Study Area

This study was conducted across 10,000 km² in Amapá State, eastern Brazilian Amazon (Figure 1). Amapá has an extensive network of protected areas, including indigenous lands, strictly protected and sustainable-use reserves (International Union for Conservation of Nature [IUCN] & UN Environment Programme World Conservation Monitoring Centre [UNEP-WCMC], 2019). These protected areas account for over 105,000 km² (IUCN & UNEP-WCMC, 2019), which represents ca. 74% of the 142,000 km² total area of Amapá state (Instituto Brasileiro de Geografia e Estatistica [IBGE], 2019a). Amapá has by far the lowest deforestation rates compared with all other states of the Legal Brazilian Amazon. From 1988 to 2018, Amapá had an accumulated deforestation rate of 1559 km², which is 99% lower than the accumulated deforestation for the same period of its neighbour Pará State, the leader in deforestation rates among all Legal Brazilian Amazon states (Instituto Nacional de Pesquisas Espaciais, 2019).

Figure 1.

Study Area. (A) State of Amapá in Brazil. (B) Location of the study area within Amapá. (C) Location of the 106 interviews (circles) conducted with local wage-earners between March to December 2011, and May to August 2015. Dark and light grey represent Uatumã-Trombetas moist forests and Guianan savanna ecoregions, respectively.

This new deforestation frontier in the Brazilian Amazon is facing several threats including mercury contamination from mining activities (Akagi et al., 1995; Fostier et al., 2000; Guimaraes et al., 1999; Hacon et al., 2020), construction of new hydroelectric dams (Fearnside, 1995; Norris et al., 2018), and increased agriculture production with consequent land cover change (IBGE, 2019a). The Brazilian savanna ecoregion in the south and east of the Amazon rainforest is in rapid decline due to agriculture expansion (Brannstrom et al., 2008), coupled with an increased demand for crop (cereal, beans and oil seeds) in eastern Amazonia, with Amapá state producing 56,542 tons per year by December 2017 (IBGE, 2019b). These crops are mainly produced in savanna habitat as almost all forest habitat in Amapá state is covered by some type of protected area (IUCN & UNEP-WCMC, 2019). Additionally, population growth in Amapá has been increasing rapidly, with a population of ca. 500,000 people in 2000 projected to double to over one million people by 2030 (IBGE,2019a). Thus, as per many regions across Amazonia, interactions between local people and wildlife species has been recorded in and around protected areas (Michalski et al., 2012), and the consumption of protein from wildlife by local landowners has also been recorded in this region (Norris & Michalski, 2013).

Study Design

Locations of interviews across the 10,000 km² study area were initially selected using Google Earth (GE) to obtain approximate coordinates, location of villages and local communities supported by key landmarks such as rivers, roads, and other visual features that could be clearly distinguished by GE images. As a prerequisite, all candidate sampling sites that had been previously identified using GE images were associated with one local informant selected at random, usually a long-term resident or landowner, who was (1) willing to be interviewed, (2) was thoroughly familiar with the history of the area, (3) had knowledge of the local wild fauna, and (4) had been living in the property or close to it for at least one year. Ethical approval to conduct interviews with humans was obtained from the Ethics Committee in Research from the Federal University of Amapá (CAAE 42064815.5.0000.0003, Permit number 1.013.843).

