Socioeconomic correlates of global mammalian conservation status

The main causes of biodiversity decline are related to human use of resources, which is ultimately triggered by the socioeconomic decisions made by individuals and nations. Characterizing the socioeconomic attributes of areas in which biodiversity is most threatened can help us identify decisions and conditions that promote the presence or absence of threats and potentially suggest more sustainable strategies. In this study we explored how diverse indicators of social and economic development correlate with the conservation status of terrestrial mammals within countries explicitly exploring hypothesized linear and quadratic relationships. First, comparing countries with and without threatened mammals we found that those without threatened species are a disparate group formed by European countries and Small Island Developing States (SIDS) with little in common besides their slow population growth and a past of human impacts. Second, focusing on countries with threatened mammals we found that those with a more threatened mammalian biota have mainly rural populations, are predominantly exporters of goods and services, receive low to intermediate economic benefits from international tourism, and have medium to high human life expectancy. Overall, these results provide a comprehensive characterization of the socioeconomic profiles linked to mammalian conservation status of the world’s nations, highlighting the importance of transborder impacts reflected by the international flux of goods, services and people. Further studies would be necessary to unravel the actual mechanisms and threats that link these socioeconomic profiles and indicators with mammalian conservation. Nevertheless, this study presents a broad and complete characterization that offers testable hypotheses regarding how socioeconomic development associates with biodiversity.


INTRODUCTION
Biodiversity loss has accelerated in recent times and many voices argue that we may be entering the Earth's 6th mass extinction event (Barnosky et al. 2011). The main threats to biodiversity are human-induced and include habitat loss, fragmentation, overexploitation, spread of exotic species and diseases, pollution, and climate change (Soulé 1991, MA 2005. Understanding why distinct species and sites are vulnerable to extinction is essential to reduce biodiversity losses occurring now and those that will likely occur in the future (Hoffmann et al. 2010).
Comparative studies of extinction risk have focused on identifying differences in vulnerability at the species (or taxonomic group) level. These studies have associated species' vulnerability with a diversity of life-history and ecolog-ical factors, such as body size, geographic range size, ecological and social specialization, and phylogenetic-lineage in mammals (Cardillo et al. 2008, Davidson et al. 2009, González-Suárez and Revilla 2013 and other taxa (Webb et al. 2002, Cushman 2006. However, species-focused studies have been criticized for their lack of applicability to management and for ignoring the role of distinct threats (Cardillo and Meijaard 2012, Murray et al. 2014; but see Owens andBennett 2000, González-Suárez andRevilla 2014). An attempt to overcome these weaknesses has been to explore human activities (which are potentially manageable) occurring within each species' geographic ranges. Studies using this approach have found that more endangered species tend to overlap with mosaic villages and residential croplands, densely populated areas or with increasing human population growth (Harcourt andParks 2003, Pekin andPijanowski 2012).
Species' intrinsic traits make some taxa more vulnerable to extinction, but also there are inherent properties associated with particular areas that make them more likely to harbor higher numbers of threatened species. The number of threatened species on a given site directly depends on the total species richness (how many species actually occupy that area) and the threats that affect those species (Fig. 1;Essl et al. 2013). In turn, species richness is determined by historical, biogeographical and environmental conditions, as well as by human activities that may have caused past local extinctions. Threats can include natural hazards (influenced by environmental conditions, Fig. 1), but today the key threats for most species are of anthropogenic origin. For example, natural processes such as volcanoes, avalanches, or earthquakes are only considered to represent a threat for 1% of the 2551 terrestrial mammals with described threats, and these species are also at risk from anthropogenic activities (IUCN 2012a, González-Suárez andRevilla 2014). Humanrelated threats are associated with human activities which are mainly determined by socioeconomic development. Although analyzing the causes leading to observed development is beyond the scope of this study, it is worth noting v www.esajournals.org that development is often influenced by environmental conditions (which in turn can be modified by development) and by the biogeographic history of a given area, both of which also influence its natural biodiversity. Eventually, if conditions change or nothing is done to stop it, threatened species become extinct (Fig. 1).
