Not just the Big Five: African ecotourists prefer parks brimming with bird diversity

Ecotourism helps sustain protected areas (PAs) that in turn conserve Africa's declining fauna. Identifying ecotourist preferences and which species and landscapes benefit from ecotourism could therefore support African biodiversity conservation efforts. Due to historic associations with trophy hunting and subsequent ecotourism marketing efforts, ecotourist preferences have been thought to traditionally center around the ‘Big Five’: elephant, lion, buffalo, leopard, and rhinoceros. But these preferences may be evolving. Here, we ask two questions, one about the drivers and one about the consequences of ecotourism: (1) Which species and landscapes do ecotourists most prefer based on realized visitation data? And (2), differently, which species and landscapes benefit most from ecotourism? We gathered data on average annual tourist visits, the occurrence of nine mammals, bird species richness, forest cover, national wealth, local human population and accessibility for 164 Sub‐Saharan African PAs. To address our first question, we used a Bayesian multivariable model to identify whether bird and megafaunal diversity explain visits to PAs while controlling for other factors. To address our second question, we used Bayesian univariate models to analyze the relationships between park visitation and each species/landscape. We found that tourist preferences extend beyond the Big Five to include bird diversity. We also observed that ecotourism may be well suited to conserve bird diversity, lion, cheetah, black and white rhinoceros, African wild dog and giraffe species. Collectively, our results may help inform how to leverage ecotourism to conserve African fauna.


