnelly, 2016), searching through roads for households with dogs, in which owners were present.We used semi-structured interviews, with a partially defined script (Albuquerque et al., 2010), to check our study's feasibility and possible interventions.For example, the pretest showed that attaching the sticker to the property could perform better (n = 11) than broadcasting reminders on the local radio (most did not listen to the radio frequently; n = 10).Also, although only two owners did not have cell phones, the reception in rural Joan opolis is very poor, which would hinder calling or sending texts to remind owners of the management practices.In the same pretest, we also observed that most owners thought it was correct to allow dog visits to forest remnants (n = 10).Therefore, we believe it would not be possible to address the conservation issue without presenting them reasons why they should engage in management practices, which we did in our informational condition (i.e., showing the harms of not containing their dogs).
For the field experiment, we chose a posttest-only control group design (Campbell & Stanley, 1963).This design does not take measurements before the intervention but requires at least one control group and one experimental group to be compared against one another.This choice is preferred when the results of the posttest could be affected by the prior condition, which we believe was our case (either because dog owners would be more prepared to behave according to our expectations or because they would not comply at all).
We combined the experiment with a survey through in-person interviews.The survey allows direct comparison of answers through statistical analyses in the search for patterns and relations between variables (Newing, 2011).Before the delivery of interventions, we used structured interviews to collect data on the control (i.e., sex, age, school level, and attitudes toward dog management) and descriptive variables (i.e., number of dogs raised, neutered, treated for ticks, reasons to own dogs, working outside, and income).At the end of the 30-day interval, we also interviewed the owners to measure intention and collect data on other variables (i.e., reasons to adopt or not management behaviors, and reasons not to fill in the diaries).

| Procedures
We sampled households in the three enumeration areas with the largest forest cover in Joan opolis (Figure 1b).Through satellite imaging, we first numbered each identified building within an enumeration area (598 buildings in total).Then, we randomized a sequence of those numbers by area and visited the corresponding geographic coordinate to check whether: (i) it was a permanent household, (ii) dogs were invariably kept on the property and provided care for (i.e., we asked owners not to include dogs that visited sporadically-e.g., to play), and The commitment sticker reads in Portuguese "Dogs from this house do not wander in the woods."Note that the commitment treatment included the informative video.
(iii) at least one dog was raised uncontained at all times.We attributed the experimental conditions following the randomization order in each enumeration area.Therefore, the first building that met our criteria was assigned to the control condition, the second to the informationonly condition, the third to the commitment condition, and so forth.
In total, we visited 264 locations, but only 79 were eligible.We searched for the dogs' owners at least three times, at different hours and dates.If the owners were not available, we would move to next randomized building.Once we found the owners at their homes, we explained the scope of our study and asked for their informed consent.However, 20.25% did not agree to participate.For those who agreed to take part in our study, we would proceed to the interview, followed by the delivery of one of the experimental conditions.In the control condition, owners watched a video with general information about dogs (n = 16).In the information-only condition, owners watched a conservation video highlighting dogs' impacts on wildlife (n = 20).In the commitment condition, owners watched the same video as the second condition and were presented with the option to commit to dog management by attaching a sticker to their building, door, or fence signaling their pledge (n = 21) (Figure 2).We asked owners to attach the sticker in the presence of the interviewer.The videos were played on a tablet and were 05 0 10 00 (control) and 05 0 16 00 (informative/ commitment) long.Both are narrated by Tagarela (Chatty, in English), a fictitious dog that lives on a farm near the forest.They differ in content, but we chose a similar duration, related scenes, and the same teller to avoid biases associated with their delivery.The scripts translated into English are available in Appendix S1.
Our final sample was reduced to 52 owners because 7.6% could not be reached (in person, by telephone, or text message) for the experiment follow-up, and 5 out of the 21 participants in the commitment condition did not agree to commit when asked.The limitations of our reduced sample size are presented in Section 4.
All owners received a 30-day diary to fill in if they performed any recommended management practices as follows.While the owner went on forest incursions or its surroundings, we suggested they (1) ask another household member to hold the dogs to avoid the immediate following; (2) keep the dogs in the kennel, if available, (3) restrained on a yard leash at home, or (4) inside the house.We also suggested ( 5) training the dogs and (6) building or repairing fences, if possible.The houses in our rural context usually have large land plots, which give dogs plenty of space to move.However, most frequently, fences are either absent or inadequate for dog containment.The suggested behaviors were depicted in a pre-section of the diary.Each of them was associated with a colored sticker.Thus, if the owners performed any of the behaviors, they were asked to attach the corresponding sticker to the day it occurred (Figure S2).Direct observation was not a feasible measurement of behavior because of the remote context of our study area.For example, households could be as far as 20 kilometers from each other by dirty road.After 30 days from the first visit, we returned to collect the diaries and measure the owners' intentions to perform those behaviors for the following month.Note that despite all 52 dog owners having handed in their diaries, we could not interview two of them on our second visit, leaving the control and the information groups with 15 and 19 participants, respectively.

