Audience research as a cornerstone of demand management interventions for illegal wildlife products: Demarketing sea turtle meat and eggs

CONTRIBUTED PAPER Audience research as a cornerstone of demand management interventions for illegal wildlife products: Demarketing sea turtle meat and eggs Diogo Veríssimo 1,2 | Sara Vieira 3 | Domingas Monteiro 3,4 | Joana Hancock 5 | Ana Nuno 6 1 Department of Zoology, University of Oxford, Oxford, UK 2 San Diego Zoo Institute for Conservation Research, Escondido, California 3 Programa Tatô, Largo do Bom Despacho, Cidade de S~ ao Tomé, S~ ao Tomé e Príncipe 4 MARAPA –Mar, Ambiente e Pesca Artesanal, Largo Bom Despacho - CP 292, Cidade de S~ ao Tomé, S~ ao Tomé e Príncipe 5 cE3c - Centre for Ecology, Evolution and Environmental Changes, Faculdade de Ciências, Universidade de Lisboa, Edifício C2, 5Piso, Sala 2.5.41 Campo Grande, Lisboa, Portugal 6 Centre for Ecology and Conservation, College of Life and Environmental Sciences, University of Exeter, Penryn, Cornwall, UK Correspondence Sara Vieira, Avenida Amílcar Cabral no 27, Cidade de S~ ao Tomé, S~ ao Tomé, S~ ao Tomé and Príncipe. Email: [email protected] Funding information Oxford Martin School, University of Oxford, Grant/Award Number: Oxford Martin Programme on the Illegal Widllife Tr Abstract The unsustainable wildmeat trade has been long recognized as a threat to wildlife. Yet, its impact on marine species has been underresearched. Sea turtles have been historically threatened by unsustainable trade, but there are few studies of consumer profile, motivations, and preferences. We conducted a survey in S~ ao Tomé Island, Gulf of Guinea, West Africa, including eight rural communities (n= 1,160) and the capital city of S~ ao Tomé (n= 628). We estimated prevalence of consumption, preference and availability of sea turtle meat and eggs. About 25% of rural and 32% of urban respondents had consumed meat in the past year. Around 25% of rural respondents were egg consumers, whereas little consumption was found in urban areas. Social norms were a predictor of consumption of sea turtle meat and eggs in rural communities but not in S~ ao Tomé city. Regarding influencers, teachers and religious leaders were the most trusted sources in rural communities, while teachers and NGOs were most trusted in S~ ao Tomé city. Radio and television were the most trusted channels in rural communities, while urban dwellers trusted radio the most. This research showcases how a structured approach to audience research can obtain behavioral insights that can guide behavior change efforts. KEYWORDS aquatic, behavior change, bushmeat, demand reduction, illegal behavior, marketing, S~ ao Tomé and Príncipe, unmatched count technique, wildlife trade, wildmeat 1|INTRODUCTION All major threats to biodiversity are a consequence of the daily decision-making patterns of billions of humans worldwide (Schultz, 2011). Thus, conservation practitioners and scientists are increasingly focused on influencing human behavior and understanding the drivers behind it (Veríssimo, 2019). This has led to the increasing interest in fields such as conservation marketing, which is the ethical application of marketing strategies, concepts, and techniques to influence attitudes, perceptions, and behaviors of individuals, and ultimately Received: 6 September 2019 Revised: 20 November 2019 Accepted: 11 December 2019 DOI: 10.1111/csp2.164 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Conservation Science and Practice published by Wiley Periodicals, Inc. on behalf of Society for Conservation Biology Conservation Science and Practice. 2020;2:e164. wileyonlinelibrary.com/journal/csp2 1of14 https://doi.org/10.1111/csp2.164 societies, with the objective of advancing conservation goals (Veríssimo, 2013, 2019; Wright et al., 2015). Marketing strongly emphasizes the key role of having an indepth understanding of the people using a given resource (i.e., the consumers), a group often called the target audience (Kotler & Armstrong, 2010; Noel, 2009). Through consumer or audience research, marketers are able to better understand the drivers and context in which consumers make decisions about resource use. These drivers can be psychographic (e.g., attitudes, social norms, and behavior) and/or demographic (e.g., gender, age, income), and influence not only the recognition of a need but also the decision to act on it (Figure 1). This knowledge allows marketers to segment key audiences, define messages, and select channels and influencers through which to reach the target consumers (Kotler & Armstrong, 2010; Noel, 2009) (Figure 1). Yet, in the context of biodiversity conservation in general and the illegal wildlife trade in particular, many behavioral