Data Collection on People’s Attitudes

From March to December 2011 and May to August 2015 we used pre-elaborated semi-structured questionnaires to interview local residents within the study area (Figure 1). Interviews conducted in the two ecoregions were distributed between the two sampling periods, and were carried out on a one to one basis with researchers asking questions and taking notes of the responses. We recorded socioeconomic characteristics of the respondents, such as gender, age, level of education, number of years living in the property, total number of people living in the property, and total monthly income (i.e., predictor variables). To understand attitudes towards wildlife species we focused on replies to three questions (i.e., response variables): (1) list five wild animal species they like, with species ranked according to the order in which they were mentioned (e.g., first mentioned species was the most liked); (2) list five wild animal species that they dislike, with species ranked according to the order in which they were mentioned (e.g., first mentioned was the most disliked); and (3) which five wild species damage their family income, with species ranked according to the order in which they were mentioned (e.g., first mentioned species was the most damaging income). Respondents were also questioned as to the reason why they liked, disliked or perceived the species as damaging their income. Our sampling focused on wage-earners, to understand attitudes towards wildlife causing damage to income, we therefore interviewed only the person responsible for the major income of the household, who was male in almost all cases (n = 101, 95.3% of all respondents, 96.9% and 92.7% forest and savanna ecoregions respectively). As differences in gender can cause bias related with attitudes (Allendorf & Allendorf, 2013; Miller & Jones, 2006) we do not attempt to evaluate attitudes in relation to different genders in our study. As there was no difference in respondent gender between ecoregions no systematic bias was introduced.

We obtained the ecoregion type where each interview was conducted (Figure 1) by overlapping the interview coordinates on a map of the Earth’s terrestrial ecoregions and biomes (Dinerstein et al., 2017) using ArcGIS 10.1 (Environmental Systems Research Institute, 2011).

Data Analyses

All analyses were performed in R (R Development Core Team, 2019). We performed descriptive analysis examining households and interviewees characteristics comparing their age, number of years living in the property, and the number of people living in the property between forest and savanna ecoregions using Kruskal-Wallis test. For the categorical variables [level of education (with four levels: non-literate, elementary, secondary school, or graduated) and the total monthly income reported for the family (with four levels: <1.0, 1.0–2.0, 2.1–3.0, and >3.0 Brazilian minimum wages) we tested the null hypothesis that proportions of responses were equal between ecoregions using the “prop.test” R function. The common names reported by wage-earners were summarized into 11 faunal groups (amphibians, ants, bats, birds, fishes, medium-bodied mammals, large-bodied mammals, primates, snakes, tortoises and turtles, and other invertebrates). These groupings were established based on documented preferences and attitudes (Alves et al., 2012; Batt, 2009; Dohm et al., 2011; Massad et al., 2013; Peres & Palacios, 2007) and to include species that are likely to require similar conservation management actions. When respondents unambiguously identified the species or congeners they liked, disliked or that damaged their income we obtained the proportion of interviews reporting the same species/congeners. We used Kruskal-Wallis test to determine significant differences between wage-earners interviewed from the two different ecoregions and total number of wildlife species reported as liked, disliked and that cause damage to their income. For all tests, a p value lower or equal to 0.05 was considered statistically significant.

To test for differences in the total number of wildlife species reported as liked, disliked and that cause damage and socioeconomic and ecoregion variables we used generalized Linear Models (GLMs, error distribution family = poisson). To assess the total number of wildlife species liked, disliked and that cause damage to income we examined the effects of (i) socioeconomic variables (e.g., age, time since living in the property, number of people living in the property, level of education, and total monthly income), and (ii) ecoregion where the interview was conducted. We controlled for high levels of inter-dependence between socioeconomic variables by performing a Spearman correlation matrix. This preliminary analysis showed that there were no strong correlations (Spearman r < 0.70) between the socioeconomic variables, with pair-wise correlation values ranging between 0.08 and 0.35, we therefore retained all variables in subsequent analyses. The influence of these predictors on the response variables was tested with separate GLMs to understand how these predictors could affect interviewee’s responses. We adopted a stepwise selection (R function step) applying the program defaults to reach the most parsimonious, “best” model. We compared variable slope estimates in both the full and the “best” models selected, which enables us to avoid well known caveats of stepwise approaches, such as inflated Type I error rates (Mundry & Nunn, 2009).

A PERMANOVA (Permutational Multivariate Analysis of Variance) (McArdle & Anderson, 2001) was used to test the hypothesis that species composition differed based on the attitudes (liked, disliked, and damaging income), ecoregion (forest vs savanna) and socio-economic factors (number of years living in the property, education level and monthly income) and their interaction. The similarities in species composition of respondents attitudes was visualized using Non-Metric Multidimensional Scaling (NMDS). Both NMDS and PERMANOVA analysis were performed in the vegan package (Oksanen et al., 2019) using the function metaMDS and anosim2, with dissimilarities calculated from Jaccard (Jaccard, 1901) distance matrix. An indicator species analysis (Cáceres & Legendre, 2009) was then used to identify the species most strongly associated with different attitudes (like, dislike, and damage income).