Focusing on the spatial distribution of human threats, many studies have identified sensitive areas based on the correlation of different human pressure indicators and different measurements of biodiversity status. Since habitat destruction is one of the main causes of biodiversity decline, measurements of human land use are those most-commonly employed to quantify human impact (Pimm and Raven 2000). In particular, conversion to arable land appears to be a key factor associated with greater risk at a regional and global scale (Kerr and Cihlar 2004, Lenzen et al. 2009, Lotz and Allen 2013. Another widely used indicator of human pressure is human population density, an aggregated proxy which has been positively correlated with abundance of threatened species at different scales (Burgess et al. 2007, Luck 2007. Within countries, some studies have shown that economic growth, energy use, human birth rate and different measures of national income or income inequality are associated with the number of threatened species (Kerr and Currie 1995, Naidoo and Adamowicz 2001, Hoffmann 2004, Holland et al. 2009) and with other measures of environmental damage (Grossman andKrueger 1995, Bradshaw et al. 2010). More recently, transborder impacts have also been suggested as risk factors found to be associated with the conservation status of the biota in developing countries (Meyfroidt et al. 2010, Lenzen et al. 2012. While providing some insights, most of these past studies have only evaluated a few indicators at a time (but see Hoffmann 2004, Lotz and Allen 2013), without taking into account the many diverse aspects that describe socioeconomic development. This diversity is reflected, for example, in the 800 indicators included within the World Development Indicators book (World Bank Group 2005). One reason why past studies have focused on few indicators is that aside from population data, land use cover, and a few derived economic metrics, global socioeconomic data are only available at the country level, especially for indicators related to the trade of goods and services, and human life quality (CIESIN 2005, Nordhaus 2006, Asselen and Verburg 2012. Socioeconomic data do exist for finer political units (e.g., counties, states) in some areas, but in many cases data are not gathered or made public at such fine scales. Therefore, in order to conduct a global study that captures the diverse aspects of socioeconomic development, using country resolution is the most feasible solution. Country-based analyses are also relevant for management and policy implementation because political decisions influencing biodiversity at large scales are usually enacted at this level (Forester andMachlis 1996, Chape et al. 2005). Results from these analyses can also be helpful to make countries aware of how their political, social and economic decisions may be influencing the conservation status of their biodiversity.
Here we present a comprehensive global analysis aimed to identify which indicators of socioeconomic development correlate with terrestrial mammalian conservation status at the country level. To explicitly include the diverse facets of socioeconomic development we considered indicators representing nine distinct categories defined by the World Bank (Appendix A: Table A1) including those which have been previously linked to biodiversity status within countries (see references above). Some of the explored indicators can be more directly associated with threats (e.g., percentage of arable land) or benefits (e.g., number of protected areas) to biodiversity, while others are aggregated descriptors of development (e.g., human population density) which may capture more complex or indirect associations between socioeconomic processes, threats and biodiversity. Exploring this broad suite of indicators we first identified the socioeconomic characteristics that differentiate countries with and without threatened mammals. Then, considering only countries with threatened mammals, we determined which indicators are associated with higher relative richness of threatened mammals. In both sets of analyses we explicitly explored relationships proposed by the two general hypotheses relating socioeconomic development and environmental v www.esajournals.org damage. The first hypothesis proposes a linear response, as human populations increase and become more industrialized the damages to biodiversity increase, with the greatest impacts associated with the most developed areas (e.g., Hettige et al. 2000, Clausen andYork 2008). The second hypothesis proposes a quadratic relationship in which the greatest impacts occur at intermediate stages of development (inverted-U-shaped, or environmental Kuznets curve as defined by Grossman and Krueger [1995]). Initially, population growth and industrialization would be associated with increased damages to biodiversity, but as societies become more technological and educated, they would also become more environmentally concerned and reduce their impact (McPherson and Nieswiadomy 2005). Because we explored different relationships and a broad range of indicators, our results present a new global and comprehensive characterization of the key socioeconomic correlates of mammalian conservation status.

Socioeconomic indicators and species data
We used socioeconomic indicators compiled by the World Bank from different official sources grouped according to these simplified thematic categories: agriculture, economy, education, environment, health, infrastructure, labor and social protection, population and private sector (Appendix A: Table A1). These categories were used in the analyses as non-redundant blocks, as explained below. Relative indicators (percentages and per capita values) were chosen over absolute values to facilitate comparison among nations. We used information from the year with the most available data in the last decade (year 2005) and excluded indicators considered a priori as relevant but with data available for ,70% of the 204 countries in our database (9 indicators out of 39; Appendix A: Table A2). As a result, no indicators from the categories education and infrastructure were included in our analyses. We did not use data imputation techniques for missing socioeconomic data because these data are not missing at random (e.g., more developed countries are more likely to have data on their development status), and the mechanisms by which data are missing can be complex and are not easily modeled (van Buuren 2012).