interventions are planned and implemented without a solid evidence base with which to identify and understand a target audience (Greenfield & Veríssimo, 2019). This is problematic because, even when there is detailed knowledge of the target audience, the high complexity and high uncertainty surrounding behavior change interventions mean that the occurrence of unintended consequences is a real possibility (Douglas & Winkel, 2014). The high complexity in which behavior change interventions occur comes from the different social actors attempting to influence the behavior of any stakeholder group while the uncertainty comes from both the feedback loops between the actions of different stakeholders but also the unpredictable external factors such as macroeconomic trends or fashions that can quickly change the preferences and norms regarding a given behavior. 1.1 |The case of wildmeat In the last decade, the unsustainable consumption and trade of meat from wild animals, also known as wildmeat, has been increasingly recognized as a key threat to wildlife populations (van Velden, Wilson, & Biggs, 2018). In some regions, including the tropics where most biodiversity is located, this phenomenon has been responsible for the decline and even local extirpation of several large and medium bodied species (Ripple et al., 2016). Attempts to tackle this unsustainable exploitation have focused mostly on the supply side, through improved law enforcement and tightening of harvesting regulations (Challender & MacMillan, 2014; Wright, Bhammar, Gonzalez Velosa, & Sobrevila, 2016). Yet, these actions have been shown to not be enough as they FIGURE 1 A conceptual model of consumer decision-making processes and their drivers in the context of a social marketing intervention. Blue marks the stages targeted during consumer research to better understand consumers (adapted from Middleton, Fyall, Morgan, Morgan, & Ranchhod, 2009; Milner & Rosenstreich, 2013; Noel, 2009) 2of14 VERÍSSIMO ET AL. are often undermined by the central role played by wildmeat in the livelihoods and social interactions of millions of people worldwide (Brashares et al., 2004). While on many contexts wildmeat is a key part of the diet of human rural populations, in urban contexts it is often a delicacy (i.e., a preferred and less accessible food item), but also used for important social interaction such as business meetings (Chausson, Rowcliffe, Escouflaire, Wieland, & Wright, 2019). These factors mean that a strong demand for wildmeat remains, which together with weak governance has tended to undermine many interventions focused on stifling supply of wildlife products (Challender & MacMillan, 2014). Reducing or shifting the demand from this commodity to other more sustainable alternatives is therefore likely to play a critical part of any solution to mitigate the impact of wildmeat harvesting on biodiversity (Chaves et al., 2018). This realization has led to an increase in the number of demand reduction interventions by conservation practitioners (Veríssimo, Schmid, Kimario, & Eves, 2018; Verissimo & Wan, 2019). This focus on the reduction of demand, as opposed to the more widespread use of marketing to stimulate demand, has been labeled in the social marketing literature as “demarketing,”and defined as “marketing that deals with discouraging customers in general or a certain class of customers in particular on a temporary or permanent basis”(Kotler & Zaltman, 1971). Demarketing has been used in a variety of issues by social marketers, namely around public health issues such as smoking or alcohol consumption (Lefebvre & Kotler, 2011). 1.2 |Beyond terrestrial wildmeat While there has been a number of consumer research studies on wildmeat use in the context of terrestrial species, such as rodents, ungulates, and primates (Chausson et al., 2019; Schenck et al., 2006; Shairp, Veríssimo, Fraser, Challender, & MacMillan, 2016), there is a large knowledge gap in terms of wildmeat consumption in the aquatic context, or aquatic bushmeat (Aquatic Mammals Working Group of the Convention of Migratory Species, 2017; Cosentino & Fisher, 2016). This lack of consumer research within aquatic systems is concerning given close linkages of, for example, coastal and marine areas to issues of food security, poverty alleviation, wellbeing, and health, in addition to sustainable use of natural resources and biodiversity conservation. As such it is imperative that interventions to reduce demand for wildmeat take into account the context in which they are being implemented to ensure not only that they have the highest likelihood of success but also that there are no undue burdens placed on the target audience. This could be the case for example if a group of