Introduction
As part of the global climate and ecological crisis (IPCC, 2014;IPBES, 2019), many African animals are threatened with extinction. This threat is substantial since Africa contains among the most diverse fauna in the world, many species of which are declining (Nieto et al., 2005;BirdLife International, 2018). For example, the northern white rhinoceros Ceratotherium simum cottoni is nearly extinct and other species' ranges are rapidly contracting: lions Panthera leo and elephants Loxodonta spp. will likely be extirpated from much of their historical ranges over the next two decades (Wittemyer et al., 2014;Bauer et al., 2015, but see Riggio et al., 2015). Even abundant mammals are declining precipitously (Craigie et al., 2010). Beyond mammals, Africa contains about a quarter of the world's bird diversity, and of these species, more than 10% are threatened with global extinction (BirdLife International, 2018). There are many proximate drivers of these bird and mammal declines including expansion of intensive agriculture, urbanization, timber harvesting, direct human-wildlife conflict and hunting (Wittemyer et al., 2014;BirdLife International, 2018;Balvanera et al., 2019;Nicole, 2019). Protected areas (PAs) have been shown to mitigate declines in African birds (Thiollay, 2006) and mammals (Riggio et al., 2015;Pacifici, Marco & Watson, 2020).
Ecotourism may help sustain PAs and thereby conserve African fauna. Ecotourism is defined as 'responsible travel to natural areas that conserves the environment, sustains the well-being of the local people, and involves interpretation and education' (The International Ecotourism Society, 2015). Ecotourism is a key mechanism for supporting PAs' capacity to mitigate the declines in African fauna (e.g. African wild dogs in South Africa, Higginbottom & Tribe, 2004;Lindsey et al., 2005), while also providing resources for communities to engage in non-extractive livelihoods (Nyaupane & Poudel, 2011). Beyond merely enabling alternative livelihoods, ecotourism may even improve livelihoods: for example, Snyman (2014) interviewed community members in six southern African countries and found that ecotourism was key to providing employment in rural areas. However, ecotourism can sometimes lead to negative outcomes for animals and people (Das & Chatterjee, 2015). For example, Cernea & Schmidt-Soltau (2006) found that hundreds of thousands of people have been displaced and become impoverished by the establishment of parks in the Congo Basin. Understanding the preferences, broadly construed, that underlie ecotourist visits could help decipher how to harness ecotourism to conserve African fauna.
Ecotourist preferences are not fully understood. Traditionally, the chief attractant of African ecotourists has been assumed to be the 'Big Five': lion Panthera leo, elephant Loxodonta spp., rhinoceros Dieros bicornis or Ceratotherium simum, leopard Panthera pardus and Cape buffalo Syncerus caffer (Nelson, 2010). However, this conception of ecotourist preferences hearkens back to a time when hunting was the primary motivation to visit African PAs (Draper, 2005). Recent research has confirmed the continued importance of elephants and lions in attracting ecotourists (Naidoo et al., 2016). But other species may also attract: An observational study in a Kenyan PA showed that cheetahs Acinonyx jubatus were surrounded by the largest number of stopped tourist vehicles (lions, elephants and multiple other species were also popular; Okello, Manka & D'Amour, 2008). Social media studies have also shown that less iconic species may attract tourists (Hausmann et al., 2016;Hausmann et al., 2017). Beyond mammals, stated preference surveys of visitors to four South African PAs showed that scenery and bird diversity were also important to tourists (Lindsey et al., 2007).
These studies suggest that ecotourist preferences may extend beyond the Big Five to birds and other species (Skibins, Powell & Hallo, 2016;Arbieu et al., 2018). However, only a handful of PAs were represented in these studies; their findings may not be generalizable to the rest of the continent. Moreover, stated preferences are often overstated, creating biased measures of relative value (Quaife et al., 2018). Furthermore, measures of how much time ecotourists spend watching a given species, or how many photos people take of a given species may not be indicative of actual ecotourist preferences (since they may have more to do with which species ecotourists happen to encounter, rather than which species ecotourists desired to encounter). Deciphering tourists' preferences for biodiversity, therefore, requires a broader analysis of ecotourist visits across Africa. Newly available tourist visit datasets (Balmford et al., 2015;Naidoo et al., 2016) could enable extensive empirical investigations of ecotourist preferences as revealed by people's actual choices about where to visit (i.e. revealed preferences). Specifically, examining what parks people visit, and what sorts of attributes those parks have, might reveal the tourist preferences that shape decisions. Furthermore, because such analyses account for actual choices that many people made about where to visit, rather than just statements, they may serve as an important complement to more local observational studies and stated preference studies. Such analyses may be important even if they do not include explicit cost or price data (which provide a clearer signal of a person's magnitude of preference, but aggregate values also can skew findings toward the wealthiest people and make benefits to individuals hard to interpret without relating an individual's expenditures to their income or net worth).
Here, we ask two questions using the information on tourism visits to parks in Africa. First, do ecotourists prefer parks with biodiversity beyond the Big Five, particularly those containing high bird diversity (Question 1; Fig. 1)? We address this question by testing whether bird and megafaunal diversity explain visits to PAs within a Bayesian model that controls for other potential drivers of ecotourism, including landscape and human geographic variables that other studies have suggested are important determinants of PA visitation. We expect that bird and mammal richness will be the key components of the revealed preferences of ecotourists.
However, visitation datasets may reveal more about ecotourism than preferences. Empirical park visit datasets could also show which species and landscapes most benefit from ecotourism funds, where visits are a proxy for funds (Higginbottom & Tribe, 2004). For example, if most ecotourists are going to parks that have species A and very few are going to parks that have species B, then species A has the potential to be conserved through ecotourism, while species B may require alternative conservation initiatives. Importantly, these inferences about A and B are true regardless of the preferences that actually direct ecotourist visits. All that is relevant is the correlation between parks that are frequently visited and parks that contain a given species or landscape. Indeed, species that are preferred by ecotourists may not necessarily be well suited to benefit from ecotourist funds (and vice versa), due to multicollinearity between species and landscape variables. For example, if tourists prefer accessible parks and African wild dogs mostly occur in inaccessible parks, then even if African wild dogs are preferred by tourists, they may receive relatively little funds from ecotourists. Consequently, identifying which species and landscapes ecotourism is best suited to conserveregardless of ecotourist preferencesrequires additional analysis of correlations between visits and different species/landscapes (Fig. 1).
Therefore, we also ask, which species and landscapes benefit most from ecotourism at the continental scale (Question 2; Fig. 1)? However, instead of using a single multivariable model, for revealed preferences, we construct a univariate model for each species/landscape feature. These models enable us to understand how well-suited ecotourism is to conserving each species/landscape, regardless of any collinearity. We expect that some species/landscapes may be negatively associated with ecotourist visits, even if our revealed preference analysis shows that tourists prefer that species/landscape. These two questions provide a complimentary analysis of both the drivers and consequences of African ecotourism.