To assess the conservation status of terrestrial mammals we used the IUCN Red List of Threatened Species version 2012.1 (IUCN 2012b) which provides a single, global status for each species. Species defined as vulnerable, endangered or critically endangered are considered as threatened; whereas least concern or near threatened species are non-threatened. The 75 mammals classified as extinct in the wild or extinct were not included in the analyses since they cannot be classified in either category. Data deficient (DD) species were initially classified as non-threatened to define a conservative minimum estimate of threatened species per country. We then repeated the analyses considering DD species as threatened, and thus defining a maximum estimate of threatened species per country. Mammal presence within each country was determined using spatial data on the current ( post 1980) global distribution of mammal species available from the IUCN (IUCN 2012b) selecting only native areas, with presence defined as extant or probably extant. We used a World Cylindrical Equal Area projection in ArcGIS 10 (ESRI 2011) and intersected species ranges (N ¼ 5014) with a current global political map. All species with any portion of their range within the boundaries of a country were defined as present in that country.
Our approach to define the number of threatened species per country assumes that the global status of a species is potentially affected by human activities within each of the countries the species occupies. Ideally, mammalian status would have been defined using national assessments (to compare national socioeconomic development and status); however, this is not possible at a global scale. National assessments are not currently available for many countries and those available do not always follow standardized criteria, which prevents comparison. For example, only 23 countries have a National Red List according to the IUCN (http://www.nationalredlist.org/) and many include only partial assessments (Appendix B:  Tables B1 and B2). Finally, we feel that the use of the global Red List to assess status is warranted if we consider that the responsibility of maintaining/menacing species should be shared by all countries that harbor them. v www.esajournals.org Analyses First, we explored which socioeconomic indicators are associated with the presence (vs. absence) of threatened mammals using generalized linear models (GLMs). GLMs were fitted with the glm procedure in R (R Development Core Team 2011) using a binomial family and a logit link. Second, considering countries with !1 threatened mammal we explored which indicators are associated the number of threatened species using the glm.nb procedure in R (package MASS; Venables and Ripley 2002) with a negative binomial family and a log link. For both questions we tested linear and quadratic relationships for all indicators to account for the two main hypotheses mentioned in the introduction.
The variable selection procedure was the same for both analyses. First, we calculated pairwise Spearman's coefficients (q) and from any pair of highly correlated indicators (q. j0.8j) we excluded the indicator with fewer data available (7 excluded, out of 30; Appendix A: Tables A2-A4). Using non-highly correlated indicators, we followed Purvis et al. (2000) to define a minimum adequate model (MAM-based approach) for each socioeconomic category. We used this approach to maximize the use of available data, as some countries have available data for some indicators but not others and generating a complete dataset (removing all cases with any missing data) to analyze all categories at once would greatly limit the available sample size. MAMs by category were defined by starting with a full model including all indicators in the category from which the least significant variable was removed (one at a time), and then a new model (potentially with a different sample size) was fitted and evaluated. After finding a model containing only significant variables (using a conservative p value ¼ 0.10) we retested the significance of previously removed variables and defined a MAM by category including any additional significant factors. Second, and in order to evaluate more systematically the relevance of the socioeconomic indicators, we determined indicator importance using an AIC c -based approach. For this approach we were limited to the subset of countries with available data for all selected indicators in the category. We estimated variable importance for each socioeconomic indicator based on variable weights (Burnham and Anderson 2002) calculat-ed from all possible model subsets in each category using the importance function in the MuMIn R package (Barton 2013). We considered indicators were supported if their variable weights .0.7. The final category model was defined using all variables included in the category MAM plus any additional variables identified as supported with the AIC c -based approach. All variables included in the final category models were used to define a final global model using the same variable selection approaches (MAM-and AIC c -based).
Model fit was estimated as the percentage of deviance explained. For the binomial model (presence of threatened mammals) we also calculated model sensitivity (true positive rate) and specificity (true negative rate; Allouche et al. 2006); setting an arbitrary 0.5 cut-off probability to define presences and absences. Furthermore, the area under the receiver operating characteristic curve (AUC) was calculated as a threshold-independent measure of model performance (Manel et al. 2001). To evaluate model predictive ability in the negative binomial model (abundance of threatened species) we calculated a standardized prediction error defined as the number of threatened species predicted minus the observed number divided by total number of mammals. Positive errors indicate that the model overestimated the number of threatened mammals, whereas negative values indicate underestimation.