To understand motives for the differences in the species mentioned we focused on the first (most important) species reported for each question in each interview. When respondents unambiguously identified the species or congeners they liked, disliked or that damaged their income. Thus we obtained the proportion of interviews reporting the same species/congeners, coupled with the reason for such a positive or negative attitude only including interviews that were unambiguous in their answers.

Interviewee’s and Households’ Characteristics

Overall respondents lived at the property for an average of 21.9 years (±SD = 16.2, range = 1.5–65), and were aged between 26 and 81 years (average ± SD = 53.3 ± 13.0). Their level of education ranged from non-literate (n = 23, 21.7%) to graduated (n = 8, 7.5%), with half of the interviewees (n = 54, 50.9%) reporting having only elementary school level. Total monthly household income was between one and two Brazilian minimum wages for almost half of the respondents (n = 51, 48.1%), and the number of people living in the property ranged from 1 to 12 people (average ± SD = 4.3 ± 2.5) (Table 1).

Table 1.

Socioeconomic Profile and Overall Attitudes on Wildlife Species of Wage-Earners in Different Ecoregions/Biomes Interviewed in Amapá State, Eastern Brazilian Amazon.

Most of the socioeconomic characteristics of the interviewees were similar between the two ecoregions (Table 1). However, there was a statistically insignificant tendency of an increased number of non-literate and fewer graduated interviewees from the Uatumã-Trombetas moist forests ecoregion (Table 1). Some significant differences in the total monthly household income were found between the two ecoregions, with families in the Uatumã-Trombetas moist forests ecoregion showing a higher number of households with less than one Brazilian minimum wage per month and fewer households receiving more than three minimum wages per month compared to the Guianan savanna ecoregion (Table 1).

Attitudes Towards Wildlife

Overall respondents unambiguously identified 57 species/congeners (Table 2) that were liked (n = 31 species), disliked (n = 36 species) or damaged income (n = 31 species). Interviewees were consistent with their responses while citing liked and disliked species, never contradicting themselves while answering our questions. For example, although it is possible for a species such as wild pigs (Pecari tajacu and Tayassu pecari) to be both liked for food and disliked due to the same species causing crop damage, none of the respondents indicated the same species as both liked and disliked. Kruskal-Wallis tests demonstrated a significant difference in the total number of liked species between the two ecoregions, with people from the Guianan savanna reporting more species liked compared with Uatumã-Trombetas moist forest (Kruskal-Wallis χ² = 15.44, df = 1, p < 0.001). Differences in total number of disliked wildlife species between the two ecoregions were marginally significant (Kruskal-Wallis χ² = 3.83, df = 1, p = 0.050), but there was no significant difference in the number of species reported to damage income between the two ecoregions (Kruskal-Wallis χ² = 0.52, df = 1, p = 0.473).

Table 2.

List of 57 Species or Congeners Perceived by Local Wage-Earners as Liked, Disliked, or Damaging Income in Forest and Guianan Savanna Ecoregions.

The Generalized Linear Models (GLMs) indicated that the explanatory power of the complete and best models were low for all attitudes on wildlife species (Table 3), with a maximum deviance explained of 15.8% (for the complete model of liked species) and a minimum of 6.2% (for the best model of disliked species). When considering the most parsimonious (“best”) models, ecoregion was the only variable retained for explaining variation in the number of wildlife species reported as liked and disliked. Although number of people living in the property was also retained in the model of disliked species it was not significant. Socioeconomic variables were retained for explaining the number of species reported as damaging income (Table 3). The significant socioeconomic variables retained for the best model were number of people living in the property and level of education (graduated), with both variables positively related to the number of species reported as causing damage.

Table 3.