people using wildmeat for either livelihood or subsistence purposes were targeted for using wildmeat without considering that there could be no other viable alternatives to them. In this article we outline how consumer research can be used to inform a behavior change intervention for biodiversity conservation. We use the demand for sea turtle meat and eggs in the island of S~ ao Tomé, Gulf of Guinea, West Africa, as a case study. Sea turtles are a biological group which has been the target of documented unsustainable exploitation for decades (Frazier, 1980; Garland & Carthy, 2010), and for which limited consumer research is available (but see, e.g., Garland & Carthy, 2010; Nuno et al., 2018) despite the fact that this group is often targeted by demand reduction interventions (Veríssimo & Wan, 2019). 1.3 |The sea turtles of S~ ao Tomé The island of S~ ao Tomé is the largest island of the archipelago of S~ ao Tomé and Príncipe, being home to more than 95% of the country's population. It is a known nesting and foraging ground for five species of sea turtles, of which the most abundant are the green (Chelonia mydas)and olive ridley sea turtles (Lepidochelys olivacea) (Castroviejo, Juste, Pérez, Castelo, & Gil, 1994). Sea turtles have been exploited for human consumption in S~ ao Tomé Island for centuries, with both meat and eggs used for both commercial and subsistence reasons (Castroviejo et al., 1994; Vieira et al., 2016). Sea turtle conservation efforts started in the 1990s, and led to the creation of the sea turtle conservation project “Programa Tatô”(www.programatato. org), (“Tatô”being the local name for olive ridley sea turtle). Run until 2018 by the local non-governmental organization (NGO) MARAPA (Graff, 1996), Programa Tatô is now run by an independent NGO of the same name. Programa Tatô was initially focused on monitoringthenestingactivityofseaturtlesaroundS ~ ao Tomé Island, but has progressively engaged a wide range of stakeholders, resulting in activities such as training workshops for law enforcement officers, education campaigns targeting schools and lobbying the national government in the case of the legislation forbidding the trade of sea turtle products (Vieira, Jiménez, Besugo, et al., 2016; Vieira, Jiménez, Hancock, et al., 2016). In 2014, the Santomean government approved national legislation (Decreto-Lei n.8/2014, of 28 April), criminalizing the possession, trade and transportation of sea turtles (dead or alive) and their by-products in the archipelago (Vieira, Jiménez, Besugo, et al., 2016). However, the law was not readily enforced, a challenge VERÍSSIMO ET AL.3of14 common to many developing nations where competent institutions often lack of the necessary technical capability and logistic means (Albuquerque & Cesarini, 2009). This lack of enforcement led to the need to shift focus from supply, and more coercive measures, to demand and voluntary behavior change, thus arising the need for understanding sea turtle consumption in the island. 2|METHODS Our data was obtained using a questionnaire survey divided into four sections (See Supporting Information S1 and S2). To guide the development and content of our questionnaire, we focused on attitudes, a positive or negative evaluation of something or someone; as well as social norms, collective understandings of group conduct as well as individual perceptions of desirable or simply actual group conduct (Lapinski & Rimal, 2005). This follows the simplified version of the Theory of Planned Behavior adopted by on Fairbrass, Nuno, Bunnefeld, and Milner-Gulland (2016). The Theory of Planned Behavior is one of the most widely used models to understand and identify the drivers of human behavior (Barber, 2012). In its full form the theory contends that attitudes toward a behavior, subjective social norms and perceived behavioral control, together shape an individual's behavioral intentions and consequently his/her behaviors (Ajzen, 1991). This survey was then refined using insights from 19 unstructured interviews with Santomean fishers and sea turtle meat sellers conducted from December 2014 to March 2015. The first questionnaire section covered sociodemographic characteristics of the respondents, including ownership of multiple household items meant to allow for the calculation of a household wealth index, as well as self-reported level of trust regarding different media and potential key influencers. The wealth index used is an asset based measure based on the presence or absence of 12 items (See Supporting Information S4). It was used as a more robust way to understand long term patterns around household expenditure, when compared to traditional direct questions on income which require extensive field work in order