In addition to the tested socioeconomic indicators, all fitted models included as control variables (additional fixed effects) a country's total land area and its total mammalian richness. Including these variables allowed us to effectively model the association of socioeconomic indicators with the presence and number of threatened mammals per country controlling for the known effects of area and richness in the response variable (we can expect more threatened mammals in large countries inhabited by more mammals). An additional control variable, the mean number of mammals shared with neighboring countries (hereafter ''shared mammals''), was included to account for the singularity of a country's fauna considering that nearby countries are generally more alike than those far apart. It is important to note that the number of shared mammals does not represent endemicity per se, but addresses issues of spatial autocorrelation among neighboring countries. It v www.esajournals.org was calculated by identifying the number of species common to all pair of countries that share a border and then calculating the mean value over all neighbors for each country, standardized by the country's total mammalian richness. Harboring a higher percentage of shared mammals implies having a greater abundance of cosmopolitan species, generally less threatened but also potentially exposed to more sources of impact. Because islands have no neighboring countries, by definition they have 0% shared mammals.

RESULTS
From the 204 countries with mammalian distribution range information and socioeconomic data, 168 host at least one globally threatened mammal (median ¼ 6 species, range 1-177), whereas 36 countries have none ( Fig. 2a; Appendix C: Tables C1 and C2). From those 36, four countries contain DD species (median ¼ 1.5, range ¼ 1-4) that could potentially be threatened. Indonesia, Brazil, Mexico and India are the countries with the highest number of threatened mammals (649, 625, 454 and 352 species, respectively, if DD are considered as non-threatened).

Presence of threatened species
We found that diverse socioeconomic indicators are associated with the presence (vs. absence) of threatened mammals. Tourism receipts and urban population exhibit an inverted-U relationship with the probability of having threatened mammals. On the contrary, the percentage of arable land by country relates with the response variable following a positive parabola. Population growth presents a positive linear effect on the probability of containing threatened species by country (Table 1). Countries with no threatened mammals have either high or low percentages of urban population and international tourism receipt values, intermediate percentages of their territory are devoted to arable lands (extreme values are more common in countries with threatened mammals), and exhibit relatively slow population growth rates (Fig. 3). Classifying DD species as threatened did not qualitatively change these results except that the percentage of urban population was no longer a relevant indicator (Appendix D).
The final model was fitted for the 162 countries with data: 135 with and 27 without threatened mammals. The latter group is formed by two distinct types of countries: 15 small islands included in the group of Small Islands Developing States (SIDS, as defined at the UNCED 1992) and 12 European countries (including island-countries Malta and Iceland). The countries with at least one threatened mammal form a more heterogeneous group, which we describe in the next section. The final model provided a good fit to the data, explaining 61.9% of the deviance, with 37.4% explained by the control variables (country land area, total mammalian richness and shared mammals) and 24.5% associated to the four socioeconomic variables identified as relevant. This final model also had high sensitivity (0.964, power to identify positives) and specificity (0.818, power to identify negatives); and excellent overall predictive ability (AUC ¼ 0.968).

Abundance of threatened species
Socioeconomic indicators also correlate with threatened mammal abundance at the country level. International tourism (receipts) and life expectancy indicators follow an inverted-U relationship with the total number of threatened mammals by country, whereas the rest of selected variables linearly correlate with the response variable. In particular, the final model shows that countries with more threatened mammals have lower percentages of urban population, intermediate to high life expectancies, generate fewer imports but more exports of goods and services, and their share in exports due to expenditures by international inbound visitors (international tourism receipts) are low to intermediate (Table  2; Fig. 4). This final model highlights the importance of transborder impacts and included data from 125 countries that have between 1 and 177 threatened mammals (the full range of observed values in the World; Appendix A: Table  A3), and explained 79.8% of the deviance, 72.0% corresponding to the control variables and 7.8% to the selected socioeconomic indicators. Model predictions for each country were generally accurate with only small errors in prediction representing 60.14% of the total mammalian richness of the country. Only six countries were predicted to have considerably more threatened mammals (.0.14%) than those currently listed: Cyprus, Indonesia, Barbados, Seychelles, New v www.esajournals.org  All fitted models (Tables 1 and 2)-evaluating presence and abundance of threatened mammals-include three control variables (total mammalian richness, country land surface and shared mammals). As expected, the presence and abundance of threatened mammals are always positively associated with total mammalian richness. Once richness is taken into account, the total land area does not significantly influence the presence and abundance of threatened mammals. The percentage of shared mammals has no significant effect on the probability of presence of threatened species, but in the abundance model countries with more shared species tend to have fewer threatened mammals.

