Socio-economic predictors of environmental performance among African nations

Socio-economic changes in Africa have increased pressure on the continent’s ecosystems. Most research investigating environmental change has focused on the changing status of specific species or communities and protected areas, but has largely neglected the broad-scale socio-economic conditions underlying environmental degradation. We tested national-scale hypotheses regarding the socio-economic predictors of ecosystem change and degradation across Africa, hypothesizing that human density and economic development increase the likelihood of cumulative environmental damage. Our combined environmental performance rank includes national ecological footprint, proportional species threat, recent deforestation, freshwater removal, livestock density, cropland coverage, and per capita emissions. Countries like Central African Republic, Botswana, Namibia, and Congo have the best relative environmental performance overall. Structural equation models indicate that increasing population density and overall economic activity (per capita gross domestic product corrected for purchasing-power parity) are the most strongly correlated with greater environmental degradation, while greater wealth inequality (Gini index) correlates with better environmental performance. This represents the first Africa-scale assessment of the socio-economic correlates of environmental degradation, and suggests that dedicated family planning to reduce population growth, and economic development that limits agricultural expansion (cf. intensification) are needed to support environmental sustainability.


Results
A non-parametric (Kendall's τ) correlation matrix among the component environmental metrics demonstrated only weak or moderate (most τ ≤ |0.385|) relationships among variables (Table 1), so we elected to keep all hypothesized correlates in the saturated (i.e., including all hypothesized correlates) structural equation model. However, there was a reasonably strong correlation (−0.523) between freshwater removal and forest loss among countries -a lack of an obvious mechanistic link between the two suggests that neither can be excluded (Table 1). After calculating the geometric mean rank of countries for which there were at least seven component environmental variables, the top five countries for best environmental performance were (in order) Central African Republic, Botswana, Namibia, Congo, and Democratic Republic of Congo (Table 2; Fig. 1). The five worst environmental performers (in order of worst to less bad) were: Morocco, Algeria, Swaziland, South Africa, and Ghana (Table 2).
We also tested the sensitivity of the final structural equation model results to variation in the minimum number of environmental indices used to construct the composite environmental performance index. Including all 8 environmental indices (ecological footprint 48 , megafauna conservation index 49 , relative species threat; freshwater removals; recent proportional forest loss 50 , livestock per hectare of arable land, extent of permanent croplands, greenhouse-gas emissions in 2013) reduced the number of countries in the ranking from 48 ( Table 2) to 41 (Table 3). the strongest predictor (i.e., appearing the most often in highest-ranked and highest goodness-of-fit models) of the composite environmental rank among African countries was population density (Table 3; see also Supplementary Information Methods and Results Section 3, Fig. S1 and Tables S3-S5, and Section 11, Fig. S5 for results from general linear mixed-effects models and boosted regression trees, respectively; these alternative modelling approaches takes either potential spatial autocorrelation or continuous responses into account, respectively), such that environmental performance (smaller rank) increased as a country's population density decreased (Fig. 3a). While the top-ranked models with sufficient goodness-of-fit indicated that the land area under protection, wealth (GDP), and wealth disparity explained some additional variation in environmental rank (Table 4), the single-parameter explanatory models for these variables indicated weak relationships (Table 4; see  also Supplementary Information Methods and Results Section 3, Tables S3-S5). Nonetheless, environmental rank improved to some extent as the proportion of the land area under protection increased (Fig. 3b), and it decreased as wealth distribution become more even (Fig. 3c) and per-capita GDP (wealth) increased. Re-running the structural equation models using the original configuration of the environmental performance index, but requiring all eight environmental variables in the calculation of the environmental performance rank (from Table 3), there was a slight shift in the top-ranked model (Table 5), but overall the main conclusions were still supported. This analysis resulted in 34 countries (cf. 38 countries for the less-stringent criterion of 7 of 8) environmental variables being considered (Table 5).

Discussion
It is simultaneously telling and disconcerting that none of the Sustainable Development Goal targets, nor any of the Aichi Biodiversity Targets, mentions reducing human population size as a pathway to achieving their goals, even though the United Nations promotes family planning as a means to empower people and develop nations 51 . Our finding that the strongest predictor of environmental performance among nations in Africa is population density means that countries with the most people per unit area suffered relatively more environmental degradation on average. This result brings into question the reality of the United Nations' Sustainability Development Goals (www.un.org/sustainabledevelopment) -particularly Goal 15 ('Sustainably manage forests, combat desertification, halt and reverse land degradation, halt biodiversity loss'), as well as the Convention on Biological Diversity's Aichi Biodiversity Targets (www.cbd.int/sp/targets) Strategic Goals A (' Address the underlying causes of biodiversity loss by mainstreaming biodiversity across government and society') and B ('Reduce the direct pressures on biodiversity and promote sustainable use') -without dedicated, well-funded, and large-scale family planning rolled out across the African continent. Indeed, the targets for human development are becoming increasingly connected with those for natural systems and biodiversity 52 , and so we concur that the "… next generation of [human development and policy] scenarios should explore alternative pathways to reach these intertwined targets, including potential synergies and trade-offs between nature conservation and other development goals" 52 .