to capture what are often multiple income streams, self-employment, and in-kind transactions (Vyas & Kumaranayake, 2006), not to mention challenges due to social desirability bias and the potential sensitives around openly disclosing income level. The second section focused on the consumption behavior for sea turtle eggs and meat, and used the unmatched count technique (UCT) also known as a list experiment; this is a specialized questioning technique developed to improve the willingness of respondents to answer truthfully to possibly embarrassing or selfincriminating questions by providing greater levels of privacy and anonymity and reducing question sensitivity (Glynn, 2013; Nuno & St John, 2015). In UCT, survey respondents are allocated into a control or a treatment group, in which control group members receive a list of nonsensitive items (e.g., food items such as chicken or fish), whereas the treatment group receives the same list but with the addition of the sensitive item (i.e., sea turtle). Group allocation (control or treatment) for the UCT was assigned pseudo-randomly at the start of the UCT, based on the time of the day; if the watch of the enumerator marked an even number of minutes, the respondent was assigned to the treatment, if it was odd the respondent was assigned to the control. Every respondent was thus shown multiple UCT cards (one pair per question; see Supporting Information S3) and asked to indicate how many, but not which, items applied to them (Glynn, 2013). UCT questions were preceded by a nonsensitive training question about household occupations. The third and fourth sections focused on, respectively, attitudes and social norms toward sea turtle conservation and consumption. These two last sections used five-point Likert scales to measure self-reported levels of agreement with specific statements. Here, we used direct questioning rather than a specialized questioning technique such as the UCT, as the concern in the context of S~ ao Tomé was with potential legal repercussions stemming from law breaking, something that can only happen through actual behavior. We piloted our questionnaire in the city of S~ ao Tomé Island (n= 10), based on which minor adjustment to language of several questions were made. This pilot data was not included in any further analysis. The length of each questionnaire was ~20 min. Participation in the survey was voluntary. The use of UCT for asking potentially sensitive questions makes it impossible to directly link incriminating data to an individual, and respondents were informed they could refuse to answer any question and withdraw at any time. Only respondents over 18 years old were eligible. The anonymity of participants was fully protected, and the study and its methodology were approved by the College of Life and Environmental Sciences (Penryn) Ethics Committee at the University of Exeter (reference 2017/1755), United Kingdom. Data collection focused on two priority groups: the inhabitants of rural and coastal fishing communities and the residents of the capital city of S~ ao Tomé. Regarding the former, the survey was conducted by two enumerators affiliated with the Santomean NGO Monte Pico, an organization with no institutional affiliation with Programa Tatô, the only sea turtle conservation program in the island. This survey took place between May and 4of14 VERÍSSIMO ET AL. October 2016, in five communities where Programa Tatô has worked since 2014, and three communities where Programa Tatô has never worked (Figure 2). The survey was conducted in Portuguese, with occasional use of the Fôrro (local dialect) when necessary. These fishing communities were selected because they neighbor important turtle nesting beaches where sea turtle poaching is known to happen. We set out to conduct a household census in the target villages, with questionnaires being conducted at the respondent's home. Within a household, we identified the respondent pseudo-randomly, using the house number. Where the house number was even, we surveyed the male head of the household, where it was odd, we surveyed the female head of the household. The second priority group was the inhabitants of the S~ ao Tomé city, where the key produce market of the country is located and where sea turtle is more regularly available for purchase. Our aim was to target all 17 neighborhoods of the city, to achieve a broad picture of behavioral patterns of urban dwellers. This survey took place between October and November 2017 and was carried out in Portuguese by two enumerators affiliated with University of S~ ao Tomé and Príncipe. We used a streetintercept survey method, with one busy location in each neighborhood selected and each third person approached to participate. Only people over 18 years old were eligible to participate. 