Marketing Approach to Selecting Flagship Species
We followed the first 4 steps of the systematic framework developed by Veríssimo et al. (2011bb) to select a flagship species to represent the biological diversity of the Serra do Urubu. First, we identified the conservation problem to be tackled as the lack of awareness of the uniqueness and importance of the avifauna of the Serra do Urubu, a problem previously identified by SAVE Brazil. Second, we identified the target audience. We selected the communities living around the main forest fragments of the Serra do Urubu because they directly affect the forest biological diversity through subsistence harvesting of natural resources, and the students and teachers of both high schools in Lagoa dos Gatos, the largest town in the region, because SAVE Brasil had worked with these schools on previous projects. Third, we investigated the relation between the selected audience groups and the birds of the Serra do Urubu. We did this by reviewing previous research carried out by SAVE Brasil and through interviews with key informants. Fourth, we focused on identifying the best species for promoting the campaign as one aspect of the campaign's “marketing mix” (Veríssimo et al. 2011a). For this, we used discrete-choice experiments because these have a strong grounding in behavioral theory and so reflect real-life choices (Louviere et al. 2010).
Choice experiments provide an understanding of how individuals’ value particular attributes of a product or service by presenting respondents with choice questions. Each choice presents 2 or more hypothetical alternative scenarios described by different levels of the same attributes (Mangham et al. 2009). We applied this approach to the choice of flagship species by asking respondents to choose between hypothetical species profiles (Veríssimo et al. 2009). Respondents’ choices then allowed us to identify preferred species characteristics.
We selected the species’ attributes to be included in the choice experiments and decided how they would vary (hereafter attribute levels). The attributes were species characteristics, and we selected these according to the information gathered through interviews with key informants and a review of the relevant literature (e.g,. Stokes 2007; Veríssimo et al. 2009). We selected direct-use, indirect-use, and nonuse attributes to determine how these aspects contributed to species valuation. We used a limited number of attributes so the survey would not be confusing to participants or lengthy. The 5 attributes were appearance, population size, geographic distribution, visibility, and ability to survive in captivity. These are described below with details on how we set attribute levels (Table 1).
Table 1. Species attributes and corresponding levels used in choice experiment and socioeconomic data collected (coding in parentheses) in the examination of respondents’ preferences for a bird flagship species for the Serra do Urubu, Brazil
|Species variables|| |
|Appearance||perception of how visually attractive a species is, presented through drawings of bird species divided into 2 levels: attractive (1) and unattractive (0)|
|Population size||number of individuals of a given species worldwide divided into 3 levels: <50 (0), 250–1000 (1), and 2500–10,000 (2).|
|Geographic distribution||number of sites, in addition to the study area, where the species occurs divided into 2 levels: study area plus 2 other sites (0) and study area plus 10 other sites (1)|
|Visibility||whether the given species would on an average day will not be heard or seen (0), only heard (1), or be both seen and heard (2)|
|Captivity||whether individuals of a given species would survive as a cage bird (1) or not (0)|
|Socioeconomic variables|| |
|Gender||female (0) or male (1)|
|Respondent type||whether a respondent was a high school student or teacher (0) or a rural community member (1)|
To determine preferences regarding bird appearance, we organized 3 workshops with students. Each student was randomly given 1 of 2 pages with 20 colored bird drawings of similar sizes. The birds were identified only by letters (Veríssimo et al. 2009). Students were asked to rate every drawing on a scale from 0 (not attractive) to 10 (very attractive). We selected the drawings to represent every taxonomic bird order that occurs in the Serra do Urubu to ensure, when possible, that every order had both a species with colorful and drab plumage. We standardized the resulting scores across the respondents score range and compared each species’ median score. Given the distribution of the scores across the scale, we decided to divide the appearance attribute into not attractive and attractive categories. We selected the 3 bird drawings with the lowest scores as not attractive species, and each time we used a not attractive species on a choice experiment card we selected one of these 3 species at random. We followed the same process for attractive species on the basis of the 3 bird drawings with the highest scores. Respondents were not informed whether a species had previously been considered attractive or not.