Materials and methods
We began by gathering data on park extents, tourist visits to PAs, animal distributions and human and geographic variables. Next, we spatially joined these datasets so that data were associated with each park. To investigate ecotourist preferences (Question 1), we constructed a Bayesian model   of ecotourist preferences for biodiversity, while controlling for other park attributes (see Fig. 1). Conversely, to investigate what species and landscapes most benefit from ecotourism (Question 2), we constructed 14 univariate Bayesian models of the distribution of ecotourist visits; these models are not meant to explain tourist visits or control for any variables, but to explore the correlations between visit and different park attributes (see Fig. 1). Our investigation of ecotourist visits reflects our underlying assumptions about human action, consistent with the Economic Needs and Independent Structure metatheories, with discussion of the Top-Down metatheory; in combination, our two questions reflect the Interdependent metatheory (Eyster et al., 2022).

Park and tourist data
We gathered the spatial extent and location of 164 PAs in 25 Sub-Saharan continental African countries from The world database on protected areas (WDPA) (2013). We only included PAs classified in IUCN Category II-IV because Category I PAs usually formally exclude tourists. For each PA, we obtained average annual tourist visits between 1998 and 2007 from Balmford et al. (2015) and Naidoo et al. (2016); such data had been collected through a variety of methods, including targeted studies, gate receipts and automated trail and road counts. PAs without visitation data were omitted. Visitation data for the same set of specific years does not exist for most parks (i.e., panel data), so average annual visit data were used.

Biodiversity data
We selected nine megafauna taxa, including the Big Five: lion Panthera leo, elephant Loxodonta spp., leopard Panthera pardus, Cape buffalo Syncerus caffer, giraffe Giraffa spp., cheetah Acinonyx jubatus, African wild dog Lycaon pictus, black rhinoceros Diceros bicornis and white rhinoceros Ceratotherius simum (Nelson, 2010). We included species beyond the Big Five because we hypothesized that the Big Five may not encapsulate the species that are most attractive. We used the following information to select the taxa that may be most attractive to tourists: park and tourism websites, personal experience, personal conversations, availability of distribution data, and published literature (e.g., Lindsey et al., 2007). Given the wide geographic extent of our study and the diversity of ecotourists, we acknowledge that these decisions were subjective and reflect the authors' positionalities. The third author has worked extensively in African PAs for the last 20 years, with a primary focus in southern Africa and a secondary focus in East Africa. This knowledge and experience informed the subjective decisions about what species to include. Future studies might build on our study by including even more mammals.
We gathered presence/absence distributions of each species across Sub-Saharan Africa from existing datasets because comprehensive density data do not exist across the range of parks where they are present. These existing datasets were created by a variety of methods, including expert elicitation, observational surveys, community scientist observations, etc. Distribution data for African wild dogs, leopard, Cape buffalo and cheetah were taken from the IUCN (2008). Black and white rhinoceros' data at the country level were taken from IUCN (2008), and were resolved to park level using park websites and other gray literature. Elephant distribution data were taken from Blanc et al. (2013) and lion data were taken from Bauer et al. (2015) and Riggio et al. (2013) via Naidoo et al. (2016). Giraffe data were provided by Stephanie Fennesey (Giraffe Conservation Foundation, unpublished). All giraffe species were lumped into a single variable. In addition to these charismatic mammals, we examined bird species richness (maximum number of species in each PA), using data extracted from Jenkins, Pimm & Joppa (2013, based on community science data, surveys and a wide variety of other census and observation methods) (see Supporting Information, Figure S7). We used bird species richness instead of other bird metrics because richness is relevant for species conservation, mega-diverse regions are important to birders who are trying to increase their life lists and richness could be easily calculated for each park; future studies might examine specific iconic bird species.