| Data analysis
We adopted a treatment-on-the-treated design.That means we analyzed only owners who received the treatment, removing those who did not agree to commit (but see an alternative specification using an intention-to-treat [ITT] design in Table S1).Although we offered diaries for daily behavior registration, we did not ask owners to record the frequency with which they went on forest incursions.Therefore, we could not estimate an accurate adherence rate.However, in the second phase, we asked owners whether they had engaged in the suggested behaviors.Thus, we used this binary variable to measure behavior.That is, whether owners performed at least one of the management practices.Nevertheless, we provide a comparison of the number of days in which owners performed management behaviors across groups in Figure S1.
For the response variable measuring intention, we developed a 7-point Likert scale.Likert scales are used to evaluate psychological phenomena from answers ranging in a given category (e.g., totally agree/totally disagree) (Likert, 1932).Thus, we considered the mean score of three items with answers ranging from 1 (totally disagree) to 7 (totally agree) (i.e., I have plans/I will try/I will make an effort to perform one or more suggested behaviors to prevent my dogs from gaining access to forest remnants in the next 30 days).
In our analyses, we also included four control variables because previous research showed they correlate with pro-environmental behaviors and attitudes (e.g., Casal o & Escario, 2018;Meyer, 2015;Zelezny et al., 2000): sex (male = 0/female = 1), age (in years), schooling (in years), and attitudes toward the proposed management behaviors.Attitudes correspond to a positive or negative evaluation of a given psychological object, including a behavior (Fishbein & Ajzen, 2011).
This variable was included because it is a determinant of human behavior, which could be associated with dog management, as studies show that more positive attitudes relate to higher chances of performing responsible ownership practices (e.g., Rohlf et al., 2015).Attitudes were estimated through a semantic differential scale, often adopted to measure human behavioral dimensions, departing from a set of items in which answers constitute opposite adjectives (Osgood et al., 1978) (e.g., Doing one or more suggested behaviors to prevent my dogs from gaining access to forest remnants in the next 30 days is 1 [very harmful] to 7 [very beneficial]).For variables measured in scales, we confirmed the one-dimensionality of the constructs with a Principal Component Analysis (see Table S2).We also adopted Cronbach's alpha to check internal consistency between items (i.e., whether items measured a single construct).The minimum scale reliability was set to α = 0.70 (Table 1).
We adopted a binary logistic regression and a multiple linear regression to test the hypotheses that commitment would increase owners' chances of performing management practices and the intention to keep them, respectively.
Due to the COVID-19 pandemic, we had to call off our field expeditions.Thus, our sample size reflects what was practically attainable.Therefore, we performed a sensitivity power analysis to estimate what effect size our study was able to detect given a power of 80.00%, n = 52, and α = 0.05 (see more in Table S3).Power is the probability that the null hypothesis will be rejected when it is actually false and, therefore, find real differences across groups (i.e., control and treatment).We obtained an effect size of OR = 15.97, which means the study would not be able to reliably detect an odds ratio smaller than this.Although our results show a higher odds ratio, we advise caution when interpreting them (see Section 4).

| RESULTS
Owners had an average of 2.67 dogs (SD = 1.38), but this number ranged from one to seven dogs per owner, which is high, considering the total of dogs (n = 139) exceeded that of residents in the sampled households (n = 135).The main reasons to own dogs were affection (96.50%), companionship to themselves or family members (93.00%), and property safety (80.70%)(Figure S3a).Regarding responsible ownership practices, 31.84% of the dogs were sterilized, and 29.61% received tick treatments up to 3 months before the interviews.After our interventions, 69.23% (n = 36) of the owners informed us they had performed at least one of the suggested management behaviors.The proportion of owners who engaged in them was 50.00% in the control group, 80.00% in the informational group, and 75.00% in the commitment group (Table 1; Figure S3b).Running the binary logistic regression (n = 52) without control variables indicates that none of the treatments had significant effects (all >0.05;Table 2).Nevertheless, once we controlled for confounders, watching the informative video alone yielded a significant increase in compliance (Wald(1) = 6.21, p < .05)compared to the control condition.The interviewees' commitment also increased their odds of performing dog restraining practices (Wald(1) = 6.90, p < .01).However, attaching the public commitment sticker did not statistically differ from the informational intervention, OR = 1.93 (95% CI [0.17, 22.34], Wald(1) = 0.28, p = .599).Schooling and attitudes were significant predictors of performing management practices (Table 2; also see Table S1 for the ITT alternative specification).