2.1 |Data analysis To measure the wealth of a household we used IBM SPSS 22.0 to produce an index. This index was made by first carrying out descriptive statistics for all variables and then using a Principal Component Analysis to determine the weights to be given to each variable, after which the first component was used to assign each item a weight (see Vyas & Kumaranayake, 2006). Lastly all weights were summed per respondent. For subsequent statistical analysis, this wealth index was converted into categorical with two levels: one including values above, and the other values below or equal to the median wealth. Likertscale statements were visualized using the Likert R package v.1.3.5 (Bryer & Speerschneider, 2016). To compare FIGURE 2 Locations in S~ ao Tomé island surveyed about the consumption of sea turtle meat and eggs VERÍSSIMO ET AL.5of14 differences in agreement with statements on attitudes, social norms and key influencers between study groups, we used the Mann–Whitney–Wilcoxon test; this allows comparing distributions of answers in Likert items while accounting for independence between the two study groups (Jamieson, 2004). To estimate behavior prevalence, generalized linear models in R 3.5.1 (R Core Team, 2018) were fitted only with card type (treatment or control). Then, UCT answers to sensitive questions were fitted with a dummy variable indicating whether the respondent received the baseline or the treatment card, the demographics, and interactions of the card membership dummy variable with each demographic (Holbrook & Krosnick, 2010); the interactions between sociodemographic variables and treatment status indicate differences between the reported number of behaviors in the two conditions for each predictor variable. Akaike information criterion (AIC) was used to select the most parsimonious models and to rank models according to their log-likelihood penalized for the number of parameters (Burnham & Anderson, 2002). When analyzing the number of reported activities to identify characteristics of individuals engaged in sensitive behaviors, only models with interactions included were considered for comparison. We averaged parameter estimates across models with ΔAIC <4 using the MuMIn package v.1.42.1 (Barto n, 2018). ΔAIC ≥4 indicates considerably less support for the model (Burnham & Anderson, 2002). 3|RESULTS 3.1 |Study participants Regarding the eight rural communities, we surveyed a total of 1,160 respondents. Between 73 and 378 households per community were sampled, with the number of surveys representing a mean 69% of total households per community. This is a conservative estimate given uncertainty in some community boundaries and unclear match to national census data. In addition, this average estimate is disproportionately influenced by the low proportion of households surveyed in Praia Gamboa (28%), an outcome due to hostility of local residents toward the subject of sea turtle conservation. Thirty respondents (2.6% of total sample size) did not provide answers to some of the sociodemographic questions and were removed from further analysis. While most nonresponse was due to the absence of respondents in many households, we unfortunately we did not collect data on response in the rural communities, due to a miscommunication with the field team, which hampers our ability to understand the importance of potential nonresponse bias. Regarding the residents of S~ ao Tomé city, we approached 672 respondents of which 628 agreed to complete the questionnaire (nonresponse rate = 6.5%). Respondents were from 15 of the 17 neighborhoods of S~ ao Tomé city, with a mean of 42 (SD = 7) respondents per neighborhood. However, 15 surveys had missing demographic data and were discarded, putting the final sample size at 613. This sample represents about 1% of the population of S~ ao Tomé city according to the estimates of the 2012 national census (INE, 2012). A summary of key sociodemographic characteristics of participants for both study groups is provided in Supporting Information S4. 