Combined with the stagnation of natural fertility decline in Africa compared to other developing regions of the world 10 , there has therefore never before been a more important time to re-invigorate the need for long-term, culturally sensitive, and meaningful family-planning measures if many African nations are to have any hope of stemming the decline of their biodiversity. This is particularly urgent for countries such as Nigeria (~187 million inhabitants in 2016; fertility = 5. Fertility rates particularly in sub-Saharan Africa remain high, in part due to high poverty, low education 53 , and high child mortality 10 , thus resulting in a desire for large family sizes 54 . In Western Africa in particular, the adoption of contraception has been slow due to pervasive attitudinal resistance 55 , even though there is still considerable unmet demand 56,57 . As such, many national governments in Africa have not prioritised family-planning programs 54 ; yet, well-designed family planning with regionally and culturally specific approaches (e.g., traditional www.nature.com/scientificreports www.nature.com/scientificreports/ methods, spacing designs) 56 allows people to regulate their reproduction, with well-established benefits for family welfare 58 , national economies 58 , and the environment 54 . For example, countries like Botswana, South Africa, and Zimbabwe benefited from early adoption of population policies and family-planning programs 56 . One culprit for slow or stalled implementation elsewhere is that early deaths from the HIV/AIDS epidemic -while having limited demographic impact partly because of antiretroviral availability -have nonetheless shifted emphasis away from family planning 54 . It is therefore undeniable that African citizens and their governments would benefit from placing greater emphasis on quality family planning, a conclusion that we have also reached with respect to Africa's environmental integrity 54 .
Some past investigations of the relationships between human population size/density and measures of environmental status have been equivocal [22][23][24][25][26][27][28][29][30][31][32]37 suggesting that issues of spatial and temporal scale, as well as the choice of environmental indicator, have bearing on the strength of evidence arising. At the national scale in Africa, human population density most likely reflects the current state of environmental performance because of the relative uniformity among the sample of nations regarding the timing of principal environmental change, as well as the rapid recent expansion of human populations in many countries in that region 9,10 . A fundamental tenet of population ecology is that per-capita resources decline as populations near carrying capacity 59 , so the absolute pressure on the environment is dictated more by variation in a country's 'carrying capacity' than absolute population size or per capita resource use per se 37 . Nonetheless, population density in the African context appears to be a reasonable reflection on average of an individual country's proximity to this moving carrying-capacity target, despite localized improvements in biodiversity following fence construction 60 , for example.
Previous country rankings for environmental performance 37 have not incorporated indices of leakage (externalizing environmental damage via pollution trading and outsourcing environmentally intensive production processes), although it is debatable whether it would make a large difference in the African context because of the relatively lower developed state of many of its nations compared to large consumers such as China, USA, and Brazil 37 . However, because we included each nation's ecological footprint in our derivation of a composite environmental performance indicator, this should at least partially account for some aspects of leakage. Another potential caveat is that our modern 'snapshot' of the trends driving environmental degradation among African nations is likely to vary temporally, such that older comparisons could reveal alternate patterns. However, data for the variables we used to construct our analyses are largely unavailable and/or incomparable for periods vastly older than our current dataset.
It is unsurprising that per-capita wealth (GDP) had the hypothesised effect on a country's relative environmental performance rank, especially considering that at the global scale at least, rising GDP reduces environmental performance among nations 37 . That same analysis 37 also found no evidence to support the environmental Kuznets curve 61 -the hypothesis that a U-shaped relationship exists between environmental degradation and per-capita wealth. This hypothesis predicts that beyond a certain threshold, wealthier societies begin to reduce their environmental footprints. However, the evidence for the environmental Kuznets curve is equivocal 62 , depending on which metrics are measured, countries examined, and periods of development history 61,[63][64][65][66][67][68][69][70] . Examining the bivariate plot between environmental performance rank and per-capita GDP rank (Fig. 3d) might suggest columbia.edu/data/collection/grump-v1/methods). Each country (3-letter ISO country codes given in Table 2) is also shown with its approximate mid-2016 total human population size (Population Reference Bureau; www. prb.org) in millions. (b) African countries shaded according to relative environmental performance (darker green indicates better relative environmental performance; see Table 2  www.nature.com/scientificreports www.nature.com/scientificreports/ a U-shaped relationship; however, examined appropriately by partialling the effects of the other socio-economic variables using a boosted regression tree approach that can identify nonlinearities, there is no evidence of a U-shaped relationship (Supplementary Information Methods and Results Section 12, Fig. S6).