DISCUSSION
Our results show that both presence and abundance of threatened mammalian species correlate with particular socioeconomic features at a global scale. (Appendix E provides maps representing the observed values per country for all indicators identified as relevant.) While our analyses do not evaluate how these socioeconomic conditions associate with the actual processes and threats that affect mammals, our results offer interesting follow-up questions and hypotheses regarding those aspects of socioeconomic development which could be more influential for mammalian conservation.
Interestingly, our results show two clearly distinct types of countries that lack threatened mammals: SIDS (Small Islands Developing States) and well-developed and relatively small European countries. SIDS have suffered relatively minor changes in land use judging by their low levels of arable land and urban population, but tourism constitutes an important part of their economies (Fig. 3) and also a potential threat to their biodiversity (Gö ssling et al. 2002, McElroy 2003. Although SIDS are characterized by economic and environmental vulnerability (Kier et al. 2009, Teelucksingh andWatson 2013), these small islands have generally low mammalian richness due to their small size and isolation (Whittaker and Fernández-Palacios 2007), thus limiting the number of potential mammals that could be at risk. Small European countries, on the other hand, have higher percentages of arable land and urban population, the result of a history of land transformation that is not reflected in the amount of mammals currently at risk at this scale (Falcucci et al. 2007, Mortelliti et al. 2010. For both groups of countries, the recent record of extinct species (post-15th century) does not seem to explain the absence of threatened species. The amount of extinct and extinct in the wild species reported by the IUCN within these countries is four species in four SIDS (one on each), and one in one European country (Appendix C: Table C1),  9.840) no more than the number of extinct species in other areas. An alternative explanation for the lack of threatened species in these countries could be that their most vulnerable, and probably scarce, mammals became extinct long ago and/or that currently extant species have been extirpated (are locally extinct) from these territories (Ceballos and Ehrlich 2002, Morrison et al. 2007). Additionally, some of these countries have nowadays the resources and will to implement conservation policies to protect their remaining fauna which could reduce the number of species listed as threatened (Pullin et al. 2009). Although a priori we could expect that the lack of threatened mammals would be associated with the ''most pristine'' or ''less humanized'' countries, our results do not reflect that trend. By exploring for the first time the socioeconomic profiles of countries harboring no threatened mammals our study offers new, testable, hypotheses to explain these absences including the effects of increased conservation actions, local extirpations and ancient global extinctions. Among countries with one or more threatened Fig. 3. Observed values for the key socioeconomic variables associated with differences in the probability of presence of threatened mammals for countries with threatened mammals (dark grey bars; present) and countries without threatened mammals (light grey bars; absent). Abbreviations are Isl, subgroup of SIDS; and Eur, subgroup of European countries.
v www.esajournals.org mammals, we identified diverse indicators, with both linear and quadratic relationships, as associated with the number of threatened mammals (Fig. 4). All else being equal, states with a higher proportion of rural population appear to be associated with higher numbers of threatened species, which suggests that more threats could be associated with rural development than with predominantly urban countries. Future research would be necessary to explore this association, but threats associated with nations with higher proportions of rural population are probably related to land transformation for agriculture and the resulting habitat loss for many mammals, as well as side effects of land use intensity such as pollution and exotic species introductions (Laurance et al. 2014). In addition, more urbanized countries could have already lost many of their most vulnerable species and thus could present apparently better conservation status. We found that human life expectancy, an indicator of overall socioeconomic development, is also associated with mammalian conservation status; with intermediate to high life expectancies being associated with more threatened species. This non-linear relationship often described as an environmental Kuznets curve was also reported in a previous study that used another aggregated indicator, per capita income by country, which is highly correlated with life expectancy (McPherson and Nieswiadomy 2005). Finally, an interesting result from our analyses is the identified importance of trade and flux of services, goods and people among countries (Fig. 4), all of them linked to the fast globalization process we are witnessing.