3.2 |Attitudes and social norms Respondents from rural communities generally agreed with attitudinal statements regarding the need for enforcement and legislative measures as well as the need for sea turtle conservation, while statements about potential reasons justifying sea turtle use (e.g., cultural and food security)weremorecommonlydisagreedwith(Figure3). While respondents generally agreed with statements about measures penalizing harvesters and sellers (e.g., 86% agreed that capture and sale of sea turtles should be outlawed and 76% agreed that arresting poachers is the best way to protect turtles), survey participants were more evenly split about penalizing consumers (e.g., 53% agreed vs. 45% disagreed with punishing those who eat turtle meat). These findings were mirrored by our respondents in S~ ao Tomé city, with an even stronger tendency toward the need for tougher sanctions on those that break rules on use and trade of sea turtles. Urban respondents felt more strongly about penalizing consumers (e.g., 77% agreed with punishing those who eat turtle meat, in contrast with only 53% of rural respondents who agree with the same statement). The importance of the role of the government in terms of protecting sea turtles was generally agreed by both urban and rural study groups (88% for both groups). While sea turtles were widely recognized as part of the Natural Heritage of S~ ao Tomé among both study groups, a substantial part of respondents in both urban (20%) and rural (31%) settings believed that eating sea turtle meat is part of the national culture; levels of agreement with this statement were identical between both study groups (W= 347,330, p= .17). The majority of respondents in rural communities strongly supported the statement that sea turtles will never go extinct, whereas urban respondents where nearly evenly divided on the topic. Moreover, there was alignment between rural and urban audiences, with overwhelming agreement that sea turtles where not a source of protein without which 6of14 VERÍSSIMO ET AL. communities may see their food security jeopardized; again, levels of agreement with this statement were identical between both study groups (W= 343,250, p= .10). In terms of availability, in both rural and urban contexts sea turtle was considered to be hard to find, although less so by rural respondents. In relation to the normative statements, there was agreement among residents of rural communities that both the consumption of sea turtle meat and eggs is a minority behavior. The same was true of turtle poaching. Nonetheless, two-thirds of rural respondents and more than half (56%) of urban respondents agreed that sea turtle meat was a delicacy. The situation was however different for sea turtle eggs, with only a minority preferring them to chicken eggs, particularly in urban areas. In terms of the context where sea turtle meat is consumed, the perceived norm was that this was not restricted to special occasions; levels of agreement with this statement were identical between both study groups (W= 361,060, p= .59), with other normative statements presenting significant differences between study groups (all p< .042). It should be noted that urban dwellers demonstrated much more uncertainty in the responses, with some questions reaching 20% of respondents neither agreeing nor disagreeing, which may also account for some of the differences described above (Figure 4). 3.3 |Prevalence of sensitive behaviors Regarding rural community residents, based on UCT, close to half (43%) of the respondents considered turtle meat to be very tasty, a number that was reduced to about one quarter in S~ ao Tomé city (Figure 5). In terms of actual consumption of sea turtle meat, we estimate that about a quarter of rural respondents and 32% of urban respondents were consumers during the 12 months prior to the study. In terms of regular consumption, we estimate that about 16–20% of respondents in both areas consumed sea turtle meat more than once a month on average. The scenario changes when sea turtle eggs are considered with an estimated quarter of respondents in rural areas being consumers but virtually no consumption in urban areas (UCT produced an unrealistic negative prevalence, although overlapping with zero). The availability of sea turtle meat is estimated to be higher in rural communities, with about 25% of residents seeing the product for sale, a number 8% higher than that estimated for urban areas. 3.4 |Potential predictors of sensitive behaviors In the context of rural communities, social norms and being or not a Programa Tatô target community were the most important variables explaining variation in involvement in target behaviors related to sea turtle consumption (Table 1; See Supporting Information S5 and S6). Level of agreement with social norm statements was a significant predictor of involvement in several target behaviors (Table 1); the stronger people agreed with these statements, the more likely they were involved in these behaviors. Furthermore, communities where FIGURE 3 Self-reported levels of agreement among survey respondents according to attitudinal statements on sea turtle conservation and consumption for both study groups: (a) residents of rural communities and (b) residents of S~ ao Tomé city. Statements are shown in the order they were presented to respondents during survey. All questions were answered using a Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Percentages represent: general disagreement (on the left; combining “strongly agree”and “agree”), neutral answers (central) and general agreement (on the right; combining “strongly disagree”and “disagree”) VERÍSSIMO ET AL.7of14 Programa Tatô had worked previously were more likely to consume sea turtle eggs. This scenario changed when it came to respondents in S~ ao Tomé city, where no variables where significant at the 5% level (See Supporting Information S6). 