Landscape and human data
To account for non-animal drivers of tourism, we included variables that have previously been shown to be important in explaining PA visits (Balmford et al., 2015;Naidoo et al., 2016). Specifically, we included four variables related to humans and landscapes: whether a park was more than 50% forested (binary variable; forests may hinder wildlife viewing: Naidoo et al., 2016), park accessibility (measured in minutes of travel time from the nearest city with more than 50 000 inhabitants), real national income (2006 US$, adjusted by purchasing power parity) and local population size (number of people living within 100 km of the edge of each park). Previous studies have shown that the local population can lead to increased tourism due to local visitation (Brainard, 2001;Sen et al., 2012;Ghermandi & Nunes, 2013). We measured local population size as the population within 100 km because previous work has shown that this distance captures this effect (Balmford et al., 2015). PA size was not included, since it has previously been shown to be unimportant for explaining African ecotourism (Balmford et al., 2015). PA accessibility and population size were taken from Balmford et al. (2015) and Naidoo et al. (2016) based on data from UN FAO (2005); national income was taken from Euromonitor International (2011) via Naidoo et al. (2016). We chose these datasets because they provided estimates for all of our sites and are contemporaneous with the visitation data. Forest cover data based on satellite remote sensing were taken from the global vegetation cover dataset GLOBCOVER (Bontemps et al., 2010) via Naidoo et al. (2016). These data are based on independent data gathering efforts reflecting particular landscape and biodiversity classifications, and so may not necessarily represent the experiences of tourists (Tobler, 1963;Drucker, 2009), or the imagined vacation at a PA (Bergmann & Lally, 2020

Spatial processing
The presence/absence of each of the nine charismatic megafaunas at each park was determined by overlapping the park extents with each species' distribution (except for rhinoceroses; see above). If any part of a species' range overlapped the park, then the species was marked as present. We marked species as present even if they were primarily nocturnal or occurred at low density in a given park since tour operators and parks still often highlight the possibility of observing such species. In contrast, we marked species as absent for the preference analysis if they were either believed to have been extirpated or if they only occurred in sections of the park that were closed to visitors. However, species that only occurred in sections of the park that were closed to visitors were marked as present for the univariate model analysis (since visits may still benefit these unobservable species).
Bird richness was calculated for each park by overlaying the bird richness raster dataset (Jenkins et al., 2013) on top of each park's extant. The maximum bird species richness value within each park was used for further analysis. Some PAs were too small to automatically extract raster values, so extraction was done manually, including for Mgahinga Gorilla (Uganda), Bobiri (Ghana), Bontebok (South Africa), Wilderness National Lakes Area (South Africa), Nairobi (Kenya), Shai Hills (Ghana), Simien (Ethiopia) and Kalamaloue (Cameroon).
Spatial processing was carried out in ArcGIS version 10.5, QGIS version 3.16.3 and R version 4.0.3 (R Development Core Team, 2020).

Statistical analysis
National income, inaccessibility and local human population were highly skewed so we log-transformed these variables. The response variable, average annual visits, was also highly skewed so we also log-transformed this variable to satisfy regression assumptions associated with the normal error distribution. To prevent undefined value creation, average annual visit values were increased by 1 before logtransformation. Finally, each continuous input variable was centered and scaled by 2 standard deviations; each binary variable was centered (following Gelman, 2008).

Modeling ecotourist preferences: Question 1
High collinearity in the park occupancy of many of the taxa (Supporting Information, Figure S1) prevented us from quantifying preferences for each mammal species separately. Instead, we summed the presence/absence of each of the nine megafaunas in each park (we also explored using Bayesian leave-one-out cross-validation to deal with the collinearity in our variablessee Supporting Information for details). We found that the resulting megafauna richness variable was not highly correlated (r < 0.3) with bird richness or the four control variables (income, inaccessibility, local human population and forest). Consequently, we modeled tourist preferences as depending on bird and megafauna richness, while controlling for effects of national income, inaccessibility, local human population and forest according to the following equation: where PAs are indexed by i, μ is the mean and σ is the standard deviation of a normal distribution and where α is the intercept coefficient and β are the slope coefficients. We did not include country as a random effect but instead attempted to directly account for differences using, for example, national income. We used this model to infer what species and landscapes are likely important for determining tourist preferences, but note that this analysis cannot infer causation between the presence of different species/landscapes and tourist visits. We used the median parameter estimates from this model to the additional ecotourist visits attributable to additional bird richness.