Recent studies suggest that international trade is associated with 30% of global species threats (Lenzen et al. 2012) and some authors have equated the imports of certain goods to the exports of ecological impacts (Meyfroidt et al. 2010). While our results support these ideas, further research would be necessary to assess the actual impacts caused by this trade including conversion of land to exportable key crops (e.g., coffee, soybean, oils, etc), logging, and overhunting for pet trade (Lenzen et al. 2012). In the meantime, given the apparent importance of trade, we propose that land use classifications and assessments of threats should explicitly consider international trade, for example separating land use changes associated with internal production from those destined to exports. In addition to the importance of trade of goods and services we found that international tourism (visiting) is also correlated with the number of threatened mammals but with a perhaps unexpected pattern. Apparently, countries whose economies highly depend on international tourism have fewer threatened mammals than those with intermediate levels. Within this group we can find many SIDS (e.g., Netherlands Antilles, Barbados) which have high levels of tourism but, as explained above, are areas naturally poor in mammals.
By considering a diversity of indicators we also show that neither of the two proposed general hypotheses linking biodiversity and socioeconomic development is consistently supported as both linear and quadratic relationships are observed (Fig. 4). For some indicators our results suggest that the effect of development on biodiversity is non-linear supporting the hypothesis that fewer threatened mammals in more developed countries can be a consequence of the increasing environmental concern and stricter environmental regulations that often accompany socioeconomic development. In other cases, the relationships are linear with more development associated with more threatened species and no subsequent improvement. This diversity of association patterns highlight why using a single development indicator is not advisable (Moran et al. 2008, Nielsen 2011, and also advocates for Our results also deliver a useful message for conservation planning highlighting countries where the observed number of threatened mammals is smaller or greater than expected by their socioeconomic profile. For example countries such as Brazil or United States (countries in green in Fig. 2c) have fewer threatened mammals than predicted perhaps because they have a mechanism that is acting to decrease threats to mammals (such as effective conservation measures), and/or because they are areas naturally occupied by less susceptible species (e.g., more cosmopolitan/resilient mammals). Conversely, countries such as India or Australia (countries in brown in Fig. 2c) harbor more threatened mammals than predicted by their socioeconomic characteristics. In these countries human threats may be especially intense and fast changing (not yet be accounted for in available assessments) Fig. 4. Predicted relationships between key socioeconomic variables and the abundance of threatened mammals by country (DD species classified as non-threatened). Model predictions were based on the final model (Table 2) and estimated by exploring the range of observed values for each indicator while using the median observed value for other variables in the model (Median values: total mammalian richness ¼ 130; land area ¼ 196,800 km 2 ; shared mammals ¼ 0.7646; urban population ¼ 56.20%; international tourism, receipts ¼ 8.56%; imports of goods and services ¼ 42.64%; exports of goods and services ¼ 36.45%; life expectancy ¼ 71.38). Shadowed area represents the confidence intervals (95%). Singapur was removed from graphs C and D to facilitate visualization. Singapur has extremely high values for these two indicators (imports of goods and services (% GDP) ¼ 200,452; exports of goods and services (% GDP) ¼ 228,007). The distribution of the observed data for each indicator is indicated over the x-axis by small bars.
v www.esajournals.org and/or mammals occupying these regions are particularly sensitive (e.g., endemic or intrinsically vulnerable). Future studies that aim to disentangle the role that these mechanisms play at finer scales are important and would be useful to complement previous global prioritization scenarios (Eklund et al. 2011, Visconti et al. 2011. Finally, we would like to discuss some limitations of our study. First, we could not explore causal relationships or establish which specific human actions associated with socioeconomic development are directly responsible for the increased vulnerability. Nevertheless, our results lead to interesting follow-up questions such as: What are the threats and processes that occur in rural countries that lead to increase mammalian vulnerability? What are the specific activities related to the exports of goods and services that are so damaging for mammals? What underlying factors make countries with high levels of international tourism less likely to contain threatened mammals? Although we do not know the answers, and often lack the data to explore the questions, our study provides guidance on key issues that need to be addressed. Second, our analyses are based on countries that comprise widely different areas (2-16,380,000 km 2 ) that may not be well-represented by average values of direct descriptors of land use or environmental characteristics. This could be the reason why our final models do not include indicators, other than percentage of arable land, more directly linked to local land use changes. Lotz and Allen (2013) conducted a similar country-level study of vulnerability to socioeconomic factors and identified some land use variables as relevant, including agricultural intensity and surface of protected area. Our results likely differ from those of Lotz and Allen (2013) due to methodological differences: we use a hierarchical model building approach to maximize data use, tested both linear and quadratic relationships, and analyzed countries with and without threatened species separately. In addition, Lotz and Allen (2013) evaluated a different subset of indicators using a different subset of socioeconomic indicators and also including variables that summarized ecological features of analyzed countries, highlighting the importance of careful variable selection and hypotheses driven analyses. Finally, we would like to note that the lack of socioeconomic information for some countries is likely limiting our full understanding of reality, as analyses may exclude potentially key factors for which information is simply not currently available.