3.5 |Key influencers and communication channels When asked to rate their levels of trust on multiple information sources, rural community residents selected radio, teachers, television, and religious leaders as the most trusted sources, with 87% or more of respondents reporting trusting these groups, while less than 8% reported distrusting them (Figure 6). At the other end of the spectrum, politicians were the least trusted group, with 46% of respondents reporting distrust. Surprisingly, “friends”were also scored low, although having the highest prevalence of neutral respondents (22%). Regarding the residents of S~ ao Tomé city, the levels of trust were generally lower, with teachers and NGOs with 63% or more of respondent stating that they trust them. It should be noted however that for religious leaders the FIGURE 5 Estimated prevalence (SE) of sensitive behaviors and attitudes related to the sale and consumption of turtle meat and eggs during the 12 months prior to the study among: (a) residents of rural communities and (b) residents of S~ ao Tomé city. Behaviors presented from potentially least sensitive (“turtle is tasty”)to potentially most sensitive (“eating turtle > 1×month”) FIGURE 4 Levels of agreement among survey respondents according to normative statements on sea turtle conservation and consumption for both study groups: (a) residents of rural communities and (b) residents of S~ ao Tomé city. Statements are shown in the order they were presented to respondents during survey. All questions were answered using a Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Percentages represent: general disagreement (on the left; combining “strongly agree”and “agree”), neutral answers (central) and general agreement (on the right; combining “strongly disagree”and “disagree”) 8of14 VERÍSSIMO ET AL. TABLE 1 Effects (Unconditional estimates; SE) of sociodemographic variables on estimated prevalence of sensitive behaviors related to the sale and consumption of turtle meat and eggs during the 12 months prior to the study for both study groups: residents of rural communities and residents of S~ ao Tomé city Residents of rural communities Residents of S~ ao Tomé city Parameter Eating turtle Turtle is tasty Eating turtle >1×month Seeing turtle for sale Eating turtle eggs Eating turtle Turtle is tasty Eating turtle >1×month Seeing turtle for sale Eating turtle eggs Community: Programa Tatô target area −0.27 (0.18) 0.36 (0.20) + −0.05 (0.18) −0.08 (0.19) 0.35 (0.07)** Not included Not included Not included Not included Not included Social norms 0.18 (0.08)*0.18 (0.09)*0.10 (0.08) 0.10 (0.08) 0.07 (0.03)*0.15 (0.17) 0.18 (0.18) 0.24 (0.18) 0.30 (0.17) + −0.01 (0.01) Gender: Female −0.03 (0.10) 0.01 (0.14) −0.01 (0.14) −0.02 (0.12) 0.01 (0.02) 0.05 (0.15) −0.01 (0.05) 0.03 (0.11) −0.16 (0.23) 0.15 (0.09) + Age −0.01 (0.01) 0.01 (0.01) −0.01 (0.01) −0.01 (0.01) −0.01 (0.01) −0.02 (0.02) −0.01 (0.01) 0.01 (0.01) —−0.01(0.01) Education level a Primary Secondary or above −0.66 (0.38) + −0.50 (0.42) 0.24 (0.37) 0.28 (0.40) −0.38 (0.35) −0.09 (0.38) −0.55 (0.37) −0.30 (0.39) — — −0.06 (0.26) — 0.01 (0.19) — 0.25 (0.40) — 0.41 (0.45) — 0.03 (0.10) Main occupation: Other b 0.01 (0.01) 0.14 (0.20) −0.01 (0.04) −0.01 (0.07) 0.01 (0.03) — 0.07 (0.21) −0.03 (0.14) 0.10 (0.25) 0.16 (0.16) Wealth: Above median 0.01 (0.01) −0.01 (0.06) −0.01 (0.04) 0.06 (0.14) 0.01 (0.02) −0.01 (0.11) 0.08 (0.25) −0.04 (0.24) −0.18 (0.23) −0.01 (0.5) Religion c Evangelical Other — 0.08 (0.22) 0.05 (0.15) 0.14 (0.26) 0.06 (0.16) ———— — —— Notes: For the social norm's category, we used the mean of the seven items in the category to obtain a score per respondent. Only averaged estimates from interactions between variables and treatment status in the linear mixed models fitted to UCT answers are presented, indicating differences between the reported number of behaviors in the two conditions for each predictor variable. Statistical significance is coded as **p< .001, *p< .05, + p< .1; “—” represents absent from the best performing models. a Reference level is no formal education. b Reference level is fisher or fish seller. c Reference level is Catholic. VERÍSSIMO ET AL.9of14