We divided population size into 3 levels, less than 50 individuals, between 250 and 1000 individuals, and between 2500 and 1000 individuals. Given the diminishing marginal value of an extra individual as population increases, we established the levels on the basis of population estimates for species occurring in the region and presented them in the form of nonlinear bounded ranges (IUCN 2010). We did this to convey some of the uncertainty surrounding population estimates in conservation and, because no consecutive population intervals were used, to ensure that levels could be differentiated by respondents.
We divided geographic distribution into 2 levels, species that could be found in the Serra do Urubu and 2 other regions and species that could be found in the Serra do Urubu and 10 other regions. With this design we sought to capture the fact that Serra do Urubu has no endemic bird species, but several species have very restricted ranges that include the area.
We divided visibility into 3 levels, species that are usually not heard or seen, species that are generally heard, and species that are usually heard and seen. This design took into account both the visual and auditory parts of bird watching. We excluded the level that considered seeing but not hearing the bird because this combination was unlikely.
We set survival in captivity as a binary attribute to determine whether respondents were more interested in bird species that could be kept as pets. There is widespread interest in keeping wild-caught birds, especially song birds, and this trade has had substantial negative effects on several bird species in the region (Pereira & Brito 2005).
We did not include a cost attribute because it tends to dominate responses in choice studies conducted in developing countries (Hope 2006) and because our aim was not to estimate willingness to pay.
Given the number of attributes types and levels, we sought to limit the potential cognitive burden by presenting each respondent with 4 dichotomous unlabeled choices. Each choice was a comparison of 2 hypothetical species. We then asked respondents which species should have its conservation priority increased, assuming that flagship-species campaigns are most successful when highlighting species the target audience is most concerned about. We did not include an explicit no-choice option because we believed some respondents would be uncomfortable with the survey process and automatically answer no choice if that was the quickest option. Although this differs from standard procedures for use of choice experiments to measure willingness to pay, we thought it would not affect the relevance of our results because we measured respondents’ relative preference for different species. Nonetheless, surveyors were instructed not to force respondents to choose and to assign any choices that respondents seemed unwilling or unable to select to an implicit no-choice category. In rural communities the survey was administered orally because of low target audience literacy rates, whereas in schools we explained the survey to the entire class and then asked students to complete it individually. In the absence of a reference demographic data set for the target population, we tried to achieve a representative sample in the rural areas by visiting communities on different weekdays and times of the day and by sampling all available residents in each visit, which was possible given the small size of communities. To ensure representativeness in schools, we sampled one randomly chosen class from each high school year and, given the small group size, all available teachers.
To test the survey, we used PASW Statistics (version 18.0, IBM, Chicago, IL, USA) to design scenarios with attribute combinations so that main effects of attributes on preferences could be estimated from orthogonal independent attribute variables. We then used a shifted or cyclic design to pair these scenarios in which a constant was added to each attribute level of an orthogonal design to produce one or more additional alternatives. The test survey given to one high school class and one rural community, both outside the study area so as to avoid members of the test study affecting the choices of later respondents. We used results of the test survey to produce the Bayesian prior distributions needed for the choice experiment. We used Ngene (version 1.0.1, ChoiceMetrics, Sydney, Australia) to produce a D-efficient Bayesian design for the main survey (Jaeger & Rose 2008). We chose this design type because it maximizes statistical efficiency in estimating preference parameters by minimizing D-error over the prior distribution of the parameters while accounting for uncertainty (Jaeger & Rose 2008). To allow for uncertainty, we used 500 Halton draws from normal distributions for each parameter prior distribution. We then compared the average Bayesian Dp error of over 30,000 Bayesian designs, selecting the one with the lowest error at 0.3303. This design had 8 choice situations, one of which is shown in Figure 1. The design was attribute balanced, meaning each attribute level occurred equally often, which minimizes the variance in parameter estimates (Mangham et al. 2009).
Figure 1. Example of one page, translated from Portuguese, of the choice-experiment survey used to select a bird flagship species for the Serra do Urubu.
Download figure to PowerPoint
We then used NLogit 4.0 to construct a multinomial logit model to determine aggregate preferences and explore preference heterogeneity among the respondents through latent-class models of flagship suitability, which divided respondents into groups according to their preferences and socioeconomic characteristics (see Supporting Information for further details). Lastly, we used Fishbein's multiattribute attitude model (Chiang et al. 2006) to combine the preference results from the latent-class model with data on each species to calculate a flagship suitability score for every bird species in the Serra do Urubu.