In conclusion, our results provide a global comprehensive characterization of the socioeconomic profiles of countries with more (and less) threatened mammalian fauna. Future work would be necessary to identify the specific human actions that cause increased number of threatened species and thus, to provide direct management recommendations. It would also be enlightening to explore the historical processes that have triggered current conservation status. Some of those countries lacking threatened mammals may actually have lost their most vulnerable species and now appear as better preserved areas. Conversely, some of the countries with many threatened species could in fact be acting as refuges for species that were originally more widespread and now can only persist in these areas. Meanwhile, these profiles can help us identify human development issues that may be particularly worrisome but are not yet well-recognized. For example, our analyses emphasize the role of globalization for mammalian conservation status. Our attention is often focused on human activities occurring at the same site as the environmental damage, while we forget that in today's globalized world, drivers located far away may be responsible for many of the observed changes. Many developed countries have a relatively well-protected fauna; however, the impact of their activities and policies extends to other countries. The effect of transborder impacts has only been explicitly addressed recently (Meyfroidt et al. 2010, Lenzen et al. 2012), yet these impacts likely play an important role in conservation.

ACKNOWLEDGMENTS
We are grateful to the IUCN Red List and the World Bank for making their databases freely available online. We also acknowledge two anonymous reviewers for comments on a previous version of this manuscript. This work was funded by the program 'Junta para la Ampliación de Estudios' (JAEPre022. BOE-A-2011-10745, co-  v www.esajournals.org Table A4. Definitions and sources of the variables considered for analyses grouped by modeling categories, including those excluded due to their high correlation with other indicators (see Table A2). All data can be accessed on http://data.worldbank.org/. For socioeconomic variables, we provide the World Bank's definition. World Bank national accounts data and OECD National Accounts data files Imports of goods and services Imports of goods and services represent the value of all goods and other market services received from the rest of the world. They include the value of merchandise, freight, insurance, transport, travel, royalties, license fees, and other services, such as communication, construction, financial, information, business, personal, and government services. They exclude compensation of employees and investment income (formerly called factor services) and transfer payments.

Definition and description of biodiversity conservation status estimates and socioeconomic indicators used in the analyses
World Bank national accounts data and OECD National Accounts data files Exports of goods and services Exports of goods and services represent the value of all goods and other market services provided to the rest of the world. They include the value of merchandise, freight, insurance, transport, travel, royalties, license fees, and other services, such as communication, construction, financial, information, business, personal, and government services. They exclude compensation of employees and investment income (formerly called factor services) and transfer payments.
World Bank national accounts data and OECD National Accounts data files GDP per capita growth Annual percentage growth rate of GDP per capita based on constant local currency. GDP per capita is gross domestic product divided by midyear population. GDP at purchaser's prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products. It is calculated without making deductions for depreciation of fabricated assets or for depletion and degradation of natural resources.
World Bank national accounts data and OECD National Accounts data files GDP per capita, PPP GDP per capita based on purchasing power parity (PPP). PPP GDP is gross domestic product converted to international dollars using purchasing power parity rates. An international dollar has the same purchasing power over GDP as the U.S. dollar has in the United States. GDP at purchaser's prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products. It is calculated without making deductions for depreciation of fabricated assets or for depletion and degradation of natural resources. Forest area Forest area is land under natural or planted stands of trees of at least 5 m in situ, whether productive or not, and excludes tree stands in agricultural production systems (for example, in fruit plantations and agroforestry systems) and trees in urban parks and gardens.
Food and Agriculture Organization, electronic files and web site Mineral rents Mineral rents are the difference between the value of production for a stock of minerals at world prices and their total costs of production. Minerals included in the calculation are tin, gold, lead, zinc, iron, copper, nickel, silver, bauxite, and phosphate.
Estimates based on sources and methods described in ' Total fertility rate represents the number of children that would be born to a woman if she were to live to the end of her childbearing years and bear children in accordance with current age-specific fertility rates.
( Health expenditure per capita, PPP Total health expenditure is the sum of public and private health expenditures as a ratio of total population. It covers the provision of health services (preventive and curative), family planning activities, nutrition activities, and emergency aid designated for health but does not include provision of water and sanitation. Data are in international dollars converted using 2005 purchasing power parity (PPP) rates.