It is not clear why governance quality consistently emerges as a weak predictor of environmental performance 37 . This conclusion exists even after using an African-specific indicator of governance quality 71 , possibly because governance problems in environmental custodianship might only become clear at finer spatial scales, perhaps only at regional or protected-area levels 32 . Alternatively, because governance quality tends to be ubiquitously low across the African continent relative to elsewhere 72 , the low inter-country variation in this metric likely diminishes the power to identify a correlation with environmental performance. The weak, yet statistically  www.nature.com/scientificreports www.nature.com/scientificreports/ supported relationship between environmental rank and wealth disparity was as predicted -increasing wealth disparity leads to better environmental performance. This relationship might seem counter-intuitive, but there is evidence that when democratic processes are restricted, a less equal income distribution generates less environmental degradation 73,74 . The observed relationship most likely arises because greater inequality in wealth among citizens likely engenders fewer opportunities for development of natural resources, thus hindering or at least delaying environmental damage 45 .
In conclusion, our results strongly support the idea that a sustainable approach to biodiversity conservation in Africa over the coming decades cannot be limited by a narrow perspective that treats different development goals of well-being and environmental custodianship as separate entities if they ignore issues of sustained human population growth 52 . Indeed, with the mounting pressures facing Africa's ecological systems, continued environmental degradation will impose further negative feedbacks on human well-being, because human quality of life is fundamentally tied to the healthy functioning of ecosystems 52 . Of course, better education, poverty alleviation, technological advances, and participation in multilateral environmental agreements could restrict land-use change and consumption rates and patterns; however, while there are many policy levers that African nations can use to improve the future state of their environments and the societies that depend on them, limiting excessive human population growth will, on average, likely facilitate better environmental custodianship. Methods environmental data. Our goal was to define an African-relevant composite environmental indicator rank for each nation on the continent. While there are many ways to measure a nation's environmental performance, there are more regionally and temporally relevant measures that attest to the specific environmental histories of regions. We therefore reasoned that given the recent colonial history of many African nations, the recent spike in human population sizes, rapid development investment over the last few decades, a rich diversity of megafauna under substantial threat from agricultural expansion and poaching 39 , and an emphasis on primary production  Table 3) where a nation's environmental performance rank (ENV; low rank = best relative environmental performance) is positively correlated with population density (POPD), and negatively correlated with gross domestic product (GDP, corrected for purchasing-power parity), and Gini wealth inequality index (GINI). Numbers on the directional pathways indicate standardized coefficients for each relationship. (b) There is also some modest evidence for a positive effect of proportion of land area under protection (PROT) (see third-ranked model in Table 3). One-way and two-way correlations among predictor variables also shown. POGR = population growth rate. www.nature.com/scientificreports www.nature.com/scientificreports/ (cropping, livestock husbandry), that the following available indicators would be ideal to construct a composite environmental index for African nations: ecological footprint (footprintnetwork.org), megafauna conservation index 49 , IUCN Red List species threat (iucnredlist.org), freshwater removal (data.worldbank.org), forest loss 50,75 , livestock density (fao.org/faostat), cropland extent (data.worldbank.org), and greenhouse-gas emissions (data. worldbank.org). We provide a full description of each indicator in the Supplementary Information (Section 1).

Combined environmental performance indicator.
For each environmental variable, we made simple hierarchical rankings (i.e., we did not consider the magnitude of the differences among absolute values between countries to avoid issues related to heteroscedasticity, non-linearity, and non-Gaussian distributions) using the rank function (means averaged) in R 76 . To construct a mean rank across all seven variables, we calculated geometric mean rankings for countries 37 where at least seven of the eight variables were available to provide a measure of relative distance between countries in the final composite rank. We argue that a 'seven out of eight' criterion maximizes sample size (number of countries) without compromising the meaningfulness of the combined index (see Tables 3 and 5 for a sensitivity analysis of this choice). This ranking approach also avoids the undue influence of outliers (i.e., analogous to a geometric mean) 77 Table 2.
www.nature.com/scientificreports www.nature.com/scientificreports/ socio-economic data. For a detailed description of the socio-economic variables and associated hypotheses, see Supplementary Information Section 2. In summary, we accessed the World Bank database for the estimated human population size for African nations in 2015, dividing this value by total land area per country to calculate a human population density (data.worldbank.org). We hypothesized that increasing human density would lead to greater pressure on environmental resources 25 , thus lowering a country's environmental performance rank. We also calculated the mean annual human population growth rate from 1960 to 2015 for African nations from the World Bank (data.worldbank.org), hypothesizing that faster mean population growth would hasten the exploitation of a country's resources relative to slower-growing nations 25 .