Modeling the distribution of ecotourist benefits: Question 2
To test which species (e.g., lions) and landscapes (e.g., inaccessible areas) are conserved by current levels of ecotourism (where ecotourist visits are a proxy for conservation funds; Higginbottom & Tribe, 2004), we built Bayesian univariate normal models for each of the 14 park features, including each of the nine megafauna taxa, bird richness, national income, local population, inaccessibility and forest. These univariate models allowed us to examine how each of the nine megafauna taxa was independently related to tourism visits, rather than as part of an aggregate index as in the multivariable model. The multivariable model for Question 1 used an explanatory modeling approach (Shmueli, 2010) to explain the drivers of ecotourist visits and attempts to isolate which factors are most important for determining what parks people visit. In contrast, our second question used univariate models and a predictive modeling approach (Shmueli, 2010) to predict the potential consequences of ecotourism on various species and landscapes, even those that are collinear. Thus, these univariate models are not intended to explain visitor numbers, understand the importance of different variables to tourists or measure tourist preferences. Instead, these models were merely meant to predict what sorts of species and landscapes tourism might benefit. Models were of the form (model for lion described here): where PAs are indexed by i, μ is the mean and σ is the standard deviation of a normal distribution and where α is the intercept coefficient and β is the slope coefficient. Note that equations (1 and 3) are identical. All models were built with Stan (Carpenter et al., 2017) using brms version 2.14.8 (Bürkner, 2018) in R version 4.0.3 (R Development Core Team, 2020). All models were estimated using four chains, each with 2000 iterations (with half devoted to warm-up) and weakly informative priors. Chain convergence was confirmed using R̂< 1.01 (Vehtari et al., 2019). We verified model fits using posterior predictive checks; we explored using zero-inflated negative binomial distributions, but we found that the normal distribution model was superior at recovering the relationships in the data (see posterior predictive check analysis in Supporting Information, Figure S5). Consequently, we relied on normal distributions for this analysis. We assessed the significance of slope coefficients using 89% credible intervals. Our hypotheses, our interpretation of these data and our choices about what data to include likely reflect our own positionalities (see, e.g., our decisions about what mammals to include, above) (Montana et al., 2020;Pascual et al., 2021). The first author is a White settler in the USA and Canada. Whether in his backyard or far away from home, experiences with birds stand out in his memory. The salience of these memories shaped the direction of this study and the hypothesis that birds may be more important than presently regarded in the literature on ecotourism in African PAs.

Results
Average annual visits to PAs ranged from 0 to nearly 1.5 million, with visitation highest in southern and eastern Africa, at least among our samples' parks (Fig. 2). The richness of our nine study mammals in PAs ranged from 0-9 (Fig. 3), while bird richness ranged from 178-655 species (Fig. 4). White and black rhinoceroses were present at the fewest of the 164 PAs, while leopard and elephant were present at the most (see density plots in diagonal of Supporting Information, Figure S1). Most of the PAs were largely unforested (Supporting Information, Figure S1).
In our model of revealed preferences of ecotourists (addressing Question 1), bird and megafauna richness were significant predictors of ecotourist visits after controlling for the forest, national income, human population and inaccessibility (for a depiction of this result and the associated uncertainty, see Figs 5 and 6; Supporting Information, Table S1). Of these control variables, national income had a positive effect on ecotourist visits, while forest and the local human population had negative effects; inaccessibility did not have a significant effect (Fig. 5; Supporting Information, Table S1). In our univariate models of the distribution of ecotourist benefits (Question 2), we found significant positive associations between ecotourist visits and black rhinoceros, white rhinoceros, cheetah, giraffe, bird richness, African wild dog, lion and national income (for a depiction of this result and the associated uncertainty, see Fig. 7; see also Supporting Information, Tables S3, S4 and S8-S13). In contrast, ecotourist visits were significantly negatively associated with the local human population, inaccessibility and forest ( Fig. 7; Supporting Information, Tables S14-S16). Leopard, elephant and Cape buffalo were not significantly associated (89% credible intervals included 0) with ecotourist visits (Fig. 7; Supporting Information, Tables S5-S7).

Discussion
Our results suggest that all else equal, ecotourists likely prefer to visit PAs with high bird diversity, in addition to high megafauna diversity (Figs 5 and 6). These findings extend the growing evidence from social media, stated preferences and in-park observations that ecotourist preferences currently extend beyond the Big Five (Lindsey et al., 2007;Okello et al., 2008;Hausmann et al., 2016;Hausmann et al., 2017). We also found that people likely prefer to visit unforested parks, perhaps because forests mask animal visibility (Naidoo et al., 2016). Meanwhile, our univariate model results suggest that ecotourism may be suited to conserve bird diversity and some megafauna (lion, both rhinoceros, cheetah, African wild dog and giraffe). Ecotourist funds are more likely to go to parks in wealthy countries, but less likely to go to those that are forested, near large human populations or inaccessible.