When asked about their motivation to adopt the management practices, owners most reported caring for their dogs' health (62.00%), wildlife health (56.00%), and complying with our study (46.00%) (Figure S3c).Eight owners who performed the suggested behaviors did not complete the diaries because they did not understand how to (62.50%), did not find it important, did not have time (both 25.00%), or found it laborious (12.50%).Among those owners who filled in the diaries (n = 28), management practices were held for an average of 17.96 days (SD = 11.99;min.= 1, max.= 30).The most frequently adopted practices by them were restraining dogs on a leash (40.40%), followed by keeping them in a kennel (32.70%), asking another household member to hold the dogs, and dog training (both 28.80%) (Figure S3d).The most frequent reasons not to perform any management practice was lack of time (56.25%),followed by feeling pity to restrict their dogs' movements, and unwillingness to collaborate (both 31.25%)(Figure S3e).
Unlike the behavioral change results, the multiple linear regression outcomes indicate none of our interventions effectively predicted changes in owners' intentions to restrain their dogs for 30 more days after the study interview (F(2, 43) = 1.60, p = .214)(Figure S3f).More years of T A B L E 2 Results of multiple logistic regression analysis for the treatment-on-the-treated design on the adoption of management behaviors to avoid the presence of dogs in forest areas (n = 52).

Contrast of margins [OR]
Std. err.Z p> jZj 95% CI Note: For the treatment variables, we report both results from pairwise comparisons of marginal linear predictions, requesting Šid ak's adjustment for unbalanced arms, and standard logistic regression in square brackets.The correct classification rates are estimated to be 87.50% in the control group, 100.00% in the information group, and 75.00% in the commitment group.Abbreviations: CI, confidence interval; OR, odds ratio; Std.Err., standard error.
schooling and positive attitudes toward avoiding dog visits were the only significant predictors of higher intention to adopt management behaviors in the future (Table 3).
Behavior change is key to improving our response to the current biodiversity crisis, although our knowledge of its effectiveness in conservation science is little (but see Thomas-Walters et al., 2022).By testing the effects of a behavior change intervention to prevent dogs' impacts on wildlife, our study, therefore, contributes three main findings to this emerging research topic.First, our results suggest both the added commitment and informational interventions increase the adoption of restraining practices to avoid dog visits to forest remnants.Current knowledge of the effectiveness of behavior change interventions on dog management is scarce, but studies have shown similar results.Even with a small sample size, David et al. (2019) found their training program significantly increased wildlife aversion-related behaviors in dogs and was a positive experience for owners.The elevated cost (AUD$150/owner) of such training, however, severely restricts its viability in poor contexts such as our study region.Walsh's study ( 2021) is more similar to ours but tested a different nudge (i.e., informing the percentage of registered dogs), which was not effective.Other messages tested showed mixed results with only the conservation and dog attacks ones yielding increases in registration but in distinct owner groups.These studies and our results suggest behavioral change interventions are promising tools to engage dog owners in biodiversity conservation.Yet, strategies may work differently depending on the targeted behavior and social groups' characteristics.
Second, we cannot state that our commitment intervention performed better or worse than watching the video alone.Nevertheless, previous evidence seems to suggest commitments work differently across environmental behaviors, sometimes effective and sometimes not.Lokhorst et al. (2010) found providing information and asking farmers to pledge did not increase conservation quality and habitat diversity in Dutch farms.In contrast, in the same study, for other behaviors (e.g., time spent in conservation), a commitment was superior to information only.In a hotel, commitment proved to be more effective than other conditions (e.g., reading an informative message) to promote towel reuse (Baca-Motes et al., 2013).
Even though we cannot infer the effectiveness of commitments in our study, we believe two issues could be addressed in further investigations to yield more compliance.First, during our pilot study, we identified among owners the fear of being reported to animal welfare agencies if they restrained dogs.Therefore, owners might be responding to social pressure when not complying with dog restraining practices, as social norms (group values and expectations) are powerful behavioral determinants (Cialdini et al., 1991).We have tried (but failed) to partner with the local NGO that provides dog shelter and care in Joan opolis to overcome this issue.Nevertheless, we informed the owners that the local Environmental Secretary endorsed our project.Future studies should aim for solutions to build trust between stakeholders.Second, the commitment sticker might not be a salient symbol of commitment because it is neither vivid enough, nor repeated in time.Salience corresponds to how we draw T A B L E 3 Results of multiple linear regression analysis for the treatment-on-the-treated design on the intention to perform management behaviors for 30 days after the study (n = 50).our attention to certain aspects of the environment, which can affect subsequent decisions (e.g., flashlights, reminders) (Taylor & Thompson, 1982).We did not specifically test salience effects because of our qualitative phase results; however, perhaps a more visually prominent commitment or other forms of repeated reminders (e.g., frequently passing by owners' houses) would have promoted higher adoption of dog management practices.Third, we cannot infer the interviewees' intention to perform dog management practices in the future.However, some factors may hinder owners' intentions and deserve consideration in forthcoming studies.For example, in our case, owners may have experienced concrete (e.g., lack of time) or psychological (e.g., pity for containing dogs) difficulties while restraining dogs during the 30-day experiment.Furthermore, the long-term effects of behavioral change interventions, such as nudges, apparently depend on their ability to promote habit formation (Frey & Rogers, 2014).For dog owners, an alternative intervention could be asking them to commit to performing management behaviors at a repeated schedule for a longer period to make this a routine habit.