GLM Model Results (Slope Coefficients With Associated ± SE in Parentheses) of Predictors of the Total Number of Wildlife Species Reported as Liked, Disliked and That Cause Damage to Their Income From 106 Interviews With Wage-Earners in Amapá state, Eastern Brazilian Amazon.

Medium and large-bodied mammals were the most liked faunal groups in both savanna and forest ecoregions (Figure 2). Although snakes were the most commonly reported disliked group in the two ecoregions, the second most disliked group in the Uatumã-Trombetas moist forest was large-bodied mammals, while in Guianan savanna the second most disliked faunal group was medium-bodied mammals (Figure 2). However, the largest difference between the attitudes of wage-earners in the two ecoregions was related to faunal groups considered to cause damage (Figure 2). Interviews from Uatumã-Trombetas moist forest reported medium- and large-bodied mammals as the two most commonly damaging faunal groups, and interviews from the Guianan savanna reporting medium-bodied mammals and ants as the faunal groups causing most damage to their income (Figure 2).

Figure 2.

Proportion of Interviews That Reported Attitudes of “Like”, “Dislike” and “Cause Damage” to 11 Faunal Groups in 106 Interviews Conducted With Local Wage-Earners in the Eastern Brazilian Amazon. Dark and light grey bars represent Uatumã-Trombetas moist forests (forest) and Guianan savanna (savanna) ecoregions, respectively.

Overall, from the total of 57 species/congeners mentioned there were generally more positive responses (“like”) regarding five species (Pecari tajacu, Mazama spp., Cuniculus paca, Dasyprocta sp., and Dasypus spp.) (Figure 3). With paca (C. paca) and deer (Mazama spp.) liked by the majority of respondents (57.5% and 56.6, respectively). Yet, the majority of species (54.3%; 31/57) were rarely mentioned i.e. cited less than 1% of responses (Figure 3). After excluding those species with less than 1% of responses, the majority of the remaining 26 species (84.6%; 22/26) tended to be mentioned in both ecoregions (forest and Guianan savanna), but with the proportion of responses sometimes differing between ecoregions (Figure 3). For example Dasyprocta sp. and Dasypus spp. were liked more by respondents in the savannah ecoregion, whereas Dasyprocta sp. was more likely to be perceived as damaging income in the forest ecoregion. In contrast there was less consistency in disliked species and species perceived as damaging income, with relatively low number of responses per species (<30% per species) (Figure 3).

Figure 3.

Attitudes of Species/Congeners. Percentage of respondents citing liked, disliked and damaging species/congeners across forest (black) and savannah (grey) ecoregions. Symbols indicate significant differences in proportion of responses between ecoregions (Significance levels: ⁺<0.10, ^*<0.05, ^**<0.01, ^***<0.001).

The PERMANOVA results revealed significant differences in species composition between respondents (Table 4). There were significant differences in species between attitudes, region and interactions (Table 4). Of the socio-economic factors, only number of years since living in the property was associated with differences in species reported. Additionally, a significant interaction was found between number of years living in the property and education level (Table 4). The species reported also varied depending on attitude with region and number of years living in the property (Table 4). Ordinations showed a clear pattern in the composition (dissimilarities) of species/congeners associated with different attitudes (Stress value = 0.042, Figure 4). For example, species reported as liked were clearly different from the ones reported as disliked or damaging to income (Figure 4A). Eight species were most strongly indicative from the 31 species cited as liked (Figure 4B). These eight included mammals (Cuniculus paca, Dasypus spp., Mazama spp. and Tayassu pecari), birds (Amazonas spp., Ara spp. and Ramphastos spp.) and a tortoise (Chelonoidis spp.). A total of 6 species were most strongly indicative of the 36 disliked species, including mammals (Panthera onca, Puma concolor and Tamandua tetradactyla), other invertebrates (Aedes sp., Tityus spp.) and snakes (Bothrops spp.) (Figure 4B). In contrast only leafcutter ants (Atta sp.) were strongly indicative of the 31 species cited as damaging income (Figure 4B).

Table 4.

PERMANOVA Results.

Figure 4.