World Health Organization National Health Account database (www.who. int/nha/en) supplemented by country data Improved sanitation facilities Access to improved sanitation facilities refers to the percentage of the population with at least adequate access to excreta disposal facilities that can effectively prevent human, animal, and insect contact with excreta. Improved facilities range from simple but protected pit latrines to flush toilets with a sewerage connection. To be effective, facilities must be correctly constructed and properly maintained.

APPENDIX B
Number of threatened and non-threatened species per country and degree of agreement between national and global Red Lists (TOT, total of species listed for that country by the IUCN; DD, data deficient; CR, critically endangered; EN, endangered; VU, vulnerable; LC, least concern; NT, near threatened; Thr (Threatened), addition of the species classified as vulnerable, endangered and critically endangered; Non-thr (Non-threatened), addition of the species classified as least concerned and near threatened; Pred Prsc (prediction of presence), predicted probability of harboring threatened mammals provided by the fitted model. Bold values indicate probabilities lower than 0.5, meaning that these countries are predicted to have no threatened species. Pred Abdc (prediction of abundance), predicted abundance of threatened mammals according to the fitted model. (A) Arrows indicate if the abundance model overestimates (") or underestimates (#). Dashes indicate lack of data for the variables included in the final fitted model. Extinct in the wild (EW) and extinct (EX) species are out of the analyses. SIDS refers to countries classified as Small Island Developing States.

APPENDIX D
Results of model predicting presence and abundance of threatened species by country considering data deficient species (DD) as threatened

Presence of threatened species
The variable selection procedure did not allow obtaining a global minimum adequate model (MAM), since the number of selected variables from categorical models (7 linear; 5 quadratic) was too high for the reduced sample size of countries with none threatened mammals (considering DD species as non-threatened 27 countries had zero threatened species; with DD as threatened that number is reduced to 24). Therefore, we used the selected variables for the model including DD species as non-threatened (Table 1 of the main text) and fitted it for the new set of data to check for coincident results.
This model (Table D1) explains 61.3% of the deviance: 34.2% by control variables and 27.1% by the socioeconomic indicators. Performance measures were satisfactory, but lower than in the conservative model (DD as non-threatened) (sensitivity ¼ 0.950; specificity ¼ 0.759; AUC ¼ 0.967).

Abundance of threatened species
The final MAM (Table D2) includes the same variables as the MAM considering DD species as non-threatened (Table 2 of the main text) and adds three more: population growth (% annual), CO 2 emissions and international expenditures on tourism (% imports). A rapidly growing country, with relatively low CO 2 emissions and extreme (either very low or very high) levels of international tourism expenditures appears also more susceptible to harbor higher numbers of threatened mammals (Fig. D1), which generally agrees with the profile of countries described in the main text.
This model (Table D2) explains 85.3% of the deviance: 77.3% by control variables and 8.0% by socioeconomic indicators. Predictions errors are within a 6 0.33% of total mammal richness per country, with four countries being estimated over this value: Indonesia, Seychelles, New Zealand and Mauritius.  D1. Predicted relationships between key socioeconomic variables and the abundance of threatened mammals by country (DD species classified as threatened). Model predictions were based on the final model (Table D2) and estimated by exploring the range of observed values for each indicator while using the median observed value for other variables in the model (median values: total mammal richness ¼ 129; land area ¼ 192,530 km 2 ; percentage of shared species ¼ 0.759; urban population ¼ 55.90%; population growth ¼ 1.32%; international tourism, receipts ¼ 9.18% exports; international tourism, expenditures ¼ 5.43% imports; exports of goods and services ¼ 36.45% GDP; imports of goods and services ¼ 42.64% GDP; CO 2 emissions ¼ 1.872 metric tons per capita; life expectancy ¼ 71.38). Shadowed area represents the confidence intervals (95%). Singapur was removed from graphs E and F to facilitate visualization, given the extremely high values it presents for these two indicators (imports of goods and services (% GDP) ¼ 200.452; exports of goods and services (% GDP) ¼ 228.007). Kuwait was removed from graph G (CO 2 emissions ¼ 35.42 metric tons/capita).

APPENDIX E
Geographic representation of the estimated values for all variables included in any final model. We show values for all countries with data available on the World Bank database even those not included in the final models (due to missing data on some of the selected variables).        v www.esajournals.org