Also from the World Bank, we accessed each country's gross domestic product (GDP) per capita (corrected for purchasing-power parity) as an index of total wealth. Some countries were missing GDP estimates for certain    Table 5. Structural equation models considered in the model set correlating socio-economic variables to the composite geometric mean environmental ranking among countries (n = 34; reduced set of countries from Table 3 www.nature.com/scientificreports www.nature.com/scientificreports/ years, so we took the mean of values from 2011-2015 as an indication of mean per-capita GDP to maximize the sample size of countries considered. Previously, we showed that a country's total wealth leads to a lower environmental performance (i.e., more degradation) 37 . Also from the World Bank, we accessed an index of wealth distribution using the Gini index from 2005 to 2014 (again, taking the mean of values across this period to maximize sample size). We hypothesized that the greater a country's inequality in wealth across its citizenry, the lower the environmental damage that would ensue due to higher poverty and less overall development 45 .
We also hypothesized that poorer overall governance would lead to higher likelihood of environmental exploitation based on previous work linking it to environmental degradation 46,47 (although at a global scale, declining governance quality had little impact on national-scale environmental performance) 37 . We used the Overall Governance Score from the 2015 Ibrahim Index of African Governance 71 , which includes measures of safety and rule of law, participation and human rights, sustainable economic opportunity, and human development indicators in its normalized overall score.
Finally, we hypothesized that a country's commitment to protecting its native species, expressed through the proportion of its total land area under some form of protection, would lead to great environmental performance 33 . However, it is not part of the composite environmental performance indicator because the amount or number of protected areas does not necessarily translate into lower extinction rates 33 . To this end, we accessed the percentage of land under protected-area status for each country from the Population Reference Bureau (pbr.org), which is originally sourced from the World Database of Protected Areas (protectedplanet.net).

Structural equation models.
To account for inter-correlations among hypothesized socio-economic explanatory variables 37 , we applied structural (path) equation models to model the hypothesized relationships 78 . We constructed thirteen candidate models (see Results Table 3) to examine the socio-economic drivers of environmental rank among African countries, keeping the hypothesized relationships between socio-economic variables constant in all. These were: (a) a two-way correlation between human population density and growth rate, based on the assumption that compensatory density feedbacks operated between these two population variables; (b) a two-way correlation between governance score and per-capita GDP; (c) a two-way correlation between per-capita GDP and wealth distribution; (d) a one-way correlation between population density and per capita GDP; and (e) a one-way correlation between governance quality and the proportion of the landscape under some form of protection (see Results for schematic). Prior to fitting, we investigated the non-parametric ordinal rank correlations using Kendall's τ because we used ranks in all models. We fitted the candidate path models to the data using the sem function 79 implemented in the R Package 76 , calculating Bayesian information criterion (BIC) weights to assign relative strength of evidence to each model in the set. We evaluated the goodness-of-fit of each model using McDonald's non-centrality index 80 and Bollen's incremental fit index 81 using the semGOF library in R, both of which should be >0.90 to consider a model's fit to be acceptable 81 . We also considered structural equation models using single environmental indicators to examine which elements of environmental change were most influenced by variation in socio-economic conditions (Supplementary Information Methods and Results Section 8, Table S10). We also considered only the 'biodiversity' components (i.e., megafauna conservation index 49 , IUCN Red List species threat, and forest loss 50,75 ) to create a second composite environmental rank to determine its relationship to the socio-economic correlates in isolation from the other 'agricultural' (freshwater removal, livestock density, and cropland extent) and economic (ecological footprint, and greenhouse-gas emissions) components of the environmental performance rank (Supplementary Information Methods and Results Section 9, Table S11). We also considered a country's poverty gap (percentage of people below the relevant country's poverty threshold -data from the World Bank) instead of the Gini index as a measure of wealth inequality (Supplementary Information Methods and Results Section 10, Table S12). These models included fewer countries (34), had generally poorer fits, but supported the dominance of population density as the most important correlate (Table S12).
Boosted regression trees. Finally, we considered the absolute differences between the values comprising the environmental performance metric, as well as those between the predictor values (cf. ranks) to examine whether ranking -despite its advantages for avoiding unequal variances, non-linearities, and non-Gaussian behaviour -resulted in substantially different conclusions. We therefore used the same data that we obtained to derive the rankings, but instead scaled and centred the data for each composite environmental metric, and then took the median value to derive a new, continuous-variable environmental-performance metric. Next, we scaled and centred the socio-economic predictor variables in the same manner, and then tested for relationships as we did for the ranked data. However, even scaling and centring could not remove potential problems of non-Gaussian distributions (Supplementary Information Methods and Results Section 11, Figs S2-S4), so we employed boosted regression trees 82 instead to test the relationships (Supplementary Information Methods and Results Section 11, Fig. S5).