Ordinations. Two-dimensional non-metric multi-dimensional scaling (NMDS) ordination of (A) interviews and attitudes and (B) indicator species most strongly associated with attitudes. Lines represent convex hull encompassing all responses from different ecoregions.

First Cited Species and Motives

A total of 17 species/congeners were first cited as liked. People interviewed in the forest cited more species as liked compared with the savanna ecoregion (12 and 9 species/congeners, respectively, Figure 5A). There were also differences in the groups cited as liked between ecoregions. For example, more large-bodied mammal species were reported as liked in the forest, whereas the majority of the species perceived as liked in the savanna were medium-bodied mammal species (Figure 5A). The reasons for respondents perceiving species as “liked” were consistent among answers for the different species (Figure 5A). The most frequently first cited species (e.g. Cuniculus paca, Dasyprocta sp., Dasypus spp., Mazama spp., and Pecari tayassu) were all liked because they were used for food. Indeed the majority of species (58.9%, 10/17) were liked because of their use as food, yet seven species (3 birds, 2 mid- and 2 large-bodied mammals) were liked because of their beauty (Figure 5A).

Figure 5.

Wage-Earners' attitudes towards wildlife. Proportion of interviews that reported attitudes of (A) like, (B) dislike, and (C) damage income to first cited species or congeners in faunal groups (invertebrates, birds, large-bodied mammals, medium-bodied mammals, turtles and tortoises, and snakes) for 106 interviews conducted with local wage-earners in Uatumã-Trombetas moist forests (forest) and Guianan savanna (savanna) ecoregions. Bars shaded based on the motivation for the attitude. Colours represent: (A) black and grey bars indicate species used for food or considered beautiful, respectively, (B) black, grey, and white bars represent species not used for food, dangerous, or that cause damage to subsistence, respectively, and (C) black and grey bars represent species perceived as predators of domestic animals/livestock, or causing damage to plantation/crops, respectively.

A total of 17 species/congeners were also cited as disliked. Overall, people in the forest disliked more species compared with people in the savanna (13 and 9 species, respectively) (Figure 5B). Species were disliked as they were not used for food (7 species), dangerous (5 species) or damaged subsistence (6 species). Reasons for disliking species were the same between forest and savannah ecoregions except for howler monkeys (Alouatta spp.), which were disliked as they damaged subsistence or were not used as food in forest and savannah areas respectively (Figure 5B).

Fewer species/congeners (n = 15) were cited as damaging income (Figure 5C). Species first cited as damaging income were linked directly with predation on domestic animals (7 species) or causing damage to plantations/crops (8 species, Figure 5C). Three species were most frequently cited as damaging income, all due to damage caused to plantations/crops: mid- (Dasyprocta sp.) and large-bodied (Pecari tajacu) mammals and leafcutter ants (Atta sp., Figure 5C).

Socioeconomic Variables and Ecoregion Effects on Wage-Earners Attitudes

Our results indicate that ecoregion type was the most important determinant for the number of wildlife species liked and disliked reported by respondents. However, socioeconomic variables such as number of people living in the property and education level were also significant positive determinants on the number of species reported as causing damage to the income of respondents. An increased number of people living in a property will generate an increased demand on food resources and generate more acute perception of losses that can cause negative attitudes. As wildlife can cause crop damage (Abrahams et al., 2018; Pérez & Pacheco, 2006) as well as prey on livestock (Boulhosa & Azevedo, 2014; Carvalho et al., 2015; Michalski et al., 2006) and compete for protein resources (Michalski et al., 2012) it is possible that larger families perceive more species as causing damage compared with smaller families that will need smaller food supply.

Education level is a well-established social factor that can affect people’s attitude towards wildlife (Fort et al., 2018; Hariohay et al., 2018; Shibia, 2010; Vodouhê et al., 2010). However, contrary to what we expected, respondents with higher education level (graduated) were the group reporting significantly more species as causing damage to their income. Our results should be looked carefully as our sample size was not equally distributed and the majority of our respondents had relatively low levels of education. However, it is interesting to highlight that, based on our results, higher levels of education do not necessarily equate with more positive attitudes towards wildlife species.