Lastly, it is paramount to highlight that our study has limitations.Although we were able to detect effect sizes, we advise caution when interpreting our results.Given our small sample size, readers should be mindful of the extent to which those effects are likely to be replicated in similar future studies.In addition, despite sample randomization, we could not avoid owners' voluntary adherence, which means self-selection bias affected our study.Therefore, the distribution of the characteristics of our final sample may not describe those of the population (see Table S4 for baseline characteristics of all participants).To mitigate this issue, Keeble et al. (2015) suggest including variables that could be associated with the selection in the analysis.Sex is an observable variable that could have influenced sampling because we expected more women would be home during our attempts to get interviews.However, at least with our sample size, sex was not a significant variable in our analyses.Thus, we call for caution when interpreting our results since they are still preliminary and demand further investigations with larger samples-a possibility hindered during the COVID-19 pandemic.
Furthermore, our measure of dog management is self-reported behavior, which might contain biases to actual behavior due to (Kormos & Gifford, 2014): (i) people's tendency to report socially desirable behaviors; (ii) the subjective nature of the measures employed to assess behavior and its determinants (i.e., the survey instrument and diaries); and (iii) failure to recall past behaviors.To partly mitigate self-reporting bias, we adopted paper diaries to help owners keep track of their behaviors.Although direct observation is more accurate, self-monitoring can contribute to pro-environmental behavioral changes (e.g., Lange et al., 2023;Marrocoli et al., 2018).Future studies should aim for improved ways to measure owners' behavior directly or indirectly (e.g., photo diaries).The advantages and disadvantages of each measurement tool should be considered, especially in remote and poor contexts like ours.For instance, photo diaries might not be feasible if owners do not have smartphones or internet access.
Given the current turning of attention to promoting human behavior change for biodiversity conservation, our study contributes to the literature on choice architecture interventions in this realm.Dog management is complex because it entangles human and animal welfare.Our study suggests that simple behavioral interventions may help to reduce the severe dog impacts on wildlife.They might be of practical relevance to conservation stakeholders who deal with threats that are not prone to mitigation through financial incentives/disincentives or command-and-control policies, inter alia, because of equity, legitimacy, or long-term maintenance; for example, in the poor tropics, where areas coincide with extensive forest remnants.Therefore, designing cost-effective strategies to mitigate dog impacts is essential to biodiversity conservation in such places.Furthermore, future research is needed to properly understand how behavior shifts toward conservation could last longer or become permanent.

F
I G U R E 1 Study area.(a) Brazil and the state of São Paulo.(b) Joan opolis municipality, highlighting the three enumeration areas with the largest forest cover and sampled locations.Note that the image size might limit the households' identification since some were too close (e.g., nextdoor neighbors).
Isabel Tostes Ribeiro, Carla Morsello, and Renata Pardini designed the study.Isabel Tostes Ribeiro collected the field data and analyzed it.Isabel Tostes Ribeiro drafted the paper under Carla Morsello's supervision, and Carla Morsello and Renata Pardini revised it thoroughly.ACKNOWLEDGMENTS This study was approved by the School of Arts, Science and Humanities of the University of São Paulo (EACH/ USP) Ethics committee (Permit 3.566.533)and supported by a studentship to I.T.R. (Finance Code 001) and funding to I.T.R and C.M. from the Coordination of Superior Level Staff Improvement (CAPES/PROEX); to R.P. from the National Council for Scientific and Technological Development (Conselho Nacional de Desenvolvimento Científico e Tecnol ogico Research Fellowship 311051/2018-9), and the São Paulo Research Foundation (Fundação de Amparo à Pesquisa do Estado de São Paulo 2013/23457-6).We thank Dr. Patricia Carignano Torres for her contribution to the elaboration of the study area map (Figure 1).
Descriptive statistics table of the variables included in our multiple regression models.
T A B L E 1