Ecotourist preferences
Bird diversity may be an important component of ecotourist preferences (Fig. 6). Based on empirical data from 164 parks, our results generalize Lindsey et al.'s (2007) finding that bird diversity is part of tourist preferences in four South African PAs. The strength of the relationship (Fig. 5), the inclusion of controls in our model and the consistency with prior stated preference studies suggest that ecotourists prefer bird diverse parks. However, the lack of panel data prevents us from ruling out other explanations. Specifically, the effect of bird diversity (for example) may be confounded by an unmeasured park attribute. For instance, perhaps ecotourists have preferences for a particular type of landscape that also coincidentally supports higher bird diversity. Thus, by choosing to visit a park with that landscape, ecotourists may only indirectly choose to visit a park with high bird diversity. For example, it is possible that distributions of species and visits are both influenced by human warfare (Daskin & Pringle, 2018). The apparent preference for bird and megafauna diversity observed in our study adds to the growing literature demonstrating the attraction of biodiversity. For example, research in cities has shown that people prefer to live in neighborhoods with higher biodiversity (the 'luxury effect'; Melles, 2005;Leong, Dunn & Trautwein, 2018). Spanning the idealized nature of wildlife safaris, to the quotidian nature in one's front lawn, these and other studies (e.g., Millenium Ecosystem Assessment, 2005;Boeri et al., 2020) suggest that biodiversity is attractive in many disparate situations. This importance has implications for conservation marketing and planning: conservation campaigns that communicate about and target biodiversity more broadly might be more successful, both in garnering support and in satisfying constituents. For instance, perhaps if a conservation organization, agency or park prioritized conserving a biodiverse area over an area with a charismatic megafauna species, and highlighted the biodiversity of said area, it may get more public support and ultimately be more successful.
Our findings show the utility of large visitation datasets for understanding the attractiveness of various park attributes. However, these visit data do have several limitations. First, they are not individual specific, so our models are not able to explain how preferences differ between individuals. Second, the origins of each visitor are unknown. This lack of data prevented us from more explicitly accounting for the travel costs to various park destinations, forcing us to instead account for costs via using more general proxies such as inaccessibility and national income. Third, this dataset does not include information on how much money each tourist spends, which prevents direct analysis of associations between tourist visits and money spent. Future surveys of visitors, including the origin and individualspecific data, might complement the tourism dataset used here. Future studies might also analyze only those parks that have accurate visitor information collected every year, thereby enabling an understanding of how preferences may evolve over time in response to the variables we considered.

Distribution of ecotourist benefits
Our results suggest that ecotourism provides funds to conserve various taxa and landscapes, including bird diversity (Fig. 7). This may mean that ecotourism contributes to the sustenance of highly bird-diverse landscapes across Sub-Saharan Africa. However, the negative association between ecotourism and forested parks indicates that birds most at risk of extinctionforest birds (Wotton et al., 2017) may benefit less from ecotourism (Fig. 7). Future research could test this further by examining relationships between ecotourist visits and threatened birds.
In addition to forest birds, some mammals may benefit less from the bulk of ecotourist funds. Leopard, elephant and Cape buffalo all displayed only haphazard relationships with tourist visits, suggesting that ecotourism should be but one component  Table S1 for intercepts and model details.  437 of a holistic conservation program. The relative commonness of these three species (Supporting Information, Figure S1) may also have contributed to this haphazard relationship. Like forest birds, forest elephants are experiencing sharp declines (Wittemyer et al., 2014). Perhaps conservation programs that do not rely on ecotourists could be particularly important for conserving forest species. Our findings may help inform conservation initiatives, particularly at the continent scale. However, these results should be interpreted with caution because our PA sample, including only those with visitor data, may not be representative of all PAs across Sub-Saharan Africa. More specifically, for example, our univariate models estimated that PAs with black rhinoceros are likely to receive 7.5 × 10 4 more tourist visits than those without. In contrast, forested PAs are likely to receive 1081 fewer tourist visits than those that are unforested. These estimates suggest that species like black rhinoceros are likely to occur in parks that are heavily visited. We, therefore, predict that ecotourists and the funds they bring may be well suited to conserve black rhinoceros. In contrast, forests are less likely to occur in heavily visited parks, and so we predict that forested landscapes may not benefit from ecotourists and the funds they bring; instead, forests may require other conservation tools. Our univariate results, therefore, demonstrate how well-suited ecotourism is for funding the conservation of particular species and landscapes at the continent scale. However, these species that present the most potential to benefit from ecotourism may also be the most likely to be negatively impacted by human disturbance and tourist infrastructure development (Steven, Pickering & Castley, 2011). Future research might focus on these species when looking for possible negative effects of ecotourism.
Our analysis assumed that tourist visits are associated with conservation investment. Some nations have fee-sharing systems, which complicates this assumption. However, studies show that the bulk of entrance fees are reinvested into the park in question. For example, Chidakel, Eb & Child (2020) found that 81% of revenue generated at Kruger National Parkone of South Africa's most visited National Parkswas invested back into the park for tourism and ecological management. Visits to parks also encourage foreign investment and much tourist money stays within local areas, incentivizing the maintenance of parks that are attractive to tourists (World Bank, 2011). Furthermore, lodging, meals, souvenirs, etc. would likely be bought and paid locally. Thus, while tourist visits may benefit other parks within a country or area, much of the benefits likely accrue to the visited park. Nonetheless, the COVID-19 pandemic has shown the inadequacy of relying chiefly on tourism to support conservation (Lindsey et al., 2020;Mitchell & Philips, 2021;Moore & Hopkins, 2021;Waithaka et al., 2021), even for those species that occur in heavily visited parks. Developing alternative funding mechanisms will be key for maintaining biodiversity in the face of public health crises.
While we have explicated how ecotourist visits might help conserve various species and landscapes (Fig. 7), future research might analyze the impacts of these ecotourist-landscape relationships on local sovereignty and neocolonialism. Specifically, these relationships could further be interpreted as a map of the landscapes and species that receive the largest neocolonial impacts from wealthy travelers (Devine, 2017;Wondirad, Tolkach & King, 2019). That is, the very mechanism by which ecotourism can aid conservationthe provision of external fundsmight also constrain local sovereignty and hinder local visions for sustainability and conservation (Ojeda, 2012). For example, our finding of a negative association between tourist visits and nearby human population (Fig. 7) could mean that ecotourism funds the maintenance of parks in places where people prefer not to live, but it could mean that tourist funds incentivize the displacement of local communities and the conservation of landscapes that are not locally desirable (analogously to green gentrification in cities; Anguelovski, 2018). While comprehensive multinational data have shown that on average certain types of PAs have positive effects on the well-being of nearby residents (Naidoo, 2019), localized case studies can demonstrate mixed or negative effects on communities (Lepp & Holland, 2006;Kibicho, 2008;Appiah-Opoku, 2011). These negative effects often stem from physical and governance exclusion from parks (Neumann, 1998(Neumann, , 2001Brockington, 2002;Nelson, 2003). For example, when Uganda's Kibale national park was created in 1993, people were evicted (sometimes forcefully; Government of Uganda, 1992, cited in Lepp & Holland, 2006) and the park became 'out of bounds' for residents (Lepp, 2004). Future research might therefore help clarify relationships between ecotourist preferences and those of local community members across Sub-Saharan Africa.

Multivariable and univariate analyses complement each other
Our study showed how a multivariable model of visitation data and animal/landscape features can elucidate revealed preferences. Such ecotourist preference are typically used to represent relationships between ecotourists and species/landscapes (e.g., Park, Bowker & Leeworthy, 2002). However, ecotourist preferences only represent one facet of ecotourist-ecosystem relationships. Investigating the other facetthat is, which species and landscapes are correlated with tourist visitsis necessary for understanding what types of ecosystems ecotourism will promote (via transfer of funds from ecotourists to park planners, etc.). While univariate models may be dismissed as overly simplistic (Haslam & McGarty, 2001), our results showed how combining univariate and multivariable models enabled the investigation of both facets of ecotourist-ecosystem relationships. For example, our multivariable models showed that ecotourists have preferences for parks with diverse megafauna, while our univariate analysis revealed that some megafauna species and landscapes are not positively associated with tourist visits; these species and landscapes may require conservation techniques beyond ecotourism.
As the rise of social media creates abundant and accessible visitation data (da Mota & Pickering, 2020), revealed preference analyses will also become increasingly feasible. We suggest that future research leverage this growing trove of visitation data by complementing multivariable analyses of revealed preferences with univariate analyses of the distribution of ecotourist benefits.